PTEN Loss Shapes Macrophage Dynamics in High-Grade Serous Ovarian Carcinoma

Macrophages with elevated HMOX1 expression are enriched in PTEN-deficient high-grade serous ovarian carcinoma, promote tumor growth, and represent a potential therapeutic target.

Abstract

High-grade serous ovarian carcinoma (HGSC) remains a disease with poor prognosis that is unresponsive to current immune checkpoint inhibitors. Although PI3K pathway alterations, such as PTEN loss, are common in HGSC, attempts to target this pathway have been unsuccessful. We hypothesized that aberrant PI3K pathway activation may alter the HGSC immune microenvironment and present a targeting opportunity. Single-cell RNA sequencing identified populations of resident macrophages specific to Pten-null omental tumors in murine models, which were confirmed by flow cytometry. These macrophages were derived from peritoneal fluid macrophages and exhibited a unique gene expression program, marked by high expression of the enzyme heme oxygenase-1 (HMOX1). Targeting resident peritoneal macrophages prevented the appearance of HMOX1^hi macrophages and reduced tumor growth. In addition, direct inhibition of HMOX1 extended survival in vivo. RNA sequencing identified IL33 in Pten-null tumor cells as a likely candidate driver, leading to the appearance of HMOX1^hi macrophages. Human HGSC tumors also contained HMOX1^hi macrophages with a corresponding gene expression program. Moreover, the presence of these macrophages was correlated with activated tumoral PI3K/mTOR signaling and poor overall survival in patients with HGSC. In contrast, tumors with low numbers of HMOX1^hi macrophages were marked by increased adaptive immune response gene expression. These data suggest targeting HMOX1^hi macrophages as a potential therapeutic strategy for treating poor prognosis HGSC.

Significance: Macrophages with elevated HMOX1 expression are enriched in PTEN-deficient high-grade serous ovarian carcinoma, promote tumor growth, and represent a potential therapeutic target.

Introduction

High-grade serous carcinoma (HGSC), the commonest type of ovarian cancer, remains a disease of poor prognosis, especially for patients whose tumors are classified as having proficient homologous recombination (1). The immune microenvironment has a strong prognostic effect on HGSC (2). The presence of intra-epithelial CD8⁺ T cells (3) and immunoreactive gene expression signatures (4) both correlate with improved overall survival (OS), whereas intratumoral regulatory T cells are associated with poor survival (5). However, responses to immune checkpoint inhibitors are poor with little correlation between response and either platinum-free interval or tumor-cell PD-L1 expression (6). Putative neoantigens can be identified in HGSC (7), but the average mutational burden is low (8).

PTEN is a tumor suppressor and negative regulator of the PI3K signaling pathway. Although deleterious single nucleotide variants and deletions in PTEN are rare in HGSC (4), inactivation through complex rearrangements is more frequent (9), and complete (13%–59%) or partial (13%–55%) PTEN protein loss is common (10, 11), suggesting transcriptional dysregulation. Furthermore, genomic alterations in the PI3K/RAS signaling pathway occur in 45% of HGSC (4), and enhanced PI3K pathway signaling is observed in the presence of PTEN protein expression (12). Thus, the PI3K pathway represents an important therapeutic target in HGSC. However, clinical trials of small molecule inhibitors have been largely negative so far (13), and there remains a need to identify effective therapeutic strategies for these tumors.

We hypothesized that loss of PTEN and activated PI3K signaling supports HGSC growth in part through an interaction with the tumor microenvironment. Using murine models, we identified that PTEN loss drives the expansion of a resident macrophage population in omental tumors, marked by high expression of the enzyme heme oxygenase 1 (HMOX1), which likely derives from peritoneal fluid macrophages. We demonstrated that similar HMOX1^hi macrophages can be identified in human HGSC samples and are associated with aberrant PI3K pathway activity and poor survival. Finally, we also show that targeting this population of macrophages may have therapeutic potential in HGSC.

Materials and Methods

Cell culture

Generation of ID8 cells with deletions in Trp53, Pten, and Brca2 has been described previously (14, 15). ID8 cells were cultured in high-glucose 4.5 g/L DMEM (Life Technologies #21969035) supplemented with 4% heat-inactivated FBS (Sigma and Life Technologies), 2 mmol/L glutamine (Life Technologies #25030024), ITS (Life Technologies #41400045; 10 μg/mL insulin, 5.5 μg/mL transferrin, and 6.7 ng/mL sodium selenite). HGS2 cells were purchased from Ximbio and were grown as previously described (16) in DMEM:F12 GlutaMAX (Life Technologies #31331028), 4% FBS, murine epidermal growth factor (20 ng/mL; Sigma #E4127) and hydrocortisone (100 ng/mL; Sigma #H0135), ITS. Early experiments were also performed on 100 U/mL penicillin, 100 μg/mL streptomycin, 250 ng/mL amphotericin B (Life Technologies 15240096), or 100 U/mL penicillin/100 μg/mL streptomycin. All cells were grown in 5% CO₂, 37°C with humidity, and used for a maximum of 10 passages. Cells were passaged using 0.1% Trypsin-EDTA (Gibco #15400054). Cells were regularly tested for Mycoplasma using the Lonza MycoAlert Detection Kit and were always negative.

In vivo experiments

All in vivo work was performed at the Central Biological Services facility, Imperial College London in accordance with the U.K. Animals (Scientific Procedures) Act 1986 under Project Licenses 70/7997, P2FEA2F22, and PA780D61A and following approval by the Imperial College AWERB (Animal Welfare and Ethical Review Body). Female C57BL6/J mice aged 6 to 7 weeks were purchased from Charles River, U.K. B6.129 (Cg)-Ccr2tm2.1Ifc/J; Ccr2^RFP/RFP) mice were purchased from JAX (strain #017586). Both aged-matched C57BL6/J mice and in-house–bred WT mice were used as controls for Ccr2^RFP/RFP mice. Female 494C57BL/6L Y5.1 (CD45.1) mice aged 6 to 7 weeks were purchased from Charles River (strain #494) and used at 17 weeks. HO-1-Luciferace-eGFP knock-in (Hmox1^GFP) mice were generated as previously (bioRxiv 2022.02.03.478952). All mice were acclimatized for at least 1 week prior to experiments. Mice were injected intraperitoneally with 1 × 10⁶ ID8 cells in 200-μL PBS or 10 × 10⁶ HGS2 cells in 200- to 300-μL PBS. Mice were monitored regularly and killed upon reaching the moderate severity limit as permitted by the Project License limits, which included weight loss, reduced movement, hunching, jaundice, and abdominal swelling.

In vivo treatments

Sn (IV) mesoporphyrin IX dichloride (SnMP; Inochem #SNM321) was dissolved in 0.1 M sterile NaOH and 0.5 M NaHCO₃ and injected intraperitoneally at 25 μmol/kg. For fixed time point and flow cytometry analysis, mice were injected with ID8 cells on day 0 and SnMP was administered once daily from day 14 to 28, and mice were harvested on day 28. For the survival experiment, mice were injected with ID8 cells on day 0, and SnMP was administered once daily from day 14 to 28, and mice were harvested upon when reaching humane endpoints. Mice with abdominal swelling but not yet at humane endpoint stopped receiving treatments (usually 1–3 days) before being culled to avoid bleeding.

