Repurposing Tin Mesoporphyrin as an Immune Checkpoint Inhibitor Shows Therapeutic Efficacy in Preclinical Models of Cancer

Unparalleled clinical responses have been achieved in cancer patients using antibodies which block the immune checkpoint molecules PD-1 and CTLA-4 to prevent CD8⁺ T-cell anergy in the tumor microenvironment (1). Combining PD-1 and CTLA-4 blockade therapies improves the overall clinical response compared with the monotherapies (2), highlighting the complementary effects of these checkpoint molecules. However, the absence of a clinical response in a significant proportion of patients suggests that further therapeutic interventions might be required to account for the number and hierarchy of immune checkpoint pathways (2, 3), as well as for the adaptive immune checkpoint response of the tumor (4).

The heme oxygenase (HO) family of proteins is responsible for the breakdown of heme, which is released from dying cells, to the biologically active products biliverdin, ferrous iron (Fe²⁺) and carbon monoxide (CO; ref. 5). There are two members of the family expressed in humans and mice: HO-1 which is induced in response to stress stimuli, and HO-2 which is constitutively expressed at basal levels in all cells (5, 6). HO-1 has been implicated in many studies to have pro-tumoral properties, including cytoprotection and immune suppression (5, 7–12). Many of these effects have been attributed to CO due to its ability to modulate several signaling pathways including p38 MAPK (13), STAT1/3 (14), and NFκB (15, 16). NFκB in particular is vital for CD8⁺ T-cell effector function (17). As such, HO activity can compromise antitumor CD8⁺ T-cell responses in the tumor microenvironment (7).

HO activity has yet to be targeted in the clinic for the treatment of cancer. SnMP is a potent HO inhibitor that targets both HO-1 and HO-2 (18), and has been administered to infants to target hepatic HO activity to control excessive serum bilirubin levels for conditions such as neonatal jaundice and Crigler-Najjar syndrome (19, 20). However, despite this, very few published studies have used SnMP to target HO-1 in the cancer field. SnMP does not affect tumor growth in ectopic Lewis lung carcinomas as a single agent in murine models, unless ovalbumin is artificially introduced into the tumor to raise a robust antitumor immune response (7).

As such, repurposing SnMP for the treatment of cancer is an attractive route to target the pro-tumoral properties of this enzyme. However, for clinical efficacy it is expected that SnMP would need to be combined with an immune-stimulating agent or therapy. Chemotherapeutics can both prime antitumor CD8⁺ T-cells and elicit their infiltration into the tumor microenvironment (21–24). Because of their widespread administration in cancer, these compounds may provide the appropriate immunological setting to target the immune-suppressive activity of HO-1.

In this preclinical study we demonstrate, using a chemotherapy-elicited antitumor immune response in a spontaneous murine model of mammary adenocarcinoma (MMTV-PyMT), that SnMP can be used as an immune checkpoint therapy that targets myeloid-derived HO-1 to permit immunological control of tumor growth when combined with chemotherapy. We show that HMOX1 expression is a poor prognostic factor in patients receiving chemotherapy and present data to suggest that HO-1 can be hierarchically equal or more important than the clinically targeted immune checkpoint PD-1 in certain tumor microenvironments, underlining the need to evaluate SnMP as an immune checkpoint therapy in the treatment of cancer.

MMTV-PyMT mice used in this study were on a FVB/n background. The Hmox1^fl/fl (25) and Itgam^cre(26) transgenic mice were on a C57Bl/6 background and a gift from Professor George Kollias, Biomedical Sciences Research Center “Alexander Fleming,” Athens, Greece. These mice were crossed to generate the MMTV-PyMT, Hmox1^fl/fl, Itgam^cre+/− line, which was of mixed background. Cohort sizes were informed by prior studies (7). All mice used for experiments were female and randomly assigned to treatment groups. Mice were approximately 26 g when tumors arose. Experiments were performed in at least duplicate and for spontaneous tumor studies individual mice were collected on separate days and all data points are presented. All experiments involving animals were executed in compliance with Institutional and Home Office UK guidelines and regulations (licence 70/7654).

