Epigenetic regulator CXXC5 recruits DNA demethylase Tet2 to regulate TLR7/9-elicited IFN response in pDCs

By detecting the presence of foreign nucleic acids, plasmacytoid DCs (pDCs) play crucial roles in the induction of both innate and adaptive immune responses (Reizis et al., 2011a; Lewis and Reizis, 2012; Merad et al., 2013; Karrich et al., 2014; Mildner and Jung, 2014; Swiecki and Colonna, 2015). Despite their rare presence in the lymphoid organs and circulating blood, pDCs are superior IFNα producers and immediate responders to viral infection (Reizis et al., 2011b; Ng et al., 2013; Swiecki and Colonna, 2015; Webster et al., 2016). After pDC depletion, the early IFN response to both DNA viruses (HSV, mouse hepatitis virus, and murine CMV) and RNA viruses (vesicular stomatitis virus [VSV] and respiratory syncytial virus) was severely impaired (Lund et al., 2006; Smit et al., 2006; Swiecki et al., 2010, 2013; Takagi et al., 2011; Cervantes-Barragan et al., 2012). Moreover, the pDC-induced IFN response has been shown to be critically involved in the induction of T cell response required for controlling chronic infection by lymphocytic choriomeningitis virus (Blasius et al., 2012; Cervantes-Barragan et al., 2012). In this regard, pDC depletion or dysfunction has been linked to a variety of human chronic infections, especially the infections caused by HIV (Chehimi et al., 2002; Li et al., 2014; Zhang et al., 2015b), hepatitis B virus (Duan et al., 2004), and hepatitis C virus (Decalf et al., 2007). In addition to viral nucleic acids, pDCs can recognize self-DNAs/RNAs released by necrotic cells, and thus have been implicated in the pathogenesis of autoimmune diseases including systemic lupus erythematosus and psoriasis (Lande et al., 2007; Ganguly et al., 2009, 2013; Rowland et al., 2014; Sisirak et al., 2014). However, how pDCs are genetically and epigenetically programmed for IFN response remains incompletely understood.

Compared with other immune cells, the most remarkable feature of pDCs rests in their unique expression of endosomal sensors TLR7/9 and massive IFN-producing capacity. Unlike TLR3/4 signals, which depend on transcriptional factor IRF3 to induce IFN response, TLR7/9 signals activate IRF7, as well as transcriptional factors NFκB and AP-1, to elicit IFNα/β production (Kawai et al., 2004). Interestingly, it seems that IRF7 has a more fundamental role than IRF3 in mounting IFN response to a broad spectrum of viruses (Honda et al., 2005), and IRF7 deficiency can lead to recurrent influenza infection in humans (Ciancanelli et al., 2015). Although most cells express IRF7 only after viral infection through feedback IFNα/β signaling, pDCs are capable of expressing IRF7 in the steady state (Sato et al., 2000; Izaguirre et al., 2003; Ning et al., 2011). Because both pDCs and IRF7 play such a pivotal role in IFN response, how pDCs manage to express IRF7 constitutively needs to be precisely defined. Indeed, some recent findings have begun to shed light on this issue. Transcriptional factors E2-2 and STAT3 not only are essential for pDC development and maintenance, but also are critically involved in IRF7 expression in the steady state (Laouar et al., 2003; Ghosh et al., 2010; Li et al., 2012). Conversely, DNA methylation of CpG-containing island (CGI) encompassing Irf7 promoter has been shown to be involved in silencing Irf7 expression in human fibroblasts (Lu et al., 2000). However, whether this layer of epigenetic regulation also operates in immune cells remains uncharacterized.

CpG islands enriched for CpG dinucleotides are present in ∼70% of mammalian gene promoters. Not surprisingly, CGIs can profoundly shape chromatin structure, thereby participating in the regulation of gene expression. Although CpG islands are frequently hypomethylated or unmethylated in somatic cells, and mostly associated with active gene transcription, mounting evidence suggests that CGIs can become hypermethylated in certain immune cells, and more frequently in malignant cells, leading to gene silencing instead (Blackledge and Klose, 2011; Li and Zhang, 2014; Smale et al., 2014; Ko et al., 2015). Recent identification and elucidation of zinc finger CXXC proteins (ZF-CXXC proteins) have provided better understanding of the dynamic nature of CGIs and their intricate link with histone modifications (Long et al., 2013). Despite their conserved function in recognizing unmethylated CpGs, ZF-CXXC proteins are very much diversified structurally and functionally. Although DNA methyltransferase Dnmt1 and DNA dioxygenase Tet1/3 are involved in DNA methylation (Gu et al., 2011; Li and Zhang, 2014), histone methyltransferases Mll1/2, histone demethylases Kdm2a/b, and Cfp1 mainly participate in the regulation of histone methylation (Farcas et al., 2012; Blackledge et al., 2013). Conversely, CXXC5 and its close homologue CXXC4 do not possess any enzymatic activity, and their physiological functions have just begun to unfold (Long et al., 2013). Recent findings imply that CXXC5 can be involved in a variety of physiological processes including myelopoiesis, osteopoiesis, and neural cell development (Pendino et al., 2009; Kim et al., 2015, 2016) and also takes part in DNA damage response, skin repair, and hematological malignancy (Pendino et al., 2009; Zhang et al., 2009; Lee et al., 2015). In these processes, CXXC5 primarily acts as a signaling component regulating ATM–p53 or wnt–β-catenin pathways (Andersson et al., 2009; Zhang et al., 2009; Kim et al., 2010, 2015; Lee et al., 2015). Recently, CXXC5 has been reported to be associated with epigenetic regulators Tet2 and SUV39H1 (Ko et al., 2013; Tsuchiya et al., 2016), implying a possible role for CXXC5 in epigenetic regulation of gene expression. In this study, we provide compelling evidence supporting a crucial role for CXXC5 in the regulation of DNA methylation and IFN response in pDCs, thereby demonstrating an additional layer of regulation integral to pDC function and antiviral defense.

