Tumour hypoxia promotes melanoma growth and metastasis via High Mobility Group Box-1 and M2-like macrophages

Hypoxia is a hallmark of cancer that is strongly associated with invasion, metastasis, resistance to therapy and poor clinical outcome. Tumour hypoxia affects immune responses and promotes the accumulation of macrophages in the tumour microenvironment. However, the signals linking tumour hypoxia to tumour-associated macrophage recruitment and tumour promotion are incompletely understood. Here we show that the damage-associated molecular pattern High-Mobility Group Box 1 protein (HMGB1) is released by melanoma tumour cells as a consequence of hypoxia and promotes M2-like tumour-associated macrophage accumulation and an IL-10 rich milieu within the tumour. Furthermore, we demonstrate that HMGB1 drives IL-10 production in M2-like macrophages by selectively signalling through the Receptor for Advanced Glycation End products (RAGE). Finally, we show that HMGB1 has an important role in murine B16 melanoma growth and metastasis, whereas in humans its serum concentration is significantly increased in metastatic melanoma. Collectively, our findings identify a mechanism by which hypoxia affects tumour growth and metastasis in melanoma and depict HMGB1 as a potential therapeutic target.

tumour cell proliferation, angiogenesis and metastasis 17,18 . Among the diverse immune cell populations present in the tumour microenvironment, macrophages are the most abundant 19 and are referred as Tumour-Associated Macrophages (TAMs). Macrophages differentiate or polarize in response to inflammatory stimuli and secrete a distinctive set of cytokines and amongst the different types of polarized macrophages described to date, the so-called M1-like or M2-like macrophages appear to have opposing effects on tumour progression 20 . M1-like macrophages are associated with type 1 T-cell responses and anti-tumour immune responses, while M2-like macrophages are reported to display tumour-promoting properties including the production of proteolytic enzymes, the suppression of anti-tumour immune responses and the promotion of angiogenesis 21 . Finally, in advanced tumours, TAMs have been reported to be preferentially skewed towards an M2-like phenotype 22-24 . Given the strong association between tumour hypoxia, tumour progression and poor clinical outcome, deciphering the precise molecular mechanisms by which hypoxia regulates tumour behaviour is of great interest and relevance. Here, we show that hypoxic melanoma cells release the alarmin HMGB1, which promotes tumour growth and metastasis through the accumulation, within tumours, of TAMS bearing an M2-like phenotype.

Results
Serum HMGB1 levels are elevated in metastatic melanoma patients. We first determined whether levels of HMBG1 are altered in cancer patients. To this end, we analysed HMGB1 levels in the serum of patients with primary melanoma, metastatic melanoma and in age-matched healthy volunteers. Significantly, increased levels of HMGB1 were found in the serum of patients with metastatic melanoma when compared to patients with primary melanoma and to healthy donors (Fig. 1a). This observation suggests that the extent of HMGB1 release correlates with the stage of the disease.

Tumour cell hypoxia drives HMGB1 release in vitro and in vivo.
Hypoxia in the context of liver ischemia and hepatocellular carcinoma has been previously reported to induce HMGB1 release 13,25 . To determine if metastatic melanoma cells release HMGB1 in response to hypoxia, we analysed metastatic melanoma cell-lines grown in hypoxic versus normoxic conditions in vitro. When compared to their counterparts grown in normoxic conditions, all tested melanoma cell-lines (n = 7) released enhanced levels of HMGB1 when grown in hypoxic conditions in which increased expression of the hypoxia target gene VEGF-A was also observed 26 (Fig. 1b,c). Notably, under the hypoxic conditions tested no increase in cell mortality was observed as revealed by the consistently low levels of LDH release. The induction of HMGB1 expression by hypoxia in melanoma cells in culture was only significantly induced after 72 hours of hypoxia, but not at shorter exposure times to hypoxia.
