**The Versatile Roles of the tRNA Epitranscriptome during Cellular Responses to Toxic Exposures and Environmental Stress**

#### **Sabrina M. Huber 1,†, Andrea Leonardi 2,3, Peter C. Dedon 1,4 and Thomas J. Begley 2,5,\***


Received: 4 March 2019; Accepted: 21 March 2019; Published: 25 March 2019

**Abstract:** Living organisms respond to environmental changes and xenobiotic exposures by regulating gene expression. While heat shock, unfolded protein, and DNA damage stress responses are well-studied at the levels of the transcriptome and proteome, tRNA-mediated mechanisms are only recently emerging as important modulators of cellular stress responses. Regulation of the stress response by tRNA shows a high functional diversity, ranging from the control of tRNA maturation and translation initiation, to translational enhancement through modification-mediated codon-biased translation of mRNAs encoding stress response proteins, and translational repression by stress-induced tRNA fragments. tRNAs need to be heavily modified post-transcriptionally for full activity, and it is becoming increasingly clear that many aspects of tRNA metabolism and function are regulated through the dynamic introduction and removal of modifications. This review will discuss the many ways that nucleoside modifications confer high functional diversity to tRNAs, with a focus on tRNA modification-mediated regulation of the eukaryotic response to environmental stress and toxicant exposures. Additionally, the potential applications of tRNA modification biology in the development of early biomarkers of pathology will be highlighted.

**Keywords:** epitranscriptomics; tRNA modifications; stress response mechanisms; codon-biased translation

#### **1. Introduction**

In an ever-changing environment, living systems are subjected to stresses such as temperature fluctuations, nutrient limitations, and exposures that damage intracellular biomolecules. Cells respond to these stresses by activation of response, repair, and adaptation pathways or, if the damage is too severe, by initiation of cell death systems. Among the well-studied stress response mechanisms are the heat shock, unfolded protein, DNA damage, oxidative, and nutrient stress responses, which are reviewed in detail elsewhere [1–3]. The molecular underpinnings of these mechanisms have been extensively explored at the level of signaling pathways, changes in transcription, and the proteome. Here, we focus on recently emerging mechanisms of cellular stress response involving the dynamic regulation of ribonucleoside modifications that control transfer RNA (tRNA) metabolism, structure, and function, which, in turn, modulate protein translation and a cell's ability to cope with stress. We discuss how tRNA modifications can affect tRNA stability, maturation, and codon recognition in

mRNA, all of which regulate cell survival and adaptation during environmental stress and xenobiotic exposures. Aberrant modification of tRNAs is also directly linked to human diseases, including neurological disorders, metabolic diseases, and cancer [4–6]. As such, tRNA-based stress response mechanisms have the potential for diagnostic applications and could be useful biomarkers, as they have been shown to be linked to increased levels of reactive oxygen species (ROS), mitochondrial dysfunction, DNA damage, and changes in metabolism [7–12].

#### **2. tRNA-Based Regulation of Cellular Processes**

tRNAs are best known as the adaptor molecules of the translation machinery, in which they deliver the appropriate amino acids to the ribosomes according to the interaction of their anticodons with the codons of messenger RNAs (mRNAs). tRNAs are extensively post-transcriptionally modified by epitranscriptomic writer enzymes, with these RNA modifications having structural and functional roles, as well as playing downstream roles in different biological processes. A detailed description of these RNA modifications is given in Section 3. Beyond their role as adaptors in protein synthesis, tRNAs have been shown to be critically involved in many other cellular processes. Aminoacylated-tRNAs (aa-tRNAs) act as amino acid donors not only to nascent peptide chains during translation, but also to membrane lipids, to peptidoglycan precursors, and to the amino-terminus of proteins [13]. In addition, tRNAs act as biosynthetic precursors of antimicrobial molecules and tetrapyrroles [13,14]. Uncharged tRNAs are used by retroviruses to prime DNA synthesis and are thus essential for viral replication [15–17]. During the priming of DNA synthesis in human immunodeficiency virus type 1 (HIV-1) infections, the 3 -end of the primer tRNA is complementary to a region of the viral RNA called the primer binding site [18].

