Hyperlipidemia May Synergize with Hypomethylation in Establishing Trained Immunity and Promoting Inflammation in NASH and NAFLD

We performed a panoramic analysis on both human nonalcoholic steatohepatitis (NASH) microarray data and microarray/RNA-seq data from various mouse models of nonalcoholic fatty liver disease NASH/NAFLD with total 4249 genes examined and made the following findings: (i) human NASH and NAFLD mouse models upregulate both cytokines and chemokines; (ii) pathway analysis indicated that human NASH can be classified into metabolic and immune NASH; methionine- and choline-deficient (MCD)+high-fat diet (HFD), glycine N-methyltransferase deficient (GNMT-KO), methionine adenosyltransferase 1A deficient (MAT1A-KO), and HFCD (high-fat-cholesterol diet) can be classified into inflammatory, SAM accumulation, cholesterol/mevalonate, and LXR/RXR-fatty acid β-oxidation NAFLD, respectively; (iii) canonical and noncanonical inflammasomes play differential roles in the pathogenesis of NASH/NAFLD; (iv) trained immunity (TI) enzymes are significantly upregulated in NASH/NAFLD; HFCD upregulates TI enzymes more than cytokines, chemokines, and inflammasome regulators; (v) the MCD+HFD is a model with the upregulation of proinflammatory cytokines and canonical and noncanonical inflammasomes; however, the HFCD is a model with upregulation of TI enzymes and lipid peroxidation enzymes; and (vi) caspase-11 and caspase-1 act as upstream master regulators, which partially upregulate the expressions of cytokines, chemokines, canonical and noncanonical inflammasome pathway regulators, TI enzymes, and lipid peroxidation enzymes. Our findings provide novel insights on the synergies between hyperlipidemia and hypomethylation in establishing TI and promoting inflammation in NASH and NAFLD progression and novel targets for future therapeutic interventions for NASH and NAFLD, metabolic diseases, transplantation, and cancers.


Introduction
Several metabolic diseases significantly drive the development of cardiovascular disease, nonalcoholic fatty liver disease (NAFLD) [1], and nonalcoholic steatohepatitis (NASH) [2]. NASH is characterized by macrovascular steatosis, hepatocellular ballooning, lobular inflammation, and pericellular fibrosis. NAFLD constitutes a major health concern, and NAFLD prevalence is estimated between 25% and 30% in the general population [3]. However, the exact underlying inflammatory mechanisms that trigger the transition from fatty liver to NASH remain unclear.
Innate immune system inflammation is considered a landmark of NASH pathogenesis. Mouse models of NASH have demonstrated that increased danger/pathogen-associated molecular pattern (DAMP/PAMP) receptor and Tolllike receptor (TLR) signaling [4] are associated with disease progression. Moreover, an intracellular DAMP/PAMP receptor and the inflammasome serve as a metabolic stress sensor and bridge metabolic distress to the initiation of pyroptosis, a proinflammatory programmed cell death [4][5][6][7]. In addition, inflammasomes serve as sensors of metabolic stresses for vascular, liver, and other inflammations [5,6,[8][9][10][11][12][13][14][15]. Inflammasome pathways can be classified into the canonical caspase-1 or the noncanonical caspase-11-dependent inflammasome pathways [16]. The canonical inflammasome pathway has been shown to play a significant role in NASH and NAFLD pathogenesis [17,18]. In addition to the molecular mechanisms of innate immunity, cellular mechanisms also play a significant role in NASH development.
Innate immune cells can develop an exacerbated immunologic responses and long-term inflammatory phenotypes following brief exposure to endogenous or exogenous insults (the first challenge), which leads to an enhanced inflammatory response after a second challenge, which is known as trained immunity (TI) [4,[26][27][28]. TI is not only important for host defense and vaccine response but also for chronic inflammatory processes such as cardiovascular inflammation, EC activation [26,27,29], and liver injuries [30]. However, the role TI plays in the development of NASH/NAFLD has yet to be determined.
DNA methylation plays a significant role in the progression of NAFLD. Previous studies have shown that there are differences in DNA methylation levels of multiple genes between liver samples from NAFLD patients with those from healthy individuals [39,40], and further studies reveled that hypomethylation of different genes was more than the hypermethylation in advanced NAFLD versus mild NAFLD indicating that genes related to steatohepatitis, fibrosis, and tumorigenesis may be demethylated as NAFLD processes and the methylation levels of these genes may reflect the severity of NAFLD [41].
