Activation of the STING-Dependent Type I Interferon Response Reduces Microglial Reactivity and Neuroinflammation

SUMMARY
Brain aging and neurodegeneration are associated with prominent microglial reactivity and activation of innate immune response pathways, commonly referred to as neuroinflammation. One such pathway, the type I interferon response, recognizes viral or mitochondrial DNA in the cytoplasm via activation of the recently discovered cyclic dinucleotide syn-thetase cGAS and the cyclic dinucleotide receptor STING. Here we show that the FDA-approved antiviral drug ganciclovir (GCV) induces a type I inter-feron response independent of its canonical thymi-dine kinase target. Inhibition of components of the STING pathway, including STING, IRF3, Tbk1, extra-cellular IFNb, and the Jak-Stat pathway resulted in reduced activity of GCV and its derivatives. Impor-tantly, functional STING was necessary for GCV to inhibit inflammation in cultured myeloid cells and in a mouse model of multiple sclerosis. Collectively, our findings uncover an unexpected new activity of GCV and identify the STING pathway as a regulator of microglial reactivity and neuroinflammation.

INTRODUCTION
Neurodegenerative diseases such as Alzheimer’s disease, Parkinson’s disease, Huntington’s disease, multiple sclerosis, fronto-temporal dementia, and amyotrophic lateral sclerosis are associated with activation of predominantly innate im-mune pathways, referred to as neuroinflammation. During this process, microglia and other brain cells and, in some cases, infiltrating cells from the systemic environment secrete inflammatory cytokines and chemokines with positive and negative effects on the brain (Glass et al., 2010). Interferons are one such class of cytokines with neuroprotective and neurotoxic properties (Deczkowska et al., 2016). For example, the type II interferon, IFNg, previously considered proinflam-matory, has recently been shown to contribute to immune surveillance in healthy brains (Kunis et al., 2013). Similarly, the type I interferon IFNb was shown to negatively affect brain function during aging (Baruch et al., 2014), and on the other hand, IFNb can serve a protective function and is used to dampen inflammation in active, relapsing multiple sclerosis (Group, 1993). Additionally, the lack of IFNb signaling in neurons resulted in Lewy body and Parkinson’s disease-like dementia in mice (Ejlerskov et al., 2015). Together, these studies suggest that the relative levels of type I and type II interferons and the context in which they act have a profound effect on neuroinflammation and neurodegeneration (Decz-kowska et al., 2016).The production of type I interferons can be induced by a number of pattern recognition receptors (Takeuchi and Akira, 2010), which trigger signaling cascades and targeted immune responses. The presence of double-stranded (ds) viral DNA in the host cytoplasm, for example, is recognized by the recently discovered cyclic GMP-AMP synthetase (cGAS), which catalyzes production of the second messenger 2030-cy-clic-GMP-AMP (cGAMP), a potent ligand of the signaling adaptor known as stimulator of interferon genes (STING/ MPYS/MITA/ERIS, encoded by TMEM173).

This cascade further elicits activation of IKK and TBK1 kinases, NF-kB, and IRF3 transcription factors and production of IFNb (Ishikawa and Barber, 2008; Okabe et al., 2009; Schoggins et al., 2011; Sun et al., 2009). STING has thus emerged as an attractive target for drug discovery, especially for cancer treatment (Ahn et al., 2015; Fu et al., 2015), but little is known about the role of STING in the brain and whether it has a role in neuroin-flammation and neurodegeneration.Fold change is based on control treatment for the experiment. All GCV treatments were performed with 200 mM unless otherwise noted. Statistical tests: one-way ANOVA followed by Dunnett’s multiple comparison test (G–I) and unpaired Student’s t test (D–F). Error bars represent mean + SEM from 3 (cell lines) or 2 (primary cells) independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.Here, we show that the antiviral drug ganciclovir (GCV) induces a type I interferon response in microglia that depends on a functional STING pathway. In vivo, STING is expressed in microglia in the CNS and is upregulated in experimental autoimmune encephalomyelitis (EAE), a mouse model for multi-ple sclerosis. Treatment with GCV reduces STING expression, the microglial inflammatory signature, immune cell infiltration,and paralysis in the EAE mouse model in a STING-dependent fashion.