RNA extraction and cDNA synthesis

Cell medium was removed from 24-well plates and 350 μL RLT buffer was added and frozen at −80°C. Plates were thawed on ice and 70% ethanol was added, gently mixed, and transferred into an RNeasy Micro Kit column (Qiagen #74104). RNA extraction was performed as per the manufacturer’s instructions, including the DNase step (Qiagen #79254). RNA was eluted in 30 μL nuclease-free H₂O and concentrated estimated using a NanoDrop; 2 μg of RNA was input into each 20 μL cDNA reaction using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems #4368814) under cycling conditions 25°C 10 minutes, 37°C 120 minutes, and 85°C 5 minutes. The cDNA was diluted in 140 μL nuclease-free H₂O. qRT-PCR reaction was set up using 9 μL cDNA, 1 μL primer, and 10 μL TaqMan Universal Master Mix II no UNG (Thermo Fisher Scientific #4440040). TaqMan primer probes were purchased from Thermo Fisher Scientific, Rpl34 (Mm01321800_m1), Ccl2 (Mm00441242_m1), Ccl7 (Mm00443113_m1), Csf1 (Mm00432686_m1), Raldh1 (Mm00657317_m1), Il6 (Mm00446190_m1), Vegfa (Mm00437306_m1), Il33 (Mm00505403_m1), and Actb (Mm02619580_g1). Samples were loaded in a 96-well plate (Applied Biosystems #4311971) and sealed with an Optical plate seal (Applied Biosystems #4346907) and analyzed on a StepOnePlus (Applied Biosystems).

SMART-Seq2 single-cell RNA sequencing

Briefly, mice were injected with ID8 Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) ID8 cells and omental tumors harvested at day 28, n = 4 mice per genotype. 44 macrophages per tumor were flow sorted based on DAPI⁻ (live), CD45⁺, CD11b⁺, Dump⁻ (CD3, CD19, Gr1), SiglecF⁻, F4/80⁺MHCII⁺ (Supplementary Table S1). SMART-Seq2 single-cell library preparation was performed by the Genomics Pipelines Group (Earlham Institute) and RNA-sequenced on Illumina NovaSeq 6000 SP Lane (150-bp paired end) with the aim for at least 1 million reads per cell. Further details on the analysis are provided in Supplementary Methods.

CD45.1 adoptive transfer

CD45.2 mice were injected with ID8-F3 or Pten1.14 cells on day 0 (n = 6 per group). Mice then received an adoptive transfer of peritoneal fluid cells either on day 1 (n = 3 for both groups) or 13 days after intraperitoneal injection (n = 3 for F3 and n = 2 for Pten1.14). A sterile peritoneal lavage (2 mmol/L EDTA in PBS) was performed on two CD45.1 mice, samples were combined and centrifuged at 330 g 4 minutes 4°C. Cells were incubated with 5 mL sterile RBC lysis buffer for 5 minutes at room temperature and washed in 15 mL PBS. A total of 605,000 cells per intraperitoneal injection were injected on day 1, and 400,000 cells, on day 13. Omental tumors were harvested at day 28 and stained for flow cytometry.

HMOX1^GFP macrophage adoptive transfer

A sterile peritoneal lavage (2 mmol/L EDTA, 0.5% FBS in PBS) was performed on healthy Hmox1^GFP mice (bioRxiv 2022.02.03.478952) and F4/80^hi CD102⁺ peritoneal macrophages were flow sorted. A total of 375,000 cells were adoptively transferred (AT) on day 21 into Hmox1^wt littermates previously injected with Trp53^−/−;Pten^−/− (Pten1.14) on day 0. Omental tumors were harvested on day 28 for flow cytometry.

Peritoneal resident macrophage ex vivo stimulation

A sterile peritoneal lavage (2 mmol/L EDTA, 0.5% FBS in PBS) was performed on healthy female C57BL/6 mice. Lavage cells were centrifuged 330 g for 5 minutes, and the pellet was resuspended in 10% FBS, 2-mmol/L glutamine in RPMI (Sigma #R5886); the cells were then plated in a 12-well plate overnight in 20-ng/mL M-CSF (BioLegend #576404). The next day nonadherent cells were washed off and peritoneal resident macrophages were stimulated with 50-ng/mL IL33 (BioLegend #580502) for 24 hours, following which, RNA extraction was performed.

Analysis of single-cell RNA sequencing data

The data from a previous single-cell RNA sequencing (scRNA-seq) study of 42 patients with high-grade serous ovarian cancer were analyzed (17) utilizing the Seurat v4.3.0 R package (18). HMOX1^hi macrophages were defined as macrophages that expressed HMOX1 > 1 SD above the mean. The built-in FindMarkers function in the Seurat package was used to identify differentially expressed genes (DEG), and those adjusted P values < 0.05 were considered as differentially expressed. Adjusted P values were calculated based on Bonferroni correction using all features in the dataset following the Seurat manual (https://satijalab.org/seurat/v3.0/de_vignette.html). Genes retrieved from Seurat analysis were displayed in a volcano plot using the enhancedVolcano package v1.14.0. MSigDB enrichment analysis of DEG between HMOX1^hi macrophages versus HMOX1^lo macrophages and between tumors with a high proportion of HMOX1^hi macrophages versus tumors with a low proportion of HMOX^hi macrophages was performed using msigdbr package v7.5.1 and clusterProfiler package v4.4.4 for Hallmark, Gene Ontology, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways.

Analysis of bulk ID8 RNA sequencing data

Bulk RNA sequencing (RNA-seq) data from Trp53^−/−;Pten^−/− and Trp53^−/− ID8 cells were generated as previously described (19). Briefly, raw fastq files were downloaded from GEO under accession GSE242835. These were checked for quality using FASTQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and aligned to mouse genome version GRCm38 (mm10) using STAR (20). Raw counts were generated using R package Rsubread (21) and differential gene expression was performed using DESeq2 (22). Ingenuity Pathway Analysis (IPA) Upstream Regulator Analysis software was used by supplying cluster 2 DEGs as input. IPA uses a z-score algorithm to make predictions, which has been described in detail (23). The results were overlapped with DEGs from bulk RNA-seq analysis of Trp53^−/−;Pten^−/− and Trp53^−/− ID8 cells. This naturally excluded chemicals and drugs while retaining transcription factors, cytokines, microRNAs, receptors, kinases, and so forth.

Prognostic value of HMOX1 in an independent validation cohort

The prognostic value of HMOX1 mRNA expression was evaluated using the Kaplan–Meier Plotter (http://kmplot.com/analysis/; ref. 24). To analyze the OS of patients with HGSC (defined as ovarian cancer with serous histology and TP53 mutation), we categorized the patients into two groups according to the best cutoff (high expression vs. low expression) of HMOX1 (ID 203665_at) and assessed differences by using a Kaplan–Meier survival plot with hazard ratios, 95% confidence intervals, and logrank P values.

Statistical analyses

A P value ≤ 0.05 was considered statistically significant. For IHC, statistical analyses were performed in R (v4.2.1). HMOX1^hi expression was categorized using the optimal threshold from the maximally selected rank statistics (survminer package v0.4.9). Comparison of survival curves was performed using the logrank test. We performed COX multivariate regression of HMOX1^hi expression on clinical parameters such as age and FIGO stage and drew a forest plot for visualization using the survival package v3.5.5. All other statistical analyses were performed using Prism v.9.4.1 (GraphPad).

Data availability

Publicly available data generated by others were used by the authors—the RNA-seq data analyzed in this study were obtained from GEO at GSE242835. All data, code, and materials are available upon request. ID8 Trp53^−/− and ID8 Trp53^−/−;Pten^−/− cells are available under the material transfer agreement via IAMcN. scRNA-seq data are available via ENA (accession number PRJEB67876).

Results

Pten-null cells are dependent on a tumor microenvironment for accelerated tumor growth

To address how PTEN loss influences HGSC growth, we utilized matched ID8 cells with inactivating mutations in Trp53 alone or both Trp53 and Pten that we generated previously (14, 15). Using multiple separate clones, we confirmed that Trp53^−/−;Pten^−/− ID8 cells lead to significantly shortened survival compared with Trp53^−/− following intraperitoneal injection (Fig. 1A). Pten deletion, as previously (25), did not decrease the in vitro doubling time in 2D high-attachment conditions (Fig. 1B and C), including those with an additional Brca2 mutation (Fig. 1D), and under low-serum and serum starvation conditions (Fig. 1E; Supplementary Fig. S1A), suggesting that enhanced intraperitoneal growth was not tumor-cell intrinsic.

Figure 1.