To fill in important knowledge gaps, we performed a panoramic database mining analysis on both human NASH microarray data and microarray data from various NAFLD mouse models and examined a total of 4249 genes by using the strategy we pioneered [9,42,43]. We made significant findings, which provide novel insights on the roles of proinflammatory cytokines and chemokines, canonical and noncanonical inflammasome pathways, TI enzymes, and lipid peroxidation in promoting NASH/NAFLD progression.

Materials and Methods
2.1. Expression profile of innatomic and secretomic genes, canonical and noncanonical inflammasome pathway regulators, trained immune genes, and lipid peroxidation genes in patients' nonalcoholic steatohepatitis (NASH) microarray data and microarray/RNA-seq data from various mouse models of nonalcoholic fatty liver disease. The 10 microarray datasets were collected from the National Institutes of Health-(NIH-) National Center for Biotechnology Information-(NCBI-) Gene Expression Omnibus (GEO) database and analyzed with an online software GEO2R (Figures 1(a) and 1(b)). One dataset was from the high-fat diet (HFD) and methionine-and choline-deficient (MCD) diet model (GSE35961). One dataset was from the high-fat-cholesterol diet-(HFCD-) induced diet model. One dataset was from the glycine N-methyltransferase-(GNMT-) KO genetic model, (4) the liver-specific methionine adenosyltransferase 1A-(MAT1A-) KO (GSE63027) genetic model. There were two human NASH microarrays (GSE63067 and GSE17470) [49]. For the mechanism studies, we collected one dataset from caspase-11-KO mice (GSE115094) (Figures 1(a) and 1(b)). One dataset was from caspase-1-KO mice (GSE32515). One human dataset was for trained immunity pathway (GSE24187). For datasets not formatted for GEO2R, analysis was performed using DESeq2 in R Studio as described by Mistry et al. (2015) (Figure 1(c)). In brief, the expression data obtained from NCBI GEO database was converted to an expression set R script element. Differential gene expression was analyzed using the DESeq2 library. The numbers and detailed information of the 7 GEO datasets are listed in Supplementary  Tables 2A and 2B. 2.2. Ingenuity Pathway Analysis (IPA). IPA was used to characterize the clinical relevance and molecular and cellular functions related to the identified genes. Differentially expressed genes were identified and uploaded into IPA for analysis. Core and canonical pathway analysis was used to identify molecular and cellular pathways [14,15].

Statistical Analysis.
We applied a statistical method similar to a meta-analysis [14,27]. GEO dataset integrity used 7 housekeeping genes (Supplementary Tables 1 and 2) [44]. The target genes with expression changes more than 1.5fold were defined as upregulated genes, while genes with their expression decreased more than 1.5-fold were defined as downregulated genes. Fold change calculations were performed in Excel as previously described by our group.  3 Journal of Immunology Research Briefly, differential gene expression data were imported into Excel and then analyzed by four Excel Macros: (1) organize the DEG data from the DESeq2 and GEO2R, (2) filter data for significant p values, (3) filter data for fold change ± 1:5, and (4) retrieve expression data for various genes of interest (Figure 1(c)).

Human NASH and NAFLD Mouse Models Upregulate
Both Cytokines and Chemokines Classified in the Innate Immune Database and Canonical Secretome. At least 11 NAFLD mouse models have been established [45,46] including diet-induced and genetic-induced models [47,48], but no model accurately reflects the disease's progression in humans [3]. We include four mouse models that best represent the metabolic and inflammatory characteristics of human NAFLD [3,[46][47][48].
