Cooperation between NRF2-mediated transcription and MDIG-dependent epigenetic modifications in arsenic-induced carcinogenesis and cancer stem cells

Abstract

Environmental exposure to arsenic, a well-established carcinogen linked to a number of human cancers, is a public health concern in many areas of the world. Despite extensive studies on the molecular mechanisms of arsenic-induced carcinogenesis, how initial cellular responses, such as activation of stress kinases and the generation of reactive oxygen species, converge to affect the transcriptional and/or epigenetic reprogramming required for the malignant transformation of normal cells or normal stem cells remains to be elucidated. In this review, we discuss some recent discoveries showing how the transcription factor NRF2 and an epigenetic regulator, MDIG, contribute to the arsenic-induced generation of cancer stem-like cells (CSCs) as determined by applying CRISPR-Cas9 gene editing and chromosome immunoprecipitation followed by DNA sequencing (ChIP-seq).

1. Introduction

Well-documented epidemiological evidence in recent decades has unequivocally pointed to the carcinogenic effect of arsenic, especially, the inorganic trivalent arsenic (As³⁺), through environmental exposure, including but not limited to air pollution, drinking water contamination, agricultural or industrial activities [1]. It is estimated that about 200 million people worldwide are exposed to environmental arsenic, mostly from drinking water with arsenic levels exceeding the recommended limit of 10 ppb by the guidelines of the US Environment Protection Agency (EPA) and the World Health Organization (WHO) [2]. In United States, a new study by US Geological Survey and Centers for Disease Control and Prevention (CDC) found that about 2.1 million people in this country may expose to high drinking water arsenic due to consume water from private domestic wells that have high concentration of arsenic from natural sources [3]. In recent years, a new concern has emerged regarding the direct consumption of rice, fruits and/or vegetables that are enriched with arsenic due to their cultivation in fields with high levels of arsenic in the soil or irrigation systems with arsenic-contaminated water [4]. Some studies have revealed that rice can absorb 10-20-fold more arsenic than wheat or barley because of the abundance of silica- and phosphate-transporters that move arsenic into rice grains [5]. It is assumed that approximately one-half of the world’s population consumes rice daily [4, 6, 7]. Accordingly, the true number of people exposed to arsenic may be larger than previously estimated.

1.1. Arsenic exposure and human cancer

In a number of earlier reports using large-scale epidemiology, case control studies, ecological analyses, and cohorts of a selected population who were exposed to moderate or high levels of environmental arsenic, a clear trend of dose-response-related odds ratios of cancer diagnosis was found among people with arsenic exposure [7]. The most common cancers associated with arsenic exposure include lung, skin, bladder, liver, breast, and prostate cancer. By analyzing US Geological Survey data on the sediment levels of arsenic and the data of lung cancer incidences in state-wide cancer registries in a combination of databases in the Surveillance, Epidemiology, and End Results (SEER) Program, Putila and Guo revealed a significant association between arsenic and an increased incidence of lung cancer in the US, and they estimated that more than 5,000 lung cancer cases diagnosed in the US annually can be attributed solely to arsenic exposure [8]. Similarly, data from Chile, Taiwan, Bangladesh, and other regions of the world support the idea that arsenic in drinking water, even at low levels, causes human lung cancer [9].

1.2. Possible link between arsenic exposure and cancer stem-like cells

Emerging evidence indicates that arsenic is highly capable of inducing the generation of cancer stem-like cells (CSCs) in several experimental settings. The first experimental data supporting this notion were obtained in vitro from a cell culture model, which showed that arsenic blocks the differentiation of progenitor cells/stem cells [10]. Using an animal model of skin cancer, Waalkes et al. showed that arsenic increased the number of CD34-positive CSCs during skin carcinogenesis, suggesting malignant transformation of normal stem cells and a change in stem cell dynamics in response to arsenic [11]. Considering that lung cancer is the most common cancer associated with environmental arsenic exposure, we used human bronchial epithelial cells as models to test the CSC hypothesis of arsenic carcinogenesis. By using arsenic concentrations similar to those to which humans are exposed in the environment, we found that consecutive treatment of BEAS-2B bronchial epithelial cells with 0.125 to 0.25 μM arsenic for 3-6 months not only induced their malignant transformation but also induced some of the transformed cells to acquire CSC features, including asymmetric division, self-renewal in vitro and in vivo, and elevated expression of stermness genes. Interestingly, in addition to the expression of well-known genes critical to stem cells and CSCs, such as Oct4, Sox2, Myc, and Klf4, the expression of a number of genes involved in metabolism and differentiation were also increased in these arsenic-induced CSCs, such as genes involved in glycolysis and Wnt family genes [12, 13].

