Replicating viral vector platform exploits alarmin signals for potent CD8+ T cell-mediated tumour immunotherapy

Viral infections lead to alarmin release and elicit potent cytotoxic effector T lymphocyte (CTLeff) responses. Conversely, the induction of protective tumour-specific CTLeff and their recruitment into the tumour remain challenging tasks. Here we show that lymphocytic choriomeningitis virus (LCMV) can be engineered to serve as a replication competent, stably-attenuated immunotherapy vector (artLCMV). artLCMV delivers tumour-associated antigens to dendritic cells for efficient CTL priming. Unlike replication-deficient vectors, artLCMV targets also lymphoid tissue stroma cells expressing the alarmin interleukin-33. By triggering interleukin-33 signals, artLCMV elicits CTLeff responses of higher magnitude and functionality than those induced by replication-deficient vectors. Superior anti-tumour efficacy of artLCMV immunotherapy depends on interleukin-33 signalling, and a massive CTLeff influx triggers an inflammatory conversion of the tumour microenvironment. Our observations suggest that replicating viral delivery systems can release alarmins for improved anti-tumour efficacy. These mechanistic insights may outweigh safety concerns around replicating viral vectors in cancer immunotherapy.

T he clinical efficacy of checkpoint blockade, oncolytic viruses and adoptive T-cell therapy heralds success in harnessing the immune system in the combat against cancer [1][2][3] . Conversely, active immunization has not yet demonstrated consistent efficacy in clinical Phase III trials [4][5][6] , raising an urgent need for improved vaccine formulations that should aim at delivering large numbers of CD8 þ cytotoxic effector T lymphocytes (CTL eff ) to the tumour site while simultaneously establishing a pool of self-replenishing memory cells for durable tumour control [7][8][9] . The immunological response profile of replicating viral vaccines represents an excellent match to these requirements 10 . By delivering tumourassociated antigens (TAA) in the context of an acute viral infection, such delivery systems should supply critical alarmin signals, also referred to as damage-associated molecular patterns , as well as pathogen-associated molecular patterns (PAMPs) for optimal CTL induction and differentiation 11,12 .
Classical tumour vaccination regimens such as peptides-inadjuvant showed only marginal clinical benefit, despite the induction of sizeable tumour antigen-specific CTL responses 13 . Major impediments include inefficient tumour infiltration and efficacy of specific CTL eff (refs 14,15). Overcoming these hurdles will critically depend on innate immune activation, which can be achieved by virus-induced inflammation 16 .
Lymphocytic choriomeningitis virus (LCMV), the prototype member of the arenavirus family, elicits CTL eff responses of high magnitude and cytolytic capacity, in combination with life-long CTL immunity. These features, together with a low hazard profile for humans, have rendered it a primary workhorse of immunologists since the 1930s (ref. 17). Experimental LCMV infections in humans have documented a systemic inflammatory reaction, accompanied by a lymphoblastic reaction in peripheral blood 18 , which was reminiscent of the massive CTL eff response in infectious mononucleosis 19 . Accordingly, studies in accidentally LCMV-infected laboratory workers have confirmed that, analogously to mice, high frequencies of effector memory CTL are maintained for several years after a single acute infection 20 . These features in combination with low seroprevalence in the human population 21,22 raised our interest in LCMV as a live-attenuated cancer immunotherapy platform, to deliver TAA-specific immunization alongside with potent innate immune activation.
We and others have previously reported that interleukin-33 (IL-33) is a key driver of potent and protective CTL eff responses to several replicating RNA and DNA viruses 11 . Subsequently, IL-33 was found to also be essential for antiviral Th1 CD4 responses 23 and for graft-versus-host disease 24 . Collectively, these reports underscore the global significance of IL-33 in promoting type 1 immune responses (reviewed in ref. 25), besides an additional undisputed function in type 2 immune responses such as in the context of allergy and immunity to parasites. Upon release from necrotic cells, IL-33 signals through its receptor ST2 on activated CD8 þ T cells, thereby enhancing clonal expansion, effector differentiation and ultimately CTL efficacy 11 . By contrast to wild-type LCMV (LCMVwt), however, replication-deficient LCMV-based vectors (rLCMV 26 ) fail to trigger the IL-33 alarmin pathway 11 .
