TLR4 Ligands Induce IFN-α Production by Mouse Conventional Dendritic Cells and Human Monocytes after IFN-β Priming¹

The type I IFNs, IFN-α and IFN-β, are critical cytokines for innate immune responses against viral and bacterial infection and contribute to the linking of innate and adaptive immunity (1, 2). They constitute a family that includes multiple IFN-α isoforms and one IFN-β (3). IFN-α is implicated in the pathogenesis of the autoimmune disease systemic lupus erythematosus (SLE),⁴ in the initiation as well as the exacerbation of the disease (3, 4, 5).

IFN-α production in SLE may derive from both endogenous and exogenous stimuli (6). The major endogenous stimuli are thought to be immune complexes consisting of autoantibodies and self-nucleic acid that induce IFN-α production by activation of TLR7 or TLR9 in plasmacytoid dendritic cells (pDCs). This occurs through uptake of the immune complex by FcγRIIa on the pDC cell surface and the subsequent internalization and delivery of the self-DNA or self-RNA within the complex to intracellular TLR9 or TLR7, respectively (7, 8, 9). Viral and bacterial infections are frequently associated with clinical flares of human lupus (10, 11, 12, 13, 14) and represent potential exogenous stimuli (6). Large amounts of type I IFN, mainly IFN-α, are produced after exposure to viruses (15, 16), with pDC being the major source (17). The ability of bacteria to induce robust IFN-α or IFN-β protein production is less well established. However, the TLR4 ligand LPS, a major component of Gram-negative bacteria, can induce IFN-α gene expression in splenocytes in vivo (18), although it generally only induces IFN-β and not IFN-α gene expression in dendritic cells (DCs) in vitro (18, 19, 20, 21).

The pDCs produce the greatest amount of IFN-α on a per cell basis. However, conventional DCs (cDCs) can produce appreciable levels under certain conditions including infection by cDC-tropic dsRNA viruses (22) or after internalization of class A CpG oligodeoxynucleotides (ODNs) by liposomal transfection (23). IFN-α production by pDC is dependent on activation of TLR7 or TLR9 by ssRNA or CpG-DNA, respectively. IFN-α production by cDCs induced by viral infection depends on viral dsRNA engagement of the cytosolic RNA helicases RIG-I (retinoic acid-inducible gene I) and MDA5 (melanoma differentiation-associated gene 5) or on that of TLR3 (24, 25).

TLRs use distinct signaling pathways to induce type I IFN production. TLR7 and TLR9 transmit signals solely through the adaptor molecule MyD88 (26). In contrast, TLR3 induction of IFN-α and IFN-β and TLR4 induction of IFN-β gene expression are MyD88 independent (27, 28, 29, 30, 31). The Toll/IL-1R domain-containing adaptor inducing IFN-β (TRIF) (32) is required for this MyD88-independent type I IFN production (28) and, in the case of TLR4, in association also with TRAM (TRIF-related adaptor molecule) (33). The MyD88-dependent and MyD88-independent pathways then interact with members of the IFN regulatory factor (IRF) family of transcription factors that serve to regulate type I IFN gene transcription (34). Four IRFs—IRF1, IRF3, IRF5, and IRF7—have been implicated as positive regulators of the transcription of type I IFN genes (35). The MyD88-dependent pathway, recruited by TLR7 and TLR9 ligation, interacts directly with IRF7 (but not with IRF3) at the endosomal compartment to induce robust IFN-α production (36). The MyD88-independent pathway, recruited by TLR3 and TLR4 ligation via the adaptor TRIF, uses mainly IRF3 for IFN-β gene induction (18), although IRF7 can substitute for IRF3 under certain conditions (18). The role of IRF5 in type I IFN production is less clearly understood, although studies in human cell lines and in IRF5-deficient mice suggest a meaningful contribution (37, 38, 39, 40, 41).

The term “priming” as used in this report refers to the pretreatment of cells, originally described with type I IFN pretreatment, such that the cells are converted from nonproducers to producers of type I IFN upon a subsequent stimulus (42). We hypothesized that when DCs are already primed, as may frequently be the case in SLE, LPS would then induce the production of IFN-α. In this study we show that supernatants from DCs activated by RNA-containing immune complexes can effectively prime cDCs to make substantial amounts of IFN-α on subsequent LPS stimulation. We further show that this effect can be recapitulated using recombinant IFN-β and we use this system to obtain novel insights into the pathways responsible for IFN-α production in cDCs. This study also suggests one possible mechanism whereby bacterial infection might induce disease flares in patients with SLE.

