Despite impressive clinical success, cancer immunotherapy based on immune checkpoint blockade remains ineffective in many patients due to tumoral resistance. Here we use the autochthonous TiRP melanoma model, which recapitulates the tumoral resistance signature observed in human melanomas. TiRP tumors resist immunotherapy based on checkpoint blockade, cancer vaccines or adoptive T-cell therapy. TiRP tumors recruit and activate tumor-specific CD8+ T cells, but these cells then undergo apoptosis. This does not occur with isogenic transplanted tumors, which are rejected after adoptive T-cell therapy. Apoptosis of tumor-infiltrating lymphocytes can be prevented by interrupting the Fas/Fas-ligand axis, and is triggered by polymorphonuclear-myeloid-derived suppressor cells, which express high levels of Fas-ligand and are enriched in TiRP tumors. Blocking Fas-ligand increases the anti-tumor efficacy of adoptive T-cell therapy in TiRP tumors, and increases the efficacy of checkpoint blockade in transplanted tumors. Therefore, tumor-infiltrating lymphocytes apoptosis is a relevant mechanism of immunotherapy resistance, which could be blocked by interfering with the Fas/Fas-ligand pathway.
Inhibitory antibodies against immune checkpoint molecules CTLA4 and PD1 induce durable tumor responses in a number of cancer patients, and have become standard of care in a number of metastatic malignancies. Yet clinical benefit of anti-CTLA4 and anti-PD1 remains limited to a fraction of patients and the priority is to understand why the majority of patients fail to respond. There is mounting evidence indicating that immunotherapy resistance is largely dependent on the tumor microenvironment, whose immunosuppressive nature is progressively shaped during the long-term process of tumor development in an immunocompetent host. There are numerous cellular and molecular mechanisms at play, and the challenge is to define those that are clinically relevant. This is traditionally based on preclinical studies, which often rely on murine transplanted tumor models. However, transplanted tumors do not recapitulate the tumor microenvironment as it progressively develops during the growth of an autochthonous tumor. This is better modeled using genetically engineered mouse models (GEMM), in which autochthonous tumors develop following the induction of oncogenic events within host tissues. However, most available GEMM models either do not express defined tumor antigens, precluding in-depth analysis of the anti-tumor immune responses in the course of immunotherapy, or express model antigens such as ovalbumin or viral antigens, which are highly immunogenic and do not reflect the poor immunogenicity of tumor antigens that are naturally expressed on human tumors. To circumvent these issues, we created the GEMM model of inducible melanomas expressing P1A, a defined mouse tumor antigen of the MAGE type55, which we chose as the best representative of the clinically relevant group of human MAGE-type tumor antigens encoded by cancer-germline genes1. This model, named TiRP, is based on the tamoxifen-induced and Cre-lox-mediated induction of H-RasG12V and deletion of Ink4a/Arf in melanocytes2. The H-RasG12V transgene is followed by an internal ribosome entry site (IRES) and the P1A coding sequence (Trap1a), so that all melanomas express P1A and are recognized by P1A-specific CD8+ T cells directed against the dominant P1A epitope, a nine-amino acid peptide presented by H-2Ld2. To ensure H-2Ld presentation, we backcrossed the model from B6 to B10.D2 mice, which are B6 congenic for H-2d. This TiRP model now gives a high incidence of melanomas (±95%) with a short latency (30–60 days depending on the tamoxifen batch). Induced TiRP melanomas are initially well differentiated and highly pigmented (Mela tumors), but then rapidly undergo dedifferentiation resulting in white aggressive tumors (Amela), characterized by Epithelial-to-Mesenchymal Transition (EMT)-like and TGFβ signatures3. This mesenchymal transition is accompanied by abnormal inflammation, involving recruitment of immature myeloid cells, which appear to impair anti-tumor immunity4. The Amela tumor signature shares many genes with the innate anti-PD1 resistance (IPRES) signature that was recently described in human melanomas resisting anti-PD1 therapy, including EMT-like genes (Axl, Twist2, Loxl2), angiogenesis genes (Vegfa) and monocyte/macrophage recruitment genes (Ccl2, Ccl7)3, 5. The TiRP model therefore provides an ideal platform to study immunotherapy resistance in a manner that is more relevant than transplanted tumors and recapitulates human tumors resisting immunotherapy.
In this report, we confirm that TiRP tumors are resistant to different forms of immunotherapy, and we link this resistance to the ability of TiRP tumors to induce tumor-infiltrating lymphocytes (TIL) apoptosis. Anti-tumor T cells recruited to TiRP tumors quickly undergo apoptosis and disappear. Therefore, they fail to reject the tumor and to persist. This happens with autochthonous TiRP tumors, but not with isogenic transplanted tumors. TIL apoptosis is triggered by Fas-ligand, which is expressed by polymorphonuclear myeloid-derived suppressor cells (PMN-MDSC). The latter are enriched in induced TiRP tumors as compared to transplanted isogenic tumors. Administration of soluble Fas-Fc can prevent TIL apoptosis and improve the efficacy of immunotherapy.
Induced TiRP melanomas resist immunotherapy
Immunotherapy based on checkpoint inhibitors anti-CTLA4 and anti-PD1 proved unable to alter the growth of tamoxifen-induced TiRP tumors, whether given alone or in combination (Fig. 1a). Because TiRP tumors express MAGE-type tumor antigen P1A, we also tested a preventive vaccination against P1A using an established prime/boost immunization scheme based on P1A-recombinant adenovirus and Semliki Forest virus (SFV), which induces high levels of P1A-specific CD8+ T cells and significant protection against a challenge injection of P1A-positive tumor cells6. However, immunization of TiRP mice before tumor induction with tamoxifen failed to prevent tumor onset, and this was not improved upon combination of vaccine with anti-CTLA4 and anti-PD1 (Fig. 1a). This was seemingly not due to loss of antigen expression, as the tumors that appeared in vaccinated mice still expressed high amounts of P1A transcripts (Supplementary Fig. 1). These results suggested a strong immunosuppressive mechanism preventing T-cell attack of TiRP tumors.