tRNAs also function as stress sensors and are key for initiating stress responses [19]. For example, eukaryotic cells respond to nutrient deprivation by global inhibition of protein synthesis, which can be mediated by tRNAs acting as signaling molecules. The limited availability of extracellular amino acids during starvation leads to the accumulation of uncharged cytosolic tRNAs that directly bind to the protein kinase GCN2 [20,21]. This tRNA binding activates the GCN2 kinase activity and results in the phosphorylation of elF2, which in turn regulates translation in amino acid-starved cells and promotes increased synthesis of the transcription factor GCN4 that activates amino acid biosynthetic genes. Interestingly, the GCN4-regulated transcripts exhibit a codon bias that can be linked to regulation by modified wobble uridines in tRNA [22], as discussed shortly. Furthermore, cytosolic and mitochondrial tRNAs can inhibit apoptosis by binding to cytochrome c, thereby blocking the formation of the apoptosome [23,24]. Furthermore, tRNAs serve as a source of small noncoding RNAs called tRNA-derived fragments (tRF), which vary in length, biogenesis, sequence, and function [25,26]. Examples of functions of tRFs in the literature include the repression of gene expression in a miRNA-like fashion, inhibition of translation through ribosome binding, and modulation of cell proliferation [25,27–29]. Interestingly, the formation of tRFs can be controlled by ribonucleoside modifications which is discussed in more detail in Section 4.3.

Here we explore how tRNA modifications play a role in this tRNA-based regulation of cellular processes in the face of stressful conditions.

#### **3. Ribonucleoside Modifications**

Post-transcriptional RNA modifications were first described in 1957, when the presence of pseudouridine was demonstrated in yeast RNA extracts [30]. Since then, more than 140 chemically distinct ribonucleosides have been characterized, some of which are conserved throughout all domains of life. Modified ribonucleosides occur in all RNA classes, with tRNAs containing the most numerous and chemically diverse modifications [31,32]. On average, ~17% of the 80–90 ribonucleosides in tRNA are modified, ranging from relatively small structural changes, such as methylation or acetylation, to highly intricate chemical alterations that give rise to so-called hypermodified bases such as queuosine (Q) (Figure 1A). The type, location, and abundance of modifications not only vary between different tRNAs, but also differ within similar isoacceptors (tRNAs bearing the same amino acid but different anticodons). Ribonucleoside modifications are located throughout tRNA molecules and can affect tRNA stability, folding, localization, transport, processing, and function [33]. The anticodon loops of nearly all tRNAs are heavily modified, predominantly at positions 34 and 37 [31,32]. The ribonucleotide at the wobble position 34 pairs with the third mRNA codon base in the aminoacyl-tRNA binding site (A-site) of the ribosome during decoding and thus is crucial for accurate reading of the genetic code. Modifications at the wobble position of the tRNA anticodon have been shown to allow decoding of multiple mRNA codons differing by the third nucleoside (synonymous codons) or restrict pairing with noncognate codons [34]. Position 37 in tRNA is adjacent to the 3 -side of the anticodon and has been shown to prevent frame shifting during translocation by stabilizing codon–anticodon interactions [35]. It is important to note that tRNA modifications play an important role during HIV-1 infections. There is an additional interaction between an A-rich loop located upstream of the primer binding site region and the lysine tRNA anticodon loop dependent on a sulfur in the wobble base modification 5-methoxycarbonylmethyl-2-thiouridine (mcm5s2U). This demonstrates the importance of modified ribonucleosides in tRNA function [18]. There is also evidence that tRNA modifications play important roles in organelle function. For example, decreased tRNA modification has been directly linked to the mitochondrial diseases myoclonus epilepsy associated with ragged-red fibers (MERRF) and mitochondrial encephalopathy, lactic acidosis, and stroke-like episodes (MELAS) [8,36,37]. These diseases are caused by a mutation in mitochondrial tRNA genes, with the change in sequence preventing formation of the tRNA modifications s<sup>2</sup> and 5-taurinomethyluridine (tm5U), both disturbing codon–anticodon interactions to disrupt protein synthesis [37,38]. As mitochondrial translation is used to synthesize key enzymes involved in metabolism, the resulting mistranslation defects can alter proteins involved in the electron transport chain and ATP synthesis. The resulting defect in energy production can be linked to the muscle weakness and neurological dysfunction associated with MERRF and MELAS, and it provides a mechanistic link between tRNA modifications and energy metabolism.