Hence we engineer a TAA delivery platform based on live-attenuated LCMV (artLCMV). Unlike replication-deficient rLCMV vectors, artLCMV targets not only dendritic but also lymphoid stromal cells, thereby triggering the IL-33-ST2 alarmin pathway and inducing superior TAA-specific CTL eff responses. The activation of critical innate immune pathways including alarmins represents a discriminating feature of replicating microbial delivery systems, which might be decisive for the success of active cancer immunotherapy.

Results
Generation of genetically stable and live-attenuated artLCMV. We aimed at combining tumour antigen vaccination with infection-induced alarmin signals. Hence we sought a strategy how to stably incorporate transgenes into replicating LCMV. The virus' genome consists of two negative-stranded RNA segments designated L and S, respectively, encoding two viral genes each (LCMVwt, Fig. 1a). 5 0 and 3 0 untranslated regions (UTR) and an intergenic region (IGR) flank the ORFs on each segment (Fig. 1a). Reverse genetic tools are available to efficiently tailor the infectious virus' genome for medical application 26,27 . As reported 28 , transgenes of interest can be accommodated in the LCMV genome by segregating the viral glycoprotein (GPC) and nucleoprotein (NP) genes onto artificially duplicated S segments (S NP , S GP in r3LCMV, Fig. 1a). According to the originally published vector design strategy 28 , both NP and GPC remain under control of their respective regulatory RNA elements (natural positioning; r3LCMV). Genetic and phenotypic stability represent key criteria for manufacturing and clinical translation of live-attenuated viral vector systems 29,30 . Accordingly, observations on transgene loss in r3LCMVinfected animals (see below) prompted us to search for molecular strategies to stabilize r3LCMV genomes. By placing GPC under 3 0 UTR control, we generated viruses with an artificial genome organization (artLCMV, Fig. 1a). The reasoning was that, although a rare event in negative strand RNA viruses 31 , inter-segmental recombination 32 would reunite NP and GPC on a single S segment, thereby deleting the transgenes (Fig. 1b). The arenavirus promoter consists of an intra-segmental RNA hybrid formed by a highly conserved stretch of 19-21 terminal nucleotides in each segment's 5 0 UTR and 3 0 UTR. Therefore, viral promoter activity requires both, a 5 0 UTR and a 3 0 UTR on each RNA template (Fig. 1b) 33 . Unlike in r3LCMV, a recombination event reuniting the NP and GPC ORFs of artLCMV would occur at the expense of losing the 5 0 UTR, thus creating an inactive recombination product devoid of a viral promoter (Fig. 1c). Both, r3LCMV and artLCMV were significantly attenuated in cell culture (Fig. 1d). Flow cytometric studies with green (GFP) and red fluorescent protein (tomato) co-expressing viruses (r3LCMV-GFP/tom, artLCMV-GFP/tom) revealed that red/green double-positive virions were outnumbered by a 4tenfold excess of bi-segmented replicationdeficient particles, carrying either only S GP or S NP in combination with the L segment (  Fig. 2a). Hence, attenuation was, at least in part, due to inefficient co-packaging of S NP and S GP segments. To study the genetic stability of artLCMV and r3LCMV we exploited AGR mice. Owing to targeted deletions of RAG1 (T-cell and B-cell deficiency) as well as of the type I and type II interferon genes, AGR mice readily reveal an attenuated virus' reversion to virulence 34 . We measured viral loads in blood by IFF, relying on the detection of the viral structural protein NP (NP-IFF, Fig. 2a). Within the first 20 days after infection, r3LCMV and artLCMV remained at considerably lower titres in blood than LCMVwt, which was in line with published data documenting in vivo attenuation of r3LCMV 35 . After 30-40 days, however, r3LCMV-infected AGR mice reached NP-IFF titres identical to those infected with LCMVwt. Conversely, artLCMV viremia remained at low levels throughout the observation period of 120 days (Fig. 2a). In artLCMV-infected animals, the assessment of total infectivity by NP-IFF and the quantification of GFP-expressing viruses by GFP-IFF yielded comparable titres on day 5 as well as on day 120 after infection, indicating that the transgene had been retained (Fig. 2b,c). In contrast, r3LCMV-GFP-infected mice turned GFP-IFF-negative while abundant NP-IFF infectivity persisted, suggesting that the virus had lost its transgene. Indeed, the viral NP was abundant in the liver and spleen of r3LCMV-GFP-infected mice, as expected, while GFP was absent (Fig. 2d, Supplementary Fig. 2b). In contrast, liver and spleen of artLCMV-GFP-infected animals exhibited dense green fluorescence (Fig. 2d, Supplementary  Fig. 2b; analogous data for blood monocytes in Supplementary  Fig. 2c). To verify whether r3LCMV-GFP had inactivated its transgene by recombining its two S segments, as hypothesized (Fig. 1b), we relied on an PCR with reverse transcription (RT-PCR)-based sequencing strategy ( Supplementary Fig. 2d). Recombined S segment RNA species (S rec , Supplementary Fig. 2d) reuniting NP and GPC sequences in a single RNA molecule were detected in ten out of ten r3LCMV-GFP-infected AGR mice from two independent series of experiments. In contrast, the blood of 11 artLCMV-infected mice was free of detectable S rec RNA.
