Endothelial RBPJ Is Essential for the Education of Tumor-Associated Macrophages

High-grade serous ovarian cancer is the deadliest of all gynecologic cancers (1). Most women have already developed peritoneal metastasis at the time of diagnosis. Epithelial ovarian cancer (EOC) cells can directly infiltrate the peritoneal cavity to seed metastases, a process called transcoelomic metastasis (2). Metastatic EOC cells initially reside in the omentum, where they undergo adaptations allowing them to spread throughout the whole peritoneal cavity. Importantly, this microenvironment is so immunosuppressive that even the infiltration of effector T cells does not per se correlate with better prognosis (3).

Peritoneal spread of EOC cells is accompanied by monocyte-derived macrophages (MN-derived macrophages) recruitment, which eventually become the most abundant myeloid cell type (4) and are a major contributor to the immunosuppressive tumor microenvironment (TME; ref. 3). Upon recruitment from blood to the tumor, monocytes differentiate into macrophages, which are further educated by the TME. Eventually, tumor-associated macrophages (TAM) strongly promote progression of metastatic ovarian cancer (5). However, little is known about the contribution of other stromal cells in this process.

Monocytes must cross the vascular endothelial barrier before infiltrating peritoneal organs or the peritoneal fluid. Endothelial cells (EC) not only form tubes for the transport of blood, but also produce soluble factors controlling the differentiation and function of adjacent cells. These angiocrine functions are highly context and organ specific (6, 7). In cancer, ECs provide angiocrine factors that influence tumor progression (8). Therefore, we hypothesized that monocytes get influenced by ECs while infiltrating into peritoneal tissues.

Notably, tumors can alter the transcriptome of ECs (9, 10) and this may influence transmigrating monocytes. For example, endothelial Notch signaling activity is frequently higher in tumors like EOC and in the metastatic niche compared with ECs from nontumorous tissue (11). Notch signaling is a highly conserved cell-to-cell communication system. Ligand binding induces cleavage of the transmembrane Notch receptors releasing their intracellular domain (ICD), which enters the nucleus to alter gene transcription. This canonical signaling pathway relies on the DNA-binding protein RBPJ, which turns into a transcriptional activator upon binding of a Notch receptor ICD (12). Sustained endothelial Notch1 signaling is associated with increased myeloid cell infiltration and metastasis (11). The Notch pathway in ECs is a major regulator of angiogenesis, angiocrine functions, and tumor cell transmigration (11, 13–18). Although EOC cells do not necessarily have to cross the endothelial barrier to spread throughout the peritoneum, we hypothesized that endothelial Notch signaling could still influence EOC progression via transmigrating myeloid cells.

Here, we provide insights into the essential role of endothelial RBPJ for the recruitment of monocytes and their proper education into protumoral TAMs in metastatic EOC.

Tissue microarray of EOC samples were provided by the tissue bank of the National Center for Tumor Diseases (NCT, Heidelberg, Germany) in accordance with the regulations of the tissue bank and the approval of the ethics committee of Heidelberg University (Heidelberg, Germany).

Female C57BL/6 mice were group housed under specific pathogen‐free barrier conditions. All animal procedures were approved by the local Institutional Animal Care and Use Committee (RP Karlsruhe, Germany and DKFZ) and performed according to the guidelines of the local institution and the local government.

Administration of tamoxifen diluted in sterile peanut (P2144, Sigma-Aldrich) in 8 to 12 weeks old randomized mice was performed by oral gavage once with 100 μL (1 mg tamoxifen; ref. 19). Control mice, which did not express Cre^ERT2 were treated with tamoxifen.

Model of ovarian cancer: Three weeks after gene recombination, 5 × 10⁶ ID8-luciferase (ID8-luc) ovarian cancer cells were administered intraperitoneally in PBS. For peritoneal lavage after sacrificing the mice, 5 mL of ice-cold PBS (Gibco/Thermo Fisher Scientific) was injected intraperitoneally, and after a careful massage to mobilize cells, peritoneal fluid was collected. For analysis of ID8-luc tumor growth, the cell suspension was centrifuged and supernatant was collected. The cell pellet was suspended in 1 mL PBS and 100 μL were used to determine the luciferase activity. A total of 100 μL cell suspension were centrifuged and the cell pellet was suspended in 100 μL lysis buffer (Promega) and 20 μL of lysed cells were pipetted into white 96-well plate in triplicates. A total of 50 μL of LAR substrate (Promega) was added to the lysed cell suspension and luminescence signal was determined using the plate reader (ClarioStar, BMG Labtech). For analysis of immune cell recruitment into the peritoneal cavity, the collected peritoneal cell suspension was centrifuged and red blood cells in the cell pellet were lysed with 1 mL ACK (Thermo Fisher Scientific). After washing, the cell suspension was counted and 1 × 10⁶ cells were used for flow cytometry staining.

