ATF3 Is a Key Regulator of Macrophage IFN Responses

Recognition of danger signals derived from bacteria and viruses by innate immune receptors leads to activation of specific transcriptional programs, culminating in the production of potent inflammatory mediators, including proinflammatory cytokines, chemokines, and type I IFNs (1). The activation of numerous innate immune receptors, including several TLRs (TLR3, 4, 7, 8, and 9), retinoic acid–inducible gene I, melanoma differentiation-associated protein 5, stimulator of IFN genes (STING), and DNA-dependent activators of IFNs, leads to type I IFN induction (2). Production of type I IFNs is particularly important for mounting a rapid antiviral response, including restricting viral replication and modulating adaptive immunity (3, 4). Type I IFNs signal via the IFN-α/β receptor (IFNAR), which induces the expression of IFN-stimulated genes (ISGs), encoding proteins with diverse antiviral functions (5, 6). Genome-wide transcriptional analysis of IFN-treated cells has revealed the existence of hundreds of ISGs, of which only a few have been characterized to date (7–10). The production of type I IFNs by pattern recognition receptors (PRRs) and the subsequent expression of ISGs are therefore critical for host protection, and, not surprisingly, mice and humans with defects in IFN responses are more susceptible to viral infections (11–13).

Inappropriate activation of PRRs can lead to excessive production of inflammatory mediators, which can damage local tissues and lead to systemic disease pathologies. High levels of type I IFN production are implicated in several inflammatory conditions and autoimmune diseases (e.g., systemic lupus erythematosus), in which patient leukocytes display an IFN signature of increased ISG expression (14). Type I IFNs have also been shown to be immunosuppressive in various chronic viral and bacterial infections, mediated in part by their induction of IL-10 (3). Therefore, the magnitude and duration of signals emanating from immune receptors must be tightly controlled at multiple steps, including at the receptor, signaling, transcriptional, posttranscriptional, and posttranslational levels (15). Activating transcription factor 3 (ATF3) is a transcriptional modulator that can repress target genes by directly binding specific nucleotide motifs within promoter loci (16). ATF3 is induced during TLR-dependent immune responses and negatively regulates numerous proinflammatory cytokines (e.g., IL-6, IL-12p40) and chemokines (e.g., MIP-1β), and, as such, mice deficient in ATF3 are more susceptible to endotoxin shock owing to excessive cytokine production (16–19). ATF3 can also modulate the expression of IFN-γ in NK cells during the antiviral response to murine CMV infection (20), and it has recently been shown to directly regulate TLR-induced IFN-β expression following its induction by an inhibitor of the class III lipid kinase, PIKfyve (21). Whereas the ability of ATF3 to repress proinflammatory cytokine expression is firmly established, its role in regulating type I IFN responses requires further investigation.

In this study, we show that in the absence of Atf3, primary mouse macrophages display significantly greater basal and PRR-inducible IFN-β expression. We identify a new important enhancer site distal to the Ifnb1 (the gene encoding IFN-β) promoter and demonstrate that ATF3 binding to this regulatory region appears to control the magnitude of IFN-β promoter activity. We also report that type I IFN induces ATF3 in both human and mouse immune cells and that a subset of inducible genes downstream of IFNAR signaling is further modulated in an ATF3-dependent manner. Finally, we show that ATF3 modulates the antiviral response by impacting viral replication. Our findings further demonstrate the role of ATF3 in controlling the IFN response of macrophages, and help to establish ATF3 as a key innate immune regulator.

Reagents included: ultrapure LPS (Escherichia coli 0111:B4), poly(I:C), and R848 (Invivogen); human and murine IFN-α and IFN-β (R&D Systems); human and murine IFN-γ (ImmunoTools and PeproTech); anti-ATF3 (C19; sc-188, Santa Cruz Biotechnology); anti–p-IFN regulatory factor (IRF)3 (Ser³⁹⁶) (4D4G; 4947, Cell Signaling Technology); anti–β-actin (926-42210) and secondary Abs (LI-COR Biosciences); anti-IFNAR1 Ab (BioLegend); purified mouse IgG1 (Life Technologies); actinomycin D, BSA, anti-hemagglutinin (HA; H6908), and 10-carboxymethyl-9-acridanone (CMA; also known as 9-oxo-10(9H)-acridineacetic acid) (Sigma-Aldrich); and 5′-triphosphate RNA (pppRNA), which was synthesized as described previously (22).

