Chronic neurodegeneration induces type I interferon synthesis via STING, shaping microglial phenotype and accelerating disease progression

Abstract Type I interferons (IFN‐I) are the principal antiviral molecules of the innate immune system and can be made by most cell types, including central nervous system cells. IFN‐I has been implicated in neuroinflammation during neurodegeneration, but its mechanism of induction and its consequences remain unclear. In the current study, we assessed expression of IFN‐I in murine prion disease (ME7) and examined the contribution of the IFN‐I receptor IFNAR1 to disease progression. The data indicate a robust IFNβ response, specifically in microglia, with evidence of IFN‐dependent genes in both microglia and astrocytes. This IFN‐I response was absent in stimulator of interferon genes (STING−/−) mice. Microglia showed increased numbers and activated morphology independent of genotype, but transcriptional signatures indicated an IFNAR1‐dependent neuroinflammatory phenotype. Isolation of microglia and astrocytes demonstrated disease‐associated microglial induction of Tnfα, Tgfb1, and of phagolysosomal system transcripts including those for cathepsins, Cd68, C1qa, C3, and Trem2, which were diminished in IFNAR1 and STING deficient mice. Microglial increases in activated cathepsin D, and CD68 were significantly reduced in IFNAR1−/− mice, particularly in white matter, and increases in COX‐1 expression, and prostaglandin synthesis were significantly mitigated. Disease progressed more slowly in IFNAR1−/− mice, with diminished synaptic and neuronal loss and delayed onset of neurological signs and death but without effect on proteinase K‐resistant PrP levels. Therefore, STING‐dependent IFN‐I influences microglial phenotype and influences neurodegenerative progression despite occurring secondary to initial degenerative changes. These data expand our mechanistic understanding of IFN‐I induction and its impact on microglial function during chronic neurodegeneration.

some degree, during bacterial infection. In the context of viral infection, they are typically induced via engagement of TLR3, TLR7, or other pattern recognition receptors (PRRs) such as RIG-1 and MDA-5 that recognize intracellular single-stranded or double-stranded RNA (Hoffmann, Schneider, & Rice, 2015). IFN-I is also induced via cytosolic DNA sensing pathways that signal via endoplasmic adaptor molecule stimulator of interferon genes (STING) (Gurtler & Bowie, 2013). First identified by expression cloning, STING was shown to activate both NF-κB and IRF3 transcription pathways to induce expression of IFN-I and exert a potent antiviral state following expression (Ishikawa, Ma, & Barber, 2009).
The principal cells responsible for IFN-I production during systemic viral infection include macrophages and plasmacytoid dendritic cells, but it is clear that most cell types in the central nervous system (CNS) can mount IFN-I responses (Blank et al., 2016;Owens, Khorooshi, Wlodarczyk, & Asgari, 2014). These responses may arise in response to systemic viral infection, circulating IFNα, or viral mimetics such as poly inosinic: poly cytidylic acid Wang, Campbell, & Zhang, 2008) or in response to brain injury and neurodegeneration (Field, Campion, Warren, Murray, & Cunningham, 2010;Hosmane et al., 2012;Khorooshi & Owens, 2010;Main et al., 2016;Minter et al., 2016;Wang, Yang, & Zhang, 2011). Recently, both astrocytes and microglia have been shown to express a number of DNA sensors and to respond to DNA stimulation with robust IFN-I responses (Cox et al., 2015).
There is evidence for both protective and deleterious roles of IFN-I in different disease states (Owens et al., 2014). The predominant view of IFN-I in the brain is that they are anti-inflammatory. IFN-I induce the anti-inflammatory cytokine IL-10 ( Lin et al., 2013) and in this way may inhibit IL-1 production (Guarda et al., 2011). Consistent with this, in a CNS context, IFN-β is a first-line therapy for multiple sclerosis and limits lymphocyte infiltration into the brain and therefore also decreases relapse rate (Owens et al., 2014;Prinz et al., 2008).
Moreover, axonal degeneration in the perforant path has been shown to induce a robust type I interferon response, which appears to limit CCL2 and MMP9 expression and prevent exaggerated cell infiltration in the injured area (Khorooshi & Owens, 2010). Similarly, an in vitro model of axon injury showed IFN-I induction in microglia. Blocking this response via disruption of Toll/interleukin-1 receptor domaincontaining adapter inducing interferon-β (TRIF), impaired microglial clearance of axonal debris, and inhibited axon outgrowth after dorsal root axotomy (Hosmane et al., 2012). Conversely, detrimental actions have also been ascribed to IFN-I. Transgenic overexpression of IFN-α in the brains of mice is associated with increased inflammation and neurodegeneration (Akwa et al., 1998;Campbell et al., 1999). IFNβ has been shown to be elevated with age and direct administration of antibodies against IFNAR1 protected against age-dependent cognitive impairment and increased neurogenesis in the sub-ventricular zone (Baruch et al., 2014).
All type I IFNs can signal via the heterodimeric interferon α/β receptor (IFNAR), which classically signals via a JAK/STAT pathway leading to the upregulation of IFN-stimulated genes (ISGs) (Schoggins & Rice, 2011), and generation of IFNAR1-deficient mice (Hwang et al., 1995) has facilitated attempts to understand the role of IFN-I in neurodegenerative disease. With respect to chronic neurodegenerative processes, IFNAR1 deficiency resulted in modestly slowed disease progression in SOD1(G93A) model of amyotrophic lateral sclerosis (Wang et al., 2011) and reduced dopaminergic cell death in the MPTP model of Parkinson's disease (Main et al., 2016).
Conversely, loss of dopaminergic neurons was observed in IFNβ −/− mice (Ejlerskov et al., 2015). Both human Alzheimer's disease (AD) and in vitro and in vivo models of AD show increased IFN-I expression (Mesquita et al., 2015;Taylor et al., 2014), and in vivo studies showed decreased pathology and altered microglial phenotype in IFNAR1 −/− × APP SWE /PS1 ΔE9 mice (Minter et al., 2016). Whether effects on microglia were secondary to effects on Aβ is unclear.
Therefore, IFN-I have divergent roles in the degenerating brain, but cellular sources and mechanisms of induction as well as downstream functions remain incompletely understood. Here, we characterized the IFN-I response in the degenerating brain in the ME7 model of prion disease and then studied both its pathway of induction and the impacts of this IFN-I response on microglial activity and on progression of neurodegeneration. We reveal a STING-mediated IFN-I response that significantly alters microglial phenotype and disease progression.

