IFN-γ extends the immune functions of Guanylate Binding Proteins to inflammasome-independent antibacterial activities during Francisella novicida infection

Guanylate binding proteins (GBPs) are interferon-inducible proteins involved in the cell-intrinsic immunity against numerous intracellular pathogens. The molecular mechanisms underlying the potent antibacterial activity of GBPs are still unclear. GBPs have been functionally linked to the NLRP3, the AIM2 and the caspase-11 inflammasomes. Two opposing models are currently proposed to explain the GBPs-inflammasome link: i) GBPs would target intracellular bacteria or bacteria-containing vacuoles to increase cytosolic PAMPs release ii) GBPs would directly facilitate inflammasome complex assembly. Using Francisella novicida infection, we investigated the functional interactions between GBPs and the inflammasome. GBPs, induced in a type I IFN-dependent manner, are required for the F. novicida-mediated AIM2-inflammasome pathway. Here, we demonstrate that GBPs action is not restricted to the AIM2 inflammasome, but controls in a hierarchical manner the activation of different inflammasomes complexes and apoptotic caspases. IFN-γ induces a quantitative switch in GBPs levels and redirects pyroptotic and apoptotic pathways under the control of GBPs. Furthermore, upon IFN-γ priming, F. novicida-infected macrophages restrict cytosolic bacterial replication in a GBP-dependent and inflammasome-independent manner. Finally, in a mouse model of tularemia, we demonstrate that the inflammasome and the GBPs are two key immune pathways functioning largely independently to control F. novicida infection. Altogether, our results indicate that GBPs are the master effectors of IFN-γ-mediated responses against F. novicida to control antibacterial immune responses in inflammasome-dependent and independent manners.

Introduction demonstrated for IFN-γ-inducible Immunity-Related GTPases (IRG) in their antimicrobial role against Toxoplasma gondii [17], GBPs might disrupt the Salmonella-containing vacuole leading to the release of this bacterium and its associated LPS into the host cytosol [10]. Alternatively, GBPs might orchestrate the recruitment of IRGB10 onto cytosolic bacteria to liberate LPS for sensing by caspase-11 [16]. Finally, as GBPs can promote caspase-11 activation without any detectable recruitment around Chlamydia muridarum inclusions and upon Legionella LPS transfection into the host cytosol, Coers and colleagues suggested that GBPs might directly facilitate caspase-11 activation [11,12].
The link between GBPs and the AIM2 inflammasome is less controversial and has been mostly studied by our group and others in macrophages infected with Francisella novicida [13,15,18]. F. novicida is a close relative of F. tularensis, the agent of tularemia. The virulence of Francisella strains is linked to their ability to rapidly lyse the phagosome, escape into the host cytosol [19,20] and replicate within this compartment. This process is dependent on a cluster of genes in the Francisella-pathogenicity island [21,22], which encodes an atypical type VI secretion system [23]. GBP2 and GBP5 are recruited onto cytosolic F. novicida and are required to lyse bacteria and release the bacterial genomic DNA into the host cytosol. DNA in the host cytosol is then recognized by AIM2 [13,15,[24][25][26]. Recently, GBPs were demonstrated to be required for IRGB10 recruitment onto cytosolic F. novicida to mediate cytosolic bacterial killing, DNA release and AIM2 inflammasome activation [16].
Chromosome 3-encoded GBPs and the AIM2 inflammasome are both equally required to resist F. novicida infection in vivo [13,15]. Yet, whether GBPs might have anti-F. novicida functions dependent on other inflammasomes or inflammasome-independent antibacterial responses remain unclear [10,16]. In this work, we demonstrate that GBPs control F. novicidamediated host cell death in a hierarchical manner implicating at least 3 different canonical and non-canonical inflammasome complexes as well as apoptotic caspases 8, 9 and 3. Furthermore, we demonstrate that upon IFN-γ treatment, GBPs control F. novicida replication independently of the canonical and non-canonical inflammasome pathways and independently of macrophage cell death. IFN-γ-, GBPs-mediated inhibition of intracellular bacterial growth was also effective during F. tularensis spp. holarctica Live Vaccine Strain infection, while IFN-γ was inefficient to block the replication of the highly virulent F. tularensis SCHU S4 strain. Finally, we demonstrate in vivo that IFN-γ-mediated host protection against F. novicida is largely GBPs-dependent and inflammasome-independent. Our work thus positions the GBPs as the master effectors of the IFN-γ-mediated anti-F. novicida responses.