C57BL/6 wild-type (WT) mice and TRIF-deficient mice (Trif^Lps2/Lps2 backcrossed four generations to C57BL/6) were purchased from The Jackson Laboratory. TLR4-deficient mice (backcrossed eight generations to C57BL/6) and MyD88-deficient mice (backcrossed twelve generations to C57BL/6) have been described previously (43, 44). IRF5-deficient mice (backcrossed eight generations to C57BL/6) and IRF7-deficient mice (backcrossed three generations to C57BL/6) were provided by Dr. T. Taniguchi (University of Tokyo, Tokyo, Japan) and Dr. T. Mak (Campbell Family Institute for Cancer Research and University of Toronto, Toronto, Canada) (36, 45). FcR common γ-chain-deficient mice (B6.129P2-Fcerg1^tm1. backcrossed 12 generations to C57BL/6) were obtained from Taconic. All mice were maintained at the Boston University School of Medicine Laboratory Animal Sciences Center (Boston, MA) or at Charles River Laboratories in accordance with the regulations of the American Association for the Accreditation of Laboratory Animal Care. All experimental procedures were approved by the Institutional Animal Care and Use Committee at Boston University School of Medicine.

CpG-A oligodeoxynucleotide 2336 (GGg gac gac gtc gtc gtg GGGGGG) was purchased from Coley Pharmaceutical Group. Capital letters show phosphorothioate backbone, and lowercase letters show phosphodiester backbone. Polyinosinic:polycytidylic acid (Poly(I:C)) and ultra-pure LPS were purchased from InvivoGen. (S)-[2,3-bis(palmitoyloxy)-(2-RS)-propyl]-N-palmitoyl-(R)-Cys-(S)-Ser-(S)-Lys₄-OH, 3HCl (Pam₃Cys) was a gift from Prof. G. Jung, University of Tübingen, Tübingen, Germany.

Bone marrow cells were seeded at 1.5 × 10⁶ cells/ml in complete RPMI 1640 (10% FBS, 2 mM l-glutamine, 50 μM 2-ME, 100 U/ml penicillin, and 100 μg/ml streptomycin) supplemented with 7.5% conditioned medium from Fms-like tyrosine kinase 3 ligand (FL)-transfected B16 cells. The FL B16 cells were originally made by Dr. H. Chapman (46) and provided by Dr. U. Von Andrian (both from Harvard Medical School, Boston, MA). The cells were used for experiments after 8 days, at which time >90% were CD11c⁺, of which 15–40% displayed a pDC phenotype (CD11c⁺CD45RA^highB220^highCD11b^low) and the remainder displayed a conventional DC phenotype (CD11c⁺CD45RA^lowB220^lowCD11b^high).

Bone marrow cells were seeded at 0.5 × 10⁶ cells/ml in complete RPMI 1640 together with 6.7 ng/ml recombinant mouse GM-CSF (BD Biosciences) and 400 pg/ml recombinant mouse IL-4 (R&D Systems). On day 6, the cells were collected and the CD11c⁺ cells were isolated using magnetic bead positive selection with anti-CD11c beads (Miltenyi Biotec). Purity was assessed by flow cytometry after staining with anti-CD11c-PE, anti-CD11b-FITC, and biotinylated anti-B220 followed by streptavidin-PE-Cy5 (BD Biosciences). Consistently, >90% of the cells were CD11c⁺CD11b⁺B220⁻ in keeping with a conventional DC phenotype.

Ribonucleoprotein (RNP)-reactive IgG from two lupus patients (50 μg/ml), IgG from a healthy volunteer (50 μg/ml), Pam₃Cys (1 μg/ml), or medium only were added to FL-DC cultures (bone-marrow cell-derived DCs cultured in FL). After 24 h, the supernatants (lupus IgG supernatant 1, lupus IgG supernatant 2, control IgG supernatant, Pam₃Cys supernatant, and medium supernatant; Table I) were removed, aliquotted, and frozen at −80°C. The sera collection and the IgG purification from the sera have been described previously (40). The study was approved by the Institutional Review Board at Boston Medical Center (Boston, MA) and written informed consent was obtained from all patients and volunteers.