Primary bone marrow–derived macrophages (BMDMs) were obtained by culturing bone marrow of 6- to 8-wk-old wild-type (WT) or Atf3^−/− C57BL/6 mice in DMEM supplemented with 10% FBS, 10 μg/ml ciprofloxacin, and 40 ng/ml recombinant human M-CSF (R&D Systems) for 6 d. Human PBMCs were purified from buffy coats over a Ficoll density gradient (GE Healthcare). Human monocytes or plasmacytoid dendritic cells (DCs) were further isolated using specific negative selection kits (Miltenyi Biotec). Cell purity was >75% on average for isolated cell populations as determined by flow cytometry: CD14⁺ monocytes (61D3; eBioscience) and CD123⁺ and CD303/BDCA-2⁺ plamacytoid DCs (AC145 and AC144; Miltenyi Biotec). Human cells were maintained in RPMI 1640 with 10% FBS and 10 μg/ml ciprofloxacin. WT, Atf3^−/−, and Ifnar1^−/− immortalized BMDM (iBMDM) cell lines were generated as previously described (23) and, along with HEK293T cells, were maintained in DMEM with 10% FBS and 10 μg/ml ciprofloxacin.

A retroviral plasmid expressing a C-terminal HA-tagged version of ATF3 was generated by amplifying ATF3 (pCMV6-mATF3; NM_007498, MC201919, OriGene Tech) using specific primers (forward, 5′-TTTTTGGCGCGCCTATGATGCTTCAACATCCAGGCCAG-3′, reverse 1, 5′-GGGGACATCGTATGGGTAGCTCTGCAATGTTCCTTCTTTTATCTG-3′, and reverse 2, 5′-AAAAAAGCGGCCGCTTCACGCGTAGTCGGGGACATCGTATGGGTAGC-3′) before being cloned into pRP_CMV-HA-IRES-mCherry with AscI and NotI. WT and Atf3^−/− immortalized macrophages were subjected to retroviral transduction with the retroviral plasmid above using a defined protocol (24). Additionally, the empty pRP_CMV-HA-IRES-mCherry parental vector was used to generate control cell lines.

RNA was isolated according to the manufacturers’ protocol (Qiagen) and synthesized into cDNA. Quantitative real-time PCR (qPCR) was performed on cDNA using the Maxima SYBR Green/ROX qPCR Master Mix (Fermentas) on a 7900T thermocycler (Applied Biosystems). The mouse and human primer sequences used can be supplied upon request. Human ATF3 was examined using predeveloped probe/primer combinations (ATF3; Hs00910173_m1 and HPRT1; Hs01003267_m1; Applied Biosystems) with the TaqMan qPCR assay system (Applied Biosystems). Expression of target genes was normalized to respective housekeeping genes.

Levels of IFN-β in culture supernatants were measured using a custom-made ELISA protocol as described elsewhere (25).

Cells were lysed on ice with 1× RIPA buffer (20 mM Tris-HCl [pH 7.4], 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 10% glycerol, 0.1% SDS, and 0.5% deoxycholate) supplemented with 0.1 μM PMSF, cOmplete protease inhibitors, and PhosSTOP (Roche). Lysates were clarified by centrifugation at 13,000 × g for 10 min at 4°C before protein concentration was measured by BCA assay (Pierce). Protein expression was measured by immunoblotting as previously described (26).

The coding sequence of murine Atf3 was amplified (forward, 5′-CTGTCTCGAGACCATGATGCTTCAACATCCA-3′, reverse, 5′-CTGTCCCGGGTTAGCTCTGCAATGTTCC-3′) and cloned into the pIRES2Ac-GFP1 (Invitrogen) expression vector via XmaI and XhoI restriction sites with an additional insertion of a Kozak sequence in front of the start codon. The murine Ifnb1 promoter was amplified (forward, 5′-CTGTCTCGAGTTTCTCTTATAGTACACT-3′, reverse, 5′-CTGTAGATCTGAGCTGCTTATAGTTGAT-3′) and inserted into the promoterless pGL4.14 vector (Promega) directly in front of the firefly luciferase cassette (Luc2) using the XhoI and BglII restriction sites. Additionally, the potential enhancer region and ATF3 binding site BS2 (mm9, chr4: 88182925-88183823) was amplified (forward, 5′-CTGTGGTACCTCTTACAAGGAAGAGGACGAGAGAAC-3′, reverse, 5′-CTGTCTCGAGTCATTTCTGGGTATTCTG-3′) and inserted in front of the Ifnb1 promoter via KpnI and XhoI. In parallel, a mutated BS2 was designed and synthetized (Integrated DNA Technologies). Insertion of the mutated BS2 was enabled via In-Fusion HD reaction (Clontech) according to the manufacturer’s protocol. Resequencing of the pGL4.14 reporter construct confirmed the directed insertion of the DNA fragment. HEK293T cells (InvivoGen) were plated at 25 × 10⁵ cells per well in a 96-well plate. Cells were transfected the following day using GeneJuice (Novagen) in OptiMEM with 50 ng IFN-β promoter firefly luciferase construct, 50 ng either ATF3 or an empty vector control plasmid (pIRES2Ac-GFP1), and finally an internal Renilla luciferase control vector pGL4.74 (Promega). The following day cells were stimulated with Lipofectamine (Invitrogen)-transfected pppRNA (also known as IVT4). Cells were lysed 16 h after stimulation in passive lysis buffer (Promega) and firefly luciferase was measured using luciferin (Promega), whereas Renilla luciferase was measured using coelentrazine (Promega). The firefly signal was then normalized to the Renilla signal.