In studies using MMTV-PyMT mice the primary tumor growth was presented. Sn (IV) mesoporphyrin IX dichloride (SnMP; Frontier Scientific) was dissolved and administered as previously described (7). 5-fluorouracil (Sigma-Aldrich) was dissolved in saline. Anti-mouse PD-1 (RMP1-14; Biolegend) was administered at 12 mg/kg every 3 days. Immune-depleted mice were injected every 5 days, starting 48 hours before the commencement of treatment, with 400 μg of either anti-CD8α (53-6.7) or isotype control rat IgG2a (2A3Clone; eBioscience). All drugs were freshly prepared on the day of injection, and administered by intraperitoneal injection. Tumor tissue was enzyme-digested to release single cells as previously described (27).

Sections of fresh-frozen human breast adenocarcinoma or mouse mammary tumors embedded in OCT were fixed in 4% paraformaldehyde in PBS for 10 minutes at RT. For histology, sections were washed in Tris Buffered Saline; Tween 20, 0.05% (TBST) and blocked with 10% donkey serum (Sigma-Aldrich), 0.2% Triton X-100. HO-1 was detected using 1:100 anti-HO-1 (412811; R&D Systems) and 1:200 anti-rat IgG HRP (Thermo Fisher Scientific) and developed using 3,3′-diaminobenzidine (Thermo Fisher Scientific). Slides were counterstained with hematoxylin, dehydrated, and scanned using a NanoZoomer Digital Slide Scanner (Hamamatsu). Immunofluorescence was performed as previously described (10). The following antibodies and dilutions were used: F4/80, 1:100 (C1:A3-1, Bio-RAD), HO-1, 1:100 (EP1391Y, Origene), CD3, 1:100 (SP7, Abcam), CD8, 1:100 (1:1 ratio of YTS169.4 and 53-6.7; Thermo Fisher Scientific), CD11b, 1:100 (ICRF44, eBioscience), CD14, 1:100 (61D3, eBioscience). Primary antibodies were detected using antigen-specific Donkey IgG, used at 1:200: AlexaFluor 488 anti-rabbit IgG, AlexaFluor 488 anti-rat IgG, AlexaFluor 647 anti-mouse IgG, AlexaFluor 647 anti-rabbit IgG, all of which were purchased from Thermo Fisher Scientific. Nuclei were stained using 1.25 μg/mL 4′,6-diamidino-2-phenylindole, dihydrochloride (DAPI; Thermo Fisher Scientific). Images were acquired using a Nikon Eclipse Ti-E Inverted spinning disk confocal with associated NIS Elements software.

Flow cytometry was performed as previously described (7). The following antibodies were purchased from eBioscience and were used at 1 μg/mL unless stated otherwise: Anti- B220 APC (RA3-6B2), CD3ϵ APC and PE (145-2C11), CD4 FITC (RM4-5), CD8α eFluor 450 (53-6.7), CD8β FITC and eFluor 450 (H35-17.2), CD11b eFluor 450 (M1/70), and CD11b Brilliant Violet 510 (M1/70; Biolegend), CD11c APC (N418), CD16/32 (2.4G2; Tonbo Biosciences), CD19 APC (6D5; Biolegend), CD45 APC-eFluor 780 (30-F11), CD90.2 eFluor 450 (53-2.1), CD273 PE (B7-DC), CD274 PE (B7-H1), CD279 APC (J43), F4/80 PE (BM8), Granzyme B PE (NGZB) IFN-γ APC (XMG1.2), Ly6C PE (HK1.4), Ly6G APC and FITC (1A8; Biolegend), NK1.1 APC (PK136), TNF PE (MP6-XT22). Where stated, the following corresponding isotype control antibodies at equivalent concentrations to that of the test stain were used: goat IgG APC and PE (Bio-techne), rat IgG2b APC and eFluor 450 (eB149/10H5), rat IgG2a APC and PE (eBR2a) and Armenian Hamster IgG APC (eBio299Arm). Intracellular stains were performed as previously described (7). Dead cells and red blood cells were excluded using 1 μg/mL 7-amino actinomycin D (Sigma-Aldrich) or Fixable Viability Dye eFluor 780 (eBioscience) alongside anti-Ter-119 PerCP-Cy5.5 (Ter-119; eBioscience). Data were collected on a BD FACSCanto II (BD Biosciences). Data were analyzed using FlowJo software (Freestar Inc.). Immune cells (CD45⁺) were separated based upon the following surface characteristics: CD11c⁺ F4/80⁻ (dendritic cells), CD11b⁺ F4/80^hi (macrophages), CD11b⁺ F4/80^low Ly6G⁻ Ly6C⁺ (monocytes), CD11b⁺ Ly6G⁺ (neutrophils), NK1.1⁺ CD3ϵ⁻ (NK cells), NK1.1⁺, CD3ϵ⁺ (NKT cells), CD3ϵ⁺ (T cells), CD3ϵ⁺ CD4⁺ (CD4⁺ T cells), CD3ϵ⁺ CD8α/β⁺ (CD8⁺ T cells), CD19⁺ (B cells). Cancer-associated fibroblasts (CAF) were identified as CD45⁻ Thy1⁺ cells.