Before setting up each assay, the FL-DC were routinely checked by flow cytometry for the relative percentages of pDC and cDC and the respective DC activation status after staining with anti-CD11c-PE, biotinylated anti-CD45RA followed by streptavidin-PE-Cy5, anti-B220-FITC, anti-CD11b-FITC, and anti-CD62L-FITC Abs (all from BD Biosciences). The GM-CSF DCs were checked as described in the preceding paragraph. FL and GM-CSF DCs were seeded in 96-well, round-bottom plates (5 × 10⁵ cells/well), pretreated with IFN-β (product no. 12400-1; PBL Biomedical Laboratories) or with supernatants (see preceding paragraph) for varying periods (5 h in most experiments), then cultured in complete RPMI 1640 with the appropriate TLR ligands in a total well volume of 200 μl. In certain experiments, the cells were washed before the addition of the TLR ligands. After varying periods, 24 h in most experiments, the supernatants were collected for cytokine measurement. To obtain purified pDC (CD11c⁺B220^highCD11b^low) and cDC (CD11c⁺B220⁻CD11b^high) populations for certain experiments, FL-DCs from 8-day cultures were stained with anti-B220-PE-Cy5, anti-CD11c-PE, and anti-CD11b-FITC Abs (all from BD Biosciences), and the DC populations were separated using a MoFlo cell sorter (DakoCytomation).

PBMCs were isolated from buffy coats by Ficoll gradient centrifugation. To obtain monocytes, initial depletion of T cells was performed by sheep RBC rosetting, after which the remaining T cells, B cells, NK cells, and red cells were depleted by negative immunomagnetic selection with anti-CD2 (clone 95-5-49), anti-CD19 (clone FMC63), and anti-glycophorin (clone 10F7) mAbs (all provided as hybridomas by Dr. S. Shaw, National Cancer Institute-National Institutes of Health, Bethesda, MD) to obtain a >95% pure population of CD11b⁺ monocytes. PBMCs and monocytes were seeded in 96-well, round-bottom plates (1 × 10⁶ cells/well), pretreated with IFN-β for 10 h, and then cultured in complete RPMI 1640 with the appropriate TLR ligands in a total well volume of 200 μl. After 24 h, the supernatants were collected for cytokine measurement. The 10-h pretreatment was used instead of the 5-h pretreatment used in the mouse cDC experiments because in preliminary studies we found that although LPS-induced IFN-α production was seen with the 5-h pretreatment, higher levels of IFN-α production were seen with the 10-h pretreatment.

Cytokine levels were measured using in-house ELISAs developed using commercially available Abs. The protocols for mouse cytokines have been previously described (40). The detection limits of the mouse IL-6, IFN-α, and IFN-β ELISAs are 16, 80, and 130 pg/ml respectively. For the human IFN-α ELISA we used an anti-human IFN-α mAb as the coating Ab (product no. 21112–1; PBL Biomedical Laboratories) and a rabbit polyclonal anti-human IFN-α Ab (product no. 31101-1; PBL Biomedical Laboratories) as the detecting Ab. The detection limit of this assay is 15 pg/ml.

FL-DCs were seeded in complete RPMI 1640 in 24-well, flat-bottom plates (3 × 10⁶ cells/well) in a total well volume of 1 ml and pretreated or not pretreated with 250 U/ml IFN-β (product no. 12400-1; PBL Biomedical Laboratories) for 5 h. The FL-DCs were then washed, resuspended in complete RPMI 1640, and stimulated or not stimulated with 1 μg/ml LPS (InvivoGen) for 1 h. Total RNA from each group of FL-DCs was isolated using the RNeasy Micro kit (Qiagen). Seven-hundred nanograms of RNA was reverse transcribed using SuperScript II reverse transcriptase (Invitrogen) and quantitative real time PCR (Applied Biosystems StepOnePlus instrument and software) using TaqMan probes and primers (Applied Biosystems) was performed to determine the expression levels of IFN-β, IFN-α2, IFN-α4, and IFN-α5 target genes. The Δ-ΔCt threshold cycle method was used for analysis. All genes of interest were normalized against the housekeeping gene GAPDH and changes were expressed as fold change relative to the non-IFN-β-primed, unstimulated control FL-DC sample.

Data are depicted as mean ± SEM. Statistical significance of differences was determined by the paired. two-tailed Student’s t test.

