TRIM33 switches off Ifnb1 gene transcription during the late phase of macrophage activation

Despite its importance during viral or bacterial infections, transcriptional regulation of the interferon-β gene (Ifnb1) in activated macrophages is only partially understood. Here we report that TRIM33 deficiency results in high, sustained expression of Ifnb1 at late stages of toll-like receptor-mediated activation in macrophages but not in fibroblasts. In macrophages, TRIM33 is recruited by PU.1 to a conserved region, the Ifnb1 Control Element (ICE), located 15 kb upstream of the Ifnb1 transcription start site. ICE constitutively interacts with Ifnb1 through a TRIM33-independent chromatin loop. At late phases of lipopolysaccharide activation of macrophages, TRIM33 is bound to ICE, regulates Ifnb1 enhanceosome loading, controls Ifnb1 chromatin structure and represses Ifnb1 gene transcription by preventing recruitment of CBP/p300. These results characterize a previously unknown mechanism of macrophage-specific regulation of Ifnb1 transcription whereby TRIM33 is critical for Ifnb1 gene transcription shutdown.

I n response to viral or bacterial infections, germ line-encoded pattern recognition receptors induce type I interferon (IFN-I, encoded by 14 Ifna and 1 Ifnb genes in mice) gene transcription in myeloid and non-myeloid cells 1 . The secreted type I IFNs act early and transiently during the innate immune response to prime adaptive immunity and then type I Ifn gene transcription is shut down 2 . This transient expression of IFN-I is physiologically important. Whereas early IFN-I expression is important to control infection, pathogens that can overcome this initial control benefit from IFN-I mediated immune system suppression 3,4 .
Among type I Ifn genes, Ifnb1 gene regulation has been a model system to study inducible transcription. After pattern recognition receptor-mediated activation, a group of transcription factors, including IRF3/IRF7, NF-kB and ATF2/c-jun, cooperatively binds to a regulatory sequence located between À 102 and À 47 bp upstream of the Ifnb1 transcriptional start site (TSS) to form the Ifnb1 enhanceosome. This enhanceosome recruits CBP/p300 for acetylation of histone H3 and permits recruitment of chromatin-remodeling complexes for initiation of transcription 5 .
Monocyte/macrophage-specific regulation of Ifnb1 gene transcription has only been observed in the modulation of the transcriptional activity of IRF3. In monocytes, a PU.1-binding site located in the enhanceosome is constitutively bound by a IRF8/PU.1 myeloid complex that creates a preformed activation complex. This myeloid complex can recruit IRF3 through direct IRF3/IRF8 interaction and is involved in the rapid transcription of Ifnb1 observed in monocytes after pathogenic stimulation 6 . Conversely, transcriptional regulator MafB can impair the interactions between IRF3 bound to the enhanceosome and coactivators, and thus negatively regulates Ifnb1 transcription 7 . Finally, binding of transcriptional regulator YY1 to different sites within the enhanceosome has been shown to result in activation (at the beginning) or repression (at the end) of Ifnb1 gene transcription in fibroblasts [8][9][10] . Altogether, these results point to the enhanceosome as a regulator of both initiation and termination of the Ifnb1 transcription cycle and of monocyte/ macrophage-specific regulation of Ifnb1 transcription.
The tripartite motif (TRIM) protein family is characterized by a shared N-terminal structure consisting of a RING, two B-box domains, a coiled-coil domain and, for most of the TRIM family members, an E3 ubiquitin ligase activity 11 . TRIM33, together with TRIM24, TRIM28 and TRIM66, form a sub-family of TRIM proteins characterized by a C-terminal plant homeodomain juxtaposed to a bromodomain 12 . These TRIM proteins do not directly bind to DNA but can be recruited by DNA-binding proteins, such as nuclear receptors 13 , SMAD 14 , PU.1 or TAL-1 (refs 15,16), to act as transcriptional regulators.
A large number of TRIM proteins are innate immune enhancers and act at multiple levels in signalling pathways, including production and action of IFN-b to confer antiviral and antibacterial cytokine production 17,18 . As for the TRIM33 sub-family, TRIM24 represses the viral-defense IFN response in hepatocytes as a consequence of repression of VL30 transcription 13,19 . Although these TRIM24 mediated repressions are potentiated by TRIM33, TRIM33 deficiency alone cannot elicit the IFN response in hepatocytes 13,19 . TRIM24 also acts as a negative regulator of the IFN/signal transducers and activators of the transcription signalling pathway 13,19 , and TRIM33 is essential for cytosolic RNA-induced NLRP3 inflammasome activation 20 . However, up to now, no TRIM protein has been shown to directly regulate Ifnb1 gene transcription after recruitment to an Ifnb1 regulatory sequence.