Model of peritoneal inflammation: To obtain newly recruited peritoneal macrophages, 1 mL thioglycolate (2 mg/mL in H₂O; B2551, Sigma-Aldrich) was injected into the peritoneum 3 weeks after gene recombination. MN-derived macrophages were isolated 24 or 72 hours after thioglycolate injection by their adherence to nontreated plastic petri dishes. Briefly, after incubation of single cells in a petri dish for 30 minutes at 37°C, nonadherent cells washed away with PBS.

Lewis lung carcinoma (LLC) model: One week after gene recombination, 5 × 10⁵ syngeneic LLC cells in PBS were subcutaneously injected into the flank of mice. Tumor volume was calculated by following formula: V = (length)(width²/2).

Mouse cardiac ECs (MCEC) were purchased from tebu-bio and cultured on gelatin-coated surfaces in in DMEM containing 1 g/L d-glucose, 5% FCS, 5% HEPES, 100 units/mL penicillin, and 100 μg/mL streptomycin.

Human umbilical vein EC (HUVEC) were grown and maintained until passage 5 in Endopan-3 growth medium containing 3% FCS and supplements (Pan-Biotech).

SK-OV-3 cells (ATCC) were cultured in DMEM containing 1 g/L d-glucose, 10% FCS, 5% HEPES, 100 units/mL penicillin, and 100 μg/mL streptomycin (Last authentication March 2022, Eurofins Genomics).

ID8-luc cells, kindly donated by Prof. Frances Balkwill (Barts Cancer Institute, London, United Kingdom) were cultured in DMEM containing 1 g/L d-glucose, 10% FCS, 5% HEPES, 100 units/mL penicillin, and 100 μg/mL streptomycin.

Cell cultures were tested by PCR on a regularly basis for Mycoplasma contamination periodically and before injecting the tumor cells into the mice.

HUVECs were infected with lentivirus constructs for CRISPR/Cas9 (Addgene plasmid #52961, lentiCRISPR v2) to knockout for Rbpj (20). After 48 hours of puromycin selection, cells were infected with adenovirus (Life Technologies; pAD/CMV-V5-DEST) overexpressing GFP as control or N1ICD.

RNA isolation from cell culture was performed using the InnuPrep Mini Kit (Analytik Jena). RNA isolation from tissues was performed using PicoPure RNA Isolation Kit (Acturus, Life Technologies). A total of 1 mL TRIzol was added to the tissue and homogenized for 1 minute and a frequency of 30/second. After disruption of the tissue, 200 μL chloroform were added and mixed by inverting several times followed by a centrifugation step for 15 minutes 12,000 × g at 4°C. Further RNA isolation steps were performed using the manufacture's protocol. Reverse transcription of isolated RNA into complementary DNA was performed using High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific). qPCR was performed with SYBR Green PCR mix (Applied Biosystems) on a QuantStudio3 Real-time PCR system (Applied Biosystems). Resulting fold changes were calculated using the 2^ΔΔC_T method and mRNA expression was normalized to the housekeeping gene (Cph for murine and HPRT for human samples). For specific sequences, see Supplementary Tables S1 and S2.

To analyze the T-cell killing potential by the lactate dehydrogenase (LDH)-Cytotoxicity Assay Kit (Ab65393, Abcam), ID8 cells (7,500 ID8 cells in 100 μL DMEM) were seeded in a 96-well plate one day before T-cell sorting. T cells were sorted from tumor-bearing Rbpj^iΔEC and control mice after 6 weeks of tumor growth and 10,000 CD3⁺ cells were cocultured with ID8 cells in technical triplicates including untreated control and blank. After overnight incubation, cytotoxicity and killing potential was measured by LDH amount in the cell supernatant using the LDH-Cytotoxicity Assay Kit (Ab65393, Abcam).

To determine RBPJ binding sites (5´-GTGGGAA-3´; ref. 21) in the promotor region of the murine (NM_009140; chr5+:90902580-90903927) and human (NM_002089; chr4-:74100502-74099123) CXCL2-encoding gene, we used ApE plasmid Editor by M. Wayne Davis (https://jorgensen.biology.utah.edu/wayned/ape/).