We then used H-2Ld-P1A tetramers to monitor P1A-specific CD8+ T cells in the spleen of vaccinated mice stimulated one week in vitro with the P1A peptide (LPYLGWLVF) (Fig. 1b)49. All vaccinated mice developed an anti-P1A CD8+ response. In control mice without tumor, this response persisted until at least 70 days after vaccination. However, in mice developing a TiRP tumor, the P1A-specific CD8+ T-cell response plummeted to the levels of unimmunized mice at day 70. Hence, the presence of a TiRP tumor induced complete disappearance of the P1A-specific anti-tumor CD8+ response. We also observed that non-vaccinated mice developing a TiRP tumor spontaneously mounted a P1A-specific immune response, confirming that TiRP tumors express the P1A antigen and indicating that the tumor is immunogenic. However, as observed in vaccinated mice, this anti-tumor immune response did not persist. This lack of persistence was restricted to anti-tumor CD8+ T cells, because when we immunized mice against both P1A and an irrelevant antigen named P91A (an H-2Ld-restricted mutated peptide7), we observed that the CD8+ response against the latter antigen persisted in tumor-bearing mice (Fig. 1c). Moreover, tumor-bearing mice were able to mount a normal primary CTL response against minor histocompatibility antigens (Supplementary Fig. 2), indicating that tumor development did not induce a general CTL unresponsiveness, as reported previously in a spontaneous sporadic tumor mouse model8, 9. In the latter model, increased levels of TGFβ1 were measured in the serum of tumor-bearing mice, and correlated with the general T-cell unresponsiveness observed after a long latency period8, 9. Therefore, we also measured TGFβ1 levels in the serum of tumor-bearing TiRP mice (Supplementary Fig. 3). Total TGFβ1 levels were high in both groups and increased in tumor-bearing mice as compared to tumor-free mice. However, the levels of active TGFβ1 were low and similar in both groups, indicating that most of the seric TGFβ1 is in its latent, inactive form. Altogether, our data do not suggest a lack of general T-cell responsiveness, but rather indicate a selective suppression of tumor-specific CTL in tumor-bearing TiRP mice.
Induced but not transplanted tumors resist adoptive cell therapy
To better track the fate of tumor-specific CD8+ T cells in this model, and to evaluate another clinically relevant immunotherapy approach, we resorted to adoptive transfer of P1A-specific CD8+ T cells isolated from mice transgenic for the anti-P1A T-cell receptor (TCR) (TCRP1A mice in the Rag1−/− B10.D2 background)10. Mice bearing induced Amela TiRP tumors received an intravenous injection of 10 million CD8+ T cells isolated from TCRP1A mice and stimulated 4 days in vitro (Fig. 2a). Strikingly, this adoptive cell therapy (ACT) was unable to alter the growth of induced TiRP tumors (Fig. 2b). In parallel, we used tumor cell line T429, which was previously established from an induced Amela TiRP tumor and grown in vitro4. We established isogenic transplanted tumors by subcutaneous injection of cells from a T429 clone, and we treated these mice with ACT as above. In contrast to induced tumors, those transplanted tumors were efficiently rejected by adoptively transferred TCRP1A CD8+ T cells (Fig. 2c). Similar results were obtained with distinct clones from line T429 and with tumors transplanted in the subcutaneous (Fig. 2d) or the intradermal space, which is the natural niche for melanoma development (Fig. 2e). Because of the identical genetic background of the tumor cells, this striking difference between induced and transplanted tumors likely resulted from differences in the tumor microenvironment, and underlines the relevance of autochthonous tumor models as opposed to transplanted tumor models for preclinical evaluation of cancer immunotherapy.
We then analyzed the persistence of TCRP1A CD8+ T cells in the spleen and lymph nodes of mice 22 days after ACT, using H-2Ld-P1A tetramers ex vivo. P1A-specific CTL were easily detected in tumor-free mice and in mice bearing transplanted tumors. However, they were undetectable in mice bearing an induced tumor (Fig. 2f). We then used TCR-specific clonotypic PCR as a more sensitive approach, but we also failed to detect TCRP1A T cells after 22 days in the spleen of mice bearing induced tumors, while we easily detected them in tumor-free and tumor-transplanted mice (Supplementary Fig. 4a). At this time point, TCRP1A T cells were also absent in most of the induced tumors themselves (Supplementary Fig. 4a). These results indicated that tumor-specific T cells were actively deleted in mice bearing induced Amela TiRP tumors, while they were not in mice bearing transplanted tumors.
Induced tumors trigger apoptosis of tumor-specific T cells
To understand the fate of tumor-specific T cells, we then analyzed mice earlier, 4 days after ACT. At this time, both induced and transplanted tumors were infiltrated by CD8+ T cells (Fig. 3a), and transferred TCRP1A CD8+ T cells were easily detected in induced tumors by clonotypic PCR (Supplementary Fig. 4b). Ex vivo tetramer analysis further showed that transferred T cells were enriched in the tumors as compared to the spleen and draining lymph nodes, in both induced and transplanted tumors (Fig. 3b). These results indicated that transferred TCRP1A CD8+ T cells effectively migrated into the induced tumors in the first days after adoptive transfer. Furthermore, those tumor-infiltrating T cells expressed activation marker CD69, indicating that they recognized the P1A tumor antigen on tumor cells (Fig. 3c). To further evaluate the functional capacity of transferred T cells in the early days after ACT, we tested their ability to kill P1A-peptide pulsed target cells in an in vivo killing assay performed 3 days after transfer (Supplementary Fig. 5). We observed efficient killing of P1A-positive targets in mice bearing induced tumors, even though it was slightly reduced as compared with tumor-free mice. Thus, the results so far indicated that transferred TCRP1A CD8+ T cells retained their functional capacity, infiltrated the induced tumors, recognized their antigen and became activated. This raised the question: why did they fail to reject the induced tumors?
