Heme oxygenase-1 orchestrates the immunosuppressive program of tumor-associated macrophages

Tumor-associated macrophages (TAMs) contribute to the maintenance of a strong immunosuppressive environment, supporting tumor progression and resistance to treatment. To date, the mechanisms that drive acquisition of these immunosuppressive features are still poorly defined. Heme oxygenase-1 (HO-1) is the rate-limiting enzyme that catabolizes free heme. It displays important cytoprotective, antiinflammatory, and antioxidant properties. A growing body of evidence suggests that HO-1 may also promote tumor development. Herein, we show that HO-1 is highly expressed in monocytic cells in the tumor microenvironment (TME) once they differentiate into TAMs. Deletion of HO-1 in the myeloid compartment enhances the beneficial effects of a therapeutic antitumor vaccine by restoring CD8+ T cell proliferation and cytotoxicity. We further show that induction of HO-1 plays a major role in monocyte education by tumor cells by modulating their transcriptional and epigenetic programs. These results identify HO-1 as a valuable therapeutic target to reprogram the TME and synergize with current cancer therapies to facilitate antitumor response. Research Article Immunology Oncology


Introduction
Immunotherapy represents a paradigm shift in the treatment of cancer. In the past few years, approaches such as immune checkpoint inhibition or adoptive transfer of engineered T cells have produced durable responses and long-term survival of many patients for whom previous therapeutic options were ineffective (1)(2)(3)(4)(5)(6)(7)(8)(9). These clinical successes demonstrate the essential role of the cancer-immunity interface in tumor progression and therapy. However, only a subset of patients responds to immunotherapies, and some of them acquire resistance to the treatment. The cellular and molecular determinants of responsiveness versus resistance to immunotherapy are incompletely understood (10,11). This may be partly due to the current focus of therapies on the T cell compartment only, with little attention to the side of the antigen-presenting cell. It is likely that a better understanding of how tumors shape their microenvironment, and alter myeloid cell functions, would reveal novel principles of cancer immunotherapy and enable innovative clinical applications, benefiting patients refractory to current immunotherapies (12,13).
In this context, tumor-associated macrophages (TAMs) generally play a deleterious role by supporting multiple aspects of tumor progression (14). They were shown to contribute to the resistance of tumors to biologic therapies, chemotherapies, and radiotherapies through direct trophic support to cancer cells. For example, TAMs may promote angiogenesis, the maintenance of cancer stem cells, and metastatic processes (15). Macrophage infiltration may also interfere with immunotherapy, hampering efforts to reactivate CD8 + T cells by targeting immune checkpoints. Indeed, macrophages express inhibitory receptors (such as programed cell death ligand 1 [PD-L1] or B7-H4) and represent an important source of cytokines (such as IL-10 or TGF-β), chemokines, and enzymes (such as arginase 1 [Arg-1], inducible nitric oxide synthase [iNOS], or indoleamine 2,3-dioxygenase) that contribute to the maintenance of a strong immunosuppressive environment by inhibiting effector T cells while inducing regulatory T cells' recruitment and expansion (16,17).
Tumor-associated macrophages (TAMs) contribute to the maintenance of a strong immunosuppressive environment, supporting tumor progression and resistance to treatment. To date, the mechanisms that drive acquisition of these immunosuppressive features are still poorly defined. Heme oxygenase-1 (HO-1) is the rate-limiting enzyme that catabolizes free heme. It displays important cytoprotective, antiinflammatory, and antioxidant properties. A growing body of evidence suggests that HO-1 may also promote tumor development. Herein, we show that HO-1 is highly expressed in monocytic cells in the tumor microenvironment (TME) once they differentiate into TAMs. Deletion of HO-1 in the myeloid compartment enhances the beneficial effects of a therapeutic antitumor vaccine by restoring CD8 + T cell proliferation and cytotoxicity. We further show that induction of HO-1 plays a major role in monocyte education by tumor cells by modulating their transcriptional and epigenetic programs. These results identify HO-1 as a valuable therapeutic target to reprogram the TME and synergize with current cancer therapies to facilitate antitumor response.