Our earlier studies unexpectedly found diminished generation or induction of reactive oxygen species (ROS) in arsenic-transformed cells, in contrast to a long-standing presumption that we and others held: the transformed or cancer cells must have an elevated capacity for ROS generation based on the speculation that ROS cause genetic mutations in cells [14]. Transcriptomic data from arsenic-induced CSCs further suggested that the expression of genes critical for mitochondrial oxidative phosphorylation and the tricarboxylic acid (TCA) cycle are substantially repressed in CSCs, which explains the observed diminishment of ROS generation in arsenic-transformed cells, since metabolism in mitochondria is the major source of ROS generation [13]. Indeed, it has been well-documented nowadays that the compromised function or under-maturation of mitochondria, along with reduced ROS generation, is a common feature of many normal stem cells and CSCs [15, 16]. It is believed that this metabolic feature can maintain stem cells and CSCs in an undifferentiating state.

An interesting question open for discussion is how does arsenic alter the metabolic status to enable the formation of CSCs? By applying CRISPR-Cas9 gene-editing technology and ChIP-seq, our most recent data revealed the critical contribution of NRF2 and NRF2-dependent HIF1α to arsenic-induced metabolic reprogramming. We found that either parallel to or in cooperation with HIF1α, NRF2 is essential for the metabolic shift from the mitochondrial TCA cycle to glycolysis, the so-called Warburg effect, in the arsenic-induced transformation and generation of CSCs [17]. Similarly, NRF2 has been implicated as a central player in arsenic-induced transformation of normal human prostate stem-progenitor cells [18]. Knocking down NRF2 not only reduced the rate of arsenic-induced transformation but also facilitated the differentiation of prostate stem progenitor cells. It was believed that arsenic may interfere with the autophagic process by suppressing the expression of ATP6V1C1 (VMA5), leading to the accumulation of SQSTM1 (p62), which sequesters Keap1 from NRF2, sustaining the activation of NRF2 [18].

1.3. MDIG and cancer epigenetics

Mineral dust-induced gene (MDIG) was originally identified in coal miners’ alveolar macrophages [19] and the T98G human glioblastoma cell line expressing an estrogen-inducible chimeric protein consisting of human c-Myc and the estrogen-binding domain of the human estrogen receptor [20], respectively. Considering the sources of identified gene and their biochemical/biological function, MDIG was also named mina53/mina [20], NO52 [21], JMJD10, and RIOX2 [22] in the literature [23]. Some environmental cancer risk factors, including silica, tobacco smoke, arsenic, World Trade Center dust, PM2.5, etc., have been shown to be highly capable of inducing MDIG in a wide range of cells [23]. Since the MDIG protein contains a conserved JmjC domain that is commonly found in a number of JmjC family histone demethylases, MDIG was originally thought to be an epigenetic regulator of histone protein demethylation. Indirect evidence supporting this speculation is the inverse relationship between MDIG expression and the level of histone H3 lysine trimethylation (H3K9me3) in human lung cancer samples [24]. The involvement of MDIG in histone demethylation was indeed observed in lung epithelial cells [24], glioblastoma cells [25] and hepatocellular carcinoma cells [26, 27] through experiments of overexpression and siRNA/shRNA silencing of MDIG, although a structural analysis did not support histone demethylase activity but suggested that this protein has hydroxylase activity toward the ribosomal protein L27a (RPL27a) [28]. However, it is known that hydroxylation is the first step during the chemical reaction of histone demethylation [29], and both hydroxylation and histone demethylation require co-factors from cellular metabolism, which may play critical roles in the development or generation of the CSCs in response to some environmental factors.

2. NRF2 in arsenic carcinogenesis, the known and unknown

Human NRF2 (NFE2L2) was first identified by Moi et al. using the AP1 and NFE2 elements in the β-globin gene as a probe to screen a cDNA library generated on the basis of K562 cells [30]. The predicted open reading frame (ORF) of NRF2 encodes a 589 amino acid protein with a molecular mass of 66.1 kD. NRF2 has been a subject of intensive studies since its discovery and was revealed as an essential regulator of the expression of many antioxidant and detoxification enzymes. A wide range of extracellular and intracellular stress signals activate NRF2 by dissociating Keap1 from NRF2 and subsequently stabilize and enable the nuclear translocation of the NRF2 protein. Keap1, in complex with CUL3 and RBX1, acts as a primary E3 ubiquitin ligase for the ubiquitination and degradation of NRF2 [31]. In the nucleus, NRF2 forms a heterodimer with either ATF4 or Maff and binds to the antioxidant response element (ARE) to activate the transcription of a number of genes. Evidence that emerged in the past decade has also suggested that NRF2 is a hallmark of cancer [32]. NRF2 has been shown to promote cancer progression and metastasis. Elevated NRF2 activity leads to cancer cell resistance to chemotherapy or radiotherapy. Furthermore, NRF2 is one of the key contributors to metabolic reprogramming during the generation of CSCs [17, 33].