Pten-null cells are dependent on a tumor microenvironment for accelerated tumor growth. A, Survival curve for mice injected with ID8 Trp53^−/− (clones F3, M20, and C7) and Trp53^−/−;Pten^−/− (clones Pten1.12, Pten1.14, and Pten1.15), n = 6 per clone. Statistical significance was tested using the log-rank (Mantel–COX) test. B, ID8 Trp53^−/− (F3) and Trp53^−/−;Pten^−/− (Pten1.14) cells grown in flat high-attachment plates were imaged every 4 hours over 72 hours. Each data point represents the average of three to four technical replicates per clone and four images per replicate. Data were generated as phase object count per well normalized to the first scan (“0 hour”). Representative data from the experiment are shown. C, Mean doubling time of ID8 cells grown in conditions as in B for 72 hours. Each data point represents a clone grown at a different passage or different clone; an average of three to four wells, and four images per well were used to generate each data point. Clones were plated as follows: Trp53^−/− ID8-F3 (circle), ID8-M20 (square), and Trp53^−/−;Pten^−/−, ID8-F3; Pten1.14 (circle), ID8-F3;Pten1.15 (triangle). Significance was tested by an unpaired t test. D, Mean doubling time of ID8 subclones grown in 2D under same conditions as in C. Clones used were Trp53^−/− (F3), Trp53^−/−;Pten^−/− (Pten1.14), Trp53^−/−;Brca2^−/− (Brca2 2.14), and Trp53^−/−;Brca2^−/−;Pten^−/− (Brca2.14 Pten22). Statistical significance was tested using an ordinary one-way ANOVA, with Šidák multiple comparison test on selected pairs. E, ID8 Trp53^−/− (F3), Trp53^−/−;Pten^−/− (Pten1.14) cells were seeded and grown in 4%, 0.4%, or 0% FBS for up to 72 hours, and the doubling time was calculated as in C. Each symbol represents the average of technical triplicates, performed over three passages (P1, P4, and P5). Statistical significance was tested using an ordinary one-way ANOVA, with the Šidák multiple comparison test on selected pairs. F, ID8 clones Trp53^−/− (F3), Trp53^−/−;Pten^−/− (Pten1.14) were grown in low-attachment u-bottomed plates for 168 hours. Each symbol represents the average of technical triplicates, repeated on two passages. The largest brightfield object area (μm²) per image was quantified and shown over time. G, Mice were injected with either PBS, Trp53^−/− (F3), or Trp53^−/−;Pten^−/− (Pten1.14) ID8 cells on day 0 and a peritoneal lavage was performed on days 1, 2, 7, and 14. The ID8 cell count was estimated by flow cytometry (gated on as CD45-, SSC-A^hi, Live). Each point represents an individual mouse. Statistical significance was tested using an ordinary one-way ANOVA, with the Šidák multiple comparison test on selected pairs. H, Mice were injected with Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) ID8 cells on day 0 and the omental tumors harvested at days 14, 25, and 28. Tumor weights shown are pooled from several experiments, including control groups from other studies. Each point represents an individual mouse. Triangles indicate mice received artificial sweeteners in their drinking water for 14 days prior to ID8 injection. Significance was tested by an ordinary one-way ANOVA, with the Šidák multiple comparison test on selected pairs. I, Mean doubling times of HGS2 lentivirus–transduced clones grown for 72 hours in the same conditions as in C. Each circle represents an average of two technical replicates from one passage, with nine images taken per well. Clones E1, F6, and F8 were transduced with control GFP lentivirus, and clones C9, E11, and F3 were transduced with Pten GFP lentivirus. Significance was tested by an ordinary one-way ANOVA, with the Tukey multiple comparison test. J, HGS2 subclones were grown in low-attachment u-bottomed plates for 90 hours. Each symbol represents the average of technical triplicates per subclone, performed at a different passage. The largest brightfield object area (μm²) per image was quantified and is shown over time. K, HGS2 parental cells, control lentivirus, or Pten lentivirus–transduced subclones were injected intraperitoneally into mice, and omental tumors were harvested. Each symbol represents an individual mouse. Omental tumor weights are shown when harvested on days 56 (white) or 59 (filled). Control lentivirus clones F6 (triangle) and F8 (circle), and Pten lentivirus clones C9 (triangle), E11 (circle), and F3 (square). Significance was tested by one-way ANOVA, with the Šidák multiple comparisons test. In all experiments, results are considered significant when P < 0.05. ns, not significant.

Tumor cells must resist anoikis and then grow in low-attachment conditions to facilitate peritoneal dissemination in HGSC. Both Trp53^−/− and Trp53^−/−;Pten^−/− ID8 clones formed spheroid-like clusters in low-attachment conditions in vitro (Supplementary Fig. S1B), but contraction rates were equal over time in both genotypes (Fig. 1F). In vivo, Pten loss conferred no immediate survival advantage following intraperitoneal injection (Fig. 1G; Supplementary Fig. S1C). However, 14 days following injection, there was a significant expansion of Pten-null cells in peritoneal fluid (Fig. 1G). Moreover, tumor burden in the omentum, the dominant site of metastasis in HGSC, was greater by day 14, and significantly greater by days 25 to 28 (Fig. 1H) in mice injected with Trp53^−/−;Pten^−/− cells.

To ensure our findings were not ID8-specific, we utilized HGS2, a cell line generated from tumors arising in a Trp53^fl/fl;Brca2^fl/fl;Pten^fl/fl;Pax8^Cre transgenic mouse (16). We re-expressed Pten in HGS2 using a lentivirus (Supplementary Fig. S1D and S1E), which did not impact doubling time in 2D high attachment (Fig. 1I) or the ability to form spheroids in low-attachment (Fig. 1J; Supplementary Fig. S1F) but produced smaller omental tumors in vivo compared with control virus-infected cells (Fig. 1K). Together, these data suggest strongly that the peritoneal microenvironment supports accelerated growth of Pten-null tumors.

Pten-null tumor cells enhance the accumulation of resident-like macrophages within the omentum

We hypothesized that macrophages support Pten-null tumor seeding and growth. Two dominant macrophage populations exist in the peritoneal cavity and omentum in mice; F4/80^loMHCII^hi monocyte-derived macrophages, which are constantly replenished by blood Ly6C^hi monocytes, and embryonically derived F4/80^hiMHCII^lo resident macrophages, which are both self-maintained and replenished from the local F4/80^loMHCII^hi pool (26, 27). Having confirmed the specificity of our gating strategy (Supplementary Fig. S2A and S2B), we first assessed how macrophage/monocyte populations altered during tumor growth. ID8 cell injection caused an influx of Ly6C^hi monocytes into the peritoneal fluid within 1 day. This infiltration significantly increased by day 14 in Trp53^−/−;Pten^−/−-injected mice (Fig. 2A). An increase in F4/80^loMHCII^hi macrophages, likely to derive from this monocyte pool (Fig. 2B), was also evident 14 days after Trp53^−/−;Pten^−/− injection. The omental resident F4/80^hiMHCII^lo population increased at day 14, in Trp53^−/−;Pten^−/−-injected mice (Fig. 2C). Interestingly on day 28, when both genotypes had substantial tumor burden, Trp53^−/−;Pten^−/− omental tumors contained significantly more resident-like macrophages across multiple subclones (Fig. 2D). Furthermore, approximately 40% of resident-like macrophages expressed the long-term residency marker TIM4⁺, indicating that they are not newly recruited (Fig. 2E and F).

Figure 2.