We used the Search Tool of Interacting Genes/Proteins database (STRING v11.5) to perform in silico protein–protein interaction analysis (22). Given CXCR2 protein as input, STRING can search for their neighbor interactors, the proteins that have direct interactions with the inputted proteins; then STRING can generate the protein–protein interaction network consisting of all these proteins and all the interactions between them. All the interactions between them were derived from high-throughput lab experiments and previous knowledge in curated databases at medium level of confidence (score ≥ 0.40).

Gene set enrichment analysis (GSEA; Broad Institute) was used to determine whether a list of genes (gene signature) was enriched between different groups. A defined list of genes exhibits a statistically significant bias in their distribution (FDR) within a ranked gene list using the software GSEA (23), showing enrichment in one of the compared groups (normalized enrichment score) obtained from microarray or public available datasets as indicated in the figure legend.

Ingenuity pathway analysis (IPA) software (Qiagen) was used to identify predicated upstream regulators and differentially regulated pathways in newly recruited macrophages (based on microarray data). For the analysis of data, fold changes were uploaded to the software. Differentially regulated pathways and upstream regulator analysis was performed from obtained microarray data.

Pathway analysis were obtained from public external databases (EnrichR) and analyzed as −log₂-fold changes.

Human ovarian cancer patient bulk tumor RNA-sequencing data were obtained from The Cancer Genome Atlas (TCGA) database. Stratification in CXCL2^high and CXCL2^low patients was performed using R Studio software. Patients were assigned to the different groups using CXCL2 expression below the first or higher than the third quartile. Normalized raw counts of CD44 were plotted comparing the two different groups. For prognosis assessment, kmploter tool was employed (24). We evaluated gene expression in metastatic lesions with the online tool TNMplot (25).

Schematics were created using BioRender.com.

Normality was tested when sample size allowed it. Those samples with normal distribution were compared using Student t test (with Welch correction when groups had different sizes). When normality was rejected, Mann–Whitney U test was used. Comparison analysis was performed with ANOVA with Tukey post hoc test when more than two groups were analyzed. Statistical analysis and the generation of the graphs were performed using GraphPad Prism 9 (GraphPad Software, Inc.).

Microarray data generated in this article have been deposited in Gene Expression Omnibus repository under the accession no. GSE213371.

Details about flow cytometry, whole mount staining, isolation of peripheral blood mononuclear cells, IHC, Western blot analysis, transwell assay, ELISA, bone marrow–derived macrophages (BMDM) differentiation are provided in the Supplementary Materials and Methods.

Metastatic EOC cells seed initially in the omentum and later spread within the peritoneum. During these steps, tumor cells undergo transcriptional changes that allow further growth and colonization (Fig. 1A). The latter is strongly influenced by MN-derived macrophages (26). To determine the contribution of canonical endothelial Notch signaling to myeloid cell infiltration and EOC progression, we used the tamoxifen-inducible VE-cadherin (Cdh5) Cre^ERT2 strain to delete Rbpj specifically in ECs of adult mice (Rbpj^iΔEC). This model is well established and leads to robust gene recombination in ECs of several organs (14, 15, 17, 18) without changes in permeability (Fig. 1B; Supplementary Fig. S1A and S1B; refs. 17, 27). Under physiologic conditions, there were no differences in blood vessel density in omentum upon Rbpj deletion in ECs compared with controls (Supplementary Fig. S1B and S1C). Upon intraperitoneal injection of ID8 cells mimicking metastatic EOC, omenta showed evidence of tumor growth (Fig. 1C) and had significantly higher vessel density in Rbpj^iΔEC mice compared with control animals (Fig. 1D). Endothelial coverage with α-smooth muscle actin-positive mural cells was unchanged (Supplementary Fig. S1D). Despite increased vessel density, 4 weeks after injection tumor burden in omenta of Rbpj^iΔEC mice was significantly lower than that in their littermate controls (Fig. 1E). Moreover, peritoneal spread of tumor cells was significantly reduced in Rbpj^iΔEC mice (Fig. 1F) concluding that deletion of Rbpj in ECs reduces EOC progression and metastasis. Although, tumor burden in omenta was no longer significantly smaller after 6 weeks (Supplementary Fig. S1E), we still observed decreased tumor burden in the peritoneal cavity of Rbpj mice compared with control (Supplementary Fig. S1F).