Mouse bone marrow cells cultured in vitro with FL develop into a mixed population of pDCs and cDCs, collectively referred to as FL-DC (47). We have previously shown that when RNP-reactive IgG from lupus patients are added to FL-DC cultures, RNA-containing immune complexes are formed that induce DC activation in a FcγR- and TLR7-dependent manner with the consequent production of IFN-α, IFN-β, and IL-6 (40). We hypothesized that these cytokines would be able to prime DC and thereby enable them to produce IFN-α on subsequent LPS stimulation, analogous to the situation that might be present in lupus patients exposed to bacterial infection.

To test this hypothesis, we first added RNP-reactive IgG from two lupus patients and IgG from a healthy volunteer to FL-DC cultures. We also added the TLR2 ligand Pam₃Cys as an additional control because it induces strong DC activation but no type I IFN (31, 48). After 24 h, we removed the supernatants (lupus IgG supernatant 1, lupus IgG supernatant 2, control IgG supernatant, Pam₃Cys supernatant, and medium supernatant; Table I) and discarded the cells. We then added the supernatants to new FL-DC cultures and, after 5 h, stimulated the new FL-DC with the TLR3 ligand poly(I:C), the TLR4 ligand LPS, and the TLR9 ligand CpG-A. We used Fc receptor common γ-chain-deficient mice as a source of these new FL-DCs (referred to as FcRγ^−/− FL-DCs) to exclude any direct effects from residual lupus IgG remaining in the supernatants. Consistent with our hypothesis, we found that LPS induced the production of substantial amounts of IFN-α from FcRγ^−/− FL-DCs pretreated with the lupus IgG supernatant 1 and the lupus IgG supernatant 2 (Fig. 1,A). In contrast, pretreatment of FcRγ^−/− FL-DC with Pam₃Cys supernatant containing high levels of IL-6, control IgG supernatant, or medium supernatant (Table I) did not effectively prime the FcRγ^−/− FL-DC for LPS-induced IFN-α production (Fig. 1 A). This led us to postulate that type I IFN might be required for the priming effect, as this was present in the lupus IgG supernatant 1 and the lupus IgG supernatant 2 but was absent in the Pam₃Cys supernatant and the control IgG supernatant. Poly(I:C)-induced IFN-α production was similarly markedly enhanced by pretreatment of the FcRγ^−/− FL-DC with the lupus IgG supernatant 1 and the lupus IgG supernatant 2, although IFN-α production was also seen in the absence of pretreatment consistent with the known ability of TLR3 activation to induce IFN-α as well as IFN-β (19, 49).

To determine whether type I IFN could prime FL-DCs for LPS-induced IFN-α secretion, we pretreated FL-DCs with different concentrations of IFN-β before LPS activation, using the same 5-h pretreatment time that was used in the supernatant experiments. We found that the IFN-β was indeed able to effectively prime the FL-DC, with optimal IFN-α secretion induced by LPS seen when the dose of IFN-β was >75 U/ml (∼2.5 ng/ml); however even a low dose of IFN-β (2.5 U/ml, 85 pg/ml) was sufficient for this effect (Fig. 1 B). We observed a very similar priming effect if the FL-DCs were pretreated with recombinant IFN-α instead of with IFN-β (data not shown).

We next determined how the duration of IFN-β pretreatment affected the priming response (Fig. 1,C). No priming was seen if the IFN-β and the LPS were added to the FL-DC concurrently. Priming was observed with 1 h of IFN-β pretreatment (the earliest time point examined), with the maximal effect evident at 5 h. The IFN-β-primed DCs produced IFN-α in a dose-dependent manner in response to LPS (Fig. 1,D). IFN-β pretreatment, although not absolutely required, also enhanced IFN-α production induced by the TLR3 ligand poly(I:C) and the TLR9 ligand CpG-A (Fig. 1 E). In contrast, Pam₃Cys failed to induce IFN-α, with or without IFN-β pretreatment.

pDCs produce the largest amount of IFN-α on a per-cell basis (17) and can express TLR4 (50, 51). cDCs can also produce IFN-α under certain circumstances (22, 23) and express TLR4 constitutively (52). To identify which DC subset was responsible for the LPS-induced IFN-α production in the IFN-β-primed FL-DC cultures, we separated FL-DCs into cDC (CD11c⁺CD11b⁺B220⁻) and pDC (CD11c⁺CD11b⁻B220⁺) populations by cell sorting (Fig. 1,F). The IFN-β-primed cDCs produced high levels of IFN-α in response to LPS and poly(I:C), whereas these ligands did not induce IFN-α production from pDCs (Fig. 1 G). The TLR9 ligand CpG-A induced IFN-α production only in pDCs.

