DNA methylation involves the transfer of a methyl group to the C5 position of the cytosine to form 5mC, an event mediated by DNMTs, which primarily silence the expression of target genes by preventing the binding of TFs to DNA. DNA demethylation is mediated by TETs, which catalyzes oxidation of 5mC to 5hmC, 5fC, and then 5caC, thereby removing this inhibitory epigenetic mark to allow gene transcription. In AS, DNA methylation is specifically associated with disease type and progression, as well as disease onset, vascular events, and plaque stabilization. For example, there is a unique whole-genome landscape of DNA methylation in AS lesions, compared to surrounding normal vessel tissue (38); a disease- or location-specific methylation pattern is associated with MI, rather than ischemic stroke (39); disease progression-specific CpG methylation profiles correlate with the grade of lesions (40); after cerebrovascular events, global demethylation is associated with up-regulation of anti-inflammatory genes and likely contributes to plaque stabilization (41). Moreover, DNA methylation of specific target genes (e.g., TRAF3 that encodes TNF receptor-associated factor 3, PPM1A that encodes protein phosphatase 1A, among many others) is also closely associated with AS or its treatment (42). With rapid advances in the technology of single-cell analysis such as scRNAseq and mass cytometry by the time-of-flight (CyToF) (43, 44), cell type-specific profiles of either global or gene-specific DNA methylomes (e.g., macrophage, endothelial cell/EC, smooth muscle cells/SMC, lymphocyte, etc.) may soon be available for more precisely monitoring the dynamic changes of DNA methylation as well as understanding their clinical significance in AS.

To date, understanding of the functional roles of the epigenetic modifiers involving DNA methylation is relatively limited in AS. Functionally, DNMT3A and DNMT3B catalyze de novo DNA methylation. DNMT3A and TET2 represent the most commonly mutated genes in patients with coronary heart disease carrying clonal hematopoiesis of indeterminate potential (CHIP) (45). Transplantation with bone marrow of Dnmt3a−/− mice significantly increases plaque size of Ldlr−/− mice (46), suggesting the anti-AS role of DNMT3A. However, Dnmt3a expression is suppressed in M2 [M(IL-4)] macrophages, probably via an lncRNA called DNMT3aos located on the antisense strand of Dnmt3a (47). Interestingly, the effect of TET2 resembles that for DNMT3A, although these two epigenetic enzymes have opposite functions in regulating DNA methylation. Partial reconstitution of bone marrow clonal hematopoiesis by transplanting Tet2-mutant cells increases plaque size in Ldlr−/− mice, in association with increased IL-1β production by Tet2-mutant ASAMs via NLRP3 inflammasome (48). TET2 specifically represses IL-6 expression in macrophages, an event for inflammation resolution (48, 49). DNMT1 is a DNA methyltransferase maintaining DNA methylation. Macrophage-specific expression of Dnmt1 promotes AS in Apoe−/− mice, via increasing M1 cytokine production but suppressing the M2 gene expression, in association with promoter methylation and thus down-regulation of PPAR-γ or KLF4 (Kruppel-like factor 4) (50, 51). Following the paradigm shift from profiling epigenomics to functional epigenetics in the epigenetics field (52), the functions of these or other epigenetic enzymes involving DNA methylation will be defined more precisely and thus become a potential targets for anti-inflammatory therapy.

H3K4me3 is one of activating histone PTMs, which is reciprocally regulated by MLL (KMT2A) and KDM5B, respectively. Emerging evidence supports the involvement of H3K4me3 in the regulation of macrophages involving AS. H3K4 methylation stepwisely increases in macrophages during AS progression, in association with disease severity (53). Training monocytes by oxLDL, rather than native LDL, results in up-regulation of multiple pro-inflammatory and pro-AS genes, in association with H3K4me3 on their promoters (Figure 1A), while this event can be abolished via H3K4me3 inhibition by pan-KMT inhibitors (54). MLL is up-regulated in M1 [e.g., M(IFNγ)] macrophages, in association with increased H3K4me3 (Figure 1A); in contrast, its enzyme activity is reduced in M2 [M(IL-4 + IL13)] macrophages (55). Increased H3K4me3 up-regulates the pro-inflammatory genes, which can be prevented by an inhibitor of the MLL-Menin interaction (MI2-2). Consistently, pan-KMT inhibition (e.g., by 5’-methylthioadenosine/MTA) also inhibits LPS-induced expression of M1 genes and LPS + IFNγ-triggered secretions of pro-inflammatory cytokines (56).