| MATERIALS AND METHODS
Female C57BL/6 (Harlan, Bicester, UK) were housed in groups of five and given access to food and water ad libitum. IFNAR1 −/− mice on a C57BL/6 background were kindly provided by Professor Paul Hertzog (Monash University, Clayton, Australia). Generation of mutant mice was as previously described (Hwang et al., 1995): 129Sv ES cells were transferred into the Balb/C background and offspring backcrossed onto a C57BL6/J background for more than seven generations.
Females were used to avoid fighting and injury, which has significant effects on behavior. Tmem173 <tm1Camb> (STING −/− ) mice, on a C57BL/6 (Charles Rivers) background, were initially provided by Dr. Jin Lei (University of Florida), and generation of STING −/− mice is as described (Jin et al., 2011), and these mice were then maintained as a homozygous colony in the TCD comparative medicine unit. STING −/− mice had agouti fur color as selection marker. Animals were kept in a temperature-controlled room (21 C) with a 12:12 hr lightdark cycle (lights on at 0700 hr). All animal experimentation was performed in accordance with Republic of Ireland Department of Health & Children and Health Products Regulatory Authority licenses, with approval from the local ethical committee and in compliance with the Cruelty to Animals Act 1876 and the European Community Directive, 86/609/EEC. All efforts were made to minimize both the suffering and number of animals used.

| Surgery and animal treatments
Mice were weighed, anesthetized intraperitoneally (i.p.) with 1.2% Avertin (2,2,2-tribromoethanol solution; 0.2 mL/10 g body weight) and positioned in a stereotaxic frame (David Kopf Instruments, Tujunga, CA). The incisor bar was set at −1 mm, to give an approximately level head. The scalp was incised, and the skull exposed. Two small holes were drilled in the skull either side of the midline to allow for bilateral injection of 1 μL of a 10% w/v scrapie (ME7 strain)-infected C57BL/6 brain homogenate made in sterile phosphate-buffered saline (PBS).
Following intrahippocampal injection, the needle was left in place for 2 min before being withdrawn slowly to minimize reflux. Mice were then placed in a heated recovery chamber and finally re-housed.
Sucrose (5% w/v) and carprofen (0. 05% v/v;Rimadyl,Pfizer,Ireland) were added to drinking water for 2 days following surgery to provide post-surgical analgesia and to optimize post-surgical recovery. One group of ME7 animals were treated with SC-560 (30 mg/kg in 24% DMSO) to assess which isoform of cyclooxygenase (COX) was responsible for prostaglandin E2 synthesis. Animals were euthanized at preplanned time points based on prior publications of neuropathology, except in survival experiments. In those experiments, due to regulatory authority licensing restrictions, animals were euthanized when they reached a predetermined humane endpoint of 15% loss of peak body weight in conjunction with overt clinical signs/terminal symptoms: kyphosis, incontinence, hunched posture, and ruffled fur.

| Behavioral and motor coordination testing
For all behavioral experiments, mice were moved from their homeroom and left in the test room for 15 min before beginning the task to ensure that they were in an optimal state of arousal.

| Open field
The open field arena consisted of a plastic base (58 cm × 33 cm) surrounded by walls of 19 cm. The floor of the box is divided into a grid of equal sized squares. Measurement was made of distance traveled (in terms of grid squares crossed) and total number of rears.
Weekly measurements of open field activity were recorded for 3 min from when animals were placed facing any corner of the arena.

| Burrowing
Black plastic burrowing tubes, 20 cm long, 6.8 cm diameter, sealed at one end were filled with 300 g of normal mouse diet food pellets, and placed in individual mouse cages. The open end was raised by 3 cm above the floor by a wooden support to prevent nonpurposeful displacement of the contents. Mice were placed individually in the cages for 2 hr, at which point the food remaining in the cages was weighed and the amount displaced (burrowed) was calculated. Weekly measurements were taken during prion disease progression.

| Horizontal bar
The horizontal bar was designed to assess forelimb muscular strength and coordination. It consisted of a 26 cm long metal bar, 0.2 cm diameter, supported by a 19.5 cm high wooden column at each end. Each mouse was held by the tail and allowed to grip the central point of the bar with its front paws only. The tail was rapidly released, and mice were scored based on whether they fell, held on for 60 s, or reached a platform on a supporting column, with the latter two results scoring the maximum of 60 s.

| Inverted screen
The inverted screen assessed muscular strength for all four limbs. It consisted of a wooden frame, 43 cm square, covered with wire mesh (12 mm squares of 1 mm diameter wire). The mouse was placed on the screen, which was then slowly (2 s) inverted. The time it took for the mouse to fall was measured, up to a criterion of 60 s. Padding was provided to cushion mice falling from both bar and screen apparatus.