**Figure 1.** (**A**) Clover-leaf structure of eukaryotic tRNA formed through base pairing in the acceptor stem, D-loop, anticodon arm, and TψC-loop. Structures and positions of some modifications discussed in the review are indicated. (**B**) Hierarchical cluster analysis of average fold-change values for tRNA modifications in total tRNA from *S. cerevisiae* exposed to equitoxic (LD80) doses of various alkylating (EMS, ethyl methansulfonate; MMS, methyl methansulfonate; IMS, isopropyl methanesulfonate; MNNG, *N*-methyl-*N* -nitro-*N*-nitrosoguanidine; MNU, *N*-nitro-*N*-methylurea) and oxidizing agents (γ-Rad, γ-radiation; H2O2, hydrogen peroxide; TBHP, tert-butyl hydroperoxide; ONOO−, peroxynitrite). Yeast cells were exposed to LD80 doses of the agents and tRNA modifications were quantified by liquid chromatography—tandem mass spectroscopy (LC-MS/MS). The fold-change values were derived from the average of normalized MS signal intensity data from five biological replicates relative to unexposed controls, and hierarchical clustering analysis was performed in log space (log2) and visualized as a heat map. Reproduced from [11], ACS publications, 2015.

#### *3.1. tRNA Modification as Dynamic Marks*

RNA modifications were long considered to be static and stable marks after their post-transcriptional enzymatic introduction by "writer" proteins and their removal from the transcriptome was thought to occur passively via degradation of the modified RNA followed by transcription of new unmodified RNA. However, the findings that exposure of wild type yeast to various alkylating and oxidizing agents caused signature patterns of changes in the relative quantities of numerous modified ribonucleosides in tRNAs (Figure 1B) [9,11], demonstrated the responsiveness of modified ribonucleosides to environmental cues and highlighted the potential for reversibility. The pattern of up- and downregulation of these modifications were unique to each stressor and predictive of exposure. For example, in response to the alkylating agents, mcm5U, mcm5s2U, m3C, and m7G were all increased, but were relatively unchanged in response to oxidizing agents. In contrast m5C, i6A, ncm5U and, to some extent, m2 2G, were increased in response to oxidizing agents, but show little change in response to alkylating agents.

Technological advances throughout recent years lead to the identification of a number of "eraser" enzymes that are able to catalyze the active removal of modified residues, in particular methylation marks, and opened the door for dynamic regulation of modified ribonucleosides in tRNA (Figure 2). The reversibility of tRNA modifications, via the eraser ALKBH1, was recently shown to mediate the demethylation of m1A in tRNAs [39]. Notably though, reversibility was first shown in 2011, when the fat mass and obesity-associated protein (FTO) and ALKBH5 were both shown to robustly remove N6-methyladenosine (m6A) on polyadenylated RNAs in vitro through an oxidative demethylation mechanism and contribute to m6A levels in cellular mRNA (Figure 2) [40–42]. Since then, reversible N6-adenosine methylation has been demonstrated to play key roles in a number of biological processes, including mRNA nuclear export, the association of nuclear speckle proteins, splicing, cap-independent translation, UV-induced DNA damage response, leukemogenesis, and drug response [42–45].