Sequence analysis of S rec segments revealed that individual non-homologous recombination events had occurred in each animal, with breakpoints in the GFP-IGR border region (Fig. 1b,  Supplementary Fig. 2e). Genetic tags in the IGR and GPC sequences of r3LCMV-GFP excluded laboratory contaminations as potential confounder in these analyses and assigned genetic elements in S rec to its parental S NP or S GP segments (Supplementary Fig. 2f,g). We found that a cDNA-derived bi-segmented virus with an exemplary S rec segment (S rec #1) grew to LCMVwt titres in cell culture (Fig. 2e). In combination with the inefficient copackaging of S NP and S GP in tri-segmented particles (Fig. 1d), this finding explained the selective advantage and consistent outgrowth of S rec segments in r3LCMV-infected AGR mice.
Apathogenic artLCMV induces potent CTL eff responses. In vivo attenuation represents a prerequisite for the use of live vaccine delivery platforms in immunocompromised cancer patients.   Infection of wt mice with LCMVwt, but not with artLCMV resulted in viremia (Fig. 3a). Moreover, low level artLCMV infectivity was only transiently detected in spleen and liver on day 4 and was cleared by day 7, whereas LCMVwt persisted for 10 days at high titres (Fig. 3b). LCMVwt can cause choriomeningitis in accidentally infected humans and in mice the virus is invariably lethal at intracranial (i.c.) doses of r10 plaque forming units (PFU, Fig. 3c, ref. 36). In contrast, artLCMV failed to cause disease at doses up to 10 4 PFU i.c., and only one of five animals developed terminal disease when given 10 5 PFU i.c. Despite its attenuation, intravenous artLCMV-induced substantial levels of serum type I interferon (IFN-I, Fig. 3d) for at least 48 h. rLCMV triggered a comparably minor and short-lived IFN-I release, whereas neither rAd-based nor replicating vaccinia virus-based vectors induced detectable levels of systemic IFN-I. Analogous results were obtained when rAd vectors were given intramuscularly or upon subcutaneous and intradermal administration of vaccinia virus-based vectors ( Supplementary Fig. 3a). Although we used replication-competent vaccinia virus for our study, the lack of systemic IFN-I was expected owing to multiple virally encoded antagonists of the IFN-I response 37 . Taken together, these results indicated that artLCMV was genetically stable and substantially attenuated, both in vitro and in vivo, but retained the ability to efficiently activate the innate immune system. To assess artLCMV-induced CD8 þ T-cell responses to transgenes of choice, we generated an ovalbumin-(OVA-) expressing vector (artLCMV-OVA). It induced OVA epitope-specific CTL numbers, which were substantially higher than those elicited by an OVA-recombinant adenovirus 5-based vector (rAd-OVA) or by a replication-deficient LCMV-based vector (rLCMV-OVA, Fig. 4a). Most notably, artLCMV-OVA-triggered KLRG1 þ CD127 À effector CD8 þ T-cell (CTL eff ) responses in blood and spleen exceeded those of the other delivery systems by Btenfold (Fig. 4b,c), while CD127 þ CTL mem memory precursor cell numbers were comparable (Fig. 4d). OVA-specific total CTL and CTL eff responses to artLCMV-OVA immunization were also significantly higher than those induced upon intramuscular administration of rAd-OVA and either subcutaneous or intradermal vaccination with rVACC-OVA ( Supplementary  Fig. 3b,c). Intracellular cytokine staining showed that artLCMV-OVA elicited ten to 20-fold higher numbers of IFN-g-producing and IFN-g/TNF co-producing CD8 þ T cells than rAd-OVA or rLCMV-OVA, and also IFN-g/TNF/IL-2 triple producers were significantly more numerous (Fig. 4e). These differences in magnitude and functionality of OVA-specific CTL populations persisted in the memory phase (Fig. 4f). Superior functionality of artLCMV-induced CTL eff responses was also evident in primary ex vivo Cr 51 release assays, with Ztenfold more lytic units in spleen after artLCMV infection than after rLCMV-OVA, rAd-OVA or recombinant vaccinia virus (rVACC-OVA) immunization ( Fig. 4g). Importantly, artLCMV-based vaccines elicited also robust CD8 þ T-cell responses against the tumour self-antigens Her2 and P1A (Fig. 4h,i). Analogously to artLCMV-OVA immunization, P1A-specific CD8 þ T cells in blood of artLCMV-P1A-immunized mice were Btenfold more numerous than in animals receiving rLCMV-P1A ( Fig. 4i). LCMVwt evades antibody neutralization by means of its envelope glycan shield 38 , and even two sequential artLCMV immunizations failed to induce detectable neutralizing serum activity (Fig. 4j). Accordingly, artLCMV-induced CTL responses were considerably augmented when artLCMV-OVA was re-administered in homologous prime-boost vaccination (Fig. 4k). These observations do not, however, rule out the likely possibility that alternative mechanisms of interference such as T-cell immunity to viral backbone epitopes 39 may compete with and thereby attenuate booster responses to artLCMV-vectorized transgenes.