2.1. Some confusions in the molecular weight of NRF2 protein

Structural characteristics of NRF2, including various functional subdomains with the ability to bind DNA or interact with positive and negative regulators, have been well documented [34]. As a member of the cap’n’collar subfamily of basic leucine zipper transcription factors, NRF2 contains NRF2-ECH homology (Neh) domains, among which Neh1 mediates heterodimerization with the Maf protein through the CNC-bZIP region of DNA binding [35], whereas Neh2 is critical for Keap1 binding through two degrons, DLG and ETGE motifs [36]. The neh3-5 domains function as transactivation domains by binding to various components of transcriptional components [37, 38], and Neh6 is critical for redox-independent regulatory action of β-TrCP E3 ubiquitin ligase and glycogen synthase kinase-3β [39], Neh7 has a domain that interacts with the retinoic X receptor α, which represses NRF2 activation [40]. Some of these domains may be essential in determine the transcriptional regulation on target genes and the pro-oncogenic roles of NRF2 in carcinogenesis in response to environmental risk factors.

Despite these detailed structural-functional characterizations, some unsolved questions remain regarding the molecular characteristics of NRF2 protein inside a given living cell. The transcript of NRF2 is approximately 2.2 kb and encodes the NRF2 protein with a predicted molecular weight of 55 to 65 kDa [30]. However, considerable confusion and inconsistency are presented in the literature and datasets of many commercial sources that provide polyclonal and monoclonal antibodies against human NRF2. In fact, during the first cloning of NRF2, Moi and colleagues noted that the major band of the NRF2 translated in vitro was approximately 96 kDa in the SDS-PAGE gel, which is much greater than the predicted size of 66 kDa [30]. This discrepancy in protein size was attributed to the abundance of acidic residues found in the NRF2 protein. However, some later studies identified 65 kDa or about 70 kDa NRF2 proteins in human cell lines [41-43]. To clarify the size issue of the NRF2 protein in a variety of experimental systems, Zhang and colleagues investigated the migration of endogenous NRF2 in an SDS-PAGE gel using several cell line and wild-type or NRF2-knockout mouse tissue samples, and the recombinant NRF2 protein purified from E. Coli using different sources of antibodies [44]. They provided concrete evidence that the actual size of NRF2, regardless of endogenous or recombinant NRF2, is approximately 95 to 110 kDa.

Our recent gene-editing experiments using CRISPR-Cas9 to delete NRF2 in epithelial cells supported the observation of Zhang et al [44] but raised new questions on the possibility of alternative splicing of NRF2 mRNA or partial cleavage of the NRF2 protein that produces NRF2 proteins of different sizes. By using sgRNA targeting the 2^nd exon of the NRF2 gene for CRISPR-Cas9 gene editing, successful NRF2-knockout clones were established. When we first used rabbit monoclonal antibody 12721 from Cell Signaling Technology, which detects an NRF2 protein with a molecular weight of 95 kDa, to screen NRF2-knockout (KO) cell clones, we detected two bands, at positions corresponding to ~95 kDa and ~110 kDa, in the wild-type (WT) clones (Fig. 1A, left panel). The KO cells showed complete deletion of the 110 kDa band but not the 95 kDa band. Thus, the 95kDa band detected by this antibody may not be the NRF2 band suggested by the antibody datasheet. To further validate our finding, we treated WT and KO cells with arsenic at different concentrations. In the WT cells, a dose-dependent induction of the 110 kDa NRF2 band was clearly established. However, there was no significant change in the 95 kDa band after cells were exposed to arsenic (Fig. 1B), further supporting the conclusion that the 110 kDa band, but not the 95 kDa band, is the authentic NRF2 protein band. We also used an antibody from Santa Cruz Biotechnology, mouse monoclonal antibody sc-365949, which detects the 70 kDa NRF2 band, to screen the WT and NRF2-KO clones (Fig. 1A, right panel). In the WT clones, three major bands at 140 kDa, 70 kDa and 40 kDa were detected by this antibody. Knocking out NRF2 by CRISPR-Cas9 did not remove the 140 kDa or 70 kDa band, but the 40 kDa band was not detected, suggesting that the 40 kDa band may represent the NRF2 protein. Accordingly, neither the 70 kDa band, as claimed by the antibody datasheet, nor the 140 kDa band can be considered to be an NRF2 protein band. Since the molecular weight of the 40 kDa band is much smaller than the theoretically calculated molecular weight 66 kDa of the NRF2 protein, it is very likely that this band is either derived from the alternative splicing of NRF2 mRNA or a product generated by the partial cleavage of the NRF2 protein.

Fig. 1.