**Figure 2.** Reversibility and dynamics of RNA modifications. (**A**) The epitranscriptomic writer for the tRNA m1A modification is a TRMT6/TRMT61A complex, which uses S-adenosyl methionine (SAM) as a methyl donor. ALKBH1 removes the methyl group and requires oxygen, iron and α-ketoglutarate cofactors for demethylase activity. (**B**) N6-Methyladenosine (m6A) is added to mRNA via the METTL3-METTL14 heterodimer, which along with accessory proteins forms the N6-methyltransferase complex using SAM as methyl donor. The modification is removed by ALKBH5 or FTO requiring oxygen, iron, and α-ketoglutarate cofactors. (**C**) 5-Methylcytidine (m5C) on cytosolic tRNAs is added by DNMT2, NSUN2, and NSUN6, while m5C on mRNA is introduced by NSUN2 only. It undergoes further oxidative metabolism mediated by TET enzymes (and possible other unknown enzymes) to 5-hydroxymethylcytidine (hm5C), requiring oxygen, iron, and α-ketoglutarate. Further TET activity results in formation of 5-formylcytidine (f5C) using hm5C as precursor. These modifications have not been shown to be fully reversible although it is predicted that there are erasers for m5C, or its metabolites [46–48].

Interestingly, recent studies show that in vivo FTO preferentially targets N6-2 -*O*-dimethyladenosine (m6Am), a highly prevalent mRNA modification found adjacent to the N7-methylguanosine cap at the first encoded nucleotide position, with nearly 100-times greater catalytic activity compared to m6A [49,50]. Using a transcriptome-wide map of m6Am, it was shown that the presence of this modification in the extended cap confers increased mRNA stability by reducing the susceptibility to DCP2-mediated decapping, thus influencing mRNA abundance and protein synthesis [49]. The application of stable isotope metabolic tracing demonstrated that oxidative processing is not limited to methylated adenosine residues and that 5-methylcytosine in RNA undergoes similar

oxidative metabolism via 5-hydroxy- and 5-formylcytosine [47]. While the ten-eleven translocation (Tet) family of Fe(II)- and α-ketoglutarate-dependent dioxygenases, well known for their oxidation of m5C in DNA, have been shown to also promote the formation of hm5C in RNA [46], there is substantial evidence that TETs are not the only/main family of enzymes able to catalyze the oxidation of m5C in RNA. For instance, significant amounts of hm5C are not only found in TET triple knockout mouse embryonic stem cells but also in organisms that do not express any TET enzymes, such as *C. elegans* [46,47]. In support of this hypothesis, the formation of 2 -*O*-methyl-5-hydroxymethylcytidine (hm5Cm), a modification closely related to hm5C, has been shown to be TET-independent [48]. It remains unclear whether hm5C is a metabolic intermediate in the demethylation pathway and dynamic regulation of m5C, or whether it is an epitranscriptomic mark with its own function. The involvement of hm5C in an active demethylation pathway is supported by metabolic labeling studies in human embryonic kidney 293 cells that demonstrated that in small RNAs (<200 nt) hm5C itself or hm5C-containing transcripts are subject to enhanced turnover [48]. However, sucrose gradient fractionation followed by dot blotting in *Drosophila* S2 cells revealed that mRNAs heavily loaded with ribosomes have a high hm5C content, suggesting that the function of RNA hydroxymethylation is to promote mRNA translation in vivo [51]. While RNA methylations have been at the forefront of dynamic examples of ribonucleoside modifications, it is likely that other examples of dynamically-regulated or reversible modifications will follow. These initial examples of RNA demethylases point towards the responsiveness of RNA modifications to environmental stimuli that allow organisms to react and adapt to changing environments.