We have previously shown that TRIM33 can interact with PU.1 to repress its transcriptional activity during haematopoiesis and suggested that TRIM33 may regulate myeloid fate and have a role in macrophages 15 . Here, we show that TRIM33 deficiency in macrophages results in sustained expression of Ifnb1 during the late phases of macrophages activation. Our results indicate that TRIM33, recruited by PU.1 on a distal regulatory element of the Ifnb1 gene, regulates Ifnb1 enhanceosome loading and shuts down Ifnb1 gene transcription during the late stages of lipopolysaccharide (LPS) activation of macrophages by preventing recruitment of CBP/p300.

Results
TRIM33 controls Ifnb1 expression in activated macrophages. Trim33 fl/fl mice were crossed with mice expressing the Cre recombinase from the endogenous Lysozyme2 locus (LyzCre) to generate Trim33 À / À mice. Trim33 À / À mice did not display developmental abnormalities and were healthy. PCR genotyping of purified haematopoietic cell populations showed deletion of Trim33 in monocytes, bone-marrow-derived macrophages (BMDM), peritoneal macrophages (PMs) and neutrophils but not in myeloid progenitors, nor in B lymphocytes (Fig. 1a, upper panels and Supplementary Fig. 1a). The deletion of the Trim33 gene allele was associated with a loss of TRIM33 expression in BMDM (Fig. 1a, lower panel).
The effects of Trim33 deletion on Ifnb1 expression in macrophages were studied by comparing kinetics of Ifnb1 mRNA levels in wild-type (WT) or Trim33 À / À BMDM and PM, upon TLR4 activation by LPS or upon TLR3 activation by poly(I:C). These two activation protocols mimic bacterial (LPS) or viral (poly(I:C)) infections. A rapid early increase of Ifnb1 mRNA levels was observed in WT and Trim33 À / À BMDM or PM activated by LPS (Fig. 1b). Thereafter, Ifnb1 mRNA levels decreased from 2 h in WT BMDM (or 4 h in WT PM) and returned to basal levels by 6 h in WT BMDM and PM. In contrast, although Ifnb1 mRNA levels decreased between 2 and 4 h, Ifnb1 mRNA levels began to rise significantly by 6 h and remained high 24 h after LPS activation in Trim33 À / À BMDM or PM (Fig. 1b). This late sustained expression of Ifnb1 was also observed when BMDM and PM were stimulated by poly(I:C) with Ifnb1 mRNA levels being much higher at all late time points in poly(I:C) treated Trim33 À / À BMDM compared with WT ( Supplementary Fig. 1b). The increased Ifnb1 mRNA levels in Trim33 À / À BMDM activated with LPS correlated with increased levels of secreted IFN-b (Fig. 1c). Finally, using a sub-optimal concentration of LPS (0.1 ng ml À 1 ) which did not activate Ifnb1 in WT BMDM, we showed, in LPS-activated Trim33 À / À BMDM, an increased expression of genes whose transcription is dependent on IFN-b and occurs late during LPS activation (Fig. 1d). This increased expression was not accounted for by different responses of WT and Trim33 À / À BMDM to IFN-b ( Supplementary Fig. 1c) and could be obtained after addition of IFN-b during the late phases of LPS activation of WT BMDM ( Supplementary Fig. 1d). Altogether, these results show that TRIM33 deficiency is associated with increased Ifnb1 mRNA level associated with IFN-b secretion during the late stages of LPS activation of BMDM.
TRIM33 protein was added back to LPS-activated immortalized Trim33 À / À macrophages (iM) 21 . As anticipated, Trim33 À / À iM bore the expected deletion of Trim33 gene and these cells expressed no TRIM33 protein ( Supplementary Fig. 1f). Ifnb1 mRNA levels increased 1 h after LPS activation of WT and Trim33 À / À iM and peaked at 2 h with twofold higher observed expression in WT iM than in Trim33 À / À iM (Fig. 1g). As anticipated, Ifnb1 mRNA levels in WT iM then dropped, whereas in Trim33 À / À iM, Ifnb1 mRNA levels remained high (Fig. 1g). Thus, LPS activation of Trim33 À / À iM mimicked the late phase of the Ifnb1 mRNA kinetics seen in Trim33 À / À BMDM. To address whether expression of ectopic TRIM33 could rescue Ifnb1 downregulation at late time points, a flag-TRIM33-IRES-GFP (or GFP alone) construct was expressed in Trim33 À / À iM and cells were stimulated with LPS (Fig. 1h). The full-length exogenous TRIM33 protein restored Ifnb1 downregulation at late time points of LPS activation of Trim33 À / À iM (Fig. 1h) indicating a role for TRIM33 in the regulation of Ifnb1 gene expression during LPS activation of macrophage.