| Tissue collection
Animals were terminally anesthetized with sodium pentobarbital (40 mg per mouse i.p., Euthatal, Merial Animal Health, Essex, UK). The skin overlying the sternum was incised, and the chest cavity opened to expose the heart. A butterfly needle connected to a peristaltic pump (Gilson, Villiers le Bel, France) was inserted into the left ventricle. The right atrium was cut, and the animal was perfused transcardially with heparinized saline (0.1% v/v heparin [LEO Pharma, Buckinghamshire, UK] in 0.9% saline) for approximately 2 min to clear blood from all tissues. Brains were rapidly removed, and an area encompassing the dorsal hippocampus and thalamus (area of maximum pathology in ME7) was punched from thick coronal sections at the appropriate rostro-caudal position. For tissues for homogenization, tissues were placed in Eppendorf tubes, snap frozen in liquid nitrogen and stored at −80 C until further processing. Those tissues used for cell isolation were prepared as below.
2.4 | Fluorescence-activated cell sorting (FACS) of microglia and astrocytes 2.4.1 | Enzymatic digestion and myelin removal The area encompassing the dorsal hippocampus and thalamus (area of maximum pathology in ME7) was punched from thick coronal sections at the appropriate rostro-caudal position and kept in ice cold 1 mL HBSS. This tissue punch was minced and dissociated in 5 mL of enzyme mixture containing collagenase (2 mg/mL), DNase I (28 U/mL), 5% FBS, and10 μM HEPES in HBSS, followed by a filtering step using a 70-μm cell strainer (BD Falcon) to achieve a single-cell suspension. Myelin from single-cell suspension obtained was removed by subsequently incubating with Myelin Removal Beads II for 20 min and passing through LS columns mounted over QuadroMACS magnet.

| Staining and cell sorting
The myelin-depleted single-cell suspension obtained above was incubated (15 min on ice) with anti-mouse CD16/CD32 antibody to block Fc receptors and subsequently incubated with anti-CD11b PEcy7 using 100 μm nozzle. 7AAD (Becton Dickinson) was used to gate out nonviable cells. Sorted cells were collected in 1.5 mL Lobind RNAse/DNAse free tubes containing 350 μL of sorting buffer (HBSS without Phenol Red supplemented with 7.5 mM HEPES and 0.6% glucose). Of note, 40-50,000 CD45 low CD11b + microglia and 80-100,000 GLAST+CD45-astrocytes were sorted (gating strategy described in Figure 1a Qiagen RNeasy ® Plus mini kits (Qiagen, Crawley, UK) were used for hippocampal and thalamic homogenates according to the manufacturer's instructions. Samples were disrupted in 600 μL Buffer RLT using a motorized pestle followed by centrifugation at 14,800 rpm for 6 min through Qiagen Qiashredder columns to complete homogenization. The flow-through was collected and transferred to the genomic DNA (gDNA) Eliminator spin column and centrifuged at 14,800 rpm for 30 s. The column was discarded, and an equal volume of 70% ethanol was added to the flow-though and mixed until homogenous.
Samples were placed in RNeasy mini spin columns in 2 mL collection tubes and centrifuged at 14,800 rpm for 15 s. On-column DNase digestion (Qiagen) RNase free DNase I incubation mix (80 μL) as an extra precaution to ensure complete removal of contaminating gDNA.
RNA was well washed before elution with 30 μL of RNase-free water.
RNA yields were determined by spectrophotometry at 260 and 280 nm using the NanoDrop ND-1000 UV-Vis Spectrophotometer (Thermo Fisher Scientific, Dublin, Ireland) and stored at −80 C until cDNA synthesis and PCR assay.
RNA was reversed transcribed to cDNA using a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Warrington, UK). Four hundred nanograms of total RNA was reverse transcribed in a 20 μL reaction volume. Of note, 10 μL master mix (for each sample, master mix contained: 2 μL 10× RT Buffer; 0.8 μL 25× dNTP mix, 100 mM; 2 μL 10× RT random primers; 1 μL MultiScribe™ Reverse Transcriptase; 4.2 μL RNase-free water) was added to 10 μL RNA for each sample in a nuclease-free PCR tube (Greiner Bio-One, Monroe, FIGURE 1 Gating strategy and purity of fluorescence-activated cell sorted (FACS) microglia and astrocytes from normal brain homogenate and ME7-injected. (a-c) Gating strategy used for sorting of microglia and astrocytes, (d) quantitative polymerase chain reaction (qPCR) results of Gfap and Itgam expressions in FAC-sorted astrocytes, depicting purity of sorting, (e) qPCR results of Itgam and Gfap expressions in FAC-sorted microglia depicting purity of sorting; all data are plotted as mean ± SEM from mice at 19 weeks postinoculation [Color figure can be viewed at wileyonlinelibrary.com] NC). No reverse transcriptase and no RNA controls were also assessed by PCR. PCR tubes were placed in a DNA Engine ® Peltier Thermal Cycler PTC-200 (Bio-Rad Laboratories, Inc., Hercules, CA), and samples were incubated at 25 C for 10 min, 37 C for 120 min, and 85 C for 5 min (to inactivate reverse transcriptase). Samples were held at 4 C until collection and then stored at −20 C until assay. . For all assays, primers were designed using the published mRNA sequences for the genes of interest, applied to Primer Express™ software. Where possible, probes were designed to cross an intron such that they were cDNA specific. In some cases, the fluorescent DNA binding probe SYBR green has been used in place of a specific probe. Primer and probe sequences, along with accession numbers for mRNA sequence of interest may be found in Table 1. Oligonucleotide primers were resuspended in 1× TE buffer (Tris Base 10 mM, EDTA 1 mM; pH 7.5-8.0) and diluted to 10 μM working aliquots. All primer pairs were checked for specificity by standard reverse transcription (RT)-PCR followed by gel electrophoresis, and each primer pair produced a discrete band of the expected amplicon size. Table 1 lists the sequences for primers and probes for those assays that have not been published in our prior studies (Cox et al., 2015;Cunningham, Campion, Teeling, Felton, & Perry, 2007;Field et al., 2010;Hughes, Field, Perry, Murray, & Cunningham, 2010;Palin, Cunningham, Forse, Perry, & Platt, 2008).  primers and probe, 4.0 μL RNase-free water. Where SYBR green was used, RNase-free water was substituted in place of the probe. To this, 1 μL of cDNA was added to give a final reaction volume of 14 μL.
Samples were run in the 7300 Real-Time and QuantStudio 5 Real-Time PCR System (Applied Biosystems, Warrington, UK) under standard cycling conditions: 95 C for 10 min followed by 95 C for 10 s and 60 C for 30 s for 45 cycles. A standard curve was constructed from serial one in four dilutions of the cDNA synthesized from total RNA isolated from mouse brain tissue 24 hr after intra-cerebral challenge with 2.5 μg LPS, which is known to upregulate most target transcripts of interest in this study. A standard curve was plotted of C t value versus the log of the concentration (assigned an arbitrary value since the absolute concentration of cytokine transcripts is not known).
All PCR data were normalized to the expression of the housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH).

