Bleomycin Exerts Ambivalent Antitumor Immune Effect by Triggering Both Immunogenic Cell Death and Proliferation of Regulatory T Cells

Bleomycin (BLM) is an anticancer drug currently used for the treatment of testis cancer and Hodgkin lymphoma. This drug triggers cancer cell death via its capacity to generate radical oxygen species (ROS). However, the putative contribution of anticancer immune responses to the efficacy of BLM has not been evaluated. We make here the observation that BLM induces immunogenic cell death. In particular, BLM is able to induce ROS-mediated reticulum stress and autophagy, which result in the surface exposure of chaperones, including calreticulin and ERp57, and liberation of HMBG1 and ATP. BLM induces anti-tumor immunity which relies on calreticulin, CD8+ T cells and interferon-γ. We also find that, in addition to its capacity to trigger immunogenic cell death, BLM induces expansion of Foxp3+ regulatory T (Treg) cells via its capacity to induce transforming growth factor beta (TGFβ) secretion by tumor cells. Accordingly, Treg cells or TGFβ depletion dramatically potentiates the antitumor effect of BLM. We conclude that BLM induces both anti-tumor CD8+ T cell response and a counteracting Treg proliferation. In the future, TGFβ or Treg inhibition during BLM treatment could greatly enhance BLM anti-tumor efficacy.


Introduction
The aim of anti-cancer therapy is the eradication of tumor cells in the patient's body, including the smallest metastasis or singlecell localization that could not be removed by surgery. Chemotherapy and radiotherapy are currently used to complement surgery to kill residual disease or for the treatment of metastatic disease. However, tumor cells can develop some resistance and escape cytotoxic treatment, causing tumor relapse. In this context, immunotherapy is seen as one of the ultimate anti-cancer strategies, as the immune system could reshape its actions against an evolving malignancy and eliminate tumor cells resistant to chemotherapy. Accumulating evidence shows that conventional chemotherapies, in addition to their direct cytotoxic effect, could trigger an antitumor immune response. In particular, we have shown that some chemotherapies activate natural killer (NK) cells [1,2], some drugs target suppressive cells such as regulatory T cells (Treg) or myeloid-derived suppressor cells (MDSC) [3,4] and some other drugs also induce a particular mode of cancer cell death called ''immunogenic cell death'' (ICD) [5]. ICD relies on the capacity of drugs to induce three main checkpoints. The first one is the exposure of 'eat-me' signals on cell surface, such as calreticulin (CRT), caused by endoplasmic reticulum (RE) stress [6], which favors cancer cell phagocytosis by dendritic cells [7]. The second is the secretion of an endogenous Toll-like receptor 4 (TLR4) ligand High-mobility group box 1 (HMGB1) [8], which is required for efficient processing of tumor antigen by dendritic cells. The third signal is the release of ATP which induces the production of interleukin (IL)-1b by dendritic cells and favors CD8 + T cell immune response [9]. Bleomycin (BLM) is an anti-tumor antibiotic glycopeptide produced by the bacterium Streptomyces [10]. BLM causes breaks in DNA, similar to those obtained with radiotherapy [11]. This DNA damage has been demonstrated to be mediated through induction of oxidative stress [12,13]. BLM is indicated, in association with other cytotoxic agents, for the treatment of cancer testis and Hodgkin disease [14,15]. The particularity of these two diseases is the high rate of cure obtained by chemotherapy. So, we proposed to test the antitumor immune response induced by BLM to determine if this mechanism could participate in the antitumor effect of bleomycin and may explain the high capacity of BLMcontaining regimens to cure cancer.
In this study, we analyzed the implication of ICD, effector cells (such as CD8+ T and NK cells), as well as modulator cells (such as dendritic cells, MDSC and Treg) in BLM antitumor efficacy. We describe how BLM treatment induces an immunogenic apoptosis of tumor cells, anti-tumor CD8 + T cell response and the production of the tolerogenic cytokine transforming growth factor beta (TGFb) by tumor cells which mediates regulatory T cell accumulation in vivo.