To confirm these results in a different cDC type, we used CD11c⁺ DCs obtained from bone marrow cells cultured for 6 days with GM-CSF (Fig. 1,H). These cDCs have been extensively used by many investigators, including for the analysis of IFN-β gene induction after LPS stimulation (18, 20, 30), although they differ phenotypically and functionally from the cDCs present in FL-DC cultures (47). We found that LPS and poly(I:C) were able to induce IFN-α production by IFN-β-primed GM-CSF cDCs, whereas the TLR9 ligand CpG-A was unable to do so (Fig. 1 I). The inability of CpG-A to induce IFN-α was not due to the absence of TLR9 in the GM-CSF DC, as CpG-A induced IL-6 production. Thus, only the TLR3 ligand poly(I:C) and the TLR4 ligand LPS induced IFN-α production in IFN-β-primed FL-cDCs and GM-CSF cDCs. These TLRs are unique in that they both use the TRIF adaptor in signaling (28, 29, 30), suggesting that TRIF might be required for LPS-induced IFN-α secretion by IFN-β primed cDCs.

We used ultra-pure LPS for all our experiments and so it was likely that all observed effects would be mediated through TLR4 (53). However, to confirm this, we compared LPS-induced IFN-α production by IFN-β-pretreated FL-DCs from WT and TLR4^−/− mice and found that, as expected, the IFN-α production was completely TLR4 dependent (Fig. 2 A).

In DCs, IFN-α production induced by TLR3 ligands and IFN-β gene induction by TLR3 and TLR4 ligands is independent of MyD88 (27) but requires the adaptor TRIF (20, 28, 29, 30). We initially assumed that IFN-α production induced by LPS in IFN-β-primed FL-DCs would also be TRIF dependent and MyD88 independent. To test this, we compared type I IFN production by WT and TRIF-deficient IFN-β-pretreated FL-DCs and found that TRIF was indeed required for IFN-α and IFN-β production induced both by LPS and poly(I:C) (Fig. 2,B). However, although MyD88 was not required for IFN-α production induced by poly(I:C), IFN-α production induced by LPS was substantially MyD88 dependent (Fig. 2,C). Furthermore, IFN-β protein production induced by LPS was also substantially MyD88 dependent, whereas that induced by poly(I:C) was MyD88 independent (Fig. 2 C).

These unexpected results led us to hypothesize that in the case of TLR4-induced type I IFN production there might be sequential use of the TRIF and MyD88 pathways, with TRIF being required initially and MyD88 being involved subsequently. To test this, we compared the kinetics of IFN-α and IFN-β production in IFN-β-pretreated FL-DCs from WT, MyD88-deficient, and TRIF-deficient mice. We collected supernatants at different periods of time between 0 and 2 h, 2 and 12 h, and 12 and 24 h following LPS activation (Fig. 2,D). In the first 2 h, IFN-α and IFN-β production induced by both TLR3 and TLR4 ligands was MyD88 independent. In contrast, between 2 and 12 h, although type I IFN production induced by poly(I:C) remained MyD88 independent, IFN-α and IFN-β production induced by LPS was mostly MyD88 dependent. No IFN-α or IFN-β production was detected between 12 and 24 h. In the case of IFN-β-pretreated FL-DCs from TRIF-deficient mice, no TLR3 or TLR4-induced IFN-α or IFN-β production was detected at any of the time points examined (Fig. 2 E). Thus, although TRIF is absolutely required for the initial production of IFN-α induced by LPS in IFN-β-primed cDC, MyD88 also plays an important role later in the activation pathway.

The transcription factor IRF7 contributes partially to TLR4-induced IFN-β gene activation, although it is essential for IFN-α/β gene induction by TLR7 and TLR9 in pDC and for virus-mediated MyD88-independent IFN-α/β gene induction in cDC (36). IRF5 contributes partially to IFN-α and IFN-β production elicited both by viruses and TLR3, TLR7, and TLR9 agonists (37, 39, 40, 41). To determine the role of these transcription factors in TLR4-mediated IFN-α production, we compared responses of IFN-β-primed FL-DCs from WT, IRF7^−/−, and IRF5^−/− mice.