Molecular weight verification of the NRF2 protein. (A). Western blotting to detect NRF2 protein in NRF2 knockout (KO) cells and wild-type (WT) cells using different antibodies as indicated. Red arrow: NRF2; green triangle and asterisk: nonspecific; (B). Arsenic (As³⁺) induces NRF2, shown with the 110 kDa band in WT cells (red arrow) but not the 100 kDa band in WT cells or NRF2-KO cells (green triangle). (C). The human NRF2 gene has two promoters, one proximal promoter and one distal promoter, as indicated by the H3K4me3 peaks. The distal NRF2 promoter is adjacent to the AGPS gene promoter at a distance of 50 bp.

In human non-small cell lung cancer (NSCLC) and head and neck squamous carcinoma (HNSC), studies by Goldstein et al [45] revealed that exon 2 was missing in an NRF2 variant generated by alternative splicing in cancer cells. Alternative splicing also resulted in deletion of the Keap1 interaction domain of the NRF2 protein, leading to sustained activation of NRF2 in cancer cells. It is currently not known whether the aforementioned 40 kDa NRF2 protein is the product of the NRF2 mRNA missing exon 2. An additional layer of complexity of NRF2 is established by the presence of a distal promoter of the NRF2 gene, as indicated in our ChIP-seq experiment. The proximal promoter drives an NRF2 transcript with five exons. If the distal promoter is active in transcription, the transcript would have eight exons. The presence of distal and proximal promoters was evident by the detection of two strong and sharp H3K4me3 peaks (Fig. 1C), since H3K4me3 is an active promoter mark. The distal promoter of NRF2 is only 50 bp from the promoter of AGPS, a gene that encodes alkylglycerone phosphate synthase for ether lipid biosynthesis. It remains to be further determined whether the new transcript from this distal promoter and the protein derived from this promoter are generated under both physiological and pathological conditions.

2.2. NRF2 is a primary driving force for arsenic-induced metabolic reprogramming

A considerable number of studies have addressed the defensive role of NRF2 during the processes of molecular carcinogenesis based on the important contribution of NRF2 to the basal and inducible expression of numerous antioxidant and detoxifying genes [46]. Many xenobiotics are chemical carcinogens with a certain degree of electrophilic and oxidative properties. Detoxification of these xenobiotics, accordingly, is an important step for protection against carcinogenesis. Therefore, it is not surprising that some scientists explored chemoprotective reagents that activate NRF2 and proposed that the consumption of vegetables containing high levels of sulforaphane or curcumin has the capacity to activate NRF2 [47]. However, increasing evidence suggests that the protective role of NRF2 in carcinogenesis is either very limited or context-dependent. There is no shortage of reports that revealed the pro-carcinogenic role of NRF2, as many cancers exhibit significantly elevated NRF2 activity, which is associated with sustained proliferation of cancer cells or escape of cancer cells from chemotherapy-induced apoptosis [48].

A number of earlier studies demonstrated NRF2 activation by arsenic in several cell lines [49]. In contrast to classic NRF2 activators, arsenic-induced NRF2 activation occurs through a noncanonical mechanism. Rather than S-alkylating the pivotal cysteine residues C273, C288 and C151 in Keap1, arsenic disrupts the later stage processes of autophagy and induces the accumulation of SQSTM1 (p62), which can sequester Keap1. This sequestration of Keap1 by SQSTM1 allows NRF2 to evade ubiquitination by the Keap1-Cul3-Rbx1 complex [50, 51]. It is believed that this noncanonical activation of NRF2 by arsenic causes prolonged NRF2 activation, leading to the transcription of a number of oncogenes or cancer stemness genes for the acquisition of malignant transformation of normal or noncancerous cells.

What is the molecular basis of NRF2-mediated carcinogenesis? The answer to this question obviously involves many more factors than the classic NRF2-target genes encoding detoxification and antioxidant molecules. Rather than focusing on a single signaling pathway or gene set that is linked to cancer development, we recently studied the impact of NRF2 in arsenic-induced CSCs through global genomic ChIP-seq and an untargeted metabolomic analysis to dissect NRF2 signaling during the malignant transformation and reprogramming of cancer cells [17]. As we had reported previously, consecutive treatment with environmentally relevant concentrations of arsenic for three to six months induced the transformation of noncancerous epithelial cells. Further analysis suggested that some of these transformed cells showed increased expression of Oct4, Sox2, Klf4, Myc, Tcf4, and several Wnt family genes, indicative of CSCs. Pathway assays using the Enrichr program indicated that many upregulated genes in arsenic-induced CSCs are target genes of stemness transcription factors, such as KLF11, KLF and TCF3. An additional bioinformatics analysis performed by evaluating transcription factor perturbations followed by expression (TFPFE) revealed that NRF2 is one of the key transcription factors of the upregulated genes in CSCs.

