The roles of inducible chromatin and transcriptional memory in cellular defense system responses to redox-active pollutants

Abstract

People are exposed to wide range of redox-active environmental pollutants. Air pollution, heavy metals, pesticides, and endocrine disrupting chemicals can disrupt cellular redox status. Redox-active pollutants in our environment all trigger their own sets of specific cellular responses, but they also activate a common set of general stress responses that buffer the cell against homeostatic insults. These cellular defense system (CDS) pathways include the heat shock response, the oxidative stress response, the hypoxia response, the unfolded protein response, the DNA damage response, and the general stress response mediated by the stress-activated p38 mitogen-activated protein kinase. Over the past two decades, the field of environmental epigenetics has investigated epigenetic responses to environmental pollutants, including redox-active pollutants. Studies of these responses highlight the role of chromatin modifications in controlling the transcriptional response to pollutants and the role of transcriptional memory, often referred to as “epigenetic reprogramming”, in predisposing previously exposed individuals to more potent transcriptional responses on secondary challenge. My central thesis in this review is that high dose or chronic exposure to redox-active pollutants leads to transcriptional memories at CDS target genes that influence the cell’s ability to mount protective responses. To support this thesis, I will: (1) summarize the known chromatin features required for inducible gene activation; (2) review the known forms of transcriptional memory; (3) discuss the roles of inducible chromatin and transcriptional memory in CDS responses that are activated by redox-active environmental pollutants; and (4) propose a conceptual framework for CDS pathway responsiveness as a readout of total cellular exposure to redox-active pollutants.

Graphical Abstract

1. Introduction

People are exposed to wide range of redox-active environmental pollutants [1, 2]. Air pollution, heavy metals, pesticides, and endocrine disrupting chemicals can disrupt cellular redox status [3, 4]. Over the past two decades, the field of environmental epigenetics has investigated epigenetic responses to environmental pollutants, including redox-active pollutants [5]. Epigenetic phenomena are responsible for changes in phenotype without changes in genotype [6, 7]. The majority of these phenotypic changes are achieved by differential regulation of gene expression by chemical modifications to DNA and its associated proteins (collectively, chromatin), as well as the three-dimensional structure of chromatin [7, 8]. The original definition of “epigenetics” required these modifications to be heritable across cell division [7], although the common usage has expanded to include non-heritable, trans-acting transcriptional regulators, including non-coding RNA [9]. The field of environmental epigenetics has focused on two primary areas: the role of chromatin modifications in controlling the transcriptional response to pollutants and the role of transcriptional memory, often referred to as “epigenetic reprogramming”, in predisposing previously exposed individuals to more potent transcriptional responses on secondary challenge [10]. Past and current work in this field primarily characterizes the epigenetic effects of individual pollutants, with the goal of identifying specific signatures of exposure and effect that can be used to evaluate past exposures and predict future disease [11]. However, individuals are rarely exposed to a single pollutant. Most people are exposed to multiple mixtures of pollutants that may act additively or synergistically to affect health [12]. Therefore, an individual’s total burden of pollutant exposure and their combined effects is likely the best predictor of disease.

Redox-active pollutants in our environment all trigger their own sets of specific cellular responses, but they also activate a common set of general stress responses that buffer the cell against homeostatic insults, including oxidative damage and protein misfolding [13]. These general stress responses occur at the cellular level, rather than the organismal level [13, 14]. In order to avoid confusion with organismal effects of psychosocial stressors (e.g., through the hypothalamic-pituitary-adrenal axis), in this review, I will refer to these pathways as cellular defense system (CDS) pathways. CDS pathways include the heat shock response, the oxidative stress response, the hypoxia response, the unfolded protein response, and the DNA damage response, as well as the general stress response mediated by the stress-activated p38 mitogen-activated protein kinase (p38-MAPK) [13, 14]. Disease can result when these systems are overwhelmed, develop resistance (e.g., fail to respond) over time, or if their sustained activation is harmful to the cell [13, 14]. Therefore, an individual’s ability to mount protective CDS responses by turning on CDS genes is a critically important protection against disease. In addition, because many pollutants activate CDS responses, transcriptional memory at CDS target genes likely reflects an individual’s total redox-active pollutant burden.

My central thesis in this review is that high dose or chronic exposure to redox-active pollutants leads to transcriptional memories at CDS target genes that influence the cell’s ability to mount protective responses. This work builds directly on prior work by Simmons and Ramabhadran that leveraged transcriptional readout of CDS pathways as biomarkers for high-throughput toxicity screening of environmental pollutants [13]. To support this thesis, I will: (1) summarize the known chromatin features required for inducible gene activation; (2) review the known forms of transcriptional memory; (3) discuss the roles of inducible chromatin and transcriptional memory in CDS responses that are activated by redox-active environmental pollutants; and (4) propose a conceptual framework for CDS pathway responsiveness as a readout of total cellular exposure to redox-active pollutants. This review is not intended to be a comprehensive review of the literature. I have included selected examples to illustrate discussed concepts.

2. Chromatin features of inducible genes

CDS pathways sense cellular stress and transmit that information through intracellular signaling pathways that terminate in activation of specific transcription factors [13, 14]. CDS-activated transcription factors translocate to the nucleus and bind their respective transcription factor binding sites, or response elements, to activate target genes [13, 14]. CDS target genes are inducible genes that are newly transcribed or transcribed to a greater degree in response to a stimulus [15, 16]. Inducible genes can be contrasted with constitutively transcribed “housekeeping” genes that are responsible for the baseline functioning of the cell [17, 18]. Inducible genes require flexible and responsive chromatin states that can prevent inappropriate transcription at baseline but can respond rapidly and transiently to stimulus by significantly increasing or decreasing transcription [17, 18]. Target gene sets must remain inducible across cell division, implying mitotic heritability of some component(s) of inducible chromatin at these loci [19]. In this section, I will briefly review the sequence of events required for transcriptional activation and then discuss the evidence for chromatin features that confer inducibility.

2.1. Two-dimensional and three-dimensional control of gene transcription

Gene expression is controlled at the level of gene transcription by chromatin modifications and protein binding to the two-dimensional (2D) genome (reviewed in [20]) and by looping dynamics of the three-dimensional (3D) genome [8, 21]. The eukaryotic 2D genome, also called the linear genome, refers to the DNA double-helix wound around protein complexes called nucleosomes, in a configuration that resembles “beads on a string” (Figure 1.1) [20]. Nucleosomes are protein octamers comprising two heterodimers of histones H2A and H2B and two heterodimers of histones H3 and H4 [20]. Chromatin that is reversibly repressed is called facultative heterochromatin and is formed when nucleosomes are densely packed together, blocking transcription factors from binding DNA and turning on genes [20]. (In contrast, constitutive heterochromatin is not dynamic; it stably silences chromatin near centromeres, telomeres and some repetitive elements [22].) In accessible chromatin, or euchromatin, nucleosomes are sparser and can be easily shifted or removed from DNA to allow binding of transcription factors [20]. Regulatory regions, or elements, within DNA (e.g., promoters, enhancers, silencers, and insulators) contain clusters of transcription factor binding sites that can be exposed to transcription factor binding in euchromatin or restricted from transcription factor binding in facultative heterochromatin [20]. Both DNA and histone proteins can be modified with chemical groups that affect chromatin accessibility [23]. Histone proteins can be modified at both their globular domains and their amino (N-) and carboxy (C-) terminal tails (most commonly at lysine, arginine, or serine residues) with a wide variety of chemical groups, most commonly acetyl and methyl groups [23]. Histone acetylation (e.g., acetylation of lysine 27 on histone H3, or H3K27ac) is generally permissive of transcription (with the notable exception of H4K20ac [24]) and functions both to relax the affinity of DNA for nucleosomes [23], which makes DNA more accessible for binding of proteins required for gene transcription, as well as to recruit chromatin remodeling proteins that further promote DNA accessibility [23]. Histone methylation can enable or repress transcription, depending on the residue that is modified and the protein complexes that are recruited by that specific modification [23]. For example, trimethylation of lysine 4 on the N-terminal tail of histone H3 (H3K4me3) is permissive of transcription, but trimethylation of lysine 9 or lysine 27 on the N-terminal tail of histone H3 (H3K9me3 or H3K27me3) are repressive [25, 26]. Specific nucleotides within DNA can be methylated, too [27]. Methylation of cytosine residues within CG dinucleotides is the most common form of DNA methylation in mammalian genomes but other dinucleotide or trinucleotide sequences can be methylated, as well [28].) DNA methylation within regulatory sequences can repress transcription both by directly blocking binding of some transcription factors [29] and by recruiting other proteins, including those that modify histones, to promote a transcriptionally repressive state [30]. Histone modifying enzymes include histone methyltransferases (HMTs), histone acetyltransferases (HATs), histone demethylases (HDMs), and histone deacetylases (HDACs) [23]. DNA modifying enzymes including DNA methyltransferases (DNMTs) and ten-eleven-translocation (TET) proteins, which promote DNA demethylation [31]. Active regulatory elements are enriched for permissive histone modifications and deficient in DNA methylation and repressed or silenced regulatory elements are enriched in repressive histone modifications and DNA methylation [32]. Elements that contain both permissive and repressive histone modifications are termed “bivalent”; these regulatory regions can pivot quickly to either a more repressive (by losing permissive modifications) or a more permissive (by losing repressive modifications) state [33].

Figure 1. Chromatin characteristics of inducible genes.

Inducible genes use chromatin strategies to regulate their responsiveness to stimuli. Inducible genes maintain accessible chromatin at transcription factor binding sites that enable transcription factors to bind and initiate a sequence of events, including remodeling of inducible chromatin to promote accessibility along the gene body, which is required for transcription. Inducible genes with poised RNA polymerase II can be induced very rapidly, since transcription factor binding triggers pause-release of the polymerase directly into elongation, rather than triggering initial assembly of the transcriptional machinery. Inducible genes with poised RNA polymerase II can be induced synchronously, as well.

The first step in transcription of an inducible gene is the binding of a stimulus-responsive transcription factor to an accessible promoter or enhancer [18]. When a transcription factor binds to a regulatory element that is relatively near the target gene’s transcription start site, it recruits a chromatin remodeling complex that increases chromatin accessibility at the core promoter [18]. Chromatin remodeling proteins or complexes (e.g., the SWI/SNF family, the ISWI family, the CHD family and the INO80 family) [34] are multi-subunit complexes that contain catalytic units that require ATP for activity and translocase subunits that regulate chromatin accessibility by controlling nucleosomes’ histone composition, position along the chromatin, and partial or total eviction from or incorporation into chromatin [35–41]. Chromatin remodeling complexes often contain subunits that can modify histone modifications [20].

Chromatin remodeling of the core promoter enables RNA polymerase II (RNAPII) binding and formation of the pre-initiation complex (PIC) [42, 43]. In addition to RNAPII, the PIC contains general transcriptional factors, as well as the positive elongation factor P-TEFb, the HAT complex SAGA and the transcriptional co-activator Mediator [42, 43]. If a transcription factor binds an accessible distal enhancer, then the chromatin must form a loop to bring the enhancer in close proximity to the core promoter of the target gene [44]. These enhancer-promoter contact loops represent 3D control of gene transcription [20, 44]. Enhancer-promoter contact loops form within larger chromatin loops called topologically active domains (TADS) [8, 21]. TADS are anchored at the loop base by the presence of two CCTC-binding factor (CCTF) insulator proteins, which binding orientation at the base of the loop are critical for loop formation, as well as cohesin complexes through which DNA is threaded to form the loop [8, 21]. TADS form genomic “neighborhoods” of gene regulation, and enhancers can only contact promoters that are contained within the same TAD [21, 45].

Transcription factor binding and PIC formation is not sufficient for a gene to be transcribed. First, RNAPII must be released into active elongation to transcribe the gene (reviewed in [42, 43]). RNAPII contains a C-terminal domain (CTD) that can be modified to regulate transcription [42, 43]. Once RNAPII is bound to the core promoter, its CTD is phosphorylated at serine 5 by the Cdk7 kinase, a subunit of the PIC [42, 43]. This initial phosphorylation triggers RNAPII poising: RNAPII escapes the PIC, transcribes a short stretch of DNA (20–50bp), and then pauses [42, 43]. Poised RNAPII is bound by negative elongation factors that stabilize the paused state [42, 43]. When the CTD of RNAPII is phosphorylated additionally at serine 2, positive elongation factor P-TEFb phosphorylates negative elongation factors DSIF and NELF which then dissociate from chromatin, and RNAPII is released into productive elongation, in a process termed “pause-release” [42, 43]. The chromatin remodeling complex bound at the promoter then shifts, modifies (either via chemical modification or by removal of a single dimer to form less stable hexamer nucleosomes), or evicts nucleosomes in the target gene body to enable RNAPII to process along the full length of the gene [42, 46].

All of the steps described above are required for any gene to be transcribed. At constitutively transcribed genes, the core promoter is maintained in an accessible state due to continuous, sequential binding of RNAPII [18, 20]. However, inducible genes respond to specific cellular cues or stressors [18]. Their responsiveness relies on three key characteristics. First, an initial regulatory element must be accessible to transcription factor binding [47], followed by a subsequent increase in chromatin accessibility at the target gene, both at the core promoter to enable PIC binding and along the gene body to enable RNAPII processivity [18] (Figure 1.1). Second, the speed at which these proteins bind and turn on genes is critical for a timely transcription [48] (Figure 1.2). Third, the ability for multiple genes to be expressed in synchrony is a critical factor for an effective cellular response [49] (Figure 1.3).

2.2. Chromatin accessibility at inducible genes

Transcription factors that are activated by environmental stressors or cues can only bind chromatin at accessible transcription factor binding sites [47]. Current evidence supports a model in which the full repertoire of regulatory elements with accessible transcription factor binding sites is established and maintained within the cell during differentiation [50]. Regulatory element accessibility is stably maintained through cell division by pioneer transcription factors, which bind to these regions and hold them open (i.e., serve as “bookmarks”) (Figure 1.1) [50]. A subset of pioneer transcription factors called lineage-determining transcription factors establish and maintain cellular identity through differential chromatin accessibility at transcription factor binding sites, which enables cell type-specific transcriptional programs [50].