Cells were lysed in TRIzol reagent (Invitrogen) and total RNA isolation, quality control, and purification were performed as described (27). Generation of biotin-labeled cRNA was performed using the TargetAmp Nano-g biotin-cRNA labeling kit for the Illumina system (Epicentre). cRNA (1.5 μg) was hybridized to MouseWG-6 v2.0 BeadChips (Illumina) and scanned on an Illumina iScan device.

Raw intensities of expression data were imported into BeadStudio 3.1.1.0 (Illumina) and exported as log₂-transformed expression values into Partek Genomics Suite v6.6 (Partek). Batch effects were removed from normalized data. Differentially expressed genes with a fold change ≥ 2 (p ≤ 0.05) were included for further analysis. Identification of ISGs was performed by screening differentially expressed genes against the Interferome database v2.0 (28).

Binding of ATF3 to promoters was predicted using Genomatix v3.2 (Supplemental Table III). For identification of ATF3 binding motifs in the Ifnb1 and cis-regulatory regions, Vector NTI v10.3.0 was used with degeneration settings (V$ATF.01) of 62.5%.

SRA files of used chromatin immunoprecipitation sequencing (ChIP-seq) data were converted to FASTQ files using fastq-dump 2.2.0 and subsequently aligned with Bowtie against the mm9 reference genome using the best match options. Respective bedGraph files were either downloaded from Gene Expression Omnibus or SAM files were converted by HOMER into tag directories to perform peak calling and conversion into bedGraph files. Visualization of normalized tag counts for transcription factor and histone modification ChIP-seq data at the murine Ifnb1 locus and the 13 genes encoding IFN-α were performed with the Integrative Genomics Viewer (v2.3.34).

Cell supernatants were collected and subjected to gas chromatography–mass spectrometry–selected ion monitoring to determine levels of 25- and 27-HC.

WT or Atf3^−/− primary BMDMs were infected with lymphocytic choriomeningitis virus (LCMV) clone 13 as indicated. For intracellular detection of LCMV nucleoprotein 24 h postinfection, cells were first stained with a fixable viability dye (Life Technologies), treated with fixation and permeabilization solutions (BD Biosciences), and then stained with an Ab specific to LCMV nucleoprotein (clone VL4).

Control or ATF3-expressing Atf3^−/− iBMDM cells were infected with vesicular stomatitis virus (VSV)*ΔG(Luc) replicon virus particles (29) as indicated. Cells were lysed in passive cell lysis buffer (Promega) and firefly luminescence measured using a SpectraMax reader (Molecular Devices).

Microarray data are accessible at Gene Expression Omnibus under GSE61055 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE61055) and GSE44034 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE44034). ChIP-seq data are accessible under GSE36104 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE36104) (30), GSE55317 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE55317) (31), GSE38379 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE38379) (32), and GSE63339 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE63339) (33).

Data are typically presented as mean ± SEM, where a p value ≤0.05 was considered significant as determined by an unpaired (mouse experiments) or paired (human experiments) two-tailed Student t tests, unless otherwise described in individual figure legends. Analyses were performed with Prism (GraphPad Software), and for microarray data with Partek Genomics Suite using ANOVA models.

ATF3 is an inducible transcriptional repressor in innate immune cells that regulates the magnitude and duration of inducible proinflammatory gene expression. We performed transcriptome analysis of resting WT and Atf3-deficient (Atf3^−/−) BMDMs to identify other potential pathways regulated by ATF3. As presented in Fig. 1, this approach revealed a strong IFN signature of spontaneous ISG expression in Atf3^−/− BMDMs. Indeed, of the 50 genes most highly expressed in Atf3^−/− BMDMs compared with their WT counterparts, most appeared to be ISGs, including Ifit2, Irf7, Isg15, Ch25h, and Usp18 (Fig. 1A). We next examined the genes that were expressed 2-fold greater in Atf3^−/− BMDMs than WT cells (157 genes) for known ISGs (Fig. 1B) (28), revealing 90 ISGs (57%), 33 of which were specific to type I IFN, 10 specific to type II IFN, and 47 common to both type I and II IFNs (Fig. 1C). We validated this general observation by qPCR, finding significant increases in the basal expression of several well-characterized ISGs (Irf7, Isg15, Ch25h, and Usp18) in Atf3^−/− BMDMs compared with WT cells (Fig. 1D). Of note, we also detected significantly higher basal expression of Ifnb1, the gene encoding IFN-β (Fig. 1E). These findings led us to hypothesize that in addition to regulating proinflammatory cytokines, ATF3 may also be a negative regulator of type I IFN responses in macrophages.