Results
Gbp chr3 and the inflammasome control in vivo host resistance to F. novicida in a non-redundant manner We and others have previously reported that Gbp chr3 -deficient and inflammasome-deficient mice (Aim2 -/-, Asc -/and Casp1/Casp11 -/mice) were highly susceptible to F. novicida infection [13,15,24,25,27,28]. In vitro studies have demonstrated that GBPs act upstream of the AIM2 inflammasome [13,15,16] suggesting that the Gbp chr3 deficiency should phenocopy deficiencies in the AIM2 inflammasome. In a survival experiment, Gbp chr3 -KO and Asc -/were almost as susceptible to F. novicida infection (Fig 1A) although we consistently noticed that Gbp chr3 KO mice died slightly faster than Asc -/mice. Similarly, Gbp chr3 -KO and Asc -/mice displayed very high bacterial burden both in the spleen and in the liver at day 2 post-infection (PI) with an average of 40-fold more bacteria than WT mice ( Fig 1B). Surprisingly, at 48 h PI, IFN-γ level in the serum of Gbp chr3 -deficient mice reached WT levels while in agreement with previous work [28], IFN-γ level in the serum of Asc -/mice was strongly decreased (Fig 1C).
We and others have previously reported that early (16 h PI [13], 24 h PI [16]) IL-18 production in vivo is GBPs-dependent, a finding which we reproduced here ( Fig 1D). However, at later time points (48 h PI), IL-18 levels were similar in Gbp chr3 -KO and in WT mice ( Fig 1E). As expected, IL-18 levels in infected Asc -/mice were not statistically different from the levels observed in uninfected mice. These results indicated that inflammasome activation was only delayed in vivo in Gbp chr3 -deficient mice. The increase in IL-18 levels observed over 48h in infected Gbp chr3 -KO mice likely explained the high IFN-γ serum level observed in these mice. The high susceptibility of Gbp chr3 -KO mice to F. novicida infection despite high levels of circulating IFN-γ is remarkable since IFN-γ is considered to be one of the most important cytokine to fight Francisella infection [29][30][31]. This conundrum led us to hypothesize that GBPs might be the main IFN-γ effector in F. novicida-infected mice. Furthermore, this result indicates that while the overall susceptibility of Gbp chr3 -deficient mice and Asc -/mice to F. novicida infection are similar, the in vivo antibacterial mechanisms of GBPs are, at least partially, independent of the inflammasome.

IFN-γ priming induces a quantitative shift in GBPs levels
The action of GBPs against F. novicida has been mostly studied in unprimed macrophages [13,15,16]. Under these conditions, GBPs induction relies on the endogenous recognition of nucleic acids by the cGAS pathway, secretion of type I IFN, signaling through the type I IFN receptor (IFNAR1) and activation of the IRF-1 pathway [15,24,32]. Importantly, there was a drastic quantitative shift in GBP2 and GBP5 transcripts levels upon priming with IFN-γ compared with F. novicida-mediated endogenous induction or to induction following IFN-β priming (S1A Fig). Indeed while GBP2 transcript levels increased by a factor of 15 upon F. novicida infection, IFN-γ priming of infected macrophages, led to a 325-fold increase in GBP2 transcript levels relative to its level in uninfected macrophages. As previously reported, this very strong induction likely results from synergistic NF-κB and IFN signaling [14,33,34]. Indeed, we obtained comparable levels of GBP2 induction when BMDMs were primed with both IFNγ and Pam 3 CSK 4 , a TLR2 agonist (S1A Fig). ProIL-1β transcript levels were not impacted by IFN-γ priming (S1C Fig) indicating that this synergy was specific for GBPs induction. Similar results were obtained while investigating GBP5 transcript levels (S1B Fig) or while monitoring GBP2 and GBP5 protein levels (S1D Fig). This quantitative shift in GBPs levels upon IFN-γ treatment led us to investigate the impact of IFN-γ on the F. novicida-mediated cell death.  GBPs control F. novicida-mediated cell death in an AIM2 inflammasomedependent and -independent manner In F. novicida-infected bone marrow-derived macrophages (BMDMs), the only reported cell death pathways are dependent on the AIM2/ASC complex [24,25,27,35]. By monitoring, in real time, propidium iodide influx and fluorescence over 24 h in unprimed macrophages deficient for various inflammasome components, we observed substantial differences in the kinetics of propidium iodide between various knock-out macrophages suggesting that several cell death pathways were engaged following F. novicida infection (Fig 2A). These differences in the kinetics of propidium iodide incorporation/ fluorescence were demonstrated to be statistically significant by calculating the area under the curve corresponding to each kinetics ( Fig 2B). As previously described [13,15,24,25], in the absence of IFN-γ priming, F. novicida-infected BMDMs died in an AIM2-dependent manner. Indeed, at MOI 10, propidium iodide incorporation/fluorescence sharply increased around 6 h post-infection in WT macrophages while Aim2 -/-BMDMs presented cell death kinetics delayed by more than 7 h compared with that of WT macrophages. Interestingly, incorporation/fluorescence increase of propidium iodide was significantly further delayed in Gbp chr3 -KO BMDMs compared to Aim2 -/-BMDMs suggesting that GBPs control both AIM2-dependent and -independent cell death pathways. Accordingly, by monitoring macrophage cell death in real time and in single cells using time-lapse microscopy (Fig 2E and 2F; S3 Fig), we clearly observed that the number of propidium iodide-positive cells increased significantly later in Gbp chr3 -KO BMDMs compared to Aim2 -/-BMDMs. This difference in cell death kinetics was exacerbated in the presence of IFN-γ priming (Fig 2C, 2D and 2F). Finally, to exclude any bias associated with propidium iodide incorporation, we used a luminescent cell viability assay based on the quantitation of ATP, a signature of metabolically active cells (CellTiter-Glo; Fig 2G). This assay confirmed that Gbp chr3 -KO BMDMs survived longer than Aim2 -/-BMDMs upon F. novicida infection. Altogether, our data strongly suggest that GBPs control both AIM2-dependent and -independent cell death/survival pathways.