Because of the observed downregulation of the genes associated with oxidative phosphorylation and the TCA cycle in arsenic-induced CSCs, an indication of a metabolic shift in these cells, we performed untargeted global metabolomic analysis on control cells and arsenic-induced CSCs. A pronounced decrease in the levels of metabolites in the TCA cycle, including those citrate, isocitrate, α-ketoglutarate, succinate, fumarate, and malate, was noted in the arsenic-induced CSCs. Additional tests suggested that even after short-term treatment of noncancerous epithelial cells with 0.25 μM arsenic for 3 days, most of the metabolites in the TCA cycle were decreased, suggesting that arsenic is effective in compromising the function of mitochondria. In contrast, metabolites in the two main subpathways of glycolysis, the hexosamine biosynthetic pathway (HBP) of protein O-GlcNAcylation and the serine-glycine pathway associated with one-carbon metabolism, were substantially increased in the CSCs, clearly indicating metabolic reprogramming from the mitochondrial TCA cycle to glycolysis.

The regulation of NRF2 signaling on genes involved in detoxification and antioxidant responses has been well documented. Our ChIP-seq data revealed that NRF2 is also a master regulator of genes critical in glycolysis and subpathways of glycolysis. Most of these genes exhibited a sharp NRF2 peak that was further enhanced by arsenic treatment. Indeed, analyzing these NRF2 peak regions among genes using the findMotifs Genome program showed that the majority of the peaks represent the known NRF2-binding motif TGCTGAGTCAT or the de novo motif TGACTCAGCA. An interesting finding obtained from the ChIP-seq data indicated the self-amplification of NRF2 signaling. There was an arsenic-enhanced NRF2 binding peak at the proximal promoter region of the NRF2 gene (Fig. 2). In addition, several key regulators of NRF2 activation appear to be regulated by NRF2, such as Keap1, ATF4, SQSTM1, VCP, and BACH1. The Wiki pathway assay confirmed that most of these NRF2-enriched genes are in the NRF2, AhR, oxidative stress, and ferroptosis pathways (Fig. 2, upper-right panel).

Fig. 2.

Arsenic induces self-amplification of NRF2 and the pathways of NRF2-regulated genes. Upper left: Screenshot of ChIP-seq data from control and arsenic-treated cells for the NRF2 gene as determined with Genome Browser. The arsenic-enhanced NRF2 peak at the NRF2 gene promoter is indicated by a red arrow. Upper right: Wiki pathway (WP) assay of arsenic-induced NRF2-regulated genes. Bottom left: Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway assay of arsenic-induced NRF2-regulated genes. Bottom-right: Biological process (BP) assay of arsenic-induced NRF2-regulated genes.

Our ChIP-seq data of arsenic-treated cells also showed how NRF2 has a “dark side” effect on oncogenesis. First, NRF2 is a direct transcriptional regulator of HIF1α, which is critical for the nonhypoxic induction of HIF1α by arsenic. In addition to HIF1α, arsenic-induced NRF2 also promotes the gene expression of some key HIF1α signaling molecules, including Akt3, EPAS1, ARNTL, and GLI2. Second, NRF2 controls the expression of several important oncogenes and stemness transcription factors. These genes include Myc, Sox2, KLF4, TCF19-Oct4, NAMPT, BACH1, ZEB1, CD44, and EGFR. Indeed, “pathways in cancer” was the top-ranked pathway in the KEGG pathway analysis of these arsenic-stimulated NRF2-induced genes (Fig. 2, bottom left). Finally, arsenic treatment enriched NRF2 binding to genes in a wide range of developmental processes, such as morphogenesis and programmed cell death (Fig. 2, bottom right), all of which are important contributors to environmental carcinogenesis and the generation of CSCs.

3. MDIG and environmental carcinogenesis

Increased expression of MDIG has been detected in a number of human cancers [23, 52]. The expression of MDIG can be induced by a number of environmental hazards, including arsenic [53], silica [19], tobacco smoke [24], and building dust [54]. Functionally, MDIG has been linked to oncogenic activity [23], migration and invasion of cancer cells [55], specialization of Th17 cells [56, 57], the Th2-type response [58], anthelmintic effects [59], HIV-1 latency [60], etc. Some MDIG activities may be linked to its potential histone demethylase property. It remains to be determined whether these effects of MDIG are also associated with its recently identified hydroxylase activity on ribosomal protein RPL27a [28].