Production of type I IFNs from immune cells constitutes an important part of host defense against infection and occurs following the activation of several PRRs. To examine the role of ATF3 in the context of PRR-driven IFN-β production, we compared responses between WT and Atf3^−/− BMDMs following activation of TLR4 (LPS), TLR3 [poly(I:C)], or STING (CMA). We saw significantly elevated Ifnb1 mRNA expression in Atf3^−/− BMDMs in response to PRR activation (Fig. 2A–C), which was also observed in response to LPS over time (Fig. 2D). The increase in mRNA correlated with greater IFN-β protein release (Fig. 2E–H). Interestingly, whereas the kinetics of IFN-β mRNA induction remained unchanged between WT and Atf3^−/− BMDMs (Fig. 2D), at the protein level IFN-β production was notably delayed in WT cells (Fig. 2H). Consistent with these observations, LPS-induced mRNA expression of the ISGs Irf7 and Isg15 was greater in Atf3^−/− BMDMs (Fig. 2I). However, no difference in Irf7 and Isg15 mRNA was observed between WT and Atf3^−/− BMDMs when directly stimulated with IFN-β (Fig. 2J). This finding suggested that elevated production of IFN-β by TLR4 activation (Fig. 2G, 2H) is likely to account for the increased expression of these ISGs in Atf3^−/− cells. Consistent with this notion, elevated levels of Irf7 and Isg15 mRNA in Atf3^−/− BMDMs compared with WT BMDMs were lost in the presence of an IFNAR blocking Ab (Fig. 2K). Taken together, these results indicate that ATF3 is important for regulating production of IFN-β levels downstream of innate immune receptors and are in line with recent findings from Cai et al. (21) who observed that induction of ATF3 by a PIKfyve inhibitor suppressed subsequent TLR-induced type I IFN production in RAW264.7 cells.

We next performed studies to assess the effects of ATF3 overexpression on inducible IFN-β levels in macrophages. We generated Atf3^−/− and WT iBMDMs overexpressing HA-tagged murine ATF3 or an empty vector control (Fig. 3A–C) and tested TLR4- or STING-induced IFN-β production in these cells. In contrast to ATF3 deficiency, ATF3 overexpression resulted in either significant decreases or a trend toward reduced LPS- and CMA-induced Ifnb1 in both Atf3^−/− and WT iBMDMs (Fig. 3D, 3E). Consistent with the ability of ATF3 to limit IL-12p40 production in macrophages (16), we also observed a significant reduction in LPS-induced Il12b (the gene encoding IL-12p40) and a trend toward a decrease in CMA-mediated Il12b expression in cells overexpressing ATF3 (Fig. 3F). The findings that ATF3-deficient macrophages displayed a basal IFN signature and increased PRR-inducible IFN-β, whereas ATF3 overexpression led to reduced IFN-β induction, suggest that ATF3 is a key regulator of macrophage IFN-β production and downstream IFNAR responses.