To assess whether the non-canonical caspase-11 inflammasome could be involved in the AIM2-independent detection of F. novicida, we used siRNA against caspase-11. We observed a consistent and significant delay in Aim2 -/macrophage death upon treatment with a caspase-11 siRNA (Fig 3C and 3D    of caspase-11 in the immune response of Aim2 -/macrophages was further strengthened by investigating IL-1β release. Indeed, caspase-11 expression knock-down strongly decreased the late secretion of IL-1β observed in Aim2 -/macrophages ( Fig 3F). Nlrp3 expression knockdown or knock-out also consistently decreased IL-1β release in Aim2 -/macrophages (Fig 3F  and 3G). This decrease may be partly due to its involvement downstream of caspase-11 [37].
Altogether, our data demonstrate the involvement of at least two sequential cell death pathways (mediated by the sensors AIM2, NLRP3 and caspase-11) elicited in response to F. novicida infection. The extensive survival of Gbp chr3 -KO BMDMs strongly suggests that GBPs contribute to activation of these three inflammasome complexes during F. novicida infection. Of note, at 24 h post-infection in absence of IFN-γ, propidium iodide incorporation/ fluorescence was similar in Asc -/and Gbp chr3 -KO BMDMs. The late cell death occurring in Gbp chr3 -KO BMDMs was associated with IL-1β release while as expected no IL-1β was observed in the supernatant of Asc -/-BMDMs (S5 Fig). This result suggests that while chromosome 3-encoded GBPs are instrumental in promoting fast inflammasome activation upon F. novicida infection, they are not strictly required to trigger inflammasome activation likely explaining the bi-phasic dependence of IL-18 serum level on GBPs (Fig 1D and 1E). Interestingly, analysis of macrophages deficient for both Asc and Gbp chr3 demonstrated a strong delay in propidium iodide incorporation/fluorescence increase compared to that of Asc -/and Gbp chr3 -KO BMDMs (S6A and S6C Fig) providing genetic evidence that GBPs can act independently of the canonical inflammasomes.
This conclusion was further strengthened by comparing cell death kinetics, morphological and molecular features of IFN-γ-primed Asc -/and Gbp chr3 -KO BMDMs. Priming with IFN-γ accelerated the cell death kinetics in the different macrophages with one notable exception ( Fig 2C-2F). Indeed, we did not detect any substantial propidium iodide incorporation/fluorescence increase in IFN-γ-primed Gbp chr3 -KO macrophages while IFN-γ-primed Asc -/-BMDMs died although with delayed kinetics compared with that of WT macrophages ( Fig  2C-2F, S3B Fig). The death of Asc -/-BMDMs was confirmed by addition of triton (TX100) at the end of the 24 h kinetics. Indeed, TX100 treatment did not further increase propidium iodide incorporation/fluorescence in Asc -/-BMDMs (S2B Fig). In contrast, TX100 addition in Gbp chr3 -KO BMDMs led to a strong increase in propidium iodide fluorescence indicating that the plasma membrane of most Gbp chr3 -KO BMDMs was still intact (and not permeable to propidium iodide) at 24 h PI. The relatively lower intensity of propidium iodide fluorescence in Asc -/-BMDMs compared to WT BMDMs might be related to intrinsic differences in propidium iodide incorporation/fluorescence upon incorporation in apoptotic vs. pyroptotic nuclei. Cell death of IFN-γ-primed Asc -/-BMDMs was associated with cell retraction (S2 Movie and Fig 4C) and was reminiscent of the morphological cell death observed in Casp1/Casp11 -/-BMDMs although it progressed with a %6 h delay compared to the latter BMDMs. Accordingly, apoptotic caspase-8/9/3 processing and DEVDase activity were clearly detectable in Asc -/-BMDMs infected for 16 h corroborating the morphological features and demonstrating that at late time points of infection, Asc -/-BMDMs die by apoptosis.