RESULTS
GCV (Figure 1A) and other nucleoside analogs of 20-deoxy-guanosine are effective anti-viral drugs for the treatment ofcytomegalovirus and herpesvirus infections (Faulds and Heel, 1990). We recently reported that GCV, at therapeutic doses equivalent to those in humans, ameliorates the disease course and pathology of EAE in mice (Ding et al., 2014). GCV exerted these effects, in part, by reducing immune cell infiltration and inhibiting the proliferation of microglia, the immune cells of the CNS. To understand the molecular basis of GCV activity, microglia-like BV-2 cells were stimulated with GCV, and 38 secreted proteins were measured using a Luminex-based array (Figures S1A and S1B). GCV treatment led to the upregulation of several antiviral proteins, and CXCL10 was most significantly over-produced (Figure S1B). Gene expression analysis of GCV-treated BV-2 cells (Figures S1C and S1D) using a microfluidic qRT-PCR panel that we created (consisting of 86 microglia genes; Table S1) showed upregulation of CXCL10 and typeIinterferons (Figures S1C and S1D). To corroborate these findings, primary microglia isolated from adult mice were treated with GCV and analyzed using the microfluidic panel. As with BV-2 cells, primary microglia showed prominent induction of type I interferon-dependent gene expression after GCV treatment (Figures 1B and 1C), including CXCL10 and IFNb (Figure 1D). In addition, these genes were increased at the protein level as well (Figure 1F). Impor-tantly, GCV not only activated this interferon response in mouse microglia but also in human induced pluripotent stem cell (iPSC)-derived microglia (iMGLs) (Abud et al., 2017; Figure 1E). We chose to use CXCL10 and IFNb as outcomes for GCV activity because these proteins were upregulated at transcript as well as protein levels across multiple microglial and myeloid cell types. GCV exhibited time- and dose-dependent activity without detectable toxicity (Figures 1G–1I). Hence, we conclude that these immune-modulatory effects of GCV are unlikely to be due to growth inhibition or cell death.In its canonical mechanism of action, GCV is phosphory-lated by viral thymidine kinases (e.g., herpes simplex virus type 1 thymidine kinase, HSVtk) (Littler et al., 1992) and incorporated into cellular DNA, inhibiting replication (Matthews and Boehme, 1988). In contrast, the GCV activity we describe did not require HSVtk or endogenous tk.

The cells used in this study did not express viral tk (Figures S1E and S1F). Additionally, microglia isolated from adult tk1 knockout mice treated with GCV also produced CXCL10 and IFNb (Figures S1G and S1H), suggesting that thymidine kinase is dispensable for this activity of GCV.In a cell-based model of inflammation where primary micro-glia or BV-2 cells were stimulated with IFNg and lipopolysac-charide (LPS), GCV led to significant transcriptional inhibition of several proinflammatory genes (Figure S2A). One of the most significantly reduced transcript and protein was NOS2/ iNOS (Figures S2A and S2C–S2E), which further led to a reduction in neurotoxic microglial nitric oxide production (Figures S2B and S2F).We tested if other compounds and antiviral drugs could induce a type I interferon response like GCV. The structurally related FDA-approved GCV analogs acyclovir (ACV) and penci-clovir (PCV), or the endogenous molecules guanine and guano-sine and structurally unrelated anti-viral drugs (structures shown in Figure S3) failed to induce CXCL10 mRNA (Figures 2B and 2C), suggesting that the 1,3-dihydroxy-2-propoxymethyl group at N9 of the guanine ring is necessary for activity. In sup-port of this notion, methylating the 1,3-dihydroxyl groups in GCV (MethylGCV) abrogated CXCL10-inducing activity, whereas providing 4 hydroxyl groups in GCV dimers synthesized using a reducible disulfide linker (thiol-GCV) or non-reducible polyeth-ylene glycol (PEG) linkers (diGCV) at C6 of the guanine ring (structures shown in Figure 2A) increased potency to induce CXCL10 (Figures 2D and 2E). Like GCV, diGCV as well as the PEGylated GCV monomer (monoGCV) dose-dependently induced CXCL10 without causing considerable toxicity (Figures 2F–2I). Additionally, monoGCV and diGCV potently reduced iNOS transcript and protein (Figures S2C, S2D, and S2G) and nitric oxide production in IFNg/LPS-stimulated BV-2 cells (Figure S2H), and they induced CXCL10 independent of endog-enous tk1 (Figure S2I).Interferons activate the Jak/Stat signaling pathway to induce CXCL10 (Liu et al., 2011), and we observed that GCV and diGCV similarly depend on this pathway (Figure 3).