We investigated the mechanisms underlying exceptionally potent CTL eff responses upon artLCMV immunization. First we studied its tropism for antigen-presenting cells in vivo as a basis for potent CTL induction. While both, rLCMV and artLCMV, efficiently targeted plasmacytoid DCs (pDCs), artLCMV infected a significantly higher number of conventional dendritic cells and macrophages than its replication-deficient counterpart rLCMV (Fig. 5a). In addition we hypothesized that artLCMV, analogously to LCMVwt, triggered the IL-33-ST2 alarmin pathway, thereby potentiating CTL responses and CTL eff responses in particular 11 . When immunized with artLCMV-OVA, wt mice mounted Btenfold higher OVA-specific CTL responses than animals lacking the IL-33 receptor ST2 (Il1rl1 À / À ), both in spleen and blood (Fig. 5b,c). These differences were particularly pronounced in the KLRG1 þ CD127 À CTL eff subset (Fig. 5b,c). Conversely, the responses to rAd-OVA, rVACC-OVA and rLCMV-OVA were unaffected by ST2 deficiency, suggesting that artLCMV-OVA immunization but neither replication-deficient rLCMV-nor rAdor vaccinia-vectored vaccination triggered the IL-33-ST2 axis. The capacity to replicate in vivo differentiates artLCMV and rLCMV and was apparently required to activate this pathway. Bioactive IL-33 is released from non-haematopoietic stromal cells 11 , which are a target of replicating LCMV infection 40 . To test the hypothesis that artLCMV triggered ST2 signalling by infecting IL-33-expressing stromal cells inside secondary lymphoid organs, we developed and validated a green fluorescent IL-33 reporter mouse (IL-33 gfp/wt , Supplementary Fig. 4a,b). More than half of the splenic CD45 neg (non-haematopoietic) gp38 þ fibroblastic reticular cells (FRC) reported IL-33 in naïve animals (IL-33-GFP þ ), whereas CD31 þ blood endothelial cells (BEC) were virtually IL-33-GFP-negative ( Supplementary Fig. 4c). Red fluorescent artLCMV-tom but not rLCMV-tom immunization yielded a distinct population of infected and IL-33-expressing (tomato þ IL-33-GFP þ ) FRCs, confirming that replicating artLCMV but not replication-deficient rLCMV infected IL-33-expressing FRCs in spleen (Fig. 5d). In addition, both rLCMV and artLCMV infected a sizeable number of splenic BECs (Fig. 5d). A subset of artLCMV-infected BECs also reported IL-33, whereas rLCMV-infected BECs were virtually uniformly IL-33-GFP-negative (Fig. 5d). IL-33-reporting BECs were highly overrepresented amongst those infected by artLCMV ( Supplementary Fig. 4d), suggesting that artLCMV infection either had a predilection for a rare subset of IL-33-expressing BECs or that it triggered IL-33 expression in infected BECs. Both artLCMV-infected (tom þ ) IL-33-GFP þ BEC and FRC frequencies declined sharply between day 3 and 7 after immunization (Fig. 5d). This observation lent support to the hypothesis that bioactive IL-33 was released from dying artLCMV-infected lymphoid tissue stromal cells, offering a mechanism whereby replicating viral delivery systems supply IL-33 to the ensuing CTL response inside secondary lymphoid organs.