The extensive survival of Gbp chr3 -KO BMDMs upon IFN-γ priming was confirmed using CellTiter-Glo assay ( Fig 2G). Furthermore, we observed a limited number of cell death events in infected Gbp chr3 -KO BMDMs using time-lapse microscopy ( Fig 2F, S3 Fig and S2 Movie) suggesting that a limited GBPs-independent cell death pathway occurs at late time points of infection. The cell survival effect of IFN-γ on Gbp chr3 -KO BMDMs (compare Figs 2A and 2B with 2C and 2D; S3A with S3B and S6A with S6B, 2F and 2G, S3C and S6C) correlates with a decrease in IL-1β release (S5 Fig). This result suggests that in infected Gbp chr3 -KO BMDMs, IFN-γ might inhibit inflammasome activation possibly via nitrosylation, an IFN-inducible mechanism previously reported to inhibit inflammasome activation [42,43]. While the three different cell death/survival assays showed subtle differences [44], these assays converge to indicate that upon IFN-γ priming, Gbp chr3 -KO BMDMs survive better than Casp1/Casp11 -/and Asc -/-BMDMs suggesting that GBPs control both pyroptotic and apoptotic pathways. Importantly, the GBPs-mediated control of apoptosis was confirmed by generating mice doubly deficient for Asc and Gbp chr3 . Macrophages deficient for both Asc and Gbp chr3 failed to demonstrate caspase-8/9 and 3 maturation and DEVDase activity in contrast to Asc -/macrophages (S6D and S6E Fig). Altogether, our results demonstrate that GBPs act as major cell death regulators by controlling numerous pyroptotic and apoptotic cell death pathways.

IFN-γ-induced GBPs control bacterial killing independently of inflammasomes
The high IFN-γ levels observed in Gbp chr3 -KO mice in the mouse model of tularemia coupled to the inability of Gbp chr3 -KO mice to control F. novicida burden (Fig 1) led us to investigate if GBPs contribute to the IFN-γ-mediated growth restriction observed in vitro [45]. As previously described [13,15,24,35], we observed a robust F. novicida replication in unprimed macrophages ( Fig 5A corresponding to the Raw data presented in S7 Fig), which was partially controlled in a GBP-dependent manner by the inflammasome. IFN-γ priming led to F. novicida killing as visualized by a net decrease in the recovered intracellular colony forming units ( Fig 5A) as soon as 6 h PI. Bacterial killing was highly dependent on GBPs. Indeed, F. novicida was not killed in IFN-γ-primed Gbp chr3 -KO macrophages, but robustly replicated by a 40-fold over 12 h of infection. We did not observe any substantial effect of IFN-γ on phagosomal rupture (S8 Fig) suggesting that IFN-γ does not act by restricting F. novicida access to its cytosolic niche. This finding is consistent with a previous study demonstrating a direct activity of IFN-γ on cytosolic F. tularensis [45]. Furthermore, the IFN-γ-mediated bacterial growth inhibition was independent of the NADPH oxidase and of the IFN-γ-inducible NO synthase (iNOS also known as NOS2), two immune effectors described to act downstream of GBPs [46] (S9C- S9E  Fig). In WT macrophages, the GBPs-mediated bacterial killing was difficult to segregate from the antibacterial effects of host cell death due to the very rapid inflammasome activation (see Fig 2B). Yet, bacterial killing was also observed in IFN-γ-primed Asc -/macrophages infected for 6 h at a MOI of 1 in the absence of substantial pyroptotic and apoptotic cell death. To assess whether IFN-γ-mediated inhibition of bacterial growth was dependent or independent of cell death we took advantage of flow cytometry using GFP-expressing F. novicida and propidium iodide to exclude dead cells.
In the absence of IFN-γ priming, replication was observed in %30% of live Asc -/and %50% Gbp chr3 -KO macrophages (as determined by the number of propidium iodide -GFP + cells). In the presence of IFN-γ, this number dropped to less than 4% in Asc -/macrophages indicating a robust bacterial growth restriction in these cells. In contrast, a large number of Gbp chr3 -KO macrophages (>40%) sustained F. novicida replication despite IFN-γ priming (Fig 5B). These results suggest that this anti-bacterial GBP-dependent mechanism proceeds independently of cell death at least in Asc -/-BMDMs.