Under physiologic conditions, HMGB1 is known to be sequestered in the nucleus, whereas cytoplasmic relocalization of HMGB1 has been reported to occur prior to active secretion or passive HMGB1 release 11 . First, we observed that melanoma metastasis cell lines expressed both cytosolic and nuclear HIF1α when cultured under low oxygen conditions, whereas HIF1α was not detectable when cells were kept in normoxic conditions. The co-labelling with an anti-HMGB1 antibody revealed that HMGB1 is located in the cytosol of HIF1α -positive cells in hypoxic conditions whereas HMGB1 was confined to the nucleus of cells kept in normoxic conditions (Fig. S1). When the intracellular localization of HMGB1 was assessed in tissue sections of nevi, primary melanoma (superficial spreading melanoma) and melanoma metastases, nuclear HMGB1 staining was observed in nevi and primary melanomas, whereas in melanoma metastases large areas of the tumour were composed of cells with cytoplasmic HMGB1 staining (Fig. 2a). Interestingly, the latter areas were also Hif1α -positive indicating that HMGB1 is released within hypoxic areas of metastatic melanomas (Fig. 2a). Co-labelling with an anti-melanosome antibody (clone HMB-45) showed that, in metastasis, HMGB1 is released by HMB-45 + melanoma cells in contrast to primary tumours where HMGB1 was exclusively found in the nucleus of HMB-45 + melanoma cells (Fig. 2b). These observations indicate that HMGB1 is released by hypoxic tumour cells within metastases in melanoma patients. HMGB1 release is partly dependent on HIF1α in cells in hypoxic condition. To assess whether HMGB1 release under hypoxic culture condition was dependent on HIF1α or not, we knocked HIF1α down using siRNA in two human melanoma cell lines. After seventy-two hours under low oxygen conditions, increased levels of stabilized HIF1α were detected in both cell lines, either unmodified or transfected with a control siRNA (and Fig. 3a). In contrast, little or no expression of HIF1α was detected in both cell lines transfected with 2 independent HIF1α sequences (Fig. 3a). Interestingly, a reduced HMGB1 secretion under hypoxia was observed in both cell lines transfected with both HIF1α -siRNA as measure by ELISA (Fig. 3b,c). Under the hypoxic conditions tested, no increase in cell mortality was observed as revealed by the consistently low levels of LDH release and VEGFA was always found to be upregulated (Fig. 3b). Notably, HMGB1 secretion decrease was of only 50%, suggesting that mechanisms, other than HIF1α , are involved (Fig. 3b,c). By inhibiting oxygen-sensitive prolyl hydroxylases (PHDs) with DMOG, we observed that the resulting increased stabilization of HIF1α ( Fig. 3d) was associated with a higher HMGB1 release in melanoma cell lines (Fig. 3e). DMOG exposure of siRNA-transfected melanoma cells did not result in significant increase of HMGB1 release (Fig. 3e), therefore reinforcing the role of the hypoxia/HIF1α axis in HMGB1 secretion. HMGB1 promotes melanoma growth and metastasis. To assess whether HMGB1 plays a functional role in melanoma progression, we diminished HMGB1 expression in B16 melanoma cells using shRNA technology. Four HMGB1-specific shRNA sequences and 2 lamin-specific shRNA sequences were transduced into B16 cells. B16 cells showing the highest downregulation of HMGB1 expression (shRNA sequence 5, Fig. S2a, top panel) were subsequently cloned by limiting dilution and expanded under selection pressure (puromycin). B16 cells transduced with shRNA to lamin showing no variation in HMGB1 expression when compared to wild-type B16 cells (shRNA sequence 2 to lamin, Fig. S2a, top panel) were chosen as controls and were also cloned by limiting dilution and expanded under selection pressure. The silencing of HMGB1 was highly stable during in vitro expansion (Fig. S2b) and after 13 days of in vivo growth (Fig. S2c). The B16 clones transduced with HMGB1-specific (a) HMGB1 was measured by ELISA in the serum of healthy individuals (n = 10), and patients with primary (n = 9) or metastatic melanoma (n = 11). Data are expressed as the mean + /− SEM are presented. * * * P < 0.001 using ANOVA followed by Turkey test. (b) Cell-lines derived from human melanoma metastases were cultured under normoxic or hypoxic conditions for 72 h. As a marker of hypoxia, VEGF-A expression was assessed by qPCR (top panels). Cell viability was assessed in cell culture supernatants with an LDH release assay (middle