| Western blotting
WT and IFNAR1 −/− mice inoculated with NBH or prion disease (ME7) were perfused with sterile saline. Tissue punches containing dorsal hippocampus and posterior thalamus weighing 20-30 mg were homogenized in 100 μL of lysis buffer which contained 50 mM Tris-HCl, 150 mM NaCl, 1% Triton ×100, and protease and phosphatase inhibitors (Roche). A DC protein Assay (Biorad) was carried out, and samples were equalised to 8 mg protein/mL. Typically, 40 μg of protein were run on 10% SDS-PAGE gels after boiling at 90 C for 5 min in sample buffer containing Tris-HCl, glycerol, 10% SDS, β-mercaptoethanol, and bromophenol blue. Of note, 5 μL of molecular weight markers (Santa Cruz: SC-2361/Sigma: P-1677) were also loaded onto the gel. Gels were run at 60 mA for 1 hr and transferred onto PVDF membrane at 225 mA for 90 min. Membranes were then blocked in 5% milk or BSA in TBS-T for 2 hr at room temperature. Table 2 lists the antibodies and dilution and incubation times.
All blots except those for PrP used the above protocol. Blotting was carried out, and samples were equalized to 10 mg/mL. The equalized cerebellar homogenate was then diluted 1:5 into PBS with protease inhibitors ± proteinase K (1,000 μg/mL). Proteinase K treatment was used to determine protease resistance of PrP Sc . Samples containing proteinase K were then incubated at 37 C for 45 min. Samples were boiled in sample buffer at 90 C for 5 min and then spun at 20,800 g for 5 min. Of note, 10% SDS-PAGE gels were loaded with 10 μg total protein and run at 60 mA. Membranes for PrP blots were incubated in 5% BSA in PBS-T for 1 hr at room temperature before probing with the primary antibody. After incubation with the primary and secondary antibodies, all membranes were then exposed for various times using Supersignal West Dura Extended Duration ECL (Pierce). The blots were quantified using Image J software.

PrP Sc
After dewaxing and rehydration, slides were placed in a container of distilled water and autoclaved at 121 C for 20 min. Following a brief PBS wash, the slides were placed in 90% formic acid for 5 min.
Following further PBS washes, the slides were quenched in 1% H 2 O 2 for 15 min and then incubated with Proteinase K and distilled water for 30 min. These steps were taken to ensure that the PrP antibody only stained for protease-resistant PrP Sc . Thereafter, the reaction continued as above.

Synaptophysin
After rehydration, sections were treated with 0.2 M boric acid (pH 9) at 65 C for 30 min and cooled to room temperature thereafter.
The DAB reaction had the additional component of 0.15 g ammonium nickel sulfate in the DAB solution to enhance intensity.

| Quantification of synaptophysin and NeuN
Synaptophysin density was assessed using transmittance of synaptic layers assessed in ImageJ after image capture using a Leica DM3000 microscope and CellA software (Olympus). A mean transmittance was calculated for corpus callosum (cc), overlying cortex, stratum oriens, CA1, stratum radiatum, and the stratum lacunosum. The relatively unstained corpus callosum (cc) served as an internal control, and synaptophysin density was calculated according to the formula: cc − rad/cc − cortex.
Synaptophysin density was measured in two thalamic areas in the same Slides were coated with APS by immersion in a 2% v/v solution of APS in methanol for 10 to 15 s. Finally, slides were rinsed in methanol followed by distilled water and dried overnight at 37 C.
Slides were air dried for 1 hr after removal from the −20 C freezer, fixed in absolute alcohol at 4 C and washed in 0.1 M PBS before blocking in 10% normal rabbit serum. Primary antibody against CD68 (FA11) was applied at 1/100 for 1 hr. Thereafter, the reaction was continued as per the general protocol. For each section (2-4 sections/animal, −1.86 to −2.3 mm from Bregma), the cc was acquired using a ×20 objective lens in optical microscope (Leica DM3000) coupled with a digital camera (Olympus) and Cell A software. CD68 +stained area fraction quantification was performed using ImageJ 1.49v Software (NIH) after creating binary images and set at 144 (area 1) and 146 (area 2) upper average threshold. CD68 + positive cells were counted using particle analyzer with pixel size = 280-Infinity and circularity = 0.05-0.75. Data normality was accessed by Shapiro-Wilk test and the following statistical analysis by independent t-test. The significance level assumed in all test was 95% (p < 0.05).

