Disruption of Prostaglandin E2 Signaling in Cancer-Associated Fibroblasts Limits Mammary Carcinoma Growth but Promotes Metastasis

Reciprocal interactions between cancer cells and stromal cells in the tumor microenvironment shape the fate of cancer both, at primary and metastatic sites (1). Among stromal cells, cancer-associated fibroblasts (CAF) are known to affect tumor growth (2, 3). Although most studies suggest a tumor-promoting role of CAF, their influence on tumor progression may be tumor stage and tissue-dependent. Breast cancer CAFs change their phenotype across tumor developmental stages toward tumor-promoting, correlating with disease outcome (4). CAFs were reported to suppress growth of noninvasive melanoma cells, whereas they supported growth of metastasis-competent melanoma cells (5). Moreover, CAF improved tumor immune control and produced tumor-restraining extracellular matrix (ECM) in pancreatic ductal carcinoma (6, 7).

The origins of CAF are diverse, ranging from tissue-resident fibroblasts, mesenchymal stem cells, pericytes, and pre-adipocytes, to monocytic cells (8–10). Tissue-resident fibroblasts are ubiquitous cells in connective tissue where they shape the ECM (11). These usually quiescent cells can be activated when tissue homeostasis is disturbed. In response to wounding, fibroblasts differentiate into myofibroblasts, which are characterized by enhanced proliferation, shedding of rigid collagen, and the acquisition of smooth muscle cell properties that enable the contraction of wound edges (12). When unchecked, these repair properties may lead to fibrosis and cancer development (13, 14). Thus, a significant proportion of the CAF population shares features with myofibroblasts. The activation of resident fibroblasts to become CAF occurs via different signaling pathways, including continuous production of TGFβ by tumor cells (15), Notch signaling (16), mechanophysical, and physiological stress in the ECM, as well as a number of inflammatory mediators (11).

The expression of COX and their products, the prostanoids, is a major inflammatory factor in carcinogenesis (17). Tumor-promoting effects were mainly attributed to thromboxane A₂ (18), and prostaglandin E₂ (PGE₂). PGE₂ is produced by the concerted action of COX enzymes and prostaglandin E synthases. Main sources of PGE₂ in mouse breast tumors are COX-2 and the terminal synthase mPGES-1 (19, 20), which are induced upon inflammation (10). PGE₂ acting on tumor cells promoted tumor initiation and growth (17), and, when acting on tumor cells potently suppressed antitumor immunity. Accordingly, combining COX inhibitors with immune checkpoint blockade induced superior antitumor immunity (21). However, the impact of PGE₂ signaling on CAF activation is unclear (10). Therefore, we set out to investigate whether and how PGE₂ affects CAF activation and how this in turn shapes tumor progression.

PyMT mice express the polyomavirus middle T antigen driven by the mouse mammary tumor virus long terminal repeat (22). Female mice develop adenocarcinomas starting from 6 to 8 weeks of age and develop lung metastases from 18 weeks onward. PyMT mice were crossed with syngenic mPGES-1 KO mice (23). EP1, EP2, EP3, and EP4 KO mice were generated as previously described (24). All mice were backcrossed into a C57BL/6 background for at least 10 generations. Female mice at an age of 10 to 20 weeks were used for all experiments. In PyMT model, all 10 mammary glands were palpated thrice weekly from week 14. Tumor size was measured with. PyMT mice were sacrificed at 20 weeks of age or earlier if one tumor reached a size of 1.5 cm in diameter. The weight of every single tumor was determined for calculation of tumor mass and tumor burden (mouse body weight/tumor weight). At the endpoint, tumors and lungs were harvested for downstream analyses. For orthotopic tumors, C57BL/6 WT and mPGES-1 KO mice were injected with 5 × 10⁴ E0771 mammary carcinoma cells into the mammary fat pad of two mammary glands, with or without addition of 1.5 × 10⁵ WT or EP3 KO mammary gland fibroblasts (MGF). Tumor was monitored by caliber thrice weekly, and mice were sacrificed after four weeks.

Mouse care and experiments involving mice were approved by and followed the guidelines of the Hessian animal care and use committee (FU/1092, FU/1095, and FU/1200).

All cells used in this study were tested regularly for Mycoplasma contamination and maintained at low passage number (<15). E0771 murine breast adenocarcinoma cells originally isolated as a spontaneous tumor from C57BL/6 mouse were purchased from the ATCC and maintained in RPMI media supplemented with 10% FCS and 1% penicillin/streptomycin in a humidified 5% CO₂ atmosphere at 37°C. PyMT cells were isolated from a primary PyMT tumor as described before (18). For MGF isolation, mammary glands of C57BL/6 WT or EP3 KO mice were collected and minced under aseptic conditions. Minced tissues were processed to single cells with the tumor dissociation kit mouse (Miltenyi Biotec) for 20 minutes at 37°C, according to the manufacturer's instruction. Cell suspensions were passed through 70 μmol/L nylon meshs and resuspended in DMEM/F12 media supplemented with 10% FCS and 1% penicillin/streptomycin, and cultured in 6-well plates, pre-coated with 1% Matrigel, at a humidified 5% CO₂ atmosphere at 37°C. Cells were grown until confluency before passaging. Cells were used at passages 3–10, and purity of the isolated fibroblasts was confirmed by FACS analysis (CD45⁻, CD326⁻, CD31⁻, CD49f⁻, CD90.2⁺, CD140a/b⁺ cells; Supplementary Fig. S4B). Before treatment, MGF were cultured until confluency of 70%–80% was reached, and starved in DMEM/F12 medium containing 0.2% BSA for 24 hours. MGFs were activated with 20 ng/mL TGFβ (R&D Systems) for 72 hours (or 1 hour for Western analysis), with or without 1 μmol/L PGE₂ (Cayman Chemical), 1 μmol/L VX-702 (Tocris Biosciences), or 5 μmol/L Talmapimod (R&D Systems).