We then observed that 4 days after adoptive transfer, a high proportion of P1A-specific CD8+ T cells infiltrating the induced tumors were apoptotic, as defined by staining with Annexin V (Fig. 3d). This was not the case in transplanted tumors (Fig. 3d). The non P1A-specific CD8+ T cells that infiltrated induced tumors were not stained with Annexin V (Fig. 3e). This suggested that induced tumors resisted immune rejection by triggering apoptosis of tumor-specific T cells. Because of the high background staining observed in spleen and lymph nodes with Annexin V, we used a series of other approaches to reveal apoptosis. 4 days after T-cell transfer, we injected mice with FLIVO, a fluorescent inhibitor-based pan-caspase marker that selectively binds active caspases and was developed for in vivo use11. After 4 h, we killed the mice and analyzed TIL by FACS after staining with H-2Ld-P1A tetramer. We observed that approximately 50% of P1A-specific TIL were stained with FLIVO in induced tumors, while only about 10% were in transplanted tumors (Fig. 3f). We also analyzed sections of those tumors by immunofluorescence, monitoring cells positive for FLIVO and for CellTracker Blue CMAC Dye, which was used to stain T cells before adoptive transfer. This confirmed the higher proportion of apoptotic TCRP1A CD8+ T cells in induced tumors (Supplementary Fig. 6a). Similar results were obtained with tumors from mice adoptively transferred but not treated with FLIVO, which were stained with antibodies to cleaved caspase-3 and to V-alpha8, the alpha chain of the TCR used in TCRP1A CD8+ T cells (Supplementary Fig. 6b, c). Lastly, we tried to recapitulate apoptosis induction by collecting tumors ex vivo and incubating freshly cut tumor slices with CMAC-labeled TCRP1A CD8+ T cells for 24 h. We then stained cryosections of these tumor slices with FLICA, a fluorescent inhibitor-based marker of activated pan-caspases, and again we observed more apoptotic TCRP1A CD8+ T cells in induced as compared with transplanted tumors (Fig. 3g).
We then performed ACT in pigmented (Mela) and unpigmented (Amela) induced TiRP tumors, and compared P1A-specific CD8+ TIL 4 days after transfer. We observed equal T-cell infiltration and activation, but T-cell apoptosis was restricted to Amela tumors, suggesting that the ability to induce T-cell apoptosis was linked to the inflammatory microenvironment that is typical of Amela tumors in this model (Fig. 3h)4.
TIL apoptosis is dependent on Fas-ligand
We then tried to understand what triggered apoptosis of anti-tumor T cells in induced tumors. Interferon-gamma (IFNγ) can play a dual role in the course of the immune response, contributing to an efficient response by promoting effector T-cell differentiation and MHC class I expression on the one hand, but also, on the other hand, inducing inhibitory molecules that contribute to the negative feedback of the immune response, including IDO, FasL and PDL112,13,14,15. Four days after T-cell transfer, induced tumors contained high levels of IFNγ mRNA and secreted protein, confirming the functionality of transferred T cells (Fig. 4a, b). To determine whether this IFNγ played a role in the induction of T-cell apoptosis, we injected an IFNγ-neutralizing antibody into mice bearing induced tumors 1 day before ACT. Four days after ACT, we observed a strong reduction in the number of apoptotic TILs, indicating the involvement of IFNγ in triggering T-cell apoptosis at the tumor site (Fig. 4c).
A number of genes induced by IFNγ encode proteins that can trigger lymphocyte apoptosis. These include IDO12, 15, PDL112, 16 and Fas-ligand17. IDO was only weakly expressed in this model, and only barely increased after T-cell transfer. In contrast, PDL1 was well expressed in this model4. However, it was not involved in TIL apoptosis because the injection of an anti-PD1 blocking antibody failed to prevent apoptosis of transferred T cells (Supplementary Fig. 7a). This was in line with the lack of therapeutic effect of PD1 blockade in this model (Fig. 1a). Expression of Fas-ligand (FasL) transcripts was observed in TiRP tumors, increased after T-cell transfer, and was dramatically reduced after IFNγ neutralization (Fig. 4d). Moreover, FasL expression was higher in induced TiRP tumors as compared to transplanted T429 tumors (Fig. 4e). We therefore investigated the Fas-FasL pathway as potentially involved in triggering apoptosis of tumor-infiltrating T cells. We first used siRNA to silence Fas in TCRP1A CD8+ T cells and make them insensitive to FasL-mediated apoptosis, and we optimized the conditions so that Fas silencing was effective for at least 7 days (Supplementary Fig. 8). We then transferred Fas-silenced TCRP1A CD8+ T cells into mice bearing induced TiRP tumors, and analyzed TIL apoptosis after 4 days, by FACS staining with Annexin V and for activated caspases (FLICA). We observed a strong reduction of apoptosis of Fas-silenced TIL (Fig. 5a). We obtained similar results when we incubated Fas-silenced TCRP1A CD8+ T cells 24 h in vitro with fresh slices of induced tumors (Fig. 5b). To confirm this result, we then used soluble Fas-Fc to neutralize FasL in vivo. We injected mice bearing induced tumors with 150 µg Fas-Fc every day starting 1 day before ACT. Again we observed a clear reduction in the number of apoptotic TILs after 4 days (Fig. 5c). Fas-Fc also reduced apoptosis of TCRP1A CD8+ T cells incubated in vitro with fresh slices of induced tumors (Fig. 5d). We conclude that the apoptosis of anti-tumor CD8+ T cells observed in induced tumors is mostly driven by FasL. Tumor necrosis factor alpha (TNFα) is another factor known to induce T-cell apoptosis18. Even though TNFα is not clearly induced by IFNγ, we tested its involvement in T-cell apoptosis in the TiRP model by injecting an anti-TNFα neutralizing antibody 1 day before and 2 days after ACT. We observed no reduction of T-cell apoptosis on day 4 after ACT (Supplementary Fig. 7b). These results supported a dominant role of FasL in triggering T-cell apoptosis in the TiRP model.