Heme oxygenase-1 (HO-1) is the rate-limiting enzyme that catalyzes free heme into 3 major biologically active by-products: carbon monoxide, ferrous iron, and biliverdin (converted to bilirubin). In numerous pathological contexts, HO-1 displays important cytoprotective, antiinflammatory, antioxidant, and antiapoptotic properties (18)(19)(20)(21)(22)(23)(24)(25)(26). In the context of alloreactivity, we previously demonstrated that HO-1 contributes to the immunosuppressive properties of myeloid cells (27). A growing body of evidence suggests that HO-1 also promotes tumor development. It is expressed in a wide variety of cancers and is generally associated with poor prognosis (28)(29)(30). In preclinical models, the administration of HO-1 pharmacologic inhibitors displays antitumor effects (31,32) and improves the response to chemotherapy (33). This effect is at least partially mediated by CD8 + T cells, but the underlying mechanisms of immunomodulation by HO-1 remain unclear (34). Furthermore, expression of HO-1 by TAMs was recently shown to promote transendothelial migration and metastatic spread (35).
Herein, we investigated the role of HO-1 in TAMs. We show that deletion of HO-1 in the myeloid compartment enhances the beneficial effects of a therapeutic antitumor vaccine by restoring T cell proliferation and cytotoxicity in the tumor microenvironment (TME). We further show that induction of HO-1 plays a major role in monocyte education by tumor cells by modulating their transcriptional and epigenetic programs. Taken together, these results identify HO-1 as a valuable target to reprogram TAMs and improve current strategies of immunotherapy.

Results
Monocytic cells express HO-1 upon differentiation into macrophages in the TME. In order to evaluate the expression of HO-1 in the myeloid compartment during tumor development, we implanted thymoma cells (EG7-OVA) intradermally in C57BL/6 mice because this model was shown to strongly promote the expansion of myeloid suppressive cells (36). We first assessed HO-1 expression in the TME by immunofluorescence staining ( Figure 1A and Supplemental Figure 1A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.133929DS1). We observed HO-1 staining in a portion of CD11b + and F4/80 + infiltrating cells. Of note, HO-1 was found in the cytoplasm and/or the nucleus of these cells. This is consistent with previous reports indicating that HO-1 can migrate to the nucleus and modulate transcriptional activity independently of its enzymatic activity (37)(38)(39). To further define the cellular sources of HO-1, we evaluated its expression by flow cytometry. Ly6C hi monocytes that are recruited in the TME gradually differentiate into Ly6C lo MHCII + TAMs (40). The proportion of Ly6G hi granulocytes or LyC6 hi monocytes (MHCIIor MHCII + ) expressing HO-1 was low ( Figure 1B). In sharp contrast, it was expressed by a significant proportion of Ly6C lo MHCII + TAMs, suggesting that it is part of the program induced in monocytic cells upon exposure to tumor-derived factors. Expression of HO-1 by TAMs was associated with strong expression of classical phenotypic markers of macrophage differentiation, such as F4/80, CD64, CD206, and CD163 ( Figure 1C). As compared with HO-1 -TAMs, HO-1 + counterparts had decreased MHCII expression. Tumors also influenced myelopoiesis and induced the accumulation of immature CD11b + Gr1 + myeloid cells in the bone marrow and the spleen ( Figure 1D). Expression of HO-1 in immature CD11b + Gr1 + myeloid cells from naive and tumor-bearing mice was comparable ( Figure  1E). These results indicate that HO-1 expression is specifically induced upon differentiation of monocytic cells in the TME. It was previously shown that STAT3-inducing cytokines, such as IL-6 and IL-10, are potent inducers of HO-1 in macrophages (41). We therefore evaluated the role of these cytokines in our experimental setting using neutralizing antibodies. As shown in Supplemental Figure 1B, HO-1 expression by TAMs was found to be independent of IL-6 and IL-10 signaling, indicating that other tumor-intrinsic factors are responsible for HO-1 induction by the TME. To define the pattern of expression of HO-1 in the context of human cancer, we analyzed single-cell RNA-Seq data from tumor-infiltrating CD45 + cells collected from patients with breast cancer (42). We observed clear separation of myeloid (characterized by LYZ and HLA-DR expression) and lymphoid populations ( Figure 2). HMOX1 expression was largely restricted to myeloid cells ( Figure 2). Next, we identified genes that were upregulated in HMOX1 + cells. Many of these genes, such as APOE1, FTH, or FTL1, are characteristic of macrophages or serve as prognostic markers in TAMs (42)(43)(44). Gene Ontology analysis revealed several relevant pathways, including the following 3: negative regulation of immune system process, inflammatory response to wounding, and macrophage activation. Trajectory analysis of monocytic clusters identified major components associated with activation and differentiation of macrophages and with monocytes once they enter the TME (42). We identified the cells that exhibited the greatest enrichment for TAMs or monocyte activation signatures and evaluated their expression of HMOX1 (Supplemental Figure 2). These data indicate that HO-1 expression is a feature of monocytes and macrophages once activated in the TME.