In the past decade, resources generated by immune cell gene profiling have led to the identification of lineage-specific regulators and the elucidation of principles governing immune cell specification. By searching these databases, we found that the putative epigenetic regulator CXXC5 is preferentially expressed in human and mouse pDCs (Robbins et al., 2008; Ghosh et al., 2010; Miller et al., 2012). Moreover, CXXC5, along with MAVS/IPS-1/VISA, has been previously linked to IFN response through a cDNA library screening (Kawai et al., 2005). These clues prompted us to investigate the role of CXXC5 in pDCs.

By quantitative real-time PCR (qPCR) and Western blotting, we confirmed that CXXC5 is expressed predominantly in pDCs, much less in lymphocytes, and barely detectably in conventional DCs (cDCs) or macrophages (Fig. 1, A and B). Next, we generated CXXC5-deficient mice by conventional gene-targeting strategy on 129/Ola strain (Fig. S1, A and B), and then backcrossed onto C57/B6 background for six generations. By Western blotting, we confirmed that CXXC5 expression was completely lost in various organs of CXXC5^−/− mice, especially in the spleen and thymus (Fig. S1 C). By flow cytometry analyses, we found that the percentage and number of pDCs and cDCs in CXXC5^−/− spleens were comparable to those in WT controls (Fig. 1 C). Moreover, B and T lymphocytes, neutrophils, and monocytes were normally generated in CXXC5^−/− mice (Fig. S1, D and E).

Because CXXC5 does not appear to affect pDC development, we went on to investigate its involvement in pDC function, especially in antiviral IFN response. We first isolated pDCs from WT and CXXC5^−/− spleens by FACS and stimulated them ex vivo with HSV-1, a DNA virus known to engage TLR9 on pDCs, for 24 h. By ELISA, we detected large amounts of IFNα and IFNβ secreted by WT pDCs after HSV-1 stimulation; however, the induction of IFNα/β by HSV-1 was diminished in CXXC5^−/− pDCs (Fig. 1 D). Next, we infected both WT and CXXC5^−/− mice with HSV-1 i.v. for 9 h and then purified splenic pDCs by FACS sorting (Fig. S1 F). Real-time PCR analyses demonstrated that the induction of Ifnα4 and Ifnβ mRNAs by in vivo HSV-1 infection was severely impaired in CXXC5^−/− pDCs (Fig. 1 E). Furthermore, after i.v. injection of A-type CpG DNAs (CpG-A, mainly engages TLR9 on pDCs) complexed with liposome DOTAP, CXXC5^−/− mice also exhibited attenuated IFNα response (Fig. 1 F). Together, these results uncovered a novel role for CXXC5 in pDC-elicited IFN response to viral infection.

pDCs are rarely present in vivo and account for <0.5% of total spleen cells. Alternatively, in vitro pDCs generated from BM by cytokine Flt3L closely resemble in vivo pDCs and have been widely used for studies on IFN response. Subsequently, we used in vitro pDCs (CD11c^intB220⁺CD11b⁻) sorted from Flt3L-differentiated BMDCs (Flt3L-BMDCs) by FACS for the following experiments. After stimulation by HSV-1 for 24 h, Flt3L-pDCs recapitulated splenic pDCs with abundant IFNα/β production (Fig. 2 A). Similarly, CXXC5^−/− pDCs produced much less IFNα/β than WT pDCs after stimulation by HSV-1 or CpG-A (Fig. 2, A and B). Concomitantly, the induction of chemokine CCL-5, as well as cytokines TNF and IL-12, was also dampened in CXXC5^−/− pDCs (Fig. 2, A and B).

Although TLR9 is involved in the recognition of viral DNAs in pDCs, TLR7 is a designated sensor for viral single-stranded RNAs. After stimulation by RNA virus VSV, the induction of IFNα/β, CCL-5, TNF, and IL-12 was universally attenuated in CXXC5^−/− pDCs (Fig. 2 C). Moreover, CXXC5^−/− pDCs also produced less IFNα/β, CCL-5, TNF, and IL-12 than WT pDCs after stimulation by TLR7 ligand R848 (Fig. 2 D). Subsequently, we sought to test whether restoring CXXC5 expression in CXXC5^−/− pDCs could rescue the defective IFN response. To this end, we reexpressed CXXC5 in CXXC5^−/− pDCs by lentiviral vector–mediated transduction. Notably, restored CXXC5 expression largely rescued the defective IFN response exhibited by CXXC5^−/− pDCs, as they now produced levels of Ifnα4, Ifnβ, and Tnf comparable with WT pDCs in response to CpG-A stimulation (Fig. 2 E). Collectively, these results demonstrated that CXXC5 has a crucial role in TLR7/9-induced IFN response.

Unlike CpG-A, which specifically operates in pDCs, B-type CpG DNAs (CpG-B) can trigger a TLR9-dependent proinflammatory response in a diversity of immune cells including pDCs, cDCs, macrophages, and B cells. After CpG-B stimulation, CXXC5^−/− pDCs produced less TNF and IL-6 than WT pDCs (Fig. 3 A). Moreover, CpG-B–induced TNF and IL-6 were also decreased in CXXC5^−/− B cells (Fig. 3 B), further supporting a positive role for CXXC5 in TLR9 signaling. However, in CpG-B–stimulated BMDCs (GM-CSF–differentiated BMDCs) and BMDMs, CXXC5 deficiency did not affect the induction of TNF or IL-6 (Fig. 3, C and D), underscoring a cell type–specific role for CXXC5 in TLR9 signaling. Similarly, although CXXC5 deficiency resulted in attenuated TNF and IL-6 expression in B cells upon LPS stimulation (Fig. 3 E), LPS-induced IFN type I and proinflammatory responses were largely intact in CXXC5^−/− BMDCs (Fig. 3 F). Therefore, although CXXC5 is involved in a broad range of TLR signals featured by TLR4, TLR7, and TLR9, it only operates in a narrow spectrum of immune cells, particularly in pDCs and B cells that express large amounts of CXXC5 proteins (Fig. 1 B).