PMN-MDSC trigger TIL apoptosis
We then compared the cellular composition of the tumor tissue of induced (Amela) and transplanted tumors. Both tumor types contained CD45+ cells (hematopoietic origin), which were more abundant in induced tumors. This likely resulted from a significant enrichment of induced tumors in polymorphonuclear-myeloid-derived suppressor cells (PMN-MDSC) (P < 0.0001, t-test), while monocytic myeloid-derived suppressor cells (M-MDSC) were equally represented in both tumor types (Fig. 6a). MDSC are a subset of immature myeloid cells that can be recruited at tumor sites and display immunosuppressive activity19. This PMN-MDSC enrichment was observed in Amela but not in Mela tumors (Fig. 6b). As compared with M-MDSC, PMN-MDSC produced more ROS, expressed less iNOS and produced less NO, in line with previous reports20, 21 (Supplementary Fig. 9). They also expressed higher levels of arginase, another immunosuppressive factor that is often expressed in M-MDSC but also in PMN-MDSC20,21,22,23. We confirmed the immunosuppressive function of these PMN-MDSC by showing their ability to suppress in vitro proliferation and cytolytic activity of anti-tumor CD8+ T cells at a 1/1 ratio (Fig. 6c, b). Moreover, when we established transplanted tumors by co-injecting T429.11 tumor cells with PMN-MDSC (isolated from induced tumors) at a 4 to 1 ratio and then treated mice with ACT, the co-injected tumors resisted immune rejection, while control tumors were rejected (Fig. 6e). These results confirmed the key role of PMN-MDSC in resistance of tumors to immunotherapy in this model. To explain the higher recruitment of PMN-MDSC in induced as compared with transplanted tumors, we measured the expression of a series of cytokines and chemokines known to play a role in the recruitment or differentiation of MDSC24, 25. We found that most of them were expressed at a higher level in induced tumors as compared with transplanted tumors (Fig. 6f, g). Among these factors, Csf3, Cxcl1 and Cxcl5 are known to specifically recruit PMN-MDSC24,25,26. Although Ccl2, Cxcl12 and TGFβ are mostly known for their ability to recruit M-MDSC, they can also promote accumulation of PMN-MDSC in some tumor settings21, 24,25,26. In addition, it has been suggested that M-MDSC recruited at the tumor site can be subsequently converted into PMN-MDSC in the tumor microenvironment27. These results could therefore explain the accumulation of PMN-MDSC in induced Amela TiRP tumors, and corroborate our previous observations of increased expression of such factors, including Ccl2, Cxcl5 and Ccl7, in Amela TiRP tumors as compared to Mela TiRP tumors, which recruit much less MDSC3, 4. Interestingly, overexpression of CCL2 and CCL7 was also reported in human metastatic melanomas that resist anti-PD1 therapy, and is a cardinal feature of the IPRES signature5.
Having shown that TIL apoptosis in the TiRP model was mediated by FasL, we then analyzed FasL expression on cells from the tumor tissue to determine which cell type was responsible for inducing TIL apoptosis. We consistently observed high levels of FasL on MDSC (Fig. 7a, b). Lower levels of FasL were also observed on endothelial cells and, although inconsistently, on tumor cells. This expression profile was identical whether we analyzed tumors collected before or after ACT (Fig. 7a, b). FasL expression was equally high on PMN-MDSC and M-MDSC (Fig. 7c), but only the former were enriched in induced as compared with transplanted tumors (Fig. 6a). This result suggested that PMN-MDSC were responsible for TIL apoptosis. This conclusion was supported by in vitro experiments showing that PMN-MDSC induced apoptosis of co-cultured TCRP1A CD8+ T cells, the extent of which was reduced by FasL neutralization with soluble Fas-Fc (Fig. 7d, e). Moreover, TIL apoptosis was reduced in vivo when we depleted PMN-MDSC by intra-tumoral injection of anti-Ly6G antibody (Fig. 7f).
To determine whether in vivo FasL neutralization could increase the efficacy of ACT, we treated mice bearing induced tumors with soluble Fas-Fc starting one week before ACT, and followed tumor growth. We observed better tumor control in mice receiving ACT combined with Fas-Fc, as compared with mice receiving ACT alone (Fig. 8a, b). Even though Ly6G-antibody mediated depletion of PMN-MDSC combined with ACT showed some effect (Fig. 8b), when combined with Fas-Fc and ACT, PMN-MDSC depletion did not further improve tumor control (Fig. 8a, b), indicating that FasL-mediated apoptosis was the dominant suppressive mechanism of PMN-MDSC in this model system. Those results indicate that FasL neutralization can improve the efficacy of immunotherapy based on ACT.
To explore whether FasL neutralization can also increase the efficacy of immunotherapy based on immune checkpoint inhibitors, we first set up experiments in which we treated mice bearing induced TiRP tumors with ACT combined with anti-CTLA4 and anti-PD1 antibodies. The addition of immune checkpoint inhibitors failed to improve tumor rejection, unless it was combined with FasL neutralization and Ly6G-depletion (Fig. 8c). This improved tumor control was, however, similar to the one observed with ACT combined with Fas-Fc ± anti-Ly6G (Fig. 8a, b). These results indicate that TiRP tumors display additional immunosuppressive mechanisms that make them insensitive to immune checkpoint inhibitors. To further explore potential synergy between immune checkpoint inhibitors and FasL neutralization, we treated mice bearing transplanted T429.11 tumors with anti-CTLA4 and anti-PD1. We observed no tumor growth inhibition (Fig. 8d). However, when we combined anti-CTLA4 and anti-PD1 with Fas-Fc we observed a significant tumor growth inhibition (P = 0.0004, two-way ANOVA). When we also depleted PMN-MDSC in those mice we observed no further increased tumor growth inhibition, indicating that MDSC do not contribute other major immunosuppressive mechanisms in this setting. Although their recruitment is lower than in induced TiRP tumors, MDSC are also present in transplanted T429.11 tumors (Fig. 6a) and they express high levels of FasL (FasL MFI was 2879 and 2758 for transplanted and induced tumors, respectively, mean of 10 mice each). Therefore, they seem to contribute to resistance to immune checkpoint therapy, although it is possible that other FasL-expressing cells also contribute. Altogether, these results indicate that FasL neutralization has the potential to improve the efficacy of immunotherapy based not only on adoptive cell therapy but also on immune checkpoint inhibitors.
A striking feature of this immunosuppressive mechanism is its antigen-specific nature: only anti-tumor CD8+ T cells are affected by apoptosis and disappear. This specificity likely results from the fact that antigenic activation strongly increases Fas expression at the surface of CD8+ T cells. This was confirmed when we isolated spleen cells from TiRP-tumor-bearing mice and incubated them with L1210.P1A.B7-1 cells to activate P1A-specific T cells: after 48 h, we compared Fas expression on CD8+ T cells that were P1A-specific or not (Supplementary Fig. 10a). Fas expression was much higher on P1A-specific CD8+ T cells, which were the only CD8+ T cells that were activated in these experimental conditions, as indicated by CD69 expression. Moreover, when we incubated these T cells with FasL, only P1A-specific CD8+ T cells underwent apoptosis (Supplementary Fig. 10b). We also observed that IFNγ further increased Fas expression on activated CD8+ T cells (Supplementary Fig. 4c). The latter finding likely contributes to explain the reduced TIL apoptosis we observed in mice treated with IFNγ-neutralizing antibodies (Fig. 4c). As FasL is already expressed by MDSC before ACT, and therefore probably does not depend on IFNγ produced by TIL, the pro-apoptotic effect of IFNγ in this setting likely results from increased expression, on activated T cells themselves, of both Fas and FasL, according to the well-described phenomenon of activation-induced cell death (AICD), an immune checkpoint process that prevents excessive T-cell activity by inducing Fas/FasL-mediated suicide/fratricide killing of activated T cells28, 29. AICD, however, is expected to occur in transplanted tumors, as well as in induced tumors, and therefore cannot account for the increased TIL apoptosis observed in induced tumors, which rather appears to be triggered by FasL-expressing MDSC that are enriched in these tumors.