H3K9me3 and H3K27me3 represent two inhibitory PTMs. Both of these PTMs are markedly increased in ASAMs and lymphocytes in AS plaques, but they are undetectable in healthy vessel tissues (57). However, these PTMs markedly decrease during disease progression (53), while they do not correlate with the expression of their KMTs, including PRC2 (EZH2 or KMT6, which inhibits M1-like polarization in AS lesions (58)) (Figure 1B), the H3K27me3 methyltransferase G9a (KMT1C), and the H3K9me3 methyltransferases SETDB1 (KMT1E) or SUV39H1/2 (KMT1A/B). SETDB2, a member of the KMT1 family that methylates H3K9, is up-regulated in M1 [M(LPS)] but not M2 [M(IL-4)] macrophages (59). While SETDB2 is highly expressed in murine AS lesions, its genetic deletion in hematopoietic cells promotes inflammation and accelerates AS in Ldlr−/− mice, in association with enhanced expression of pro-inflammatory genes but attenuated efferocytosis in CD45⁺ ASAMs (Figure 1C).

Jumonji domain-containing 3 (JMJD3) (KDM6B), which specifically demethylates H3K27me3, is involved in the reprogramming of M1 [M(LPS)] polarization (60) (Figure 2A). More than 70% of LPS-inducible genes are JMJD3 targets in macrophages. However, although Jmjd3 deletion increases H3K27me3, but does not markedly affect most of those LPS-induced genes, suggesting that JMJD3 only acts to fine-tune the transcriptional program for inflammatory gene expression in LPS-activated macrophages and this action is independent of its H3K27me3 demethylase function. Although the enzymatic activity of JMJD3 seems unable to determine the direction of macrophage polarization by itself, it works together with the H3K9me2/H3K27me2/H4K20me1-specific demethylase KIAA1718 and certain transcription elongation-regulated proteins to demethylate H3K27me3 of pro-inflammatory genes. In contrast, down-regulation of either JMJD3 or KIAA1718 attenuates mRNA elongation of these genes (61). In this case, JMJD3 may require collaboration with an additional epigenetic modifier (e.g., KIAA1718) to act as an H3K27me3 demethylase in M1 reprogramming, at least in response to LPS. Serum amyloid A (SAA), another pro-inflammatory factor involved in AS, induces JMJD3 expression, in association with reduced H3K27me3 in macrophages (62). Unlike LPS-induced genes, SAA-induced expression of pro-inflammatory genes can, however, be blocked by Jmjd3 knockdown or inactivation, via restoration of H3K27me3, suggesting that the JMJD3 functions may vary in a stimulus-specific manner. Considering JMJD3 and H3K27me3 as a potential target in inflammation (63), the selective inhibitors of H3K27-specific demethylases have been developed as a novel anti-inflammatory therapy. For example, GSK-J1 (an inhibitor of JMJD3 and UTX, both belonging to the KDM6 subfamily), which specifically binds to the catalytic pocket of JMJD3 to inhibit its demethylase activity, sharply inhibits pro-inflammatory gene expression in macrophages, via increasing H3K27me3 (64).

On the other hand, IL-4 also induces JMJD3 expression and thus activates M2 genes by reducing H3K27me2/3 at their promoters (65), suggesting a role of JMJD3 in M2 [e.g., M(IL-4)] polarization (Figure 2B). JMJD3 also mediates M2-like polarization in response to M-CSF, but is not involved in M1-like polarization induced by GM-CSF. In this context, helminth infection or chitin fails to induce M2-like polarization in Jmjd3−/− mice (66). Although Jmjd3 deficiency results in a global increase in H3K27me3 at the promoters of numerous genes, only a small number of them are specifically affected by Jmjd3 deletion. In the latter, IRF4 (interferon regulatory factor 4) represents one downstream target responsible for M2-like polarization, expression of which is associated with the demethylase activity of JMJD3. While Irf4 knockout copies the phenotype of Jmjd3 deletion, restoration of IRF4 expression can rescue the M2-like polarization that is impaired in Jmjd3−/− macrophages. However, GM-CSF also induces IRF4 expression via enhancing JMJD3 demethylase activity, leading to the production of pro-inflammatory CCL17 in a murine model of arthritis, an event blocked by GSK-J1 (67). In addition, the JMJD3-IRF4 axis may also contribute to repolarization of M1-like to M2-like phenotype mediated by adiponectin (68) (Figure 2C). Therefore, JMJD3 and its downstream targets such as IRF4 play dual functions context-specifically in both M1 and M2 polarization.