Myeloid HO-1 promotes tumor growth by an immunosuppressive mechanism. To determine the impact of HO-1 expression by TAMs on tumor growth, we invalidated Hmox1 in myeloid cells (LysMCre +/wt Hmox1 fl/fl Hmox1 ΔM mice). We evaluated the growth of intradermally implanted EG7-OVA tumors at Representative flow cytometry plots of the accumulation of immature myeloid cells compatible with myeloid-derived suppressor cell phenotype (CD11b hi Ly6C + Ly6Gand CD11b hi Ly6C int Ly6G + summarized as CD11b + Gr1 + cells) in the bone marrow (BM) and spleen from tumor-bearing WT mice. Data representative of 3 independent experiments. Each point represents an individual mouse. Horizontal bars indicate median ± interquartile range (E) HO-1 expression measured by flow cytometry among CD11b + Gr1 + cells from bone marrow, spleen, and EG7-OVA tumor from tumor-bearing WT mice, compared with tumor-free WT mice (naive). Horizontal bars indicate median ± interquartile; n = 3 (naive), and n = 6 (tumor-bearing group). Statistical analysis was performed with Mann-Whitney U test. ***P < 0.001; ****P < 0.0001. regular intervals. Tumor growth in Hmox1 ΔM mice was comparable to Hmox1 fl/fl littermates ( Figure 3A). Next, we repeated these experiments and induced an antitumor T cell response by immunizing the mice against OVA at day 7 and day 14 postimplantation. As an adjuvant, we used poly(I:C) because this dsR-NA analog was shown to promote CD8 + T cell responses in preclinical tumor immunotherapy settings (45). As expected, using this regimen, we observed a delay in tumor growth in Hmox1 fl/fl mice. However, in most cases, mice had to be sacrificed at later time points because of tumor escape. In contrast, we observed complete tumor regression upon therapeutic immunization in a majority of Hmox1 ΔM mice. This antitumor effect was abrogated upon depletion of CD8 + T cells by antibody treatment ( Figure 3B). The effect of HO-1 invalidation in the myeloid compartment on tumor growth was also observed in the absence of immunization, when mice were treated with cyclophosphamide (Supplemental Figure  3), suggesting that HO-1 inhibition could favor antitumor responses in the context of chemotherapy-induced immunogenic cell death. To further evaluate the antigen-specific nature of this enhanced antitumor response, we implanted EG7-OVA tumor cells on one flank and parental EL4 cells on the other flank of the same animal. Upon immunization and transfer of OVA-specific CD8 + T cells (OT-1), progression of EG7-OVA tumors was reduced in Hmox1 ΔM as compared with Hmox1 fl/fl mice ( Figure  3C). However, in the same animals, growth of EL4 tumors was not restrained in the Hmox1 ΔM group. Taken together, this set of experiments shows that myeloid-specific inactivation of HO-1 potentiates antigen-specific antitumor CD8 + T cell responses in the context of therapeutic immunization.
Myeloid HO-1 controls antitumor T cell proliferation and cytotoxicity in the TME. To further assess the effect of myeloid HO-1 on antigen-specific T cell responses, we performed adoptive transfer of CFSE-labeled OT-1 cells 10 days after EG7-OVA implantation. Mice were immunized concomitantly. Two days after, we assessed OT-1 cells' frequency. Although the proportions of OT-1 cells in the spleen or the draining lymph nodes were comparable in both groups, they were strongly increased within the tumors of Hmox1 ΔM mice as compared with Hmox1 fl/fl controls ( Figure 4A). This was accompanied by high proliferation rate, as assessed by CFSE dilution ( Figure 4B) and Ki-67 staining ( Figure 4C). Importantly, the proliferation rate of OT-1 cells in the draining lymph nodes and in the spleen was comparable in both experimental groups (Supplemental Figure 4). To evaluate the functionality of adoptively transferred T cells, we analyzed granzyme B and IFN-γ expression following ex vivo stimulation of tumor-infiltrating lymphocytes with OVA SIINFEKL peptide. We observed a modest but statistically significant increased expression of these cytotoxic mediators in OT-1 cells transferred in tumor-bearing Hmox1 ΔM mice ( Figure 4, D and E). This was accompanied by higher T-bet levels ( Figure 4F). In contrast, expression of Eomes was similar in both groups ( Figure 4G). Taken together, these EG7-OVA tumor cells were inoculated intradermally at day 0 on the right flank of Hmox1 ΔM mice (n = 11). Their tumor volumes were compared with Hmox1 fl/fl littermates (n = 10) at regular intervals following implantation. (A) There was no significant difference between the groups of tumors. However, a blockade of tumor growth was observed in Hmox1 ΔM mice (n = 11) compared with Hmox1 fl/fl littermates (n = 8) after therapeutic immunization with subcutaneous injection of ovalbumin protein (10 μg/mouse) and poly(I:C) (50 μg/mouse) 7 days after tumor inoculation and boost 7 days later on the right flank of the animals. (B) Intraperitoneal administration of isotype control or CD8 + T cell-depleting monoclonal antibody (clone YTS169) 1 time/wk (500 μg/mouse). (C) Bilateral tumor model, where EG7-OVA tumor cells were inoculated on the right flank and EL4 cells on the left flank from Hmox1 ΔM mice (n = 8) and Hmox1 fl/fl mice (n = 10) that were therapeutically immunized as described above and adoptively transferred with OT-1 cells (10 6 cells/mouse) at day 10. Data are representative of 3 independent experiments. Statistical analysis was performed with Mann-Whitney U test (A and B) and Wilcoxon matched-pairs signed rank test (C). *P < 0.05; **P < 0.01; ***P < 0.001. data indicate that HO-1 expression by TAMs leads to strong immunosuppressive activity in the TME that limits antigen-specific CD8 + T cell effector function against tumor cells.