PyMT organoids were generated from PyMT cancer cells as previously described (25), with modifications. Briefly, cells were seeded in Cultrex Basement Membrane Extract Type 2 (R&D Systems) and growth factor–reduced (GFR) Matrigel (Corning), at a 1:1 mixture (at a density of 3 × 10⁵ cells per 300 μL gel mixture). Cultures were maintained in organoid media composed of DMEM/F12 media supplemented with 1% penicillin/streptomycin, 10 mmol/L HEPES, 1× Glutamax, 1× B27 (all from Gibco), and 20 μmol/L N-acetylcysteine, 5 μg/mL insulin, 100 ng/mL hydrocortisone (all from Sigma) were supplemented with 50 ng/mL murine recombinant EGF, 5 ng/mL murine recombinant FGF2 (both from Immunotools), and 10 ng/mL murine recombinant FGF10 (BioLegend).

For IHC analyses, tissues were fixed overnight in zinc fixative solution at 4°C, dehydrated, and embedded in paraffin. For multispectral immunofluorescence (PhenOptics), Opal 7-Color Fluorescent IHC Kits (Akoya Biosciences) were used as described previously (26). Briefly, antigen was retrieved from deparaffinized 10-μm sections through microwave treatment. Sections were blocked via incubation with blocking buffer for 30 minutes at room temperature. This was followed by primary/secondary antibody treatment. Opal signal generation was then performed by adding opal fluorophore working solution (10 minutes, room temperature), followed by mounting with DAPI. The following antibodies were used for staining PyMT tumors and mammary glands: anti-αSma (α-smooth muscle actin; Sigma), anti-Vim (Abcam), anti-Col-1 (Bio-Rad), anti-Fsp1 S100A4 (Dako), anti-Ki67 (Abcam), anti–Pan-Ck (Abcam). For metastasis evaluation, metastases were counted in 10 nonconsecutive lungs sections spaced throughout one lung lobe and were added up. Analysis of CAF subsets and tissue compartments in invasive breast cancer tissue microarrays (TMA) specimens provided by the Cooperative Human Tissue Network and the Cancer Diagnosis Program (other investigators may have received specimens from the same subjects) was also done via PhenOptics. Cancer cores were evaluated on the basis of tissue integrity and quality after staining, as well as on the presence of myoepithelia [to discriminate (DCIS ductal carcinoma in situ) from IDC]. On the basis of these criteria, 123 individual IDC cores were suitable for analysis. Human TMAs were stained with primary antibodies against TAK1 L (GeneTex), αSMA (Sigma), KI67 (Abcam), VWR (DAKO), and SERPINH1 (Novus Biologicals). Nuclei were counterstained with DAPI. The BOND RX Automated IHC Research Stainer (Leica Biosystems) was used for staining of human TMAs. All PhenOptics samples were acquired at ×20 magnifications using the Vectra3 automated quantitative pathology imaging system (Akoya Biosciences). The InForm v2.5 software (Akoya Biosciences) was used to identify and quantify tissue compartments and cell subsets.

For IHC phospho-staining, tissue sections were deparaffinized and processed using a Bond Max (Leica). The Bond Polymer Refine Detection kit from Leica was used for antibody staining and detection according to the manufacturer's instructions. Antigen was retrieved using citrate buffer, and the following antibodies were used: anti-pp38 (Cell Signaling Technology), anti-pAkt (Cell Signaling Technology), anti–β-catenin (MerckMillipore). Stained sections were scanned with a ScanScopeCS2 (Leica) using a ×20 objective. Aperio ImageScope software (Leica) was used for staining quantification.

For immunofluorescence analysis of PyMT organoids, Organoids were washed and fixed overnight in 4% PFA at 4°C, then permeabilized/blocking using 0.3% Triton X-100 (Applichem) with 5% donkey serum (Sigma-Aldrich) in PBS for 12 hours at 4°C. Organoids were stained in PBS with 0.5% donkey serum and 0.1% Triton X-100, with the following antibodies: Anti–E-cad-AF647 (BioLegend), anti–vimentin-APC (R&D Systems), Ki67 (Abcam), phalloidin-AF488 (Thermo Fisher Scientific), and Hoechst 33342 (Sigma-Aldrich). Secondary AF546 coupled antibody to visualize Ki-67 was used (Thermo Fisher Scientific). Stainings were imaged with a Plan-Apochromat 20x/0.8 M27 objective using a LSM780 (Zeiss) and further analyzed with Imaris 9.1 (Bitplane). Mean distance between the surface of the entire rendered spheroid and the volume rendered vimentin-positive cells within the spheroid was calculated using the Imaris distance calculation plug-in.

Tumors were minced and digested using tumor dissociation kit mouse (Milteyni Biotec) according to the manufacturer's instructions. Single-cell suspensions were blocked with FcR blocking reagent (Miltenyi Biotec) in PBS supplemented with 0.5% BSA. The following fluorochrome-conjugated antibodies were used to identify fibroblasts. Anti-CD140a PE (BD Biosciences), anti-CD140b APC (BD Biosciences), anti-CD90.2 PE (Miltenyi Biotec), anti–CD31 PE-Cy7 (eBioscience), anti-CD326 BV711 (BD Biosciences), anti-CD45 VioBlue (Miltenyi Biotec), anti–CD49f PE-CF594 (BD Biosciences). For intracellular staining with anti-αSMA FITC (Sigma), cells were fixed and permeabilized using BD Biosciences' Cytofix/Cytoperm Fixation/Permeablization Kit. For identifying tumor-infiltrating lymphocytes, the following antibodies were used. Anti–CD3-PE-CF594, anti–CD4-BV711, anti–CD8-BV650, anti–NK1.1-BV510 (BD Biosciences), anti–CD45-VioBlue (30-F11.1), anti–MHC-II-APC (Miltenyi Biotec), anti–CD11b-BV605, and anti–GITR-FITC (BioLegend). Samples were acquired with a LSRII/Fortessa flow cytometer, or FACS-sorted using a FACS Aria III cell sorter (both from BD Biosciences). Data were analyzed using FlowJo software VX (Treestar). Antibodies were titrated to determine optimal concentrations. Dead cells were excluded from the analysis on the basis of FSC and SSC gating. For single-color compensation, CompBeads (BD Biosciences) were used to create multicolor compensation matrices. Flow Cytometry Absolute Count Standard (BangsLabs) beads were used as an internal counting standard.