Next, we analyzed the immune cell compartment of our EOC model because it is a major contributor to tumor progression. High Notch signaling activity in ECs induces VCAM1 expression, which promotes leukocyte extravasation (11, 28). However, VCAM1 expression was unchanged between Rbpj^iΔEC and their controls littermates (Supplementary Fig. S2A). Nevertheless, whole-mount staining of omenta revealed that there was a reduction in immune cells inside tumor nodules of Rbpj^iΔEC mice compared with controls (Fig. 2A; Supplementary Fig. S2B). Interestingly, tumors in control mice contained more cells with a large vacuolated cytoplasm, reminiscent of macrophages. Therefore, we analyzed this cell population in detail. Tumor-bearing omenta from Rbpj^iΔEC mice contained less CD11b⁺ cells within tumor nodules, although this was not statistically significant (Fig. 2B). To better determine the macrophage subpopulations responsible for the observed decrease in CD11b⁺ cells, we studied the myeloid and macrophage populations in the peritoneal cavity. After 4 weeks of tumor growth, there were no significant differences in the total amount of myeloid cells or macrophages (Fig. 2C and D). However, we found decreased numbers of small peritoneal macrophages (SPM), which are MN-derived macrophages characterized as MHC-II^hi/F4/80^low (Fig. 2E) and CCR2⁺/Tim4⁻ (Fig. 2F). Notably, in naïve (tumor-free) conditions, endothelial Rbpj deletion had no effect on peritoneal macrophage populations (Supplementary Fig. S2C–S2F). These data suggest that the tumor-driven recruitment of monocytes into the peritoneum is impaired upon endothelial Rbpj deletion.

To analyze chemotaxis in a more simplified in vitro system, we measured transmigration of human CD14⁺ monocytes through a monolayer of human ECs in a transwell insert. Chemotaxis was stimulated by SK-OV-3 human ovarian cancer cells below the insert (Fig. 2G). RBPJ deletion in HUVECs (Supplementary Fig. S3A) resulted in significantly decreased monocyte transmigration rates compared with control HUVECs (Fig. 2H). This further suggests that canonical Notch signaling in ECs is important to facilitate monocyte recruitment into the TME.

We considered the possibility that the reduced MN-derived macrophage recruitment might only be attributed to the lower amount of EOC cells present in Rbpj^iΔEC mice. To test this hypothesis, we performed a subcutaneous tumor model, where we injected LLC cells in the flank of Rbpj^iΔEC mice and their control littermates and let them grow for 2 weeks (Fig. 3A). We monitored tumor growth to avoid that tumor sizes where different between the two groups at the time of analysis (Fig. 3B). As expected (17), vessel density was increased while vessel coverage was comparable in tumors grown in Rbpj^iΔEC mice (Fig. 3C and D). Nevertheless, these tumors contained significantly less macrophages, in particular MN-derived macrophages (Fig. 3E–H). This indicates that the reduced MN-derived macrophage recruitment into tumors of Rbpj^iΔEC mice is not restricted to a certain organ bed or cancer entity and is not a simple consequence of reduced tumor growth.

In addition, we validated this Notch-mediated myeloid cell recruitment in a tissue microarray of 50 human ovarian cancer samples. There was a positive correlation between NOTCH1 activity (nuclear staining indicating active Notch1 intracellular domain) and tumor infiltration with CD33⁺ cells (Fig. 3I). As such, endothelial Notch/RBPJ signaling is a regulator of myeloid cell tumor infiltration in mice and human.

We transduced RBPJ-deficient HUVEC and control cells with N1ICD (Supplementary Fig. S3B) to determine which of those transcriptional changes are entirely dependent on RBPJ. These were CCL1, CCL21, CXCL12, and CXCL2 (Fig. 4A; Supplementary Fig. S3C and S3D). Next, we measured the mRNA expression levels of these chemokines in peritoneal adipose tissue obtained from Rbpj^iΔEC mice. There was lower Cxcl2 expression in peritoneal adipose tissue from Rbpj^iΔEC mice compared with littermate controls, whereas the other chemokines analyzed were not changed (Fig. 4B). Analysis of the same cytokines in endothelial-specific N1ICD mice (11, 15), revealed that higher endothelial Notch1 signaling activity led to higher Cxcl2 expression in peritoneal adipose tissue (Fig. 4C). We confirmed this Rbpj-mediated gene expression regulation also in isolated lung EC (Supplementary Fig. S3E). Next, we evaluated blood serum levels of CXCL2. In ID8 tumor-bearing mice, we only detected a slight decrease in CXCL2 levels (Fig. 4D). We aimed at generating an acute inflammation in the peritoneum, similar to that happening during tumor growth, to test whether endothelial Rbpj would be required for systemic CXCL2 induction. Rbpj^iΔEC and control mice were injected with thioglycolate to induce MN-derived macrophage recruitment into the peritoneal cavity under tumor-free conditions. Mice lacking endothelial Rbpj had significantly lower CXCL2 amounts in serum (Fig. 4D). This underscores the essential role of canonical endothelial Notch signaling for the production of CXCL2 during inflammation.