Similar to the majority of histone demethylases, the MDIG protein contains a conserved JmjC signature motif. Although it is still debatable whether MDIG acts as a histone demethylase, studies from different laboratories have indicated the involvement of MDIG in the demethylation of some histone trimethylation marks, including H3K9me3, H3K27me3, H3K36me3, and H4K20me3 [60, 61]. By screening the MDIG expression status and the level of H3K9me3 in human lung cancer samples that were case-matched with adjacent normal lung tissues, we first found that most lung cancer tissues expressed a higher level of MDIG protein along with a notable decrease in H3K9me3 compared to the normal adjacent tissues. Transient overexpression of exogenous MDIG in bronchial epithelial cells partially diminished H3K9me3 [24]. Although it was marginal to moderate, the demethylase activity of the immunoprecipitated MDIG was detected in a test tube incubation with histone H3 peptide containing H3K9me3 [62].

Studying glioblastoma cell proliferation and tumorigenesis, Huang et al [63] supported the notion that MDIG downregulates H3K9me3. MDIG is commonly overexpressed in several glioblastoma cell lines. Data mining for clinical glioblastoma cases further showed that MDIG is a prognostic factor for this malignancy: A high level of MDIG predicts poorer survival of patients. Using shRNA to silence MDIG, Huang et al observed an elevation of H3K9me3 in the LN-229 glioblastoma cell line. This increase in H3K9me3 by MDIG silencing was associated with an overall decrease in several key regulatory genes for the cell cycle, including CDK2, CDK4, cyclin D1, cyclin B1, cyclin E2, etc. These data strongly indicate that MDIG contributes to the demethylation of H3K9me3 and subsequent cell cycle gene expression, cell proliferation and tumorigenesis.

Initially, we mainly focused our effort on the oncogenic role of MDIG in human lung cancer. However, at one point, when we had the opportunity to access human hepatocellular carcinoma (HCC) tissue samples [64], we also evaluated the expression level of MDIG in HCC. With few exceptions, most of the HCC tissues expressed higher levels of MDIG, as determined by RT-PCR (Fig. 3A). Analysis of HCC patient survival data provided strong evidence indicating worse disease-specific survival (DSS), relapse-free survival (RFS) and progression-free survival (PFS) in the patients with high MDIG expression levels (Fig. 3B). These observations support the recent findings by Ye et al who showed a significantly elevation of MDIG among 284 cases of HCC specimens relative to the adjacent normal hepatic tissues [65]. Additional evidence pointing to the oncogenic role of MDIG in human HCC was recently provided by Li and colleagues, who showed that among a total of 155 HCC cases examined, 69.7% cases showed an increased expression of MDIG relative to the adjacent noncancerous hepatic tissues [26, 27]. Stable overexpression of MDIG in three individual HCC cell lines resulted in a substantial decrease in H3K9me3. In contrast, knocking down MDIG by shRNA in the Huh7 and MHCC-97H HCC cell lines caused a pronounced increase in H3K9me3 levels.

Fig. 3.

MDIG is a prognostic factor for hepatocellular carcinoma (HCC). (A). RT-PCR for MDIG mRNA in human HCC samples, (C: cancer; N: case-matched adjacent noncancerous liver tissues). Red asterisk: nonspecific band. (B). Kaplan-Meier plot shows that a higher level of MDIG predicts poorer survival of HCC patients. DSS: disease-specific survival; RFS: relapse-free survival; PFS: progress-free survival.

Overexpression of exogenous genes or siRNA/shRNA silencing of the genes of interest has been frequently criticized because of the existence of artifacts or off-target effects. To circumvent this concern, we recently applied CRISPR-Cas9 gene-editing technology to knock out MDIG in the BEAS-2B human bronchial epithelial cell line, A549 lung cancer cell line, and MDA-MB-231 triple-negative breast cancer cell line [61]. ChIP-seq analyses using antibodies against the most common histone trimethylation marks revealed that depletion of MDIG in these cell lines resulted in an enhancement of repressive histone trimethylation marks, including H3K9me3, H3K27me3 and H4K20me3. Knocking out MDIG, however, had limited impact on the active histone trimethylation mark H3K4me3. In contrast, a marginal decrease in H3K4me3 level was observed in the MDIG-knockout cells. In addition, knocking out MDIG had no effect on the H3R8 me2a level. Gene pathway assays for genes enriched with the repressive trimethylation marks in ChIP-seq suggested that MDIG drives the expression of genes in cell growth and stemness, lung fibrosis, TGFβ signaling, cell adhesion, and formation of the extracellular matrix. Interestingly, MDIG appears to be inhibitory for a gene cluster of chemokines [61].