mRNA isolation and cDNA transcription were performed immediately after CAF-sorting using SMART-Seq v4 Ultra Low Input RNA Kits for Sequencing (Takara) according to the manufacturer's instructions. cDNA was amplified by 16 PCR cycles and samples were purified after cDNA and library synthesis with AMPure XP beads for PCR Purification (Beckman Coulter). For the determination of cDNA content, a Qubit 3.0 Fluorometer in combination with the Qubit 1x dsDNA HS Assay Kit (both Thermo Fisher Scientific) was used. Fragment sizes were determined using High Sensitivity DNA Kit with 2100 Bioanalyzer (both Agilent). The Covaris M220 system was used for controlled cDNA shearing (2 minutes at 20°C, Peak Power 75W, 20% Duty Factor, Burst Cycle set at 200). Library preparation was performed with SMARTer ThruPLEX DNA-Seq Kit (Takara). Ten samples were subjected to an indexed single-read sequencing run with 1 × 75 cycles on an Illumina NextSeq 500 system.

Initial sequence quality was monitored with FastQC (V 0.11.5). Potential 3′ end degradation biases were visualized using PicardTools CollectRnaSeqMetrics (V 2.17.2). Using Flexbar (V 3.0.3; ref. 27), adapter sequences were removed from the 3′ ends, and resulting reads were subjected to a window-based quality trimming (Phred score <20, 5-nt window). Processed reads were mapped to the mouse genome (assembly GRCm38/mm10) based on GENCODE gene annotations (release m16) using STAR (28). Reads were allowed to map with up to 5 mismatches, while considering no multi-mapping reads. Differential expression analysis was performed using the R/Bioconductor package DESeq2 (V 1.22.2; ref. 29). Read overlaps were counted within annotated exons using GenomicAlignments (V 1.18.1) in “union” mode (30). Differential testing compared the effect between mPGES-1 KO and WT. Resulting log₂-transformed fold changes were adjusted for gene expression levels using DESeq2's “shrinkage estimator” functions. The P values were adjusted for multiple testing using Benjamini–Hochberg correction (significance threshold 0.01). This analysis revealed a total of 2,200 significantly regulated genes (differentially expressed genes) upon mPGES-1 KO. For visualization purposes (Fig. 4C), k-means clustering (k = 2) was used to separate genes in two groups according to the direction of expression change.

Gene expression values (in transcripts per million, TPM) and clinical data were retrieved from The Cancer Genome Atlas (TCGA) breast cancer dataset (RTCGA version 1.18.0, BRCA dataset). The combined effect of the expression levels of genes encoding PGE₂-metabolizing enzymes COX1/2 (PTGS1/2), mPGES1/2/3 (PTGES/2/3), and HPGD (HPGD) on patient survival was analyzed using Cox regression (multivariate proportional hazard model; R survival package version 3.2). Genome-wide screening for significant associations of gene expression with patient survival was addressed by univariate Cox regression. Resulting P values were adjusted for multiple testing via Benjamini–Hochberg correction and a threshold of 0.05 was applied to call significance.

PGE₂ production was modeled as multiple linear regression and compared with the expression of individual CAF marker genes. First, a model was built for the fibroblast marker gene PDGFRB (platelet-derived growth factor B) using the expression levels either of PTGS1, PTGS2, PTGES, PTGES2, PTGES3, and HPGD (“core” model) or of all genes annotated with the Reactome pathway—Synthesis of Prostaglandins (PG) and Thromboxanes (TX; “extended” model; Supplementary Fig. S2A). The estimated effect that each gene expression change on the PDGFRB expression level was evaluated by comparing the coefficient estimates for all genes in the core and extended model (Supplementary Fig. S2D). The coefficients of the resulting models were correlated against the expression levels of all of its components, to reveal the direction of association captured by the models. Because a positive correlation with HPGD expression indicated that the models captured PGE₂ degradation rather than production, model coefficients were multiplied by −1. Coefficients of the PGE₂ production models were then correlated to additional fibroblast marker genes that revealed a consistent negative correlation (Supplementary Fig. S2D).

All genes expressed in CAF of WT versus mPGES-1 KO CAF were used as input to analyze gene sets in the Molecular Signatures Database (v6.0) using gene set enrichment analysis (GSEA) after applying a cutoff value (raw counts <3) to filter out genes related to contaminating cell populations such as lymphocytes. To quantify CAF phenotypes in PyMT tumors, we used two sets of published CAF signatures: (i) inflammatory CAF (iCAF) and myofibroblastic CAF (mCAF; compared against quiescent CAF) from (31), and (ii) refined signatures for iCAF, myCAF and antigen-presenting CAF (apCAF) from (32) as previously described (4, 31). The P values of all GSEA analyses were jointly adjusted for multiple testing using Benjamini–Hochberg correction.

The expression levels of all expressed genes (average TPM > 1) in the TCGA dataset were compared with the expression levels of MAP3K7CL using pairwise Pearson correlation. A total of 927 genes were found significantly correlated (Benjamini–Hochberg-corrected P value < 0.05) while also showing an absolute correlation coefficient >0.3. These genes were selected for an enrichment analysis using REACTOME and GO Biological Process, with all expressed genes considered as background universe. The 15 most significant terms were selected for plotting are shown in Supplementary Fig. S6.