Specifically, the Stat1 inhibitor fludarabine (Frank et al., 1999) or the Jak kinase inhibitors ruxolitinib and TG101348 (Zhou et al., 2014) strongly inhibited CXCL10 production in response to GCV and diGCV (Figures 3A and 3F) without causing toxicity (Figures 3B and 3G). Likewise, small interfering RNA (siRNA) knockdown of Stat1 and Jak1, but not TLR3, reduced GCV activity (Figures 3C, 3H, and 3I). Primary microglia from Stat1 knockout (KO) mice also significantly reduced CXCL10 or IFNb mRNA induction by GCV, monoGCV, and diGCV (Figure 3D). Additionally, the ability of GCV to suppress the inflammatory marker iNOS was dependent on Stat1 (Figure 3E). GCV and its derivatives might activate the Jak/Stat pathway either directly or through the pro-duction and autocrine signaling of IFNb through subsequent feedback loops, activating other pattern recognition receptors (Figure 4A). Indeed, neutralization of IFNb with an antibody partly reduced CXCL10 induction mRNA by GCV, monoGCV, or diGCV in BV-2 cells (Figure 3J).Recent studies have shown that, upon sensing exogenous dsDNA in the cytoplasm, the enzyme cGAS catalyzes the for-mation of cGAMP (structure in Figure S3), which subsequently induces a potent interferon response (Hornung et al., 2014; Ish-ikawa et al., 2009). Cyclic dinucleoside monophosphates (e.g., c-di-GMP; structure in Figure S3) can induce a similar response (Chin et al., 2013). These dinucleotides activate the endo-plasmic reticulum (ER) membrane adaptor protein STING, which then activates TANK binding kinase 1 (Tbk1), NF-kB, and IRF3 (Barber, 2015; Ishikawa and Barber, 2008) and down-stream effector genes, including IFNb and CXCL10 (Figure 4A). Accordingly, and in line with previous studies (Gao et al., 2013), cGAMP and c-diGMP strongly induced CXCL10 and IFNb in microglia (Figure 4B), as did the reported STING agonists 5,6-Dimethylxantheonone-4-acetic acid (DMXAA) and 10-carboxymethyl-9-acridanone (CMA) (Cavlar et al., 2013; Gao et al., 2013; Prantner et al., 2012; Figure 4C).

GCV and diGCV also induced CXCL10 in the human monocyte cell line THP-1, whereas DMXAA, which is specific for mouse STING (Conlon et al., 2013), did not (Figure 4D). Due to itsFold change is based on control treatment. All monoGCV and diGCV treatments were performed with 200 mM unless otherwise noted. Statistical tests: one-way ANOVA followed by Dunnett’s multiple comparison test (B–I) and unpaired Student’s t test (I, right). Error bars represent mean + SEM from 3 independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.apparent structural similarity, we hypothesized that diGCV, and possibly GCV cellular metabolites, may mimic cyclic dinucleo-tides and activate the STING pathway.Excitingly, siRNA-mediated knockdown of STING in BV-2 cells largely abrogated the capacity of diGCV and, to a lesserextent, monomeric GCV to induce CXCL10 (Figure 4E). More-over, primary microglia from STINGgt/gt mice, which lack func-tional STING protein (Sauer et al., 2011) and do not respond to cGAMP, failed to induce CXCL10 and IFNb mRNA in response to GCV, monoGCV, and diGCV (Figure 4F). Consistent with GCV targeting the STING pathway, pharmacological inhibition of Tbk1 activity using the antagonist amlexanox and siRNA-mediated knockdown of IRF3 inhibited the capacity of monomeric and dimeric GCV to induce CXCL10 mRNA (Figures 4G and 4H). Furthermore, siRNA-mediated knock-down of the upstream activator cGAS did not affect the activ-ity of GCV and diGCV to induce CXCL10 (Figure 4I). These data in aggregate show that, like the reported STING agonists (Burdette et al., 2011; Cavlar et al., 2013; Gao et al., 2013), the ability of GCV and its derivatives to induce a type I interferon response in microglia-like cells requires a functional STING pathway and downstream Jak/Stat signaling. Native GCV and diGCV molecules did not bind strongly to purified mouse STING protein (Figure 4J). The possibilities that these molecules are prodrugs, which are modified intracellularly to be active, or that they bind to another target in the STING pathway remain to be elucidated.