mRNA was extracted and quantitative reverse transcription PCR was performed as previously described (7) using the EXPRESS one-step Superscript RT PCR kit and the following primers/probes purchased from Thermo Fisher Scientific: Cxcl9 mm00434946_m1, Cxcl10 mm00445235_m1, Cxcl11 mm00444662_m1, Tbp mm01277045_m1. Expression of all genes is represented relative to the house-keeping gene Tata-binding protein (Tbp). Assays were performed using an ABI 7900HT Fast Real Time PCR instrument (Thermo Fisher Scientific).

Patient survival data were acquired from the BreastMark dataset (28), the Kaplan–Meier plotter (29–31), and the NCBI Gene Expression Omnibus, accession number GSE65904 (32). Expression data were acquired from the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC; ref. 33). Heat maps were created with Gene Cluster 3.0 and Treeview Software. The single-cell RNA-seq (34) dataset was extracted from the NCBI GEO, accession code GSE75688. Data were acquired as raw gene expression in transcripts per million (TPM). Low-quality cells were filtered from the dataset based upon a threshold of 2000 unique genes detected per cell. Genes that were detected in <2% of single cells were removed leaving 509 cells and 19,097 genes. Data were log transformed [log(TPM + 1)] for all further analyses, which were performed using the R software package Seurat (https://github.com/satijalab/seurat; http://satijalab.org/seurat/). See Supplementary Methods for a detailed description of the single-cell RNASeq analysis pipeline.

Normality and homogeneity of variance were determined using a Shapiro–Wilk normality test and a F-test, respectively. Statistical significance was then determined using a two-sided unpaired Students t test for parametric, or Mann–Whitney test for nonparametric data using GraphPad Prism 6 software. When comparing paired data, a paired ratio Student t test was performed. A Welch's correction was applied when comparing groups with unequal variances. Statistical analysis of tumor growth curves was performed using the “compareGrowthCurves” function of the statmod software package (35). No outliers were excluded from any data presented.

The use of animals for this study was approved by the Ethical Review Committee at King's College London and the Home Office, UK. Human breast adenocarcinoma tissue was obtained with informed consent under ethical approval from the King's Health Partners Cancer Biobank (REC reference 12/EE/0493).

To evaluate the potential clinical efficacy of SnMP on tumor growth, we used the aggressive spontaneous murine model of mammary adenocarcinoma (MMTV-PyMT; ref. 36). This preclinical model closely follows the stages to malignancy that occur in human cancer progression (37). HO-1⁺ cells could be detected by IHC in MMTV-PyMT tumors (Fig. 1A and Supplementary Fig. S1A). These cells accounted for 2.0% ± 1.1% of the tumor (Fig. 1B). HO-1⁺ cells were primarily detected as clusters across the stroma, but there was no obvious anatomical location within the tumor in which these cells were predominanly found (Fig. 1A and Supplementary Fig. S1A). All tumors analyzed expressed Hmox1, the gene for HO-1, although to varying degrees (Supplementary Fig. S1B), which was in agreement with the IHC analysis. MMTV-PyMT mice bearing established spontaneous mammary adenocarcinomas were treated with the HO inhibitor SnMP or with vehicle, and their tumor growth was monitored therafter (Fig. 1C). SnMP did not affect tumor growth in this model, consistent with that observed in ectopic murine Lewis Lung carcinomas (7). Treatment with 5-FU, a chemotherapeutic which has been used in the clinic (38), and which has been demonstrated to stimulate an antitumor immune response (39, 40), also did not affect tumor growth as a single agent (Fig. 1C). Chemotherapy has been described to upregulate HO-1 expression (8, 12), but this was not observed with 5-FU (Supplementary Fig.S1B). When SnMP and 5-FU were concurrently administered there was striking therapeutic synergy between these two drugs, resulting in immediate tumor regression and control of tumor growth (Fig. 1C). These data demonstrate that SnMP could be used to improve responses to chemotherapy.