To determine the TRIM33 protein domains involved in Ifnb1 regulation, mutant flag-TRIM33-IRES-GFP constructs lacking the coiled-coil domain (flag-DCC), the ubiquitin ligase activity (flag-DUb) or the Smad-interaction domain (flag-DSmad) were expressed in Trim33 À / À iM ( Supplementary Fig. 1g). The flag-DUb and the flag-DSmad rescued normal Ifnb1 expression at late time points of LPS-activated Trim33 À / À iM whereas the flag-DCC did not rescue normal profile of Ifnb1 expression in activated Trim33 À / À iM (Fig. 1h). These results indicate that the Smad-interaction and ubiquitin ligase domains are not necessary for TRIM33 regulation of Ifnb1 expression and suggest that TRIM33 is part of a regulatory complex that regulates Ifnb1 transcription.
Altogether, these results reveal a role for TRIM33 in the regulation of Ifnb1 expression specifically in downregulating its expression during the late stages of macrophage activation.
TRIM33 binds a distal Ifnb1 regulatory element. TRIM33 ChIP-seq data obtained from the RAW 264.7 murine macrophage cell line or from BMDM treated with LPS for 24 h showed a single TRIM33 peak at the Ifnb1 locus that overlapped with a PU.1 peak located 15 kb upstream from the Ifnb1 TSS (Fig. 2a). Interestingly, no TRIM33 occupancy was observed at the PU.1 peak located just upstream the Ifnb1 promoter corresponding to the enhanceosome-binding sites. We called this À 15 kb region the Ifnb1 Control Element (ICE). Recently, the human homologue of ICE, named L2, has been shown to be a virus-inducible enhancer of Ifnb1 in non-myeloid cells where it displays promoter and enhancer activities 22 .
Constitutive TRIM33 occupancy at ICE was observed in BMDM although occupancy appeared to increase at late time points (24 h) following LPS activation (Fig. 2b, left panel). PU.1 binding at the Ifnb1 promoter and at ICE was constitutive and detected in WT and Trim33 À / À BMDM (Fig. 2b, middle and right panels) indicating that PU.1 binding to ICE was not TRIM33-dependent. Finally, no TRIM33 binding was observed on the Ifnb1 promoter or ICE in NIH3T3 cells (Fig. 2c), indicating macrophage specificity of TRIM33 in binding to ICE.
To determine if TRIM33-mediated regulation of Ifnb1 in myeloid cells can be obtained in NIH3T3 by simultaneous expression of PU.1 and TRIM33, NIH3T3 cells were transduced with a PU.1 expressing lentivirus ( Supplementary Fig. 2a). PU.1 expressing NIH3T3 were then transduced with the lentivirus that promoted efficient shRNA-mediated depletion of TRIM33 ( Supplementary Fig. 1e) and NIH3T3 PU.1 þ TRIM33 þ and NIH3T3 PU.1 þ TRIM33 À responses to poly(I:C) were studied. Whereas TRIM33 deficiency did not modify kinetics of Ifnb1 mRNA levels in response to poly(I:C) in normal NIH3T3 cells (Fig. 1f), TRIM33 deficiency increased Ifnb1 mRNA levels in NIH3T3 PU.1 þ cells treated with poly(I:C) but did not result in sustained expression of Ifnb1 during the late stages of poly(I:C) activation (Fig. 2d). These results show that TRIM33 can repress Ifnb1 expression in presence of PU.1 and indicate that additional factors that are not present in NIH3T3 cells are required for TRIM33-mediated late downregulation of Ifnb1 expression.
To study the effect of ICE on Ifnb1 promoter activity in myeloid cells, we used RAW 264.7 cells in which LPS regulates Ifnb1 gene expression ( Supplementary Fig. 2b). RAW 264.7 cells were transfected with luciferase reporter constructs containing a 391 bp fragment of the Ifnb1 promoter preceded by a 408 bp fragment corresponding to the central region of ICE (ICE-Ifnb1 Prom-Luc) or the fragment containing the Ifnb1 promoter alone as a control sequence (Ifnb1 Prom-Luc). Transcriptional activity of ICE-reporter (ICE-Ifnb1 Prom-Luc) showed four to tenfold less activity than the promoter reporter in LPS-activated RAW 264.7 cells indicating that ICE functions as a cis-acting transcriptional repressor element in RAW 264.7 cells (Fig. 2e).
To study the role of ICE in RAW 264.7 and NIH3T3 cells, we used a CRISPR/Cas9 lentiviral system to deliver Cas9 and gRNA ( Supplementary Fig. 2c). Using this system, we had to screen 200 RAW 264.7 clones to get one RAW 264.7 clone with monoallelic deletion of ICE ( Supplementary Fig. 2d). This low efficiency of recombination in RAW 264.7 cells does not favour the use of transient transfection for CRISPR/Cas9-mediated deletion of ICE, transient transfection efficiency in RAW 264.7 being 15% compared with 90% using the lentiviral system. RAW 264.7 and NIH3T3 clones bearing this deletion on one allele (termed RAW ICE þ / À and NIH3T3 ICE þ / À , respectively; Supplementary Fig. 2d) were analysed for Ifnb1 expression after LPS or poly(I:C) activation. In LPS-activated RAW WT cells, Ifnb1 mRNA levels increased during the first 4 h and then gradually decreased ( Supplementary Fig. 2e). In contrast, whereas Ifnb1 mRNA levels increased comparably in RAW ICE þ / À cells, they subsequently remained stable ( Fig. 2f and Supplementary  Fig. 2e). In NIH3T3 WT and ICE þ / À cells activated with poly(I:C), kinetics of Ifnb1 mRNA levels were similar ( Fig. 2g and Supplementary Fig. 2f). These results indicate that ICE plays a role in the negative regulation of Ifnb1 expression at late stages of LPS activation in RAW cells.