LC-MS/MS analysis of prostanoids in tissue samples was performed essentially as described previously (33). Tumor tissue samples were homogenized in 200 μL PBS using a swing mill (Retsch). Samples were subjected to liquid–liquid extraction with ethyl acetate after adding 20 μL internal standard solution PGE₂-d4 purchased from Cayman Chemicals) and 100 μL 0.15 mol/L EDTA. The extraction step was repeated, organic layers were collected, evaporated at 45°C under a gentle stream of nitrogen, and reconstituted with 50 μL acetonitrile/water/formic acid (20:80:0.0025, v/v). 10 μL were injected into the LC-MS/MS system, an Agilent 1290 Infinity LC system (Agilent) coupled to a hybrid triple quadrupole linear ion trap mass spectrometer 6500QTrap⁺ (Sciex) equipped with a Turbo Ion Spray source operating in negative electrospray ionization mode. Chromatographic separation was done using a Synergi Hydro-RP column (2.0 × 150 mm, 4-μm particle size; Phenomenex), coupled to a precolumn of the same material; 0.0025% formic acid and acetonitrile with 0.0025% formic acid serving as mobile phases. The following Mass spectrometric parameters were used: Ionspray voltage −4500 V, source temperature 500°C, curtain gas 35 psi, nebulizer gas 40 psi, and Turboheater gas 60 psi. Both quadrupoles were running at unit resolution. For analysis, Analyst Software 1.6.3 and Multiquant Software 3.0.2 (both Sciex) were used, employing the internal standard method (isotope-dilution mass spectrometry). The precursor-to-product ion transitions used for quantification were m/z 351.2 → m/z 315.0 for PGE₂, m/z 351.2 → m/z 233.3. The calibration curves were constructed using linear regression with 1/x2 weighting.

Gel contraction assay was performed as described previously (34). PureCol (CellSystems) in BSA pre-coated 24-well plates were used. WT MGF were seeded on top of the polymerized PureCol, and starved in DMEM/F12 medium containing 0.2% BSA for 24 hours. Gel matrices were detached and treated with 20 ng/mL TGFβ and/or 1 μmol/L PGE₂ for 72 hours. Gels were scanned and contraction areas were quantified using ImageJ software.

MGF were seeded in 96-well plate pre-coated with 1% Matrigel in triplicates at a seeding density of 5,000 cells/well, and cultured under standard culture conditions in an incubator containing the IncuCyte R S3 live-cell analysis system (Sartorius). Images were taken every 24 hours and confluency was used to monitor cell proliferation.

On-Targetplus SMARTpool (Dharmacon) siRNA-targeting mouse Map3k7cl, which contains four different Map3k7cl targeting siRNAs, was used to transiently knock down Tak1L, in addition to a non-targeting control siRNA (siControl). MGF were transfected using HiPerfect transfection solution (Qiagen) at a final concentration of 50 nmol/L for 6 hours. Transfection media were exchanged with DMEM/F12 supplemented with 10% FCS for overnight culture. Afterwards, cells were used for further analyses.

Supernatants from WT and EP3 KO MGF were harvested when cultured cells were 70%–80% confluent. Supernatants of five individual cell preparations from separate animals were pooled and loaded onto Proteome Profiler Mouse Chemokine Array Kits (R&D Systems) according to the manufacturer's instructions (100 μg of protein per membrane). Pixel analysis was used for quantification with ImageStudio software V5.

RNA was extracted using the RNA isolation kit according to the manufacturer's instructions (RNeasy Mini Kit, Qiagen), and RNA concentrations were quantified using the NanoDrop spectrophotometer (Thermo Fisher Scientific). RNA was transcribed into cDNA using Fermentas Reverse Transcriptase Kit (Thermo Fisher Scientific) according to the manufacturer instructions. qRT-PCR was performed using SYBR green and the Quant5 Real-time-PCR system (Thermo Fisher Scientific). Quantitect primer assays (Qiagen) were used to detect murine Map3k7cl, Mapk14, EP3, Sox9, and Dusp6. Other murine primer sequences (obtained from biomers.net) were: Rps27a F: 5′-GAC CCT TAC GGG GAA AAC CAT-3′, R: 5′-AGA CAA AGT CCG GCC ATC TTC-3′; Acta2 F: 5′-CCC AGA CAT CAG GGA GTA ATG G-3′, R: 5′-TCT ATC GGA TAC TTC AGC GTC A-3′; Pcna F: 5′-GGA GGA GCC GCA GTC AGA T-3′, R: 5′-GCA GCG CCT CAC AAC CTC CGT C-3′; EP1 F: 5′-ACT AGA GAA TGG GCA AGC CG-3′, R: 5′-CCA TGG CAA TCT CTC ACC AGT-3′; EP2 F: 5′-CCT CCG AGC TCT TCG GTT TT-3′, R: 5′-AGG ACC GGT GGC CTA AGT AT-3′; EP4: F: 5′-TGG GAA GAG ACT GAT GGC TG-3′, R: 5′-CCT GTA GGG TGG GGT TAG GAG-3′. The data were quantified using QuantStudio (Thermo Fisher Scientific) normalized to Rps27a endogenous control to compensate for experimental variations. Fold changes were calculated using the comparative C_t (cycle threshold) method.