panels) and HMGB1 release measured by ELISA (bottom panels). (c) Statistical representation and analysis of the data presented in (b). Data expressed as the mean +/− SEM are presented. A paired Student t-test was performed. * P < 0.05, * * P < 0.01. (a) Immunofluorescence labelling, using antibodies to HIF1α (red) and HMGB1 (green), was performed on healthy skin, nevi, primary cutaneous melanoma and metastases as indicated (n = 5/group; representative pictures are presented). (b) Immunofluorescence labelling using antibodies to human melanosome (clone HMB-45, red) and HMGB1 (green) was performed on healthy skin, nevi, primary cutaneous melanoma and metastases as indicated (n = 5/group; representative pictures are presented). As an hypoxia target gene, VEGF-A expression was assessed by qPCR (bottom panels). Cell viability was assessed in cell culture supernatants with an LDH release assay (middle panels) and HMGB1 release measured by ELISA (top panel). (c) HMGB1 release summary and statistical analysis of the data presented for individual cell lines in (b). Data expressed as the mean + /− SEM are presented. A paired Student t-test was performed. * * * * P < 0.0001. (d) HIF1α expression was assessed by Western blotting in two human metastatic melanoma cell lines (MM130421 and MM131112) transfected with 2 sequences of siRNA to HIF1α and exposed to DMOG or incubated for 72 hrs in normoxic (Nor, N) or hypoxic conditions (Hyp, H). (e) HMGB1 release was measured by ELISA (top panels). Cell viability was assessed in cell culture supernatants with an LDH release assay (middle panels) and VEGF-A expression was assessed by qPCR (bottom panels).
shRNA exhibiting the highest HMGB1 knockdown (Fig. S2a, bottom panel) and the clones transduced with lamin-shRNA showing similar HMGB1 expression when compared to wild-type B16 were then injected s.c. in C57BL/6 mice. Following sub-cutaneous injection, mice implanted with HMGB1-shRNA-transduced B16 clones exhibited significantly reduced tumour growth when compared to mice implanted with the same number of lamin-shRNA-transduced B16 clones (Fig. 4a). Importantly, the transduction of lamin-shRNA did not alter the in vivo growth kinetics of B16 tumours when compared to wild-type B16 cells (Fig. S3). Since such a reduced growth was reproducibly obtained irrespective of the shRNA sequence and clonal selection of B16 cells, all experiments described below were performed with one B16 clone transduced with HMGB1-specific shRNA and one B16 clone transduced with lamin-specific shRNA (HMGB1 clone 17 (sequence 5) and lamin clone 1 (sequence 2), respectively). This growth delay was neither due to an intrinsic effect of shRNA transduction or HMGB1 knockdown on growth or apoptosis as revealed by the identical in vitro growth and apoptosis of HMGB1-and lamin-shRNA-transduced B16 cells (Fig. S4). To further validate the above observation and assess the contribution of released extracellular HMGB1 on tumour growth, we next treated mice implanted subcutaneously with lamin-shRNA-transduced B16 cells with either a recombinant HMGB1 inhibitor (BoxA, 50 μ g i.p. every 3 days) or vehicle (PBS) in an independent set of in vivo experiments. In accordance with results observed with HMGB1 silencing using shRNA, systemic treatment of tumour-bearing mice with the HMGB1 inhibitor BoxA also resulted in significantly delayed tumour growth compared to mice exposed to vehicle alone (Fig. 4b).
To determine whether HMGB1 release in this mouse model of melanoma was reminiscent of what was observed in human melanoma metastases, we assessed the intra-cellular localization of HMGB1 in tumours of mice implanted s.c. with lamin-shRNA-transduced B16 cells. Large areas where HIF was stabilized were found mainly in the centre of B16 tumours, as revealed by immunocytochemistry using a HIF1α antibody (Fig. 5a). Within HIF1α -positive tumour areas, large numbers of tumour cells displaying cytoplasmic HMGB1 expression were observed and double-labelling experiments identified numerous cells co-expressing HMGB1 and HIF1α (Fig. 5b). In contrast, nuclear HMGB1 localisation was predominantly observed in HIF-1α -negative areas of the tumour, and no evident co-labelling of anti-HMGB1 and HIF1α antibodies was observed in such areas. These observations are in line with data from human melanomas. Altogether, these results suggest that HMGB1 released by hypoxic tumour cells promotes melanoma growth and metastasis.