Sequential deletions centered on the PU.1/TRIM33-binding site of ICE were performed in RAW 264.7 cells using CRISPR/ Cas9 technology ( Supplementary Fig. 2g). Of the 96 clones screened, two carried a biallelic deletion of ICE ( Supplementary  Fig. 2d). Two RAW 264.7 clones bearing a 100 or 176 bp deletion on the two Ifnb1 alleles (termed #34 and #15; Supplementary  Fig. 2g) were analysed for Ifnb1 expression after LPS or poly(I:C) activation. In the two clones high and sustained Ifnb1 mRNA levels were found but only after LPS or poly(I:C) activation, with the highest increase being obtained during the late phase of activation (Fig. 2h).
Altogether, these results show that ICE functions as a cis-acting transcriptional repressor element of the Ifnb1 gene activation in macrophages and indicate that TRIM33 and PU.1 bound to ICE may have an important role in Ifnb1 gene transcription shutdown during the late phase of macrophage activation.  transcription factors ( Supplementary Fig. 3a, left panel). Furthermore, c-JUN, the p65 subunit of NF-kB and IRF3, three transcription factors that form the Ifnb1 enhanceosome and regulate its transcriptional activity, were also bound to ICE three hours after LPS activation of BMDM ( Supplementary Fig. 3a, right panel; data from ref. 23). We studied a possible TRIM33 dependency of c-jun and p65 binding to ICE but found that TRIM33 deficiency did not modify the kinetics of their binding to ICE (Fig. 3a). Analysis of ICE chromatin structure in macrophages showed that this region is characterized by high levels of H3K4me3 and H3K27ac modification, flanked by regions enriched for H3K4me1, all features of transcriptionally active promoters ( Supplementary  Fig. 3b, red graph, data from ref. 24). These epigenetic modifications are present in short-term haematopoietic stem cells (ST-HSCs) and in myeloid cells, but not in mature erythroid and lymphoid cells (Supplementary Fig. 3b). We studied the effect of TRIM33 deficiency on the chromatin structure of ICE in nonactivated and LPS-activated BMDM to ask whether TRIM33 has a role in histone modification of this region. Before and during LPS activation of WT and Trim33 À / À BMDM, no significant differences in H3K4me1 levels on ICE could be detected ( Supplementary Fig. 3c). ChIP and ChIP-seq analyses showed a constitutive binding of RNA Polymerase II (Pol II) and constitutively high levels of H3K4me3 on ICE (Fig. 3b). Prior to LPS activation, histone H3 acetylation (H3ac) on ICE was similar in WT and Trim33 À / À BMDM (Fig. 3c, left panel). During LPS activation, H3ac on ICE did not change in WT BMDM but was transiently increased 10 h after LPS activation in Trim33 À / À BMDM (Fig. 3c, left panel). This increased H3ac on ICE in Trim33 À / À BMDM was associated with transient increase of CBP/ p300 histone H3 acetyl-transferase (HAT) -binding on ICE 10 h after LPS activation (Fig. 3c, right panel). Collectively, database analyses and our results suggest that, in BMDM, ICE exhibits a promoter-like chromatin signature established early during myeloid differentiation. TRIM33 deficiency did not modify ICE chromatin signature and only led to transient increases of H3ac and CBP/p300 recruitment on ICE whereas a sustained high expression of Ifnb1 was observed in LPS-activated Trim33 À / À BMDM suggesting that TRIM33 bound to ICE might act at a distance.
To investigate how TRIM33 bound to ICE, that is located 15 kb upstream of the Ifnb1 transcription unit, may regulate Ifnb1 transcription, chromosome conformation capture experiments followed by deep sequencing (3C-seq) were performed in WT and Trim33 À / À BMDM before and after LPS activation. ICE (Fig. 3d, top) and the Ifnb1 gene (Fig. 3d, bottom), including its promoter, were used as viewpoints. In WT BMDM, both viewpoints showed a loop between ICE and the Ifnb1 gene before LPS activation and a twofold stronger interaction 24 h after LPS activation (Fig. 3d, blue curves). Interestingly, Trim33 deletion did not significantly modify DNA looping in this region, even after LPS treatment (Fig. 3d, compare red (Trim33 À / À ) and blue (WT) curves). These results indicate that ICE interacts with the Ifnb1 proximal region in a constitutive and TRIM33independent manner, but that this interaction is strengthened following LPS stimulation.