To determine how CXXC5 regulates the TLR9-induced proinflammatory response, we examined CpG-B–induced signaling events. Surprisingly, similar levels of p-IKKα/β, p-IκBα, p-p38, and p-ERK were detected in WT and CXXC5^−/− pDCs after CpG-B stimulation (Fig. 3 G), indicating that CXXC5 does not seem directly affect the activation of NFκB or MAPKs in TLR9 signaling. On the other hand, we found that CXXC5 deficiency influenced the recruitment of p65 NFκB to proinflammatory genes (Fig. 3, H and I). First, by chromatin immunoprecipitation (ChIP)–qPCR analyses, we detected attenuated p65 binding to the promoters of several proinflammatory genes, such as Il6, Cxcl2, and Ccl5, in CXXC5^−/− pDCs after CpG-B stimulation (Fig. 3 H). Second, the enrichment of p65 to the promoters of Tnf and IL-6 was also diminished in CXXC5^−/− B cells stimulated by either CpG-B or LPS (Fig. 3 I). Together, these data implied that CXXC5 might impinge on NFκB recruitment to influence proinflammatory gene expression.

In contrast to CpG-B, CpG-A stimulation leads to the activation of IRF7 in pDCs through IKKα (Hoshino et al., 2006) and mTOR (Cao et al., 2008), which then collaborate with IKKβ and MAPKs to induce an IFN response. By Western blotting, we found that CpG-A–induced phosphorylation and degradation of IκBα were comparable between WT and CXXC5^−/− pDCs (Fig. 4 A). Further, the phosphorylation of p38, mTOR, and S6 was similarly induced by CpG-A in both WT and CXXC5^−/− pDCs (Fig. 4 A). Overall, CXXC5 deficiency did not compromise CpG-A–induced activation of NFκB, MAPKs, and mTOR in pDCs. In contrast, we found that IRF7, but not IRF8, proteins were decreased in CXXC5^−/− pDCs that were either untreated or treated with CpG-A (Fig. 4 B). Consistently, the phosphorylation of STAT1 was also impaired in CXXC5^−/− pDCs (Fig. 4 B), likely because of decreased IFN response in these cells (Fig. 2 B). Intriguingly, the levels of CXXC5 protein and mRNA gradually declined after CpG-A stimulation (Fig. 4, B and C), a phenomenon also observed in human pDCs (Fig. 4 D). Moreover, Cxxc5 expression was also down-regulated by IFNβ stimulation (Fig. 4 C), implying feedback regulation on CXXC5 expression in pDCs. Given the down-regulation of E2-2 upon pDC activation (Ghosh et al., 2010), it is tempting to propose that these feedback mechanisms may be required to control the duration of TLR9 signaling in pDCs.

Next, we assessed the activation of IRF7 by CpG-A signal through immunofluorescence and ChIP analyses. Although IRF7 primarily resided in the cytoplasm of untreated pDCs, activated IRF7 was translocated to the nuclei after CpG-A stimulation (Fig. 4 E). Notably, compared with WT pDCs, fewer nuclear IRF7 proteins were induced by CpG-A in CXXC5^−/− pDCs (Fig. 4 E). Furthermore, IRF7 binding to the promoters of Ifna4 and Ifnβ was also lessened in CpG-A–treated CXXC5^−/− pDCs (Fig. 4 F). Concomitantly, epigenetic markers associated with active transcription (such as H3K4me3, H3K27ac, and H4ac) were decreased on Ifnα4 promoter of CXXC5^−/− pDCs (Fig. 4 G). Therefore, CXXC5 appears to play a role in the recruitment of IRF7 to Ifnα4/β genes.

Next, we conducted global gene expression profiling to gain comprehensive understanding of CXXC5-regulated genes in both steady-state and activated pDCs (Fig. 5 A and Table S1). Notably, among CpG-A–induced genes that were down-regulated by CXXC5 deficiency (>1.5-fold), Ifnα/β/λ members were highly enriched (Fig. 5 A). Remarkably, another group of genes that were down-regulated in both steady-state and CpG-A-activated CXXC5^−/− pDCs were CGI-containing genes, such as Irf7, Fos, Cxcl2, and Tnf (Fig. 5 A). The third group is composed of genes encoding cytokines and chemokines whose expression was neither driven by IRF7 nor regulated by CGI (Fig. 5 A). These data suggested an interesting scenario, that CXXC5 may be primarily involved in the regulation of CGI-containing genes, whose products (such as IRF7) then contribute to rapid and robust IFN response in pDCs. To this end, we validated that Irf7, along with other CGI-containing genes Fos, Cxcl2, and Tnfα, were indeed down-regulated in steady-state CXXC5^−/− pDCs; however, some non-CGI genes such as Tlr7/9 and Tcf4 (which encodes E2-2) were normally expressed in these cells (Fig. 5 B). By Western blotting, we further confirmed that the expression of IRF7 was attenuated in steady-state CXXC5^−/− pDCs, whereas other regulators of pDC function, such as IRF5, IRF8, and STAT3, were unaffected (Fig. 5 C). Notably, the major regulators of IFNα/β signaling, such as IFNαR1/2, STAT1/2, and IRF9, were expressed at comparable levels in both WT and CXXC5^−/− pDCs (Fig. 5 D).

Next, we sought to examine whether CXXC5 may exert its effect on CGI-containing genes, in particularly Irf7, through an epigenetic mechanism. Although CXXC5 has been shown to reside in either the cytoplasm or the nucleus, varying with cell types, we found that CXXC5 proteins were predominantly presented in the nuclei of pDCs (Fig. 6 A). By ChIP-qPCR analyses, we observed specific enrichment of CXXC5 within the CpG island located in the first intron of Irf7 gene (500–1,000; Fig. 6 B). Moreover, bisulfate sequencing revealed that this CpG-rich region (500–1,000) was hypomethylated in the pDCs (16.25%) but hypermethylated in the cDCs (57.5%; Fig. 6, C and D). Remarkably, ablation of CXXC5 led to elevated DNA methylation in pDCs (36.25%) but not in cDCs (52.5%; Fig. 6, C and D). Conversely, the hydroxylation of 5-methyl-cytosine (5hmC), the first of three steps leading to 5mC demethylation (Xu et al., 2011), was lessened not only in Irf7 CGI, but also in the CGIs of Cxcl2 and Tnfα in CXXC5^−/− pDCs (Fig. 6 E). These results therefore suggested a novel role for CXXC5 in regulation of DNA methylation in pDCs.