In silico analysis showed that the mouse and human Cxcl2 genes contain RBPJ-binding sites in promoter regions (Fig. 4E). Consistently, cultured human ECs secreted higher CXCL2 protein levels after being transduced with N1ICD-expressing adenovirus compared with control (Fig. 4F).

The data presented so far indicate that canonical Notch signaling induces CXCL2 expression in ECs. CXCL2 attracts granulocytes, but also to a lesser extend monocytes (29). As such, CXCL2 is an interesting endothelial Notch target, which could mediate the observed effects on monocyte recruitment into the TME. To evaluate whether CXCL2 was capable of attracting monocytes, we performed transwell chemotaxis experiments and observed that recombinant CXCL2 induced monocyte transmigration (Fig. 4G). Next, we inhibited CXCL2 expression in ECs using short hairpin RNA, which led to an about 50% reduction of CXCL2 protein expression (Fig. 4H). Compared with nonsilencing control, this reduction was already capable of reducing the numbers of monocytes transmigrating through ECs toward SK-OV-3 cells (Fig. 4I). In summary, the data implicate that the endothelial RBPJ/CXCL2 axis contributes to monocyte recruitment into the peritoneum during transcoelomic metastasis.

Recruited monocytes differentiate into macrophages, which are further educated by tumor cells to become tumor-promoting macrophages. Hypersensitivity toward IL4 facilitates macrophage education in EOC. This is induced by tumor cell–mediated cholesterol depletion (4). We sought to understand whether lack of Rbpj in ECs could influence macrophage phenotypes. To assess this, we isolated newly recruited MN-derived macrophages (CD11b⁺/F4/80⁺/CCR2⁺; Fig. 5A) and obtained their transcriptomic profile by microarray analysis. GSEA comparing MN-derived macrophages from Rbpj^iΔEC and control tumor-bearing mice revealed that Rbpj in ECs is necessary to acquire the typical phenotype of TAM in this model of metastatic EOC (Fig. 5B). As such, tumor cells could not fully educate MN-derived macrophages in mice lacking endothelial Rbpj.

Transcriptomic profiling via IPA determined that IL4 was the most downregulated signaling pathway in MN-derived macrophages obtained from Rbpj^iΔEC tumor-bearing mice (Fig. 5C). IPA and pathway analysis showed that genes important for cholesterol synthesis were downregulated in MN-derived macrophages from Rbpj^iΔEC mice (Fig. 5D and E). We then employed a cholesterol homeostasis gene set from TAMs (4), representing EOC-induced cholesterol metabolism in TAMs. GSEA showed that this gene set was significantly enriched in newly recruited macrophages coming from control tumor-bearing mice (Fig. 5F), indicating that Rbpj in ECs is necessary for cholesterol depletion in TAMs. We confirmed these results by studying cholesterol depletion in isolated MN-derived macrophages (thioglycolate injection) from Rbpj^iΔEC and control mice. As expected, we observed that conditioned medium from ID8 ovarian carcinoma cells (ID8-CM) was not able to reduce cholesterol from the membrane of MN-derived macrophages isolated from Rbpj^iΔEC mice (Fig. 5G). Moreover, this effect translated into a lower arginase-1 (Arg1) mRNA expression a well-known IL4 target (Fig. 5H).