To quantify F. novicida replication in a large number of macrophages, we analyzed infected macrophages using high-resolution microscopy in Flow (ImagestreamX). This single cell quantification technique further demonstrated that in the absence of Gbp chr3 , IFN-γ was almost ineffective to control bacterial replication. Indeed, in the presence of IFN-γ, there was a 100-fold increase in the number of Gbp chr3 -KO macrophages permissive for bacterial replication (containing more than 100 bacteria) compared with that of Asc -/macrophages (Fig 5C,  S9F Fig). Representative immunofluorescence images of this striking phenotype are presented in Fig 5D. While this difference was exacerbated in the presence of IFN-γ and as previously noticed using Aim2 -/macrophages [13], we observed that in unprimed macrophages, GBPs also controlled in an ASC-independent manner the bacterial burden.
GBP5-associated F. novicida loose their GFP expression, which has been associated with a loss of bacterial viability [15]. We thus quantified GFP intensity in single bacteria in propidium iodide-negative BMDMs as a surrogate marker of bacterial viability/metabolic activity ( Fig  5E). In WT macrophages, IFN-γ treatment led to a significant reduction in the average GFP intensity (-10.2%) of intracellular F. novicida with a large increase (30%, n = 640) in the number of bacteria expressing low GFP levels (as defined by a GFP intensity <80% of the average intensity of bacteria in unprimed WT macrophages). Similarly, IFN-γ priming significantly reduced the average GFP intensity of bacteria in Asc -/-BMDMs (-8.4%) with a large increase (+19%, n = 843) in the number of low GFP-expressing bacteria. Surprisingly, IFN-γ priming had a paradoxical effect on bacterial GFP expression in Gbp chr3 -KO BMDMs with an increase in the average GFP intensity (+20.6%) and a strong decrease in the number of low GFPexpressing bacteria (-30%, n = 658). While this paradoxical increase (also visible in Fig 5D) remains to be understood, this experiment demonstrates that IFN-γ-induced GBPs in live macrophages affect the metabolic activity of bacteria. The decrease GFP expression observed in the presence of IFN-γ and GBPs is likely due to GBP-mediated antibacterial activity expressing F. novicida at a MOI of 10, assessed by high-resolution microscopy in flow and presented as a comparison of both genotypes, with bacteria-per-cell values grouped by increments of 10. (D) Immunofluorescence of BMDMs infected for 10h with GFP-expressing F. novicida at a MOI of 1 (stained with DAPI (blue) and with an antibody to F. novicida (red); scale bars: 5 μm. (E) GFP intensity was quantified in single GFPexpressing F. novicida in BMDMs from the indicated genotypes primed or not overnight with 100 U/ml of IFN-γ and infected for 10 h. Cumulative data from two independent experiments are shown. The GFP intensity of single bacteria was normalized in each experiment to the average intensity of single bacteria in unprimed WT BMDMs. Dots represent the normalized GFP intensity of a single bacteria, the bar represents the mean. although we were unable to directly quantify bacteriolysis in single cells. Importantly, the antibacterial activity of IFN-γ against F. tularensis live vaccine strain (LVS) was also fully dependent on Gbp chr3 extending our results to an attenuated strain of the F. tularensis species (Fig 5F). While an IFN-γ-dependent GBPs antibacterial activity was clearly observed upon LVS infection (Fig 5F; 23x fold reduction in WT BMDMs upon IFN-γ treatment), LVS sustained a substantial replication (%100x) in IFN-γ-primed BMDMs. More strikingly, we were unable to observe a robust IFN-γ-mediated growth restriction of F. tularensis SCHU S4 in BMDMs (S10 Fig). These results suggest that F. tularensis strains have evolved mechanisms to avoid, at least partially, IFN-γ-mediated GBPs-dependent antibacterial activity.
The synergistic roles of GBP2 and 5 in promoting the antibacterial effect of IFN-γ could be demonstrated in J774.1 macrophage-like cells using CRISPR/cas9 (S11 Fig). This result rules out that the extensive replication observed in Gbp chr3 -KO BMDMs despite the priming with IFN-γ could be due to a non-specific defect of Gbp chr3 -KO BMDMs associated with their large genomic deletion. This is well in line with control experiments previously performed on Gbp chr3 -KO mice demonstrating normal induction of IFN-inducible genes and normal susceptibility/resistance to certain pathogens including L. monocytogenes [8,16]. Altogether, these findings demonstrate that, in vitro, in infected macrophages, the anti-bacterial function of IFN-γ, a cytokine known to induce a large number of antibacterial effectors relies almost exclusively on Gbp chr3 .