Because CGI methylation can lead to the recruitment of chromatin remodeling complexes such as PRC2 and HDACs, we subsequently investigated whether CXXC5 is involved in the regulation of chromatin remodeling. By ChIP-qPCR, we found that Ezh2, the core component of PRC2, was highly enriched in the CGI of Irf7 in CXXC5^−/− pDCs, whereas the enrichment of Sin3a was unaltered (Fig. 6 F). Moreover, the presence of H3K27me3, a repressive marker primarily assembled by histone methyltransferase Ezh2, was markedly enhanced in Irf7 CGI of CXXC5^−/− pDCs (Fig. 6 G). Conversely, several active histone markers (such as H3K4me3, H3K36me3, H3K27ac, H3ac, and H4ac) were concomitantly decreased in the same CGI region of CXXC5^−/− pDCs (Fig. 6 G). As a result, polymerase II (pol II) binding to Irf7 promoter was impaired in CXXC5^−/− pDCs (Fig. 6 H). Collectively, these data implied that CXXC5 might be required for the maintenance of active chromatin structure on Irf7, possibly through regulating DNA methylation and histone modifications.

Because of a lack of enzymatic activity, it is conceivable that CXXC5 has to work with other partners to regulate DNA methylation. One of the candidate partners for CXXC5 is DNA demethylase Tet2 which, unlike Tet1 and Tet3, does not possess a CXXC domain (Xu et al., 2011, 2012; Long et al., 2013). To test this scenario, we conducted coimmunoprecipitation and identified CXXC5/Tet2 complexes in WT pDCs (Fig. 7 A). We then generated a series of mutants of CXXC5 and tested their interactions with Tet2 in HEK293T cells. Consistent with a previous studies (Ko et al., 2013), although WT CXXC5 colocalized with Tet2 in the nuclei, the CXXC domain–disrupted mutant of CXXC5 (C258A/C286A) failed to colocalize with Tet2 (Fig. S2 A). Moreover, the DNA binding domain–disrupted mutant of CXXC5 (TGHQ to AGAA) also lost the interaction with Tet2 (Fig. S2 A). Notably, ChIP-qPCR analyses revealed that Tet2 was also predominantly enriched in the 500–1,000 region within the CGI of Irf7 gene, a process dependent on the expression of CXXC5 in pDCs (Fig. 7 B).

We then tested whether Tet2 is involved in the regulation of CGI methylation in pDCs. By bisulfate sequencing, we detected elevated DNA methylation in Irf7 CGI of Tet2^−/− pDCs (Fig. 7 C), a phenomenon similarly observed in CXXC5^−/− pDCs (Fig. 6 C). Moreover, histone markers H3K36me3 and H3K4me3 were less enriched in Irf7 CGI of Tet2^−/− pDCs (Fig. 7 D). However, unlike CXXC5, Tet2 did not seem to have an impact on the enrichment of H3K27me3 or Ezh2 in this region (Fig. 7, D and E). Consistent with these epigenetic changes, Tet2^−/− pDCs expressed much less IRF7 than WT controls (Fig. 7 F). Interestingly, overexpression of WT, but not Tet2 binding–deficient mutants of CXXC5, resulted in constitutive Irf7 expression in cDCs (Fig. S2 B). Consequently, only WT, but not those mutants of CXXC5, was able to elevate Ifnα4 and Ifnβ expression in cDCs after LPS stimulation (Fig. S2 C). By ELISA, we were able to detect greater IFNβ production in CXXC5-overexpressing cDCs upon LPS stimulation, although IFNα remained undetectable (Fig. S2 D).

Besides Irf7, we observed decreased 5hmCs in CGIs of Cxcl2 and Tnfα in Tet2^−/− pDCs (Fig. 7 G). Moreover, pol II binding to these promoters was attenuated in Tet2^−/− pDCs (Fig. 7 H). Consistently, the transcription of Irf7, Cxcl2, and Tnfα was also impaired in Tet2^−/− pDCs (Fig. S2 E). These results strongly imply that CXXC5-Tet2 axis has a critical role in maintaining the hypomethylation states of a subset of CpG islands in pDCs.

Given that CXXC5 plays a critical role in TLR7/9-induced responses in pDCs, we went on to test how this layer of regulation may contribute to host defense against viral infection in vivo. After i.v. infection by HSV-1, WT mice exhibited robust IFN response, secreting large amounts of IFNα/β and CCL-5 at the 6-h time point, whereas this early IFN response was severely impaired in CXXC5^−/− mice (Fig. 8 A). However, in both WT and CXXC5^−/− mice, the early proinflammatory responses, such as TNF and IL-12, were similarly induced after HSV-1 infection (Fig. 8 A). Moreover, the IFN response remained lower in CXXC5^−/− mice until 24 h after HSV-1 infection (Fig. S3 A). At that time, CXXC5^−/− mice exhibited ∼2-log more HSV-1 viruses than WT controls (Fig. 8 B) and suffered much more severe morbidity (Fig. S3 B). By day 8, nearly 90% of CXXC5^−/− mice died of infection, compared with only 25% of WT mice (Fig. 8 C). Importantly, adoptively transferred WT, but not CXXC5^−/−, pDCs were able to alleviate the mortality of CXXC5^−/− mice caused by HSV-1 infection (Fig. 8 C), suggesting that the involvement of CXXC5 in pDCs might have partially contributed to host defense against HSV-1.