Cholesterol depletion is mediated by tumor cell–secreted high molecular weight hyaluronan (HMW-HA). The interaction with HA receptors, such as CD44, in macrophages leads to cholesterol efflux (4). To understand whether Rbpj in ECs could be important for CD44 expression in monocytes, we cocultured human monocytes with HUVECs lacking RBPJ or with respective controls. Monocytes cocultured with RBPJ-deficient HUVECs, had less CD44 on their membrane than those incubated with control HUVECs (Fig. 5I). Next, when incubating BMDM with conditioned medium from immortalized MCECs lacking Rbpj (CRISPR-Cas9 mediated), macrophages expressed less Cd44 (Fig. 5J) and Mmp9 (Fig. 5K), a known CD44 target gene. This indicates that a secreted angiocrine factor regulated by the transcription factor RBPJ in ECs is necessary for the regulation of CD44 in macrophages. To understand whether this would also happen in vivo, we analyzed Cd44 mRNA expression MN-derived macrophages (after thioglycolate injection) and found that those coming from Rbpj^iΔEC mice expressed significantly less Cd44 than those coming from control mice (Fig. 5L), confirming that endothelial Rbpj is essential for Cd44 regulation in MN-derived macrophages in vivo.

Considering the important role that CXCL2 had in monocyte recruitment in Rbpj^iΔEC tumor-bearing mice and that higher level of CXCL2 in serum is associated with myeloid cell infiltration into the TME and worse prognosis for patients with EOC (30), we decided to analyze its role in regulating CD44 expression. When stimulating BMDMs with recombinant CXCL2, Cd44 mRNA (Fig. 5M) and protein (Fig. 5N) were increased. In addition, immunofluorescence staining revealed that CXCL2 stimulation significantly increased the presence of CD44 on the plasma membrane of BMDMs (Fig. 5O; Supplementary Fig. S4), which would consequently increase its accessibility to hyaluronic acid. We further confirmed this altered localization by flow cytometry (Fig. 5P). This was mediated through CXCR2, the most common CXCL2 receptor, as the CXCR2 inhibitor SB225002 blocked the CD44 increase on the plasma membrane (Fig. 5P).

The data presented so far indicate that the angiocrine factor CXCL2, which is under transcriptional control of Notch/RBPJ signaling, preconditions monocytes to be educated by tumor cells in the TME. To evaluate the relevance of this, we analyzed CXCL2 and CD44 mRNA expression levels in publicly available datasets from TCGA. Remarkably, the stratification of patients with ovarian cancer in 25% upper (CXCL2^high) and lower (CXCL2^low) CXCL2 expression showed that CXCL2^high patients displayed significantly more expression of CD44 (Fig. 6A). These data suggest a positive correlation between CXCL2 and CD44 expression in human tumors. Interestingly, CD44 expression is significantly higher in metastatic samples compared with normal or primary tumor samples (Supplementary Fig. S5A). In addition, we performed GSEA comparing CXCL2^high and CXCL2^low patients using the abovementioned TAM signature and found that it was significantly enriched in CXCL2^high patients (Fig. 6B), indicating that the increased CD44 expression in these patients consequently impacts TAM education.

When analyzing differences between CXCL2^hi and CXCL2^low patients, we found the expected upregulation of chemotactic response and myeloid cell recruitment in CXCL2^high patients, as shown by Gene Ontology term analysis (Supplementary Fig. S5B and S5C). Interestingly, we again identified cholesterol metabolism as one of the most downregulated pathways in CXCL2^low patients (Fig. 6C). CXCL2^high patients showed induction of pathways such as lipid metabolism and atherosclerosis (Fig. 6D). Indeed, GSEA showed that the same cholesterol metabolism gene set previously employed was significantly enriched in CXCL2^high cohort (Fig. 6E), indicating a similar cholesterol depletion as the one occurring in peritoneal macrophages of our mouse model.

To explore the relevance of CXCL2^high transcriptional profile for patient survival, we extracted the 10 most enriched genes in CXCL2^high compared with CXCL2^low, obtained from the GSEA for cholesterol homeostasis (Fig. 6E). We then evaluated their impact on prognosis. To discriminate mainly for metastatic events, we decided to filter samples for optimal debulking surgery, which is defined as that one where the residual tumor was <1 cm, and analyzed progression-free survival (PFS). We obtained that this CXCL2-associated cholesterol homeostasis signature correlated with a significantly faster PFS, from 31 to 19 months (Fig. 6F). To evaluate the association of our TAM signature on tumor progression, we focused on the GSEA performed in isolated TAMs from our mouse models (Fig. 5A–F). We selected the 10 most enriched genes in the controls, as a macrophage (TAM) signature. In the same conditions of optimal debulking surgery, we obtained that those TAM genes correlated again with a much shorter PFS (from 29 to 19 months), indicating their possible positive correlation to metastasis (Fig. 6G). Moreover, when analyzing the effect on prognosis of both signatures (CXCL2 and TAM) combined, we obtained an even more drastic reduction of PFS, suggesting that these genes have additive effects (Fig. 6H). These data reinforce the role that CXCL2-mediated TAM education has on tumor prognosis.