| Statistics
All statistical analyses were performed using GraphPad Prism 5 for Windows. In all cases, data are expressed as mean ± SEM. Transcripts in ME7 versus NBH were compared by Students' t-test. Assessment of RNA or protein expression was performed using two-way ANOVA with disease and strain as between subjects factors. Longitudinal Eif2ak2 (p = 0.0030, 0.0007) were significantly elevated in both microglia and astrocytes of wild-type ME7 animals; however, the fold increases in microglia were noticeably higher compared to astrocytes (Figure 2j,k). Therefore, there is a robust IFN-I signature in the priondiseased brain, including immune sensors of DNA-damage.
3.2 | Impact of IFN-I on cellular and molecular aspects of prion disease Along with Irf7 and Mx1, Eif2ak2 (PKR) is a classical IFN-dependent gene known to be induced by IFN-I. Here, we demonstrate the induction of all three genes in the ME7 brain and show that this induction is absent in IFNAR1 −/− mice inoculated with ME7 ( Figure 3a).
The Eif2ak2 gene product PKR has been shown to be capable of phosphorylation of eukaryotic initiation factor 2α (eIF2α), a translational controller which has been proposed to play a key role in the progression of neurodegeneration in models of prion disease (Moreno et al., 2012;Moreno et al., 2013). We thus assessed the expression levels of PKR and eIF2α. Western blotting showed a very robust increase in the expression of PKR protein in ME7 animals with respect to NBH animals, and this increase was abolished in IFNAR1 −/− mice (Figure 3b,c). Therefore, the ME7-associated increase in PKR is IFN-I mediated. However, contrary to previous findings, we did not detect any change in the phosphorylation status of eIF2α in prion disease, when expressed as a ratio to total eIF2α (Figure 3b,c,d).  Table 4. Il1b and, to a lesser extent, Nos2 (iNOS) were modestly increased by disease but not affected by genotype (Figure 4a,e), whereas Tnf (4b), and Tgfb1 (4c) were robustly induced in disease but suppressed in IFNAR1 −/− mice. Il10 (4d) and
These data are strongly suggestive of an altered microglial phenotype; but, because some of these transcripts may also be astrocytic in origin, we then isolated microglia and astrocytes from additional animals and assessed key markers ( Figure 5). The microglial markers Cd68, Itgax, Tmem119, and Trem2 all showed significant disease-associated increases (p < 0.01), and those for Itgax, Tmem119, and  increases that were absent in IFNAR1 −/− mice). C3 and C1qα expression was evident in astrocytes but was not significantly affected by disease. In addition, transcripts for cathepsin S and NADPH oxidase subunits (Cyba, Cybb) showed disease-associated increases in microglia (significant interaction between disease and genotype: F ≤ 13.87, df 1,21, p ≤ 0.0024) which were once again IFNAR1-dependent.
Cathepsin S and the NADPH oxidase subunits also showed increases in astrocytes, but the overall level of expression of these genes in astrocytes was much lower than in microglia. Therefore, at the transcriptional level, it is clear that the lack of IFNAR1 signaling results in significant phenotypic changes in microglia and astrocytes.

| STING-mediated IFN-I production
Given the influence of IFNAR1 on microglial phenotype, we interro-  , and ME7 (IFNAR1 −/− ) and blotted with antibodies against PKR, phospho-eIF2α and total eIF2α (exemplars for groups of n = 8). Bands were quantified by densitometry, normalized to the equivalent sample for total eIF2α, and then expressed as a fold increase from NBH (WT). ** denotes a significant interaction between disease status and strain, for PKR expression, by two-way ANOVA (F = 11.67, df 1,28; p = 0.002) . All data are plotted as mean ± SEM with n = 4 (NBH in WT or IFNAR1 −/− ) and n = 5 (ME7 in WT or IFNAR1 −/− ) and analyzed by two-way ANOVA with disease and strain as between subjects factors. Significant differences are denoted *p < 0.05, **p < 0.01, and ***p < 0.001 by Bonferroni post hoc analysis with this, Irf7 mRNA was also significantly decreased in microglia (and in astrocytes, not shown) isolated from STING −/− animals (F 1,21 = 28.14, p < 0.0001), indicating that expression of Irf7 is strongly STING-dependent. PKR mRNA, however, was not found to be reduced significantly in microglia isolated from STING −/− animals ( Figure 6). Interrogating some key microglial transcripts identified as IFNAR1-dependent in Figures 4 and 5  although there were no such significant differences in the thalamus, which is predominantly grey matter at this anterior-posterior location.

| Microglial lysosomal and phagocytic activation
Based on these white matter CD68 changes, we assessed both IBA1 and cathepsin D labeling in the cc, directly dorsal to the degenerating hippocampus. We show (Figure 8m,q) that in both cases, there is a clear increase in labeling in this white matter tract in ME7 that contains axons exiting the degenerating hippocampus. This increase is apparent in ME7 WT animals but mitigated in IFNAR1 −/− mice (Figure 8n,r). Collectively, these data suggest that neurodegeneration-associated increases in microglial lysosomal cathepsin D activity and white matter CD68 expression is significantly mitigated in IFNAR1 −/− mice, suggesting that, in the absence of IFN-I signaling, there is less microglial phagocytic activity in areas of the brain in which degenerative processes are occurring.