In addition, we found that CXXC5^−/− mice were also highly susceptible to VSV infection. Upon i.v. infection of VSV, CXXC5^−/− mice were unable to mount efficient IFN responses at the early stage of infection, exhibiting attenuated IFNα/β and CCL-5 production compared with WT mice at 6- and 24-h time points (Fig. 8 D and Fig. S3 C). Interestingly, after VSV infection, CXXC5^−/− mice also produced fewer IgM antibodies than WT mice on day 3, although IgG2 titers were undetectable by that time (Fig. 8 E). Furthermore, CXXC5^−/− mice suffered much greater weight loss than WT mice (Fig. S3 D), and eventually all died by day 5 (Fig. 8 F). In contrast, ∼60% of the WT mice survived VSV infection (Fig. 8 F). Together, these results imply an indispensable role for CXXC5 in host defense against viral infection.

Considering that CXXC5 and Tet2 work in concert to regulate DNA methylation and gene expression in pDCs, we further investigated whether Tet2 has a similar role to CXXC5 in IFN response and antiviral defense. Indeed, upon stimulation by CpG-A, Tet2^−/− splenic pDCs produced less IFNα and IFNβ than WT controls (Fig. 9 A). After infection by HSV-1 and VSV, Tet2^−/− mice on mixed 129S/B6 background suffered more severe morbidity and mortality than WT mice (Fig. 9, B and C). These results suggest that Tet2 is also involved in the regulation of pDC function and antiviral defense.

We wondered whether CXXC5 might function similarly in human pDCs, and therefore used human pDC line Gen2.2 (Chaperot et al., 2006; Di Domizio et al., 2009) to test this scenario. We infected Gen2.2 cells with lentiviral vectors expressing scrambled or Cxxc5-targeted shRNAs, and then selected positive cells by GFP for further study. Interestingly, knockdown of CXXC5 led to reduced Irf7 expression in Gen 2.2 cells (Fig. 10 A). Further, in Cxxc5-knockdown cells, HSV-1– and R848-induced Ifnα and Ifnβ mRNAs were also impaired (Fig. 10 A). In contrast, knockdown of Tet2 also decreased Irf7 expression and led to compromised IFN response in Gen2.2 cells (Fig. 10 B). Therefore, the CXXC5–TET2 axis might also operate in human pDCs, possibly through the same mechanism implicated in mouse pDCs (Fig. 10 C).

Despite sharing conserved molecular signatures with cDCs and B cells, pDCs are generally viewed as a key component of the innate immune system, contributing to the immediate IFN response upon viral infection. Although the signaling pathways by which TLR7/9 induces a robust IFN response have been largely defined, the involvement of epigenetic regulation in this process has yet to be established. In this study, we report that CXXC5 has a crucial role in TLR7/9-induced IFN response, and propose that DNA methylation might act as an epigenetic mechanism licensing pDC for antiviral defense.

Although CXXC5 has been previously linked to hematopoiesis and hematological malignancy (Pendino et al., 2009; Treppendahl et al., 2013), its role in the immune response has been largely unknown. Although CXXC4 is uniformly expressed in a diversity of immune cell subsets (Robbins et al., 2008; Miller et al., 2012), we and others (Robbins et al., 2008; Ghosh et al., 2010) found that CXXC5 is highly expressed in pDCs but barely detectable in cDCs or macrophages. The most notable feature distinguishing pDCs from cDCs and macrophages rests in their robust IFN-producing capacity. In this regard, we demonstrated here that ablating CXXC5 expression diminished IFN response in pDCs, whereas forcing CXXC5 expression in cDCs elevated their IFN response. Therefore, cell type–specific expression of CXXC5 seems to be one of the mechanisms underlying differential IFN-producing capacity between pDCs and cDCs. pDCs are one of the major cell populations contributing to immediate IFN response against viral infection. Indeed, we found that the early IFN response, but not proinflammatory response, was diminished in CXXC5^−/− mice after viral infection. Interestingly, CXXC5^−/− mice had more pronounced virus propagation at early stages and eventually succumbed to infection. Besides pDCs, CXXC5 is modestly expressed in B cells and T cells, and thus may also have functional roles in these cells. In line with this notion, our data demonstrated that CXXC5 is involved in the regulation of TLR response and antibody production in B cells. Conceivably, all these mechanisms executed by CXXC5 might have contributed to the marked morbidity and mortality exhibited by CXXC5^−/− mice after HSV and VSV infection. Nevertheless, adoptive transfer of WT pDCs to CXXC5^−/− mice was able to ameliorate the mortality caused by HSV-1 infection, indicating an integral role for pDC in host defense. However, further studies will be required to decipher the role of CXXC5 in other immune cells and how additional mechanisms may contribute to host defense against viral infection.

It is well known that pDCs uniquely express high levels of IRF7 at steady state, yet the underlying mechanism remains incompletely understood. Corroborating a previous study linking DNA methylation to Irf7 silence in human fibroblasts (Lu et al., 2000), we found that the CpG island of Irf7 is hypermethylated in cDCs as well, suggesting that CpG methylation may be broadly involved in restriction of IRF7 expression in a variety of cells, including nonimmune and immune cells. Conversely, in pDCs, Irf7 CGI is hypomethylated, a phenomenon that depends on the expression of both CXXC5 and Tet2. Mechanistically, our study suggests that CXXC5 may function as a “reader” for unmethylated CpGs, a scenario also implied in previous studies (Iurlaro et al., 2013; Spruijt et al., 2013), by recruiting DNA demethylase Tet2 to maintain hypomethylation of CGIs. In support of this hypothesis, Tet2^−/− mice and CXXC5^−/− mice exhibited similar defects with regard to pDC function and antiviral defense, despite their subtle difference in genetic background. However, Tet2 can also execute physiological functions through CXXC5-independent mechanisms. For example, Tet2 can be recruited to the genome through various transcriptional factors such as IκBζ (Zhang et al., 2015a), T-bet (Ichiyama et al., 2015), and Smad3/STAT5 (Yang et al., 2015), thereby participating in the induction of cytokines IL-6, IFNγ, and IL-10 in macrophages, cDCs, or T helper cells, respectively.