HO-1 drives transcriptional and epigenomic programs of TAMs. In order to define the role of HO-1 in myeloid cells, we examined the proportions of Ly6G hi granulocytes, Ly6C hi monocytes, and Ly6C lo M-HCII + TAMs in tumors of Hmox1 ΔM and Hmox1 fl/fl mice 12 days after implantation ( Figure 5A). An intravenous adoptive transfer of OT-1 cells (2 × 10 6 cells/mouse) was performed 10 days after tumor inoculation. This was followed by an immunization with subcutaneous injection of ovalbumin protein (50 μg/mouse) and poly(I:C) (50 μg/mouse) 1 hour later on the right flank of the animals. Two days later, EG7-OVA tumors were enzymatically and mechanistically digested and analyzed by flow cytometry. (A) Proportions of OT-1 cells in the spleen (n = 11 and n = 11), the draining lymph nodes (dLN) (n = 11 and n = 11) (axillary and inguinal lymph nodes on the right side), and the tumor (n = 24 and n = 23) of Hmox1 ΔM mice compared with Hmox1 fl/fl littermates. OT-1 cells were labeled with CFSE before intravenous adoptive transfer (2 × 10 6 cells/mouse). This was followed by an immunization of the mice as described above. (B) Tumor-infiltrating OT-1 cell proliferation assessed by CFSE dilution (n = 9 and n = 9) and (C) Ki-67 expression (n = 5 and n = 7) among OT-1 cells. (D) Granzyme B (GzmB) (n = 9 and n = 13) was analyzed by intracytoplasmic staining in tumor-infiltrating OT-1 cells. (E) Production of IFN-γ (n = 12 and n = 12) and MFI of (F) T-bet (n = 5 and n = 7) and (G) Eomes (n = 6 and n = 5) were assessed by ex vivo stimulation overnight with OVA SIINFEKL peptide (and brefeldin A added 2 hours later). Data are pooled from 3/4 experiments. Horizontal bars indicate median ± interquartile range. Statistical analysis was performed with Mann-Whitney U test. *P < 0.05; **P < 0.01; ***P < 0.001.
There were no significant changes in proportions between these 2 groups. However, we observed a larger proportion of higher MHCII hi cells in TAMs from Hmox1 ΔM mice. In addition, myeloid HO-1 deletion led to increased iNOS and decreased Arg-1 expression in TAMs ( Figure 5B). Taken together, this suggests that HO-1 could influence their differentiation status, leading to the restoration of tumor-infiltrating T cell proliferation in Hmox1 ΔM mice. Next, we performed an RNA-Seq experiment on CD11b hi CD64 + Ly6C lo MHCII + TAMs from Hmox1 ΔM and Hmox1 fl/fl mice. We showed Hmox1 invalidation through the lack of a full-length RNA of the Hmox1 gene upon LysM-driven expression of Cre recombinase in these cells ( Figure 6A) and identified more than 1000 differentially expressed genes (594 down-and 539 upregulated genes in HO-1-deficient cells as compared with their controls; FDR < 0.05, and fold change > 2) ( Figure 6B). We performed gene set enrichment analysis to evaluate the expression of genes that were shown to be upregulated in TAMs as compared with splenic monocytic cells ( Figure 6C). We observed a global decrease in the expression of these genes in HO-1-deficient cells, indicating that the core molecular signature the TME induces is dysregulated in the absence of HO-1. The profile observed in HO-1-deficient TAMs did not follow a simple M1/M2 dichotomy because both M1 and M2 signatures were found to be significantly affected. Importantly, multiple genes encoding molecules that participate in the immunosuppressive features of TAMs, such as Arg-1, iNOS, IL-10, PD-L1, and PD-L2, were downregulated in HO-1-deficient cells ( Figure 6D). Expression of IL-4-dependent genes was also globally decreased in this group. Furthermore, classical proinflammatory M1 genes, such as Il27 and Il12b, were also affected ( Figure 6D).