In addition to baseline chromatin accessibility of transcription factor binding sites, chromatin at inducible target genes must be responsive. In order to facilitate rapid increases in chromatin accessibility in response to stimuli, inducible genes often contain unstable nucleosomes [51]. Unstable nucleosomes can be more easily shifted or evicted by chromatin remodeling proteins to increase chromatin accessibility [51]. A common form of an unstable nucleosome is a hexamer nucleosome that contain three, rather than four, protein dimers [52]. Hexamer nucleosomes are formed after eviction of a single histone dimer, rather than an entire nucleosome, by a chromatin remodeling complex [52]. Unstable nucleosomes are more likely to form in CG-rich DNA sequences [53]. Therefore, nucleosomes within CG-rich sequence are inherently less stable and more easily evicted from chromatin stochastically [53]. Almost 40% of mammalian coding genes contain proximal promoter elements with regions of high CpG site density, termed CpG islands [54, 55]. Genes with CpG islands in their promoters may spontaneously activate without the assistance of chromatin remodeling proteins [53]. It is possible that these genes can also be easily induced to higher transcriptional levels in the presence of active chromatin remodeling.

Non-canonical nucleosomes containing certain histone variants are often components of inducible chromatin [56]. For example, the baseline presence of H2A.z (a variant of the canonical histone H2A that forms an unstable nucleosome) at gene promoters is essential for induction target genes of the nuclear hormone receptor estrogen receptor-α (ER-α) [57–60] and H2A.z promoter enrichment increases with the strength of the estrogen receptor binding site [61]. H2A.z is lost from enhancers during induction [57, 62–64], likely due to full nucleosomal eviction or partial eviction of only H2A–H2B dimers during the chromatin remodeling that accompanies activation [65]. However, H2A.z is likely re-incorporated into nucleosome octamers after transcription is completed, in order to maintain baseline inducibility [64]. In another example, the presence of H3.3 (a variant of the canonical histone H3) in gene bodies potentiates transcriptional elongation. The N-terminus of H3.3 contains a serine 31 residue that is absent in the canonical H3 which is phosphorylated (H3.3S31ph) in response to stimulation in nucleosomes within gene bodies of rapidly induced genes [66]. H3.3S31ph binds the histone methyltransferase SETD2 and evicts the elongation co-repressor ZMYND11, which promotes RNAPII processivity and rapid elongation [66]. H3.3 is incorporated into genes immediately after induction and requires active transcription for incorporation [67]. H3.3 turns over continuously at constitutively active ribosomal DNA genes in Drosophila but is stably incorporated at induced Hsp70 genes after they have returned to baseline states [67].

Chromatin remodeling complexes can be blocked from remodeling nucleosomes by the presence of histone modifications [68]. Specifically, mono-ubiquitination of lysine 120 on the C-terminal tail of histone H2B (mono-ubiquitination of H2BK120, or H2Bub1) can block eviction of H2A.z-containing nucleosomes or histone heterodimers by blocking INO80 binding [68]. The H2B de-ubiquitinase USP22 is required for transcription initiation of inducible genes [69]. Overexpression of the E3 ubiquitin ligases RNF20 and RNF40 that deposit H2Bub1 is sufficient to block induction of ER-α target genes in response to hormone stimulus and knockdown of RNF20/40 is sufficient to promote increased induction in response to stimulus [70]. H2Bub1 may also function to define the boundaries of new chromatin opening in response to stimulus. For example, in MCF-7 breast cancer cells, H2Bub1 levels are high overall at ER-α inducible enhancers, except at transcription factor binding sites, to a degree proportional to the strength of the ER binding site [68]. Immediately after induction with estradiol, a “valley” of H2Bub1 enrichment forms at the centers of strong enhancers and these valleys gain other histone modifications of active enhancers [68]. The “peaks” of H2Bub1 on either side of these valleys may prevent the spread of accessible chromatin [68]. In these cells, chromatin rich in H2Bub1 overlaps chromatin that is rich in H2A.z, but H2A.z levels drop when H2Bub1 valleys form and impaired H2Bub1 leads to a decrease in H2A.z both before and after induction [68]. These data suggest that H2Bub1 stabilizes H2A.z in nucleosomes and that H2A.z-containing nucleosomes can be evicted when H2Bub1 is removed, ostensibly by USP22, in response to stimulus [68]. Disruption of H2Bub1 is also associated with aberrant gene activation in primary human cancers [61]. H2Bub1 levels are low in primary tumor samples of late-stage breast cancer, and estrogen-dependent breast cancer cells can proliferate in the absence of estrogen when RNF20/40 is knocked down [61]. However, RNF20/40 knockdown has only a moderate effect on constitutive gene expression [70, 71], supporting a specific role for H2Bub1 in regulating inducible, rather than constitutively active, genes. H2Bub1 may only block chromatin remodeling when it is present in regulatory elements and have different functions in different genetic contexts. For example, H2Bub1 is required for RNAPII elongation within gene bodies in yeast [72–74]. H2Bub1 is also enriched in actively transcribed regions of human genes [75] and likely plays a similar role in transcriptional elongation in human cells [72, 75]. In support of this model, H2Bub1 increases across the gene bodies of ER-α target genes GREB1 and TFF1 in human cancer cell lines on induction with estradiol [61, 76–78]. Overall, current evidence supports H2Bub1 as a critical component of active transcription within gene bodies but as a potent repressor of initiation in promoters and enhancers [79, 80].

2.3. Speed and synchrony of gene induction

Although all transcribed genes must go through all of the same steps to achieve successful transcription, genes differ in their transcriptional speed. Transcriptional speed is partly determined by the state of RNAPII at baseline, which can be described using the phrase “ready, set, go!”[42, 43, 81] At many genes, RNAPII is unbound at baseline (“not ready”), and must be recruited to the core promoter as part of the PIC [42, 43, 81]. However, RNAPII may also be bound to the core promoter at baseline but unmodified (i.e., a “ready” state) [42, 43, 81]. Alternatively, RNAPII may be bound in the poised state, in which serine 5 of the RNAPII CTD is phosphorylated and transcription has been primed (i.e., a “set” state) [42, 43, 81]. In this third instance, binding of a transcription factor does not trigger initial PIC assembly but rather results in phosphorylation of serine 2 on the RNAPII CTD, which triggers release of RNAPII into productive elongation (“go!”) [42, 43, 81]. Genes in the unready state at baseline are the slowest to respond, and genes in the set state are the fastest [82]. In addition, a target gene set in which all genes are all maintained in the set state at baseline can be activated with a high degree of synchrony [49], enabling a specific rapid response at the cellular level.

The genes that respond fastest to an initial stimulus are termed primary response genes (PRGs) [83]. PRGs can be induced without any new protein synthesis, so any required transcription factors or cofactors are already present in the cell at baseline [83]. PRGs can be subdivided into those genes with immediate response, termed immediate-early genes (IEGs), and those with delayed response but that still do not require new protein synthesis, termed delayed PRGs [48, 84]. PRGs often encode transcription factors and signaling molecules that, once translated, trigger transcription of the second wave of target genes (secondary response genes, or SRGs) [48]. IEGs have the fastest induction kinetics of known inducible genes, which is accomplished by pre-setting genes with poised RNAPII that is then released into elongation by the positive elongation factor P-TEFb which phosphorylates serine 2 on the RNAPII CTD [48, 85]. P-TEFb is recruited to IEG promoters by the bromodomain protein 4 (BRD4), which is itself recruited by the acetylation of histone H4 at lysine 16 (H4K16ac) [86]. H4K16ac is deposited by the histone acetyltransferase MOF, which is recruited by phosphorylation of histone H3 at serine 10 (H3S10ph) [86]. Therefore, the H3S10ph modification kicks off a sequence of events that results in pause-release of RNAPII from IEG promoters [86]. However, this modification is not present at IEGs at baseline [86]. Rather, IEGs are generally bivalent at baseline and contain the repressive modification H3K27me3 and the activating modification H3K4me2/3, as well as activating acetylated histones [87]. IEGs generally produce short transcripts and their promoters are enriched for specific transcription factors, including serum response factor (SRF), nuclear factor-κB (NF-κB), cyclic AMP response element-binding protein (CREB), and Zeste-like factor [84] Therefore, a likely sequence of events entails activation of one of the above transcription factors by a stimulus, binding of that transcription factor to the IEG promoter, recruitment by that transcription factor of histone modifying enzymes that remove H3K27me2/3 and deposit H3S10ph [86]. H3S10ph then recruits MOF, which deposits H4K16ac [86]. H4K16ac recruits BRD4 and P-TEFb, which triggers RNAPII pause-release and transcriptional induction [86]. Additional chromatin modifications, including H3S28ph [86], and the enzyme PARP-1 [88] have been linked to IEG activation, although their roles are unclear.

2.4. Chromatin looping controls the speed and intensity of transcription at inducible genes

Inducible genes and their respective enhancers are enclosed within distinct chromatin loops (<200kb) that are anchored by CTCF, within which smaller enhancer-promoter contact loops can form [89]. These larger chromatin loops support efficient induction, by bringing enhancers and promoters of inducible genes closer to one another in physical space, and they also restrict the enclosed inducible genes to activation by enhancers that are also present within the CTCF loop [89]. They also insulate inducible genes from neighboring housekeeping genes which are constitutively expressed and therefore constantly recruiting RNAPII and other transcriptional activators [89]. These CTCF loops enclose dynamic sets of inducible genes that are required for cellular function [89]. In primary cells, these loops contain genes involved in transcriptional regulation, cell motility and stem cell differentiation, but in immortalized cells, the genes responsible for these functions are not contained within distinct CTCF loops [89], which tracks with these genes’ differential regulation in immortalized cells [89]. CTCF is required for these loops and when CTCF is not present, inducible enhancers lose chromatin modifications indicative of active (H3K4me1 and H3K27ac) or poised (H3K4me1) enhancers [89, 90], indicating that these inducible enhancers are inactive in the absence of CTCF. This is an important result in light of general acceptance that knockdown of CTCF has little effect on global gene expression [91]. Although this is true in unstressed cells, the phenotypic effects of CTCF knockdown are apparent when the appropriate stimuli are applied. In fact, published case reports on three unrelated patients with mutations in CTCF showed reduced expression of enhancer-dependent inducible genes [92]. These case reports support a model in which inducible gene transcription can be initiated by promoters but not amplified by contact with enhancers in the absence of CTCF-mediated loops [92]. Notably, some inducible genes also show enhancer-promoter loops at baseline [93, 94]. This pre-setting of enhancer-promoter contacts further increases speed of target gene induction [93–95]. Loop pre-setting occurs both in undifferentiated cells during embryonic development [95] and in fully differentiated cells [96]. In both mouse and Drosophila embryonic stem cells, genes that are set to be activated at later stages of differentiation are pre-looped and associated with chromatin modifications characteristic of poised enhancers [95]. In differentiated human cells, TNF-α-responsive genes show pre-looped enhancers and promoters [96]. One of these target genes shows a unique pre-loop that connects the gene’s 5’ and 3’ ends and requires CTCF; this pre-loop is required for induction of this particular gene [96]. Since TNF-α is one of the transcription factors with motif enrichment in IEG promoters [48], these data suggest the possibility that pre-looping may be a general characteristic of IEGs. Chromatin loops can be actively disassembled to repress inducible genes [97]. For example, an enhancer-promoter loop that induces the Kit gene during development is later actively removed by replacing the loop-inducing transcription factor GATA-1 with the non-loop-inducing GATA-2 at the distal enhancer, which functions to repress Kit [97].

3. Transcriptional memory

Transcriptional memory describes a stable shift in a gene’s inducibility at the chromatin level [98, 99]. Cells can develop transcriptional memories of prior stress responses that alter either active transcriptional profiles or secondary inducibility profiles [98, 99]. Transcriptional memory suggests that an individual’s baseline chromatin state at CDS genes is determined by past exposures and will influence response to future exposures. Transcriptional memory can take four forms: (1) sustained activation of a gene even after exposure ceases (Figure 2); (2) sustained repression or silencing after an exposure ceases (Figure 2); (3) positive priming, in which a gene that responds to an initial hit of a stressor will respond either more quickly or more strongly to a second hit (Figure 2); or (4) negative priming, in which a gene that responds to an initial hit of a stressor will respond less strongly or not at all to a second hit (Figure 2).

Figure 2. Transcriptional memory takes several forms.

Inducible genes that are activated (or repressed) by a stressor respond rapidly but transiently to stressor. However, in some cases transcriptional memory of stress exposure forms at stress-responsive genes. Transcriptional memory can take one of four forms. In sustained activation, genes that are induced by a stressor can remain at high levels of transcription instead of returning to a baseline state. In sustained repression, genes repressed by a stressor remain repressed, instead of returning to a baseline state. In positive priming, genes induced by a stressor return to baseline after stress ceases, but induction is faster or stronger on re-stimulation. In negative priming, genes induced by a stressor return to baseline after stress ceases, but induction is decreased or absent on re-stimulation.

All four of these forms of memory can occur in transient or relatively stable forms [100–105], although the mechanisms governing which genes within relevant target gene sets develop memories and how long those memories last are unclear. Stable memories are not necessarily epigenetic, as they may not be inherited across cell division [106, 107]. However, there is some evidence that chromatin modifications can be mitotically inherited [108], providing a mechanism for epigenetic memory. Similar to inducible genes’ dependence on DNA sequence motifs, transcriptional memory formation is also influenced by underlying DNA sequence [99, 109, 110]. Importantly, transcriptional memory can be accomplished through non-chromatin mechanisms. For example, transcription factors can be post-translationally modified to increase their residence time on chromatin to increase transcription level and duration [111] and sequestration proteins can be overexpressed or their associations with transcription factors can be stabilized to inhibit a transcription factor’s ability to translocate to the nucleus or bind DNA [112]. In addition, transcriptional regulators can be deposited by a parent cell into daughter cells during mitosis to confer parental transcriptional states on daughter cell genomes [113].

Genes that are newly expressed by a stimulus usually return to baseline after removal of the stimulus [18]. In some cases, however, induced genes show sustained activation [100, 114, 115]. Sustained activation likely depends on the strength of the initial activation [100] and may vary even among cells within a tissue [100]. For example, individual cells that showed particularly strong initial transcriptional responses to doxycycline, hypoxia and DNA damaging agents showed sustained changes in gene expression, as well as growth rates and viability, for multiple cellular generations after the initial stimulus, but cells with lower initial transcriptional responses did not [100]. Sustained gene activation can be a result of persistent, aberrant de-repression of gene sets, as well. For example, loss of repressive marks H3K27me and H3K9me3 in Caenorhabditis elegans associated with developmental bisphenol A exposure is inherited for multiple generations [114]. In another example, loss of DNA methylation at the retroelement viable yellow Agouti associated with developmental bisphenol A (BPA) [116] or genistein [115, 116] exposure persists well into adulthood. However, it is not clear that these two examples represent memories that are triggered by an initial transcriptional event.