HO-1 has been demonstrated to modulate the tumor cell stress response (8–12) and to suppress immune-mediated cytotoxicity (11) in response to chemotherapy treatment. To establish whether the tumor control was immunological, SnMP and 5-FU were administered to mice that had been depleted of CD8⁺ T-cells in vivo before the initiation of treatment. This depletion resulted in a loss of tumor-infiltrating CD8⁺ T-cells (Supplementary Fig. S2A and S2B). Administration of anti-CD8α antibodies resulted in a slowing of tumor growth that became evident at the later stages (Fig. 2A). However, the regression and long-term control of tumor growth seen with SnMP and 5-FU treatment was lost in the absence of tumoral CD8⁺ T-cells, demonstrating that the mechanism of tumor control was CD8⁺ T-cell dependent (Fig. 2A). Because tumor remission was observed within 48 hours of the initiation of treatment, a role of chemotherapy in priming an antitumor immune response was unlikely (22). This suggested that the effects were more likely mediated by modulation of an ongoing spontaneous response. MMTV-PyMT tumors which had been acutely treated for 36 hours were analyzed to characterize the earliest biological changes at the point at which tumors had started to regress (Fig. 2B). Tumors were excised and enzyme-digested to assess the prevalence of lymphocytes in the tumor microenvironment (Fig. 2C and D). In the treatment groups containing 5-FU, there was a significant influx of CD4⁺ and CD8⁺ T-cells (Fig. 2D), whereas SnMP as a single agent had no effect on the tumoral accumulation of lymphocytes (Fig. 2D). This was a specific influx of lymphocytes and not an artifact of a changing microenvironment, as there was no difference in the tumoral abundance of TAMs, monocytes, neutrophils or CAFs (Fig. 2E). CD3⁺ CD8^+/− T cells could also be observed infiltrating the tumor tissue by immunofluorescence staining of frozen tumor sections (Fig. 2F and G and Supplementary Fig. S2C). This T-cell influx coincided with a tumoral increase in expression of the T-cell chemokines Cxcl9, 10, and 11, suggesting that this was a chemotactic response (Fig. 2H). There was no increase in expression of other genes which have been associated with resistance to chemotherapy such as VEGF, cathepsin proteases or IL-12 (ref. 41; Supplementary Fig. S2D). We therefore assessed the activation status of the tumor-infiltrating CD8⁺ T-cells. We found that there was an increased prevalence of CD8⁺ T-cells capable of expressing the effector molecule granzyme B (Fig. 2I). Moreover, ex vivo inhibition of HO increased the proportion of tumoral CD8⁺ T-cells capable of producing the effector cytokines IFNγ and TNFα in response to phorbol myristate acetate (PMA)/ionomycin stimulation (Fig. 2J), indicating that HO was suppressing the effector functions of the infiltrating CD8⁺ T-cells. The therapeutic synergy of combining chemotherapy and SnMP was not tumor model or chemotherapy class specific, as SnMP also improved the immune-mediated effects of paclitaxel in the orthotopic 4T1 model of murine mammary adenocarcinoma (Supplementary Fig. S2E and S2F). These data suggest that SnMP could be regarded as an innate immune checkpoint therapy in murine models of breast cancer.

In MMTV-PyMT tumors, 87.3 ± 8.0% of HO-1⁺ events could be found colocalizing with the TAM marker F4/80 by immunofluorescence staining of frozen tumor sections (Fig. 3A and B; Supplementary Fig. S3A). TAMs were also the most abundant stromal cell type in MMTV-PyMT tumors (Fig. 3C), permitting the conclusion that TAMs were the predominant tumoral source of the enzyme. In agreement, F4/80⁺ TAMs were the primary source of HO-1 in orthotopic 4T1 tumors (Supplementary Fig. S3B and S3C). TAMs play a pivotal role in resistance to chemotherapy (42–44) and facilitate relapse after the cessation of treatment (45). We considered whether TAM-derived HO-1 might have accounted for these observations, and whether TAM-derived HO-1 mediated the therapeutic efficacy of SnMP. To investigate this, the Hmox1 gene was specifically inactivated in the myeloid lineage by crossing the MMTV-PyMT transgene to a mouse carrying a loxP flanked Hmox1 gene (25) and cre recombinase driven by the myeloid-specific integrin CD11b (Itgam) promoter (26). This resulted in a 64% ± 4.6% reduction in macrophage expression of HO-1 (Supplementary Fig. S4). Myeloid-specific inactivation of HO-1 did not affect the latency of tumor establishment (Fig. 3D) nor tumor growth thereafter (Fig. 3E). However, administration of 5-FU to cre⁺ mice in which myeloid-derived HO-1 had been inactivated resulted in tumor regression and long-term control of tumor growth, mirroring the effects of SnMP treatment (Fig. 1C). Inactivation of myeloid HO-1 did not affect the abundance of TAMs in terminal size tumors (Fig. 3F), nor the abundance of CD8⁺ T-cells (Fig. 3G). However, in line with the ex vivo effects of HO inhibition (Fig. 2J), HO-1 inactivation resulted in an increased proportion of tumoral CD8⁺ T-cells capable of producing the effector molecules IFNγ and TNFα (Fig. 3H), demonstrating that SnMP was most likely targeting myeloid HO-1 activity to alleviate the suppression of the chemotherapy-elicited CD8⁺ T-cell response.