To further determine underlying molecular processes at play, we analyzed epigenomic landscapes of these cells by assay for transposase accessible chromatin sequencing (ATAC-Seq) approaches. This technique allows us to map open chromatin regions throughout the genome (46). We observed extensive modifications in HO-1-deficient cells from Hmox1 ΔM mice. As shown in Figure 7A, 1518 and 4284 regions were found to be significantly more or less accessible in controls, respectively. Most of the differentially accessible peaks were located in enhancers rather than in promoters. We used the Binding and Expression Target Analysis (BETA) package (47) to predict the activating or repressive function of these differentially accessible regions. Regulatory regions that were more/less accessible were clearly associated with genes that were up-or downregulated in WT or HO-1-deficient TAMs, respectively ( Figure 7B). This observation indicates that the impact of HO-1 on their transcriptional profile has a strong epigenetic component. For example, we observed decreased accessibility in regulatory elements associated with the genes that encode the immune checkpoint molecules PD-L1, PD-L2, and MERTK ( Figure 7C) (48), in line with their reduced expression in the absence of HO-1. Similarly, we identified regions that were less accessible within the locus of Mmp2, which together with other matrix metalloproteinases favors neovascularization and tumor dissemination (49). Next, we performed Gene Ontology analysis using Genomic Regions Enrichment of Annotations Tool (GREAT) (50). The most relevant pathways were associated with regions that were less accessible in HO-1-deficient cells ( Figure 8A). As expected, many of these were involved in the regulation of inflammatory response or cytokine signaling. In line with the cytoprotective functions of HO-1 (51), we also observed signatures for wound healing and cell redox homeostasis pathways. Importantly, several metabolic processes were also identified along with other important tumor-related pathways, such as signaling by EGFR, VEGF, and TGF-βR ( Figure  8A). These observations suggest important and widespread functional impact of HO-1 on the epigenetic programming of TAMs. We then scanned for binding motifs at the center of ATAC peaks located in these differentially accessible regions using CiiiDER algorithm (52). We observed strong enrichment for consensus binding motifs characteristic of basic region/leucine zipper (bZIP) and zinc finger families of transcription factors in WT and HO-1-deficient cells, respectively ( Figure 8B). Among these bZIP factors, we noted motifs for C/EBPs, which are key factors involved in myeloid cell differentiation, and for Fos-and Jun-related factors. Of note, the consensus antioxidant response elements bound by nuclear factor erythroid 2-related factor 2 (NRF2) were significantly enriched in WT cells. This is of particular interest because it represents the main transcriptional pathway responsible for induction of Hmox1 in response to oxidative stress (53). Conversely, nuclear HO-1 was shown to interact with NRF2 and to promote its transcriptional activity (37). Taken together, these data indicate that HO-1 supports a major transcriptional and epigenetic reprogramming of monocytic cells once they enter the TME.

Discussion
Tissues' macrophages are highly heterogeneous and plastic and acquire specific functions in response to their environmental cues. In the TME, they integrate multiple signals that reshape their enhancer landscape and as a consequence their transcriptional and functional programs (16). Here we show that HO-1 is induced in monocytic cells that infiltrate the tumor bed upon differentiation into TAMs. Several signals could contribute to this observation. Cytokines such as IL-6 or IL-10, produced by cancer-associated fibroblasts (54), endothelial cells (55), or TAMs themselves (40,56,57), were shown to be potent inducers of HO-1 as part of a "wound healing" signature (35). However, in our tumor model, HO-1 expression by TAMs was independent of these signaling pathways. Hence, other signals, such as tissue hypoxia or accumulation of lactic acid, key metabolic features of the TME (58, 59), could induce HO-1 through the activation of HIF1α (60). Intratumor hemorrhage, commonly encountered in cancer, might also lead to extravasation of hemoglobin (61), the physiologic inducer of HO-1.