Relevance to human tumors
To determine whether FASLG expression was associated with disease progression in human tumors, we used The Cancer Genome Atlas (TCGA) database to compare the survival of patients bearing tumors expressing different levels of FasL. In most tumor types, high FASLG transcript levels were associated with a relatively better survival than low levels. This difference was statistically significant in cutaneous melanoma (P = 0.000014), head-and-neck squamous cell carcinoma (P = 0.011) and breast carcinoma (P = 0.0078) (Fig. 9a–c). In sharp contrast, high FASLG expression in renal cell carcinoma (P = 0.04) and uveal melanoma (P = 0.000059) was associated with significantly worse prognosis (Fig. 9d, e). It is noteworthy that these two tumor types are also those that diverge from most other malignancies by their shorter survival associated with higher TIL infiltration30,31,32. We therefore considered the possibility that FASLG expression in human tumors was in fact associated with T-cell infiltration. Consistently, expression of FASLG in the main TCGA tumor types was strongly correlated with the levels of T-cell-specific transcripts such as IFNG (shown for melanoma in Fig. 9f), CD3E and CD8B (not shown). FASLG transcript levels in tumors thus reflect TIL abundance and activity, in line with the selective expression of FasL in activated T cells, and cannot be used as an independent prognostic factor. Interestingly, a similar correlation with TIL infiltration was observed for the transcript levels of IDO1 and PD-L1, two well-known immune checkpoints that are induced by T-cell activation and involved in adaptive tumoral resistance, as ascertained by the clinical benefit obtained with specific inhibitors12.
Collectively, our results establish a dominant mechanism of resistance to anti-tumor immunity in the TiRP model of inflammatory melanoma, characterized by the induction of TIL apoptosis through FasL expression by PMN-MDSC infiltrating the tumor. Our results also illustrate, for the first time, the strikingly different behavior of autochthonous vs. isogenic transplanted tumors, which are efficiently controlled by immunotherapy. The long-standing interaction between the host and the autochthonous tumor growing progressively among normal tissues results in a state of abnormal inflammation4, which allows immunological tolerance of the tumor and is characterized by enrichment in PMN-MDSC. Those cells express high amounts of FasL, which then induces apoptosis of tumor-specific CD8+ T lymphocytes infiltrating the tumor and their progressive depletion from the periphery. These results not only confirm the potential of targeting PMN-MDSC to reduce tumor-induced immunosuppression, but also identify FasL as a promising target whose neutralization could increase the clinical success of cancer immunotherapy.
The role of FasL in mediating tumoral immune resistance has been discussed for many years, following an initial report about FasL expression by melanoma cells triggering TIL apoptosis33. Several reports subsequently showed that enforced expression of high-levels of FasL in tumor cells induced inflammation associated with recruitment and activation of neutrophils resulting in tumor rejection34, 35. These conflicting results initiated what was called the Fas-counterattack controversy36. The issue was further complicated by the poor specificity of the anti-FasL antibodies available at the time, the relative role of membrane-bound vs. soluble FasL, which does not trigger apoptosis, and the interference of FasL expression by other cells, including the T cells themselves in the process of AICD36,37,38. Subsequent work partly clarified the controversy, by suggesting that the pro-inflammatory role of FasL was dependent on high expression levels obtained by enforced FasL expression, and showing that it was not observed with tumors expressing FasL naturally, in which case FasL down-regulation reduced tumorigenicity, in line with the immunosuppressive role of FasL39. Another key aspect is the type of cells expressing FasL: contrary to the previous work discussed above, in the TiRP model we did not so much observe FasL expression on tumor cells but rather on PMN-MDSC.
Similarly, recent work highlighted the immunosuppressive role of FasL expressed by endothelial cells in ovarian cancer40. In this work, FasL-mediated apoptosis of anti-tumor T cells was shown to prevent access of CD8+ T cells to tumor nests in ovarian carcinoma40. This was linked to FasL expression by tumor endothelia, inducing apoptosis of T cells reaching the tumor vasculature, thereby preventing T-cell extravasation. Despite similarity in the mode of T-cell execution, the mechanism of our model is different, because TIL extravasation and infiltration is not prevented, but TILs are induced to apoptosis by PMN-MDSC within tumor nests. In practice, however, our observations concur with those of Motz et al. in defining FasL as a relevant drug target whose neutralization could improve the efficacy of cancer immunotherapy. A recent study highlighted an additional benefit of neutralizing FasL in the context of ACT41. This study compared the anti-tumor efficacy of ACT making use of naive or memory anti-tumor CD8+ T cells, or a mix of both. The latter condition would mimic the clinical situation in cancer patients. It is known that naive CD8+ T cells are more efficient due to better persistence after ACT. The authors showed that when naive CD8+ T cells are mixed with memory CD8+ T cells and used for ACT, they undergo a precocious differentiation that limits their anti-tumor efficacy. This precocious differentiation is driven by memory CD8+ T cells through non-apoptotic Fas signaling: FasL neutralization prevents this precocious differentiation of naive CD8+ T cells and increases the efficacy of ACT in this setting. We only used naive CD8+ T cells and injected them after short-term in vitro activation, so this effect does not explain the beneficial effect of FasL neutralization in our experimental setting. But it provides an additional rationale for FasL neutralization in clinical protocols of ACT, in which the autologous CD8+ T cells used are usually a mix of naive and memory CD8+ T cells.