To determine whether NAFLD mouse models share innate immune cytokines and chemokines with human NASH, we performed Venn diagram analysis. Six innate immune cytokines and chemokines including CXCL10, tumor necrosis factor (ligand) superfamily, member 13b (TNFSF13B), lymphotoxin beta (LTB, TNFSF3), TNFSF10, interleukin 1 receptor antagonist (IL-1RN), and nicotinamide phosphoribosyltransferase (NAMPT) were shared between human NASH and NAFLD mouse models  Figure 1: The flowchart of our data mining analyses. (a) Workflow included three parts: (1) identify genes and signaling pathways shared by microarrays from several mouse models of nonalcoholic fatty liver disease (NAFLD), (2) identify genes and signaling pathways common in mouse models of NAFLD and patients with nonalcoholic steatohepatitis (NASH), and (3) Figure 2: The cytokine and chemokine expressions are increased in both human nonalcoholic steatohepatitis (NASH) and mouse models of nonalcoholic fatty liver disease (NAFLD). Ingenuity Pathway Analysis results showed that 53 cytokines and chemokines were sorted out from 1376 innate immune genes (innatome, from the Innate Immune Database (https://www.innatedb.com/); also, see our paper (PMID: 33628851)) and 123 cytokines and chemokines were sorted out from 2641 canonical secretome (with signal peptide) genes (from the Human Protein Atlas database (https://www.proteinatlas.org/); also, see our paper (PMID: 32179051)). (a) A schematic figure showed the pathogenic effects of cytokines released from NASH on different organs. Two human NASH datasets (GSE63067 and GSE17470) and four mouse models of NAFLD datasets (GSE35961, GSE63027, GSE63027, and GSE53381) were analyzed. (b, c) The 53 innatome cytokines and chemokines were analyzed in human NASH and mouse models of NAFLD. (d) Venn diagram showed the significant overlapped regulated innatome cytokines and chemokines in human NASH and mouse models of NAFLD. The cytokine and chemokine expressions are increased in both human nonalcoholic steatohepatitis (NASH) and mouse models of nonalcoholic fatty liver disease (NAFLD). (e, f) The 123 canonical secretomic cytokines and chemokines were analyzed in human NASH and mouse models of NAFLD. (g) Venn diagram showed the overlapped significant regulated canonical secretomic cytokines and chemokines in human NASH and mouse models of NAFLD. The cytokine and chemokine expressions are increased in both human nonalcoholic steatohepatitis (NASH) and mouse models of nonalcoholic fatty liver disease (NAFLD). (h) The 16 cytokines and chemokines (innatome and secretome) were upregulated (upregulated in GSE63067 or GSE17470) in human NASH. PMID: 34084175. (i) The 16 cytokines and chemokines (innatome and secretome) were upregulated (GSE35961 or GSE63027) in mouse models of NAFLD. Secretome gene list detailed in our previous publication PMID: 34084175. 8 Journal of Immunology Research ( Figure 2(d)). Two NAFLD mouse models upregulated 22 (78.6%) specific innate immune cytokines and chemokines; and human NASH upregulated two (25%) specific cytokines and chemokines C-C motif chemokine ligands 20 (CCL20) and CMTM2.
Taken together, our results have demonstrated innate immune secretomic cytokines and chemokines suggesting that human NASH and NAFLD mouse models share innate immune mechanisms and there is a significant role of exosomes and caspase-4 secretomic in driving liver and systemic inflammations.

Human NASH Can Be Classified into Metabolic and
Immune NASH; MCD+HFD, GNMT-KO, MAT1A-KO, and HFCD Can Be Classified into Inflammatory, SAM Accumulation, Cholesterol/Mevalonate, and LXR/RXR-Fatty Acid β-Oxidation NAFLD, Respectively. To determine the signaling pathways mediating the transcriptomic changes in human NASH and NAFLD mouse models, we performed IPA analysis for the top 20 pathways for significantly modulated genes in each dataset. Human NASH dataset GSE17470 had top upregulated pathways including phospholipases, liver X receptor (LXR)/retinoid X receptor (RXR) activation, fatty acid β-oxidation I, and leukocyte extravasation, among which leukocyte extravasation was shared by the second human NASH and MCD+HFD, LXR/RXR activation was shared by HFCD, and fatty acid β-oxidation was shared by GNMT-KO and HFCD (Table 1, Supplementary Figure 1).
The GNMT-KO (GSE63027) had top upregulated pathways including fatty acid β-oxidation I, stearate biosynthesis I, antioxidant action of vitamin C, oxidative phosphorylation, glutaryl-CoA degradation kinase, triacylglycerol biosynthesis, and tricarboxylic acid cycle (TCA) cycle II, among which fatty acid β-oxidation I and stearate biosynthesis I were shared with the first human NASH (GSE17470). The MAT1A-KO (GSE63027) had top upregulated pathways including nicotine degradation II, superpathway of cholesterol biosynthesis, cholesterol biosynthesis I, cholesterol biosynthesis III, mevalonate pathway I, and NAD salvage pathway II, among which nicotine degradation II was shared with the first human NASH.
The HFCD model (GSE53381) has only four upregulated pathways including LXR/RXR activation and fatty acid β-oxidation I, among which LXR/RXR activation was shared with the first human NASH, fatty acid β-oxidation II was shared with the first human NASH and GNMT-KO, and nicotine degradation II was shared with MAT1A-KO.