TRIM33 represses Ifnb1 by preventing CBP/p300 recruitment. To characterize how TRIM33 might regulate Ifnb1 transcription, we studied the effect of TRIM33 deficiency on temporal binding of p65 and c-jun to the enhanceosome as these two factors are major regulators of its activity. As anticipated, in WT BMDM, c-jun and p65 were recruited to the Ifnb1 promoter at early times following LPS activation (1 h), but their binding returned to basal levels after 24 h (Fig. 4a). In contrast, in the absence of TRIM33, p65 and c-jun binding to the Ifnb1 promoter was maintained even 24 h after activation (Fig. 4a). Thus, TRIM33 deficiency maintains enhanceosome loading by c-jun and p65 during the late phases of LPS activation.
We then determined the effect of TRIM33 deficiency on Ifnb1 transcription and chromatin structure during LPS activation of BMDM. On the Ifnb1 TSS, low levels of paused Pol II were observed before LPS stimulation in both WT and Trim33 À / À cells (Fig. 4b, upper left panel). In WT BMDM, Pol II occupancy at TSS and in Ifnb1 3 0 region increased after 1 h of stimulation and then was lost at 4 h (Fig. 4b, upper panels). In Trim33 À / À BMDM, Pol II occupancy similarly increased after 1 h of stimulation but its levels were maintained up to 24 h (Fig. 4b, upper panels). H3K4me3 levels were transiently increased in WT BMDM after 4 h of LPS stimulation, whereas in Trim33 À / À BMDM, persistent and high levels of H3K4me3 were also observed up to 24 h (Fig. 4b, lower panel and right panel). These observations suggest that the paused Pol II seen in the absence of LPS treatment is used for rapid Ifnb1 transcription immediately following LPS stimulation in WT and Trim33 À / À BMDM. Then, in WT BMDM, Pol II drops off and Ifnb1 transcription ceases, whereas in Trim33 À / À BMDM, continued Pol II occupancy and H3K4me3 promote sustained Ifnb1 transcription.
Prior to LPS activation, histone H3ac on the Ifnb1 promoter was similar in WT and Trim33 À / À BMDM (Fig. 4c). This H3ac increased at 1 h after LPS activation of WT and Trim33 À / À BMDM. It then returned to basal levels in WT BMDM, whereas a 5-to 8-fold increase in H3ac was observed after 10 and 24 h in Trim33 À / À BMDM (Fig. 4c). H3ac on the Ifnb1 promoter in LPS-activated Trim33 À / À iM mimicked that observed in Trim33 À / À BMDM (Fig. 4d). Expression of ectopic full-length flag-TRIM33 in Trim33 À / À iM led to decreased H3ac on the Ifnb1 promoter 7 h after LPS stimulation (Fig. 4d), whereas the flag-DCC that could not rescue normal Ifnb1 expression in Trim33 À / À iM (Fig. 1h) did not decrease H3ac on the Ifnb1 promoter 7 h after LPS stimulation (Fig. 4d). These results indicate a link between TRIM33 and the regulation of H3ac on the Ifnb1 promoter.
Increased H3ac on the Ifnb1 promoter in LPS-activated Trim33 À / À BMDM indicated a possible role for TRIM33 in regulating the equilibrium between recruitment and/or activity of HAT and histone deacetylases (HDAC). Treatment of LPS-activated WT BMDM with trichostatine A, an inhibitor of HDACs, did not result in sustained Ifnb1 expression, whereas it did alter expression of genes regulated by HDACs (Fig. 4e, left panel and Supplementary Fig. 4a). Conversely, the increased H3ac was found to be associated with CBP/p300 HAT binding on the Ifnb1 promoter 10 and 24 h after LPS activation in Trim33 À / À BMDM (Fig. 4e, right panel). As total CBP protein levels were similar in WT and Trim33 À / À BMDM (Fig. 4f), these results indicated that TRIM33 deficiency promotes CBP recruitment. To determine the importance of CBP/p300 activity in the phenotype of Trim33 À / À BMDM, we used C646, a small-molecule inhibitor that efficiently competes with CBP/p300 substrates 25 . Adding C646 2 h after LPS induction did not modify Ifnb1 expression in WT BMDM, but reversed the high sustained Ifnb1 transcription level in Trim33 À / À BMDMs at both 10 and 24 h (Fig. 4g). This reversion was associated with decreased levels of H3ac (Fig. 4h) and reduced Pol II occupancy within the Ifnb1 promoter region in Trim33 À / À BMDM ( Supplementary Fig. 4b).
Altogether, these results show that enhanced CBP/p300 recruitment and activity at late times of activation is required for sustained Ifnb1 expression in Trim33 À / À BMDM and that TRIM33 may regulate Ifnb1 expression at late times by preventing the recruitment of CBP/p300.