Although the dominant mechanism of T-cell cytotoxicity relies on delivery of apoptosis-inducing granzymes into target cells42, the FasL pathway also contributes, mostly to bystander cell killing. Such killing of stromal cells through FasL was found to be required for complete tumor rejection in the context of ACT43. However, the major role of the FasL pathway in the immune system is to promote activation-induced cell death, which entails fratricide and/or suicide death of T cells induced to express high levels of Fas and FasL upon activation. This counter-regulatory mechanism prevents excessive T-cell activity, and as such qualifies as an immune checkpoint mechanism, whose inactivation should benefit cancer immunotherapy. This notion is supported by the phenotype of gld and lpr mice, which are genetically deficient in FasL and Fas, respectively, and display autoimmune phenotypes44,45,46,47.
In sum, the potential benefits of FasL neutralization for cancer immunotherapy, particularly in the context of adoptive cell therapy, are fourfold: (i) preventing TIL apoptosis induced by FasL-expressing PMN-MDSC, (ii) improving CD8+ T-cell infiltration into the tumor by preventing apoptosis induced by endothelial cells, (iii) improving T-cell persistence at the tumor site by inhibiting the immune checkpoint relying on AICD, (iv) improving T-cell persistence and activity by preventing precocious differentiation of naive T cells induced by co-injected memory T cells. Strategies aimed at neutralizing FasL in combination with clinical approaches of cancer immunotherapy are therefore warranted.
TiRP-10B;Ink4a/Arfflox/flox mice2 contain an inducible transgene which is controlled by the tyrosinase promoter and drives expression of H-Ras and of Trap1a, which encodes a MAGE-type tumor antigen. They were backcrossed to a B10.D2 background (TiRP mice) and bred to homozygosity (TiRP-10B+/+;Ink4a/Arfflox/flox). B10.D2;Ink4a/Arfflox/flox (TiRP-10B−/−) mice were used as controls and as recipients for tumor transplantation experiments. TCRP1A mice heterozygous for the H-2Ld/P1A35-43-specific TCR transgene10 were kept on the B10.D2;Rag1−/− background. For all the in vivo treatments, mice were randomly divided into groups. Investigators were blinded to the group allocation during the animal experiments. Age-matched littermate mice of both sexes between 5 and 7 weeks of age were used for all in vivo experiments. All mice used in this study were produced under SPF conditions at the animal facility of the Ludwig Institute for Cancer Research. All the rules concerning animal welfare have been respected according to the 2010/63/EU Directive. All procedures were performed with the approval of the local Animal Ethical Committee, with reference 2015/UCL/MD/15.
Tumor induction with 4OH-Tamoxifen
A fresh solution of 4OH-Tamoxifen was prepared by dissolving 4OH-Tamoxifen (Sigma Aldrich and Imaginechem) in 100% ethanol and mineral oil (ratio 1:9) followed by 30-min sonication, and injected twice s.c (4 mg/200 µl) in the neck area of anesthetized TiRP mice two weeks apart. The therapeutic experiments were performed exclusively with mice bearing unpigmented Amela tumors. Unless mentioned otherwise, mice with pigmented tumors (Mela) were excluded due to different tumor growth kinetics4. Tumor volume (in mm3) was calculated by the following formula: volume=length/2×width2. For occasional mice that developed more than one tumors we only considered the tumor that appeared first.
T429.11, T429.6 and T429.18 clones were derived from an induced Amela TiRP tumor referred to as T4294. They were cloned from the T429 induced melanoma primary tumor line. The expression of the transgene2 was evaluated in all three T429 clones and T-cell recognition was tested for each of them in vitro using TCRP1A CD8+ T cells. P511 is an azaguanine-resistant variant of P81548, 49. P1204 is a P815AB-negative variant carrying a deletion of gene Trap1a48, 49. L1210.P1A.B7-1 cells were obtained by transfection of L1210.P1A cells with the murine B7-1 cDNA cloned into plasmid pEFBOS50. All cells were maintained at 37 °C with 8% CO2. Unless otherwise specified, all culture media are IMDM (GIBCO) containing 10% fetal bovine serum supplemented with L-arginine (0.55 mM, Merck), L-asparagine (0.24 mM, Merck), glutamine (1.5 mM, Merck), beta-mercaptoethanol (50 µM, Sigma), 50 U ml−1 penicillin and 50 mg ml−1 streptomycin (Life Technologies) (complete medium). Cell lines were tested for mycoplasma contamination.
Adoptive transfer of TCRP1A CD8+ T cells
P1A-specific CD8+ T cells were isolated from spleens and lymph nodes of TCRP1A mice using anti-CD8α MicroBeads (Miltenyi Biotec), and stimulated in vitro by co-culture for 4 days with irradiated (10,000 rads) L1210.P1A.B7-1 cells51 (105 of each cell type per well in 48-well plates) in IMDM complete medium. After purification on a Ficoll gradient on day 4, 107 living TCRP1A CD8+ T cells were injected i.v. in 200 µl PBS.
Vaccine and antibodies treatment
TiRP mice were immunized against P1A-tumor antigen by a heterologous prime-boost regimen consisting of a first injection of an Adeno.Ii.P1At (108 PFU/mouse/100 µl) followed, 15 days later, by SFV.P1A (107 IU/mouse/100 µl)6. Both viruses were given intradermally. Anti-PD1 (BioXcell, clone RMP1-14, Rat IgG2a, 200 µg/mouse i.p.) and anti-CTLA4 (BioXcell, clone 9H10, Syrian Hamster IgG2, 40 µg/mouse i.p.) treatment were started at the time of tumor appearance every 3 days for a total of 4 injections. Control mice received isotype controls (for CTLA4: BioXcell, Polyclonal Syrian Hamster IgG; for PD-1, BioXcell, Rat IgG2a, clone 2A3).
For the control immunization against P91A (Fig. 1c), mice received 4 intramuscular injections every two weeks of P91A peptide (50 µg) in PBS, mixed v/v with AS15 adjuvant52 kindly provided by GlaxoSmithKline, Belgium. For in vivo neutralization of IFNγ, mice received 0.5 mg anti-mouse IFNγ (BioXcell, clone XMG1.2,) 1 day before ACT.
For the intra-tumoral depletion of Ly6Gh MDSC, presented in Fig. 7f, mice received 200 µg/mouse of antibody to Ly6G (BioXcell, clone 1A8, Rat IgG2a) or isotype control (BioXcell, Rat IgG2a, clone 2A3) every 3 days starting 4 days before ACT.