In addition to DNA methylation, we found that CXXC5 can also profoundly affect histone modifications in the CGI region. It has been demonstrated that CGI-containing inflammatory genes tend to adopt “loosened” or “opened” chromatin structures, thus being poised for constitutive expression or rapid induction (Ramirez-Carrozzi et al., 2009; Kaikkonen et al., 2013; Barozzi et al., 2014). In this regard, CXXC5-mediated DNA hypomethylation and histone modifications may facilitate chromatin remodeling, allowing the entry of transcriptional factor E2-2 and pol II, thereby ensuring steady-state Irf7 expression in pDCs. Potentially, this type of chromatin remodeling fueled by hypomethylated CGI may facilitate the recruitment of NFκB and pol II to a subset of proinflammatory genes, such as Tnfα and Cxcl2, as well (Fig. 10 C). Nevertheless, more studies are needed to precisely dissect the intricate interplay between DNA methylation and histone modifications in the context of gene regulation by CXXC5.

In addition to Irf7, CXXC5 is involved in the regulation of several other CGI-containing genes, such as Tnfα, Cxcl2, and Fos, in steady-state pDCs. Corroborating a previous study (Ramirez-Carrozzi et al., 2009), the methylation states of these CGIs, at least for Tnfα and Cxcl2, are well correlated with their induction by TLR signals. It is likely that CXXC5 regulates the expression of these CGI-containing genes in a direct manner, possibly via DNA demethylation. Subsequently, DNA hypomethylation may lead to active histone remodeling, thereby priming them for transcription. On the other hand, CXXC5 may exert its effect on non-CGI genes through indirect mechanisms. One possible scenario is that some CGI-containing genes encoding transcriptional factors, such as IRF7, IRF1, and Fos, could then contribute to the induction of some non-CGI genes in activated pDCs. Although IRF7 is a well-known master regulator of IFN response, Fos and IRF1 have been widely linked to proinflammatory gene expression as well. For example, IRF1 has an indispensable role in IL-12b expression (Salkowski et al., 1999), and Fos is involved in IL-10 induction by promoting the recruitment of NFκB to its promoter (He et al., 2015). Nevertheless, the role of CXXC5 in immune regulation has just begun to unfold, and further studies will be required to unravel the precise mechanisms by which CXXC5 regulates gene expression for antiviral defense.

In summary, our study not only uncovers a cell type–specific role for CXXC5 in TLR-induced IFN type I and proinflammatory responses, but also suggests DNA methylation as a novel epigenetic mechanism integral to pDC specification and antiviral defense. Given that human CXXC5 shows similar expression and function in pDCs, we anticipate that these findings will have further implications for viral infections and autoimmune diseases.

CXXC5-deficient mice were generated by deleting the second exon of Cxxc5 gene in a 129/Ola ES cell line as described in Fig. S1 A and then backcrossing onto C57BL/6 background for six generations. Tet2-deficient mice on mixed 129S/C57BL/6 background were generated previously (Hu et al., 2014). These mice were bred and maintained in a pathogen-free animal facility at Institut Pasteur of Shanghai. All procedures were conducted in compliance with a protocol approved by the Institutional Animal Care and Use Committee at Institut Pasteur of Shanghai.

Cxxc5 cDNA was amplified from C57BL/6 BM. The mutants (mutating C258/C286 to 258A/286A or TGHQ to AGAA) of CXXC5 were constructed by overlapping PCR approach. WT and mutants of Cxxc5 were cloned into pCDH vector by enzymes NheI and EcoRI or cloned into pIP-FLAG and pCMV vector by enzymes BamHI and EcoRI or EcoRI and BamHI, respectively.

CpG-A (ODN1585), CpG-B (ODN1826), CpG-A (ODN2216 for human), LPS, R837, and R848 were purchased from InvivoGen. Antibodies for Tet2 (#94580), H3K4me3 (#8580), H3K27ac (#4729), H3K27me3 (#6002), and Ezh2 (#3748) were from Abcam. Anti-CXXC5 (I-6513-I-AP) was from Proteintech, and anti-5hmC (#39769) was from Active Motif. Antibodies against p-IκBα (#2859), p-p38 (#9211), p-JNK (#4668), JNK (#9252), p-IKKα/β, p-STAT3 (#9145), STAT3 (#9139), p-mTOR (#2971), mTOR (#2983), IRF5 (#4950), p-S6 (#4858), S6 (#2317), p-STAT1 (#7649), and STAT1 (#9172) were purchased from Cell Signaling Technology. Antibodies against ERK (SC-93), p-ERK (SC-7383), p38 (SC-728), Sin3a (sc-994x), IκBα, Myc-tag, p65 (C-20), IRF8 (SC-6058), and GAPDH were from Santa Cruz Biotechnology, Inc. Anti-Flag M2 was from Sigma-Aldrich, and anti-ERK (H-72), anti-LaminB1 (332000), and anti-IRF7 (51–3300) were from Invitrogen.

BM cells were isolated from femurs and tibias of 6–8-wk-old C57BL/6 mice, and red blood cells were lysed using ACK lysis buffer (0.15 M NH₄Cl, 1 mM KHCIO₃, and 0.1 mM Na₂EDTA, pH 7.3). BM cells (10⁶/ml) were seeded on RPMI 1640 medium containing 10% FBS (HyClone) and 50 ng/ml Flt3L (R&D Systems) and cultured for 5–7 d. Every 2 d, half of the media were replaced by fresh RPMI containing 10% FBS and 50 ng/ml Flt3L. Before use, suspension cells were sorted by FACS (Aria II; BD), and CD11c^int/B220⁺/CD11b⁻ pDCs were collected.