Cells were lysed using RIPA buffer, and 50–100 μg of protein was run on 10% polyacrylamide gels, followed by transfer to nitrocellulose membranes. Membranes were blocked with 5% BSA, and then incubated with primary antibodies overnight at 4°C. Blots were visualized by IRDye 800- and IRDye 680-coupled secondary antibodies (LI-COR Biosciences) using the LI-COR Odyssey imaging system (LI-COR Biosciences). Protein bands were quantified using Image Studio software version 5. The following antibodies were used: Anti-p38 (Cell Signaling Technology), anti-phospho38 (Cell Signaling Technology), anti-Tak1L (LifeSpanBiosciences), and anti-Tubulin (Sigma).

Unless stated otherwise, data are presented as means ± SEM for at least 2–3 independent experiments with multiple biological replicates. Statistical comparisons between two groups were performed using Mann–Whitney test, paired or unpaired two-tailed Student t test. One- or two-way ANOVA, or multiple t tests, followed by appropriate post hoc tests were used for multiple comparisons. Data were pre-analyzed to determine normal distribution and equal variance with D'Agostino-Pearson omnibus normality test. Normalized data were analyzed using the one-sample t test. Statistical survival analysis was performed using the log-rank test. Correlation was analyzed using the Spearman r test. Statistical analysis was performed with GraphPad Prism v8. Differences were considered significant if *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001. The sample sizes used were based on level of changes and consistency expected. No statistical test was used to predetermine sample size, and all samples were included in the analysis. Details on statistical tests used in individual experiments can be found in the figure legends.

RNA-sequencing data that support the findings of this study have been deposited in GEO with the accession code GSE169297. The token for anonymous reviewer access is irkrsmkqvnsbbsr. Main R scripts for the analysis and necessary data are available at GitHub [https://github.com/ZarnackGroup/Elwakeel-et-al-2021].

We previously observed delayed tumor growth in mPGES-1 KO mice in the polyoma middle T oncogene (PyMT) mammary carcinoma background (19), but the underlying reason remained unclear. Histomorphological assessment and PhenOptics multispectral imaging revealed that stromal content inversely correlated with tumor burden. A low tumor burden in mPGES-1 KO PyMT mice lacking tumor PGE₂ (Supplementary Fig. S1A) was associated with increased stroma (Fig. 1A–D). The latter was due to an increase in CAF, indicated by elevated levels of the CAF markers αSma, vimentin (Vim), and collagen-1 (Col-1; Fig. 1E–H; Supplementary Fig. S1B). The increase in CAF observed in mPGES-1 KO PyMT tumors was associated with increased Ki67 levels in stromal cells, indicating CAF proliferation (Fig. 1E and I). At the same time, Ki67-positive tumor cells were markedly decreased in mPGES-1 KO tumors (Fig. 1E and J). To approach whether a negative correlation between PGE₂ and CAF was also true in human breast cancer, we used transcriptomic data of 1,065 patients with breast cancer from TCGA database (35). Among the enzymes mainly linked to PGE₂ production (Supplementary Fig. S2A), PTGS2 (encoding COX-2) and PTGES (mPGES-1) were expressed at low levels. PTGS1 (COX-1) expression was higher than that of PTGS2, and PTGES3 (cPGES) was particularly highly expressed (Fig. 2A). Thus, dominant enzymatic sources of PGE₂ in human and murine mammary tumors may differ. Accordingly, high PTGS1 and PTGES3 levels were negatively associated with patient survival, whereas expression of the PGE₂-metabolizing enzyme 15-hydroxyprostaglandin dehydrogenase (HPGD) was positively associated with survival (Fig. 2B). Of note, PTGES3 showed the second highest hazard ratio of all genes in the TCGA dataset (Supplementary Fig. S2B). We next generated a core model based on the expression of PGE₂-generating enzymes and an extended model on prostanoid metabolizing enzymes in relationship to the fibroblast marker PDGFRB to test the relationship between PGE₂ production and fibroblast content/activation (Supplementary Fig. S2C and S2D). Using both models, high virtual PGE₂ levels correlated negatively with the CAF markers ACTA2 (αSMA), PDGFRB, collagen type 1A1 (COL1A1), fibroblast-specific protein (FSP1/S100A4), and fibroblast-activating protein. There was no apparent correlation with the general fibroblast marker SERPINH1 (Fig. 2C). Given the negative correlation between PGE₂ and fibroblast activation markers in both mouse and human breast tumors, a direct effect of PGE₂ on MGF activation was studied. MGF were activated with TGFβ for 72 hours, upregulating αSma protein expression (Supplementary Fig. S3A–S3C). TGFβ–induced αSma expression was prevented by PGE₂ (Fig. 2D). Accordingly, collagen contraction induced by TGFβ in MGF was significantly reduced by cotreatment with PGE₂ (Fig. 2E). In conclusion, these data indicate that PGE₂ inhibits fibroblast activation.

PGE₂ signals via four G-protein–coupled receptors (EP1–4). In FACS-sorted MGF, the expression of EP3 was most pronounced (Fig. 3A). Interestingly, mPGES-1 and EP3 KO mammary glands contained higher MGF levels than WT, EP1, EP2, and EP4 KO mammary glands (Fig. 3B). The higher numbers of MGF in EP3 KO mammary glands were likely due to enhanced proliferation, because EP3 KO MGF expressed significantly higher levels of proliferating cell nuclear antigen (Pcna) compared with WT MGF (Fig. 3C) and showed massive proliferation in a growth factor–deprived setting in vitro (Fig. 3D). An increased abundance and activation status of EP3 KO MGF was also suggested by high stromal Col-1 expression relative to EP1, EP2, or EP4 KO mammary glands (Fig. 3E and F). These findings emphasize the importance of EP3 for PGE₂ signaling in MGF of nontransformed mammary glands. mPGES-1 KO MGFs were not viable in culture.