Materials and Methods
Cell Culture MCA205 (fibrosarcoma), B16F10 (melanoma) cell lines, syngeneic of C57BL/6 mice, and CT26 (colon carcinoma) cell line, syngeneic of Balb/c mice, were cultured at 37uC in 5% CO 2 in RPMI 1640 (Lonza) enriched with 10% fetal bovine serum, 0.4 mmol/L of sodium pyruvate, 4 mmol/L of HEPES, and antibiotics (penicillin, streptomycin, and amphotericin B). All cell lines were obtained from the American Type Culture Collection, excepted for B16-OVA, which was a kind gift from Pr Laurence Zitvogel.
Tumor models. CT26 tumor cells were injected into the right flank of Balb/c mice (1.10 6 cells). Tumor size was monitored three times a week with a caliper. In some experiments, 1.10 6 cells were injected intraperitoneally to induce peritoneal carcinosis.

In vitro Treatments
Tumor cells were treated for 24 hours with 30 mM Mitomycin C, 0.25 mM doxorubicin, 30 mM N-acetyl-cystein, or 15 mg/mL BLM for CT26 and MCA205 cells, 150 mg/mL BLM for B16F10.
Cell transfer. CD45.2 Foxp3-EGFP mice were sacrificed then lymph nodes and spleens were harvested. These organs were mechanically disrupted and the single cell suspension was enriched in CD4 + lymphocytes using MACS CD4 + T cells isolation kit II (Miltenyi). Then CD4 + GFP 2 CD62L + cells (naïve CD4 + T lymphocytes) and CD4 + GFP + cells (Treg) were sorted using FACS Aria III (BD). In vivo, sorted naive T cells and Treg cells were injected intraveinously into CD45.1 C57BL/6 CT26 tumor bearing mice. CD45.1 mice which had received Treg were then treated with PBS, BLM or IL-2. CD45.1 mice which had received naïve CD4 + T cells were treated with PBS or BLM. The day before sacrifice, CD45.1 mice which had received Treg were treated by EdU (70 mg/kg, intraperitoneal). Five days after BLM treatment, all CD45.1 mice were sacrificed and spleens and tumors harvested. Organs were mechanically disrupted, and a CD4 CD45.2 Foxp3 EdU (Click-iT EdU kit, Invitrogen, Saint Aubin, France) staining was performed. Cells were then analysed by flow cytometry. In vitro, Treg were cultured with supernatant from PBS-or BLM-treated CT26 cells in presence of anti-CD3 and anti-CD28 antibodies. Three days later, cells were incubated 3 h with BrdU, then handled with BD BrdU Flow Kit, according to the manufacturer's protocol and FACS-analysed.

Autophagy Detection
Cyto-ID kit (Enzo) was used according to the manufacturer's protocol for flow cytometry and immunofluoresence. Anti-LC3 (clone 5F10, Nanotools) was used on cells fixed and permabilized with BD fix/perm kit and revealed with alexa 488 anti-mouse (InVitroGen).

ROS Detection
Cells were incubated 2 hours with the indicated drugs, washed, then incubated with 100 mM DCFH 2 /DA (Sigma) for 50 min at 37uC, and analysed by flow cytometry.

Immunoblot Analysis
Samples were lysed in lysis buffer containing 1% SDS (Promega), 10 mM Tris HCl pH 7.5, 1 mM NaF (Sigma), 1 mM EDTA (Promega), 2 mM sodium orthovanadate (Sigma) and antiprotease cocktail (Roche, Mannheim, Germany). The protein concentration for each sample was determined by the DC protein assay according to the manufacturer's protocol (Bio-rad, Marnes-la-Coquette, France). After immunoblotting, immunochemical detection was revealed with an appropriate horseradish peroxidase-labeled secondary antibody (Jackson ImmunoResearch). Specific bands were visualized using a chemiluminescence detection system (Santa Cruz, Heidelberg, Germany), according to the manufacturer's protocol. Controls of equal loading of proteins were performed with anti-actin antibody. The following antibodies were used: monoclonal rabbit anti-p-eIF2a (Cell Signaling), monoclonal rabbit anti-eIF2a (Cell Signaling), monoclonal mouse anti-b-actin (Sigma).