Pten-null tumor cells enhance the accumulation of resident-like macrophages within the omentum. A–C, Mice were injected with either ID8 Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) cells. Peritoneal fluid and omenta were harvested 1, 2, 7, and 14 days later. Flow cytometry was performed for indicated cell populations. Counting beads were used to estimate absolute cell numbers, normalized to either the total lavage fluid (mL) or per omentum. A, Gating strategy for monocytes: Zombie Yellow⁻, CD45⁺, CD11b⁺, Ly6C^hi. B and C, Macrophages: Zombie Yellow⁻, CD45⁺, CD11b⁺, Ly6C⁻, Ly6G⁻, SiglecF⁻, F4/80^lo, MHCII^hi (monocyte-derived; B), or F4/80^hi, MHCII^lo (resident-like; C). Every data point represents an individual mouse. Statistical significance was tested using a one-way ANOVA with the Šidák multiple comparison test on selected samples. D–F, Mice were injected with individual ID8 Trp53^−/− (F3, C7, and M20) or Trp53^−/−;Pten^−/− clones (Pten1.12, Pten1.14, and Pten1.15) on day 0 and omental tumors harvested at day 28 for flow cytometry. D, Resident macrophages were defined as Zombie Yellow⁻, CD45⁺, CD11b⁺, Ly6C⁻, Ly6G⁻, SiglecF⁻, F4/80^hi, and MHCII^lo and normalized to omental tumor weight (mg). E, Representative gating strategy used to define TIM4⁺ cells within the F4/80^hiMHCII^lo population. F, Quantification of percentage of TIM4⁺ cells out of total F4/80^hiMHCII^lo macrophages. Statistical significance was tested using a one-way ANOVA, with the Šidák multiple comparison test on selected samples. G, Mice were injected with ID8 Trp53^−/− (F3), Trp53^−/−;Pten^−/− (Pten1.14), Trp53^−/−;Brca2^−/− (Brca2 2.14), or Trp53^−/−;Brca2^−/−;Pten^−/− (Brca2.14 Pten22) clones and the number of resident macrophages quantified by flow cytometry as in D. H, Density of monocyte-derived, defined as Zombie Yellow⁻, CD45⁺, CD11b⁺, Ly6C⁻, Ly6G⁻, SiglecF⁻, F4/80^lo, and MHCII^hi cells in omental tumors of same mice as in D. I, Density of T cells, defined as Zombie Yellow⁻, CD45⁺, and CD3⁺ in omental tumors of same mice as in D. J, Density of monocyte-derived, defined as Zombie Yellow⁻, CD45⁺, CD11b⁺, Ly6C⁻, Ly6G⁻, SiglecF⁻, F4/80^lo, and MHCII^hi cells in omental tumors of same mice as in G. K, Density of T cells, defined as Zombie Yellow⁻, CD45⁺, and CD3⁺ in omental tumors of same mice as in G. L, CEL (n = 6 mice) or PBS (n = 3) were injected intraperitoneally into mice on −14, −7, and −1 days prior to Trp53^−/−;Pten^−/− (Pten1.12) tumor-cell injection. CEL or PBS was then administered on days +7, +14, +21. Mice were harvested on day 26 (circles), apart from one PBS-treated mouse that reached the endpoint at day 23 (triangle), and the omental tumor weight (mg) and ascites fluid volume (mL) were analyzed. Statistical significance was tested using an unpaired t test. M,Ccr2^+/+, Ccr2^RFP/+, Ccr2^RFP/RFP (clear symbols), or in-house wild-type (filled symbols) age-matched mice were injected with either Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) ID8 cells on day 0 and culled on day 28. Omental tumor weight (mg) and ascites fluid volume (mL) were measured. Statistical significance was tested using a one-way ANOVA, with the Šidák multiple comparison test on selected samples (omental tumors) or with the Tukey multiple comparison test (ascites). In all experiments, results are considered significant when P < 0.05.

We also used Trp53^−/−;Brca2^−/− ID8 cells with or without Pten deletion. Loss of Brca2 alone did not impact resident macrophage expansion (Fig. 2G). However, the additional deletion of Pten again significantly increased the density of resident macrophages (Fig. 2G), which was combined with a relative paucity of monocyte-derived macrophages and T cells (Fig. 2H and I). This is coupled with an increase in in vivo aggressiveness (15). Deletion of Brca2 in addition to Pten rescued the recruitment of monocyte-derived macrophages and T cells observed with Pten loss alone, (Fig. 2J and K), suggesting that Pten deletion alters resident macrophages specifically rather than inducing global changes in the immune microenvironment (Supplementary Fig. S3A and S3B).

To determine if resident macrophages are drivers of Pten-null omental tumor growth, we first depleted all macrophages using intraperitoneal clodronate-encapsulated liposomes (CEL) prior to tumor inoculation (Supplementary Fig. S4A). This pan-macrophage depletion completely prevented Trp53^−/−;Pten^−/− omental tumor formation and ascites production (Fig. 2L). Unfortunately, high mortality observed following CEL injection, as previously reported (28), precluded further studies using CEL in our institution. To dissect macrophage contribution to Pten-null tumor growth further, we utilized mice that lack either one (Ccr2^RFP/+) or both copies (Ccr2^RFP/RFP) of Ccr2 and consequently have markedly reduced bone marrow monocyte egress. Monocytes and monocyte-derived macrophages were significantly reduced in the peritoneal fluid and omentum in both Ccr2^RFP/+ and Ccr2^RFP/RFP mice as expected, with no alteration in the resident macrophage pool (Supplementary Fig. S4B–S4E) or consistent alterations in other populations (Supplementary Fig. S4F–S4J). Strikingly, deletion of Ccr2 significantly increased tumor burden and ascites volume in Trp53^−/−;Pten^−/−-injected mice (Fig. 2M), indicating that the monocyte-derived macrophages have a protective antitumoral role during Pten-null tumor seeding and growth.

Pten-null tumors drive the appearance of a unique HMOX1^hi macrophage subpopulation

Macrophage phenotypes cannot be simplified to binary M1/M2 marker expression and multiple distinct subtypes exist in omental tumors (29, 30). Thus, we performed scRNA-seq on flow sorted macrophages from omental tumors using the SMART-Seq2 protocol (Supplementary Fig. S5A; ref. 31). Uniform Manifold Approximation and Projection clustering revealed five distinct clusters (Fig. 3A and B). Cluster 0 expressed genes classically found in monocyte-derived macrophages, including MHCII-associated molecules (H2-Eb1, H2-DMb2, H2-DMb1, H2-Ab1, Cd74, H2-Oa, and H2-Aa), chemokine receptors Ccr2, Cx3cr1, and costimulatory molecule Cd86 (Supplementary Table S2). Cluster 0 also localized in the F4/80^loMHCII^hi region by flow cytometry (Fig. 3C). Clusters 1 and 3 both expressed genes defined in peritoneal macrophages. Cluster 1 expressed Fcna (ficolin 1), Fn1 (fibronectin 1) and the retinoid X receptor Rxra (Supplementary Table S2) and was predominantly located in the F4/80^loMHCII^hi region by flow cytometry (Fig. 3C). Cluster 3 expressed many more canonical peritoneal resident genes, including Ltbp1, Garnl3, Serpinb2, Alox15, Selp, F5, Timd4, Icam2 (CD102), and Gata6 (Supplementary Table S2). Cluster 3 localized in the F4/80^hiMHCII^lo region by flow cytometry (Fig. 3C), which, taken together with the expression of TIM4 (Timd4), suggests that cluster 1 is a monocyte-derived precursor that transitions into cluster 3. Cluster 4 expressed genes are normally found in epithelial or mesothelial cells (Krt18, Krt19, Msln, Wt1), which suggests that they may be phagocytic macrophages (Supplementary Table S2).

Figure 3.