To investigate whether EP3 signaling in MGF was also important in a mammary tumor context, we orthotopically co-grafted E0771 mammary carcinoma cells with syngenic EP3 KO or WT MGF into the mammary fat pad of WT and mPGES-1 KO mice, respectively (Fig. 3G). Tumor growth was monitored for four weeks and compared with tumor growth in WT and mPGES-1 KO mice that only received E0771 mammary carcinoma cells. Tumor volume in co-grafted (E0771 + EP3 KO MGF) WT mice was low at the endpoint, recapitulating the low tumor volume in mPGES-1 KO mice that received E0771 cells alone. In contrast, mPGES-1 KO mice receiving E0771 cells + WT MGF showed higher tumor burden, similar to WT mice receiving E0771 cells alone (Fig. 3H). Moreover, co-transplanted tumors in WT mice (E0771 + EP3 KO MGF) and tumors in mPGES-1 KO mice contained elevated numbers of CAF compared with co-transplanted tumors in mPGES-1 KO mice (E0771 + WT MGF) and WT tumors (Fig. 3I). Thus, stromal mPGES-1 deficiency, that is, low PGE₂ levels in tumors, was phenocopied by injecting EP3 KO MGF into WT mice, and this effect was reversed when WT MGF were injected into mammary glands of mPGES-1 KO mice. EP3 KO in MGF was sufficient to induce CAF expansion and to restrict tumor growth in mice, similar to the phenotype in mPGES-1–deficient PyMT mice. These data suggest that PGE₂ signals through EP3 to modulate MGF activation, thereby promoting tumor growth.

To dissect how PGE₂ depletion influences fibroblast phenotypes in mammary tumors, we used bulk mRNA sequencing of FACS-sorted CAF from late-stage WT and mPGES-1 KO PyMT tumors (Fig. 4A). The isolated fraction comprised approximately 0.05%–0.5% of the total acquired cells with significant increase in the mPGES-1 KO group (Fig. 4B), supporting PhenOptics data (Fig. 1). RNA sequencing identified 2,200 differentially expressed genes between CAF isolated from WT and PGE₂-low tumors (P_adjusted value < 0.05, |log₂ fold change| > 1; Fig. 4C; Supplementary Table S1). This gene set was used to classify CAF phenotypes in PyMT tumors based on published CAF signatures (31, 32). GSEA using a broad gene signature (31) suggested that CAF from mPGES-1 KO tumors resembled inflammatory CAF (iCAF; Fig. 4D; Supplementary Fig. S4A). Similar gene expression patterns as in CAF from mPGES1 KO tumors were observed upon EP3 ablation when WT and EP3 KO MGF were cultured with PyMT tumor organoids to induce an iCAF phenotype (Fig. 4E; ref. 31). Here, EP3 KO MGF, showed increased expression of chemokine ligand 5 (Ccl5), the pro-fibrotic transcription factor Sox9 (36), and dual specificity phosphatase 6 (Dusp6), whereas Map3k7 C-terminal like (Map3k7cl) was expressed at low levels compared with WT MGF (Fig. 4F–I). An iCAF-like secretory phenotype of EP3 KO CAF was validated using cytokine microarrays (Supplementary Fig. S4B; Supplementary Table S2). However, when using a more concise set of gene signatures in GSEA (32), the iCAF phenotype was not substantiated. Rather, a shift toward antigen-presenting CAF (Fig. 4J; Supplementary Fig. S4C and S4D) became apparent. Accordingly, mPGES-1 KO CAF expressed higher levels of the costimulatory molecule Cd83 (Supplementary Fig. S4E) and mPGES-1 KO PyMT showed higher CD4⁺ T-cell infiltrates (Fig. 4K; Supplementary Fig. S5). Moreover, CD4⁺ T cells numbers were increased in E0771 tumors of mono-transplanted mPGES-1 KO and co-transplanted WT (E0771 + EP3 KO MGF) mice (Fig. 4L), indicating a role of EP3 signaling in the antigen-presenting features of CAFs. Thus, PGE₂ signaling via EP3 limited the establishment of a secretory and potentially immunostimulatory CAF phenotype.

Further analysis of the PyMT CAF gene signature by GSEA revealed a number of altered pathways upon mPGES-1 depletion (Supplementary Table S3). Among cellular signaling pathways, AKT, WNT, and MAPK signaling, particularly p38 MAPK activation via TAK1, were identified (Fig. 5A). Analyzing expression of key players in these signaling pathways in WT and mPGES-1 KO tumor sections by IHC revealed an increased expression of phosphorylated p38 (pp38) MAPK in the stroma of mPGES-1 KO tumor sections, also in CAF (Fig. 5B and C; Supplementary Fig. S6A). AKT and WNT signaling were not altered at protein level (Supplementary Fig. S6B and S6C). To validate the modulation of p38 MAPK signaling in MGF by PGE₂, WT and EP3 KO MGF were treated with TGFβ as the prototypical CAF activation factor that also induces p38 MAPK signaling (37). Already under basal conditions, EP3 KO MGF expressed higher levels of pp38 compared with WT MGF (Fig. 5D), and TGFβ treatment induced pp38 only in WT MGF (Fig. 5E). At a functional level, regulation of p38 MAPK signaling by PGE₂ affected MGF activation and proliferation. EP3 KO MGF demonstrated high Acta2 expression compared with WT MGF at baseline (Fig. 5F) and TGFβ induced Acta2 expression strongly in WT but only weakly in EP3 KO MGF. The p38 inhibitors VX-702 and Talmapimod prevented the TGFβ–dependent induction of Acta2 mRNA expression in WT MGF, and reduced Acta2 mRNA expression even in unstimulated EP3 KO MGF (Fig. 5G). Moreover, both compounds inhibited EP3 KO MGF expansion in cell culture, without affecting the TGFβ–induced growth of WT MGF (Fig. 5H and I). These data suggest that PGE₂ restricts p38 MAPK signaling, to limit MGF activation and expansion.