Extracellular ATP Concentration
Supernatants were centrifugated twice (425 g, 3 min, room temperature) to eliminate cellular debris. Then ATP concentration was determined with the Cell Titer Glo Viability Assay Kit (Promega), according to the manufacturer's protocol.

ELISA
The following kits were used according to the manufacturer's protocol: HMGB1 detection kit (Chondrex), IFNc ELISA kit (BD), TGFb1 Ready-SET-Go ELISA kit (eBiosciences). shRNA Transduction CT26 cells were transduced with MISSION Lentiviral Transduction Particles (Sigma-Aldrich) to silence CRT or with negative control particles according to the manufacturer's instructions.

Statistical Analysis
Data were analysed with GraphPad Prism 5 Software. Comparisons between control and test conditions were performed with student t test or Mann&whitney test; group comparison were performed with Kruskal-Wallis test followed by Dunn's post-test. Tumor growth comparisons were performed by two-way ANOVA method.

BLM Induces Stigmata of Immunogenic Cell Death
BLM antitumor effect is due to its capacity to induce radical oxygen species (ROS) production [19,20]. Because ROS generation could lead to reticulum stress [21], we hypothesized that BLM could be able to cause ROS-dependent ER stress and the induction of ICD. We found that BLM, like doxorubicin (DOX), an anthracycline and immunogenic drug [7], triggers eIF2a phosphorylation, a marker of reticulum stress in CT26 cells (Figure 1a). Chemotherapy-induced reticulum stress could trigger calreticulin (CRT) and the endoplasmic reticulum chaperone ERp57 translocation at the cell surface [22,23]. In vitro, BLM induced translocation of CRT and ERp57 onto tumor cell plasma membrane in three mouse cancer models ( Figure 1b). As expected, BLM effects are comparable to DOX effects, whereas mitomycin C (Mito C), a non-immunogenic drug [7], did not induce CRT or ERp57 exposure, (Figure 1b). Importantly, we showed that these translocations are aborted by using N-acetylcystein (NAC) (Figure 1c), a ROS inhibitor ( Figure S1). The second determinant of ICD is the capacity of chemotherapy to induce the release of HMGB1 which triggers TLR4 on antigen presenting cells [8]. Using ELISA, we observed that BLM induces similar release of HMGB1 as doxorubicin (Figure 1d). The third element of ICD is the capacity of the chemotherapeutic agent to trigger autophagy and subsequent ATP secretion which activates dendritic cells via P2RX7 [9,24]. We observed that BLM induced, like DOX, LC3 puncta in immunofluoresence and Cyto-ID labeling in flow cytometry, two markers of autophagy vacuoles (Figure 1e, 1f and Figure S2). Of note, mitomycin C was unable to induce these signs of autophagy ( Figure S2). Similarly to DOX, BLM also induced ATP release from CT26 cells (Figure 1g). The ROS inhibitor NAC also blunted both autophagy and ATP release (Figure 1f and 1g).
Together, these data demonstrate that BLM triggers all characteristic events of immunogenic cell death [7][8][9] in a ROSdependent manner.