Genes that are newly repressed by a stimulus can show sustained repression after removal of the stimulus [101]. Sustained repression does not imply lack of responsiveness on secondary challenge, only that repression induced by an initial stressor is persistent. Over time, transient repression, which is characterized by the loss of activating histone modifications (e.g., H3K4me3, H3K9ac, H3K14ac) and the presence of repressive histone modifications (e.g., H3K27me3), can be stabilized into a fully heterochromatic silenced state, which is characterized by the addition of H3K9me3 and DNA methylation [117], which trigger 3D chromatin compaction [101]. Heterochromatin that is artificially induced via recruitment of the transcription factor heterochromatic protein-1 (HP-1) [118] or the repressive histone modification H3K9me3 [119] is epigenetically maintained, in that it can be inherited across cell division [101, 118, 119]. Sustained repression may even be inherited across generations. In C. elegans, the repressive modification H3K27me3 can be meiotically inherited [120] and newly repressed genes may remain repressed for many generations [120]. In Drosophila, repressive Polycomb group proteins (including the histone methyltransferase EZH2, which deposits H3K27me3) form repressive loops between an endogenous homeotic gene Abdominal-B and a cis-regulatory Polycomb response element elsewhere in the genome [121]. This repressive loop can be reconstituted between the response element and an artificial transgene and abolishment of the loop is sufficient to de-repress the gene [121]. De-repression of Abdominal-B is inherited by several Drosophila progeny, suggesting that loss (or gain) of sustained repressive memory may be meiotically heritable in metazoans [121]. Polycomb group proteins form similar repressive loops in humans [122]. Until recently, it was thought that humans did not have Polycomb response elements to serve as loop anchors; however, a recent study reported four human Polycomb response elements [123]. DNA methylation is often deposited at H3K27me3-modified heterochromatin, as well [124]. DNA methylation is a particularly attractive candidate for this form of memory, since there is a clear mechanism for its propagation during DNA replication [125]. DNA methyltransferase 1 (DNMT1) associates with the DNA replication machinery and adds new methyl groups to newly synthesized strands at locations that are methylated in the parent template strand [125].

Inducible genes that show positive priming are characterized on the 2D level by baseline increases in active histone modifications (e.g., H3K4me1/2/3) [102, 103] RNAPII occupancy and/or RNAPII poising [85, 103]. An increase in active histone modifications and RNAPII occupancy or poising generally increases the rate and intensity of transcriptional activation on secondary stimulus challenge [102, 103, 126]. Positive chromatin priming has been observed in repeated studies of immune system cells (reviewed in [127]). For example, human macrophages exposed to Candida albicans show increased H3K4me3 at pro-inflammatory genes and signal transducers of the immune response that lasts for a week after exposure and is associated with faster transcriptional response of those genes on secondary exposure [128]. H3K4me3 is newly deposited after activation of inducible immune genes that have no basal activity [129, 130]. However, limited studies in other cell types suggest that priming mechanisms are common across cell types in response to multiple stimuli. H3K4me3 is newly deposited at genes associated with prostate cancer in prostate tissue of rats following postnatal exposure to the endocrine disrupting chemical bisphenol A [131]. Subsequent to this initial postnatal exposure, these prostate cancer genes show elevated expression at baseline, as well as increased expression intensity in response to hormone in adulthood [131]. A series of studies in yeast provides more detail on a potential mechanism. Yeast develop positive priming at the inositol-1-phosphate synthase (INO1) gene by recruiting a remodeled form of the Set1/COMPASS histone methyltransferase complex that deposits only H3K4me2 and cannot deposit H3K4me3 [132]. The H3K4me2 modification recruits Set3, a component of the SET3C HDAC complex, which is in turn required for both RNAPII recruitment and H3K4me2 maintenance at INO1 [103]. In addition, a modified form of the transcriptional co-activator Mediator complex is required for the recruitment and poising of RNAPII at INO1 [103]. A Mediator complex that contains the Cdk8 kinase controls RNAPII recruitment and poising at the INO1 gene under memory conditions, but not under active conditions, suggesting a highly specific role for this remodeled Mediator complex [103]. In rare cases (or, at least, infrequently reported cases), positively primed enhancers that were previously not readily accessible to transcription factor binding can become newly and stably accessible, even in differentiated cells with established enhancer repertoires [133]. Fully differentiated human macrophages show sequential binding of stimulus-activated transcription factors (STAT1 and STAT6) and a lineage-determining transcription factor (Pu.1), followed by acquisition of active enhancer modifications, including H3K4me1 and H3K27ac, to previously inactive but not silenced enhancers in response to stimulus [133]. These new enhancers lose Pu.1 binding after stimulus ceases, likely because of a low affinity Pu.1 binding site that mandates cooperative binding of Pu.1 with stimulus-responsive transcription factors [133]. However, H3K4me1 is retained at ~30% of these enhancers and mediates a faster and stronger response on re-stimulation [133]. In addition, some regulatory elements that are active during embryonic development retain a hypomethylated state (low levels of DNA methylation) in fully differentiated cells, despite those elements’ inactivity in differentiated cells, which suggests a mechanism for reactivating developmental programs, given the right stimulus [134]. How does H3K4me1/2/3 accomplish positive priming? H3K4me may recruit histone acetyltransferases that accelerate transcriptional response [135]. Alternatively, H3K4me may not be responsible for priming, but rather the enzyme complexes that deposit or maintain it. In the above yeast example, Set3 recruits RNAPII to the INO1 gene under memory conditions [103]. In mouse embryonic stem cells, the MLL2 subunit of the MLL2/COMPASS complex can bind developmental genes and prevent their repression by Polycomb repressive complexes (PRCs) and DNA methyltransferases [136]. Although acetylated histones are generally lost when inducible genes are repressed following stimulus removal, there are some examples of retained histone acetylation at genes with positive priming. RNAPII and the HAT p300, as well as acetylated histones, are newly bound at baseline in some IEGs in T-cells after stimulation, and their binding correlates with faster re-activation of those genes [85]. Retained histone modifications can also increase the sensitivity of regulatory elements to a stimulus. For example, after initial stimulation, T-cells with newly bound RNAPII and the HAT p300 are able to respond to stimuli that previously were unable to induce a T-cell response [85].

Positive priming events can also occur on the 3D level, in the form of new chromatin loops or stabilized chromatin loops, and/or by association with the nuclear pore complex (reviewed in [127, 137]. In yeast, when inducible genes are activated by specific transcription factors, those genes form loops that link their 5’ and 3’ ends, and these loops move to the nuclear periphery and interact with proteins contained within the nuclear pore complex [98]. Heterochromatic regions containing silenced or repressed genes are often found on the nuclear periphery of eukaryotic cells and euchromatin is localized to the center of the nucleus [138]. Therefore, the nuclear pore complex represents an exception to this nuclear organization rule [139]. These induced genes generally return to the nucleoplasm when gene activation ceases, but in cases of positive priming, they remain looped and associated with the nuclear pore complex and more rapidly recruit RNAPII to their promoters, yielding faster and stronger gene transcription on secondary stimulus challenge [98]. Similar nuclear pore complex association is required for the induction of the Drosophila heat shock protein gene hsp70 [140], as well as in mammalian genes modified with acetylated histones [141], suggesting that localization to the nuclear pore complex is a common step in gene induction. An additional study in Drosophila demonstrate that the nuclear pore protein Nup98 has a distinct architectural role in forming enhancer-promoter loops at target genes [126] and a conserved role for Nup98 is apparent in humans, as well, where its presence is associated with H3K4 methylation and poised RNAPII [126]. Histone variants may play a role in 3D positive priming, as well. In yeast, the histone variant H2A.z is incorporated into the promoters of two genes, GAL1 and INO1, just after transcriptional activation [142]. These genes localize to the nuclear pore complex during activation, and H2A.z is required to retain them there after repression [142]. This continued association with the nuclear pore is required for faster reactivation of these genes on secondary challenge [142]. However, the roles of H2A.z and nuclear pore complex proteins in transcriptional memory are not supported by all evidence. At least one report shows that these chromatin factors do not control memory, but rather mitotically inherited transcription factors and a classic signaling feedback loop are responsible for apparent memory at the yeast GAL gene cluster [143]. The controversy about the relative roles of feedback signaling loops and structural chromatin features in transcriptional regulation is not a new one [107]. It may be possible that both phenomena are at play and must be carefully dissected mechanistically. An additional study in yeast showed that mild salt stress led to later resistance to hydrogen peroxide that persisted for four to five generations [144]. This resistance phenotype was entirely due to long-lived catalase that was synthesized during the initial salt stress and distributed to daughter cells [144]. However, two forms of memory may operate simultaneously; in a separate effect from hydrogen peroxide resistance, these cells also showed faster gene expression response at stress-induced genes, which response required the nuclear pore protein Nup42p [144].

Inducible genes that show negative priming show baseline loss of active histone modifications [104, 105]. Negative priming is sometimes referred to as transient silencing, although this alternate term does not capture the lack of responsiveness to secondary challenge. A classic example of negative priming is endotoxin tolerance, in which target genes have a refractory period after an initial exposure to a pathogen or pathogen-like (i.e., lipopolysaccharides, or LPS, present on pathogen cell membranes that stimulate immune responses) [104, 105, 145]. Negative priming likely avoids adverse consequence of exaggerated, chronic or repeated activation of target gene sets, but it can also present serious consequences. For example, human monocytes in people recovering from sepsis or other immune trauma cannot respond to a subsequent pathogen challenge for a substantial period of time, often several days, placing these patients at risk for mortality due to immunosuppression [104, 145]. Endotoxin tolerance is associated with suppression of inflammatory cytokines, including IL-1β, and upregulation of negative regulators of the inflammatory response, including the IL-1R-associated kinase-M (IRAK-M) which regulates TLR4 signaling, soluble triggering receptor expressed on myeloid cells-1 (TREM-1), and suppressor of cytokine signaling 3 (SOCS3), among others, which temporarily block an inflammatory reaction to LPS and other immune stimuli (reviewed in [145]). Endotoxin-tolerant human monocytes that suppress IL-1β do not show the increase in induction-associated H3S10ph and decrease in repressive H3K9me3 at this gene that endotoxin-sensitive cells do in the presence of LPS [146]. This negative priming effect extends to a full set of inflammatory cytokine genes [105]. Specifically, the SWI/SNF remodeler BRG1 was not recruited to these genes in response to LPS in tolerant cells and these genes show decreased baseline activating histone modifications (total H4 acetylation, H3K4m3) and decreased chromatin accessibility; however, these genes do not carry histone modifications that indicate gene silencing (H3K9me3, H3K27me3) [105]. Instead, repression is mediated by the high mobility group box 1 protein (HMGB1) and the linker histone H1 [147]. Specifically, HMGB1 binds the TNF-α promoter and promotes assembly of the RelB repressor complex [147]. H1, which displaces HMGB1 from chromatin under baseline conditions, co-binds with HMGB1 to the repressed promoter [147]. HMGB1 and H1 are recruited independently to the promoter and their binding requires H3K9me2 [147]. This mechanism is gene-specific; a similar repression is seen at the pro-inflammatory IL-1β promoter, but not at the anti-inflammatory IκBα promoter [147]. In addition, IEGs, delayed PRGs and SRGs showed differences in chromatin remodeling requirements during endotoxin tolerance [148]. Under normal conditions, chromatin remodeling by SWI/SNF is required for activation of SRGs and delayed PRGs, but not IEGs [149]. The Mi-2β remodeling protein co-binds with SWI/SNF at some of these genes to repress their activation, providing an alternate mechanism for negative priming in endotoxin tolerance [149]. It is interesting to note that endotoxin tolerance also involves positive priming of antimicrobial genes, which were characterized by higher baseline and inducible histone acetylation and faster BRG1 recruitment [105], highlighting that multiple types of transcriptional memory may regulate a biological response.

4. CDS pathways respond to redox-active pollutants

CDS pathways sense shifts in cellular homeostasis caused by redox-active pollutants and response by activating inducible CDS target genes [13]. In this section, I will briefly review each CDS pathway, how it senses cellular stress, the sequence of events required for induction of its target genes and evidence for transcriptional memory formation in response to pathway stimulation. Although I have not reviewed them here, redox-active pollutants can trigger induction and transcriptional memory at genes not within CDS pathways (reviewed in [150]) and redox-active pollutants can also affect chromatin state through general mechanisms that do not specifically target CDS genes [151].

4.1. Stress-activated p38 MAP kinase signaling

In addition to specific CDS responses, mammals have a general stress response pathway governed by p38 mitogen-activated protein kinases (MAPKs) that can respond to a wide range of stressors (reviewed in [152]). (P38-MAPKs also have many functions in the cell that are unrelated to CDS pathways, including involvement in immune and inflammatory responses [152].) Much of our understanding of the mammalian general stress response is based on research on the yeast general stress response [18]. Unlike humans, yeast only have this general stress response system (in yeast, called the Environmental Stress Response or ESR [153].) The ESR is activated by a single stress-activated MAPK, Hog1, that regulates expression of a general set of ESR target genes (~300 genes are induced and ~600 are repressed, although increasing stress leads to induction of a larger gene set) [154]. The classic Hog1 inducer is osmotic stress, although the ESR responds to a variety of stressors (reviewed in [15]). Hog1 is rapidly activated by phosphorylation after osmotic stress, after which it translocates to the nucleus [153]. Hog1 affects transcription both through transient phosphorylation and subsequent activation of transcription factors and by physically associating with target gene promoters [153, 155]. Hog1 is tethered to chromatin by its interaction with transcription factors, which enables Hog1 to phosphorylate other transcriptional co-activators [153]. For example, Hog1 can activate the Sko1p-Cyc8p-Tup1p complex to recruit the HAT complex SAGA, the HDAC Rpd3, SWI/SNF chromatin remodelers, and RNAPII [155–157]. However, chromatin remodeling may not be required for stress gene induction [158], suggesting that Hog1 binding to gene promoters may only be necessary under some conditions. When H3K4 methylation, particularly H3K4me1, is present at stress gene promoters, chromatin remodeling is required for induction [158]. When this modification is removed, SWI/SNF is no longer required for target gene induction and alternative remodeling complexes, including the Swr1 complex that deposits H2A.z, can bind to these genes [158]. These data suggest that yeast stress response genes are controlled by two remodeling mechanisms: SWI/SNF remodels gene promoters in the presence of H3K4 methylation and Swr1 deposits the unstable histone H2A.z in its absence [158], which may promote accessible chromatin in the absence of active remodeling.