While our ChIP-seq data from three different cell lines with MDIG knocked out indicated that MDIG is most likely an antagonist of repressive histone trimethylation marks, we could not exclude the possibility that MDIG may target some active histone trimethylation marks. Using the MDA-MB-231 triple-negative breast cancer cell line, we added H3K36me3 to our ChIP-seq antibody panel and indeed noted a detectable enhancement of H3K36me3 in the MDIG-KO cells (Fig. 4A). H3K36me3 is an essential epigenetic mark for transcript elongation during gene transcription. Indeed, the enrichment pattern of H3K36me3, as shown in a Genome Browser assessment of our ChIP-seq data, is unique. In contrast with H3K4me3, which shows sharp peaks centered on gene promoters, and H3K9me3, H3K27me3 and H4K20me3, which show some scattered peaks on promoters, exons, introns, and intergenic regions, H3K36me3 predominantly marks the entire gene body, as shown for the representative genes TGFBI, PMEPA1 and COL5A1, which exhibited a pronounced enrichment of H3K36me3 in MDIG-KO cells (Fig. 4B and 4C). The possible effect of demethylase-like activity of MDIG on H3K36me3 was also observed in the CRISPR-Cas9 screening for HIV-1 latency-promoting genes [60]. In an in vitro demethylation assay, recombinant MDIG protein showed strong demethylation of H3K36me3 and, to a lesser degree, H3K9me3. It is believed that demethylation of H3K36me3 on the HIV-1 LTR by MDIG prevents KAT8 from binding to and acetylating nucleosomes, leading to the latent state of the HIV-1 proviruses.

Fig. 4.

MDIG regulates H3K36me3. (A). Trend plot of H3K36me3 in ChIP-seq. Data show merged peaks. (B & C). Screenshot of H3K36me3 on the TGFBI, PMEPA1 and COL5A1 genes in WT and MDIG-KO cells.

Studies by Ge et al. provided the first evidence that MDIG (RIOX2) is a histidine hydroxylase on ribosomal protein RPL27a [28]. It is currently unknown whether, in addition to RPL27a, other targets serve as substrates for MDIG-catalyzed histidine hydroxylation. The biological outcome of MDIG-mediated protein hydroxylation also remains to be elucidated further. In our recent posttranslational proteomic analysis of wild-type and MDIG-knockout (KO) MDA-MB-231 cells, we found that RPL27a hydroxylation disappeared in MDIG-KO cells, suggesting that MDIG indeed functions as a histidine hydroxylase. In addition to RPL27a, we identified target proteins that may be hydroxylated by MDIG, such as RPL22L1 and H2AC21. In MDIG-KO cells, we noted enhanced nuclear translocation of RPL27a and other ribosomal proteins, along with an overall increase in a large number of ribosomal proteins. Thus, the hydroxylation of ribosomal proteins may delay the nuclear translocation and destabilization of these proteins. Because arsenic, as well as certain other environmental factors, can induce MDIG expression, either MDIG-mediated histone demethylation or ribosomal protein hydroxylation, or both, may play important roles in arsenic-induced carcinogenesis and the generation of CSCs.

4. Cross talk between NRF2 and MDIG in arsenic carcinogenesis

The accessibility of transcription factors, such as NRF2, HIF1α NF-κB, AP-1, AHR, SP1, Myc, Oct4, Sox2, Klf4, etc., to the target genes is mainly determined by the configuration of chromatin. Posttranslational modification of histone proteins, such as methylation, acetylation, phosphorylation, ubiquitination, and DNA methylation, can either tighten or loosen the chromatin structure to affect transcription factor binding to the promoter or enhancer regions of regulated genes. In general, H3K9me3, H3K27me3, H4K20me3, and DNA methylation are repressive methylation marks for the formation of heterochromatin and silencing of gene transcription by preventing access of transcription factors to their targets. In contrast, H3K4me3, H3K36me3, and acetylation of some lysine residues in histone H3 loosen the condensed chromatin to make the promoters or enhancers of the genes more accessible to transcription factors [66, 67]. Our MDIG gene knockout studies followed by ChIP-seq clearly indicated an antagonistic effect of MDIG on repressive histone trimethylation marks, suggesting that the presence of MDIG favors the open chromatin conformation and active gene transcription by transcription factors.

Although there is no evidence of physical interaction between MDIG and NRF2 [68], there are certain levels of cross-talk between these two factors. The ChIP-seq data suggested that MDIG and NRF2 undergo some measurable mutual regulation. Knocking out MDIG in epithelial cells caused the enrichment of H3K9me3 in the first intron region next to the distal promoter of the NRF2 gene (Fig. 5, indicated by a red arrow). In addition to NRF2, MDIG knockout also increased the level of H3K9me3 on several key NRF2 regulators. In MDIG-KO cells, both Keap1 and SQSTM1 were showed increased levels of H3K9me3, a repressive chromatin methylation mark, and downregulation of H3K4me3, an active transcription mark. Accordingly, MDIG appears to be able to positively regulate NRF2 and its regulators. In contrast, NRF2 might influence the gene expression of MDIG indirectly. In bronchial epithelial cells treated with arsenic, a notable enhancement of HIF1α in the MDIG gene promoter was observed (Fig. 6, indicated by the red arrow). As we reported recently, HIF1α is a direct transcriptional target of NRF2 [17]. There is a strong NRF2-binding element upstream of the HIF1α gene promoter, and arsenic treatment significantly increased NRF2 binding at this site. Thus, the enhanced HIF1α binding on the MDIG promoter in response to arsenic can be partially attributed to the activation of NRF2 by arsenic. This notion is in agreement with our previous report showing the induction of MDIG mRNA and protein by arsenic in epithelial cells [53]. Arsenic treatment also induced self-amplification of NRF2 signaling, as evidenced by the elevated NRF2 and HIF1α enrichment peaks on theNRF2, Keap1 and SQSTM1 genes (Fig. 6).