Although IRF7 contributes only partially to LPS-induced IFN-β gene activation (36), we found that LPS-induced production of IFN-α was abolished in IFN-β-primed FL-DCs from IRF7^−/− mice (Fig. 3,A). Similarly, poly(I:C)-induced IFN-α production, which is MyD88-independent (Fig. 2,C), was entirely IRF7 dependent (Fig. 3,A). As expected, the robust production of IFN-α by pDC induced by the TLR9 ligand CpG-A was absent in FL-DCs from IRF7^−/− mice. These studies demonstrate that not only is IRF7 the master regulator of type I IFN production induced by TLR7 and TLR9 ligands and by viral infection (36), but it is also required for IFN-α production elicited by TLR3 and TLR4 ligands. The absence of robust IFN-α production in the IRF7-deficient cDCs was not due to an inability of these cells to produce cytokines, as Pam₃Cys-induced IL-6 production by the IRF7-deficient cDC was at least as high as that in the WT cDC (Fig. 3,A). The somewhat lower IL-6 production observed with poly(I:C), LPS, and CpG-A activation likely reflects a decrease in the type I IFN feedback loop required for IL-6 production (40). In IRF5^−/− FL-DC, IFN-α production induced by LPS was significantly, but not completely, reduced as compared with WT FL-DC (Fig. 3 B). IFN-α production induced by both CpG-A and poly(I:C) was decreased by more than two-thirds in IRF5^−/− FL-DCs, demonstrating an important role for IRF5 in IFN-α production via MyD88-dependent and TRIF-dependent/MyD88-independent pathways.

Following viral infection of fibroblasts, induction of the IFN-α4 gene is evident after 4 h, whereas the induction of other IFN-α genes including IFN-α2, IFN-α5, IFN-α6, and IFN-α8 is delayed, with gene induction being evident only after 8 h (54). We therefore evaluated whether a similar sequential activation of the IFN-α genes would be seen in the IFN-β-primed FL-DCs stimulated by LPS. After IFN-β priming, FL-DCs were stimulated with LPS for 1 h and IFN gene expression was measured by quantitative real-time PCR. We found that even at that early time point IFN-α2, IFN-α4, and IFN-α5 as well as IFN-β were all strongly induced by LPS after IFN-β priming (Fig. 4). In contrast, in the absence of IFN-β priming, LPS did not induce IFN-α2, IFN-α4, or IFN-α5, although IFN-β gene induction was seen, consistent with previous reports demonstrating that the IFN-β gene is induced in macrophages 1 h after TLR4 activation (Fig. 4) (29). Thus, IFN-β priming enables the rapid induction of multiple IFN-α genes by LPS.

To determine whether human cells would respond similarly as mouse cells, we started by testing human PBMCs obtained from healthy donors. We found that PBMCs pretreated with IFN-β produced IFN-α in response to both LPS and poly(I:C) (Fig. 5,A). Of the various DC subsets and DC precursors present in human peripheral blood, only monocytes express TLR4 and respond notably to LPS stimulation (52). We therefore repeated these experiments using purified monocytes and found that IFN-β-pretreated monocytes produced IFN-α when stimulated with LPS (Fig. 5 B). In contrast, IFN-β pretreated monocyte-derived DCs generated in vitro with GM-CSF and IL-4 did not produce IFN-α in response to LPS or poly(I:C), although they did produce IL-6 in response to LPS (data not shown). This suggests that mouse cDCs are functionally more similar to human monocytes than to human monocyte-derived DCs, at least in terms of IFN-α production in response to TLR4 and TLR3 ligands.

pDCs are thought to be the principal sources of IFN-α production in most situations (17). However, cDCs are also able to produce appreciable levels of IFN-α under certain conditions, such as with viral infection by cDC-tropic dsRNA viruses (22) or after internalization of class A CpG ODNs by transfection with liposomes (23). In this study we show that TLR4 activation by LPS is able to induce robust IFN-α production by mouse cDC and human monocytes that have been pretreated with type I IFN.