HMGB1 promotes the accumulation of tumour-associated M2-like macrophages. To further investigate the tumour-promoting effects of HMGB1 release, as a consequence of tumour hypoxia and given the absence of evidence for a direct effect of HMGB1 on tumour cell growth or apoptosis in vitro, we analysed the immune cell infiltrate within the tumour microenvironment. Analysis of dissociated tumours by flow-cytometry revealed a significant increase in the total number of TAMs and a slight but significant decrease in the total number of neutrophils in HMGB1-shRNA-transduced B16 tumours (Fig. 6a). To assess the phenotype of these TAMs, we analysed the in vivo expression of markers discriminating M1-like and M2-like macrophage subpopulations 27 . Quantitative PCR of macrophages accumulating at the tumour site revealed that the downregulation of HMGB1 in B16 cells was associated with a significant induction of the expression of the M1 macrophage marker CD80 28 , whereas lamin-shRNA-transduced B16 tumours were associated with a significant upregulation of the M2 markers YM1, Fizz1, and IL-10 (Fig. 6b). These results suggest that HMGB1 expression and release within tumours favours the accumulation of M2-like macrophages in the microenvironment. HMGB1 induces IL-10 in M2-like macrophages through RAGE. Given the reported role of the cytokine milieu of the tumour microenvironment on tumour progression 29,30 , we evaluated the effect of HMGB1 on TAM cytokine expression. While HMGB1 had no effect on IL-6, TNF or IL-1β expression, it significantly increased IL-10 expression in bone marrow-derived M2-like macrophages (BMM2, Fig. 7a). Furthermore, HMGB1-induced upregulation of IL-10 expression was not observed in BMM2 from RAGE −/− mice while it was retained in TLR2-and TLR4-deficient BMM2 (Fig. 7b), suggesting that HMGB1 induces IL-10 in TAMs through RAGE-dependent signalling. To assess the effect of IL-10 on tumour development in the model used herein, mice implanted with lamin-shRNA transduced B16 tumours were treated with an anti-IL-10 neutralizing antibody. Similar to mice in which tumour expression of HMGB1 was silenced or inhibited with recombinant BoxA, mice treated with anti-IL-10 exhibited significantly delayed tumour growth compared to controls, whereas anti-IL-10 did not significantly affect the growth of HMGB1-shRNA-transduced B16 tumours (Fig. 7c). Collectively, these results indicate that HMGB1-dependent production of IL-10 by tumour-associated M2-like macrophages contributes to tumour progression in our mouse melanoma model. In further support for a potential role of TAMs and IL-10 in melanoma is the presence of IL-10-producing TAMs in human melanoma metastases as revealed by the presence of CD163 + IL-10 + cells infiltrating human metastases in areas with high cytoplasmic HMGB1 expression (Fig. 7d). In contrast, lower expression and nuclear localisation of HMGB1 in nevi was associated with a very discrete presence of CD163 + cells and an absence of IL-10 production. Taken together, these results are supportive of a tumour-promoting role for HMGB1 dependent on its ability to induce IL-10 secretion by M2-like macrophages in a RAGE-dependent manner.

Discussion
Tumour-associated macrophages (TAMs) have emerged as key components of the tumour microenvironment with a crucial role in tumour progression 31 . TAMs are reported to be associated with poor prognosis in several types of cancer [32][33][34][35] , and it has been proposed that monocytes, which continuously infiltrate tumours, once polarized to M2-like macrophages preferentially accumulate in hypoxic tumour areas [36][37][38][39][40] . By promoting angiogenesis and metastasis, M2-polarized TAMs can "assist" the tumour in overcoming a hostile hypoxic environment and thus sustain its progression 24,41 . Therapeutic targeting of this process in cancer is currently limited by an inadequate understanding of factors released by hypoxic cells that may be associated with M2-like macrophage accumulation in the tumour microenvironment. Here, we show that the alarmin HMGB1, which is released from tumour cells under hypoxic conditions, plays a critical role in promoting tumour progression by triggering the accumulation of M2-like TAMs. We demonstrate that HMGB1 released by hypoxic tumour cells favours the accumulation of M2-like macrophages at the tumour site and their secretion of IL-10. The relevance of the in vitro and in vivo data generated in this mouse model of melanoma is substantiated by the observation that HMGB1 release also occurs in human melanoma cell-lines under hypoxic conditions in vitro, particularly in Hif-1α -positive areas within human melanoma metastases and by the fact that melanoma metastases but not benign melanocytic nevi are infiltrated by CD163 + IL-10 + TAMs.