BLM-induced Immunogenic Cell Death Drives CD8 + Dependent Antitumor Immunity
To determine whether BLM-treated tumor cells induced antitumor immune response in vivo, we vaccinated naïve mice by injecting B16-OVA cells treated with PBS, BLM or DOX in the footpad. We observed that BLM-and DOX-treated cells are able to induce ovalbumin-dependent IFNc production by T cells from the draining lymph node (Figure 2a). We also performed in vivo vaccination with control or CRT-knocked down BLM-treated dying tumor cells, followed by a challenge with live tumor cells. We could observe that the BLM-treated control CT26 cells were able to induce protective immunity (100% protection, n = 10), in contrast to mitomycin-treated CT26 (20% protection, n = 10) and CRT-invalidated BLM-treated cells (30% protection, n = 10). Furthermore, in the setting of growing s/c CT26 tumors, we observed that 5 injections of BLM delayed the growth of control shRNA transfected CT26 cells but had no effect on CRT shRNA transfected CT26 cells (Figure 2b). Finally, IFNc or CD8 depletion completely prevents BLM antitumor effect in vivo (Figure 2c and  2d). Together, these data strongly suggest that BLM exerts a calreticulin-, IFNc-and CD8-dependent antitumor effect.

BLM Shows No Effect on Innate Immune Cells
In addition to its capacity to elicit ICD of tumor cells, chemotherapy could harness the host immune system by inducing dendritic cell maturation [25], NK cell activation [1] or by eliminating immunosuppressive cells such as Treg and MDSC [3,4,26].
To address these possibilities in the BLM treatment, we have first assessed if BLM injection did significantly modify the number of B cells, CD4+ T cells, CD8+ T cells, NK cells, dendritic cells, monocytes/macrophages or MDSC in spleen of tumor-bearing mice (Figure 3a). We observed that BLM treatment did not change the proportion of all the cell types analyzed. We further tested whether BLM promoted the maturation of antigen presenting cells by testing the effect of a single systemic injection of BLM or lipopolysaccharide (LPS) on the maturation pattern of splenic dendritic cells (DC). While LPS induced as expected a massive upregulation of CD86 and CD40, BLM administration did not affect the expression of these markers, ruling out any direct DCactivating properties of BLM (Figure 3b).
We also checked NK cell activation in tumor draining lymph nodes. We did not observe any NK activation, assessed by CD69 expression, after either PBS, one or five BLM injections (Figure 3c).
Finally, we evaluated the frequency of MDSC in the spleen of BLM-treated tumor-bearing mice. We found that BLM did not affect the number of myeloid or granulocytic MDSC subsets ( Figure 3d) and did not affect MDSC immunosuppressive function, as assessed by MDSC capacity to blunt OT-I capacity to produce IFNc (Figure 3e).
Altogether, these data suggest that BLM does not affect innate immune cells such as dendritic cells, NK cells or MDSC.

BLM Mediates Regulatory T Cell Proliferation via Tumor Cell-derived TGFb
While we detected no change in most immune cell populations (Figure 3a), we observed an accumulation of Foxp3 + Treg in lymphoid organs of mice treated with BLM and bearing CT26 (Figure 4a) or EL4 (not shown) tumors.We then determined the biological function of these Treg. We sorted CD4 + CD25 + (Treg) and CD4 + CD25 2 (conventional T) T cell subsets from the spleen from PBS-or BLM-treated tumor-bearing mice, and tested their capacity to blunt IFNc production by OT-I cells in an in vitro assay. We observed as expected that conventional T cells were unable to blunt OT-I capacity to produce IFNc. In contrast, Treg from untreated or BLM-treated mice had strong and similar capacities to blunt OT-I IFNc production (Figure 4b).
Treg cell proliferation is known to be driven by IL-2 production [28,29] or TGFb [30,31]. We tested the presence of IL-2 and TGFb in ascites of a CT26 peritoneal carcinosis model. While we could not detect any IL-2 in ascites from PBS or BLM-treated mice (not shown), we observed an induction of TGFbproduction in the ascites of BLM-treated animals (Figure4f). In vitro, we also observed that BLM treatment induced TGFb production by CT26 both at the mRNA ( Figure S3) and protein level (Figure4g). Similarly, we observed that BLM could induce tgfb mRNA expression in B16F10, EL4 and MCA205 cells (not shown).
To validate that TGFb was also produced in vivo by tumor cells, we determined the expression of latency-associated protein (LAP), a protein associated with TGFb, in cancer and immune cells. Only tumor cells isolated from subcutaneous tumor showed increased LAP expression after BLM treatment (Figure 4h and Figure S4).
To test if this TGFb production by tumor cells may be responsible for Treg accumulation in BLM-treated mice, we sorted CD4 + GFP + from Foxp3-EGFP mice and stimulated them with T cell receptor (TCR) triggering. In some conditions, supernatant from PBS-or BLM-treated CT26 cells was added, with or without anti-TGFb blocking antibody. After 72 h of stimulation, the proliferation of Treg was assessed by a BrdU incorporation assay. We could observe that supernatant from BLM-treated CT26 induces proliferation of Treg, and that this effect is dependent on TGFb (Figure 4i). In vivo, we also observed that inhibition of TGFb with a blocking anti-TGFb antibody prevents the BLM-induced accumulation of Treg in tumor-bearing mice (Figure 4j).
Taken together, these data strongly suggest that BLM induces Treg accumulation through enhancement of tumor-produced TGFb.