To determine the involvement of STING in regulating micro-glial reactivity and neuroinflammation in vivo, we induced the autoimmune disease EAE in wild-type (WT) and STINGgt/gtmice and treated them with GCV (Figure 5A). We found that STING was specifically expressed in microglia and notNeuron 96, 1290–1302, December 20, 2017 1293Drug treatments were performed with 200 mM unless otherwise noted. Statistical tests: one-way ANOVA followed by Dunnett’s multiple comparison test (A–C and F–I) or unpaired Student’s t test (C, right; D and H, right; I, right; and J). Error bars represent mean + SEM from 3 (cell lines) or 2 (primary cells) independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.detectable in other CNS cell types (Figure 5B; Figure S4). Interestingly, EAE induction led to a dramatic increase in STING expression in Iba1+ myeloid cells as well as in Tmem119+ microglia, and GCV reversed this phenotype almost completely (Figure 5B). We next asked if STING was required for the therapeutic effects of GCV in the EAE mouse model. As we reported previously (Ding et al., 2014), GCV drastically reduced disease severity in WT mice in three inde-pendent experiments (Figure 5C), lowering disease incidenceby 60%–70% (Figure 5D) and lethality from 20% to 0% (Fig-ure 5E). Although mice lacking STING showed a very similar disease course as WT mice (Figures 5C–5E), GCV failed tosignificantly reduce disease severity (Figure 5C), incidence (Figure 5D), and lethality (Figure 5E) in STINGgt/gt mice atadvanced stages of the disease. However, during the early phase of disease, GCV only partially ameliorated EAE, possibly because of compensatory mechanisms and other unknown complexities of EAE progression in STINGgt/gtStatistical tests: one-way ANOVA followed by Dunnett’s multiple comparison test (C and D) or unpaired Student’s t test (E–I). mRNA fold change was determined by qRT-PCR. Error bars represent mean + SEM from 3 (cell lines) or 2 (primary cells) independent experiments. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.mice.

It was shown that STINGgt/gt mice exhibit attenuated EAE development compared with WT mice (Lemos et al.,2014). We hypothesize that the intermediate effect on EAE scores by GCV in STINGgt/gt mice is due to this slow andpossibly altered EAE pathology. In support of this, STINGgt/ gt mice with EAE show higher numbers of proliferating cells overall (Figure 5F), proliferating T cells (Figure 5K; Figure S6E), and activated microglia (Figure 6C).(F–K) Quantification of the average number of BrdU+ proliferating cells (F), Iba1+ myeloid cells (G), Tmem119 expression (H), CD68 expression (I), percent Iba1+BrdU+ proliferating myeloid cells (J), and CD3+BrdU+ proliferating T cells (K). For histology, n = 6–10 mice/group.Error bars represent mean + SEM. Statistical tests: two- way ANOVA followed by Sidak’s multiple comparisons test between the indicated groups. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.Consistent with our published data (Ding et al., 2014), GCV significantly decreased the numbers of bromodeoxyuridine (BrdU)+ and PCNA+ proliferating cells (Figure 5F and S5), Iba1+ myeloid cells (Figure 5G), Iba1+BrdU+ and Iba1+PCNA+ proliferating myeloid cells (Figure 5J; Figure S5) but did not change CD3+BrdU+ proliferating T cells (Figure 5K) in cere-bella and spinal cords (Figure S6) of WT mice with EAE. GCV treatment did not increase the number of cleaved cas-pase-3+ cells (Figure S7), suggesting that the inhibition of myeloid proliferation by GCV was not due to induction of apoptosis. Additionally, GCV-treated WT EAE mice showed reduced expression of the microglia-specific marker Tmem119 (Figure 5H; Figures S4B and S8) and the microglial activation marker CD68 (Figure 5I; Figure S8). In starkcontrast, and in agreement with the pre-clinical data above, GCV-treated STINGgt/gt mice with EAE showed no reductionin overall cell proliferation (Figure 5F; Figures S5 and S6), Tmem119 expression (Figure 5H), myeloid cell activation (Figure 5I), proliferating myeloid cells (Figure 5J; Figures S5 and S6), or T cells (Figure 5K; Figure S6).

To deduce whether this effect of GCV on EAE was due to inhi-bition of infiltrating myeloid cells or resident CNS microglia, weisolated CD11b+ myeloid cells from the cerebella of WT and STINGgt/gt mice, with or without EAE, and analyzed CD45 immu-noreactivity (Figure 6). GCV treatment reduced total CD45hi (Fig-ure 6A) and CD11b+CD45hi (Figure 6B) activated myeloid cells aswell as Tmem119+CD45hi activated microglia (Figure 6C) in WT EAE but not in STINGgt/gt mice, suggesting that GCV exerts its effects on both the infiltrating (Tmem119–, CD45hi) as well asresident microglia in the CNS (Tmem119+). Interestingly, STINGgt/gt mice with EAE had twice as many proliferating cellsas WT mice (Figure 5F), and showed a trend toward increasing overall T cell proliferation (Figure 5K), and CD68 and CD45hi immunoreactivity (Figures 5I and 6C), suggesting an altered EAE pathology in these mice.Finally, to elaborate these anti-inflammatory effects of GCV, we isolated CD11b+Tmem119+ microglia (Figure 7A) from GCV-treated WT and STINGgt/gt mice with EAE andanalyzed them by RNA sequencing (RNA-seq). Unsupervised clustering of significantly changed genes segregatedGCV-treated WT microglia from the other groups (Figures 7B and 7C). The most significant differences were found be-tween WT PBS- and GCV-treated microglia (Table S2), and they were all STING-dependent (Figure 7D), supporting our finding that GCV requires STING to regulate microglial activ-ity. The most significantly modulated genes by GCV in micro-glia from WT mice with EAE (Figure 7E) are known to be associated with inflammation (Alox5, Faim3, Ctsb, Lyz1, Clec2i, and Apoe), small-molecule transport (Sidt1, Fabp5, and Slc25a31), and G-protein-coupled receptors (F2rl2). Inflammatory response was the most significant GO term associated with GCV versus PBS WT microglia from mice with EAE (Figure 7F).