Pten-null tumors drive accelerated formation of unique HMOX1^hi macrophage subpopulation. A, Mice were injected with ID8 Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) ID8 cells on day 0 and omental tumors harvested at day 28, with n = 4 mice per genotype. Macrophages were single-cell flow-sorted based on DAPI⁻ (live), CD45⁺, CD11b⁺, Dump⁻ (CD3, CD19, and Gr1), SiglecF⁻, F4/80⁺MHCII⁺, singlets, and analyzed by plate-based SMART-Seq2 scRNA-seq. Following quality filtering, a Uniform Manifold Approximation and Projection (UMAP) of macrophages is shown, using the Seurat pipeline. Selected significantly DEGs (defined as adjusted P value < 0.05 and average log₂-fold change >0) are shown next to the respective cluster. Total DEGs per cluster are 1,022 genes (cluster 0); 59 genes (cluster 1); 425 genes (cluster 2); 141 genes (cluster 3); and 1,625 genes (cluster 4). B, The percentage of macrophages identified in each cluster isolated from either ID8 Trp53^−/− or Trp53^−/−;Pten^−/− omental tumors from A is shown. C, The expression of F4/80 and MHCII per macrophages, as collected during index sorting is shown with cluster identity overlaid by color. D, Data in A were reanalyzed using the Monocle 3 package and Pseudotime analysis applied (shown as heatmap), with the root node placed in cluster 0. E, Mice were injected with ID8 Trp53^−/− or Trp53^−/−;Pten^−/− ID8 cells on day 0 and omental tumors harvested at early (day 28, Trp53^−/−; day 21, Trp53^−/−;Pten^−/−, “E”) and late (day 47, Trp53^−/−; day 28, Trp53^−/−;Pten^−/−, “L”) time points. The density of macrophages in omental tumors was calculated for F4/80⁺MHCII⁺, CX3CR1⁺MHCII^hiCD86⁺CD11c⁺. Statistical significance was tested by one-way ANOVA and Tukey multiple comparison test. F, As in E, the density of macrophages in omental tumors was calculated for F4/80⁺MHCII⁺, LYVE1⁻CD102⁺TIM4⁺. Statistical significance was tested by one-way ANOVA and Tukey multiple comparison test. G, As in E, the density of macrophages in omental tumors was calculated for F4/80⁺MHCII⁺, LYVE1⁻CD102⁻TIM4⁻Arginase1⁺PD-L1⁺. Statistical significance was tested by one-way ANOVA and Tukey multiple comparison test. H, As in E, the density of macrophages in omental tumors was calculated for F4/80⁺MHCII⁺, HMOX1^hi. Statistical significance was tested by one-way ANOVA and Tukey multiple comparison test. I, ID8 Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) omental tumors harvested at day 28 were stained for HMOX1 by IHC. The number of HMOX1^hi cells was quantified using QuPath. Statistical significance was tested using an unpaired t test. J, Representative IHC images of HMOX1 (brown stain) from Trp53^−/− and Trp53^−/−;Pten^−/− tumors from I are shown. Scale bar is indicated in the image. K, HMOX1^GFP mice (n = 2) were injected with ID8 Trp53^−/−;Pten^−/− (Pten1.14) cells on day 0, and omental tumors harvested at day 25. Representative histogram of HMOX1GFP expression is shown per cell population. The CD45⁻ population will contain transgenic stromal cells as well as the GFP⁻ ID8 cells. L, Left, gating strategy used to define cluster 2; F4/80⁺MHCII⁺, LYVE1⁻, CD11c⁻, MHCII^lo, CD102⁻, F4/80^lo, HMOX1^hi, arginase1⁺, and PD-L1⁺. Right, density of cluster 2 macrophages in day 28 ID8 Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) tumors. Statistical significance was tested using an unpaired t test. In all experiments, results are considered significant when P < 0.05.

The most interesting cluster was cluster 2, which was found almost exclusively in Trp53^−/−;Pten^−/− tumors (Fig. 3A and B) and had high expression of heme oxygenase 1 (Hmox1), an enzyme that catalyzes the breakdown of heme into carbon monoxide, iron (Fe²⁺) and biliverdin. Cluster 2 also expressed genes involved in lipid accumulation, such as Trib3 (Tribbles pseudokinase-3) and Lgals3 (galectin 3), as well as genes that protect against heavy metal toxicity, such as metallothioneins Mt1 and Mt2 (Supplementary Table S2). Cluster 2 also expressed genes associated with immunosuppression, including Cd274 (PD-L1), Arg1 (arginase 1), and Vegfa (Supplementary Table S2). Cluster 2 localized mainly in the F4/80^hiMHCII^lo region by flow cytometry, suggesting that it may derive from resident peritoneal macrophages (Fig. 3C; Supplementary Fig. S5B). We performed pseudotime analysis using Monocle3 (32) to estimate the putative direction of differentiation. When re-clustered (Supplementary Fig. S5C) with cluster 0 selected as the starting node, pseudotime predicted that cluster 2 derived from Clusters 1 and 3 (Fig. 3D), indicating that it represents a subtype of peritoneal resident macrophage.

We confirmed the presence of these macrophage subpopulations by flow cytometry at early (proceeding ascites formation) and late (ascites present) time points (Fig. 3E and F; Supplementary Fig. S5C–S5E). This confirmed that cluster 2 macrophages (defined as either Arginase1⁺PD-L1⁺ or HMOX1^hi) were present early in Trp53^−/−;Pten^−/− tumors and increased significantly in late tumors (Fig. 3G and H; Supplementary Fig. S5D and S5E). We also validated the presence of HMOX1⁺ cells in ID8 omental tumor sections using IHC (Fig. 3I), in which they were observed surrounding adipocytes and in tumor borders (Fig. 3J).

We next assessed the selectivity of HMOX1 expression in cluster 2. Using Hmox1^GFP transgenic mice (33), we confirmed that only monocytes and macrophages express HMOX1 (Fig. 3K). Although the LYVE1⁺ mesothelial lining population, which represents a small proportion of total macrophages, had the highest expression, cluster 2 (Arginase1⁺PD-L1⁺) highly expressed HMOX1, followed by CD102⁺ peritoneal macrophages. CD11c⁺MHCII^hi monocyte-derived macrophages and monocytes had weak expression and other populations had weak/no expression. When combined, our data show that Arginase1⁺PD-L1⁺HMOX1^hi macrophages are significantly enriched in Trp53^−/−;Pten^−/− tumors (Fig. 3L).

HMOX1^hi macrophages derive from resident peritoneal fluid macrophages

To test the hypothesis that peritoneal fluid resident macrophages were the primary source of HMOX1^hi macrophages, we first AT CD45.1⁺ peritoneal fluid cells (which will include monocytes, monocyte-derived and resident macrophages) into CD45.2⁺ mice 24 hours or 13 days following ID8 cell injection. CD45.1⁺ cells were detected in omental tumors, proving that trafficking can occur between peritoneal fluid and tumor (Fig. 4A). Interestingly, resident CD45.1⁺F4/80^hiMHCII^lo cells were enriched in Trp53^−/−;Pten^−/− tumors (Fig. 4B and C, left) and correspondingly depleted in ascites (Fig. 4B and C, middle). The majority of CD45.1⁺ macrophages were TIM4⁺, indicating long-term residency (Fig. 4B and C, right). To determine further if HMOX1^hi macrophages can derive from peritoneal fluid, we sorted peritoneal fluid F4/80^hiCD102⁺ cells from healthy HMOX1^GFP mice and AT them into HMOX1^wt littermates bearing Trp53^−/−;Pten^−/− tumors (Fig. 4D). We detected HMOX1^GFP cells in Trp53^−/−;Pten^−/− omental tumors (Fig. 4E), which phenotypically copied the host’s own population (CD11c⁻MHCII^lo, CD102⁺Arginase1⁺) and almost exclusively came from long-term resident CD102⁺TIM4⁺ cells (Fig. 4F). However, a fraction of CD102⁻ cells in the host macrophage pool had both high arginase 1 and PD-L1 expression (Fig. 4G and H), suggesting that some cluster 2 macrophages may also derive from non-CD102⁺ peritoneal fluid cells.

Figure 4.