Taken together, our results have demonstrated that the two human NASH datasets have diversified signaling pathways, which allow us to classify the first one into LRX/ Table 1: Ingenuity Pathway Analysis (IPA) for top 20 pathways of significantly modulated genes in each dataset of human NASH studies and mouse models of NAFLD. Leukocyte extravasation was shared by both human NASH studies and in the MCD+HFD mouse model; LXR/RXR activation was shared by HFCD model of NAFLD; and fatty acid β-oxidation was shared by the GNMT-KO and HFCD models of NAFLD. IL-8 signaling was shared by the second human NASH study and the MCD+HFD model. Fatty acid β-oxidation I and stearate biosynthesis I were shared by the first human NASH study and GNMT-KO and HFCD models. Nicotine degradation II was shared by the first human NASH study and the MAT1A-KO model.   [56,65] inflammatory NASH. IPA of upregulated genes in each mouse models of NASH and human NASH identified top pathways for significantly modulated genes. The major differences between metabolic NASH and immune NASH include the following: first, numerous inflammation and immune pathways are identified on the top pathway list, and second, metabolic reprogramming in any given mouse models affects four metabolic pathways including (1) increased glycolysis, (ii) glutaminolysis, (iii) increased accumulation of tricarboxylic acid cycle metabolites and acetylcoenzyme A production, and (iv) increased mevalonate synthesis. Furthermore, this classification provides a criterion for mouse model stratification and selection, providing novel insight of which mouse models best represent various human cases of NASH/NAFLD.

Canonical and Noncanonical Inflammasome Pathways
Play Differential Roles in the Pathogenesis of NASH/ NAFLD. To examine the differential roles of caspase-1-and caspase-11-dependent pyroptosis, we collected 90 canonical and 14 noncanonical inflammasome pathway regulator genes and determined the expression changes of these inflammasome regulators in four NAFLD mouse models.
Statins (β-hydroxy β-methylglutaryl-CoA (HMG-CoA) reductase inhibitors) block cholesterol synthesis and mevalonate generation, preventing TI induction [69]; and statin treatments do not revert the TI phenotype in patients with familiar hypercholesterolemia [70]. We hypothesized that statins may modulate TI enzyme expression in NAFLD models. For this, we collected two microarray datasets of statin-treated human primary hepatocytes [71].
Among the 2004 differentially expressed genes, 16 genes were overlapped with 99 TI enzyme genes (16.2%) [27], of which 15 (15.2%) genes were upregulated and one (1%) gene was downregulated by atorvastatin (Figures 5(a)-5(c)). In addition, among 2448 differentially expressed genes in rosuvastatin-treated human primary hepatocytes, 19 genes were overlapped with 99 TI enzyme genes (19.2%), of which 15 (15.2%) genes were upregulated and 4 (4%) genes were downregulated by rosuvastatin. Then, we used Venn diagram to determine the causative effects of statins in TI enzyme genes upregulated in NAFLD models. The results showed that the atorvastatin treatment overlapped with 13 TI enzyme genes upregulated in human and mouse NASH/NAFLD models. Similarly, rosuvastatin treatment overlapped with 15 TI enzyme genes upregulated in human and mouse NAFLD models. These results have demonstrated that TI inhibitors, statins, promote the expression of TI enzyme genes not only in mevalonate pathways but also in glycolysis pathways and acetyl-CoA generation pathway.       . These reactive aldehydes interact with and inactivate GPx, leading to an increased rate of 12-HpETE and 15-HpETE peroxidation [72]. The nonenzymatic peroxidation mediates conversion of AA to 4-HNE and MDA metabolites [73] (Figure 6). We hypothesized that downregulation of lipid peroxidation antioxidant enzymes CYPs and GPXs is inversely associated with upregulation of cytokines and chemokines in (     NAFLD and that upregulation of proinflammatory lipid peroxidation enzymes COXs and LOXs [74] is associated with upregulations of TI enzymes in NAFLD. To examine these hypotheses, we collected 26 COX lipid peroxidation enzymes, 44 cytochrome P450 CYP lipid peroxidation antioxidant (anti-inflammatory) enzymes [75,76], 26 AA metabolism enzymes, and 8 hepatic glutathione peroxidase antioxidant (anti-inflammatory) enzymes [77].