The mammalian general stress response is much more restricted than the yeast response [159]. In mammals, ~100–150 genes are rapidly upregulated on heat stress, osmotic stress and oxidative stress but the common response is restricted to ~30% of them [152]. Human p38 MAPK has four isoforms (p38α-δ, or MAPK11–14, sometimes called stress-activated kinases or SAPKs); p38α is the isoform most commonly activated in mammalian stress response [152]. Outside of the common response, p38α interacts with specific CDS responses [152]; these interactions are highlighted in later sections of this review. A significant fraction of specific mammalian CDS responses rely on p38α [159]. For example, p38α is recruited to target genes in response to osmotic shock, anisomycin, or TNF-α activation by their respectively activated transcription factors, and p38α is then required for RNAPII recruitment [157]. In another example, exposure of mammalian cells to heat shock and oxidative stress for 30 minutes induced around 100 genes [160]; exposure of mammalian cells to osmotic stress, the inflammatory cytokine TNF-α and the protein synthesis inhibitor anisomycin for 45 minutes induced more than 120 genes; the majority of both gene sets were dependent on p38α, including a set of common response genes (~20% of total target genes) activated by different stimuli [159]. Like Hog1, p38α can be recruited to target gene promoters and recruit SWI/SNF complexes [161]. P38α phosphorylates mitogen- and stress-activated protein kinases MSK1/2, which in turn phosphorylate both the chromatin architectural protein high mobility group protein 14 (HMG-14) and serine 10 on histone H3 (H3S10ph), as well as the IEG transcription factors CREB and NF-κB to trigger stress gene transcription [162]. MSKs themselves can be recruited to IEG promoters, as well, to initiate transcription [163]. P38α also phosphorylates MAP kinase-activated protein kinase 2 and 3 (MK2/3), which phosphorylate the IEG transcription factors CREB and SRF [164]. Last, p38α directly phosphorylates the transcription factors p53 and ATF6, which have distinct roles in CDS pathways (DNA damage response and unfolded protein response, respectively) (reviewed in [152]). P38α associates both with regulatory elements and along the entire length of transcribed gene bodies [165, 166], suggesting that it also plays a role in transcriptional elongation of stress response genes.

4.2. Heat shock signaling

The heat shock response is an evolutionarily conserved cellular defense that is triggered by proteotoxic stress caused by protein misfolding or denaturation and subsequent protein aggregation [167]. The heat shock response was first described in 1962 by Ferrucio Ritossa, who discovered that increased temperature induced chromosomal puffs to form in the large chromosomes of the salivary glands of Drosophila busckii [168]. These puffs were later identified as regions of actively transcribed genes encoding heat shock proteins, which are protein chaperones that support protein folding and clear protein aggregates [168]. Current data show that the heat shock response is characterized by more complex transcriptional changes than the upregulation of a small number of heat shock proteins [82]. The mammalian heat shock response triggers the rapid induction of several hundred genes and repression of several thousand genes [169]. Genes that are upregulated are required for maintaining cellular homeostasis, including heat shock proteins and cytoskeletal proteins required to maintain cellular integrity and cellular transport [170].The cell downregulates all other transcription globally, including constitutively expressed genes that control cell cycle, general transcription and metabolism, to prevent misfolding of newly synthesized proteins and an exacerbation of proteotoxicity [169]. Under heat shock conditions, ubiquitination of misfolded nascently translated proteins is detected by p38α which then translocates to the nucleus and directly binds the promoters of repressed genes [169]. In addition to sensing denatured proteins, increased temperature is sensed directly via thermosensory structures in DNA, RNA, proteins and lipids. DNA topology can be altered, RNA hairpins can melt, heat shock proteins can shift conformations, and ion channels are activated in mammalian cell membranes [171–173]. A sustained heat shock response can be harmful to the cell. In Drosophila, induction of the transcription factor heat shock factor (HSF) is restricted within minutes of heat stress, and when this restriction is lost (during HSF overexpression or overexpression of HSF targets), cells show reduced fitness and compromised viability [174].

4.2.1. Redox-active pollutants activate the heat shock response

Proteotoxicity can be triggered by stimuli other than heat. For example, heavy metals and metalloids (e.g., lead, mercury, cadmium, arsenite) can bind thiol groups within proteins, disrupting their structures and function, causing proteotoxicity [150, 175]. In addition, any pollutant that increases reactive oxygen species in the cell can cause damage to proteins, triggering a heat shock response [176] (see Oxidative Stress Response.) Heat shock signaling occurs in response to misfolding of nascently translated proteins, rather than existing proteins present in the cell [169]. This suggests that the heat shock response is triggered partly by misfolding of proteins upregulated by pollutant exposure, rather than by inactivation of proteins present in the cell at the time of exposure.

4.2.2. Heat shock response in the 2D genome

At the 2D genome level, several heat shock transcription factors orchestrate a rapid change in global transcription in the cell in response to proteotoxic stress [14, 169]. The majority of upregulated heat shock response genes are activated by increased release of poised RNAPII from target gene promoters [14, 169]. This release is directly triggered by the positive elongation factor P-TEFb, which phosphorylates and releases negative elongation factors NELF and DSIF, as well as phosphorylating serine 2 on the CTD of RNAPII [169]. P-TEFb is recruited to target gene promoters when transcription factors bind [169]. This regulation implies that the initial steps of transcriptional activation (increased chromatin accessibility at the core promoter, assembly of the PIC, poising of RNAPII) all occur at baseline, prior to proteotoxic stress exposure. Unlike heat shock gene promoters, heat shock enhancers are generally unbound by RNAPII at baseline, and their chromatin accessibility is bookmarked by pioneer transcription factors (GAGA-associated factor, or GAF, in Drosophila, and GATA-2 and TAL1 in humans) [14, 18]

A recent time-course study in mouse embryonic fibroblasts demonstrates that three groups of genes are activated, and two groups of genes are repressed, in response to heat shock, but they differ in their transcription factor control and timing [82]. Heat shock factor 1 (HSF1), originally considered the “master regulator” of the heat shock response [177–179], controls only a small number of heat shock-induced genes, primarily heat shock proteins and other chaperones, by increasing release of poised RNAPII from heat shock promoters [82]. These genes are upregulated rapidly (within 2 hours of heat shock) and maintained at a high transcriptional level for at least four days (60 hours) [82]. The transcription factor SRF induces IEGs, primarily cytoskeletal genes, also by triggering release of poised RNAPII from target gene promoters [82]. A subset of these genes is activated rapidly (within 2.5 hours of heat shock), and another subset is activated more slowly (within 12 hours of heat shock) [82]. Both subsets of genes are activated only transiently; all genes return to baseline by the 60-hour mark [82]. A third group of genes, primarily apopotic genes, is upregulated, likely by the transcription factor NRF2 [82]. This group of genes is also induced by increased release of poised RNAPII, but their induction occurs late after heat shock [82]. Although all three of these groups of induced genes are primarily regulated by increased release of poised RNAPII, there is some evidence that all three are also regulated to a lesser degree by an increase in new RNAPII recruitment and PIC formation [82]. For example, HSF1 can recruit the chromatin remodeling complex SWI/SNF to mammalian target gene promoters to enable PIC binding (reviewed in [14]). The majority of the heat shock response involves global transcriptional repression of active genes, classified into two groups: those involved in cell cycle and metabolism (repressed early, within 2.5 hours of heat shock) and those involved in mRNA processing (repressed late, within 12 hours of heat shock) [82]. Both groups show sustained repression for at least four days (60 hours) [82]. It was previously thought that HSF1 was responsible for transcriptional repression, too, by binding gene bodies and creating a physical obstacle to transcript elongation [180], but current evidence shows that HSF1 bound at intragenic (and intergenic) sites is more likely binding to enhancers [180]. Transcriptional repression is accomplished by reduced release of poised RNAPII, mediated by increased recruitment of the negative elongation factor NELF to target gene promoters [82], in some cases by p38α [169]. Heat shock also activates transcription of non-coding transposable elements (Alu in humans, SINE B2 in mice) that prevent initiation of transcription on some repressed genes by interfering with PIC formation [181–183].

Histone modification by the chemical groups PAR (PARylation) and SUMO can promote heat shock transcriptional responses [184–186]. PARylation is a stress-inducible histone modification deposited by the poly[ADP-ribose] polymerase 1 (PARP-1) in the promoter and gene body of the target heat shock protein gene HSPA1A on heat shock [148, 184]. PARylation promotes increased chromatin accessibility and allows HSF1 binding to the gene promoter [184, 187]. SUMO is a complex of small ubiquitin-like modifier proteins that is increased at the promoters of highly transcribed genes and enhancers during heat shock [185, 186, 188]. SUMO restricts the activity of these genes and promotes their repression [185]. Both positive (P-TEFb) and negative (NELF) elongation factors are also modified with SUMO groups, which targets them for proteasomal inhibition [185]. This result suggests that SUMOylation of elongation factors can affect release of poised RNAPII, although not in a single clear direction. Heat shock triggers a decrease in SUMOylation of regions containing CTCF-binding sites, which suggests a role for SUMO in TAD shifts during heat shock (discussed below) [185].

In addition to programmed gene responses, heat shock (among other forms of stress, including oxidative and osmotic stress), leads to failed transcriptional termination in several genes [189, 190]. The resulting transcripts are called “downstream of gene-containing transcripts” (DoGs), because they include readthrough transcription that can span thousands of base pairs downstream of target genes’ ends [189, 190]. The function of DoGs is unknown, although they may be needed for nuclear scaffold maintenance during stress [189, 190].

4.2.3. Heat shock response in the 3D genome

At the 3D genome level, both TAD reconfiguration and formation of chromatin loops that bring enhancers in contact with heat shock target gene promoters are important parts of the heat shock response [191, 192]. The involvement of enhancers in the heat shock response provides a model for tissue-specific heat shock transcriptional programs, either through differential target gene sets, or differential speed, intensity or stability of target gene responses. Pluripotency factors OCT4, NANOG and KLF4 are involved in loop formation between heat shock enhancers and target genes, as well as target gene regulation, in human embryonic stem cells [191], suggesting that developmental timing of proteotoxic stress may influence the transcriptional response, as well. Higher order TADs are reorganized during the heat shock response by re-localization of CTCF to new sites and formation of stress-induced TADs in human embryonic stem cells [191]. New TADs enable new enhancer-promoter contacts that were not possible in baseline TAD conformations; new TADs also disrupt baseline enhancer-promoter contacts, which may result in gene repression [89, 137]. In contrast to the human embryonic stem cell data, Drosophila cells exposed to heat shock do not disrupt baseline loops and form new ones, but rather re-localize CTCF and cohesin from TAD boundaries to enhancers and promoters contained within the original TADs, which weakens TAD boundaries and increases inter-TAD interactions [193]. These inter-TAD interactions function to repress certain genes during heat shock, by enabling new interactions between re-localized cohesin and the repressive Polycomb complex [193]. In addition to TAD dynamics, heat shock protein genes located on different chromosomes in yeast can physically interact with one another through HSF1-mediated contacts, in a function independent from its role as a transcription factor [192]. These inter-chromosomal contacts may support formation of transcriptionally active foci, or transcription factories, to drive synchronous induction of heat shock target genes [192]. These foci suggest that heat shock response requires phase separation [194–196], or formation of biochemically distinct nuclear sub-compartments, to accomplish a rapid, synchronous response.

4.2.4. Transcriptional memory in heat shock response

Heat shock signaling can lead to transcriptional memory in plants and animals [173, 197, 198]. In plants, temperature shock leads to H3K4me2/3 marking of induced heat shock genes that is associated with hyper-induction on secondary heat shock [25]. This memory requires the transcription factor HSFA2, which appears to be dedicated to memory formation, since it has no role in initial induction [25]. Heat shock leads to transcriptional memory in rodents and chicks [197, 198]. Chick exposure to heat stress three to five days after hatching leads to long-lasting transcriptional changes at the heat shock protein hsp70 in chick anterior hypothalamus [197]. Harsh heat stress triggered both a sustained induction response and a positive priming response. Specifically, harsh heat yielded an immediate 13.7-fold increase in hsp70 transcription which increased to 16.5-fold induction at 6 hours and remained high (4.8-fold over baseline) for the next 18 hours [197]. In contrast, mild heat stress triggered a negative priming response. Specifically, mild heat yielded only a 2.3-fold initial increase in hsp70 transcript levels that remained stable through 24 hours and then returned to baseline levels [197]. On secondary challenge with mild heat, chicks conditioned with harsh heat showed a stronger hsp70 expression response as compared to mild heat-conditioned chicks (3.1-fold vs 1.7-fold increase after 6 hours), indicating a positive priming response in the harsh heat group, but the mild heat group showed a blunted transcriptional response, as compared to its initial response to a heat stress of comparable intensity [197]. These data suggest that mild heat triggered conditioning, in which the chick acclimatized to a higher heat environment, but that harsh heat triggered a stress response. The harsh heat-conditioned group showed increased DNA methylation at repressive elements within the hsp70 promoter; this increased DNA methylation decreased binding of repressive transcription factors, as well as a chromatin remodeling complex containing a histone deacetylase [197]. Therefore, this increased promoter DNA methylation led to increased promoter histone acetylation at baseline, which is responsible for higher inducible hsp70 transcription in high heat-stressed chicks [197]. In rodents, mild heat stress triggered a positive priming response. Specifically, mild heat stress in rats led to transiently increased H3S10ph but sustained increase in histone H4 acetylation in the promoters of heat shock elements of heat shock genes hsp70 and hsp90 [198]. In this case, H4 acetylation levels predicted faster transcription of these genes on secondary challenge [198]. The authors hypothesize that the initial H3S10ph is required for deposition but not maintenance of histone acetylation [198]. This response extends to physiological increases in heat that occur during physical exertion. Specifically, monocytes isolated from female mice 30 days after exertional heat stress showed changes in promoter DNA methylation and increased intensity of transcription of heat shock genes following secondary heat shock [160]. Together, these data underscore the complexity of cellular responses to differential exposure duration and stressor intensity, as well as the importance of standardized stress exposure designs to clarify response patterns.