HMOX1^hi macrophages are partially derived from resident peritoneal fluid macrophages. A, CD45.2 mice were injected with ID8 Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) cells on day 0 (n = 6 per group). Mice then received an AT of CD45.1 peritoneal fluid cells on either day 1 (n = 3) or day 13 (n = 3 for F3 and n = 2 for Pten1.14) after ID8 intraperitoneal injection. Tumors and ascites were harvested at day 28. One mouse was excluded as there were insufficient cells and thus became a negative control. Representative flow cytometry gating strategy for live CD45.1 and CD45.2 cells in omental tumors. An FMO-CD45.1 control and no AT control are also shown. B, The omental tumors from A were analyzed by flow cytometry. The percentage of resident F4/80^hiMHCII^lo macrophages of all CD45.1 cells in omental tumor (left) and ascites (middle) are shown for mice that received CD45.1 AT 24 hours post ID8 injection. The percentage of TIM4⁺ out of F4/80^hiMHCII^loCD45.1⁺ macrophages in omental tumors is also shown (right). One mouse had no detectable F4/80^hiMHCII^lo macrophages, therefore the %TIM4⁺ value was not able to be analyzed. Black values are from ID8 Trp53^−/− (F3)-injected mice and pink are from Trp53^−/−;Pten^−/− (Pten1.14)-injected mice. Statistical significance was tested by the unpaired t test. C, The omental tumors from A were analyzed by flow cytometry. The percentage of resident F4/80^hiMHCII^lo macrophages of all CD45.1 cells in the omental tumor (left) and ascites (middle) are shown for mice that received CD45.1 AT 13 days post ID8 injection. The percentage of TIM4⁺ out of F4/80^hiMHCII^loCD45.1⁺ macrophages in omental tumors is also shown (right). Black values are from ID8 Trp53^−/− (F3)-injected mice and pink are from Trp53^−/−;Pten^−/− (Pten1.14)-injected mice. Statistical significance was tested by the unpaired t test. D, F4/80^hi CD102⁺ peritoneal macrophages were FACS sorted from healthy Hmox1^GFP mice and AT into Hmox1^wt littermates bearing ID8 Trp53^−/−;Pten^−/− (Pten1.14) tumors on day 21. Omental tumors and ascites were harvested on day 28. E, Representative FACS plot of GFP⁺ cells in the CD45⁺ live gated cells in an omental tumor. The relative fluorescence of GFP⁺ cells (green) compared with GFP⁻ cells (gray) is shown for markers CD11b, MHCII, CD11c, CD102, arginase1, and PD-L1. F, Macrophages were gated as previously, and the percentage of cells within each gate out of total detected GFP⁺ cells is shown. G, The percentage of each macrophage population gated within GFP⁺ (green) or GFP⁻ (gray) cells that is positive for arginase1. H, The percentage of each macrophage population gated within GFP⁺ (green) or GFP⁻ (gray) cells that is positive for PD-L1. In all experiments, results are considered significant when P < 0.05. ns, not significant.

HMOX1 inhibition extends survival

To understand if HMOX1 is a potential therapeutic target in HGSC, we used the HMOX1 inhibitor tin mesoporphyrin (SnMP; ref. 34). SnMP treatment did not impact the omental density of MHCII^hiCD11c⁺ monocyte-derived, or CD102⁺F4/80^hi macrophages (Fig. 5A and B), but did increase the density of Arginase1⁺PD-L1⁺HMOX1⁺ macrophages (Fig. 5C). This is likely to be a compensatory mechanism in response to HMOX1 inhibition, as SnMP is known to stimulate HMOX1 upregulation while still blocking enzyme activity (35). Interestingly, HMOX1 inhibition depleted LYVE1⁺ macrophages (Fig. 5D), which suggests they are susceptible to heme-induced toxicity, in line with their proximity to tumor vasculature (36). SnMP treatment for 14 days reduced ascites formation, albeit nonsignificantly (P = 0.078; Fig. 5E). However, extended SnMP, given as a 5 days on/2 days off regime, significantly increased survival of Trp53^−/−;Pten^−/− tumor-bearing mice (Hazard ratio 0.32, 95% CI, 0.11–0.94, P = 0.002; Fig. 5F).

Figure 5.

HMOX1 inhibition extends the survival in mice bearing Pten-null ID8 tumors. A, Mice were injected with ID8 Trp53^−/− (F3) or Trp53^−/−;Pten^−/− (Pten1.14) cells on day 0. From day 14, mice received 25 μmol/kg SnMP (n = 6) or vehicle control (n = 6) daily for 14 days. Omental tumors were harvested on day 28 and analyzed by flow cytometry. The density of CD11c⁺MHCII^hi macrophages in the omental tumors is shown. Statistical significance was tested using one-way ANOVA and with the Šidák multiple comparisons test with selected comparisons. B, The density of CD102⁺F4/80^hi macrophages from A is shown per mg omental tumor. Statistical significance was tested using one-way ANOVA and Šidák multiple comparisons test, with selected comparisons. C, The density of Arginase1⁺PD-L1⁺HMOX1⁺ macrophages from A is shown per mg omental tumor. Statistical significance was tested using one-way ANOVA and Šidák multiple comparisons test, with selected comparisons. D, The density of LYVE1⁺ macrophages from A is shown per mg omental tumor. Statistical significance was tested using one-way ANOVA and Tukey multiple comparison test. E, The ascites volume from A. Statistical significance was tested using one-way ANOVA and Šidák multiple comparisons test, with selected comparisons. F, Mice were injected with ID8 Trp53^−/−;Pten^−/− (Pten1.14) cells on day 0. From day 14, mice received 25 μmol/kg SnMP (n = 9) or vehicle control (n = 9) daily on 5 days on/2 days off schedule until mice were harvested reached the humane endpoint, which included advanced abdominal swelling. One mouse was censored in the SnMP group as it was killed before reaching the endpoint at the end of the study (day 42); it had minimal disease present. Statistical significance was tested using a logrank (Mantel–COX) test. G, Volcano plot depicting log₂-normalized fold change and FDR-adjusted P values for gene expression differences between Trp53^−/−;Pten^−/− and Trp53^−/− ID8 cells. Cluster 2 activators are individually labeled. H, List of cluster 2 activators and inhibitors as identified through the IPA analysis. I, Venn diagram showing the overlap of genes upregulated in Trp53^−/−;Pten^−/− cells and genes that are predicted to activate cluster 2 genes identified through the IPA analysis. Among these 25 predicted activators, 16 were found common among the genes upregulated in Trp53^−/−;Pten^−/− bulk RNA-seq analysis. J, Bicycle wheel diagram showing the 29 genes expressed by cluster 2 macrophages that are activated or inhibited by Il33, identified through the IPA analysis. K,Il33 gene (left) and IL33 protein expression (right) in ID8 clones. Each data point represents a clone. Clones plated as follows: Trp53^−/− ID8-F3 (circle), ID8-M20 (square), ID8-C7 (triangle) and Trp53^−/−;Pten^−/−, ID8-F3; Pten1.12 (square) ID8-F3; Pten1.14 (circle), ID8-F3;Pten1.15 (triangle). Significance was tested by an unpaired t test. L, Peritoneal fluid macrophages were cultured in vitro for 24 hours in the presence of 50-ng/mL recombinant IL33 protein or untreated media control. Hmox1 gene expression was then assessed. In all experiments, results are considered significant when P < 0.05. ns, not significant.

IL33 as a potential driver of cluster 2 macrophages

To identify how Pten-null tumor cells drive resident macrophage expansion, we initially screened chemokine and cytokine expression. This identified significantly increased Ccl2 and Ccl7 expression in some Pten-null cells (Supplementary Fig. S6A–S6C) but not in cells additionally lacking Brca2 (Supplementary Fig. S6B and S6C). We also analyzed the expression of retinoic acid–producing enzymes Raldh1, 2, and 3 as peritoneal and omental resident macrophages are supported by retinoic acid, which drives their resident gene expression program, including Gata6. However, Raldh1 expression was not consistently altered by Pten deletion (Supplementary Fig. S6D), whereas Raldh2 and 3 expression was negligible in all cells. Furthermore, Trp53^−/−;Pten^−/− cells did not demonstrate an enhanced ability to recruit bone marrow–derived macrophages (Supplementary Fig. S6E), whereas deletion of Ccr2 caused a reduction in bone marrow–derived macrophage recruitment to both genotypes (Supplementary Fig. S6F). PTEN deletion can drive IL6 production in prostate cancer (37). However, Il6 expression was weak in Trp53^−/−;Pten^−/− ID8 cells (C_T values 34–37) with no statistical difference between clones (Supplementary Fig. S6G), although IL6 protein was undetectable by ELISA (data not shown). Vegfa, encoding vascular endothelial growth factor, was also unaltered by Pten loss (Supplementary Fig. S6H). Taken together, these data suggested that one or more factors beyond Ccl2 and Ccl7 are produced in vivo that support resident macrophage recruitment and expansion. The gene signature of cluster 2 allowed us to dissect out potential drivers of HMOX1^hi macrophages further. We first analyzed bulk RNA-seq analysis of Trp53^−/−;Pten^−/− ID8 cells (19). Using this approach, we found 4,505 genes significantly upregulated in Trp53^−/−;Pten^−/− compared with Trp53^−/− ID8 cells (Fig. 5G). Using IPA (Fig. 5H), we identified 25 potential upstream regulators of cluster 2, of which, 16 overlapped with genes also upregulated in Pten-null cells (Fig. 5I). Among these genes, we identified Il33 as a likely candidate to stimulate cluster 2 gene expression, as it is both a secreted factor and predicted to activate one of the largest groups of genes in cluster 2 (Fig. 5J). We confirmed increased Il33 gene and IL33 protein expression in Pten-null ID8 lines (Fig. 5K). We then cultured peritoneal fluid resident macrophages in the presence of IL33 and observed significantly increased Hmox1 gene expression (Fig. 5L).