We next explored mechanisms of EP3-dependent restriction of p38 MAPK signaling in MGF. Tak1-mediated p38 signaling was enriched in mPGES-1 KO CAF (Fig. 5A), and Map3k7cl was affected by the deletion of both mPGES-1 and EP3 (Fig. 4C and I). Map3k7cl encodes TGFβ–activated kinase like (Tak1L) protein, a Tak1 paralogue that lacks the N-terminal kinase domain, while containing a homologous C-terminal domain required for Tab2/3 binding (38). We hypothesized that Tak1L might represent a kinase-dead variant of Tak1, to negatively regulate Tak1 signaling. To narrow down its potential role in our system, we first assessed Tak1L coexpression patterns, starting with human breast cancer data. A comparison of MAP3K7CL with all other genes expressed in the TCGA dataset revealed 925 significantly correlating genes. Gene Ontology (GO) and REACTOME pathway analysis on these 925 genes revealed enrichment in ECM structure, collagen synthesis, and vascular function, implying that MAP3K7CL may be primarily expressed in fibroblasts and vascular cells (Supplementary Fig. S7). Accordingly, Map3k7cl was highly expressed in primary MGF, but not other cell subsets abundant in PyMT tumors, such as cancer cells and macrophages (Fig. 6A). Moreover, Map3k7cl expression was low in NIH3T3 embryonic fibroblasts, and was reduced in EP3 KO MGF (Fig. 6A). Exploring a correlation of Map3k7cl with expression and activation of p38 MAPK, we unexpectedly found an inverse expression relationship between Mapk14 (encoding p38 MAPK) and Map3k7cl (Fig. 6B). More importantly, transient Map3k7cl knockdown (KD) in WT MGF increased the p38 MAPK phosphorylation in the absence of other stimuli (Fig. 6C–E). Given the role of p38 MAPK in MGF activation downstream of EP3, the impact of Map3k7cl in MGF was tested. TGFβ induced Acta2 expression in WT MGF, which was blocked by PGE₂ treatment. However, Map3k7cl or EP3 deletion significantly induced Acta2 mRNA levels at baseline, which were not further induced by TGFβ and not inhibited by PGE₂ (Fig. 6F). Map3k7cl or EP3 deletion also induced mRNA expression of Mapk14 (Fig. 6G). Thus, Map3k7cl expression is under the control of EP3 in MGF, and both Map3k7cl and EP3 depletion trigger MGF activation and unlink it from control by PGE₂.

To determine whether Map3k7cl in MGF affects tumor growth similar to the effects seen after ablation of mPGES-1 or EP3, PyMT tumor cell organoids were cultured in the presence of conditioned media from WT, EP3 KO, and Map3k7cl KD MGF. Organoid culture medium was used as control, whereas TGFβ was used as a growth inhibitor, also inducing epithelial-to-mesenchymal transition (EMT; Fig. 7A; Supplementary Fig. S8A). Fluorescence microscopy analyzing E-cadherin (Ecad) as an epithelial, Vim as a mesenchymal, and Ki67 as a proliferation marker was used to track morphological changes in organoids (Fig. 7B). Organoid size and Ki67 expression were significantly increased upon culture with conditioned media from WT MGF compared with the control group. This was not observed when conditioned media from MGF lacking either Map3k7cl or EP3 were used (Fig. 7C and D). Unexpectedly, the conditioned media from MGF that lacked either Map3k7cl KD or EP3 induced EMT as determined by the Vim/Ecad ratio (Fig. 7E) and tumor matrix invasion (Fig. 7F). The induction of EMT is associated with metastasis (39). Accordingly, conditioned media from MGF lacking lacked either Map3k7cl KD or EP3 induced the expression of the pro-metastatic transcription factors Zeb2 (40) and Lef1 (41) in PyMT organoids (Fig. 7G and H). Interestingly, significantly higher PyMT mRNA expression in lungs of 20-weeks-old mPGES-1 KO PyMT versus WT PyMT lungs was observed, being a sign of metastasis (Fig. 7I). PhenOptics analyses confirmed an increase in number of metastasis foci in mPGES-1 PyMT lungs compared with WT PyMT lungs (Fig. 7J and K). These data suggest that uncoupling CAF from PGE₂ limits tumor growth, but may promote metastasis.

To evaluate the clinical relevance of our findings concerning metastasis in human cancer, we used human CDC breast cancer progression TMAs containing cores of patients with invasive breast cancer, of which a large proportion (∼30%) had distant metastasis. TMA were stained with antibodies against TAK1L, pp38, SERPINH1 to enrich fibroblasts, αSMA for activated CAF, VWF for endothelial cells, and KI67 for proliferating cells using PhenOptics (Fig. 8A; Supplementary Fig. S8B). Cancer cores were histomorphologically assessed to remove cores with remaining myoepithelia, and tissue segmentation included vessels. In this manner, we removed SERPINH1⁺ αSMA⁺ myoepithelia and SERPINH1⁺ TAL1L⁺/αSMA⁺ smooth muscle cells from our analysis. TMA also contained separate cores of normal mammary gland and DCIS samples. TAK1 L expression by SERPINH1⁺ cells was enhanced in invasive carcinoma compared with DCIS or normal mammary tissue (Fig. 8B). More importantly, TAK1 L expression in SERPINH1⁺ cells was negatively correlated with αSMA expression, lymph node status, metastasis status, and survival, whereas it positively correlated with KI67-positive cancer cells, and tumor content (Fig. 8C). Accordingly, high TAK1 L expression in SERPINH1⁺ cells was associated with favorable patient survival, whereas αSMA expression in SERPINH1⁺ cells correlated with poor survival (Fig. 8D and E). These data indicate that TAK1 L in SERPINH1⁺ cells, which, by histological assessment, are mainly fibroblasts promotes tumor cell proliferation, but restricts metastasis, thereby promoting survival in a cohort with significant distant metastasis.