rIFN-γ administration partly rescues the in vivo antimicrobial function of Asc -/mice but fails to complement Gbp chr3 deficiency
GBPs have been tightly linked to inflammasome complexes [10][11][12][13][14][15][16]. Yet our in vitro data clearly demonstrate that GBPs have potent inflammasome-independent antimicrobial functions. Particularly, in vitro, GBPs are the main IFN-γ antimicrobial effectors. In contrast, in the presence of IFN-γ, the inflammasome complex seems largely facultative to control F. novicida replication. ASC is required in vivo to induce IFN-γ via IL-18 release [27,28,47]. To further explore these potential differences in vivo, we administered rIFN-γ to F. novicida-infected mice. Early rIFN-γ administration allowed WT mice to survive F. novicida infection (Fig 6A). rIFN-γ administration clearly extended ASC -/mice survival ( Fig 6B) and strongly delayed Asc -/mice weight loss (S12 Fig), while it had a very moderate (although statistically significant) effect on Gbp chr3 -KO mice survival and weight loss (Fig 6C and 6D, S12 Fig). To assess the functional links in vivo between IFN-γ, GBPs or the inflammasomes and their impact on bacterial replication, we analyzed the bacterial burden in the spleen and the liver at 48h PI following rIFN-γ injection at day 0 and day 1 PI. rIFN-γ was highly efficient to control the bacterial burden in both WT ( Fig 6E) and Asc -/mice (Fig 6F). In contrast, upon IFN-γ injection in Gbp chr3 -deficient mice, there was no statistical reduction in the bacterial burden in the liver and the spleen (4-fold reduction in the average splenic burden in Gbp chr3 -KO mice versus a 40-fold reduction in Asc -/mice) (Fig 6F). These results demonstrate that, as observed in infected macrophages, the antibacterial action of IFN-γ is mostly mediated by GBPs in vivo and is largely independent of the canonical inflammasomes.

Discussion
With the recent discoveries of the tight links between the inflammasome complexes and the GBPs, it became unclear whether most antimicrobial functions of GBPs in infected macrophages were mediated by various inflammasome complexes or were independent of the inflammasomes [48]. Indeed, Gbp chr3 -and inflammasome-deficient mice are similarly highly susceptible to F. novicida infection. One of the current models to explain the link between GBPs and the inflammasomes positions GBPs as the molecular platform promoting the inflammasome supramolecular complex [2,3,48,49]. This model is strongly supported by evolution since GBPs from jawed fish display inflammasome-related CARD domains [14]. During F. novicida infection, we and other have previously demonstrated that GBPs are required to trigger AIM2 inflammasome activation while they are dispensable upon direct delivery of DNA into the host cytosol [13,15]. GBPs, in cooperation with IRGB10, were demonstrated to participate in F. novicida lysis into the host cytosol [13,15,16] suggesting that GBPs mostly act to release bacterial DNA into the host cytosol. The polymeric nature of DNA might alleviate the requirement for host factors to promote AIM2 inflammasome activation. Indeed, cytosolic dsDNA could provide the scaffold for AIM2 oligomerization and subsequent inflammasome activation [50,51].
In this work, we demonstrate that GBPs are required not only for the AIM2 canonical inflammasome activation but for most of the programmed cell death pathways that can take place in F. novicida-infected macrophages. In vitro, in F. novicida-infected WT murine BMDMs, so far the only described cell death pathway was the AIM2 inflammasome. Yet, using various knock-out macrophages, we have demonstrated here that several canonical and noncanonical inflammasome complexes can be active during F. novicida infection. These alternative pathways were revealed thanks to the use of Aim2 -/-BMDMs and are masked in WT macrophages. Interestingly, Harton and colleagues recently demonstrated that F. tularensis strains, in contrast to F. novicida, elicit NLRP3 inflammasome activation in BMDMs [52]. This result suggests that F. tularensis has evolved to escape AIM2 inflammasome activation while the delayed NLRP3-dependent sensing is conserved in response to various Francisella species. The identification of a caspase-11-dependent role in mediating Aim2 -/-BMDMs death and IL-1β release in the absence of AIM2 was unexpected. Indeed, the direct delivery of Francisella LPS into BMDMs cytosol does not activate caspase-11 due to its underacylated structure ( [36]). It is still unclear whether the delayed caspase-11-dependent activation observed in Aim2 -/-BMDMs is due to sensing of F. novicida LPS or of another endogenous or bacterial ligand. The large number of inflammasome complexes activated upon F. novicida infection is reminiscent of what have been observed during infection with other intracellular bacteria [11,53,54], although F. novicida infection of murine macrophages is somewhat unique due to its high dependence on AIM2 [18]. Remarkably, IFN-γ-induced GBPs are required to trigger all these pathways suggesting that they either act upstream of the inflammasomes or that they have conserved mechanisms to facilitate activation of several (AIM2, NLRP3, caspase-11) inflammasome sensors. Such facilitation