IL-33-driven tumour control upon artLCMV immunotherapy.
To compare the efficacy of several viral vector platforms in cancer immunotherapy, we first exploited a transplantable OVAexpressing tumour model (EG7-OVA). Unlike rLCMV-OVA and rAd-OVA, which showed partial or no clinical benefit, respectively, when administered to mice with an established solid tumour, treatment with artLCMV-OVA afforded substantial tumour control and prolonged the animals' survival in a CD8 þ T-cell-dependent manner (Fig. 6a,b, Supplementary Fig. 5a,b). Importantly, an irrelevant artLCMV vector (artLCMV-GFP) LCMVwt artLCMV failed to prevent tumour progression, indicating that tumour antigen-specific T-cell induction was essential for clinical efficacy (Fig. 6a,b). Moreover, artLCMV-OVA-based tumour immunotherapy of EG7-OVA tumours was only effective in ST2-sufficient wt animals but failed in Il1rl1 À / À mice ( Fig. 6c-f). These differences correlated with OVA-specific CTLs and CTL eff , which were 4tenfold higher in the blood of artLCMV-OVA-treated wt mice than in analogously treated Il1rl1 À / À mice ( Supplementary  Fig. 5c), altogether attesting to the critical function of the IL-33-ST2 axis in artLCMV-based cancer immunotherapy. To assess the potency of artLCMV-based immunotherapy in a mouse tumour model without artificially introduced non-self antigens, we exploited the P815 mastocytoma model. Immunotherapy with artLCMV expressing the cancer testis (self) antigen P1A (artLCMV-P1A) significantly delayed the growth of established subcutaneous tumours and prolonged the animals' survival, whereas replication-deficient rLCMV-P1A and irrelevant artLCMV-GFP were ineffective (Fig. 6g,h). Of note, P815 tumours were refractory to anti-PD1 checkpoint inhibition but responded to artLCMV-P1A (Fig. 6i,j). Anti-PD1 unresponsiveness of P815 tumours has been previously documented and may be due to several mechanisms including insufficient expression of PD-1 ligands, as reported 41 . Accordingly, the combination of artLCMV-P1A and anti-PD-1 checkpoint blockade was not superior to artLCMV-P1A monotherapy ( Supplementary  Fig. 5h).
artLCMV-induced CTL drive inflammatory conversion of tumours. CTL infiltrates predict survival in many human cancers 7,9 and the recruitment of circulating specific CTLs into the tumour   represents an important goal of cancer immunotherapy. artLCMV-OVA immunotherapy yielded higher numbers of circulating OVA-specific CTLs and CTL eff in blood of EG7-OVA tumour-bearing mice than did artLCMV-GFP, rAd-OVA or rLCMV-OVA (Fig. 7a). Flow cytometry and immunohistochemistry documented also that artLCMV-OVA immunotherapy yielded significantly higher densities of tumourinfiltrating CTLs (TILs), OVA-specific TILs and tumourinfiltrating OVA-specific CTL eff (Fig. 7b,c). Analogously, artLCMV-P1A immunotherapy augmented P1A-specific CTLs and CTL eff in blood of tumour-bearing mice and in tumour tissue ( Supplementary Fig. 5d-f). The administration of irrelevant artLCMV-GFP resulted in only modestly elevated TIL numbers and was clinically ineffective, corroborating that artLCMV immunotherapy operated in a largely antigen-specific manner (Fig. 7b,c). Accordingly, immunohistochemical analysis did not reveal artLCMV antigen in the tumour, while viral antigen was clearly detected in spleen ( Supplementary Fig. 5g), arguing against an oncolytic effect of artLCMV. Inflammatory gene expression profiling of EG7-OVA tumours on day 9 after artLCMV-OVA immunotherapy identified 30 genes, which were significantly different from tumours of untreated control mice. Thereof 20 were IFN-inducible (Fig. 7d). Validation by TaqMan RT-PCR confirmed that artLCMV-OVA immunotherapy induced several inflammatory mediators namely including the chemokines Ccl5, Cxcl9, Ccl4 and Cxcl10, which are predictive of prolonged survival in human cancers (Fig. 7e) [42][43][44] . Of note, expression of CCL5 was significantly higher in tumours of artLCMV-OVA-treated animals than in those receiving rLCMV-OVA or rAd-OVA (Fig. 7e, similar trend for Cxcl9). Importantly, however, only tumours of artLCMV-OVA immunized but not of artLCMV-GFP control-treated animals exhibited substantial chemokine induction. This indicated that OVA-specific CTL eff infiltration was essential for (a-e) We implanted EG7-OVA tumour cells subcutaneously into the flank of C57BL/6 