We next investigated how ATF3 modulates IFN-β production in response to activation of distinct PRRs. A common feature leading to IFN-β induction downstream of numerous PRRs is the activation of the transcription factors IRF3 and IRF7 (34, 35). In plasmacytoid DCs, IRF7 is constitutively expressed and is the predominant IRF involved in Ifnb1 transcription, whereas in macrophages its expression is induced upon IFN stimulation. In contrast, IRF3 is constitutively expressed in macrophages and its activation leads to the production of type I IFNs following PRR stimulation. Hence, we investigated whether ATF3 affected IRF3 in mouse macrophages. However, we saw no obvious difference in basal IRF3 phosphorylation between WT and Atf3^−/− BMDMs, or following LPS or poly(I:C) stimulation (Fig. 4A, 4B). There was also no difference in the amount of total IRF3 between WT and Atf3^−/− BMDMs (Fig. 4A, 4B). This finding suggests that ATF3, in keeping with its characterized role as a transcriptional modulator, may act downstream of IRF3 activation. We also found that ATF3 deficiency did not affect LPS-induced Ifnb1, Ch25h, or Tnf mRNA stability, as assessed by mRNA degradation following actinomycin D treatment (Supplemental Fig. 1A). Because ATF3 is known to bind DNA, we hypothesized that ATF3 may regulate IFN-β at the transcriptional level. We therefore investigated whether ATF3 can directly bind to the Ifnb1 promoter in resting BMDMs. First, we examined two independent ATF3 ChIP-seq datasets, including one we generated using WT and Atf3^−/− BMDMs (26, 30), for enrichment of ATF3 at sites proximal to the Ifnb1 transcription start site (TSS). In both datasets a peak was called distal to the TSS of Ifnb1 (from −14,200 to −15,175 bp; termed site 2), whereas a site closer to the TSS (from +13 to −467 bp; termed site 1) was called only in the dataset not controlled by ChIP-seq analysis of Atf3^−/− cells (Fig. 4C). Indeed, in our dataset this region only showed a peak in Atf3^−/− samples. Of note, examination of our ChIP-seq dataset identified no significant peaks in the promoter loci (up to −20 kb) of the 13 genes encoding IFN-α (data not shown). As a second approach to assess the likelihood of ATF3 binding to the Ifnb1 promoter and the distal binding site 2, we performed bioinformatic analysis of both identified sites, revealing 11 ATF3 binding motifs at site 1 and 25 at site 2, suggesting that ATF3 binds at site 2 (Supplemental Fig. 1B, 1C, Supplemental Table I). As a potential cis-regulatory element, binding site 2 could be important for Ifnb1 gene expression. Pivotal for the binding of transcription factors to defined enhancers is an open chromatin structure, which is achieved by posttranslational modifications (e.g., methylation and acetylation) of histone proteins and the binding of pioneer transcription factors, for example, PU.1 (36–39). Indeed, so-called poised enhancers, which are required for inducible gene expression, are defined by PU.1 binding and monomethylation of the histone H3 at lysine 4 (H3K4me1) and acetlyation of histone H3 at lysine 27 (H3K27ac) (32). We used available ChIP-seq data from BMDMs stimulated for 4 and 24 h with LPS (32) to assess the enhancer classification of our novel ATF3 binding site 2 (Fig. 4D). Whereas untreated and activated macrophages showed similar H3K4me1 marks at the designated binding sites 1 and 2, basal levels of PU.1 and H3K27ac marks increased with prolonged LPS signaling at both sites. This was expected for site 1, which contains the well-characterized IFN enhanceosome (40) (e.g., IRF, NF-κB, and AP-1 binding sites); however, these data show that binding site 2 is also a poised enhancer (Fig. 4D). This observation was further corroborated by detailed analysis of ChIP-seq data for the two histone marks, H3K4me1 and H3K27ac, in resident tissue macrophages from five different tissues (Supplemental Fig. 1D) (33). Both H3K4me1 and H3K27ac marks were found in all tissue macrophages at binding site 2, supporting the hypothesis that site 2 is epigenetically accessible for transcription factors, and that ATF3 may therefore act by inhibiting transcriptional activator binding at this site. We further characterized these two potential ATF3 binding sites by performing luciferase assays in HEK293T cells. To this end, we generated three murine Ifnb1 promoter constructs containing site 1 alone, sites 1 and 2, or site 1 and a mutated version of site 2. Whereas all promoter constructs were activated upon stimulation with the retinoic acid–inducible gene I agonist, pppRNA (also known as IVT4), interestingly the promoter construct containing both the WT site 1 and site 2 was activated significantly more than was the promoter construct containing site 1 alone (Fig. 4E). Of note, cotransfection of ATF3 significantly reduced activity of the Ifnb1 promoter construct containing WT sites 1 and 2, compared with cotransfection of a control plasmid (Fig. 4F). However, cotransfection of ATF3 with the Ifnb1 promoter constructs containing either WT site 1 alone or site 1 and a mutated site 2 showed no difference (Fig. 4F). We conclude that site 2 represents an important regulatory element for controlling Ifnb1 expression and that ATF3 binds to this regulatory site to limit transcription and subsequent IFN-β production.