Interestingly, some genes (e.g., Ctsb, Apoe, and Lyz1) that were significantly downregulated with GCV treatment were recently described as disease-associated microglia (DAM) genes in microglia from Alzheimer’s and Amyotrophic Lateral Sclerosis (ALS) mouse models (Keren-Shaul et al., 2017).
We compared DAM geneswith an independent RNA-seq study from microglia in diseased EAE mice(Lewis et al., 2014) and found that the majority of DAM genes were also upre-gulated in microglia from EAE mice. Further analysis of the most differen-tially expressed DAM genes in ourstudy showed that GCV downregulated these genes in WT but not in STINGgt/gtmicroglia from mice with EAE (Figures 8A–8C). Additionally, there was an overall increase in expression of ho-meostatic microglia genes with GCV treatment (Figure 8D). Type I interferon transcripts were undetectable in theRNA-seq dataset. However, a micro-fluidic array on mRNA from cerebella of EAE mice showed an increase in type I interferons and confirmed the downregulation of inflammatory micro-glial transcripts observed by RNA-seqin WT mice treated with GCV (Figure S9). Thus, we conclude that GCV induces low therapeutic levels ofIFNs and results in downregulation of disease-associated genes in a STING-dependent way, reducing inflammation.

DISCUSSION
In aggregate, these studies show that GCV reduces EAE in a STING-dependent fashion similar to DNA nanoparticles, which were recently shown to attenuate EAE (Lemos et al., 2014). STING is highly regulated in microglia in vivo, and activation of the STING pathway reduces microglial reactivity and the neuro-inflammatory disease EAE. Because excessive IFN production is linked to interferonopathies such as STING-associated vascul-opathy with onset in infancy (SAVI) and Aicardi-Goutie`res syn-drome (Rodero and Crow, 2016), it will be important to find the optimal therapeutic levels to activate the STING pathway in a beneficial way.exploit viral thymidine kinase activity and inhibit viral replication (Elion et al., 1977). After decades of highly effective use in hu-mans, our study uncovered a remarkable non-canonical activity of GCV, but not acyclovir, that involves the innate immune recep-tor STING and a stereotypical cellular antiviral program. Weshow that GCV can exhibit dual function in microglia (Figure S10): in the naive state, GCV induces microglia to be ‘‘primed’’; on the other hand, GCV reduces inflammation in active microglia. We propose that GCV pushes microglia toward a primed state. This multi-modality of GCV is unique and may, in part, beresponsible for the continued strong success of GCV (and its pro-drug valganciclovir), in spite of many newer antiviral drugs. Lower doses of GCV elicited little to no effect in CNS demyelin-ation and viral disease models in mice (Skripuletz et al., 2015), suggesting that appropriate dosing is necessary for the novel properties observed here. Alternatively, it is also possible thatGCV does not cross the blood-brain barrier or is functional spe-cifically in the EAE mouse model.Because of its growing relevance not only in anti-viral immune responses but possibly in sensing mitochondrial damage as well (West et al., 2015), STING has become an attractive target for drug development itself (He et al., 2015). In addition, mutations in STING are associated with vascular and pulmonary syndrome (Liu et al., 2014) and other autoimmune diseases (Jeremiah et al., 2014; Sharma et al., 2015), STING-IRF3 stress is associated with alcoholic liver disease (Petrasek et al., 2013), and haploinsufficiency in the STING activating kinase Tbk1 is associated with ALS and frontotemporal dementia (FTD) (Freischmidt et al., 2015; Pottier et al., 2015). Our findings that GCV, and GCV dimers in particular, activate a type I interferon response inaSTING-dependent way and reduce microglial proliferation and neuroinflammation in vivo open the possibility to develop a new class of drugs to treat neurodegenerative and related diseases where neuroinflammation has been GSK8612 implicated.