We demonstrate that myeloid-restricted HO-1 ablation strongly improves the response toward therapeutic immunization by enhancing antitumor CD8 + T cell proliferation and cytotoxicity. In line with this observation, enzymatic inhibition of HO-1 by metalloporphyrins was previously shown to promote tumor regression or complement conventional cancer therapies and to facilitate cytotoxic antitumor immune response (31)(32)(33)(34)62). Several lines of evidence in different pathologic conditions suggest that HO-1 induction in macrophages plays a critical role in controlling the adaptive immune response by influencing their polarization (63)(64)(65)(66)(67)(68)(69). We observed that HO-1 ablation in TAMs decreased the expression of the M2-associated enzyme arginase 1 while enhancing the expression of iNOS, a classical proinflammatory marker. However, our data indicate that HO-1 ablation had a global impact on the transcriptional and epigenetic profiles of TAMs that is not limited to the classical M1/M2 polarization profile, because we observed decreased expression of both immunosuppressive and proinflammatory markers. This apparent discrepancy could be related to the heterogeneity of TAMs and should be resolved using single-cell-based approaches. Furthermore, it will be important to study these features in different experimental settings that modulate the activation status of TAMs, such as injection of TLR agonists or adoptive cell therapy (70). We suggest that the action of HO-1 on TAMs' differentiation program could be independent of its enzymatic activity because multiple pieces of evidence support a role for HO-1 as a transcriptional modulator. After exposure to hypoxia, HO-1 translocates to the nucleus in a cleaved and enzymatically inactive form, where it directly interacts with transcription factors, such as Nrf2 or JunD, and modulates their activity (37)(38)(39). We propose that similar processes could be at play in TAMs and account for their acquisition of an immunosuppressive program. Further understanding of the underlying mechanisms will be important to develop adequate pharmacologic approaches. Collectively, our data indicate that HO-1 expression in tumor-infiltrating monocytic cells represents a molecular switch that promotes their immunosuppressive functions. It could therefore represent a valuable target to reprogram the TME and potentially synergize with the current therapeutic approaches focused on the T cell compartment.

Methods
Mice. C57BL/6 mice were purchased from Envigo. Hmox1 ΔM mice were generated at the Institute for Medical Immunology by crossing Hmox1 fl/fl mice (in which the Hmox1 allele was flanked by loxP sites) with LysMCre +/wt mice, both of which had been backcrossed onto the C57BL/6 background for more than 10 generations (Instituto Gulbenkian de Ciência, Oeiras, Portugal). The LysMCre transgene causes a specific deletion of the Hmox1 gene in myeloid cells. Hmox1 fl/fl mice (Hmox1 fl/fl LysMCre wt/wt ) were used as controls for Hmox1 ΔM mice and were littermates. MHC class I-restricted, OVA-specific, TCR-transgenic OT-1 mice with a Rag1 -/background were obtained from The Jackson Laboratory. All experimental and control mice were 8-to 12-week-old animals and were of the same sex for each experiment. Tumor cell line. The EL4 lymphoma cell line was obtained from American Type Culture Collection (ATCC, TIB-39). The EG7-OVA tumor cell line was derived from the tumor cell line EL4 (ATCC CRL-2113) by transfection with a plasmid carrying the chicken ovalbumin (OVA) and neomycin phosphotransferase -G418 resistance -genes. The cells were maintained at 37°C and 5% CO 2 in RPMI 1640 medium (Lonza) supplemented with 10% fetal calf serum (FCS), 0.1 mM of nonessential amino acids, 100 U/mL of penicillin, and 100 U/mL of streptomycin (all reagents from Lonza, hereafter referred to as complete medium). The EG7-OVA tumor cells were cultured in a complete medium supplemented with 1 mg/mL of G418 sulfate (Geneticin Selective Antibiotic, Thermo Fisher Scientific) once a week. OVA peptide expression on MHCI molecules of EG7-OVA tumor cells was regularly verified by flow cytometry.