ATF3 is an early response gene, induced following activation of innate immune cells by a range of stimuli (41). As PEGylated IFN-α was shown to induce ATF3 in human PBMCs (42), we hypothesized that ATF3 may also be induced by IFN-β to act in a negative feedback loop. Hence, we examined ATF3 expression upon type I (IFN-α and IFN-β) and type II (IFN-γ) IFN treatment of BMDMs. We observed that type I IFNs induced ATF3 gene and protein expression, whereas little or no induction was seen in response to IFN-γ (Fig. 5A, 5B). The responsiveness of BMDMs to treatment with the different IFNs was demonstrated by examining mRNA expression of the ISG Irf7 (Supplemental Fig. 1E). Notably, ATF3 protein expression was induced to comparable levels in BMDMs across several doses of IFN-β (Fig. 5C). The kinetics of Atf3 mRNA induction by IFN-β appeared delayed, with significant increases first observed after 6 h (Fig. 5D). Additionally, a similar, albeit more rapid, ATF3 induction profile was observed in iBMDMs treated with IFN-β (Supplemental Fig. 1F). Increases in expression of the ISG Irf7 demonstrated that these cells responded normally to IFN-β (Supplemental Fig. 1G). Whereas ATF3 induction in response to IFN-β was fully dependent on Ifnar1, LPS-mediated ATF3 induction was only partially IFNAR-dependent, particularly at later LPS time points (Fig. 5E). To date, the immune function of ATF3 has been primarily characterized in mouse cells, where only one isoform is expressed. Humans, however, express six different isoforms of ATF3 (43), some of which are thought to act as transcriptional activators rather than repressors due to truncations in the C-terminal DNA-binding leucine zipper domain. We therefore investigated whether the full-length repressive isoform of ATF3 was induced by IFNs in human monocytes using specific qPCR primers for detection. We found that low-dose type I IFNs significantly induced expression of full-length ATF3, whereas IFN-γ treatment resulted in a more modest increase (Fig. 5F). CXCL10 was strongly induced by all IFNs, demonstrating responsiveness to IFN treatment (Supplemental Fig. 1H). Furthermore, we found that in addition to monocytes, IFN-β induced ATF3 and CXCL10 in both human plasmacytoid DCs and PBMCs (Fig. 5G–I, Supplemental Fig. 1I–K). IFN-β induction of full-length ATF3 protein in human monocytes from four individual donors was further confirmed using an Ab that recognizes a sequence close to the C terminus (Fig. 5J). We also observed reduced expression of a smaller reactive protein that likely represents the ATF3b isoform (containing an intact C terminus) (43), which remains to be functionally characterized. Taken together, these data demonstrate that ATF3 induction by type I IFNs is conserved between murine and human immune cells, potentially forming part of an IFN-β negative feedback loop.

We next investigated the functional relevance of ATF3 induction by type I IFNs with respect to ISG expression. 25-Hydroxycholesterol (25-HC) is an oxysterol that displays potent antiviral activity (44, 45). Additionally, Ch25h, the gene encoding cholesterol 25-hydroxylase, which controls the production of 25-HC, is classified as an ISG (46). Interestingly, ATF3 was identified as a negative regulator of Ch25h in response to foam cell formation and TLR4 stimulation (47). Consistent with this, we observed that Atf3^−/− BMDMs expressed significantly more Ch25h (Fig. 6A) and 25-HC (Fig. 6B) in response to LPS compared with WT cells. In contrast, we saw no difference in production of LPS-induced 27-HC (Supplemental Fig. 1L), an oxysterol not under the control of Ch25h. Unlike the ISGs tested earlier (Fig. 2K, 2L), direct IFN-β treatment of Atf3^−/− BMDMs led to significantly greater Ch25h gene expression compared with WT cells (Fig. 6C). The increase in Ch25h gene expression correlated with greater 25-HC release from Atf3^−/− BMDMs (Fig. 6D), confirming Ch25h as an ISG that is also a direct ATF3 target gene. Furthermore, Atf3^−/− BMDMs showed consistently higher Ch25h mRNA levels upon LPS or IFN treatment, even in the presence of an IFNAR blocking Ab (Fig. 6E), unlike Irf7 or Isg15 (Fig. 2K, 2L). To examine the possibility that ATF3 regulates other genes induced downstream of IFNAR signaling, we performed transcriptome analysis of WT and Atf3^−/− BMDMs stimulated with IFN-β for 6 h. Using the model presented in Fig. 6F, we identified 34 IFN-β–inducible genes that were further induced in the absence of Atf3 (Fig. 6G). Of note, several of these genes have known inflammatory functions, including the chemokines CCL3 and CCL12 (Supplemental Table II). We further examined the promoter regions of the 34 genes identified for conserved ATF3 binding sites, discovering that 23 contained one or more of these sites (Supplemental Table III). We next validated the increased expression of identified ATF3 target genes Ccl12, Ccl3, Ifitm6, and Clec4e in Atf3^−/− versus WT BMDMs upon 6 h IFN-β treatment (Fig. 6H–K), using Ch25h as a positive control (Fig. 6L). As expected, no difference was observed for Irf7 (Fig. 6M). IFN-β–induced Ccl3 and Ccl12 expression was also consistently higher over a time course in Atf3^−/− BMDMs (Supplemental Fig. 1M). These findings demonstrate that ATF3 can regulate a subset of genes induced in response to IFN-β.

Production of type I IFNs and ISGs is important for mediating host antiviral responses. We hypothesized that macrophages deficient in Atf3 would be more efficient at clearing viral infections due to increased IFN-β and consequently increased ISG expression. We therefore infected WT or Atf3^−/− primary BMDMs with LCMV and assessed viral expression 24 h postinfection by measuring LCMV viral Ag in cells by flow cytometry. We found a marked reduction in LCMV viral replication within Atf3^−/− BMDMs compared with WT BMDMs (Fig. 7A). Finally, we infected Atf3^−/− iBMDMs overexpressing ATF3 or an empty vector control with VSV replicon particles encoding a firefly luciferase reporter [VSV*ΔG(Luc)] (29). Consistent with results from primary BMDMs, we observed a significant increase in viral replication when cells expressed ATF3 (Fig. 7B). Taken together, these data demonstrate that ATF3-dependent regulation of type I IFN responses is important for an appropriate antiviral response in macrophages.