For the in vivo antibody treatment combined with adoptive cell transfer (Fig. 8a–c), anti-Ly6G (200 µg/mouse except the first injection which was 400 µg/mouse, twice a week), and Fas-Fc (150 µg/mouse, two injections per week) treatment started when the tumor size reached 500 mm3. The anti-CTLA4 (100 µg/mouse) and anti-PD1 antibodies (200 µg/mouse) were injected 4 times at a 3 days interval i.p. 1 day after the anti-Ly6G and Fc-Fas injection. The corresponding isotype controls were injected accordingly. Mice received 107 activated TCRP1A CD8+ T cells (ACT) 3 days after Fas-Fc and anti-Ly6G injection. Tumor size was monitored every 2 days.
For the in vivo antibody treatments (Fig. 8d), anti-Ly6G (BioXcell, clone 1A8, Rat IgG2a, 200 µg/mouse except the first injection which was 400 µg/mouse, twice a week), and Fas-Fc (150 µg/mouse, two injections per week) treatment started when the tumor size reached 50–150 mm3. The anti-CTLA4 (BioXcell, clone 9H10, 100 µg/mouse) and anti-PD1 antibodies (BioXcell, clone RMP1-14, 200 µg/mouse) were injected 4 times at 3-day intervals i.p. starting 1 day after the anti-Ly6G and Fas-Fc injections. The corresponding isotype controls were injected accordingly. Tumor size was monitored every 2 or 3 days.
H-2Ld/P1A and P91A tetramers (1 µM) were produced in our Institute53. Antibodies used were: CD8α-PerCP (clone 53–6.7, Biolegend, 1 µg/mL); CD69-APC (clone H1.2F3, Biolegend, 1 µg/mL); CD3ε-FITC (clone 17A2, Biolegend, 1 µg/mL); CD45-FITC, CD45-Alexa700 (clone 30-F11, Biolegend, 1 µg/mL); Gr-1-APC, Gr-1-Bv421 (clone RB6-8C5, Biolegend, 1 µg/mL); Ly6C-PerCP (clone HK1.4, Biolegend, 1 µg/mL); Ly6G-PE, Ly6G-Bv510 (clone 1A8, Biolegend, 1 µg/mL); CD11b-BV421 (clone M1/70, Biolegend, 1 µg/mL); CD178 (FasL) (clone Kay-10, Biolegend, 2 µg/mL); CD31-PE/Cy7 (clone 390, Biolegend, 1 µg/mL); CD95-PE/Cy7 (Fas) (clone Jo2, Becton Dickinson, 1 µg/mL) and corresponding isotype controls. Viability dye efluor780 (eBiosciences, 1:500) was used. Annexin V-BV421 or Annexin V-APC (Biolegend, 1:50) was used to monitor apoptosis. P1A antibody (clone P1A102B3, 1 µg/mL) was produced by immunizing P1A-KO mice54 with a P1A peptide (CEEMGNPDGFSP) coupled to ovalbumin. Hybridomas were sub-cloned and further screened for production of anti-P1A antibodies. Clone P1A102B3 was selected and confirmed to produce an IgG1 recognizing P1A specifically and applicable for western blot, IHC, Flow cytometry and ELISA.
Total RNA isolated from tumor tissues were tested by qRT-PCR using the following primers: Trap1a (P1A): forward: 5ʹ-AGC-TGA-GGA-AAT-GGG-TGC-TG-3ʹ (exon 1), reverse: 5ʹ-CAG-CAT-TTT-CAC-ACC-TAC-ACT-CCA-3ʹ (exon 2), probe: 5ʹ-FAM-CCA-TCA-TTT-AAG-GAA-GAA-TGA-AGT-GAA-GTG-TAG-GAT-GA-TAMRA-3ʹ (exon 2). Ifng: forward: 5ʹ-TCA-AGT-GGC-ATA-GAT-GTG-GAA-GAA-3ʹ, reverse: 5ʹ-TGG-CTC-TGC-AGG-ATT-TTC-ATG-3ʹ, probe: 5ʹ-FAM-TCA-CCA-TCC-TTT-TGC-CAG-TTC-CTC-CAG-TAMRA-3ʹ. Samples were assayed in triplicates and expression levels were normalized to β-actin, which was tested with the following primers: forward actin: 5ʹ-CTC-TGG-CTC-CTA-GCA-CCA-TGA-AG-3ʹ, reverse actin: 5ʹ-GCT-GGA-AGG-TGG-ACA-GTG-AG-3ʹ, probe actin: 5ʹ-FAM-ATC-GGT-GGC-TCC-ATC-CTG-GC-TAMRA-3ʹ. Faslg was amplified using ROX SYBR MasterMix blue dTTP using the following primers: forward: 5ʹ-TGA-ATT-ACC-CAT-GTC-CCC-AG-3ʹ, reverse: 5ʹ-AAA-CTG-ACC-CTG-GAG-CC-3ʹ (Fig. 4d). In Fig. 4e, Faslg was amplified using other primers and probe (Integrated DNA technologies, Mm.PT.58.28447971). Gapdh (Mm.PT.39a.1) Il6 (Mm.PT.58.10005), Csf1 (Mm.PT.58.11661276), Csf2 (Mm.PT.58.9186111), Csf3 (Mm.PT.58.7976206.gs), Cxcl1 (Mm.PT.58.42076891), Cxcl12 (Mm.PT.58.12038563), Ccl2 (Mm.PT.58.42151692), Tgfb1 (Mm.PT.58.11254750), Ccl7 (Mm.PT.58.17719534), Cxcl5 (Mm.PT.58.29518961.g) primer and probe were purchased from Integrated DNA technologies.
Fas silencing in TCRP1A CD8+ T cells
Silencing of Fas in activated TCRP1A CD8+ T cells was achieved by using Accell siRNA (1 µM) (Dharmacon), with sequence 5ʹ(P)-GUGCAAGUGCAAACCAGAC(dTdT)-3ʹ. After 1 day of activation, TCRP1A CD8+ T cells were incubated with 1 µM Accell control or Fas siRNA in IMDM complete medium without serum for 24 h, then the medium was replaced with IMDM complete medium with serum. The silencing procedure was repeated on day 3. On day 4, 107 TCRP1A CD8+ T cells were adoptively transferred into mice bearing induced tumors. Silencing efficiency was monitored by Western blot using anti-Fas antibody (Santa Cruz, FL- 335, 1:1000). Blots were stripped and reprobed with a rabbit anti-mouse monoclonal antibody to GAPDH (Cell Signaling, 14C10, 1:5000).