In light of these observations, we investigated whether HMOX1 expression had any prognostic value in patients. In the BreastMark dataset (28), HMOX1 expression was associated with poor prognosis, especially in patients receiving chemotherapy (Fig. 4A and B) with a hazard ratio of 2.15 (P = 0.003). HMOX1 was one of the genes most highly associated with poor prognosis in this patient group (Supplementary Fig. S5A). In contrast, HMOX1 had no association with survival in patients that had not received chemotherapy (Fig. 4C). The same association was found using the Kaplan–Meier Plotter dataset (ref. 29; Fig. 4D–F). This association was specific to HMOX1, and not a general result of enhanced infiltration of myeloid cells or macrophages in these tumors, since the myeloid associated genes CD14 and ITGAM had no prognostic value (Supplementary Fig. S5B and S5C). The absence of an effect of HMOX1 expression in patients that had not received chemotherapy was also not due to lower grade of disease in this subset (Fig. 4G–I). This association between high HMOX1 expression and poor prognosis also applied to gastric (Fig. 4J–L) and lung cancer (Supplementary Fig. S5D) patients receiving chemotherapy, but did not apply to melanoma patients (Supplementary Fig. S5E), suggesting that the prognostic value of HO-1 may depend on the specific tumor type. In accordance with our preclinical data, these observations suggest a detrimental role of HO-1 expression in patients receiving chemotherapy.

As HO-1 was prognostic for patients receiving chemotherapy, and because our preclinical models of breast cancer had suggested that TAMs were the major tumoral source of HO-1, we investigated whether these observations might translate to the clinic. We therefore analyzed a published single-cell RNA-seq dataset of 509 cells taken from 11 patients with invasive ductal carcinoma of the breast who had undergone breast-conserving surgery or total mastectomy (34). Using an RNA-seq analysis pipeline (see Supplementary Materials and Methods) we grouped cells based on their transcriptomic profiles, using non-linear dimensionality reduction [t-stochastic nearest neighbor embedding (tSNE)] followed by density based clustering (Fig. 5A, Supplementary Figs. S6 and S7). Clusters of cells were annotated by their respective cell lineages based upon their distinguishable preferentially expressed marker genes (Supplementary Table S1). Within each cluster group we assessed expression of HMOX1. In agreement with our observations in preclinical models of breast cancer, HMOX1 expression was most prevalently expressed in the myeloid cell population (Fig. 5B). Of the myeloid cells analyzed, 67% had detectable HMOX1 expression, whereas only 28% of epithelial/tumor cells had detectable HMOX1 expression. Although this analysis does not account for the overall abundance of the various cells types, it does demonstrate that myeloid cells are a major tumoral source of the enzyme in human breast cancer. Myeloid cells were also a major tumoral source of the other checkpoint molecules PDL1 and PDL2 in human cancer (Fig. 5C and D), suggesting that the myeloid population could play a key role in immune-regulation in these tissues. To confirm that HO-1 protein can be detected in myeloid cells within the breast tissue, tumor sections of human mammary adenocarcinoma were stained for CD11b and HO-1 by immunofluorescence. A population of CD11b⁺ (Fig. 6A and Supplementary Fig. S8A) and CD14⁺ (Supplementary Fig. S8B) myeloid cells expressed HO-1 in these tumors confirming that HMOX1 could be found translated to HO-1 protein in these cells. Using the METABRIC dataset of breast cancer transcriptomes (33), we separated the upper and lower quartiles of HMOX1 expression in the search for genes co-regulated with HMOX1. The group patient characteristics are presented in Supplementary Table S2. HMOX1 expression was found to be closely associated with myeloid markers CD163, LYZ, CD14, CD68, and ITGAM (Fig. 6B), which was in agreement with the single-cell RNA-seq dataset. This was not representative of a broader stromal association, as the neutrophil marker SELL was only weakly associated and ELANE was not associated with HMOX1 expression. Collectively, these data demonstrate that in breast cancer, myeloid cells represent a major tumoral source of HMOX1 expression as well as of other checkpoint inhibitor genes such as PDL1 and PDL2.