4.3. Oxidative stress signaling

The oxidative stress response is caused by reactive oxygen species (ROS) that disturb cellular redox status by oxidizing proteins, lipids, and nucleic acids (reviewed in [199]). A majority of endogenous cellular ROS is generated as a byproduct of mitochondrial oxidative phosphorylation (OXPHOS), but ROS is also generated by NADPH oxidases and during protein folding in the endoplasmic reticulum (ER) [199, 200]. ROS include hydrogen peroxide, superoxide radicals and hydroxyl radicals [199, 200]. ROS can serve as signaling molecules and also as undesirable cytotoxic byproducts of oxidative phosphorylation in mitochondria and other metabolic pathways in the peroxisome [199]. The mammalian cell’s oxidative stress response is primarily regulated by the NRF transcription factors, NRF1, NRF2, and NRF3 [201], although most research has focused on NRF2 [200]. The classic regulatory mechanism for NRF2 involves the Kelch-like ECH-associated protein (Keap1) [202]. At baseline, NRF2 is sequestered in the cytoplasm through association with Keap1, which makes NRF2 accessible to ubiquitination by the E3 ubiquitin ligase Cullin 3-Ring-box protein 1 (CUL3) and subsequent proteasomal degradation [112, 202–204]. When the cellular environment becomes more oxidative, cysteine residues on Keap1 are oxidized, which prevents NRF2 ubiquitination, resulting it NRF2 accumulation, nuclear translocation and chromatin binding to antioxidant response elements (AREs) in target gene promoters and enhancers [112, 202, 203, 205]. These target genes include antioxidant enzymes and enzymes required for biosynthesis of glutathione, the cell’s primary ROS scavenger (reviewed in [14]). Knockout of NRF2 increases susceptibility to many chemical pollutants and worsens diseases associated with oxidative stress (reviewed in [200]). Conversely, biochemical or genetic increases in NRF2 protects the cell against oxidative damage [200]. However, NRF2 activation can harm the cell, as well (reviewed in [206]). NRF2 signaling is associated with increased cisplatin resistance in certain cancers (reviewed in [150]). Oxidative stress signaling requires additional transcription factors, specifically members of the forkhead box O (FOXO) transcription factor family: FOXO1, FOXO3, FOXO4 [14]. FOXOs are activated by post-translational phosphorylation by upstream signaling molecules (JNK and MST1) and translocate to the nucleus, where they trigger transcription of specific antioxidant genes [14]. Other cellular signaling pathways, including NF-kB and JUN, may play a role in regulation of certain oxidative stress target genes (reviewed in [14]). Oxidative stress can also cause indirect effects by modulating levels of epigenetic substrates and mitochondrial metabolites/second messengers that themselves affect gene expression profiles (reviewed in [151, 207]). Importantly, ROS do not always cause oxidative stress and have independent functions as redox signaling molecules in cellular homeostasis and organismal development (reviewed in [199, 208]).

4.3.1. Redox-active pollutants activate the oxidative stress response

A large number of chemical pollutants can increase cellular ROS during their metabolism by cytochrome P450 proteins, by binding reactive thiol groups on certain proteins and inactivating them, by redox cycling, or by triggering mitochondrial dysfunction (reviewed in [150]). For example, non-dioxin-like polychlorinated biphenyls (PCBs) generate ROS as byproducts of metabolism by cellular cytochrome P450 enzymes [209]. Other pollutants (e.g., lead, cadmium, mercury) bind thiol groups within antioxidant defense proteins that protect the cell from free radicals like ROS or reactive nitrogen species (RNS), deactivating these proteins and causing free radical levels to rise passively in the cell (reviewed in [210]). In addition, certain toxicants (e.g., iron, cobalt, copper, and chromium) can redox cycle, or alternately function as electron acceptors and donors, which disrupts the stability of cellular proteins and lipids [210]. Last, select toxicants (e.g., lead, mercury) also accumulate in mitochondria, which have high ROS levels due to OXPHOS [210]. It is unclear whether metals directly trigger mitochondrial dysfunction, leading to ROS overproduction and leakage into the cytosol, or simply amplify their effects in an organelle rich in ROS scavengers, like glutathione, and antioxidant defense enzymes, including glutathione peroxidase and superoxide dismutase [210]. Some pesticides are known to cause mitochondrial dysfunction by inhibiting one or more complexes in the electron transport chain, which leads to electron leakage and ROS formation, with an eventual increase in cellular ROS [150].

4.3.2. Oxidative stress response in the 2D genome

During oxidative stress, NRF2 target genes are induced by new chromatin accessibility and PIC formation at the core promoters [211]. However, a subset of target genes show poised RNAPII at baseline and these genes are induced through increased release of poised RNAPII [211]. Unlike HSF1, which recruits the positive elongation factor P-TEFb to release RNAPII, during oxidative stress, the negative elongation factor NELF dissociates from chromatin and enables RNAPII release even in the presence of a P-TEFb inhibitor, suggesting that mechanisms of RNAPII pause-release may vary by stressor [211]. Chromatin remodelers can be recruited to oxidative stress target genes by DNA structures. For example, the NRF2 target gene heme-oxygenase 1 (HO-1) forms Z-DNA in response to NRF2 binding, which in turn assists in recruiting the SWI/SNF remodeler BRG1, which is required for activation of this particular gene [212]. However, chromatin remodeling complexes can also restrict oxidative stress gene expression. Decreased expression of SWI/SNF family remodelers BRG1 and BRM is correlated with increased expression of some NRF2 target genes [213]. This result indicates that SWI/SNF restricts NRF2 binding at some sites via local nucleosome remodeling that restricts accessibility. Higher order chromatin structure may also function in repressing a NRF2 response. For example, the high-mobility group box 1 (HMGB1), a chromatin architectural protein, suppresses NRF2 transcription, as well as transcription of NRF2 target genes, in response to hydrogen peroxide in human melanocytes [214]. Although the majority of the oxidative stress response is regulated by NRF2 directed to AREs, some target genes have alternate mechanisms of activation. At least two oxidative stress target genes contain DNA sequences within their proximal promoters that serve as oxidative damage sensors that trigger the genes’ activation [215]. Oxidative damage to DNA forms 8-oxo-7,8-dihydroguanine residues [215]. When these damaged residues occur within a G-quadruplex sequence in the proximal promoter elements of the vascular endothelial growth factor (VEGF) and the endonuclease III-like protein 1 (NTHL1) genes, a component of the DNA repair machinery, OGG1, is recruited to the damaged promoter elements [215] where it directly recruits transcription factors and RNAPII to the promoter and stimulates transcription of the genes [216]. The heat shock response transcription factor HSF1 is also activated in an oxidative cellular environment [150]. At baseline, HSF1 is sequestered in the cytoplasm by the heat shock proteins HSP90 and HSP70 [150]. Both of these heat shock proteins contain reactive cysteines that can be oxidized during a shift in cellular redox status [150]. When oxidized, HSP90 and HSP70 dissociate from HSF1, and HSF1 translocates to the nucleus and triggers gene expression, although it is unclear whether HSF1 mounts separate transcriptional programs during oxidative stress and heat shock responses [150].

Oxidative stress can directly damage the genome and the 2D epigenome (reviewed in [151, 217]). Peroxynitrite can nitrate tyrosines on histones, which increase their thermostability and possibly protects DNA from oxidative damage [218–220]. ROS promotes the production of reactive carbonyls that form adducts with reactive cysteines, arginines, lysines and histidines on proteins, including abundant histones; these adducts can alternately stabilize [221] or destabilize [222, 223] histones, cause histone loss from chromatin [224], or prevent programmed modifications at residues blocked by adducts [225, 226]. Adduct formation occurs more easily in free histones, which cannot be reincorporated into chromatin if they cannot be modified by chromatin remodeling enzymes [225]. Histones can also be modified with glutathione during oxidative stress [227]. Guanine residues within DNA can be oxidized, which can block binding of methyl binding proteins [228], and methyl groups bound to DNA (5-methylcytosine) can be directly oxidized to 5-hydroxymethylcytosine, which is an intermediate in the active DNA demethylation pathway and can lead to global loss of DNA methylation [229]. The methionine adenosyltransferase that catalyzes formation of the methyl donor S-adenosylmethionine required for methylation of cytosines in DNA can itself be oxidized and deactivated, leading to a passive decrease in DNA methylation through a second pathway [230, 231]. In contrast, TET DNA demethylases can also be oxidized and inactivated, which can lead to global increases in DNA methylation [232]. The DNA methyltransferase DNMT1 and the HMT EZH2 is recruited to sites of oxidized DNA damage, resulting in localized increases in DNA methylation and H3K27me3 [233, 234]. HMTs and HDACs [235, 236] can also be inhibited by oxidative stress [237] and DNA and histone demethylases can be inhibited by lack of methyl donor availability [151, 238]. These responses may cause specific effects on target gene sets or global, non-programmed effects that are a function of enzyme kinetics and substrate specificity, rather than gene functionality. One example of a specific effect is inhibition of histone deacetylases that are components of co-repressor complexes bound at baseline to promoters of redox-sensitive transcription factors [236]. Release of this repression by inhibition of these enzymes permits transcription of these transcription factors [236].

4.3.3. Oxidative stress response in the 3D genome

The chromatin remodeler Cockayne syndrome group B protein (CSB), a SWI/SNF family member, interacts with CTCF bound to chromatin under oxidative stress conditions and forms long-range chromatin loops in which CSB is bound to the regulatory element and CTCF to its sequence motif [239]. CSB deficiency increases cells’ sensitivity to ROS, suggesting that this remodeler is particularly important in the oxidative stress response [239]. CTCF deficiency also increases cells’ sensitivity to ROS [240], highlighting the importance of long-range chromatin loop formation or maintenance in mounting an effective oxidative stress response. Expression of the insulator and chromatin architectural protein CTCF is decreased in oxidative stress conditions (via NF-κB signaling, not NRF2), which leads to loss of insulator function and appropriate mono-allelic expression at the imprinted locus H19/IGF2 [241].

4.3.4. Transcriptional memory in oxidative stress response

Transcriptional memory in the oxidative stress response occurs in yeast, worms, fish and rodents [144, 154, 242–245]. Yeast cells conditioned with salt stress show positive priming at 77 genes induced by challenge with oxidative stress [144, 154, 244]. This memory persists for four generations and requires the nuclear pore protein Nup42 [144, 154, 244]. Stochastic increases in physiological levels of endogenous ROS during early development decrease global H3K4me3 and increase stress resistance and lifespan in adult C. elegans [245]. This study indicates that mild environmental exposures may be sufficient to trigger transcriptional memory. This finding is particularly interesting in light of another example that involves the transcription factor aryl hydrocarbon receptor (AhR). AhR activates its target gene cytochrome P450 1A (CYP1A) in response to PAH or dioxin [243]. In mice with a transient adult dioxin exposure, CYP1A shows positive priming that persists for forty days post-exposure [243]. However, in Atlantic killifish chronically exposed for multiple generations to PAH pollution, CYP1A loses its ability to be induced by AhR, but regains responsiveness within one generation of clean-water rearing [242]. This result suggests either negative priming or sustained repression that is lost within one generation [242]. In the AhR-CYP1A example, positive priming is the initial protective response to stress in mice and negative priming appears to be the later response to chronic stress in killifish. However, in the low-level ROS study in C. elegans, ROS levels are mild and insufficient to induce stress. The decrease in H3K4me3, if it confers negative priming, suggests differential memory formation pathways in mild, non-stressful exposures that may be protective and adaptive.

4.4. Hypoxia signaling

Healthy mammalian cells require oxygen to generate energy through OXPHOS [246]. Normal tissue oxygen levels are referred to as a state of normoxia [246]. Hypoxia refers to a state of decreased cellular oxygen levels, which leads to diminished energy production [246]. In order to survive this oxygen reduction, cells activate transcription of genes that restore oxygen levels, as well as downregulating genes related to energy production [246]. Transcriptional activation triggered by hypoxia is controlled by transcription factors called hypoxia inducible factors (HIFs), including HIF-1α, HIF-2α, HIF-3α, and their common dimerization partner, HIF-1β [246]. HIF-1β is constitutively expressed, and the levels of the other three HIFs are regulated by cellular oxygen levels [247]. HIF-1α is the best studied of the HIFs [248]. HIF-1α is prolyl hydroxylated during normoxia, which triggers ubiquitination by the E3 ubiquitin ligase von Hippel-Lindau protein (pVHL) and lysosomal degradation [248]. Under hypoxic conditions, this hydroxylation is inhibited, and HIF-1α dimerizes with HIF-1β, translocates to the nucleus, and binds to hypoxia response elements (HREs) in HIF-1α target genes [248]. HIF-1α is stabilized by PARylation deposited by PARP-1, which facilitates HIF-1α binding to chromatin [249]. HIF-2α and HIF-3α are similarly regulated by pVHL and activate separate transcriptional programs, although there is some overlap in target genes [246]. Although hypoxia does induce a subset of target genes, hypoxia primarily represses genes to reduce energy demanding processes and redirect limited resources to housekeeping functions in the cell [247]. At least ten different transcriptional repressors have bene reported for HIFs, including Repressor Element-1 Silencing Transcription Factor (REST) [246]. About 20% of genes repressed in hypoxia are downregulated by REST [250].

4.4.1. Redox-active pollutants activate the hypoxia response

Mercury and cadmium can exert toxicity by inhibiting protective hypoxia signaling even under normoxic conditions [251–255]. One class of HMTs that contain Jumonji C domains require oxygen for optimal function and can be inhibited by heavy metals/metalloids (e.g., cadmium, arsenic, chromium, nickel), so exposure to metals in the presence of cellular hypoxia likely decreases histone methylation levels (reviewed in [256, 257]). In some cases, toxicant action is potentiated by existing dysregulation of hypoxia signaling. For example, endocrine disrupting chemicals (EDCs), including BPA, benzyl butyl phthalate (BBP), and di(2-ethylhexyl) phthalate (DEHP) decreased activation of ER-α mediated transcription in the presence of hypoxia in human breast cancer cells [258]. This suggests that EDCs can potentiate cancer pathology associated with hypoxia [258].

4.4.2. Hypoxia response in the 2D genome

Genes induced by hypoxia generally display basal expression levels that are enhanced under low oxygen conditions [248] and therefore are bound by activating histone marks, like H3K4me3, at baseline [and by high levels RNAPII in core promoters and moderate levels of RNAPII along gene bodies [259], indicating active transcription. HIFs may increase pause-release of RNAPII at target genes, based on more equal levels of RNAPII binding at promoters and along gene bodies seen under hypoxic conditions [259] and based on evidence that HIFs recruit CDK8 and TIP60, which promote elongation [260, 261]. The evidence for widespread changes in chromatin accessibility in hypoxia is mixed: some studies report globally decreased [262, 263] chromatin accessibility and others report no changes or changes limited to specific target gene promoters [264].

Hypoxia leads to changes in DNA methylation but not in one clear direction. For example, in some reports, DNA methylation is globally increased in cancer cells under hypoxic conditions (by inhibiting TET activity) [265], but in others, DNA methylation is globally decreased [266]. These discrepancies may be due to experimental techniques that do not account for DNA hydroxymethylation, which is an intermediate in the DNA demethylation pathway that also serves as a regulatory modification [267]. However, there is mixed evidence for hypoxia effects on TET enzymes, which catalyze oxidation of DNA methylation to DNA hydroxmethylation [267, 268]. One study reports that DNA hydroxymethylation was reduced under normoxic conditions via TET degradation through the pVHL pathway [268], which is consistent with reports of increased TET activity under hypoxic conditions [267, 269]. However, at least two studies report TET inhibition under hypoxic conditions [265, 270]. However, these studies were all performed in cancer cell lines and the results may be specific to cancer subtypes.