3.6. As Upstream Master Regulators, Caspase-11 and Caspase-1 Partially Upregulate the Expressions of Cytokines, Chemokines, Canonical and Noncanonical Inflammasome Pathway Regulators, TI Enzymes, and Lipid Peroxidation Enzymes. It has been reported that caspase-11 mediates hepa-tocyte pyroptosis and promotes the progression of MCD diet-induced NASH/NAFLD [80]; and inflammasomegasdermin D-pyroptosis [81] may regulate the progression of NASH/NAFLD [82]. In addition to promoting inflammation by enzymatic cleavage of pre-IL-1β, pre-IL-18, and Nterminal GSDMD, caspase-1 also cleaves as many as 114 substrates, suggesting that caspase-1 may regulate inflammation indirectly via all the protein substrate-mediated signaling [15], 38 interaction protein-mediated signaling [15], and GSDMD-secretome-mediated signaling [23]. However, an important question remains whether caspase-11 modulates the expression of cytokines, chemokines, and TI enzymes. We hypothesized that caspase-11 promotes inflammatory cytokines and chemokines upregulated in NASH/NAFLD and modulates the expression of canonical and noncanonical inflammasome pathway genes and TI enzymes. To examine this hypothesis, we collected all the cytokines and chemokines, canonical and noncanonical inflammasome pathway genes, and TI enzymes upregulated in NASH/NAFLD and crossed with caspase-11-KO and caspase-1-KO datasets (Supplementary Table 2b).
Others also reported that HFD and HFD plus HCD feeding promote the progression of NASH/NAFLD [45,97], which are well correlated with our new working model that hyperlipidemic stimulations are functional as the first stimulation and second stimulation in TI to enhance innate immune responses. Very interestingly, we found that methionine deficiency diet, choline deficiency diet in MCD model, and MAT1A-KO all lead to S-adenosylmethionine (SAM) decrease or deficiency [104] in methionine-homocysteine cycle [91]. Furthermore, since glycine methylation is one of the reactions that contribute most to total transmethylation flux [105], GNMT1-KO lead to decreased glycine methylation [91] and global DNA hypomethylation [106]. Since SAM is cellular methyl donor we reported [85,[87][88][89][90][91]107] and reviewed [84,108], the SAM decrease/deficiency and weakened SAM function lead to decreased ratios of SAM/ S-adenosylhomocysteine (SAH), cellular hypomethylation, decreased histone methylation, and consequently increased histone acetylation, presumably including TI-related histone 3 lysine 27 acetylation (H3K27ac) and H3K14ac, as we reviewed [26,28] and reported [27,29,30,109].
Furthermore, increased histone acetylation by inhibiting histone deacetylase-2 (HDAC2) promotes macrophage infiltration and progression of NASH [110]. MAT1A and GNMT are relatively liver-specific enzymes; hypomethylation in the liver of MAT1A-KO and GNMT-KO can be striking. Hypomethylation may be an essential factor for liver injury but not essentially required for the establishment of TI since H3K4me3 also mediates TI [26] [111]. In support of our conclusion, it was reported that both hypermethylation and hypomethylation of lipid metabolism genes are found in obese patients with hypercholesterolemia [112]. Since hyperlipidemia can act alone to promote NASH/ NAFLD progression and presumably the establishment of TI, both hyperlipidemia and hypomethylation act as second stimuli for TI (Figure 7(c)). As we indicated [27], the TI is a novel mechanism for qualifying any stimuli in the environments and underlying how chronic metabolic diseases [27,28,113] such as inflammation progression in NASH/ NAFLD. Interestingly, we demonstrated that caspase-1 and caspase-11/4 not only serve as metabolic stress-derived DAMP sensors and inflammation initiators but also serve as upstream master regulators for at least partially upregulating inflammatory cytokines, chemokines, canonical and noncanonical inflammasome regulators, TI enzymes, and lipid peroxidation enzymes.
One limitation of the current study is that, due to the low-throughput nature of verification techniques in all the research laboratories, we could not verify every result we identified [61,114]. We acknowledge that carefully designed in vitro and in vivo experimental models will be needed to verify all the findings. Nevertheless, our findings provide novel insights on the roles of proinflammatory cytokines and chemokines [58,59] and canonical secretome [61,115,116], canonical and noncanonical inflammasome pathways, TI enzymes, and lipid peroxidation enzymes in promoting NASH/NAFLD progression as well as novel targets for the future therapeutic interventions for NASH/NAFLD, metabolic diseases, transplantation, and cancers.