Splenic pDCs were prepared as described previously (Vremec et al., 2000). In brief, spleens were minced and digested with collagenase D (1 mg/ml; Sigma-Aldrich) in RPMI 1640 at 37°C for 30 min. Digested cells were passed through a nylon mesh cell strainer (BD), and incubated with anti–PDCA-1 microbeads (Miltenyi Biotec). Subsequently, enriched pDCs were labeled with CD11c, mPDCA-1, B220, and Siglec-H for FACS sorting, and the CD11c^int/B220⁺/mPDCA-1⁺/Siglec-H⁺ population was harvested as splenic pDCs.

BMDCs and BMDMs were prepared by methods described previously (Deng et al., 2015). In brief, BM cells (10⁶/ml) were seeded on RPMI 1640 containing 10% FBS and 20 ng/ml GM-CSF (R&D Systems) and cultured for 7–9 d to generate BMDCs. BM cells (10⁶/ml) seeded on RPMI 1640 containing 10% FBS and 30% L929-conditioned medium were cultured for 5–7 d to generate BMDMs.

VSV-specific antibody titers were measured as described previously (Bach et al., 2007). In brief, 96-well plates were coated with 5 mg/ml PEG-precipitated VSV particles overnight, and then blocked with 5% BSA in PBS at RT for 2 h. Sera from uninfected and VSV-infected mice were serially diluted and incubated with precoated plates for 2 h. After extensive washing, biotin-labeled rat anti–mouse IgM, IgG2a, or IgG2b (BD) was added to detect three subtypes of antibody.

RNAs were extracted from cultured cells or tissues using TRIzol (Invitrogen) according to the manufacturer’s instruction. cDNAs were reverse transcribed from 0.5 µg total RNAs by PrimeScript RT-PCR kit (Takara Bio Inc.). Real-time PCR was performed with PrimeScript RT reagent kit (Takara Bio Inc.) on an ABI 7900HT Fast Real-time PCR System. Relative expression levels of target genes were quantitatively normalized against the expression of GAPDH using the ΔΔCT method. All real-time PCR primers used in this study are described in Table S2.

Total mRNAs (1 µg) were obtained from Flt3L-differentiated WT and CXXC5^−/− pDCs either untreated or treated with 2 µg/ml CpG-A for 3 h and analyzed by Agilent Technologies Gene Expression Hybridization kit according to the manufacturer’s protocols. Agilent Technologies Feature Extraction software (version 11.0.1.1) was used to analyze acquired array images. Quantitative normalization and subsequent data processing were performed with the GeneSpring GX v12.1 software package (Agilent Technologies). GEO accession no. is GSE79643.

Flt3L-differentiated BM DCs were either untreated or treated with CpG-A (2 µg/ml) for 2 h before cross-linking, and FACS-sorted cells were then used for immunoprecipitation. Cells were cross-linked with 1% methanol-free formaldehyde for 10 min at room temperature. Cross-linking was quenched by 125 mM glycine for 5 min, and cell pellets were washed twice with PBS. Subsequently, cells were lysed in sonication buffer (10 mM Tris-HCl, pH 8.0, 100 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 0.1% sodium deoxycholate, and 0.5% N-lauroylsarcosine) at 4°C and sonicated for 22 cycles (20 s on and 50 s off per cycle) to obtain DNA fragments averaging 500 bp. Chromatins were immunoprecipitated overnight with 1–3 µg of respective antibodies and then incubated with Dynabeads Protein A or G (Invitrogen) for 2 h at 4°C. Beads containing protein–DNA complexes were washed seven times with RIPA buffer (50 mM Hepes, pH 7.5, 500 mM LiCl, 1 mM EDTA, 1% NP-40, and 0.7% sodium deoxycholate), followed by washes with 1 mM Tris-EDTA/300 mM NaCl and 1 mM Tris-EDTA/50 mM NaCl (once each). Cross-linking was reversed, and DNAs were eluted with 100 µl extraction buffer (1% SDS, 100 mM NaHCO₃, and 300 mM NaCl) by boiling for 15 min. Next, proteins were degraded with 1 mg/ml proteinase K at 55°C, and DNAs were extracted by phenol-chloroform and precipitated by ethanol. Both immunoprecipitated DNAs and input DNAs were resuspended in 100 µl 0.1× TE buffer for real-time PCR. The real-time PCR primers used for ChIP assay are listed in Table S2.

Genomic DNAs were purified and fragmented by sonication to generate DNA fragments ranging from 200 to 500 bp. Sonicated DNAs (4 µg) were denatured and incubated with control IgG or anti-5hmC (0.2 µl) at 4°C overnight. Antibody–DNA complexes were captured by protein A beads preincubated with salmon sperm DNA. Immunoprecipitated DNAs were purified with a PCR purification kit and quantified by real-time PCR against input DNAs.

500 ng of genomic DNAs prepared by phenol-chloroform extraction and ethanol precipitation, and treated with RNase A for 15 min at 37°C, were used for bisulfite conversion in an Epitech bisulfite kit (QIAGEN) according to the manufacturer’s instruction. Bisulfite-sequencing primers were designed according to MethPrimer software. Specific fragments were amplified by PCR and cloned into pMD18T vector (Takara Bio Inc.). Isolated plasmids with expected fragments were sequenced, and the methylation states were analyzed by BiqAnalyzer.

To make whole-cell lysates, cells were lysed in lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton X-100, and 1 mM EDTA, pH 8.0) supplemented with protease inhibitor complete mini (Roche) and 1 mM PMSF, 1 mM Na₃VO₄, and 1 mM NaF for 30 min on ice, and cell debris was cleared by centrifugation at 13,000 rpm for 15 min.

To make cytosolic and nuclear extracts, cells were resuspended in hypotonic buffer (10 mM Hepes, pH 7.6, 1.5 mM MgCl₂, 10 mM KCl, and 1 mM EDTA) supplemented with protease inhibitor complete mini (Roche) and gently homogenized by Douncer (10–15 strokes). After centrifugation at 3,000 rpm for 5 min, the supernatant was removed as crude cytosolic fraction; the pellet was washed twice in hypotonic buffer and further lysed in high-salt buffer (20 mM Hepes, pH 7.6, 500 mM NaCl, 1.5 mM MgCl₂, and 1 mM EDTA) supplemented with protease inhibitors to generate crude nuclear fraction. Crude cytosolic and nuclear fractions were further centrifuged at 13,000 rpm for 15 min to remove debris, and supernatants were collected as cytosolic and nuclear fractions, respectively.