As HO-1 is an inducible protein and is expressed in response to various stimuli (5, 6), we investigated whether expression of HMOX1 could be associated with any common features in the tumor microenvironment using the METABRIC dataset (33). Interestingly, the tumors with the highest HMOX1 expression were associated with an immunologically “hot” gene signature (Fig. 6C) and had higher expression of the lymphocyte associated genes CD19, CD4 and CD8A (Fig. 6D). This association has previously been described for PD-L1 (46). Indeed, in human breast cancer, the tumors expressing the highest levels of PDL1 and PDL2 also had significantly higher HMOX1 expression (Fig. 6E). This raised the possibility that redundancy may exist between these immune checkpoint molecules, which could compromise the therapeutic efficacy of the respective monotherapies targeting HO-1 and PD-1. MMTV-PyMT tumors were therefore assessed for PD-1 and PD-L expression. PD-1 expression could be detected on 17% of tumoral CD8⁺ T-cells in this model (Fig. 6F), which paralleled the observations in human breast cancer, where on average 19% of tumor-infiltrating CD8⁺ T-cells express PD-1 (47). Administration of 5-FU did not affect the prevalence of PD-1 on the CD8⁺ T-cell population in the MMTV-PyMT model (Fig. 6F). As was observed in the human breast cancer dataset (Fig. 5C and D), PD-L1/-L2 were predominantly expressed on the TAM population (Fig. 6G). Because the MMTV-PyMT model mirrored the human disease, the biological consequence of immune checkpoint co-expression was analyzed in MMTV-PyMT mice bearing large (approximately 800 mm³) tumors, a size that would permit analysis of potential synergistic effects of combination therapy. PD-1 blockade suppressed tumor growth when compared with vehicle treatment, but did not result in remission of the disease, and this was not enhanced by the addition of 5-FU (Fig. 6H). Administration of SnMP alongside 5-FU was more efficacious than PD-1 blockade in MMTV-PyMT tumors (Fig. 6H). Combining the three therapies did not improve the response compared to SnMP/5-FU (Fig. 6H), suggesting that, despite the presence of PD-1⁺ CD8⁺ T-cells and PD-L1/2 expression in the tumor microenvironment, HO-1 represented a hierarchically more important immune checkpoint. The same experiment was conducted in orthotopic 4T1 tumors using paclitaxel (Supplementary Fig. S9). In 4T1 tumors 25.8 ± 6.4% of CD8⁺ T-cells expressed PD-1 (Supplementary Fig. S9A and S9B), and in agreement with the MMTV-PyMT tumors and the human disease, PD-L1/2 were predominantly expressed by the TAMs (Supplementary Fig. S9C). Treatment of mice bearing established 4T1 tumors with SnMP and/or PD-1 blockade alongside paclitaxel resulted in an equivalent control of tumor growth in this model (Supplementary Fig. S9D). Together, these observations suggest that HO-1 represents an important immune checkpoint which can suppress chemotherapy-elicited antitumor immune responses. This highlights the clinical need to evaluate SnMP as an immune checkpoint therapy, especially in patients receiving chemotherapy.