Tumor inoculation, therapeutic immunizations, and cytokine neutralization. Tumors were initiated by intradermal injection of 2.5 × 10 5 tumor cells (in 100 μL of sterile PBS) per mouse into the right flank at day 0. When indicated, at day 7, 10, or 14 after EG7-OVA tumor inoculation, mice were injected subcutaneously with 10 or 50 μg of OVA protein (grade VI, A2512, MilliporeSigma) associated with 50 μg of poly(I:C) (HMW, VacciGrade, InvivoGen) in 100 μL of sterile PBS per mouse on the right flank. Before the immunization, mice were anesthetized intraperitoneally with weight-adjusted amounts of ketamine (1%, 100 μg/g, Nimatek) and xylazine (2%, 10 μg/g, Bayer) solution. In other experiments, the mice were injected i.p. with 0.5 mg of depleting monoclonal antibodies in 200 μL of sterile PBS per mouse for CD8α + T cell depletion (YTS169, University of Cambridge), 1 day before tumor inoculation and then once a week. The control mice were injected with IgG2b isotype control antibodies (Bio X Cell, clone LTF-2). Peripheral blood samples were collected once a week after depletion and analyzed by flow cytometry to confirm the depletion. For cyclophosphamide experiments, mice were injected i.p. with cyclophosphamide monohydrate (C0768-1G, MilliporeSigma), with a dose of 1.5 mg/mouse in 200 μL of PBS at day 10 after tumor inoculation. For cytokine neutralizations, mice were injected every other day, starting at day 1 after tumor inoculation, with 200 ng of anti-IL-6 (clone MP5-20F3, Bio X Cell), anti-IL-10R (clone 1B1.3A, Bio X Cell), or rat IgG1 isotype control (HRPN, Bio X Cell). Mice received 6 doses by i.p. injection.
Tumor monitoring. Mice were monitored every other day for tumor growth by using fine calipers. Tumor volume (mm 3 ) is described as (A × B 2 )/2, where A and B represent tumor length and width, respectively. Mice were sacrificed for tumor analysis when specified or when the total volume of the tumor reached 3000 mm 3 .
Tissue digestion protocol. Tumors were dissected, finely chopped, and perfused with a digestion solution containing DNAse I 1 mg/mL (Grade II, MilliporeSigma, 10104159001) 20 μL and a mix of collagenase I and II 2.5 mg/mL (Liberase TL Research Grade, Roche) 20 μL in 5 mL of RPMI 1640 (Lonza) each and were incubated 30 minutes at 37°C. After 5 mL of RPMI FCS 5% with 2 mM ethylenediaminetetraacetic acid (EDTA, MilliporeSigma) were added to each sample, tumor pieces were mashed and filtered, twice. Cell suspensions from spleens were obtained by homogenizing individual spleens to release splenocytes in 5 mL of RPMI 1640 with 10% FCS each. The red blood cells were lysed briefly in 800 μL of ACK lysis buffer. Lymph node cell suspensions were prepared by dissecting inguinal, deep, and superficial axillary lymph nodes on the right flank and grinding the tissue in RPMI 1640 with 10% FCS. To obtain bone marrow cell suspensions, femur and tibia from the mice were dissected, rinsed in ethanol, and transferred to RPMI 1640 with 10% FCS. Bone ends were cut with sterile sharp scissors, and the contents of the bone marrow were flushed with the medium. Bone marrow cells were diluted by vigorous pipetting. All the cell suspensions from individual organs were filtered through a 40-μm cell strainer, centrifuged at 300 g for 10 minutes, and resuspended in RPMI 1640 with 10% FCS.
OT-1 T cell isolation and adoptive transfer. Cell suspensions from the lymph nodes of the MHC class Irestricted, OVA-specific, TCR-transgenic OT-1 mice (OT-1 cells) (8 weeks to 4 months old) were harvested (see tissue digestion protocol) and adoptively transferred by tail vein injection (2 × 10 6 cells per mouse) on the specified day after tumor inoculation. When specified, OT-1 cells were labeled before intravenous injection with 2 μM CFSE (CellTrace CFSE Cell Proliferation Kit, Invitrogen, Thermo Fisher Scientific) by incubating them for 20 minutes at 37°C according to the manufacturer's instructions. This was followed by an immunization with OVA protein (50 μg) and poly(I:C) (50 μg) 1 hour later. Cell division accompanied by CFSE dilution was analyzed by flow cytometry 2 days later (detected in the FITC channel).
TAM cell sorting. CD11b + cells were first purified from tumor cell suspensions using positive magnetic selection with a cell isolation kit (Miltenyi Biotec). LIVE/DEAD -CD11b + Ly6G -Ly6C -MHCII + CD64 + TAMs were sorted using a BD FACSAria III cell sorter (100,000 cells from Hmox1 ΔM mice and from WT mice, in triplicates) after surface staining with monoclonal antibody mix (see the section Flow cytometry).