Mouse and human HMOX1^hi macrophages share common characteristics

To ensure our murine data were relevant to patients with HGSC, we analyzed a large scRNA-seq dataset, which contained data from 160 biopsies from 42 newly diagnosed, treatment naïve patients with HGSC (17). Tumor-associated macrophages (n = 166,895) were annotated based on known marker genes including PTPRC, CD14, FCER1G, and CD68. We first categorized human macrophages based on HMOX1 expression, defining those with a scaled HMOX1 expression >1 SD above the mean as HMOX1^hi (Fig. 6A). We then compared DEGs in HMOX1^hi macrophages with those for each mouse cluster. Cluster 2 showed the highest number of overlapping genes (n = 39) with those in HMOX1^hi cells (Fig. 6B). Similarly, of the 87 DEG in HMOX1^hi macrophages, the highest proportion (44.8%, n = 39/87) was shared with cluster 2 (Supplementary Fig. S7A). HMOX1^hi macrophages were enriched in key signature genes for resident macrophages (LYVE1), heme metabolism (BLVRB), cellular response to hypoxia (HILPDA), metallothioneins (MT1E, MT1F, MT1G, MT1H, and MT1M), iron transporter, storage, and homeostasis (SLC40A1, HAMP, FTH1, and FTL) and lipid metabolism and storage (APOC1, PLIN2, and LIPA; Fig. 6C; Supplementary Table S3). Conversely, HMOX1^lo macrophages were enriched in interferon gamma response genes (CXCL9 and CXCL10), immune cell and T-cell recruitment genes (CCL5, CXCL9, CXCL10, CXCL11, and IL1B) and MHCII gene (HLA-DQA1; Fig. 6C; Supplementary Table S3). MSigDB (38) enrichment analysis of HMOX1^hi macrophage transcriptomes revealed enrichment of hypoxia response, ion homeostasis, lipid metabolism, and mTORC1 signaling pathways that were also found in mouse cluster 2 (Fig. 6D; Supplementary Fig. S7B and S7C). Thus, human HGSC contains a cluster of macrophages that share common characteristics with mouse cluster 2 and are characterized by high HMOX1 expression, tissue residency, oxidative stress response, and low expression of proinflammatory cytokines/chemokines.

Figure 6.

Mouse and human HMOX1^hi macrophages share common characteristics in HGSC. A, Scaled and centered HMOX1 expression on tumor-associated macrophages. Macrophages with a scaled HMOX1 expression above one SD from the mean were defined as HMOX1^hi. Macrophages with a scaled HMOX1 expression below one SD from the mean were defined as HMOX1^lo. B, The overlap between DEG found in human HMOX1^hi macrophages and DEG found in each mouse macrophage cluster 0–4 is shown. C, DEG in HMOX1^hi macrophages (right side of the volcano plot) and HMOX1^lo macrophages (left side) defined in A from human HGSC tumors is shown. D, Comparison of MSigDB pathway enrichment in human HMOX1^hi macrophages and mouse cluster 2 macrophages showing selected pathways of interest that were significantly enriched (Hallmark, Gene Ontology, and KEGG).

HMOX1^hi macrophages associate with poor OS and PI3K signaling pathway activation in HGSC

In mice, cluster 2 macrophages were found almost exclusively in Pten-null tumors (Fig. 3A and B). In the scRNA-seq dataset, cancer cells from tumors enriched in HMOX1^hi macrophages (Supplementary Fig. S8A) exhibited high mTOR signaling and high insulin-like growth factor signaling (Fig. 7A), which can activate PI3K/AKT signaling (39). By contrast, immune-related pathways were downregulated in tumors enriched in HMOX1^hi macrophages (Fig. 7A).

Figure 7.

A high proportion of HMOX1^hi macrophages is associated with poor OS and PI3K signaling pathway activation. A, MSigDB enrichment analysis (Hallmark, Gene Ontology, and KEGG) of HGSC tumors with high versus low proportion of HMOX1^hi macrophages showing selected pathways of interest that were significantly enriched (left) or downregulated (right). Pathways relating to PI3K signaling are highlighted in red. B, CD68 (top left) and HMOX1 (top right) IHC staining in the BriTROC-1 study tissue microarray. QuPath positive cell detection is shown (red) for CD68 (bottom left) and HMOX1 (bottom right). Scale bar, 200 µm. C, Spearman correlation between the proportion of HMOX1^hi macrophages and the proportion of CD68⁺ macrophages found in BriTROC-1 tissue microarray cores. D,pAKT staining in the BriTROC-1 study with (left) and without (right) the QuPath tumor classifier showing weak (1+), moderate (2+), and strong (3+) staining. E, Spearman correlation between pAKT tumor H-score and the average proportion of HMOX1^hi macrophages per patient in the BriTROC-1 study. F, OS of patients in the BriTROC-1 study with high (n = 76) and low (n = 50) proportion of HMOX1^hi, in which the cutoff is based on the optimal threshold. Statistical comparison was performed using the log-rank test. G, Multivariate regression forest plot of HMOX1^hi expression.

IHC on diagnostic HGSC samples from 172 patients in the BriTROC-1 study (40) demonstrated a strong correlation between HMOX1 and CD68 (Fig. 7B and C), allowing us to use high HMOX1 expression as a surrogate for HMOX1^hi macrophages. The presence of HMOX1^hi macrophages positively correlated, albeit weakly, with positive pAKT (S473) staining in tumor cells (Fig. 7D and E; Supplementary Fig. S8B) and was also independently associated with reduced survival in BriTROC-1 patients, after adjustment for age and stage [HR = 1.80 (1.07–3.0); Fig. 7F and G]. The prognostic impact of high HMOX1 expression was confirmed in a separate validation cohort at the mRNA level (Supplementary Fig. S8C).

Discussion

PTEN loss and other PI3K signaling alterations are frequent in HGSC but have proven challenging to target therapeutically. In this study, we have used mouse models and human HGSC samples to demonstrate that PI3K signaling pathway activation is associated with poor survival and the presence of HMOX1^hi macrophages. Importantly, we have shown that targeting this population with a specific HMOX1 inhibitor, SnMP, extends survival in mice. This suggests that targeting deleterious tumor-infiltrating macrophages has therapeutic potential.

Macrophages are abundant in HGSC (41) but they have thus far eluded therapeutic targeting. Multiple macrophage subtypes with diverse functions exist in HGSC (17, 29, 30), and resident macrophages, derived from embryonic precursors, dominate the pro-tumoral response (29, 30). Conversely, we show here that monocyte-derived macrophages are protective against tumor growth in Pten-null HGSC. This corroborates previous data in which anti-CSF1R treatment following carboplatin was shown to shorten survival via inhibition of the adaptive immune response (42), and also the demonstration that stromal macrophage infiltration (43) and a high intratumoral HLA-DR:CD163 ratio correlate with improved survival (44). Collectively, this indicates strongly that macrophage therapeutic approaches need to be subtype-specific.