Fig. 5.

ChIP-seq data showing the indicated methylation status of MDIG, NRF2, Keap1 and SQSTM1 in WT and MDIG-KO cells.

Fig. 6.

ChIP-seq data showing NRF2 and HIF1α binding to the genes of MDIG, NRF2, Keap1, and SQSTM1 in the control cells (Ctrl) and cells treated with 1 μM As³⁺ for 6 h.

An additional level of cross-talk between MDIG and NRF2 occurs on their common sets of target genes. As we had reported previously, knocking out MDIG resulted in an increased level of H3K9me3 and/or H3K27me3 on genes important for cell growth, stemness and inflammation, such as Sox4, KIT, WNT5A, BICC1, and TGFβ [61]. Both ChIP-seq and RNA-seq of epithelial cells treated with 1 μM arsenic for 6 h showed that many of the genes regulated by MDIG are also transcriptional target genes of arsenic-induced NRF2, NRF2-dependent HIF1α, or both [17]. Knocking out MDIG caused a significant enrichment of H3K9me3 in the gene body and the downstream intergenic region of the TGFβ2 gene (Fig. 7, upper left panel indicated by a red arrow). Although there was no obvious binding of NRF2 at the promoter or gene body of TGFβ2, there were many NRF2 peaks that were enhanced by arsenic treatment in the downstream intergenic region of the TGFβ2 gene (Fig. 7, upper right panel, indicated by multiple red arrows). In addition, arsenic also induced HIF1α binding to the promoter of the TGFβ2 gene. Bicaudal C1 (BICC1) is an RNA-binding protein important for embryo development, morphogenesis, and progenitor cell maintenance [69]. The expression of BICC1 is MDIG-dependent. Knocking out MDIG in epithelial cells not only diminished the active transcription mark H3K4me3 but also significantly enriched the repressive mark H3K9me3 (Fig. 7, bottom-left panel). The binding of both NRF2 and HIF1α to the BICC1 gene was enhanced in the cells in response to arsenic (Fig. 7, bottom-right panel). In addition to TGFβ2 and BICC1, many genes involved in development, cellular stemness and inflammation also exhibit dual regulation by MDIG and NRF2. Indeed, biological process (BP) pathway analysis unraveled that development pathway is one of the top pathways in both NRF2-regulated genes (Fig. 2, bottom right panel) and MDIG-regulated genes (Fig. 7E in [61]). Accordingly, in the arsenic-induced generation of CSCs or arsenic carcinogenesis, MDIG and NRF2 exert a concerted effect, which may accelerate the malignant transformation of normal cells exposed to arsenic and induce tumorigenesis due to the generation of CSCs.

Fig. 7.

Cross-talk between MDIG and NRF2 on their common target genes. Left panels: ChIP-seq shows MDIG KO increases in H3K9me3 on TGFβ2 and BICC1 genes. Right panels: ChIP-seq shows that arsenic enhances NRF2 and HIF1α binding to the TGFβ2 and BICC1 genes.

5. Perspective

Environmental arsenic contamination in drinking water, air, soil, and food remains a serious worldwide health concern for the public. Emerging evidence has shown that both high and subtle levels of arsenic exposure can cause health problems. Arsenic exposure in children can also predispose them to other health issues later in life [70]. Although arsenic has been classified as a group I human carcinogen, arsenic itself is a weak mutagen [71]. Thus, similar to a vast majority of environmental hazards, the carcinogenicity of arsenic is very likely achieved through an epigenetic mechanism. The contribution of MDIG to the demethylation of repressive histone trimethylation marks not only facilitates access of transcription factors, such as NRF2 and HIF1α, to their target genes for expression but may also destabilize the genome and sensitize genomic damage elicited by environmental metals and chemicals. Elucidating the interaction between MDIG and NRF2, accordingly, will help in understanding the molecular mechanisms of arsenic carcinogenesis and provide new insights into cancer therapy by targeting MDIG, NRF2, and their downstream effectors in metabolism and cell stemness.