The critical role of the type I IFN feedback loop as an enhancer of IFN-α production is well established (55, 56). This refers to the ability of small amounts of type I IFN to engage the type I IFN receptor and thereby induce a variety of receptors, adaptors, and transcription factors that serve to augment IFN-α production in response to a subsequent stimulus. Although this has been studied most extensively in fibroblasts, it also plays an important role in other cell types including DCs, although pDCs may be less reliant on this feedback loop under some circumstances (57). It has therefore been puzzling why, although TLR4 is able to strongly induce IFN-β gene expression, it does not generally induce IFN-α gene expression in vitro (other than IFN-α4) (18, 20, 27, 35, 52). This contrasts with TLR3, which is able to induce both IFN-β and IFN-α (19). Even with IFN-β pretreatment of GM-CSF-derived cDC for 16 h, a previous study found no induction of IFN-α gene expression with LPS activation (18). However, in human macrophages pretreated with IFN-α for 16 h, LPS did induce IFN-α gene expression (58). Our findings provide one explanation for this apparent discrepancy by showing that the length of pretreatment with IFN-β is critical, with the priming effect being maximal with 5 h of pretreatment and declining thereafter. Thus, at least in cDCs, a pretreatment duration of 16 h may be too long to sustain the priming effect. This could potentially be due in part to the short half-life of certain of the components of the IFN-α inducing pathway that are up-regulated by IFN-β pretreatment, for example IRF7, which has a half-life of ∼1 h (54, 59).

However, the TLR9 ligand CpG-A was unable to induce IFN-α in cDCs despite the established ability of type I IFN pretreatment to up-regulate many components of the TLR9 pathway required for IFN-α production in pDCs such as MyD88 and IRF7 (18, 60). Furthermore, CpG-A was able to induce IL-6 in cDCs, demonstrating intact TLR9 signaling pathways and a selective impairment of IFN-α induction. This indicated that additional factor(s) specific to the TLR4 signaling pathway must be induced by IFN-β pretreatment that enable IFN-α production following TLR4 activation.

IL-6 likely plays a critical role in the B cell hyperactivity and immunopathology of human SLE (61) and has been linked to disease pathogenesis in murine lupus (62). IL-6 and type I IFN also mediate protective effects against a variety of infectious organisms (63, 64). In this study we found that DC supernatants containing type I IFN (lupus IgG supernatant 1 and 2; Fig. 1,A and Table I), but not DC supernatants containing IL-6 (Pam₃Cys supernatant; Fig. 1,A and Table I), could prime cDCs for TLR4-induced IFN-α production. We have previously reported that IL-6 production by DCs in response to a number of different TLR ligands including TLR4 ligands is type I IFN dependent (40). Thus although IL-6 does not directly prime cDCs for type I IFN production, in the context of an infection the type I IFN induced by TLR4 activation could lead to enhanced IL-6 production through this feedback loop. The increased levels of IL-6 and type I IFN could then contribute to recovery from the infection but worsening of the SLE.

Most TLRs are absolutely dependent on the expression of MyD88 for signaling. However, TLR3 signaling is completely MyD88 independent and TLR4 signaling is partially MyD88 independent, with both TLR3 and TLR4 being unique in their ability to use the adaptor TRIF (26, 28). TRIF expression is up-regulated by type I IFN pretreatment (58), and this therefore represents one possible mechanism whereby IFN-β pretreatment might enable IFN-α production by cDCs in response to LPS. Consistent with this possibility, LPS failed to induce IFN-α production in TRIF-deficient cDCs.

Unexpectedly however, we found that IFN-α and IFN-β protein production induced by LPS in type I IFN-pretreated cDCs was also substantially MyD88 dependent. This contrasted with IFN-α and IFN-β induced by poly(I:C), which we found to be entirely MyD88 independent, consistent with the findings of other investigators (30, 32, 65). Most studies evaluating IFN-β production in response to LPS have focused on gene expression rather than protein production and have examined events soon after LPS stimulation. These studies clearly demonstrate that the early induction of IFN-β gene expression after LPS stimulation is MyD88 independent (20, 27, 28, 29, 30). However, where later time points have been examined, lower levels of IFN-β gene expression were found in MyD88-deficient DCs compared with MyD88-sufficient DCs (20). Our findings with IFN-β protein production demonstrate that early production is MyD88 independent (but TRIF dependent), whereas later production is largely MyD88 dependent. IFN-α protein production from IFN-β-pretreated cDCs induced by LPS followed a very similar course. The TLR4-mediated induction of proinflammatory cytokine genes such as TNF and IL-6 also depends on both the MyD88 and TRIF pathways (26). However, in the case of proinflammatory cytokines, MyD88-deficient cells lack early activation whereas TRIF-deficient cells show impairment in late activation (66), the opposite of what we found for IFN-α and IFN-β production. This concept of cooperative interaction between the TRIF and MyD88 pathways is further supported by a number of studies showing synergy between ligands for TRIF-associated TLRs and ligands for MyD88-associated TLRs for the production of inflammatory cytokines (48, 67, 68, 69).