mice. On day 7, when tumours became palpable, we treated them with artLCMV-OVA, artLCMV-GFP, rLCMV-OVA, rAd-OVA or left them untreated. Peripheral blood (a) and tumour tissue (b-e) were analysed on d8 and d9 after treatment, respectively. We enumerated OVA-tetramer-binding total CTLs and CTL eff (Klrg1 þ ) in blood (a) and tumour (b) as indicated. CTLs were gated as live CD8 þ CD4 À CD3 þ B220 À lymphocytes. For normalization to EG7 tumour cells the latter were differentiated from infiltrating inflammatory cells by size and granularity. CTL eff were identified as Klrg1 þ . Bars represent the mean þ s.e.m. of five (artLCMV-OVA) or four mice (all other groups). N ¼ 2. inflammatory activation of the tumour microenvironment in artLCMV-OVA immunotherapy whereas the accompanying non-specific inflammation, which was also induced by artLCMV-GFP, was insufficient to mediate these effects in full.

Discussion
These results identify virally delivered alarmin signals as key drivers of protective CTL eff responses in vectored cancer immunotherapy. The function of such damage-associated molecular patterns appears non-redundant with PAMPs, which can also be provided by replication-deficient viral delivery systems 45,46 . As schematically outlined in Supplementary Fig. 6, the ability of artLCMV to trigger IL-33/ST2-dependent CTL eff induction correlated with the vectors' spread into IL-33-expressing lymphoid stroma cells. Subsequent death of these cells, possibly as a consequence of CTL attack 47 , offers a likely mechanism for IL-33 release to neighbouring T cells. Activated CTLs inside secondary lymphoid organs express ST2, and IL-33 sensing is known to potentiate their expansion, effector differentiation and survival 11 . The present observations suggest that upon emigration to the tumour, these alarmin-imprinted CTLs contribute essentially to tumour microenvironment changes and to tumour control, thus providing a mechanistic rationale for the exploitation of replicating delivery platforms in the fight against cancer. Besides renewed interest in the tumour immunotherapy field, replicating live vaccine approaches currently experience a revival in indications such as Ebola hemorrhagic fever 48 , tuberculosis 49 and immunodeficiency virus infection 50 . These infectious diseases and cancer have in common that potency is rate-limiting, outweighing potential concerns related to the safety profile of replicating vector systems or to the release of genetically modified organisms into the environment. Accordingly, the safety profile of genetically engineered live viruses has become acceptable for oncolytic virus therapy, with a licensed product already on the market 1 . The development of artLCMV represents an innovative addition to a limited quiver of replicating viral vaccine delivery systems. Several mechanistic features suggest it holds promise for tumour immunotherapy. Efficient DC targeting results in potent CTL priming. In vivo spread and infection of IL-33-expressing lymphoid stromal cells unleashes the IL-33-ST2 alarmin pathway, thereby augmenting CTL function and effector differentiation 11 . Low LCMV seroprevalence in the human population 21,22 predicts high response rates. Inefficient vectorneutralizing antibody induction 38 facilitates repeated vector re-administration. The intravenous administration of LCMV is safe in humans and non-human primates 18,51 .
In recent years, checkpoint inhibitors such as anti-PD-1/PD-L1 or anti-CTLA-4 antibodies had groundbreaking success in a variety of malignancies 2 . However, immune checkpoint blockade merely disinhibits ongoing T-cell responses and, by consequence, tends to fail in tumours with a paucity of pre-existing tumourinfiltrating CTLs 52,53 . Such 'cold' tumours can be the result of immunoediting and T-cell escape 54 . Alternatively, tumours can exhibit immunogenic determinants but fail to induce clinically significant CTL responses, which is referred to as immune exclusion or ignorance 14,53 . As a powerful tool for active immunization artLCMV immunotherapy delivered significant numbers of tumour-reactive CTL eff to the tumour bed, resulting in an inflammatory conversion of the tumour microenvironment (Fig. 7, Supplementary Fig. 5). This transition is thus predicted to render the corresponding tumours more responsive to immunomodulatory therapy.