JMJD3 expression is regulated at transcriptional level by a number of TFs. LPS induces JMJD3 expression via a MyD88-dependent activation of the NF-κB pathway in ECs (69) or the activation of STAT1 and STAT3. For the latter, knockdown of both Stat1 and Stat3, rather than either of them, inhibits the expression of JMJD3 and pro-inflammatory genes in microglia (70). As one of the HIF-1-dependent genes (71), JMJD3 expression also involves the HIF pathway in response to LPS (72). Whereas IL-4 induces JMJD3 expression via STAT6 activation in macrophages, Stat6 knockout only prevents JMJD3 expression induced by IL-4, but not LPS (65). Moreover, Stat6 deficiency prevents H3K27me3 demethylation and M2 gene expression in response to IL-4, while has no effect on the pro-inflammatory effect of LPS. Similar phenomenon has also been found in SOD1-induced M2-like polarization (73). Thus, although the roles of JMJD3 and its downstream targets (e.g., IRF4) may not be specific to M1 or M2 polarization, the upstream signals (e.g., diverse TFs) of JMJD3 seem to determine the direction of JMJD3-mediated macrophage polarization (74). Of note, myeloid Jmjd3 deficiency leads to progression of AS lesions in Ldlr−/− mice, suggesting an anti-AS role of JMJD3 in ASAMs (75). However, it remains unclear whether JMJD3 acts to promote the polarization of ASAMs toward an anti-inflammatory (M2-like) phenotype in this setting.

KDM4A [JMJD2A or jumonji C-domain-containing histone demethylase 3A (JHDM3A)] demethylates H3K9me3, H3K36me2, and H1.4K26me3. We have identified KDM4A as another member of the JMJD demethylase family that regulates M1-like polarization, at least in response to oxidized low-density lipoprotein (oxLDL) (76) (Figure 2C). Exposure to oxLDL results in KDM4A up-regulation, accompanied by the expression of multiple M1 genes. While KDM4A knockdown prevents oxLDL-induced M1-like polarization, it instead promotes the expression of M2 genes. As an oxygen sensor, hypoxia increases the protein level of KDM4A, but not its mRNA level (77), by preventing its degradation via the SCF-containing ubiquitin ligase complex (78). In contrast, other JMJD family members (e.g., JMJD3) are primarily regulated at the transcriptional level via HIF-1α activation (79). However, although oxLDL also activates the NF-κB and HIF pathways (80), KDM4A expression seems to be independent of the NF-κB and HIF-1 pathways in macrophages exposed to oxLDL (76). Therefore, KDM4A inhibition may skew ASAM polarization straight from M1-like to M2-like phenotype (i.e., repolarization), thus serving as a potential target for the development of anti-inflammatory therapy against AS.

Traditionally, the epigenetic modifications were considered to act only temporarily due to their reversible features and thus not to contribute to immune memory. However, this concept has been challenged by recent discoveries that immune cells (e.g., macrophages) can memorize the cellular states and perturbations (e.g., environmental stimuli) without changing the DNA sequence. Such an epigenetic memory is able to burst a rapid and robust inflammatory response once cells encounter the same or even different stimuli (8). For example, Western diets can cause NLRP3-dependent immune memory in monocytes/macrophages, which contributes to chronic and stepwisely worsening inflammation in AS plaques (81). Later, such phenomenon has been attributed to trained immunity (e.g., by oxLDL) via histone epigenetic modifications (e.g., H3K4me1 and H3K4me3), which could be prevented by the inhibitors of HMTs (54, 82, 83). Of note, trained immunity driven by epigenetic reprogramming is closely associated with metabolic rewiring from oxidative phosphorylation to glycolysis (84–86), a well-documented hallmark of M1 macrophages. However, which HMT(s) or KDM(s) are responsible for trained immunity remains to be defined.

Histone acetylations are generally activating PTMs that promote the expression of target genes. To date, only a few HATs have been identified to be involved in macrophage polarization. EP300 (KAT3B) binds to c-Myc via protein arginine methyltransferase 1 (PRMT1) and is recruited to the promoters of target genes, resulting in M2 gene expression (87). Another HAT CBP (KAT3A) mediates transcriptional activation of IFNβ via increasing H3K56-Ac (88). However, a novel HAT named MOF (KAT8) promotes TNF-α/NF-κB-mediated expression of pro-inflammatory genes via increasing H4K16-Ac (89).