In this study, we have demonstrated that SnMP has significant potential to be re-evaluated as a novel immune checkpoint therapy in patients receiving immune-stimulating chemotherapy (illustrated in Supplementary Fig. S10). Chemotherapeutics have been used for the treatment of cancer for nearly 70 years. Although these drugs were first administered to patients for their ability to target the cell cycle, their immune-stimulating properties may underlie a significant proportion of their clinical efficacy (21–24). As we demonstrate here, the efficacy of these therapies can be compromised by tumor immune checkpoints, a concept that is supported by the observation that inhibition of the checkpoint molecules PD-1 and CTLA-4 improves the response to chemotherapy (24). Accordingly, we have shown that chemotherapies used in cancer treatment such as 5-FU and paclitaxel provide valuable tools for eliciting antitumor immune responses. Importantly, this study has demonstrated that chemotherapy-elicited CD8⁺ T-cell responses are sufficient to achieve long-term immunological control of tumor growth when combined with appropriate immune checkpoint blockade therapy. These data also suggest that HO-1 activity might contribute to unresponsiveness of a group of patients to the current immune checkpoint blockade therapies (1, 2). CO, one of the by-products of heme catabolism, has been demonstrated to have potent immune-suppressive properties through directly modulating cell signaling (48). As such, HO-1 could be regarded as an innate checkpoint compared with PD-1 and CTLA-4, as it does not require receptor-ligand interactions. As only 17% of CD8⁺ T-cells expressed PD-1 in MMTV-PyMT tumors, which mirrored the clinical setting, this might in part explain why HO-1 inhibition was more efficacious than PD-1 blockade, despite the expression of both immune checkpoints in the tumor. In breast cancer, we demonstrate that expression of the genes for both HO-1 and PD-L are most prevalent within the myeloid-stromal compartment. In other cancers PD-L1 and PD-L2 have been described to be widely expressed by tumor cells (49), highlighting the inter-tumor heterogeneity of immune checkpoint expression. Inter- and intra-tumor hierarchy and redundancy of these molecules will ultimately dictate responses to immune checkpoint blockade therapies in the clinic. These observations therefore suggest that the “tolerogenic landscape” of each tumor needs to be considered when implementing immune checkpoint blockade therapies, or that broader drug combinations might be required to achieve efficient immune checkpoint blockade therapy in the greatest number of patients.

SnMP has completed a phase II (b) trial for the treatment of neonatal hyperbilirubinemia as a single bolus dose injection (19). Because of its low toxicity, repeat dosing of SnMP has also been successfully tested clinically (3 doses/week) in two patients with Crigler-Najjar syndrome (20). Repeated SnMP dosing is well tolerated in humans with the most frequent toxicities being mild (and reversible) cutaneous photosensitivity of sun-exposed areas and anemia (20). The human equivalent of the dose of SnMP used in our study is within the range previously administered to patients, making these observations directly clinically relevant. As such, these data support the clinical evaluation of SnMP in cancer, as a novel immunotherapeutic approach to improve the immunological response to chemotherapy by targeting HO-1.

No potential conflicts of interest were disclosed.

Conception and design: T. Muliaditan, J.F. Spicer, J.N. Arnold

Development of methodology: T. Muliaditan, J.W. Opzoomer, J. Caron, P. Kosti, S. Kordasti, J.N. Arnold

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Muliaditan, J.W. Opzoomer, J. Caron, M. Okesola, A. Tutt, C.E. Gillett, J.M. Burchell, J.N. Arnold

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Muliaditan, J.W. Opzoomer, J. Caron, P. Kosti, M. Van Hemelrijck, A. Grigoriadis, S.F. Madden, J.F. Spicer, J.N. Arnold

Writing, review, and/or revision of the manuscript: T. Muliaditan, J.W. Opzoomer, J. Caron, P. Kosti, M. Van Hemelrijck, F. Dazzi, S.F. Madden, S. Kordasti, S.S. Diebold, J.F. Spicer, J.N. Arnold

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Okesola, S. Lall, A. Tutt

Study supervision: J.M. Burchell, S.S. Diebold, J.N. Arnold

The authors thank Thomas Hayday and Dr. Yasmin Haque (KCL) for cell sorting and flow cytometry assistance, Dr. Hans Garmo (senior medical statistician, KCL) for his advice on the application of statistics used in this study, Charlotte Saxby for running the Western blot shown in Supplementary Fig. S3C, Dr. Patrycja Gazinska (KCL) for scanning the histology slides, Samantha Arnold for preparing Supplementary Fig. S9, Dr. Morris Muliaditan (University College London) for critical review of the article and the Nikon Imaging Centre KCL for use of their facilities and assistance with confocal microscopy. This work was funded by grants from the European Research Council (335326) and Cancer Research UK (22253) and supported by the Experimental Cancer Medicine Centre at King's College London and the National Institute for Health Research (NIHR) Biomedical Research Centre based at Guy's and St Thomas' NHS Foundation Trust and King's College London.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.