of the activation of multiple inflammasomes by GBPs has been previously observed upon Chlamydia infection [11]. Yet, GBPs-mediated control of Chlamydia-mediated cell death was only partial in contrast to what we observed upon F. novicida infection. Importantly, Asc -/-(and Casp1/Casp11 -/-) BMDMs died by apoptosis as demonstrated by morphological and molecular analyses (Fig 4, S1 and S2 Movies) [35] indicating that GBPs control both pyroptotic and apoptotic pathways. Indeed, processing of caspase-8, 9 and 3 and DEVDase activity were clearly visible in Asc -/-BMDMs but absent in Asc -/-Gbp chr3 -KO BMDMs (S6D and S6E Fig). The GBP-dependent pathway leading to apoptotic caspase activation in Asc -/-BMDMs is still unclear. Antibiotic-mediated bacteriolysis of another cytosolic pathogen (Shigella flexneri) triggers massive caspase-9-dependent apoptosis of epithelial cells [55]. Based on the antibacterial role of GBPs ( [13,15,16], this work), we speculate that GBPsmediated action releases a PAMPs that directly or indirectly triggers apoptosis in the absence of the inflammasome adaptor ASC. The concurrent maturation of both caspase-8 and caspase-9 suggests activation of both extrinsic and intrinsic (e.g. mitochondrial) apoptotic pathways. This dual activation may be due to the cross-activation of the mitochondrial intrinsic apoptosis pathway following cleavage of Bid by caspase-8 [56]. Yet, we cannot exclude a direct GBPsdependent induction of mitochondrial dysfunction such as the one occurring in S. flexneriinfected cells of non-myeloid lineages [57]. IRGs can lyse T. gondii vacuole leading to parasite permeabilisation and a caspase-1-independent necrotic death in mouse embryonic fibroblats [58]. Due to the diversity of cell death pathways controlled by GBPs and IRGs, we favor the hypothesis that these IFN-induced GTPases act as PAMPs-shedders to release/uncover various microbial cell death-activating ligands. Future studies are needed to establish the pathways linking GBPs to the different cell death pathways.
IFN-γ is the most potent cytokine against intracellular bacteria due to its ability to induce hundreds of genes promoting host defense [1]. Remarkably, our data indicate that the antimicrobial action of IFN-γ against F. novicida and F. tularensis Live Vaccine Strain is almost exclusively dependent on GBPs. While it was previously known that IFN-γ could restrict cytosolic Francisella growth independently of cell death, reactive oxygen or nitrogen species, autophagy and IDO-mediated tryptophan degradation [45], the mechanisms responsible for this growth restriction were unknown. The IFN-γ-mediated F. novicida growth restriction is independent of caspase-1 and caspase-11 (S3 Fig). It thus differs from the recently described mechanisms responsible for growth inhibition of cytosolic Salmonella [59]. Interestingly, the highly virulent F. tularensis SCHU S4 largely escaped IFN-γ-mediated antibacterial activity in BMDMs (S10 Fig). It remains unclear whether this escape is the result of an active process or due to the failure of innate immune sensors to detect/recognize cytosolic F. tularensis SCHU S4.
In a concurrent work, Kanneganti and colleagues identified IRGB10 as an IFN-inducible GTPase recruited onto cytosolic F. novicida and required to lyse the bacterium and trigger AIM2 inflammasome activation [16]. IRGB10 recruitment is abolished in Gbp chr3 -KO macrophages indicating that GBPs and IRGB10 may act together and that the latter protein is likely involved in IFN-γ-mediated growth restriction in murine macrophages. In contrast to GBPs [6], IRGs (with the exception of the constitutively expressed IRGM) are absent in humans [60]. Yet, IFN-γ priming efficiently restricts cytosolic Francisella growth in human macrophages ( [45], S13 Fig) indicating that IRGs are facultative for the IFN-γ-mediated antimicrobial role. While our results identify that GBPs are required for IFN-γ-mediated killing of cytosolic bacteria, other host factors are likely involved upstream of GBPs to facilitate GBPs targeting onto cytosolic bacteria. The molecular mechanisms sustaining IFN-γ-dependent GBPs-mediated antibacterial activity remains to be understood.

Mouse infections
Age-and sex-matched animals (6-10 weeks old) were infected subcutaneously with 5x10 3 or 5x10 4 or 4x10 5 CFU of F. novicida in 100 μl PBS (as indicated in the figure legends). When applicable, 10 6 U/ml of rIFN-γ was injected intraperitonealy in 100 μl PBS. Blood was collected by retro-orbital bleeding at 16 h post-infection or intra-cardiac puncture at 48 h post-infection. Animals were sacrificed at the indicated time point post-infection. Mice were examined twice daily for signs of severe infection and euthanized as soon as they displayed signs of irreversible morbidity or as soon as weight loss exceeded 20%.

Replication assay
For F. novicida intracellular replication assay, macrophages were lysed with 1% (w/v) saponin (Sigma) in water for 5 min. Dilution, plating on TSA supplemented with 0.1% (w/v) cysteine and counting was performed using the easySpirale Dilute (Interscience). For LVS replication assay, cells were lysed in PBS with 0.1% deoxycholate, serially diluted in PBS and plated on modified GC-agar base plates.