Discussion
Although IFN-a and IFN-b share the same receptor, IFN-b seems to be the major immune-suppressive IFN 26 and macrophagerestricted synthesis of IFN-b can contribute to protective or pathological immune responses in the lung 27 . Surprisingly, despite the importance of IFN-b production by macrophages, no cis-or trans-acting factor that might confer a myeloid specificity to Ifnb1 gene transcription has been characterized. The data presented in this study provide evidence of a previously unappreciated regulation of Ifnb1 transcription in macrophages a **  through a macrophage-restricted TRIM33 recruitment on a regulatory sequence, designated as ICE, located 15 kb upstream from the Ifnb1 TSS.
ICE has an open-chromatin structure highly conserved during myeloid differentiation and characterized by Pol II occupancy and tri-methylation of H3K4, two features that are not typical marks of an enhancer or a repressor. In addition to a PU.1binding site, ICE contains DNA-binding sites for transcriptional regulators such as IRF3, c-jun or p65 that are also part of the enhanceosome. ICE has been recently shown to interact with the Ifnb1 promoter in human fibroblasts in the absence of any viral infection 22 . Here, we show that ICE can also interact with the Ifnb1 promoter in non-activated macrophages. These combined findings indicate that ICE/Ifnb1 loop is cell-type independent, stable and independent of viral or bacterial stimulation. Such enhancer or repressor loop stability has been described during development where they are associated with paused polymerase 28 and in human fibroblasts where TNF-a-responsive enhancers contact their target promoters before TNF-a activation 29 .
Transient transfection in fibroblast cell lines shows that ICE acts as an enhancer of Ifnb1 promoter activity, but only after viral infection 22 . In contrast, in macrophages, ICE acts as a repressor of Ifnb1 promoter activity during LPS activation. Furthermore, the high and sustained expression of Ifnb1 in LPS-or poly(I:C)activated RAW cells where both alleles of ICE are deleted using lentiviral delivery of Cas9 and gRNA indicates that ICE has a repressor activity when these cells are stimulated. Although the lentiviral system used might increase the frequency of off-target effects 30,31 , the similar results obtained with three different clones and with different gRNAs make off-target effects less likely. Use of recombinant Cas9 protein 32 or highly efficient transient transfection of RAW cells will reduce off-target effects and definitively show that ICE is a unique element that acts as an enhancer in fibroblasts, but displays promoter marks and acts as a repressor at late stages of LPS activation of macrophages. These opposite roles of ICE in fibroblasts and macrophages might be linked to recruitment of distinct proteins in the two cell types. Indeed, PU.1, which is not expressed in fibroblasts, is bound to ICE in macrophages and might be involved in the cell-specific function of ICE. Ectopic PU.1 expression in TRIM33 þ and TRIM33 À NIH3T3 cells indicates that TRIM33-mediated repression of Ifnb1 is dependent on PU.1 expression. However, in fibroblasts activated with poly(I:C), PU.1 expression associated with TRIM33 deficiency does not result in the sustained Ifnb1 expression observed in poly(I:C)-activated Trim33 À / À BMDM. This result suggests that factors present in BMDM, but not in fibroblasts, are required for Ifnb1 repression during the late stages of BMDM activation. Amongst haematopoietic-restricted PU.1 partners, IRF8 and IRF3 that regulate Ifnb1 activation in BMDM 6,33 are good candidates to cooperate with TRIM33 for Ifnb1 repression during the late stages of BMDM activation.
Several TRIM proteins regulate production and action of IFN-b during innate immune response. TRIM6 is a positive regulator of IFN-b and is critical for IFN-b-mediated antiviral response 34 , TRIM25 is essential for RIG-I-mediated-type I IFN induction 35 , TRIM21 regulates IRF3-dependent IFN-b expression 36 and TRIM56 modulates STING and subsequent IFN-b induction 37 . Here, we show the implication of TRIM33 in the transcriptional regulation of Ifnb1. Interestingly, TRIM33 deficiency did not alter the initial activation of Ifnb1 transcription after LPS activation of macrophages but impaired repression of Ifnb1 expression during the late stages of this activation. Together with the continuous binding of TRIM33 on ICE, the complementation of the Ifnb1 phenotype in macrophages by reexpression of TRIM33 in Trim33 À / À immortalized macrophages suggested a direct role of TRIM33 in Ifnb1 transcription and indicated that the activity and/or substrate availability of TRIM33 regulated Ifnb1 expression at the end of LPS activation of macrophages.