One of the genes downregulated in MN-derived macrophages from Rbpj^iΔEC tumor-bearing mice is Cd74 (Fig. 5B). CD74 is important in TAMs during brain metastasis (31), and associated with worse prognosis in metastatic ovarian cancer (32). CXCL2 is a ligand for CXCR2 (29), and CD44 is part of a receptor complex that contains CD74 and CXCR2 (33, 34). Moreover, CD44 plays an important role in CD74-mediated signal transduction (35). In silico analysis showed that CD74 interacts closely with CD44 and CXCR2 (Fig. 7A) and it is increased in metastatic samples as compared with normal tissue (Supplementary Fig. S6A).

Furthermore, CD74 was associated with an immunosuppressive phenotype in macrophages (36). We therefore investigated whether the downregulation of CD74 could have a consequence on TAM function. For that, we analyzed publicly available datasets of IL4 responses in CD74-deficient macrophages (37). GSEA showed that the TAM signature was enriched in wild-type macrophages, indicating that CD74 is not only regulated by IL4, but also necessary for TAM education (Fig. 7B). We extracted a signature with the 500 most enriched genes in control compared with CD74-deficient macrophages stimulated with IL4 (CD74-mediated signature), representing a group of genes induced by IL4 through CD74 activation. When comparing newly recruited MN-derived macrophages from our EOC model by GSEA, we found that the CD74-mediated signature was significantly enriched in macrophages isolated from control mice compared with Rbpj^iΔEC. This suggests that newly recruited macrophages from Rbpj^iΔEC mice cannot induce their immunosuppressive phenotype due to downregulation of CD74-mediated genes (Fig. 7C).

To verify the specific effect on the newly recruited macrophages, we repeated the same analysis on resident macrophages (CD11b⁺/F4/80⁺/CCR2⁻), in which Cd74 is not differentially expressed between Rbpj^iΔEC tumor-bearing and control mice, and found no enrichment of this gene set in any group. This confirms that endothelial Rbpj deletion only affects Cd74 expression in newly recruited MN-derived macrophages (Supplementary Fig. S6B). In summary, only macrophages that have crossed the EC barrier as monocytes were affected by the lack of Rbpj in the endothelium.

It has been previously reported that TME in metastatic ovarian cancer is highly immunosuppressive, which was attributed to TAMs (3). For this reason, we tested whether impaired TAM education impacts on T-cell function. We isolated T cells from tumor-bearing Rbpj^iΔEC mice and their littermate controls 6 weeks after intraperitoneal injection of ID8 cells (Fig. 7D). Results demonstrated that T cells derived from Rbpj^iΔEC mice were more efficient in killing cultured ID8 cells (Fig. 7E). This confirms that impairment in TAM education has a direct impact on T-cell cytotoxicity. In addition, the analysis of T-cell populations in peritoneal cavity from tumor-bearing mice revealed that, although the total frequency of T cells was not changed, the cytotoxic CD8⁺ T-cell population was significantly increased in Rbpj^iΔEC mice compared with their littermate controls (Fig. 7F and G). In summary, our data reveal that Rbpj in ECs is necessary for tumor cell–mediated education of MN-derived macrophages into TAMs (Fig. 8).

Collectively, this study provides evidence about a novel angiocrine axis influencing the tumor immune microenvironment. We show how Notch/RBPJ-mediated transcription in ECs, which is frequently hyperactive in tumors (11), is required for CXCL2-mediated monocyte chemotaxis, induction of CD44 expression on monocytes, and the adoption of a TAM gene signature in metastatic ovarian cancer.

Mice with EC-restricted Rbpj loss had impaired ovarian carcinoma growth in the peritoneum and lower numbers of MN-derived macrophages in the peritoneal fluid. These macrophages are essential for metastatic ovarian cancer progression (4, 38). As such, the endothelium can influence tumor progression and metastasis by altering the immune status of the TME. Specifically, we observed that monocyte recruitment is potentiated, at least in part, through the release of RBPJ-mediated CXCL2 from ECs. This effect is independent of the tumor cells, because it can be reproduced in an independent tumor model or upon thioglycolate injection. Interestingly, higher serum levels of CXCL2 in patients with ovarian cancer are associated with myeloid cell infiltration, poor prognosis, and chemoresistance (39). It should be noted that this chemokine has been traditionally associated with the recruitment of neutrophils (29). However, there is evidence that CXCL2 also plays a role in the regulation of TAMs, especially those derived from monocytes. For instance, the CXCL2 receptor CXCR2 on monocytes and macrophages is important for the education of TAMs in prostate cancer (40).