Previous studies have shown that IRF3 rather than IRF7 is essential for LPS-induced IFN-β gene induction, because induction of the IFN-β gene in response to LPS is abolished in DCs from IRF3-deficient DCs (18, 70) but is almost normal in DCs from IRF7-deficient mice (35, 36). However, when DCs are pretreated with recombinant IFN-β, LPS is able to induce IFN-β mRNA expression in DCs from IRF3-deficient mice (18). This suggests the possibility that IRF7, which is strongly up-regulated in DCs by IFN-β, might participate in TLR4 signaling under certain conditions (18, 35). Our results demonstrate that IRF7 is absolutely required for LPS-induced IFN-α production by IFN-β-pretreated DCs. This formally establishes that IRF7 can participate in TLR4 signaling and is consistent with the established role of IRF7 as the master regulator of IFN-α responses (36).

The exact role of IRF5 in type I IFN production is less well established. IRF5 was originally identified in cell lines as a regulator of type I IFN gene expression induced by infection with certain viruses (37, 38) and is a central mediator of type I IFN production induced following TLR7 activation by R848 in HEK293 cells (39). In vivo, IRF5-deficient mice are vulnerable to viral infections and have a reduced level of type I IFN in their sera (41). IRF5 also plays an important role in proinflammatory cytokine production (45). We have previously found that IRF5 participates in the production of both IFN-α and IFN-β induced by TLR3, TLR7, and TLR9 ligands in DCs in vitro (40). In this study, we confirm the involvement of IRF5 in IFN-α production induced by TLR3 and TLR9 ligands and further show that IRF5 also participates in IFN-α production induced by TLR4 signaling.

Monocytes have the capacity to differentiate into either tissue macrophages or conventional DCs. They comprise up to 10% of circulating PBMCs in humans, express high levels of TLR4, and respond to LPS (52). We found that human monocytes pretreated with IFN-β behaved similarly to mouse cDCs in terms of their ability to produce IFN-α on LPS stimulation. Although the amount of IFN-α produced per cell was relatively modest, it is possible that this could translate into substantial levels in vivo in the context of bacterial infection given the large numbers of monocytes present in the circulation. Thus, in situations where monocytes may have been exposed to type I IFN such as in SLE, bacterial infection could further increase IFN-α levels and initiate or exacerbate clinical disease. Cell types other than monocytes could also potentially play a similar role. For example, IFN-α-pretreated macrophages express type I IFN following TLR4 triggering (58). Human myeloid DCs and pDCs express little or no TLR4 and so are unlikely to contribute substantially to LPS-induced IFN-α production (52).

Studies of hospitalized patients have shown that overall disease activity measured by either the SLE Disease Activity Index (SLEDAI) or the Lupus Activity Index (LAI) correlates well with the incidence of infection, the majority of which is bacterial (11, 12, 14). In mouse models of lupus, administration of LPS greatly accelerates disease (71). Although these studies do not directly implicate type I IFN in pathogenesis, it is noteworthy that many of the toxic effects of LPS in vivo require type I IFN induction (72), and type I IFN induces sensitization to subsequent LPS challenge (73, 74). Endogenous TLR4 ligands could act similarly. In this context, modified low-density lipoproteins and fatty acids have been identified as TLR4 ligands (75, 76, 77) and could potentially contribute to the worsening of lupus disease activity seen in the setting of elevated low-density lipoprotein levels (78). TLR4 transgenic mice expressing multiple copies of TLR4 develop lupus-like autoimmune disease in the absence of exogenous stimuli, suggesting that either endogenous TLR4 ligands or commensal bacteria contribute to disease pathogenesis in this setting (79). In summary, we describe a novel pathway for TLR4-induced IFN-α production by cDCs and suggest a possible mechanism whereby bacterial infection or endogenous TLR4 ligands could precipitate disease flares in SLE.

We thank Dr. Tadatsugu Taniguchi for providing the IRF5- and IRF7-deficient mice and Dr. Tak Mak for permission to use the IRF5-deficient mice. We also thank Rocco Richards for excellent technical assistance.

The authors have no financial conflict of interest.