| Microglial COX-1-mediated prostaglandins
We have previously shown that COX-1 expression is robustly increased in microglial cells during progression of the ME7 model of prion disease (Griffin, Skelly, Murray, & Cunningham, 2013). Accordingly, PGE2 is also robustly increased and systemic administration of the COX-1-specific drug SC560 (30 mg/kg i.p.) completely abolished these levels, confirming synthesis by COX-1 (Figure 9a; p < 0.001 by Bonferroni post hoc after significant one way ANOVA). The expression of mRNA for COX-1 was significantly reduced in ME7 IFNAR1 −/− mice (Figure 9b), and consistent with this, there was also a marked reduction of PGE2 levels in ME7 IFNAR1 −/− mice with respect to wild-type ME7 animals (Figure 9c; p < 0.001 by Bonferroni post hoc after significant one-way ANOVA).
FIGURE 5 Transcriptional analysis of microglia and astrocytes isolated from normal brain homogenate (NBH) and ME7 animals on IFNAR1-deficient and wild-type backgrounds. Top panel; changes in Cd68, Itgam, Itgax, Tmem119, and Trem2 transcripts in isolated microglia. Second panel; neuroinflammatory transcripts Tgfb1, Tnfa, and Il1b in isolated microglia and astrocytes. Thereafter, panels illustrate transcript levels, from isolated microglia and astrocytes, of complement pathway components: C1q, C3, of lysosomal cathepsins Ctsd and Ctss, and of NADPH oxidase subunits Cybb and Cyba. All data are from animals at 19 weeks plotted as mean ± SEM with n = 6 (NBH in WT), n = 5 (NBH in IFNAR1 −/− ), n = 7 (ME7 in WT) and n = 7 (ME7 in IFNAR1 −/− ) and analyzed by two-way ANOVA with disease and strain as between subjects factors. Significant differences are denoted *p < 0.05, **p < 0.01, and ***p < 0.001 by Bonferroni post hoc analysis Therefore, IFN-I influences the microglial expression and action of COX-1, thus contributing to brain PGE2 levels.

| Effect of IFNAR1 deficiency on disease progression
We examined previously characterized measures of disease progression to assess the impact of IFNAR1 −/− on progression of this fatal neurodegenerative disease. Because all measures were known effects of ME7 (Betmouni, Deacon, Rawlins, & Perry, 1999;Cunningham et al., 2005;Reis et al., 2015), we have included NBH controls for comparison only and statistical analysis has been performed only on WT and IFNAR1 −/− ME7 animals.  (Reis et al., 2015), and this was particularly apparent as a decrease in the ratio between synaptic density in VP and Po. Although synaptic loss was clearly present in IFNAR1 −/− ME7 animals, this was significantly less than that apparent in WT ME7 animals. This difference in synaptic loss was statistically significant (p < 0.05 by Student's t-test) and is shown in Figure 10b.

| Presynaptic terminals
Similarly, we have previously described significant neuronal loss in the posterior thalamic nucleus (Reis et al., 2015). The extent of neu- Type I interferons (IFN-I), neuroinflammatory, and lysosomal response in isolated microglia and astrocytes from WT and STING −/− animals inoculated with normal brain homogenate (NBH) and ME7. Top panel; transcripts of IFN-I response, Ifnb1, Irf7, and Eif2ak2 (PKR) in isolated microglia. Bottom panel; transcripts of Tgfb1, C1q, C3, and Ctsd in isolated microglia. All data are plotted, from animals at 20 weeks postinoculation, as mean ± SEM (n = 6 NBH in WT, n = 3 NBH in STING −/− , n = 7 ME7 in WT, n = 6 ME7 in STING −/− ) and analyzed by twoway ANOVA with disease and strain as between subjects factors. Significant differences in pairwise comparisons by Bonferroni post hoc analysis, after significant main effects, are denoted *p < 0.05, **p < 0.01, and ***p < 0.001 compared to wild-type prion diseased mice (Figure 10g). Animals were euthanized based on reaching a humane endpoint of loss of 15% of peak body weight. Survival was calculated as days postinoculation, and Kaplan-Meier log-rank survival analysis revealed that ME7-inoculated IFNAR1 −/− mice survived on average 2 weeks longer than their wild-type counterparts (median survival 162 days vs 148 days; p < 0.0001).
STING −/− mice were also protected against development of late stage signs ( Supplementary Information S3). Thus, the absence of IFNAR1 and consequent loss of signaling of the IFN-I confers a delay not just in survival time but also in the onset of neurological dysfunction (by approximately 2 weeks) in prion-diseased animals.

| Prion protein aggregation/deposition
One possibility for differences in disease progression in IFNAR1 −/− versus WT animals is via changes in extracellular prion protein (PrP) levels or deposition. Total PrP levels and proteinase K-resistant PrP levels were assessed by Western blotting using 6D11 and quantified. When expressed as a ratio to β-actin, neither total PrP (Figure 11a,d) nor PKresistant PrP (Figure 11b,d) were different in WT and IFNAR1 −/− ME7 animals. This is also manifest when examining 6D11 labeling of extracellular PrP in histological sections in the hippocampus and thalamus ( Figure 11e). Therefore, the impact of IFNAR1 on the progression of ME7 prion disease is independent of altered PrP Sc deposition.