We have demonstrated that ATF3 modulates basal and PRR-inducible IFN-β levels in macrophages by directly regulating Ifnb1 mRNA expression. These findings are consistent with a recent study showing that TLR-induced type I IFN levels were reduced by a PIKfyve inhibitor in RAW264.7 murine macrophage cells via induction of ATF3 (21). Immune cells can rapidly respond to low systemic concentrations of type I IFNs, which are constitutively maintained under homeostatic conditions by the commensal microflora (48, 49). Indeed, low-level basal IFN-β expression induces IFNAR signaling via an autocrine loop and is important for priming immune cells for rapid responses to microbial insults (3). Thus, ATF3 modulation of basal IFN-β production represents an important threshold mechanism to ensure that appropriate responses to pathogens are elicited. In line with the ability of IFN-β to signal in an autocrine manner, we observed higher basal and LPS-inducible ISG expression in Atf3^−/− BMDMs, potentially due to higher circulating levels of IFN-β in these cells (Figs. 1D, 2I). Indeed, when we treated Atf3^−/− and WT BMDMs with LPS, we found that increased Irf7 and Isg15 mRNA expression in Atf3^−/− cells was due to increased IFN-β, as this effect was blocked by the addition of an IFNAR blocking Ab (Fig. 2K). The observation that Atf3^−/− BMDMs display higher basal ISG expression may be due to the use of M-CSF (also known as CSF-1) in our cultures. M-CSF is the major growth factor controlling the differentiation of macrophages from the bone marrow (50), and it is also known to induce low levels of basal IFN-β production from macrophages (51). In the absence of regulation by ATF3, M-CSF is likely to trigger more basal IFN-β production, subsequently resulting in increased expression of ISGs. Similar to Atf3^−/− BMDMs, spontaneous basal ISG expression is also observed in macrophages deficient in other important IFN regulators, including TREX1, SAMHD1, and the transcriptional repressor FOXO3 (52-55).

Considering the potential detrimental effects of excessive IFN-β to the host, its production is tightly controlled, including on the transcriptional level. For instance, cooperative binding of the transcription factors IRF3/7, NF-κB, and ATF-2/Jun to a highly conserved regulatory element within the Ifnb1 promoter, termed the enhanceosome, is required for optimal IFN-β induction (56). Additionally, virus infection induces nucleosome remodeling of the Ifnb1 promoter (via histone acetyl transferases, as well as the SWI/SNF chromatin remodeling complex), allowing for increased transcription (56). Consistent with the role of ATF3 as a transcriptional repressor, two recent studies suggested that ATF3 has binding sites within the Ifnb1 promoter (21, 30). Our ChIP-seq dataset (26) has identified an ATF3 binding site at a regulatory element 15 kbp upstream of the Ifnb1 TSS, which showed significant and specific enrichment of ATF3 under basal conditions (Fig. 4C). ATF3 interacts with histone deacetylase 1 (16), which can counteract the effects of histone acetyl transferases to maintain a more closed chromatin conformation to limit transcription. Of note, histone deacetylase 1 has previously been shown to repress IFN-β expression (57). Collectively, this suggests that ATF3 may regulate the magnitude of Ifnb1 expression by inducing a less transcriptionally active chromatin state, which is also reflected by higher basal levels of Ifnb1 gene expression in Atf3^−/− macrophages. The importance of the regulatory element we have termed site 2 for transcriptional regulation of Ifnb1 is underscored by its methylation and acetylation status, which suggests that it is indeed a poised enhancer in both LPS-stimulated BMDMs and in vivo in tissue macrophage populations (Fig. 4D, Supplemental Fig. 1D). Furthermore, in luciferase reporter assays, addition of site 2 increased inducible Ifnb1 promoter activity (Fig. 4E), which was reduced when ATF3 was also present (Fig. 4F). These intriguing findings suggest that other positive transcriptional regulators of IFN-β may bind this distal Ifnb1 enhancer (site 2), and potentially be displaced upon ATF3 binding. Future studies that further examine the role of this new enhancer site and ATF3 will be of great interest. Interestingly, whereas ATF3 modulated the level of PRR-inducible IFN-β mRNA expression in macrophages, it did not appear to affect its kinetics (Fig. 2D), which is likely due to the existence of additional regulatory mechanisms acting on IFN-β mRNA following immune activation (58). At the protein level, however, we saw an obvious delay in IFN-β production in WT compared with Atf3^−/− BMDMs (Fig. 2H).