CXCR2 blockade has also been shown to resensitize ovarian cancer to cisplatin treatment (41). Here we suggest that the CXCL2/CXCR2 axis might also have a role in the recruitment and education of macrophages in ovarian cancer. By separating newly recruited macrophages from macrophages that have resided in the peritoneal cavity for a longer period, we observed that endothelial RBPJ is necessary for the education into TAMs by tumor cells. Specifically, we report that MN-derived macrophages in contact with ECs lacking RBPJ had a lower expression of the HA receptor CD44. This receptor gets stimulated by HMW-HA produced by tumor cells to induce cholesterol depletion in macrophages, a crucial mechanism by which tumor cells educate TAMs (4). Therefore, we propose that through this mechanism ECs can precondition MN-derived macrophages and contribute to their immunosuppressive phenotype within the TME. Consistently, T cells isolated from this TME were more efficient at killing tumor cells.

In conclusion, we demonstrate that peritoneal ECs are critically involved in the recruitment and education of MN-derived macrophages in ovarian carcinoma.

No disclosures were reported.

E. Alsina-Sanchis: Data curation, software, formal analysis, validation, investigation, methodology, writing–review and editing. R. Mülfarth: Data curation, software, formal analysis, validation, investigation, methodology, writing–review and editing. I. Moll: Data curation, software, methodology. S. Böhn: Data curation, software, methodology. L. Wiedmann: Data curation, software, investigation, methodology. L. Jordana-Urriza: Data curation, software, investigation, methodology. T. Ziegelbauer: Data curation, software, investigation, methodology. E. Zimmer: Investigation, methodology. J. Taylor: Data curation, software, investigation, methodology. F. De Angelis Rigotti: Data curation, software, investigation, methodology. A. Stögbauer: Data curation, software, methodology. B.D. Giaimo: Resources, data curation, software, methodology. A. Cerwenka: Resources, formal analysis, funding acquisition, writing–review and editing. T. Borggrefe: Resources, formal analysis, writing–review and editing. A. Fischer: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, project administration, writing–review and editing. J. Rodriguez-Vita: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration.

The authors thank Ralf Adams (MPI Münster, Germany) for providing Cdh5-CreERT2 mice and kindly acknowledge Dr. Christiane Opitz for providing the SK-OV-3 cell line and Frances Balkwill (Barts Cancer Institute, London, United Kingdom) for providing ID8-luc cells. They thank the Light Microscopy core facility, the Microarray Unit, the Flow Cytometry core facility, and animal caretakers of the German Cancer Research Center (DKFZ) for providing excellent services. The authors would like to thank Damir Krunic (DKFZ, Light Microscopy Core Facility) in particular for his help with FIJI software data analysis. They thank the Tissue Bank of the National Center for Tumor Diseases (NCT) Heidelberg and the Institute of Pathology at Heidelberg University Hospital, both in Germany, for their support with the tissue microarray. This work was funded by the Deutsche Forschungsgemeinschaft (DFG) project 394046768 - SFB1366 projects C4, C2 (to A. Fischer and A. Cerwenka), SPP 1937 (CE 140/2-2 to A. Cerwenka), TRR179 (TP07 to A. Cerwenka), SFB-TRR156 (B10N to A. Cerwenka), TRR81-A12 (to T. Borggrefe); the Cooperation Program in Cancer Research of the German Research Cancer Center (DKFZ) and the Israeli Ministry of Science and Technology (MOST; to A. Fischer), the State of Hesse (LOEWE iCANx), von Behring-Röntgen Stiftung (65-0004 to T. Borggrefe), DFG project 419966437, Deutsche Krebshilfe project 70113888, MCIN/AEI/10.13039/501100011033 (PID2020-117946GB-I00 and RYC2019-027937-I), and RYC2019-027937-I to J. Rodriguez-Vita). The Science Ministry of Spain or the Health Ministry (ISCIII) receives support from the EU and its ERDF program. Part of the equipment used in this work has been funded by Generalitat Valenciana and co-financed with ERDF funds (OP ERDF of Comunitat Valenciana 2014–2020).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.