Areas of hypoxia in tumours have been reported to be associated with advanced stage malignancy and resistance to therapy 16 . Whether hypoxia and its by-products directly induce an M2-like phenotype in TAMs remains unclear. Our data suggest that, although HMGB1 clearly contributes to the accumulation of M2-like macrophages within tumours, it does not appear to be directly involved in M2 polarization. Indeed, and in line with a previous report showing that hypoxia is not a driver of the differentiation of TAMs 37 , we were not able to differentiate macrophages towards an M2-like phenotype in vitro using recombinant HMGB1 alone. However, a recent study showed that lactic acid, produced by tumour cells as a by-product of hypoxic glycolysis, has a critical function in tumour development by inducing VEGF expression and M2-like polarization of TAMs in a HIF1α -dependent manner 42 . Furthermore, oncostatin M and eotaxin have been suggested to promote breast cancer metastasis by favouring M2 polarization and tumour infiltration 43 . Recent evidence also suggests that HMGB1 may directly act on progenitor cells to favour the induction of myeloid-derived suppressive cells 44 .
Mechanistically, hypoxia-induced HMGB1 release seems to be associated with several tumour-promoting events. In advanced hepatocarcinomas, hypoxia has also been shown to be associated with the release of HMGB1 45 , and it has been suggested that HMGB1, via RAGE, induces the expression of NF-κ B-dependent pro-angiogenic factors including VEGF 46 and the matrix metalloproteinases MMP2 and MMP9 47 . It has also been observed that HMGB1 released from dying cells in prostate cancer induces the accumulation of tumour-infiltrating T cells and the expression of lymphotoxin-α 1β 2 on their surface, which in turn recruits macrophages to the tumour and supports angiogenesis 48 . Noteworthy, tumour cells or tumour-infiltrating immune cells seem not to be the sole source of HMGB1. Indeed, UVB radiation of the skin has been shown to induce HMGB1 release from epidermal keratinocytes, resulting in a neutrophilic inflammatory response that stimulate angiogenesis and promotes melanoma metastasis in mice 49 .
We also observed that HMGB1 directly induced the production of IL-10 in TAMs and that blockade of IL-10 with a neutralizing antibody led to delayed tumour growth in the B16 mouse melanoma model. As we cannot, at present, technically and specifically delete IL-10 in TAMs, the relative role of HMGB1-induced IL-10 in TAMs remains incompletely elucidated in our model. It is known that regulatory T cell-mediated/IL-10-dependent suppression of CD8 + T cells can be blocked by removal of tumour-derived HMGB1 50 , which is consistent with our observation that HMGB1 inhibition leads to delayed tumour growth although via an alternate mechanism. It is likely that IL-10, which can also be produced by melanoma cells 51 and tumour-associated myeloid-derived suppressor cells 44 may favour immunoregulatory responses by inducing the downregulation of molecules involved in antigen presentation to CD8 + T cells 52 , by inducing regulatory T cells 53,54 and/or by supressing the production of pro-inflammatory cytokines including TNFα , IFN-γ and IL-2 by T cells 55 . In accordance with this, elevated IL-10 production levels in melanoma patients are associated with poor prognosis 56 .
HMGB1 can signal by binding to TLR2, -4 and -9 as well as RAGE. According to our data, TLR signalling was not required for HMGB1-dependent induction of IL-10, the latter being selectively dependent on RAGE signalling. The potential importance of the HMGB1-RAGE interaction in promoting tumour progression is supported by a recent report showing that RAGE and HMGB1 are associated with the progression of prostate cancer and poor patient outcome 57 . However, since our data shows that HMGB1 downregulation or the use of a soluble inhibitor did not lead to complete inhibition of tumour growth, it is very likely that signalling by other alarmins through TLRs or RAGE also have the ability to induce a tumour-promoting microenvironment 58 . In addition, we do not exclude an incomplete inhibition of the effect of HMGB1 in our model or other HMGB1-independent mechanisms of tumourigenesis.
In conclusion, we demonstrate that HMGB1, derived from hypoxic tumour cells, significantly contributes to melanoma progression by favouring the accumulation of IL-10-secreting TAMs within the tumour. Tumour-derived HMGB1 released as a consequence of focal intra-tumoural hypoxia thus directly contributes to tumour progression and likely represents an attractive therapeutic target for tumour therapy as demonstrated here in the case of melanoma.