Two lines of evidence suggest that, during the late stages of LPS activation of BMDM, TRIM33 represses Ifnb1 expression through modulation of CBP/p300 recruitment and/or activity at the Ifnb1 promoter. First, TRIM33 inactivation led to recruitment of CBP/p300 at the Ifnb1 promoter and enhanced H3ac. Second, CBP/p300 activity is critical for enhanced Ifnb1 transcription at late stages as chemical inhibition of CBP/p300 enzymatic activity could restore the regulated expression of Ifnb1 gene in Trim33 À / À macrophages. As CBP/p300 acts as a switch to initially turn on Ifnb1 gene transcription 38 , these results indicated that TRIM33 might turn off Ifnb1 transcription during the late stages of LPS activation of macrophages by impairing the CBP/p300 recruitment and activity at the Ifnb1 promoter. This TRIM33-dependent recruitment and activity of CBP/p300 could be direct and/or mediated by trans-acting factors such as p65 that are, in Trim33 À / À BMDM, loaded onto the enhanceosome during the late phases of LPS activation. However, as PU.1 can interact with CBP/p300 (refs 39,40) and TRIM33, we propose a model where CBP/p300 and TRIM33 compete for PU.1 binding during LPS activation of BMDM. During the initial stages of LPS activation, CBP/p300 recruited by the enhanceosome might activate Ifnb1 transcription, whereas during the late stages of LPS activation, TRIM33 might interact with PU.1, impair PU.1/CBP/ p300 interaction and shut down Ifnb1 transcription (Fig. 4i).
Numerous studies have shown the complexity of Ifnb1 gene transcriptional regulation. Our data describe a new regulatory layer through identification of cis-and trans-acting elements that regulate Ifnb1 gene transcription at specific steps of macrophage activation, but not in fibroblasts. These findings not only provide Cell culture. For BMDM generation, mouse bone marrow (BM) was flushed out of the tibiae, femora and humeri using a syringe with PBS, filtered through a 70-mm mesh filter (Becton Dickinson) to remove debris and pelleted by centrifugation. Adherent cells were removed by incubating BM cells on culture dishes for 3 h in IMDM supplemented with 10% foetal calf serum (FCS), 1% penicillin/streptomycin (PS; Life Technologies) and 10 mM thioglycerol (Sigma Aldrich), then non-adherent cells were cultured for 7 days on culture dishes in the same medium supplemented with 25 ng ml À 1 mouse CSF1 (Miltenyi Biotec). On day 7, BMDM were activated with LPS or poly(I:C) in IMDM supplemented with 2.5% FCS and 1% PS. For PMs, cells were collected with cold PBS and cultured for 2 h in RPMI 1640 medium (Life Technologies), supplemented with 10% FCS. All adherent cells expressed F4/80 and were considered as PMs. PM were activated with LPS or poly(I:C) in RPMI 1640 medium supplemented with 2.5% FCS and 1% PS. WT and Trim33 À / À iM were generated as previously described 21 . Briefly, BM cells isolated from WT and Trim33 À / À mice were cultured in IMDM supplemented with 10% FCS and 50 ng ml À 1 mouse CSF1 (Peprotech). After overnight culture, the cells were transduced with a lentiviral vector expressing the SV40 large T antigen (SFFV.Tag LV). Cells were then cultured in a chemically defined medium (SFM-macrophage medium, Life Technologies) supplemented with CSF1 (50 ng ml À 1 ). Single clones were isolated that expressed F4/80, CD11b but not Gr1. The resulting clones, termed iM, were maintained in IMDM supplemented with 20% foetal bovine serum, 1% PS and 50 ng ml À 1 CSF1. iM were activated with LPS in IMDM supplemented with 2.5% FCS and 1% PS. For cell lines, RAW 264.7 and NIH3T3 cells were grown in DMEM (Life Technologies) supplemented with 10% FCS and 1% PS. RAW 264.7 cells were activated with LPS in DMEM supplemented with 1% FCS. Unless otherwise indicated, LPS (Sigma Aldrich) was used at 100 ng ml À 1 , poly(I:C) at 30 mg ml À 1 , TSA (Sigma Aldrich) at 10 nM, and C646 (Merck Millipore) at 10 mM. NIH3T3 were transfected with 10 mg ml À 1 poly(I:C) (Sigma Aldrich).
Chromatin Immunoprecipitation (ChIP). Cells were fixed with 1% formaldehyde for 10 min at 37°C, lysed in SDS lysis buffer (50 mM Tris pH8, 10 mM EDTA, 1% SDS, protease inhibitor cocktail (Roche)) and sonicated. Supernatant was diluted 10 times in IP dilution buffer (16.7 mM Tris pH8, 167 mM NaCl, 1.2 mM EDTA, 1.1% TritonX-100, 0.01% SDS, protease inhibitor cocktail (Roche)) and immunoprecipitations were carried out overnight with specific antibodies. Immunoprecipitated chromatin was collected using Protein A Agarose/Salmon Sperm DNA beads (Millipore) and, after washing and elution, reverse cross-linking was carried out with 0.2 M NaCl at 65°C overnight. The chromatin was then digested by 20 mg of Proteinase K (Invitrogen) for 1 h at 45°C and isolated by phenol-chloroform extraction. PCR reactions were performed using SYBR Green PCR Master Mix (Applied Biosystems) and specific primers sequences are listed below.