Previously, ATF3 was found to be IFN-α inducible in human PBMCs (42). In the present study, we further show that ATF3 is induced in both mouse and human immune cells in response to both IFN-α and IFN-β. Although all type I IFNs are known to signal via the same receptor containing the IFNAR1 and IFNAR2 subunits, we observed differences in ATF3 induction by IFN-α and IFN-β (Fig. 5A, 5B). This may be due to observations made that suggest IFN-β can have differential effects on IFN-α on a variety of biological processes, including gene transcription (59–61). In general, induction of ATF3 is promiscuous, with numerous studies showing that a range of stimuli (e.g., TLRs, HDL, PIKfyve inhibitors) and cellular conditions (e.g., cellular stress) can induce its expression (16, 18, 26, 41). This suggests that both the cell type and the specific context under which ATF3 is induced may dictate its functional role. Furthermore, our data support the work presented by Whitmore et al. (18) showing that TLRs that elicit IFN-β production (i.e., TLR3 or TLR4) can induce ATF3 expression in a biphasic manner (Fig. 5E) via an early IFNAR-independent wave of transcription and via a second, delayed IFNAR-dependent wave. Whereas the first, acute wave is described for regulating proinflammatory cytokine production (16), the second IFN-dependent induction of ATF3 was not investigated until now. Although it is likely that IFN-mediated ATF3 expression constitutes an important component of a negative feedback loop for regulating IFN-β, or further downstream modulation of IFN-γ expression (20), we have now discovered that IFN-β–mediated ATF3 is also important for controlling expression of a subset of genes downstream of IFNAR signaling. Given that ATF3 expression was also robustly induced in human immune cells by type I IFNs (Fig. 5F–I), it is possible that a similar regulatory system exists in humans. Of the ISGs we identified as direct ATF3 target genes, the most highly induced was Ch25h. Cholesterol 25-hydroxylase (encoded by Ch25h) catalyzes the formation of 25-HC during cholesterol metabolism, and ATF3 was previously found to directly regulate Ch25h expression (47). Additionally, separate studies have characterized Ch25h as an ISG, and 25-HC has been shown to elicit broad antiviral functions (44, 45). In this study, we show, to our knowledge for the first time, that ATF3 can regulate the expression of Ch25h following IFN signaling, a mechanism that is likely important for maintaining an appropriate immune response to viruses. Indeed, a recent study found that whereas production of 25-HC is advantageous during the acute stages of influenza viral infection, its presence ultimately results in excessive inflammation and severe tissue pathology via amplification of proinflammatory gene expression (62). Furthermore, IFN-induced 25-HC was shown to broadly suppress IL-1–activating inflammasomes (63), in line with the known immunosuppressive activity of type I IFNs (3). Consistent with a role in dampening inflammation, we also found that ATF3 regulates the expression of the IFN-inducible chemokines CCL3 and CCL12, which are required for inflammatory monocyte recruitment (64). The overrepresentation of cell surface markers/receptors and chemokines (Supplemental Table II) in addition to Ch25h suggests that ATF3 might be induced by IFN to limit subsequent inflammation and modulate the adaptive immune response.

Overall, our findings suggest that regulatory mechanisms exerted on IFN-β and ISGs by ATF3 are important for an appropriate and balanced immune response: too little ATF3 expression results in exaggerated IFN-β and ISG production, which can lead to chronic inflammation or immunosuppression, whereas too much ATF3 results in enhanced viral replication, owing to repression of IFN-β and antiviral mediators. Our work reinforces the importance of ATF3 as a key player within the gene regulatory networks of the immune system, and it highlights the potential for modulating ATF3 expression or activity therapeutically to combat chronic viral infections. Moreover, further investigation into the use of ATF3 inducers (e.g., HDL, PIKfyve inhibitors) in diseases involving unwarranted IFN production and signaling, such as in systemic lupus erythematosus, would be of great interest.

We thank B.R.G. Williams (Monash Institute of Medical Research–Prince Henry’s Institute, Clayton, VIC, Australia) and M. Röcken (University of Tubingen, Tubingen, Germany) for providing bones from Atf3^−/− mice (originally from T. Hai, The Ohio State University, Columbus, OH); P.J. Hertzog (Monash Institute of Medical Research–Prince Henry’s Institute) for bone marrow from Ifnar^−/− mice and helpful discussions; G. Zimmer (Institute of Virology and Immunology, Bern, Switzerland) for VSV*ΔG(Luc) replicon particles; and D. Kalvakolanu (University of Maryland, College Park, MD) for J2 recombinant retroviruses. We acknowledge T. Cavlar and M. Schlee (University of Bonn) for providing pppRNA and helpful discussions. We also acknowledge H. Theis, M. Kraut, and S. Martin of the Life and Medical Sciences Institute, Bonn, Germany; A. Kerksiek and B. Putschli of the University of Bonn; and P. Langhoff of the Institute of Innate Immunity for technical support.

The authors have no financial conflicts of interest.