Flt3L-differentiated pDCs were cultured on glass slides placed on 24-well tissue culture dishes (2 × 10⁵ cells/well) and stimulated with 2 µg/ml CpG-A or 100 nM CpG-B for 5 h. Cells were spun down and fixed with 4% paraformaldehyde for 10 min at room temperature. After treatment with 0.1% Triton-100 for 5 min at room temperature, cells were labeled with rabbit polyclonal anti-IRF7. For transfected HEK293T cells, mouse anti-Myc or rabbit anti-FlagM2 were used to stain attached cells overnight at 4°C. Anti–rabbit IgG Alexa Fluor 555 (Invitrogen) or/and anti–mouse IgG Alexa Fluor 488 (Invitrogen) were used as secondary antibodies. Slides were washed and mounted in Prolong Gold antifade mounting media with DAPI (Beyotime), and images were collected on a laser capture confocal microscope (FV1200, Olympus) using separate laser excitation to avoid any cross-interference between different fluorophores.

Fluorochrome-labeled antibodies for CD3 (145-2C11), CD8 (53-6.7), B220 (RA3-6B2), CD11c (N418), mPDCA1 (ebi927), Ly6C (HK1.4), and MHC-II (KH95) were from eBioscience. Antibodies for CD4 (GK1.5), Ly6G (IA8), and CD11b (M1/70) were purchased from BD. All antibodies were tested for their specificity with respective isotype controls. Cell-surface staining was performed by incubating cells with specific antibodies for 30 min on ice and washing with MACS buffer twice. All samples were processed with a Fortessa flow cytometer (BD), and data were analyzed with FlowJo software (Tree Star).

Lentiviral vectors pcdh-EGFP expressing WT or mutants of CXXC5 were transiently transfected into HEK293T cells along with packaging plasmids (Δ8.91/VSV-G), and virus-containing media were harvested in 48 h. Lentiviruses were added onto BM cells cultured with RPMI 1640 supplemented with Flt3L (50 ng/ml) or GM-CSF (20 ng/ml) on day 2. After infection for 24 h, virus-containing media were removed and substituted with pDC or BMDC differentiation media. 1 d later, GFP⁺ cells were sorted by FACS and continued in culture for 5 d.

Scrambled or Tet2- and CXXC5-specific shRNAs were subcloned into pLKO.1-EGFP vector, and lentiviral particles were prepared as described above in this section. Gen2.2 cell culture and transduction were conducted according to previous studies (Chaperot et al., 2006; Li et al., 2014). In brief, Gen2.2 cells were cultured in GlutaMax-RPMI 1640 (Thermo Fisher Scientific) supplemented with 10% FBS/2mM l-glutamine/1× MEM-non-essential amino acid/1 mM sodium pyruvate and infected by shRNA-expressing lentiviral particles in the presence of 4 mg/ml polybrene (Sigma-Aldrich) for 48 h. Subsequently, lentiviral-infected GFP-positive cells were sorted with BD FACS Aria II, followed by HSV-1 and R848 stimulation.

6–8-wk-old and sex-matched littermates were infected i.v. with HSV-1 (KOS strain, 5 × 10⁶ pfu) or VSV (Indiana strain, 5 × 10⁶ pfu) in 0.1 ml PBS. Weight loss and survival were monitored every 12 h. Sera for ELISA were collected 6 h after infection, and spleens for real-time PCR analyses were taken 24 h after infection. For plaque assays, sera samples obtained 24 h after HSV-1 infection were sonicated, serially diluted, and added into confluent Vero cells cultured on 24-well plates. 1.5 h after infection, the supernatant was removed and the cells were overlaid by 2× MEM containing 1.6% agar. 5 d later, the overlay was removed and cells were fixed with 4% formaldehyde for 2 h and stained with 0.5% crystal violet. The plaques were counted, and viral titers were calculated accordingly.

Log-rank (Mantel–Cox) test and Student’s t test were used for statistical analysis; P < 0.05 is considered significant.

Fig. S1 shows the gene-targeting strategy and immune cell development for CXXC5-deficient mice. Fig. S2 demonstrates that CXXC5 requires both CXXC and DNA-binding domains to execute its Tet2-dependent function in regulating Irf7 and Ifnα4/β expression. Fig. S3 shows HSV- and VSV-induced splenic IFN responses and morbidity in WT and CXXC5-deficient mice. Table S1 is a list of genes regulated by CXXC5 in pDCs. Table S2 is a list of primers used in this study.

We are grateful to Prof. Gang Wei and Dr. Ying Li from Institute of Computational Biology for their assistance in ChIP assay. We are also indebted to Bangguo Qian for his technical support in confocal microscopy.

This study was supported by the Ministry of Science and Technology of the People's Republic of China Key Development and Research Project (2016YFA0502102 to H. Xiao) and National Science and Technology 973 project (2014CB541902 to H. Xiao and N. Shen), the National Natural Science Foundation of China (grants 31470847 and 91542206 to H. Xiao), and the Strategic Priority Research Program of the Chinese Academy of Sciences (grant XDB19000000).

The authors declare no competing financial interests.

Author contributions: S. Ma, Z. Deng, L. Shi, C. Hao, Z. Zhou, C. Zhou, Y. Fang, J. Liu, J. Yang, X. Chen, T. Li, and A. Zang performed experiments; X. Wan and Y. Zhang generated CXXC5-deficient mice; S. Yin and L. Jiang provided assistance on ChIP assays and microarray data analysis; B. Li, J. Plumas, L. Chaperot, X. Zhang, G.L. Xu, N. Shen, S. Xiong, and X. Gao provided materials and technical supports essential for this project; H. Xiao and S. Ma analyzed the data and wrote the paper.