| DISCUSSION
The current study has shown that chronic neurodegeneration, in the ME7 prion disease model, drives a robust STING-dependent IFN-I response, Blocking IFN-I action via deletion of IFNAR1 led to a significant alteration of microglial phenotype, with significant downregulation of a large number of phagocytic, lysosomal, complement, and NADPH oxidase transcripts. Consistent with this, IFNAR1 deletion significantly suppressed disease-associated microglial cathepsin D and scavenger receptor CD68, particularly in areas of white matter pathology. The  Using cell isolations, we show here that Ifnb1 is expressed in microglial but not astrocytic cells, whereas the primary IFN-dependent genes were expressed in both microglia and astrocytes.
Although there have been several studies of IFN-I upregulation in CNS disease and in aging (Baruch et al., 2014;Khorooshi & Owens, 2010;Main et al., 2016;Minter et al., 2016;Wang et al., 2011), the molecular trigger for these events has been unclear.
Here, we show clear upregulation of the DNA damage sensors p204 and cyclic GMP-AMP synthase (cGAS), which can detect DNA damage intracellularly and generate cyclic GMP-AMP (Chen, Sun, & Chen, 2016;Hartlova et al., 2015). This can trigger IFN-I responses via interaction with the endoplasmic reticulum (ER)-associated adapter STING (Gurtler & Bowie, 2013;Ishikawa et al., 2009). These receptors are expressed by both microglia and astrocytes and can mediate robust IFN-I responses to transfected viral DNA (Cox et al., 2015).
Here, we show, in vivo, that Ifnb1 transcription occurs in microglia in chronic neurodegenerative disease and is STING-dependent. To our knowledge, this is the first demonstration that STING activation drives detrimental microglial IFN-I responses in chronic neurodegenerative disease. A recent study showed that the IFN-I response to traumatic brain injury to be STING-dependent (Abdullah et al., 2018). Interestingly, STING has been also recently been shown to be activated by the antiviral drug ganciclovir and to suppress neuroinflammation in EAE (Mathur et al., 2017). However, EAE is characterized by significant immune cell infiltration, and in that setting, IFNAR1 plays a beneficial role (Prinz et al., 2008).  Figure S1), and escape of damaged dsDNA from the nucleus can result in cytoplasmic sensing by cGAS-STING. Similarly, mitochondrial DNA is a strong stimulus for STING (Carroll et al., 2016), and mitochondrial damage is prominent in the inner and outer mitochondrial membranes in the ME7 model (Siskova et al., 2010). Finally, phagocytosis of cell-free DNA by microglia may also lead to STING activation (Marsman, Zeerleder, & Luken, 2016).
These possibilities require further study.

| Impact of IFNAR1-deficiency on microglial phenotype
We have previously shown major increases in phagocytic and lysosomal transcripts and increased phagocytic activity in ME7 animals (Hughes et al., 2010), and here, we show that a large number of these transcripts are suppressed in IFNAR1 −/− mice. Initial experiments used RNA isolated from homogenates of the hippocampus and thalamus, the major regions of pathology in the ME7 model, and the patterns observed in those crude preparations were almost entirely replicated in microglial cells isolated, by FACS sorting, from these same regions. Most inflammatory transcripts were suppressed in FIGURE 9 COX-1 mediated PGE2 production is diminished in IFNAR1 −/− . (a) PGE2 is synthesized at elevated levels in ME7 animals with respect to normal brain homogenate (NBH) (at 17-18 weeks postinoculation), and this is blocked by the COX-1 inhibitor SC-560 (30 mg/kg i.p.) (n = 17, 26, and 9 for NBH, ME7, and ME7 + SC-560 groups, respectively). (b) COX-1 mRNA, measured by quantitative PCR from RNA isolated from the hippocampus and thalamus, is significantly increased in ME7 animals with respect to NBH, and this is reduced in ME7 IFNAR1 −/− mice (n = 7, 7, and 6 for NBH, ME7, and ME7 IFNAR1 −/− groups, respectively. (c) Hippocampal/thalamic PGE2 measured by ELISA, is significantly increased in ME7 animals with respect to NBH, and reduced in ME7 IFNAR1 −/− mice (n = 4, 4, and 6 for NBH, ME7, and ME7 IFNAR1 −/− groups). All data are plotted as mean ± SEM. ** and *** denote p < 0.01 and p < 0.001 by Bonferroni analysis after a significant main effect of treatment by one-way ANOVA IFNAR1 −/− mice, although this was not the case for all transcripts examined. Il1b, Arg1, and Nos2 were not altered by IFNAR1 deletion in our study, and these changes along with the suppression of Trem2 in IFNAR1 −/− are distinct from profiles in other degenerative models: In the APP/PS1 double transgenic model of AD, disease-associated increases in mRNA for iNOS were reversed in APP/PS1 × IFNAR1 −/− FIGURE 10 Neuropathological and neurological impact of IFNAR1 deficiency during prion disease. (a) Normal brain homogenate (NBH) and ME7 mice on WT and IFNAR1-deficient backgrounds were labeled with Sy38 (anti-synaptophysin) and anti-NeuN antibodies to assess (a) synaptic and (c) neuronal loss, respectively, and these were then quantified by density ratio analysis (b) and manual cell counting (d), respectively (n = 5 for both ME7 groups and 4 for NBH). * denotes p < 0.05 by Student's t-test. (e, f ) NBH and ME7-inoculated wild-type and IFNAR1 −/− mice were assessed weekly on the horizontal bar (e) and inverted screen (f ) tests for motor coordination and muscle strength. Significant differences between ME7 groups on the horizontal bar following Bonferroni post hoc tests after two-way repeated measures ANOVA, are denoted by *.
Our observation that microglia from IFNAR1 −/− ME7 mice became activated but did not upregulate Trem2 and did attain the moredamaging phagocytic phenotype might be explained by the reduction in Trem2 expression.
Conversely, Tnfa showed similar disease-associated IFNAR1dependent induction in both APP/PS1 and ME7 models ( Figure 5 and (Minter et al., 2016). Both Il1b and Tnfa are reported to be elevated in the MPTP model of Parkinson's disease, and IFNAR1 deficiency was also shown to reverse these, although fold-increases in these transcripts were relatively small and rather variable (Main et al., 2016).
It is likely that impaired STAT1 signaling contributes to the changes in microglial activation observed here and STAT1 levels are known to be lower in IFNAR1 −/− mice . However, it has recently emerged that IFN-β may also signal through IFNAR1 without dimerization with IFNAR2 and without activation of the JAK-STAT pathway (de Weerd et al., 2013). This signaling is believed to drive a more pro-inflammatory state than canonical IFN-I signaling, inducing IL-1β and TREM1. Here, IFNAR1 deficiency did not impact