PGE₂ is a tumor-promoting lipid mediator in multiple cancer models (19, 42). A tumor-promoting role was also suggested for CAF. Therefore, the negative regulation of CAF function by PGE₂ observed in this study is conceptually challenging. Interrupting PGE₂ signaling through EP3 in CAF restricted primary tumor growth but induced metastatic features in tumor cells. Thus, PGE₂ may be involved in suppressing metastasis by limiting CAF activation. Our study so far does not allow attributing increased metastasis in mPGES-1 KO PyMT mice to CAF alone, although PGE₂ acting on cancer, vascular or immune cells, often through EP4, has so far mainly been reported to promote metastasis (43). Indeed, even CAF-derived PGE₂ promoted metastatic features in cancer cells (44). However, in contrast, blocking EP1 in murine breast cancer cells augmented pulmonary metastasis (45). Thus, the relative contribution of EP3 in CAF to modulating metastasis requires further exploration. Importantly, the ambivalent role of PGE₂ signaling was also supported in human patients with breast cancer. A composite PGE₂ signature correlated with poor survival in the TCGA cohort containing few patients with metastasis, whereas MAP3K7CL, which is under the control of PGE₂ in CAF, correlated with improved survival in a cohort of patients with significant distant metastasis. This may help to explain the conflicting results and setbacks concerning PGE₂ modulation in cancer therapy. Clinical trials showed that PGE₂ inhibition reduced colorectal cancer and melanoma risk (46, 47), but no association was found between aspirin use and breast cancer risk (48). Moreover, the COX inhibitor celecoxib combined with chemotherapy adversely affected survival of patients with breast cancer (49).

PGE₂ triggered anti-fibrotic features in lung fibroblasts by signaling through EP2 and/or EP4 (50, 51), and in synovial and dermal fibroblasts through EP2 (52, 53). Because fibroblasts are heterogeneous from one anatomical site to another, the EP receptor mediating PGE₂ signaling in these cells may differ between tissues. EP3 appeared as the dominant EP receptor in MGF. Expansion of mPGES-1 KO and EP3 KO MGF in nontransformed mammary glands may therefore suggest a homeostatic role of PGE₂ in the breast, preventing MGF expansion and, potentially, a predisposition to fibrotic disease. Interestingly, PGE₂-reducing nonsteroidal anti-inflammatory drugs are commonly used to treat pain in fibrocystic breast disease. Their impact on fibrosis has not been evaluated, which might be of interest in light of our data.

Tak1-mediated p38 MAPK activation emerged from our study as a PGE₂/EP3-dependent signaling pathway that promoted CAF activation and proliferation. This fits to recent observations connecting p38 MAPK signaling in fibroblasts to cardiac fibrosis and right ventricular hypertrophy (54, 55). Tak1 activation promoted synthesis and release of pro-inflammatory molecules via MAPK and/or NF-κB pathways (56), corresponding to the secretory phenotype of EP3 KO MGF. We show that p38 MAPK activation is regulated by Tak1L downstream of EP3. Tak1L thus emerges as a factor that restricts p38 MAPK activation to limit fibroblast activation. Its role as a kinase-dead dominant negative variant of Tak1 needs to be tested in future mechanistic studies.

In conclusion, this study challenges the concept of the strictly tumor-promoting nature of PGE₂, suggesting PGE₂ as a double-edged mediator that can promote tumor growth at the primary site yet restrict metastasis. PGE₂ targeting strategies in breast cancer may, therefore, benefit from EP-selective approaches.

D. Thomas reports grants from German Research Foundation during the conduct of the study. I. Fleming reports grants from Deutsche Forschungsgemeinschaft during the conduct of the study. F.R. Greten reports grants from HMWK during the conduct of the study. No disclosures were reported by the other authors.

E. Elwakeel: Conceptualization, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Brüggemann: Conceptualization, data curation, formal analysis, methodology, writing–review and editing. J. Wagih: Investigation, writing–review and editing. O. Lityagina: Investigation, writing–review and editing. M.A.F. Elewa: Investigation, writing–review and editing. Y. Han: Investigation, writing–review and editing. T. Frömel: Formal analysis, investigation, visualization, writing–review and editing. R. Popp: Formal analysis, investigation, visualization, methodology, writing–review and editing. A.M. Nicolas: Investigation, methodology, writing–review and editing. Y. Schreiber: Investigation, writing–review and editing. E. Gradhand: Formal analysis, validation, writing–review and editing. D. Thomas: Formal analysis, writing–review and editing. R. Nüsing: Resources, writing–review and editing. J. Steinmetz-Späh: Methodology, writing–review and editing. R. Savai: Resources, writing–review and editing. E. Fokas: Formal analysis, writing–review and editing. I. Fleming: Resources, writing–review and editing. F.R. Greten: Resources, writing–review and editing. K. Zarnack: Conceptualization, data curation, formal analysis, supervision, visualization, methodology, writing–original draft, writing–review and editing. B. Brüne: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. A. Weigert: Conceptualization, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing.

The authors thank Praveen Mathoor and Margarete Mijatovic for excellent technical assistance. This work was supported by Wilhelm-Sander Foundation (2019.082.01), Else Kröner-Fresenius Foundation (EKFS), Deutsche Krebshilfe (70114051), Deutsche Forschungsgemeinschaft (FOR 2438, TP3 and 8; SFB 1039, TP A06, B04, B06, and Z01; SFB902, TP B13; GRK 2336, TP1, 5 and 6), the LOEWE Center Frankfurt Cancer Institute (FCI) funded by the Hessen State Ministry for Higher Education, Research and the Arts [III L 5—519/03/03.001—(0015)], and “Förderung von Nachwuchsforschern” by Goethe-University Frankfurt (71000644).

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.