The hypoxia response is also limited by pre-existing DNA methylation. DNA methylation of the HRE inhibited HIF activation of target gene Epo [271]. In malignant cancer cells, DNA methylation inhibited HIF binding to the EGLN3/PHD3 gene, the protein product of which hydroxylates HIF [272, 273], suggesting that DNA methylation can inhibit this negative feedback loop in disease states.

Hypoxia leads to changes in histone acetylation. Hypoxia triggers global loss of histone acetylation, as well as specific loss of acetylation at hypoxia-repressed genes [274, 275]. Histone deacetylase 1 (HDAC1) is induced by hypoxia and represses the tumor suppressor genes VHL and p53 to enable angiogenesis (which tumor suppressor genes inhibit) [276]. The HATs CBP, p300, and SRC-1 directly interact with HIF transcription factors to promote induction of the genes activated by hypoxia (reviewed in [247, 277]). CBP and p300 bind HIF-1α under hypoxic conditions; when HIF-1α binds a target gene, these HATs acetylate the gene promoter to promote chromatin remodeling and gene induction [278–281]. Specifically, these HATs are required for induction of the target genes EPO, VEGF and lactate dehydrogenase-A (LDH-A) [282]. Inactivation of p300 abolishes HIF-1α gene induction [280, 281] and overexpression of the HIF-1α domain that binds CBP and p300 potentiates induction [283]. However, CBP or p300 HAT function is required for only 35–50% of HIF-1α-induced genes [279]. SRC-1 is an important third HAT at some target genes that functions cooperatively with CBP and p300 but may not require CBP and p300 for gene activation [282]. CBP and p300 form a complex with SRC-1 and support colocalization of SRC-1 and HIF-1α [282]. SRC-1 may form a physical bridge between HIF-1α and the PIC [284] and its presence at gene promoters enhances activation of HIF-1α genes [284, 285].

The negative feedback loop for rapid cessation of a hypoxia signaling response includes disruption of HIF-1α’s association with HATs. Factor Inhibiting HIF-1 (FIH-1) requires molecular oxygen as a cofactor and disrupts the interactions between HIF-1α and p300/CBP in the presence of sufficient cellular oxygen [286–290].

Hypoxia leads to changes in histone methylation at both repressed and induced genes. Many histone demethylases require oxygen as a cofactor, and these enzymes are inhibited under hypoxic conditions (reviewed in [256]). For example, hypoxia triggered increased H3K4me3 via inhibition of the HDM JARID1A’s demethylation capacity [291]. However, not all histone demethylases appear sensitive to oxygen. For example, hypoxia downregulates the DNA repair genes BRCA1 and RAD51 by removal of activating H3K4 methylation by LSD1 [292]. In another example, the HDM JMJD1A decreased repressive H3K9 methylation at hypoxia target genes, enabling their induction [293].

HIFs recruit SWI/SNF chromatin remodeling proteins to activate target genes. The BRM and BRG1 catalytic subunits of human SWI/SNF promote transcription of a subset of HIF-1α and HIF-2α target genes and induce transcription of the HIFs themselves [294]. BRM and BRG1 are recruited to the Epo promoter by HIF-1α during hypoxia and both are necessary for Epo transcription [295]. Both chromatin remodeling proteins are required for the overall hypoxia response, but in many cases, they are targeted to distinct promoters by different transcription factors [296, 297]. In some cases, BRM and BRG1 are recruited but are not required for induction. For example, both proteins are recruited to the VEGF promoter, but VEGF transcription does not require either protein [295], suggesting that this gene does not require chromatin remodeling for induction. Both BRM and BRG1 contain bromodomains that bind acetylated lysines, and the VEGF promoter is highly acetylated [295, 298]. Therefore, these chromatin remodelers may be recruited to the VEGF promoter based on the promoter’s chromatin structure, even if the promoter does not require chromatin remodeling for activation.

4.4.3. Hypoxia response in the 3D genome

HIF-binding sites and their target genes interact over long ranges [299] through chromatin looping [300]. These loops are constitutively established at baseline, even in the absence of transcriptional activation by HIFs [300], and hypoxia does not appear to substantially alter looping patterns on a global scale [300], which is supported by gene-specific data for the target gene PAG1 [301]. These pre-established loops may function to increase the speed of the hypoxia response. However, there is some evidence a CTCF-mediated loop may be altered at another target gene, PAX6 [302]. Specifically, CTCF binds a repressor element upstream of the PAX6 gene to repress the gene [302]. At baseline, CTCF is post-translationally modified with SUMO; SUMOylated CTCF cannot bind the PAX6 repressor element [302]. Although this study does not report on 3D looping, CTCF often exerts its transcriptional regulation through chromatin loops [89], suggesting that a new loop is involved in PAX6 repression.

4.4.4. Transcriptional memory in the hypoxia response

Cells that are pre-conditioned to hypoxia retain a cross-reactive memory of that conditioning [303]. Intra-tumoral cells reside in a hypoxic microenvironment and these cells develop a ROS-resistant phenotype that provides a survival advantage during metastasis [303]. These cells show a sustained increase in the expression of a subset of hypoxia-inducible genes after they metastasize [303]. A similar memory is reported in chick embryos with neuroblastomas [304]. Live imaging of these embryos shows that exposure of neuroblastoma cells to 1% oxygen for three days leads to these cells acquiring metastatic characteristics that are retained by these cells after hypoxia ceases; this memory effect was not seen in milder hypoxia (8% oxygen) or after shorter exposure to 1% oxygen [304]. In addition, non-hypoxic cells are influenced by neighboring cells that were pre-conditioned by hypoxia; these non-hypoxic cells can acquire the same metastatic characteristics, suggesting that regulatory molecules are passed from hypoxic to non-hypoxic cells [304], possibly via extracellular vesicles [305–307], or that hypoxic cells create a path for normoxic cell invasion by promoting matrix degradation or cell migration [247]. This phenotype requires HIF-dependent transcription of genes involved in metastasis and can be blocked by HIF inhibition [304].

Transcriptional memory of hypoxia signaling can cross-react with immune functioning, as well. HIF-1α is required for trained immunity, or the epigenetic reprogramming of myeloid cells that protects them from secondary pathogen challenge [308]. Mice with a myeloid defect in HIF-1α were unable to mount this response during bacterial sepsis [308]. Neonatal mice transiently exposed to hypoxia and challenged with bacterial LPS at post-natal day 9 or at 6–8 weeks showed sex-specific augmentation of basal expression of microglial anti-inflammatory cytokines and attenuated microglial responses to LPS stimulation [309], indicating that these memories can be stably maintained into adulthood.

4.5. Unfolded protein response

The unfolded protein response is activated when proteins misfold or aggregate in either the endoplasmic reticulum, where nascent proteins are folded, or mitochondria, where ROS levels are high and proteotoxicity more likely [310, 311]. There are unfolded protein responses attributed to each of these organelles: the endoplasmic reticular unfolded protein response (UPR^ER) [311] and the mitochondrial unfolded protein response (UPR^mt) [312].

ER stress results from accumulation of unfolded or misfolded proteins in the ER lumen, which activates transmembrane sensors on the ER: inositol-requiring protein 1 (IRE1, also known as ERN1), protein kinase RNA-like ER kinase (PERK, also known as EIF2AK3), and activating transcription factor 6 (ATF6) (reviewed in [311]). At baseline, these sensors are bound and kept inactive by the ER luminal chaperone binding immunoglobulin protein (BiP) [311]. When unfolded proteins accumulate in the ER lumen, BiP is sequestered away from the lumen, enabling sensor activation [311]. IRE1 dimerizes and removes an intron from the X-box binding protein 1 (XBP1) mRNA, which is then translated to an active transcription factor [311]. PERK phosphorylates the eukaryotic translation initiation factor-2α (eIF2α) to inhibit translation, with the notable exception of activating transcription factor 4 (ATF4), which is translated at a higher rate [311]. ATF6 localizes to the Golgi complex and is cleaved to produce an active cytosolic transcription factor [311]. The resulting three active transcription factors (XBP1, ATF4, ATF6) bind ER stress response elements (ERSEs) and unfolded protein response elements (UPREs) [313] to initiate transcription of genes encoding protein chaperones (including BiP), which increase assisted protein folding capacity, and components of the ER-associated protein degradation (ERAD) pathway, in which misfolded proteins are ubiquitinated, translocated to the cytosol and degraded by lysosomes [311]. These transcription factors can also activate the transcription factors include CCAAT/enhancer-binding protein beta (C/EBPβ) and C/EBP homology protein (CHOP, also known as DDIT3) to induce apoptosis [311]. Some UPR^ER genes are upregulated by other transcription factors, including JUN, TFEB, TFE3, NFAT, and TCF/LEF, but their activation and roles are unclear [311]. Even in the absence of any external stressor, ER homeostasis is disturbed during cell growth and during cellular differentiation, when the folding capacity of the ER is exceeded by the high protein synthesis rates [311]. Sustained ER stress is seen in diseases like cystic fibrosis and neurodegenerative diseases due to UPR dysregulation (reviewed in [14]).

Protein homeostasis in the mitochondria is sensed and regulated by the UPR^mt [310]. In the model organism C. elegans, the primary transcription factor that regulates the UPR^mt is ATFS-1, which has a functional mammalian ortholog, ATF5 [310]. ATF5 is present in mitochondria under normal conditions, but under conditions of mitochondrial proteotoxicity, it translocates to the nucleus and activates transcription of mitochondrial chaperones [310]. UPR^mt transcription factors are regulated by C/EBPβ and CHOP (reviewed in [14].) In pancreatic β cells, the pancreas/duodenum homeobox protein 1 (Pdx1) regulates cell survival and ER stress susceptibility through direct transcriptional regulation of Atf4, the UPR^ER transcription factor, and of Atf5, the UPR^mt transcription factor [314]. Pdx1 is required for Atf4 and Atf5 transcription, suggesting tissue-specific regulation of UPRs is controlled by tissue-specific transcription factors [314].

Cytoplasmic sensors of proteotoxicity are important in UPRs, too. The nascent polypeptide-associated complex α subunit (αNAC) binds to growing nascent polypeptide chains and inhibits signal recognition particles from transporting these nascent chains to the ER lumen [315]. γ-taxilin, which is involved in intracellular vesicle trafficking, binds αNAC [315]. Both of these proteins appear to function as extra-ER chaperones [315]. The αNAC/γ-taxilin pathway can be downregulated by glycogen synthase kinase 3β (GSK3β), which is activated by the kinase AKT, which is, in turn, activated via dephosphorylation by protein phosphatase 2A [315]. This downregulation triggers UPRs and leads to mitochondria-dependent apoptosis [315].

Overactivation of the UPR^ER can clearly be harmful to the cell. UPR^ER overactivation in astrocytes (both in vitro in response to ER stressors and in vivo in mice with neurodegenerative disease) caused a loss of synaptogenic function by modifying the astrocytic secretome [316]. In these astrocytes, the secretome was depleted of compounds important for synapse formation and maintenance, including extracellular matrix components, like collagen, fibronectin and filamen [316]. In addition, overactivation of the UPR^ER is an important acquired characteristic in cancer cells, particularly for drug resistance phenotypes, and its loss inhibits tumor growth [317].

4.5.1. Redox-active pollutants activate the unfolded protein response

UPRs are activated by pollutants that cause proteotoxicity of fully translated proteins that are located within the ER or mitochondria [318]. Although the sensing mechanism is different for UPRs and the heat shock response (which senses ubiquitination of nascently translated proteins in the cytosol), a similar set of pollutants activate both responses. For example, heavy metals that bind thiol groups in proteins, including lead, mercury and cadmium, activate UPRs [319]. In addition, DNA damaging agents, including ionizing radiation and topoisomerase inhibitors like etoposide, promote the proteasomal degradation of BiP to trigger UPR^ER activation [315]. Inhibition of topoisomerases, which relieve DNA torsion stress during replication and transcription, increases DNA double-strand breaks, as well as ROS, which then oxidizes DNA bases and enhances the DNA cleaving function of uninhibited topoisomerases [320]. Although synthetic topoisomerase inhibitors are used in clinical therapies, several naturally occurring compounds possess topoisomerase inhibition activity, including polyphenols, like EGCG found in green tea, genistein in soy, and resveratrol in red wine and various plants [321–323]. However, UPR^ER activation also triggers repression of the gene encoding topoisomerase IIα [324], for reasons that are unclear, highlighting the complexity of UPR response to cellular stress.

4.5.2. Unfolded protein response in the 2D genome

Histone modifications and transcription factor binding partners appear to be highly target-specific in the UPR, which likely confers granularity and specificity in responses. For example, in C. elegans, heterochromatin protein like-2 (HPL-2), a homolog of the human heterochromatin protein 1 (HP1), downregulates the UPR^ER in the intestine, perhaps by repressing XBP1 expression [325]. HPL-2 and HP1 are important chromatin binding proteins that are recruited by the repressive mark H3K9me to promote chromatin compaction and gene silencing [325]. Inactivation of HPL-2 results in enhanced resistance to UPR^ER signaling through increased basal XBP1 activation in the absence of stress [325]. H3K9 methylation regulates the UPR^ER independently of HPL-2; mutant worms lacking both hpl-2 and the histone methyltransferase met-2 (but not the histone methyltransferase set-25) showed synergistic survival phenotypes [325]. In yeast, histone acetylation is the critical factor; deletion of the histone acetyltransferase Gcn5 is sufficient to block a UPR transcriptional program [326]. In mammals, ER stress enhances chromatin accessibility in activated genes mediated by increased histone acetylation and activating histone methylation [327] but not decreased nucleosome occupancy [312]; these data indicate that the DNA relaxes to allow transcription factor binding without nucleosome eviction [40]. ATF6 is recruited to UPR target gene promoter elements by a constitutively bound transcription factor CBF/NF-Y (CBF) [328]. CBF is also required for recruitment of the TATA-binding protein to the promoter, but it is not required for H3K9 acetylation or H3K4 methylation that occur prior to target gene transcription [328]. Another constitutively expressed transcription factor, YY1, is required for expression of BiP, but YY1 binds the BiP promoter only under ER stress in a complex with ATF6 [329] YY1 recruits the arginine methyltransferase PRMT1, which methylates arginine 3 on histone H4 [329]. Histone H4 acetylation is deposited at the BiP promoter by p300, although it is unclear how p300 is recruited to the gene [329]. It is possible that ATF6 recruits p300, in a manner similar to ATF4 [330]. Chromatin regulatory mechanisms in the UPR may be specific to cell type, differentiation state, or disease state. For example, the HMT DOT1L prevents induction of key UPR genes in neural stem cells via the histone modification H3K79me; this effect is absent in mouse embryonic fibroblasts and the opposite effect is observed in human leukemic MOLM-13 cells [331].