HDAC3, a class I HDAC, deacetylates H4K9-Ac and H4K14-Ac, which is involved in the regulation of both M1 and M2 polarization. In HDAC3-deficient macrophages, LPS fails to activate nearly half of the inflammatory genes, suggesting the role of HDAC3 in M1 polarization (90) (Figure 3A). While there are near 700 genomic regions are hyperacetylated at histone H4 in Hdac3−/− macrophages, the number of hyperacetylated regions are tripled after LPS stimulation. In contrast, a large number of regions display H4 hypoacetylated in both untreated or LPS-treated Hdac3−/− macrophages, in which the recognition motifs for the IRF family and STAT1 are mostly enriched. Moreover, while IRF3 (interferon regulatory factor 3) directly controls Ifnb1 transcription, the pro-inflammatory IFNβ-STAT1 axis is, however, impaired in Hdac3−/− macrophages exposed to LPS, in association with up-regulation of PTGS1 (encodes COX-1). During NLRP3 inflammasome activation, HDAC3 translocates to mitochondria and thus restricts fatty acid oxidation (FAO) by deacetylating the non-histone protein HADHA (mitochondrial trifunctional protein subunit α) at K303, which reduces its FAO enzyme activity and promotes IL-1β production by shaping mitochondrial adaptation (91). Thus, HDAC3 seems to trigger M1 gene expression indirectly via diverse mechanisms.

HDAC3 is also involved in the regulation of M2 polarization (Figure 3B). In this case, Hdac3−/− macrophages display an M2-like phenotype analogous to IL-4-induced alternative activation, in association with increased H3K9-Ac and H3K14-Ac (92). Interestingly, although most up-regulated genes in Hdac3−/− macrophages can be induced by both IL-4 and LPS in wild-type macrophages, a large number of the down-regulated genes in Hdac3−/− macrophages are up-regulated only by LPS in wild-type macrophages. Thus, HDAC3 acts as a suppressor of M2 polarization via its deacetylase activity even in the absence of pro-inflammatory stimuli. After exposure to oxLDL, Hdac3−/− macrophages secrete more TGF-β, an M2 cytokine, than wild-type counterparts, in association with increased H3K9/14-Ac at the Tgfb locus. This suggests that HDAC3 directly binds to the regions near the Tgfb promoter and inhibit its expression. Taken together, HDAC3 activation may skew the direction of macrophage polarization from M1-like to M2-like phenotype.

HDAC3 expression is associated with plaque vulnerability in human AS (93), suggesting its pro-AS role. Moreover, conditional knockout of Hdac3 in macrophages transforms their phenotype to an anti-AS fibrotic phenotype, leading to increased collagen deposition and thus more stable plaques in Ldlr−/− mice. Interestingly, while HDAC3 overexpression in ECs promotes endothelial-to-mesenchymal transition (EndMT), HDAC3 inhibition reduces AS lesions in Ldlr−/− mice, further supporting the pro-AS role of HDAC3 (94). On the contrary, up-regulation of HDAC3 in ECs may also inhibit inflammation and AS (95). Nevertheless, whether the functions of HDAC3 described above in the regulation of macrophage polarization would be applied in AS remains uncertain.

During the differentiation of monocytes to macrophages, HDAC9 (a class IIa HDAC) are up-regulated (96), which consists of two isoforms with or without HDAC domain, respectively. The latter, named HDAC-related protein (HDRP) or MEF2-interacting transcription repressor protein (MITR), is a truncated form that lacks deacetylase activity, which functions to recruit other HDACs (e.g., HDAC1 or HDAC3). Although HDAC9 is highly expressed in macrophages, it can be further induced by LPS (via toll-like receptor, TLR), oxLDL, and acetylated LDL (Figure 3C). High levels of HDAC9 in macrophages is maintained by DNMT3A via H3K27me3 repression at it distal promoter region, while HDAC9 in turn binds to TBK1 to enhance its K241 deacetylation and kinase activity (97) (Figure 3D). In this context, Dnmt3a deficiency selectively impairs the expression of type I interferons (e.g., IFN-α and IFN-β) induced by LPS, due to inhibition of TBK1-mediated IRF3 phosphorylation. Therefore, HDAC9 could execute its pro-inflammatory functions via diverse mechanisms involving multiple upstream and downstream pathways. As relatively high basal levels of HDAC9 seems to be required for macrophages, it would be quite challenging to target only the inducible part of HDAC9 to inhibit inflammation triggered by (micro) environmental stimuli.