RNA-Seq. TAMs were isolated by FACS in RLT buffer and flash-frozen. RNA extraction was performed using RNeasy Mini Kit (QIAGEN), and sample quality was tested on a 2100 Bioanalyzer (Agilent). Libraries were prepared using Ovation SoLo RNA-Seq System (NuGEN Technologies) and underwent paired-end sequencing (25 × 10 6 paired-end reads/sample, NovaSeq 6000 platform) performed by BRIGHTcore ULB-VUB, Belgium (http://www.brightcore.be). Adapters were removed with Trimmomatic-0.36 (with the following parameters: Truseq3-PE.fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MIN-LEN:36 HEADCROP:4) Reads were then mapped to the reference genome mm10 by using STAR_2.5.3 software with default parameters. We then sorted the reads from the alignment according to chromosome positions and indexed the resulting BAM files. Read counts in the alignment BAM files that overlap with the gene features were obtained using HTSeq-0.9.1 with "--nonunique all" option (if the read pair aligns to more than one location in the reference genome, it is counted in all features to which it was assigned and scored multiple times). Genes with no raw read count greater than or equal to 20 in at least 1 sample were filtered out with an R script, raw read counts were normalized, and a differential expression analysis was performed with DESeq2 by applying an adjusted P < 0.05 and an absolute log 2 ratio larger than 1.
Regions obtained by MACS2 were subjected to differential analysis using DESeq2 provided by SeqMonk 1.43.0 (Mapped Sequence Analysis Tool, Babraham Bioinformatics, http://www.bioinformatics.babraham. ac.uk/projects/seqmonk/). First, we created an atlas containing all obtained peaks for all the populations using bedtools (74) with a minimum overlapping of 1 bp. We used DESeq2 (75) with a P-adjusted cutoff of 0.05. Resulting peaks were separated into 2 categories: peaks located in promoters (located within 2 kb around the transcription start site) and peaks located in enhancers (not located in the defined promoter regions). For downstream visualization, a scaling factor was calculated using deepTools package (76) to normalize peak intensity to fraction of reads in peaks and generate bigWig files. For Gene Ontology analysis, we introduced BED files from differential ATAC-Seq peaks to GREAT with default parameters (50). For motif analysis, CiiiDER algorithm was used to perform motif enrichment in the differentially accessible regions. We used BETA package with default parameters to integrate ATAC-Seq (differentially accessible regions) and RNA-Seq (transcriptome) data and evaluate the regulatory potential of chromatin accessibility to promote/repress genes' expression.
Single-cell RNA-Seq analysis. Preprocessed transcription-counts matrices from tumor-infiltrating CD45 + cells of 8 primary breast carcinomas were downloaded in duplicates, resulting in 21,346 cells, and loaded into R to be analyzed using Seurat package (version 3.1.2) (77). SCTransform function with default parameters was used for normalizing, scaling, and finding variable features among cells. Principal component analysis was performed with default parameters and used in UMAP analysis to identify clusters. Cell populations were identified by performing differential expression analysis between clusters (related to Figure 2). To identify cells expressing HO-1, we annotated cells with HMOX1 > 0 counts as positive (535 cells) and HMOX1 = 0 count as negative (20,811 cells). We performed a new differential expression analysis between HMOX1 + and HMOX1clusters to identify genes associated with HO-1 expression (related to Supplemental Figure 2). We used AUCell R package (78) to score the activity of gene sets in each cell. We surveyed 21,346 cells to score gene sets that include genes highly correlated with TAMs' activation and monocytes' activation (42). Threshold of the area under the recovery curve was set to 0.2 and 0.3 for the TAM activation component and the monocyte activation component, respectively.
Data availability. RNA-Seq and ATAC-Seq data that support the findings reported in this study have been deposited in the National Center for Biotechnology Information's Gene Expression Omnibus database with the accession code GSE148585.
Statistics. All data points were included with median and interquartile range. A 2-tailed nonparametric Mann-Whitney U test was used to compare 2 data sets and a Wilcoxon matched-pairs signed-rank test when different tumors from the same animal were compared. Differences were considered statistically significant as follows: P values less than 0.05 were flagged with *, less than 0.01 with **, less than 0.001 with ***, and less than 0.0001 with ****. NS means not statistically significant. All graphs were created and statistical analyses were performed using GraphPad Prism 6 software.
Study approval. All animal studies were approved by the Animal Welfare and Ethics Committee of the Université Libre de Bruxelles Institute of Molecular Biology and Medicine. All experiments were conducted in accordance with recommended guidelines and regulations.

Author contributions
EA conducted most of the experiments. BV, AD, AA, and AC contributed to some experiments. AA, M Splittgerber, and FL performed bioinformatics analysis. EA, AD, AA, and BV analyzed the data and prepared the figures. M Soares and LB provided critical reagents. ALM and SG supervised the work and wrote the manuscript. All authors were involved in critically revising the manuscript for important intellectual content. All authors had full access to the data and approved the manuscript before it was submitted by the corresponding author.