siRNA knockdown
Gene expression knockdown was done using GenMute (SignaGen laboratories) and siRNA pools (siGenome, Dharmacon). Briefly, wild-type BMDMs were seeded into 24-, or 96-well plates at a density of 1.5x10 5 or 3x10 4  Cytokine measurement and cell death assays IL-1β and IL-18 were measured by ELISA (R&D systems and platinum ebioscience, respectively). Cell viability was determined by the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Global cell death kinetics was monitored in BMDMs by assessing in real time incorporation of propidium iodide (used at 5 μg/ml) through measurement of fluorescence emission at 635 nm every 15 min on a microplate reader (Tecan, see. S2A Fig for the sensitivity of the technique). When indicated triton X100 (Sigma) at 1% (v/v) was added at the end of the kinetics to further control cell death/viability. Area under the curve were computed using Prism software (GraphPad) to obtain a single quantitative readout of the full kinetics as recently described [64]. Gliotoxin (Enzo Pharma) and Nigericin (Sigma) were used at 5 μM, the latter after a 3 h priming with LPS at 100 ng/ml. Single cell death kinetics was determined using an automated time-lapse video microscope (Arrayscan high-content system, Thermo Fisher Scientific). Image analyses of four sparse fields per well were performed using the HCS studio analysis software. Wheat-germ agglutinin (WGA)-labelled BMDMs were used. Individual dead cells were detected and numerated based on the propidium iodide fluorescence staining and normalized to the total number of cells numerated through the vital Hoechst staining at time 0. WGA intensity in propidium iodide positive cells was calculated to quantify cell retraction using HCS studio software. CO2-independent medium (ThermoFisher Scientific) was used for all cytokines dosage and cell death kinetics.

Immunofluorescence and GFP quantification in single bacteria
Macrophages were seeded on glass coverslips and infected as described above. At the desired time, cells were washed 3 times with PBS and fixed with 4% paraformaldehyde for 15 min at 37˚C. Following fixation, coverslips were washed and the fixative was quenched with 0.1 M glycine for 10 min at room temperature. Coverslips were stained with primary antibodies at 4˚C for 16 h, washed with PBS, incubated for 1 h with appropriate secondary antibodies at room temperature (1:500, AlexaFluor, Invitrogen), washed with PBS and mounted on glass slides with Vectashield containing 6-diamidino-2-phenylindole (DAPI) (Vector Labs). Antibodies used were chicken anti-Francisella (1:1000, a gift from D. Monack). Coverslips were imaged on a Zeiss LSM710.
Quantification of GFP in GFP-expressing F. novicida was performed using an automated process in ImageJ (NIH, USA). The threshold was adjusted using the moments-preserving thresholding method with the dark background option. Binary watershed process was used to separate individual bacteria. Fluorescence intensity quantification was restricted to individual particle of 0.2 to 2 μm 2 with a circularity comprised between 0.5 and 1.

Flow cytometry and microscopy in flow
For assessment of bacterial replication by flow cytometry, macrophages seeded onto non-tissue culture-treated plates were infected as described above with GFP-expressing F. novicida strains. At desired time, cells were lifted with trypsin and immediately analyzed by Flow cytometry on a Canto 2 cytometer (BD biosciences). Dead cells were excluded based on staining with propidium iodide.
For the microscopy in flow experiments, macrophages infected with GFP-expressing bacteria were fixed in PFA 4% and analyzed on ImageStream X mark II (Amnis, EMD-Millipore) using the Inspire software with the Extended depth of field (EDF) function activated to increase the spot counts accuracy. Images of single cells were analyzed with the Ideas Software (Amnis, EMD-Millipore) as previously described [13] to quantify the number of bacteria per cell.

Statistical analysis
Statistical data analysis was done using Prism 5.0a (GraphPad Software, Inc.). To evaluate the differences between three selected groups or more (cell death, cytokine release, FACS, CFU and immunofluorescence-based counts) one-way ANOVA analysis was performed with Tukey's correction for multiple analysis. Komogorov-Smirnov test was used to compare the cell distribution as determined by Imagestream and the distribution of GFP intensity in single bacteria. P values were adjusted for multiple comparisons with the Bonferroni correction approach. Animal experiments were evaluated using Kruskal-Wallis analysis with Dunn's correction except when only two groups were present in the analysis, Mann-Whitney analysis was performed. Survival experiment was analyzed thanks to log-rank Cox-Mantel test. In figures NS indicates 'not significant', P values are given according to the following nomenclature: Ã P<0.05; ÃÃ P<0.01; ÃÃÃ P<0.001.