| Phagocytic and lysosomal changes in microglia
The major pattern observed in the current transcriptional analysis was a suppression of genes associated with phagocytosis and lysosomal activity including scavenger receptors, components of the complement pathway, and NADPH oxidase and lysosomal cathepsins. There is evidence for a role for complement-mediated opsonization of synaptic elements in models of AD (Hong et al., 2016;Stevens et al., 2007), and activation of phagocytic and lysosomal pathways is conserved across multiple neurodegenerative diseases (Holtman et al., 2015). Recent studies of repeated systemic LPS-induced neurodegenerative changes in the substantia nigra showed a marked transcriptional shift toward increased complement and phagosome pathway activation, including the products of many of the genes we found to be IFN-dependent in the current study, such as p22phox, gp91, C3, cathepsin S, CD68, and components of the C3 receptor complex (Bodea et al., 2014). In the repeated LPS-model, degeneration was shown to be C3-dependent, giving credence to the notion that driving excessive phagocytic activity in microglia may be detrimental to neuronal integrity. There is also direct evidence that inhibition of microglial phagocytosis limits inflammatory neuronal death in other model systems (Neher et al., 2011). Therefore, there is evidence to support the idea that mitigating these pathways, in this case via deletion of IFNAR1, will slow degenerative processes.
Cathepsin D is a relatively ubiquitously expressed lysosomal protein. We observed strong IFNAR1-dependent upregulation of cathepsin D, and this new synthesis was dominated by increased lysosomes/lysosomal activity in the microglial population. These cathepsin D-rich microglia are very similar to those observed in Niemann-Pick C mice (German et al., 2002) and like those authors, we speculate that this represents significantly higher lysosomal degradative activity in wild-type ME7 animals and that this may contribute to the neuronal and synaptic loss seen in this region. Although cathepsin D is also important for neuronal integrity (Shacka et al., 2007), its expression appears to be preserved in surviving thalamic neurons and indeed IFNAR1 −/− mice actually show greater protection against disease-associated loss of thalamic neurons (Figure 7d).
It may be significant that ME7-associated increases in IBA-1, cathepsin D, and CD68 expression in white matter (cc and internal capsule) were mitigated in IFNAR1 −/− mice. Recently, a deficiency in the ubiquitin specific protease 18 (usp18) was shown to specifically activate white matter microglia, and this was mediated specifically by a loss of regulation of STAT1 signaling and uncontrolled IFN-I activity (Goldmann et al., 2015). This resulted in a loss of microglial "quies-

| Impact of IFNAR1-deficiency on disease progression
Prion disease-induced neurological impairments were manifest from about 17 weeks in wild-type ME7 animals, as previously reported, and the onset of these impairments was delayed by approximately 2 weeks in IFNAR1 −/− ME7 animals. Kaplan-Meier analysis showed that these mice also survive on average 2 weeks longer than wild-type controls were not different in IFNAR1 −/− mice (Supplementary Information Figure S1). This is an important observation because it indicates that neither early clearance of the surgically inoculated infectious material (Beringue et al., 2000) nor early stages in disease development are affected by non-responsiveness to IFN-I. That IFNAR1-deficiency does not affect those processes but does affect later stages of disease indicates that the IFN-I response is part of a disease-associated inflammation that occurs secondary to disease but nonetheless does contribute to the rate of decline. Such longitudinal analysis has typically not been performed in other model systems in which IFNAR1 has been shown to be influential.
In stressing the importance of IFN-I influence on microglial function, it is important to emphasize two other aspects of disease.
(1) Proteinase K-resistant PrP and extracellular formic-acid resistant PrP were equivalent in histological sections of WT and IFNAR1 −/− mice, indicating that the observed changes in microglial phenotype are not secondary to changes in tissue PrP deposition.
Thus, IFN-I and STING are drivers of chronic neurodegeneration in prion disease. This dovetails with older studies with interferon inducers (as a proposed treatment for what was then believed to be a viral disease) showing accelerated prion disease (Allen & Cochran, 1977). Moreover, our own prior studies showed that multiple peripheral challenges with the IFN-I inducer poly I:C produced repeated acute disease exacerbations and accelerated the progression of disease .

| CONCLUSION
Despite its well-established beneficial anti-inflammatory effects in viral infections (Isaacs & Lindenmann, 1957;Pestka, Langer, Zoon, & Samuel, 1987) and the widespread use of IFNβ as a therapy for relapsingremitting MS (Goodin et al., 2002) and protective effects of both STING and IFN-I in EAE (Mathur et al., 2017;Prinz et al., 2008), the results of the current study show that STING-mediated IFN-I induction has multiple effects on microglial phenotype and contributes to chronic neurodegeneration. Protective effects against disease progression in the current study could not be explained by effects on eIF2α or on PrP deposition indicating that modulation of microglial activity may be key to the protective effects observed. The demonstration that IFN-I arises via STING activation establishes this pathway as a potential target in neurodegenerative disease and the observed changes in microglial phenotype expand our understanding of the impact of IFN-I on microglial function during chronic neurodegeneration.