A genome-wide association study (GWAS) has unveiled a SNP within the HDAC9 gene, which is significantly associated with the risk of large vessel stroke (98). Another genome-wide association meta-analysis has identified two genetic loci (HDAC9 and RAP1GAP) associated with aortic calcification, an independent predictor for the risk of cardiovascular events (99). These observations in the large population-based cohorts provide a strong link between HDAC9 and AS. In this context, Hdac9 deficiency inhibits M1 gene expression via up-regulating PPAR-γ, while promotes M2-like polarization of macrophages and expression of the ATP-binding cassette transporter ABCA1 and ABCG1 via increasing H3K9-ac at their promoters in Ldlr−/− mice (96). In AS plaques, HDAC9 is associated with MMP12 expression in the regions clustered with inflammatory genes in ASAMs. The expression of both Mmp12 and Hdac9 is associated with a unique subtype of macrophages named M4 (100), an inflammatory phenotype that recruits neutrophils to form neutrophil extracellular traps (NETs) in response to oxLDL (101). In Apoe−/− mice, Hdac9 deficiency confers plaque stability via an alternative mechanism involving deacetylation of the non-histone protein IKKα and IKKβ, which leads to the activation of the NF-κB pathway in both macrophages and ECs to drive inflammatory response (102). Of note, a specific inhibitor of the class IIa HDACs (including HDAC9) limits this pro-inflammatory response and attenuates lesion formation. Thus, due to its dual roles in pro-inflammatory response and plaque vulnerability, HDAC9 is considered as a very promising target for the treatment of AS-related diseases.

HDAC7, another class IIa HDAC, is structurally similar to HDAC9. Unlike the latter, HDAC7, however, has minimal deacetylase activity, while often binds to HDAC3 to suppress gene expression. In pre-B cells, HDAC7 inhibits pro-inflammatory genes essential for macrophage functions, while HDAC7 is specifically down-regulated to release this brake during transdifferentiation of pre-B cells into macrophages (103). In differentiated macrophages, HDAC7 promotes the TLR4-induced expression of a subset of pro-inflammatory genes (Figure 3D), an event prevented by a selective inhibitor of the class IIa HDACs (104). In macrophages, HDAC7 binds to and deacetylates PKM2 (pyruvate kinase muscle isozyme 2) at K433, therefore increasing LPS-induced inflammatory responses via promoting glycolysis (105), suggesting a role of HDAC7 in immunometabolism (106). In this case, while the role of HDAC7 in M1 polarization remains uncertain, it appears to be required for maintaining the pro-inflammatory property of M1-like macrophages. Thus, targeting HDAC7 could suppress the pro-inflammatory functions of macrophages, but not reprogramming for M1 polarization. While the role of HDAC7 remains uncertain in AS, it has been demonstrated that HDAC7 plays an important role in the maintenance of vascular integrity by repressing MMP10 expression in ECs (107). The latter would be a major concern for targeting HDAC7 in AS.

The findings involving the effects of HDAC inhibition on inflammation remains controversial thus far. The pan-HDAC inhibitor SAHA (vorinostat) reduces immune cell infiltration and inflammation in AS plaques of hypercholesterolemic Apoe−/− mice (108), while another pan-HDAC inhibitor TSA increased macrophage infiltration in AS lesions, in association with increased histone H4 acetylation (109). In macrophages, TSA impairs the expression of most M1 markers and cytokines induced by LPS or IFNγ + LPS, but also inhibits the expression of M2 markers induced by IL-4 (56). In the latter case, TSA down-regulates both basal and IL-4-induced expression of Arg1 via an HDAC3-independent process (92). MS-275 (entinostat), a selective inhibitor of class I HDACs (e.g., HDAC1 and HDAC3), increases the basal level of the M2 marker Arg1, similar to the phenotype of Hdac3 deletion, while does not affect IL-4-induced Arg1 expression.

Numerous non-histone proteins also serve as substrates of deacetylation by HDACs (110), including various TFs (e.g., RelA/p65, p53, STAT1). For example, the activation of the NF-κB pathway represents a major signal required by M1 [M(IFNγ)] polarization and functions, while NF-κB inhibition can promote M2 [M(IL-4 + IL-13)] polarization (111). Inhibition of class I HDACs (e.g., HDAC3) lead to persistent activation of the NF-κB pathway via preventing deacetylation of RelA (p65) (112–114), which may induce the IKK-dependent expression of pro-inflammatory CXCL8 (IL-8) (115) or up-regulate CD47 to impair efferocytosis by M2-like macrophages (25, 116). Similarly, acetylation of STAT6 also suppresses M2 [M(IL-4)] polarization (117). Therefore, owe to such diversity of HDACs’ substrates, HDAC inhibition may have either anti- or pro-inflammatory activities, dependently upon which class or individual HDAC(s) are targeted more specifically as well as which substrate(s) are involved more preferentially (118).