Beyond the commonly benign nature of human LCMV infection and a lack of horizontal transmission in humans, inefficient co-packaging of the three artLCMV genome segments represents a molecularly defined mechanism of attenuation. The 41,000-fold increase in i.c. mouse LD 50 exceeds the B100-fold safety margin of the clinically used, live-attenuated Junin arenavirus vaccine Candid#1 (ref. 55). Last but not least, genetic stability (Fig. 2a-d) should enable industrial exploitation. During scale-up in batch production, but also upon administration to vaccinees, rearranged genomes with a fitness gain are readily selected 32,56 , compromising product safety as well as efficacy 29,30 . Accordingly, stable transgene expression and attenuation of artLCMV, together with the simplicity and rapidity of vector generation, represent critical assets for clinical translation and the vector's exploitation in personalized medicine approaches 57 .
Taken together, our study identifies alarmin signals as crucial for the induction of protective anti-tumour CTL eff responses in vectored immunotherapy. By demonstrating mechanistically that alarmin release depends on viral in vivo spread, our work suggests that live viruses deserve their place in the rapidly evolving array of cancer treatment modalities. artLCMV may represent the prototype of a novel class of live microbial delivery systems, which leverage not only PAMPs but also alarmin release for CTL eff differentiation and enhanced anti-tumour efficacy. , and GP-expressing 293T cells have previously been described 26,58 . All cell lines were tested mycoplasma negative.

Methods
Viruses and virus neutralization test. The origin, passage and titration of the LCMV strains Armstrong and Clone 13 strains have been described 11 . GP-IFF has previously been described 58 . For GFP-IFF we used rat-anti-GFP antibody (Biolegend, 1:2,000 dilution) as primary antibody. Replication-competent LCM viruses were propagated on BHK-21 cells, rLCMV vectors on 293T-GP cells 26 . Growth curves of LCM viruses were performed on BHK-21 cells at a multiplicity of infection of 0.01. The reverse genetic engineering of rLCMV, r3LCMV and artLCMV vectors has been described 26,27 . Entire ORFs of the cancer-testis antigen P1A (comprising the immunodominant LPYLGWLVF epitope), GFP, dTomato, OVA (comprising the immunodominant SIINFEKL epitope) and Cre were used for insertion into the respective vectors and viruses. artLCMV-Her2 contained a ubiquitin-based expression cassette 59 delivering the immunodominant TYLPANASL miniepitope. Recombinant vaccinia virus and E1-deleted recombinant adenovirus 5-based vectors expressing ovalbumin (rVACC-OVA, rAd-OVA) have previously been described 60,61 . A genetically tagged GPC-encoding S segment cDNA (pol-I-S-GPC tag ) was generated by codon-optimizing the 255 C-terminal nucleotides of the viral GPC ORF, creating a novel ORF named 'GPC tag '. To create a S rec #1 expression plasmid, we substituted the IGR in pol-I-S-GPC tag for a synthetic cDNA fragment (Genscript) as described in Supplementary Fig. 2e. The neutralizing capacity of mAb and immune serum was analysed in immunofocus reduction assays 11 .
Animal experimentation and ethics statement. Il33 À / À mice 62 were obtained through the RIKEN Center for Developmental Biology (Acc. No. CDB0631K; http://www.cdb.riken.jp/arg/mutant%20mice%20list.html). AGRAG mice (IFNa/bR À / À , IFNgR À / À , RAG1 À / À triple-deficient) 63 , Il1rl1 À / À (ref. 64) and ZP3-Cre mice 65 have been described. IL33 reporter mice (Il33 gfp/wt ) were generated by intercrossing IL-33 À / À mice with the ZP3-Cre germ line deleter strain to remove the neomycin cassette. C57BL/6, BALB/c and DBA/2 wild-type mice were either purchased from Charles River and Janvier Labs or were bred at the Institut für Labortierkunde of the University of Zurich, Switzerland under specific pathogen-free conditions. Unless specified otherwise, rAd-OVA vectors and rVACC-OVA were administered in their respective optimal dose range of 10 8 viral particles i.v. and 10 6 PFU i.v., respectively 66,67 . The same respective doses were used for intramuscular (rAd-OVA) and intradermal or subcutaneous (rVACC-OVA) immunizations in the experiments to Supplementary Fig. 2. artLCMV and r3LCMV were routinely given at a dose of 10 5 PFU i.v., except for virus tracing experiments in C57BL/6 mice (10 6 PFU) and experiments in AGRAG mice (10 4 PFU). rLCMV vectors were used at 10 5 PFU for standard immunization experiments or at 10 6 PFU for immunization with tumour self antigens, for tumour immunotherapy and for Cr 51 release experiments. Except for