TGFb or Regulatory T cell Depletion Potentiate the Antitumor Effects of BLM
To assess the therapeutic relevance of these observations, we tested the effect of Treg depletion on the antitumor effect of BLM.
In the colon carcinoma model CT26. We first observed that the antitumor effect of BLM was drastically improved by CD4 + T cell depletion (Figure 5a). To validate that the antitumor effect of CD4 depletion is mediated essentially by Treg, we tested the effect of an anti-CD25, which is a membranous marker for mouse Treg, depleting antibody. We observed that CD25 depletion also enhanced the antitumor efficacy of BLM (Figure 5b).
We demonstrated above (Figure4) that BLM induces TGFb production by tumor cells and Treg expansion in tumor bearing mice. To determine if this phenomenon impacts on BLM antitumor effect, we treated tumor-bearing mice with either BLM, anti-TGFb or both. While the anti-TGFb antibody did not significantly impact on tumor growth, TGFb depletion enhanced the antitumor efficacy of BLM (Figure 5c).
Taken together, these data highlight that BLM treatment causes TGFb-dependent proliferation of Treg, which decreases its antitumor efficacy.
was titrated in supernatant of CT26 treated with indicated drugs for 24 h. All data presented are representative of one out of three experiments. B-D, E and F show mean +/2 SEM. doi:10.1371/journal.pone.0065181.g001