Biological samples from melanoma patients and healthy donors. Serum and tumour biopsies were
collected from patients with primary melanoma or metastatic melanoma (stage IIIB to IV) in the Department of Dermatology of the University Hospital of Zürich. Characteristics of patients with primary and metastatic melanoma are reported in Tables 1 and 2, respectively. Serum was obtained from healthy blood donors and healthy skin was obtained as excess skin resulting from aesthetic/reconstructive procedures in the plastic surgery unit at the University Hospital of Zürich. All human biological samples were collected after written informed consent of the patient and with approval of Local Ethics Committee (Kantonale Ethikkommission Zürich, KEK-ZH authorization Nr. 2014-0425) in accordance to GCP guidelines and the Declaration of Helsinki.
Mice. Six to 8-week-old female C57BL/6 mice (Harlan, Venray, Netherlands) were used in this study. Germany). All experimental procedures were approved by the Veterinary Office of Zürich and the institutional animal care officer and were carried out in accordance to the approved guidelines.
Generation of HMGB1-shRNA-transduced B16. B16-F10 mouse melanoma cell-lines stably expressing shRNA specific to Lamin or HMGB1 were generated by transducing B16-F10 cells with a lentiviral vector. Briefly, specific shRNA were generated by inserting oligonucleotides targeting Lamin or HMGB1 into the pSU-PER vector (Oligoengine, Seattle, WA) and subsequent cloning into the lentiviral vector pSP-93 (Oligoengine). Second-generation packaging plasmids pMD2-VSVG and psPAX2 (kindly provided by Prof. J. Tschopp, Biochemistry Institute, Lausanne, Switzerland) were used for lentivirus production and infection. For each target molecule, the procedure was performed with different shRNA sequences. ShRNA-transduced B16 cells were subsequently cloned by limiting dilution under selection pressure (puromycin). For each shRNA, one clone was chosen according to knock-down efficiency and stability as well as in vitro properties (Figs S1 and S3).

Detection of HMGB1 by ELISA.
To determine the concentration of HMGB1 in the serum of healthy donors, primary melanoma patients and melanoma patients with metastasis, serum samples were analysed using the HMGB1 ELISA kit (IBL International, Hamburg, Germany) according to the manufacturers' instructions.
To determine the concentration of released HMGB1 from in vitro cell cultures, supernatants were centrifuged at 1,500 g for 5 min at 4 °C and analysed using the HMGB1 ELISA kit (IBL International) according to the manufacturers' instructions.

In vitro differentiation of M1-and M2-like macrophages from bone marrow cells. Recombinant
cytokines were all from Peprotech (Rocky Hill, NJ). To generate of M1 and M2 macrophages, bone marrow cells from the tibia and fibula of C57BL/6 mice (Harlan) were cultured at 37 °C in 5% CO 2 in RPMI supplemented with 1% L-glutamine and 10% foetal bovine serum with 10 ng/ml mouse M-CSF. Medium was replaced on days 3 and 6 and cells were harvested on day 8. For M1 phenotype induction, cells were stimulated for 24 h with 10 ng/ ml M-CSF and 100 ng/ml IFN-γ and for an additional 24 h with 10 ng/ml M-CSF and 20 ng/ml ultra-pure LPS (Invitrogen, Carlsbad, CA). For M2 phenotype induction, cells were stimulated twice for 24 h with 10 ng/ml M-CSF and 20 ng/ml IL-4. To determine the effect of HMGB1 on M1 and M2 macrophages, culture medium used to induce M1-or M2-like macrophages was supplemented with 1 μ g/ml recombinant HMGB1 (HMGBiotech).
Statistical analyses. Unless otherwise indicated, data are presented as the means ± standard error of the mean (SEM) and are representative of three independent experiments. Statistical analyses were performed using the Prism Software (GraphPad Software, San Diego, CA). P-values were calculated with paired or unpaired Student's t test. Where indicated, statistical analyses where performed using ANOVA test followed by a Tukey test. Differences were considered significant when: * P ≤ 0.05, * * P ≤ 0.01, * * * P ≤ 0.001 and * * * * P ≤ 0.0001.