For sequencing, 10 ng of purified DNA from ChIP was adapter-ligated, PCR amplified and sequenced on the Illumina Genome Analyzer IIx as single-end 50 base reads following Illumina's instructions. Sequence reads were mapped to reference genome mm9/NCBI37 using Bowtie v0.12.7.
3C-seq. 3C-seq was performed as previously described 41 , using EcoRI as first cutter and DpnII as second cutter. Briefly, cells were fixed with 2% formaldehyde for 10 min at room temperature, and nuclei were prepared by lysing cells in 10 mM Tris pH8, 10 mM NaCl, 0.2% NP-40. Nuclei were lyzed in 0.5 ml 1 Â EcoRI buffer supplemented with 0.3% SDS for 1 h at 37°C. Triton X-100 was added at 2% final for another 1 h at 37°C. DNA was then digested with 400 units of EcoRI overnight at 37°C. Samples were incubated for 25 min at 65°C with 1.6% SDS, diluted to 7 ml with ligase buffer, supplemented with 1% Triton X-100 and incubated at 37°C for 1 h. Ligation was performed for 4 h at 16°C with 100 u T4 DNA ligase, then 300 mg proteinase K were added and samples were decrosslinked overnight at 65°C. RNase (300 mg) was added for 45 min at 37°C, and samples were extracted with phenol/chloroform/isoamyl alcohol (25:24:1; PCI). After precipitation, samples were digested with DpnII, extracted with PCI and precipitated, and ligated in 15 ml ligation buffer at 16°C for 4 h before another PCI extraction and precipitation. Samples were amplified by PCR with the primers listed below, and PCR products were purified on QIAprep columns (Qiagen). The resulting libraries were sequenced on a Illumina HiSeq 2000 as single-end 100 bp reads. After trimming, reads were aligned on the mm9 version of the mouse genome using Bowtie.
Results are presented as the number of reads per million reads per EcoRI fragment, in the middle of each fragment.
Assay for IFN-b. WT and Trim33 À / À BMDM supernatants were collected after LPS stimulation. ELISA was performed with LegendMax Mouse IFN-b kit (Biolegend), according to manufacturer's instructions.
Quantitative RT-PCR. Total RNA was extracted with the RNeasy Plus Micro kit (Qiagen) and reverse transcribed with random primers and Superscript III (Life Technologies). Quantitative PCR was performed using the Power SYBR green PCR master mix (Applied Biosystems) in the 7900HT Fast Real-Time or the StepOne PCR Systems (Applied Biosystems). Primer sequences are listed below.
Lentiviral vectors for rescue and knockdown. For rescue experiments in Trim33 À / À iM, the lentiviral vector was constructed by inserting full-length flag-TRIM33 from the pSG5-flag-TRIM33 (ref. 15) into the pTRIP/DU3-MND-IRES-GFP vector, downstream the MND promoter, using the Gateway technology (Life Technologies). Deletions of the coiled-coil domain (aa 321-457), the ubiquitin ligase activity (aa 141-145) and the SMAD-interaction (aa 555-876) domain were generated by PCR. The pTRIP/DU3-MND-GFP was used in control experiments. NIH3T3 PU1 þ cells were obtained by transducing a lentiviral vector pTRIP/DU3-MND-PU.1-IRES-GFP. For TRIM33 knocking-down in NIH3T3 or NIH3T3 PU.1 þ , the sequence of the shRNA targeting mouse Trim33 was obtained from Sigma Aldrich (5 0 -CCGGCGGACTTAAATCGGTTGTTAACTCGAGTTAA CAACCGATTTAAGTCCGTTTTTG-3 0 ) and cloned in the pSuper vector (OligoEngine) downstream the H1 promoter. H1-shRNATRIM33 fragment was then cloned in the pTRIP/DU3-EF1a-GFP, where the GFP coding sequence is under the control of the EF1a promoter. All the constructs were verified by sequencing. Lentiviral vector production and cells transduction were carried out as described 42 . GFP-positive cells were sorted and used in functional assays.
Luciferase assays. A 391 bp DNA fragment of murine Ifnb1 promoter was obtained by PCR on genomic DNA and cloned into the pGL3 Basic vector (Promega) using KpnI and XhoI sites. A 408 bp DNA fragment spanning the À 15 kb PU.1/TRIM33-binding site was obtained by PCR of genomic DNA and cloned upstream of the 391 bp DNA fragment of murine Ifnb1 promoter into the pGL3 Basic vector. After sequencing of the different constructs, RAW 264.7 cells were transfected using JetPEI (PolyPlus-transfection) with the indicated constructs and the TK-renilla reporter vector (Promega), as an internal control for normalization. Twenty-four hours after transfection, cells were activated with LPS (100 ng ml À 1 ) and luciferase activity was determined using the dual-luciferase reporter assay (Promega). Firefly luciferase activity was normalized with respect to renilla luciferase activity.
All the clones were verified by sequencing of a 2 kb genomic region containing the ICE.