4.5.3. Unfolded protein response in the 3D genome

Currently, there is limited, gene-specific evidence in yeast and humans for 3D control of the UPR. In yeast, inositol starvation triggers a UPR response, including activation of the UPR target gene INO1 (highlighted earlier in this review in the section on positive priming) [132]. INO1 is activated when the transcription factor Hac1 (a homolog of XBP1) triggers dissociation of a repressor upstream of the gene, causing INO1 to be physically recruited to the nuclear membrane for activation [132]. In humans, four UPR^ER genes (CACFD1, GTF3C5, SARDH, and FAM163B) are differentially expressed in skeletal muscle in individuals carrying a SNP in intron 9 of the G-protein signaling modulator 1 (GPSM1) gene [332]. This SNP disrupts CTCF binding to the GPSM1 allele and decreases transcription of the four UPR^ER target genes, ostensibly via disruption of chromatin looping [332].

4.5.4. Transcriptional memory in the unfolded protein response

UPR activation during early development can trigger transcriptional memory in adulthood. Specifically, mild UPR^mt stress during early development increases lifespan in C. elegans, possibly through ROS resistance in adulthood [312]. In these worms, the gene lin-65 is essential for UPR^mt response in mitochondrial stress (although lifespan extension does occur in the absence of this gene, highlighting that mitochondrial stress triggers multiple pathways that contribute to the lifespan effect) [312]. The LIN-65 protein translocates to the nucleus during stress and is also required for the translocation of a UPR^mt transcription factor DVE-1 [312]. After nuclear translocation, DVE-1 binds to nuclear chromatin at accessible sites [312]. Prior to DVE-1 translocation, the UPR^mt triggers the chromatin incorporation of new histones modified by repressive H3K9me2, possibly to restrict DVE-1 binding sites to a specific gene set during stress [312]. Overexpression of the histone demethylase genes jmjd-1.2 and jmjd-3.1, the protein products of which remove repressive H3K27me3, are sufficient to trigger mild UPR^mt activation in early life (likely due to repression of mitochondrial genes necessary for function) and recapitulate the lifespan extension effect [333]. Although these results do not specifically demonstrate transcriptional memory, they do report an initial phenotype that requires a chromatin-based response and a secondary phenotype that is suggestive of memory.

Activation of UPR^ER in fully developed yeast and rodents can also trigger transcriptional memory. As outlined in the section on positive priming, yeast exposed to inositol starvation mount a UPR response that involves activating the INO1 gene by recruiting it to the nuclear membrane [132]. Yeast develop a highly specific, programmed memory at this gene by initial binding of the transcription factor Sfl1 to a Memory Recognition Sequence, followed by recruiting a modified Set1/COMPASS complex that deposits only H3K4me2 [99, 103, 132, 142]. The H3K4me2 modification recruits Set3, which in turn recruits RNAPII more quickly to the gene promoter than in the unprimed state [99, 103, 132, 142]. A UPR response in mice also yields gene-specific memories, although it is unclear if these memories are programmed or if they differ based on stochastic factors. In a mouse model of post-traumatic stress disorder, adult mice exposed to a single, prolonged stress showed transient activation of the UPR^ER, except for upregulation of IRE1α, which was sustained for seven days post-stress (peaking on days 1, 4, and 7), before returning to baseline [334]. Other target genes showed different memory patterns. The pro-apoptotic UPR^ER gene Bcl-2 showed a peak in transcription on day 1 post-stress and then returned to baseline, but the pro-apoptotic gene Bax showed sustained upregulation on days 1–3 post-stress, with a subsequent increase on day 4 that was maintained for several days [334].

4.6. DNA damage response

The DNA Damage Response (DDR) is a cellular sensor program for detecting DNA damage, activating appropriate cell cycle checkpoints to slow cell cycle progression, and activating DNA repair pathways to repair DNA lesions [335, 336]. The general DDR program is activated for all DNA damage types, although the specific responses differ by damage type [335, 336]. The regulation of these DNA repair pathways, including induction of DNA repair proteins and chromatin remodeling at sites of DNA damage to facilitate repair, is highly complex and is described elsewhere. In this review, I focus on the transcription regulation of the primary response genes that initially respond to the stress of DNA damage in healthy (non-cancerous) cells. The protein kinases ataxia-telangiectasia mutated (ATM), ATM- and Rad3-Related (ATR), and DNA-dependent protein kinases (DNA-PKs) are the primary kinases active in the DDR (reviewed in [335, 336]). ATM and ATR target the protein kinases CHK1 and CHK2 to reduce cyclin-dependent kinase (CDK) activity, partly through the action of the p53 transcription factor [335, 336]. CDK inhibition slows or arrests the cell cycle to allow for DNA repair prior to DNA replication [335, 336]. ATM and ATR also activate new transcription of DNA repair proteins, in addition to modifying these proteins post-translationally and recruiting them to sites of damage [335, 336]. If the damage is repaired, DDR signaling ceases; if the damage is unrepaired, chronic DDR signaling triggers cell death via apoptosis or cellular senescence [337].

4.6.1. Redox-active pollutants activate the DNA damage response

A large number of redox-active genotoxic pollutants trigger the DDR. These include ultraviolet light, ionizing radiation, environmental carcinogens including PAH, all of which cause specific forms of DNA damage, including single strand breaks, double strand breaks, and bulky DNA adducts, among many more [338–340]. In addition, many pollutants that are not considered genotoxic can induce ROS, including many heavy metals, endocrine disrupting compounds and pesticides [150]; ROS itself can cause oxidative DNA damage [3]. This topic has been extensively reviewed elsewhere [338–340].

4.6.2. DNA damage response in the 2D genome

The primary transcriptional regulators of the initial DDR transcriptional response include the transcription factor p53 and the kinases, ATM, ATR and p38α [341]. P53 is ubiquitinated and targeted for degradation by the E3 ubiquitin ligase Mouse double minute 2 (MDM2) under non-stress conditions [341]. Under stress, ATM disrupts the P53-MDM2 interaction by phosphorylating p53 in its MDM2-interaction domain, which stabilizes p53 [341]. P53 then translocates to the nucleus and mounts a transcriptional program to accomplish one of three outcomes: transient cell cycle arrest to enable DNA repair, permanent cell cycle arrest (senescence), or apoptosis [341]. Many transcriptional co-regulators modify p53 activity and refine its transcriptional program [341].

P53 binds DNA with the HAT p300, and p300 acetylates chromatin to recruit chromatin remodelers at activated genes [342]. P53 can also be acetylated by p300, which promotes p53’s association with other co-activators and HATs [342] and increases its residence time bound to regulatory elements to increase transcription of target genes [111]. At p53-activated target genes, chromatin is enriched with active chromatin marks (e.g., H3K4me3) [343] and p53 associates with the chromatin remodelers SWI/SNF or Remodeling and spacing factor 1 (RSF) [344], which remodel promoter chromatin and induce transcription [345]. The CSB chromatin remodeling protein can also bind to p53 directly, but in a reciprocally regulated manner [346]. CSB enables p53 binding to chromatin when p53 levels are low but at high p53 concentrations, p53 occludes CSB’s nucleosome interaction surface and prevents CSB from binding DNA [346]. P53 can also activate genes that are enriched with the silencing histone modification H3K9me3 at baseline [347]. P53 downregulates the HMT SUV39H1 that deposits H3K9me3, after which these promoters are modified with the activating mark H3K4me3 by the HMT MLL1 [347]. A subunit of the COMPASS/MLL1 complex, Ash2L, is required for p53 to activate transcription of these genes [347]. Ash2L deficiency results in decreased poising of RNAPII, reduced general transcription factor binding, and reduced gene induction, even though p53 and RNAPII binding are unaffected [347]. P53 can repress target genes in cooperation with the histone demethylase LSD1; P53 and LSD1 co-bind a regulatory element and LSD1 demethylates H3K4 to repress a classic target gene, α-fetoprotein (AFP) [348].

P53 activity is repressed by the linker histone H1.2 and by co-binding with MDM2 [349] Linker histones bind DNA between nucleosomes and are often considered transcriptional repressors because they prevent transcription factor binding [349]. The linker histone H1 directly interacts with regulatory factors to repress transcription [349]. The linker variant H1.2 is recruited to p53 target genes where it associates with p53 to repress p300-mediated chromatin acetylation [349]. Acetylation of p53 by p300 and H1.2 phosphorylation by DNA-PK disrupt the p53-H1.2 association and release p53 to activate transcription [349]. P53 bound to DNA can also be blocked from activating transcription by co-binding of MDM2 [350, 351]. MDM2 is itself a p53 target gene, which suggests both a negative feedback loop and crosstalk with the oxidative stress pathway [342, 352]. MDM2 is recruited to chromatin independently of p53 via tethering by the transcription factors ATF3/4 to regulate cellular redox state during oxidative stress [352]. Depletion of MDM2 in p53-deficient cells impairs serine/glycine metabolism, glutathione recycling, and the ratio of oxidized to reduced nicotinamide adenine dinucleotide (NAD⁺/NADH) [352].

Interestingly, p53 plays a role in limiting lineage conversion, in which one cell type switches cellular identity via epigenetic remodeling with lineage-specific transcription factors [353]. In mouse fibroblasts treated with liver-specific lineage-determining transcription factors, ATM-p53 was strongly activated and triggered proliferation arrest and cell death [353]. Most interestingly, ATM’s activation was not triggered by DNA damage, but rather responded to regions of chromatin that were newly opened by the lineage-determining transcription factors [353]. Baf60b, a subunit of SWI/SNF, recruits ATM to these newly opened sites to initiate a cell death response [353].

P38α also plays an important role in chromatin responses in the DDR. Ultraviolet radiation activates p38α to phosphorylate MK2, which, in turn, phosphorylates negative elongation factor polypeptide E (NELF-E) [164]. Phosphorylated NELF-E dissociates from chromatin, enabling release of poised RNAPII at more than 2,000 genes involved the DDR [164]. In contrast, P-TEFb is recruited to target gene promoters by RNA-binding motif protein 7 (RBM7) in response to DNA damage; this recruitment is mediated by p38α [354].

4.6.3. DNA damage response in the 3D genome

Chromatin looping is an important regulatory component of the p53 transcriptional program, as evidenced by the altered chromatin looping and differential transcription program activated by mutant p53, as compared to wild-type p53 [355]. In one example in healthy cells, p53 uses a loop to repress its target gene CDKN1A, and abolishing the loop is sufficient to activate the gene [356]. Human cells exposed to the DNA damaging agent doxorubicin showed eviction of cohesin from the CDKN1A gene body, which correlated with transcriptional induction of that gene [356]. Cohesin is evicted by RNAPII as it moves along the p53 target gene CDKN1A, and loss of cohesin disrupts a chromatin loop that connects a downstream repressor to an alternative promoter within the gene, permitting gene activation [356]. More globally, cohesin is evicted from gene bodies of multiple p53 target genes [356]. In addition, p53 target genes cluster in chromatin looped hubs at baseline, which allows faster transcription when p53 binds enhancers that are already in contact with gene promoters [356]. P53 can also regulate multiple target genes from a single distal enhancer, a process that requires physical bridging from enhancer RNA, which are transcribed from active enhancers [357].

4.6.4. Transcriptional memory in the DNA damage response

Although we have known for some time that DNA lesions can be inherited mitotically [358], it has only recently been reported that DDR signaling can be inherited across cellular generations. Transcriptional memory of DDR signaling determines whether daughter cells will immediately enter the next cell cycle or will remain in a quiescent state. Proliferative human breast cancer cells directly transmit p53 protein and mitogen-induced cyclin D1 (CCND1) mRNA to daughter cells [113]. [113]. The cyclin-dependent kinase inhibitor p21, which is induced by p53 to promote cell cycle exit, and the newly translated cyclin D1 protein are variably expressed in daughter cells and compete to determine cell cycle commitment [113]. This is a remarkable example of epigenetic (mitotically heritable) transcriptional memory directly involving key transcriptional regulators. It is unclear whether this memory is chromatin-based, since it is unknown whether mitotically inherited p53 is bound to chromatin, and, if so, whether the bound sites represent sites of transcriptional memory.

5. Crosstalk and cross-reactivity in CDS pathways

Although I have presented CDS pathways as functionally independent in this review, in reality they crosstalk extensively. I list a few examples here to illustrate these concepts. One form of crosstalk involves a key regulator of one CDS pathway subsequently activating a second CDS pathway. For example, in oxidative stress caused by arsenite or peroxide, NRF2 activates the HSF1 promoter to trigger a heat shock response [359]. Another form of crosstalk involves one CDS pathway repressing another. For example, the DNA damage response and hypoxia compete for activation. In hypoxic conditions, the DNA repair genes BRCA1 [292] and MLH1 [360] are silenced. Alternately, p53 represses the transcription of HIF-1α via competition for binding with the HATs p300/CBP [361–366]. In a third form of crosstalk, the protective response of one CDS pathway creates secondary stress that triggers a second CDS pathway. For example, hypoxia triggers the UPR^ER due to protective inhibition of protein translation, which is one way that the cell conserves energy in low oxygen conditions [255, 315, 329]. In addition to crosstalk, CDS pathways are cross-reactive. For example, early life exposure to ROS causes a more robust response to adult exposure to heat shock [245].

6. A conceptual framework for CDS responsiveness as a readout of redox-active pollutant exposure

I propose a conceptual framework in which an individual’s cumulative redox-active pollutant exposures can alter chromatin inducibility and induce transcriptional memories at CDS target genes. These memories can either be protective (in response to mild pollutant exposure) or they can cause an exaggerated subsequent response to stress or a dampened ability to mount a protective response (in response to severe or chronic pollutant exposure). This framework is supported by existing evidence that exaggerated and dampened CDS responses can harm the cell and are associated with disease. For example, comparative studies of yeast and human cells show that stress responses must be temporally restricted because their continuous induction is detrimental to cell growth [244, 367]. This slowed cell growth may be due to chronic activation of cell cycle checkpoint systems and subsequent cell cycle delays or to energy diversion into stress response [368, 369]. This effect is apparent in whole organisms, too. Loss of Nrf2 in mice increases susceptibility to cancer [150] and defects in the heat shock response disrupt mouse development [370]. Human diseases, including cardiovascular disease [371], cancer [304], diabetes [372], Alzheimer’s disease [373], and Parkinson’s disease [374], are also associated with CDS dysfunction.