Production of soluble Fas-Fc
Recombinant soluble dimers of the murine Fas death receptor extracellular domain (ECD) were produced in Drosophila S2 cells (Invitrogen). The sequence coding for the ECD of murine Fas was fused in frame to the sequence coding for the murine FcIgG2A portion to allow dimerization of the fusion protein. A sequence coding for a derivative of the human IL-9 signal peptide was added at the 5ʹ end of the fusion gene to allow secretion of the recombinant protein. The recombinant gene was cloned into a derivative of the expression vector pRMHA3, which contains a neomycin-resistance gene and a copper inducible promotor from the Drosophila melanogaster metallothionein gene. Drosophila S2 cells were transfected and stable transfectants selected using 1.5 mg/ml of G418 antibiotic (Roche). Production of the recombinant protein was induced in the cultured cells by the addition of 0.5 mM CuSO4. The protein was purified from the cell culture supernatant by affinity chromatography on a protein A sepharose column (GE Healthcare).
In vivo apoptosis staining
Four days after ACT, tumor-bearing mice received an i.v. injection of NIR-FLIVO™ 690 Control (8.5 µg/mouse) or NIR-FLIVO™ 690 (NIR-FLIVO™-DyLight® 690-VAD-FMK in vivo Apoptosis Tracer, ImmunoChemistry Technologies, 8.5 µg/mouse) 4 h before killing. TCRP1A CD8+ T cells were then identified using H-2Ld-P1A tetramers and anti-CD8α antibody and analyzed for FLIVO staining by FACS.
IFNγ secretion was evaluated by ELISA (Biolegend) in the supernatant of tumor homogenates after 24 h of culture in IMDM medium.
T-cell apoptosis in organotypic tumor slices in vitro
Fresh tumor tissue slices of 250–300 μm thickness were prepared using the microtome (LEICA, Vibratome VT1200), and cultured on organotypic inserts for 24 h (Millipore) in 4 ml IMDM complete medium. One million TCRP1A CD8+ T cells pre-labeled with CMAC (CellTracker Blue CMAC Dye) were added for 24 h. After washing and OCT embedding, cryosections (7 μm) of these organotypic tumor tissues were stained for apoptosis using FLICA pan-caspase reagent (ImmunoChemistry Technologies, 1:150) and scanned with MIRAX digital microscope. Quantification was performed using Biopix software.
After mechanical and enzymatic dissociation of induced TiRP tumors, MDSC were enriched by centrifugation on Ficoll-Paque Plus (GE Healthcare). Ly6Gh cells were further purified by magnetic sorting (Miltenyi) and confirmed as PMN-MDSC by FACS analysis (Gr-1h, CD11b+, Ly6Clo and Ly6Gh).
Tumor transplantation with MDSC isolated from induced tumors
Single cell suspensions were prepared from induced Amela TiRP tumors and from T429.11 transplanted tumors using gentleMACS™ Dissociator. MDSCs were purified from the induced TiRP tumor suspension (as described above) and mixed with the transplanted tumor suspension (MDSC/T429.11 ratio: 1/4). The T429.11 tumor cell suspensions mixed or not with isolated MDSC were injected intradermally into naive B10.D2;Ink4a/Arfflox/flox mice, on the left and right flank respectively. When the tumor volume of the left flank reached around 500 mm3, the mice received adoptive cell transfer of TCRP1A specific CD8+ T cells. The tumor size was measured every 4 days from the day of the adoptive cell transfer (n = 8 mice).
MDSC-mediated T-cell suppression assay
TCRP1A T CD8+ T cells activated in vitro for 4 days with irradiated L1210.P1A.B7-1 cells were purified, washed, counted and 100 µl of 2 × 105 cells viable cells were seeded in a 96-well plate in IMDM complete medium. Ly6G+ MDSC from induced tumors were isolated and purified as above. TCRP1A CD8+ T cells were co-cultured with Ly6G+ MDSCs at the indicated ratios for 3 days at 37 °C. TCRP1A T cells were then purified using anti-CD8 microbeads. Their proliferation was assessed in an 18-hour 3H-Thymidine incorporation assay. TCRP1A CD8+ T cells co-cultured with Ly6G+-MDSC for 3 days as above and purified using anti-CD8 microbeads were tested for cytolytic activity in a standard 4 h-chromium release assay, using P511 cells as P1A+ target cells and P1A− cells P1204 as cold target competitors.
Statistical analyses were performed using Prism 6 (GraphPad Software). Comparison between two groups was performed using the unpaired Student t-test, while two-way analysis of variance (ANOVA) was used to compare the effects of different treatments on tumor growth. P values < 0.05 were considered significant (P-value < 0.05 was annotated *; < 0.01 **; < 0.001 ***; < 0.0001 ****).
The tumoral gene expression values and patient survival data from The Cancer Genome Atlas (TCGA) database were retrieved from the cBioPortal’s Cancer Genomic Data Server using the CGDS-R package (http://www.cbioportal.org/cgds_r.jsp) and analyzed with the R software using the “survival” package. The patients were divided in two groups according to high and low FASLG expression level, separated by the median value. Their survival curves were compared using Cox proportional hazard regression model. A P-value below 0.05 was considered significant.
The TCGA data referenced during the study are available in a public repository from the TCGA website (https://cancergenome.nih.gov). The authors declare that all other data supporting the findings of this study are available within the article and its supplementary information files and from the corresponding author upon reasonable request.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank Guy Warnier, Gilles Gaudray, Julien Gossiaux and Laurent Hermans for large-scale production of TiRP and TCRP1A mice, Nicolas Dauguet for help with flow cytometry, Floriane Ribeiro and Dominique Donckers for help with mouse experiments, Madeleine Swinarska for producing soluble Fas, Pascal Schneider for advice in producing soluble Fas and for providing recombinant FasL, Jacques Van Snick and Pamela Cheou for providing antibodies, Ahmed Essaghir for patient survival analysis, Pierre van der Bruggen and Thierry Boon for critical comments, and Mandy Macharis and Auriane Sibille for editorial assistance. This work was supported by Ludwig Cancer Research, Walloon Excellence in Life Sciences and Biotechnology (WELBIO, Belgium), FNRS-Télévie (Belgium), Foundation Against Cancer (Belgium), de Duve Institute and Université catholique de Louvain (Belgium).
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.