Discussion
Our study reveals that BLM can be counted among the immunogenic anticancer drugs which trigger ICD along with oxaliplatin and anthracyclines [7,8,32]. Here, by studying the immune-based effects of BLM anti-cancer treatment, we show that BLM is capable to cause ER and oxidative stress, CRT exposure, HMGB1 release, autophagy and ATP release, providing all the elements required to induce an ICD, leading to IFNc and CD8 + T cell response against tumor cells. Importantly, BLM has a twoedged effect, because BLM also induces the production of TGFb by tumor cells and consequent Treg accumulation which limits its anti-tumor efficacy. These data demonstrate that BLM may have ambivalent immune effects which limit the antitumor efficacy of this anticancer agent. This is similar to our recent conclusions on the immune effects of 5-fluorouracile (5 FU) on tumor immunity. We indeed described that 5 FU induces MDSC death, which restores CD8 + T cell antitumor immunity and causes initial efficacy of chemotherapy [4]. But while MDSC die, they release IL-1b, which induces accumulation of deleterious Th17 cells, promoting later tumor relapse [33]. Together, these observations about 5 FU and BLM underscore the importance to perform a systematic analysis of the immune effect of anticancer agents on the different aspects of the immune response.
Besides, cardiac glycosides, such as digoxin, were recently identified as strong immunogenic cell death inducers, using a high throughput strategy [34]. In this study, it was shown in rodent  models that treatment of tumor cells with digoxin can enhance the effects of non-immunogenic cytotoxic drugs, such as cisplatin [32], due to their capacity to convert non immunogenic to immunogenic cell death. However, the effect of BLM on the immune response was not detected using this strategy. So we believe that systematic analysis of immune properties of chemotherapeutic agents on each immune cell subset is complementary to high throughput strategy.
Foxp3 + CD4 + CD25 + Treg were initially defined as a subpopulation of suppressor T cells that mediate immune tolerance by suppressing autoreactive T cells [35,36]. In human neoplasia, Treg cells accumulate in tumors, draining lymph nodes and blood, and their induction is often correlated with poor outcome both in human and rodent cancers [37]. Activation and expansion of Treg subset was determined to be dependent on IL-2 [28,29] and TGFb [30,31]. BLM treatment of lung fibroblasts results in an increased transcription of tgfb1 and increased TGFb protein [38,39]. This production of TGFb by fibroblasts in the lung after BLM injection is responsible for the induction of lung fibrosis, one of the major side-effects of BLM [40,41]. Here we made the original observation that BLM could also trigger TGFb production by tumor cells. We may suspect that this increase in TGFb expression could result in immunosubversion of anticancer immune response. In particular, we showed that BLM-induced TGFb production by tumor cells drives Treg expansion and may thus inhibit the CD8 + antitumor immune response induced by BLM-driven immunogenic cell death. In the B16F10 model, while either high or low BLM dosage could induce the production of TGFb by B16F10, only high concentration of BLM could induce ICD (not shown). We may therefore raise the hypothesis that in some cases the ICD effect which only occurs with high dosage of BLM is less likely to happen in vivo and that the TGFb-Treg driven induction would be predominant. We also show that Treg depletion or TGFb inhibition significantly enhance BLM anti-tumor efficacy in our in vivo rodent model. Taken together with the current literature, our data suggest that TGFb and Treg are responsible for both limited antitumor effect and major side effect of BLM [40,41]. This may give the rationale to disrupt this phenomenon by combining BLM and Treg inhibitors, such as cyclophosphamide [3]. In the future, the association between BLM and novel drugs like anti-human CD25 (such as daclizumab) [42] or inhibitors of TGFb signaling (such as Smad inhibitors) [43] may be beneficial for both anti-cancer efficacy and side-effect management.
In conclusion, this study supports the ambivalent immunological property of BLM by its capacity to trigger immunogenic cell death and Treg accumulation. One day after treatment, spleens and tumors were collected and the proliferation status of transferred cells was assessed by revealing EdU by flow cytometry. E: CD4 + CD62L + GFP 2 naive T cells were sorted from CD45.2 FOXP3-EGFP mice, and injected i.v. in CD45.1 mice bearing CT26 tumor. The mice then received PBS or BLM treatment. Spleens were harvested and the cells were analyzed through flow cytometry for Treg detection. F: Mice were injected with 1.10 6 CT26 cells i.p. Ten days later mice received PBS or BLM injection. The following day ascites were collected, and TGFb was assessed using ELISA method. G: CT26 cells were treated with PBS or BLM in vitro for 24 h, then the supernatant was collected and assessed for TGFb using ELISA method. H: Mice bearing CT26 tumor cells were treated with PBS or BLM. The day after, CT26 tumors were collected and tumor cells were separated from the tumor infiltrating lymphocytes. Tumor cells were stained for LAP and analyzed by flow cytometry. I: GFP + CD4 + cells were sorted from CD45.2 FOXP3-EGFP mice, then cultured in vitro under TCR-stimulating conditions. We added culture supernatant of CT26 treated for 24 h with PBS of BLM. In some wells, blocking anti-TGFb antibody was added. After three days of culture, the cells were incubated with BrdU for 3 h, and then proceeded to BrdU detection by flow cytometry. J: Same as A, and mice received injection of blocking anti-TGFb antibody. All data presented are representative of one out of two (panels B, D, E, G and I) or three experiments (panels A, F, H and J). Graphs show mean +/2 SEM. doi:10.1371/journal.pone.0065181.g004