I hypothesize that transcriptional memories at CDS genes are tissue specific. First, tissue specificity in toxicological responses can be due to a higher relative dose in a specific target tissue where a pollutant may accumulate [1]. Higher doses of exposures elicit stronger induction and subsequent memory formation at target genes [160]. In addition, the landscape of accessible transcription factor binding sites differs among cell types [47]. Different cell types and cells at different stages of differentiation show different CDS transcriptional programs in response to stress [154], suggesting that CDS gene set bookmarking may diverge during lineage-determination in early development. This suggests the possibility of alternative enhancer bookmarking in early development by a developmental redox-active pollutant exposure, which is supported by evidence that NRF2 regulates expression of the lineage-determining factor PU.1 in differentiated cells [375], which suggests that similar regulation may occur at earlier stages of development. Disruption of CDS bookmarking during development (perhaps by activation of genes critical for cellular protection in response to pollutant exposure) could permanently alter cells’ abilities to mount CDS responses, as well as the observed cell type differences in CDS gene set activation.

This framework highlights several open questions. First, the mechanisms of transcriptional memory formation are still poorly understood. Second, it is not known why some target genes form memories and others do not. Specifically, are transcriptional memories programmed, as suggested by the binding of transcription factor Sfl1 to a Memory Recognition Sequence to promote highly specific memory at the INO1 gene in yeast, or are they stochastically established [132], as suggested by evidence showing that memories preferentially form within cells that show stochastically greater initial transcriptional responses at target genes [100]? Both possibilities are supported by current evidence. Third, why are some memories retained for long periods, or through cell division, and others are not? Fourth, what factors determine positive vs. negative priming responses? Fifth, given evidence that genetic variation impacts inter-individual differences in CDS gene expression [376], does genetic variation also play a role in inter-individual differences in chromatin inducibility and transcriptional memories at CDS genes?

Within this framework, environmental stressors that exert effects in other ways can still impact CDS responses. The CDS comprises master pathways that are highly interconnected with others in the cell, many of which cannot be turned on at the same time; sustained activation of CDS pathways can potentially block other cellular responses, and vice versa (e.g., the transcription factor AhR competes for a heterodimerization partner with HIF-1α [377], so activation of one pathway blocks activation of the other).

This framework offers a new way of conceptualizing biological impacts of multiple or chronic exposures to redox-active pollutants. I note that it is possible that CDS transcriptional memories will scale in proportion to total cellular burden of redox-active pollutant exposures, which would suggest these loci as cumulative biomarkers of effect.

7. Conclusion

In summary, redox-active pollutants can activate a common set of cellular defense system pathways, in addition to those pathways activated specifically by individual pollutants. These cellular defense pathways are activated in response to an individual’s cumulative redox-active pollutant exposure and may be more likely than other pathways to develop dysfunction. CDS pathway dysfunction is a common hallmark of many chronic and complex diseases, suggesting that redox-active pollutant exposure contributes to complex disease development through activation or dysfunction of CDS pathways. Therefore, increasing our understanding of CDS pathway function is an important step in preventing disease. The majority of current evidence for redox-active pollutant effects on these pathways focuses on pollutant sensing and cellular signaling. However, cellular signaling pathways trigger different gene expression patterns in different cell types and in response to different stimuli. I suggest that part of this specificity in CDS response is due to chromatin inducibility at target genes in different cell types and that chromatin inducibility can change (i.e., transcriptional memories form) with repeated exposures. A clearer understanding of chromatin determinants of CDS responses will significantly clarify sources of inter-individual differences in cellular responses to redox-active pollutants and will support better prediction of individual disease risk in response to these pollutants.

Highlights.

• Redox-active pollutants activate cellular defense system pathways

• Inducibility of CDS genes requires a responsive chromatin state and structure

• Transcriptional memories at CDS genes influence responses to future exposure

Abbreviations:

2D

two-dimensional

3D

three-dimensional

AFP

α-fetoprotein

AhR

aryl hydrocarbon receptor

AKT

AKT serine/threonine kinase

αNAC

nascent polypeptide-associated complex α subunit

ARE

antioxidant response element

Ash2L

ASH2-like histone lysine methyltransferase complex subunit

ATF4

activating transcription factor 4

ATF5

activating transcription factor 5

ATF6

activating transcription factor 6

ATFS-1

activating transcription factor associated with stress

ATM

ataxia-telangiectasia mutated

ATR

ATM- and Rad3-Related

Baf60b

BRG1-associated factor 60B

Bax

BCL2-associated X

BBP

benzyl butyl phthalate

Bcl-2

B-cell CLL/Lymphoma 2

BiP

binding immunoglobulin protein

BPA

bisphenol A

BRCA1

breast cancer type 1 susceptibility protein

BRD4

bromodomain-containing protein 4

BRG1

Brahma-related gene 1

BRM

Brahma

C/EBPbeta

CCAAT-enhancer binding protein betea

CACFD1

calcium channel flower domain containing 1

CBF

core binding factor

CCND1

mitogen-induced cyclin D1

CDK

cyclin-dependent kinase

Cdk7

cyclin-dependent kinase 7

Cdk8

cyclin-dependent kinase 8

CDKN1A

cyclin-dependent kinase inhibitor 1A

CDS

cellular defense system

CHD

chromodomain helicase DNA-binding

CHK1

checkpoint kinase 1

CHK2

checkpoint kinase 2

CHOP

C/EBP homology protein

COMPASS

complex proteins associated with Set1

CPB

CREB-binding protein

CREB

c-AMP response element binding protein

CSB

Cockayne syndrome group B protein

CTD

C-terminal domain

CUL3

Cullin 3-RING-box E3 ubiquitin ligase

CYP1A

cytochrome P450 1A

DDIT3

DNA damage inducible transcript 3

DDR

DNA Damage Response

DEHP

di(2-ethylhexyl) phthalate

DNA

deoxyribonucleic acid

DNA-PK

DNA-dependent protein kinase

DNMT

DNA methyltransferase

DoG

downstream-of-gene-containing transcript

DOT1L

DOT1-like histone lysine methyltransferase

DSIF

DRB sensitivity inducing factor

DVE-1

defective proventriculus in Drosophila homolog 1

EDC

endocrine disrupting chemical

EGCG

epigallocatechin gallate

EGLN3/PHD3

Egl-9 family hypoxia inducible factor 2/prolyl hydroxylase domain 3

EIF2AK3

eukaryotic translation initiation factor 2-alpha kinase 3

eIF2α

eukaryotic translation initiation factor 2-α

Epo

erythropoietin

ER

endoplasmic reticulum

ER-α

estrogen receptor-α

ERAD

endoplasmic reticulum-associated protein degradation pathway

ERN1

endoribonuclease 1

ERSE

endoplasmic reticulum stress response element

ESR

environmental stress response

EZH2

Enhancer of Zeste homology 2

FAM163B

family with sequence similarity 163 member B

FIH

factor inhibiting HIF-1

FOXO

Forkhead box O

FOXO1

Forkhead box O1

FOXO3

Forkhead box O3

FOXO4

Forkhead box O4

GAF

GAGA-associated factor

GAL

galanin and GMAP prepropeptide

GATA-1

GATA-binding factor 1

GATA-2

GATA-binding factor 2

Gcn5

general control non-repressed 5

GPSM1

G-protein signaling modulator 1

GREB1

growth regulating estrogen receptor binding 1

GSK3β

glycogen synthase kinase 3β

GTF3C5

general transcription factor IIIC subunit 5

H1

histone H1

H1.2

histone H1.2

H19/IGF2

h19 transcript/insulin like growth factor 2

H2A

histone 2A

H2B

histone 2B

H2Bub1

histone 2B mono-ubiquitinated

H3

histone 3

H3.3S31ph

histone 3.3 serine 31 phosphorylated

H3K14ac

histone 3 lysine 14 acetylated

H3K27ac

histone 2 lysine 27 acetylated

H3K27me3

histone 3 lysine 27 tri-methylated

H3K4me1

histone 3 lysine 4 mono-methylated

H3K4me3

histone 3 lysine 4 tri-methylated

H3K9me3

histone 3 lysine 9 tri-methylated

H3S10ph

histone 3 serine 10 phosphorylated

H4

histone 4

H4K16ac

histone 4 lysine 16 acetylated

H4K20ac

histone 4 lysine 20 acetylated

Hac1

hyperpolarization-activated cation channel 1

HAT

histone acetyltransferase

HDAC

histone deacetylase

HDAC1

histone deacetylase 1

HDM

histone demethylase

HIF

hypoxia inducible factor

HIF-1α

hypoxia inducible factor 1α

HIF-1β

hypoxia inducible factor 1β

HIF-2α

hypoxia inducible factor 2α

HIF-3α

hypoxia inducible factor 3α

HMG-14

high mobility group protein 14

HMGB1

high mobility group protein B1

HMT

histone methyltransferase

HO-1

heme oxygenase-1

Hog1

high-osmolarity glycerol 1

HP-1

heterochromatin protein-1

HPL-2

heterochromatin protein like-2

HRE

hypoxia response element

HSF

heat shock factor

HSF1

heat shock factor 1

HSFA2

heat shock factor A2

Hsp70

heat shock protein 70

Hsp90

heat shock protein 90

HSPA1A

heat shock protein A1A

IEG

immediate early gene

IκBα

nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor α

IL-1 β

interleukin-1β

INO1

inositol-3-phosphate synthase

INO80

inositol-requiring mutant 80

IRAK-M

interleukin-1 receptor-associated kinase M

IRE1

inositol-requiring protein 1

ISWI

imitation SWI of Drosophila melanogaster

JARID1A

JumonjiC and ARID-domain containing histone lysine demethylase 1A

Jmjd-1.2

Jumonji domain protein 1.2

Jmjd-3.1

Jumonji domain protein 3.1

JNK

c-JUN N-terminal kinase

JUN

jun proto-oncogene

Keap1

Kelch-like ECH-associated protein 1

Kit

tyrosine protein kinase KIT

KLF4

Kruppel-like factor 4

LDH-A

lactate dehydrogenase-A (gene)

Lin-65

abnormal cell lineage 65

LPS

lipopolysaccharide

LSD1

lysine-specific histone demethylase 1

MAPK

mitogen-activated protein kinase

MCF-7

Michigan Cancer Foundatio-7 cell line

MDM2

Mouse double minute 2

Mi-2β

chromodomain helicase DNA-binding protein Mi-2

MK2/3

MAPK-activated protein kinase 2/3

MLH1

mutL homolog 1

MLL1

mixed-lineage leukemia histone methyltransferase 1

MLL2

mixed-lineage leukemia histone methyltransferase 2

MOF

Ortholog of Drosophila Males Absent on the First

MOLM-13

human leukemia cell line with MLL gene rearrangement

MSK1/2

mitogen- and stress-activated kinase 1/2

MST1

macrophage stimulating 1

NAD+

oxidized nicotinamide adenine dinucleotide

NADH

reduced nicotinamide adenine dinucleotide

NANOG

Homeobox transcription factor NANOG (derived from Celtic ‘Tir nan Og’)

NELF-E

negative elongation factor polypeptide E

NF-κB

nuclear factor kappa-light chain-enhancer of activated B cells

NFAT

nuclear factor of activated T-cells

NRF1

nuclear respiratory factor 1

NRF2

nuclear respiratory factor 2

NRF3

nuclear respiratory factor 3

NTHL1

Nth like DNA glycosylase 1

Nup98

nuclear pore protein 98

OCT4

octamer transcription factor 4

OGG1

8-oxoguanine glycosylase

OXPHOS

oxidative phosphorylation

P-TEFb

positive transcription elongation factor b

P300

adenovirus E1A-associated cellular p300 transcriptional co-activator protein

P53

tumor protein 53

PAG1

phosphoprotein membrane anchor with glycosphingolipid microdomains

PAH

polycyclic aromatic hydrocarbon

PAR

poly[ADP-ribose]

PARP-1

poy[ADP-ribose]polymerase-1

PAX6

paired box protein 6

PCB

polychlorinated biphenyl

Pdx1

pancreas/duodenum homeobox protein 1

PERK

protein kinase RNA-like ER kinase

PIC

pre-initiation complex

PRG

primary response gene

Pu.1

PU-box binding protein 1

pVHL

von Hippel-Lindau protein

RAD51

recA-like protein 51

RBM7

RNA-binding motif protein 7

RelB

RELB proto-oncogene

REST

repressor element-1 silencing transcription factor

RNA

ribonucleic acid

RNAPII

RNA polymerase II

RNF20

ring finger protein 20

RNF40

ring finger protein 40

RNS

reactive nitrogen species

ROS

reactive oxygen species

Rpd3

reduced potassium dependency 1 histone deacetylase

RSF

Remodeling and spacing factor 1

SAGA

Spt5-Ada-Gcn5 histone acetyltransferase complex

SAPK

stress-activated protein kinase

SARDH

sarcosine dehydrogenase

Set1

Set domain-containing 1 histone methyltransferase

SET3

Set domain-containing 3 histone methyltransferase

Sfl1

suppressor gene for flocculation 1

SINE

short interspersed nuclear element

Sko1p-Cyc8p-Tup1p

Suppressor of Kinase Overexpression 1 - Cytochrome C 8 - dTMP-Uptake 1 yeast co-repressor complex

SNP

single nucleotide polymorphism

SOCS3

Suppressor of Cytokine Signaling 3

SRC-1

Steroid Receptor Coactivator-1

SRF

serum response factor

SRG

secondary response gene

STAT1

Signal transducer and activator of transcription 1

STAT6

Signal transducer and activator of transcription 6

SUMO

Small ubiquitin-like modifier

SUV39H1

Suppressor of variegation 3–9 homolog 1

SWI/SNF

Switch/Sucrose Non-Fermentable chromatin remodeling complex

Swr1

SWi/Snf2-Related 1

TAD

topologically-associated domain

TAL1

TAL BHLH Transcription Factor 1

TCF/LEF

transcription factor/lymphoid enhancer binding factor

TET

ten-eleven translocase

TFE3

transcription factor binding to IGHM enhancer

TFEB

transcription factor EB

TFF1

Trefoil factor 1

TIP60

Tat interactive protein 60-kDa histone acetyltransferase

TLR4

Toll-like receptor 4

TNF-α

Tumor Necrosis Factor-α

TREM-1

Triggering Receptor Expressed on Myeloid Cells 1

UPR

unfolded protein response

UPRE

unfolded protein response element

UPR^ER

unfolded protein response - endoplasmic reticulum pathway

UPR^mt

unfolded protein response - mitochondrial pathway

USP22

Ubiquitin Specific Peptidase 22

VEGF

Vascular endothelial growth factor

XBP1

X-box binding protein 1

YY1

Yin Yang 1

ZMYND11

Zinc Finger MYND-Type Containing 11