prime-boost immunization experiments in Fig. 4j,k, animals were given only a single shot of viral vector or virus, respectively. LCMVwt was administered i.v. at the same dose as artLCMV was used. Intracranial LCMV challenge was performed through the skull and animals developing terminal disease were euthanized by CO 2 inhalation in accordance with the Swiss law. PD-1 blocking antibody (clone RMP1-14, from BioXcell) was administered at a dose of 12.5 mg kg À 1 intraperitoneally. CD8 þ T cells were depleted by injecting 200 mg anti-CD8 antibody (YTS169, from BioXcell) intraperitoneally. Animal experiments were performed at the Universities of Geneva and Basel in accordance with the Swiss law for animal protection. Permission was granted by the Direction générale de la santé, Domaine de l 0 expérimentation animale of the Canton of Geneva and by the Veterinäramt Basel-Stadt, respectively. Experimental groups were sex-and age-matched. Animals in tumour therapy experiments were assigned to groups in a manner to assure even distribution of tumour volumes between groups at the time of tumour therapy. The groups were neither randomized, nor were experiments conducted in a blinded manner. The Swiss law for animal protection requires that mice with wounds on the tumour or exhibiting signs of distress (evident namely in lethargy, hunchback, piloerection, emaciation and agonal breathing) be euthanized by CO 2 inhalation irrespective of tumour size and diameter. Accordingly, animals not reaching humane endpoints of tumour volume or diameter were excluded from survival curves.
Tumour transplantation and measurements. Suspensions containing 10 6 tumour cells were implanted subcutaneously in the right flank. Tumour growth was monitored three times per week and the longest ('length') and the shortest diameter ('width') were measured using a caliper. Tumour volumes (mm 3 ) were calculated as 1/2 (length*width^2). In accordance with the Swiss law for animal protection, mice were euthanized when tumour volumes exceeded 1,500 mm 3 or when the longest median tumour diameter exceeded 20 mm.
Immunostained bright-field sections and fluorescence sections of virus-infected tissues were scanned using a Panoramic Digital Slide Scanner 250 FLASH II (3DHISTECH). White balance was adjusted and contrast was linearly enhanced using the tools 'levels', 'curves', 'brightness' and 'contrast' in Photoshop CS6 (Adobe).
Primary ex vivo cytotoxicity assays. Primary ex vivo cytotoxicity of splenocytes was determined in a 5-h assay on Cr51-loaded EL-4 target cells pulsed with SIINFEKL peptide according to established procedures 69 . Cytotoxicity on unlabelled target cells was subtracted to obtain specific lysis. Lytic units were determined according to established procedures from dose-response curves 69 ; one lytic unit was calculated to represent the minimal number of spleen cells that lysed 410% of the target cells.
IFN-a ELISA. Serum interferon-alpha levels were determined by ELISA using the Verikine Mouse Interferon Alpha ELISA Kit (PBL Assay Science, cat. no. KMC4011).
Statistical analysis. For statistical analysis, GraphPad Prism software (Version 6.0, GraphPad Software) was used. Differences between two groups were assessed using unpaired two-tailed Student's t-tests. Single values of multiple groups were compared by one-way analysis of variance (ANOVA), followed by Bonferroni post hoc test or Dunnett's post hoc test when comparing against a reference group. A Bonferroni correction was made when comparing multiple parameters. Two-way ANOVA with Bonferroni post hoc test was used to compare multiple groups with multiple measurements. Survival curves were analysed by log-rank tests with Bonferroni correction. For the comparison of tumour growth curves, the area under the curve was compared 70 . The analysis ended when the first animals reached human endpoints.
Viral load data were log-converted to obtain a near-normal distribution prior to statistical analysis. Variances within different groups in a given experiment were similar. In accordance with current standard practice in the field of viral immunology, only very substantial differences were reported, obviating the need for variance testing.
Data availability. Nanostring data that support the findings of this study have been deposited in National Center for Biotechnology Information Gene Expression Omnibus (GEO) with the accession code GSE84039. The nucleotide sequences of recombinant and recombined LCMV segments have been deposited in GenBank with the accession codes KX462116-KX462128. All relevant data are available from the authors upon request.