Trans-celomic spread is the main mechanism by which HGSC disseminates around the peritoneal cavity and macrophages seem critical for this spread: gene expression in omental resident macrophages changes within hours of tumor-cell injection in mice, whereas macrophages promote seeding of ID8 cells on the omentum, and macrophage depletion prior to tumor implantation prevents tumor seeding (45). This early seeding is independent of T, B, and NK cells, and occurs equally well in immunodeficient models (46). Omental fat-associated lymphoid clusters are macrophage-rich, and resident embryonic-derived TIM4⁺CD163⁺ macrophages promote ID8 seeding and spread (29). Additionally, omental-independent mesenteric-derived resident macrophages also support dissemination (30).

PTEN loss and PIK3CA copy number alterations occur in HGSC carcinogenesis (47), whereas Pten deletion is essential for metastatic spread from the fallopian tube in transgenic murine models (48) and also accelerates intraperitoneal tumor growth (15). We show here that Pten deletion does not enhance proliferation or survival in low-attachment conditions per se. However, Pten-null cells specifically induce the expansion of resident macrophages in the peritoneal fluid and omentum. Peritoneal resident macrophages support tumor spheroid formation and spread (49) and targeting them can reduce tumor burden (50). We show that Pten-null tumors recruit peritoneal resident macrophages directly into the omentum and that this is independent of blood monocyte recruitment, as Ccr2^RFP/RFP and Ccr2^+/+ mice have equivalent omental resident macrophage numbers. This is important for considering targeting approaches, as recruitment does not occur directly from the blood. Furthermore, we show that Pten deletion accelerates formation of a unique resident macrophage population that expresses high levels of the heme-degrading enzyme, HMOX1. We find that Pten-null cells significantly upregulate known activators of the HMOX1^hi macrophage gene program, including the cytokine IL33, which is critical in inducing HMOX1 expression in red pulp macrophages (51) and drives immunosuppressive macrophage accumulation in glioblastoma (52).

HMOX1 expression is normally restricted to splenic and hepatic macrophages that remove senescent red blood cells, in which its induction is cytoprotective against the oxidative stress induced by heme accumulation. However, aberrant expression in tumor-associated macrophages drives immunosuppression (53), and HMOX1 inhibition by SnMP improves T-cell infiltration and activity when administered with chemotherapy (53). HMOX1 also induces expression of pro-inflammatory and angiogenic genes (54), whereas HMOX1^hi macrophages can also directly drive metastasis, but not primary tumor growth, partly by aiding transendothelial migration and angiogenesis (55, 56).

Crucially, we found that HMOX1 inhibition extended the survival of Trp53^−/−;Pten^−/− ID8 tumor-bearing mice. SnMP is not directly cytotoxic to either tumor cells or macrophages (53) and thus SnMP antitumoral activity is likely to be driven only via altered macrophage function. We found high HMOX1 expression in LYVE1⁺ macrophages (33), previously shown to be mesenteric-membrane resident macrophages that can also promote ovarian cancer spread (30). We did not detect this population in our scRNA-seq, most likely due to the small number of cells analyzed. However, SnMP treatment did ablate LYVE1⁺ macrophages while also driving an apparent increase in cluster 2 macrophages. This reduction in LYVE1⁺ macrophages could result from their location in the perivascular niche and consequent susceptibility to heme-induced cytotoxicity (36). This suggests that the more abundant cluster 2 macrophages may upregulate HMOX1 in part by heme, but also by other microenvironmental factors, such as, but not limited to, IL33. Future work will be required to determine whether the LYVE1⁺ population contributes in any way to the therapeutic effect here.

There are limitations to our study, not least that most of our findings are derived from the ID8 murine model of HGSC, which is of ovarian surface epithelium origin. However, ID8, as well as other OSE-derived models, such as STOSE (57), can recapitulate the dominant features of HGSC, namely, peritoneal dissemination and omental metastasis. However, it has been shown that different murine models represent the HGSC tumor microenvironment differently, as recently demonstrated (19). For this reason, we attempted to replicate our findings using the fallopian-derived HGS2 line (16). However, we could not stably restore wild-type Pten in HGS2 cells using CRISPR/Cas9 because of the Brca2 deletion and consequent defective homology-directed repair (16). Nevertheless, restoring Pten via lentivirus transduction in HGS2 cells reduced omental tumor growth. This was not consistent across all clones tested, potentially due to promoter silencing, which is known to occur in lentivirus-derived genes (58). Most importantly and reassuringly, two independent HGSC datasets validated our findings in mice: both the scRNA-seq and IHC results reinforce the finding that the presence of HMOX1^hi macrophages is associated with poor outcome and activated PI3K signaling in HGSC.

Correlating murine and human data is extremely challenging. Here we used Pten deletion to activate PI3K signaling in ID8 cells, whereas in HGSC, the pathway can be activated through multiple additional mechanisms, including PIK3CA and AKT mutation and amplification. We used pAKT staining on IHC as a surrogate for pathway activation in patient samples, but it remains unclear whether every mechanism for activating PI3K signaling will generate HMOX1^hi macrophages. Similarly, the number of scRNA-seq datasets available to interrogate tumor-specific PI3K signaling remains small, and further data will be necessary to elucidate more nuanced biomarkers of pathway activity.

In summary, we have shown that HMOX1^hi macrophages, with common gene expression programs including immunosuppression, hypoxia, cholesterol efflux, and lipid transport, can be identified in both murine and human HGSC. The function of HMOX1^hi macrophages in HGSC remains to be understood fully and the gene expression pathways may reflect both the PI3K-driven microenvironment that induces HMOX1^hi cells and overall macrophage function. Nonetheless, our study highlights that HMOX1 inhibition may provide a relevant treatment strategy for HGSC.

Supplementary Material

Supplementary Methods

Description of supplementary methods

Supplementary Figure legends

Legends to supplementary figures

Supplementary Figures

8 supplementary figures

Table S1 Antibodies

Table of antibodies used in flow cytometry and immunohistochemistry.

Table S2 Genes per cluster

Genes upregulated in each mouse macrophage cluster

Table S3 HMOX1-hi gene expression

Up- and down-regulated differentially expressed genes in HMOX1-hi human macrophages

Authors’ Disclosures

S. Spear reports receiving salary and research consumable support through a grant from the charity Ovarian Cancer Action, reference PSF687. O. Le Saux reports grants from AstraZeneca and Bristol Myers Squibb outside the submitted work. J.D. Brenton reports grants from Cancer Research UK during the conduct of the study, as well as personal fees and other support from Tailor Bio Ltd. and personal fees from GSK, GE Healthcare, and Clovis Oncology outside the submitted work. B.C. Vanderhyden reports grants from Ovarian Cancer Canada outside the submitted work. I.A. McNeish reports grants from Ovarian Cancer Action and Cancer Research UK, grants and personal fees from AstraZeneca, and personal fees from GSK, pharma&, Clovis Oncology, Roche, BioNTech, OncoC4, and Duke Street Bio outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

S. Spear: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft. O. Le Saux: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft. H.B. Mirza: Data curation, formal analysis, visualization. N. Iyer: Investigation. K. Tyson: Investigation. F. Grundland Freile: Investigation. J.B. Walton: Data curation, investigation. C. Woodman: Investigation. S. Jarvis: Formal analysis. D.P. Ennis: Investigation. C. Aguirre Hernandez: Investigation. Y. Xu: Investigation. P. Spiliopoulou: Investigation. J.D. Brenton: Supervision, funding acquisition, methodology. A.P. Costa-Pereira: Supervision, funding acquisition. D.P. Cook: Data curation, investigation. B.C. Vanderhyden: Supervision, funding acquisition, investigation. H.C. Keun: Supervision, funding acquisition. E. Triantafyllou: Supervision, investigation. J.N. Arnold: Supervision, funding acquisition. I.A. McNeish: Conceptualization, resources, formal analysis, supervision, funding acquisition, visualization, writing–original draft, project administration.