#### *3.2. tRNA Modifications Prevent Translational Infidelity and Proteotoxic stress*

Proper anticodon–codon pairing and maintaining the correct reading frame on translated mRNA are key functions linked to tRNA modifications. Studies specific to wobble uridine U34-based modifications and their corresponding writers have been published highlighting how the C5 and C2 position on U34 play key roles in preventing protein synthesis errors. For example, the writer tRNA methyltransferase 9 (Trm9) from yeast completes the formation 5-methoxycarbonylmethyluridine (mcm5U) and mcm5s2U by adding the terminal methyl group. The mcm5U and mcm5s2U modifications are found at U34 on tRNAs that decode arginine, glutamine, glutamic acid, and lysine. The arginine codons AGA and AGG are found in a split codon box with the codons AGU and AGC for serine. The mcm5U modification is needed to prevent pairing of tRNAArg, which normally decodes AGA and AGC, with the AGU and AGC codons for serine. As such, cells deficient in Trm9 and mcm5-based modifications show increased levels of arginine misincoporation at serine codons [52]. In addition, *trm9*Δ cells show increased −1 frameshifts and activation of heat shock and unfolded protein response (UPR) pathways. Studies in yeast using cells deficient in the wobble uridine writers for s<sup>2</sup> (Ncs2) and c<sup>5</sup> (elp6) have employed a novel reporter system and ribosome profiling to demonstrate that there is increased proteotoxic stress due to tRNA modification defects and perturbed translation [53]. Additional studies in yeast have also demonstrated that wobble U tRNA modifications play important roles in maintaining reading frame, with unmodified or under modified tRNA not entering the ribosome A-site efficiently [54,55]. As the levels of mcm5U and mcm5s2U change during the cell cycle and in response to specific exposures [52], there could be dynamic changes in translational fidelity during stress responses.

#### *3.3. tRNA Modification Enzymes as Essential Features of the Cell Stress Response*

Many genes encoding tRNA-modifying enzymes are essential for cell function, with losses causing severe defects in growth and development [56,57]. However, despite their conservation throughout evolution, many genes encoding tRNA modification enzymes are not essential under normal growth conditions. tRNA normally contains many modifications and is considered to be a stable RNA. Notably though, the loss of key epitranscriptomic marks can destabilize tRNA and lead to rapid degradation. For example, mature tRNA for valine missing m5C and m7G, specific to tRNA methyltransferase (Trm) 4 and Trm8, is surveyed and targeted by the rapid tRNA decay (RTD) pathway [58,59]. The ability to generally function in the absence of a tRNA modification is illustrated by the fact that tRNAs lacking modifications (e.g., in vitro transcribed tRNAs) still function in translation and that cells can compensate for lost modifications by increasing tRNA copy numbers to drive translation [60,61]. The absence of a modification often causes subtle phenotypic effects in cells, so it has been challenging to elucidate the exact biological functions of more than a few modifications. However, the deletion of writer proteins often increases the cellular sensitivity to specific stresses, demonstrating the importance of RNA modifications in adaptation to environmental changes. For instance, depletion of N1-methyladenosine (m1A) in the hyperthermophilic bacterium *Thermus thermophilus* resulted in a thermosensitive phenotype, suggesting a role of m1A in temperature adaptation [62]. In the same organism, loss of pseudouridine at tRNA position 55 (ψ55) caused abnormal increases in the levels of other modified nucleosides (Gm, m5s2U, and m1A) and led to growth retardation at lower temperatures [63]. An emerging literature now documents the critical roles of tRNA modifications in the cellular response to physiological changes, environmental changes, and stressful exposures [12,64–66].

#### **4. Mechanisms by which tRNA Modifications Function in the Cell Stress Response**

#### *4.1. m1A Affects Translation Initiation during Cell Stress*

Initiation of protein synthesis in eukaryotes is mediated by eukaryotic initiation factors (eIFs) and involves the assembly of the initiator tRNA, the 40S, and 60S ribosomal subunits into an 80S ribosome at the start codon of mRNAs, typically AUG coding for methionine [67]. The initiator methionyl-tRNA (tRNAiMet) is used exclusively during initiation of protein synthesis and is different from the elongator methionyl-tRNA, which is solely used for insertion of methionine into a growing polypeptide chain [68]. Initiator tRNAs carry a highly conserved N1-methyladenosine residue at position 58 (m1A58) that is key for the formation of a tertiary substructure not seen in elongator tRNAs (Figures 1A and 3A) [69]. In yeast, m1A58 is essential for viability. Impaired function of m1A methyltransferases resulted in growth arrest and ultimately cell death, which could be attributed to rapid and specific degradation of mature tRNAiMet in the absence of m1A58 by the use of pulse-chase experiments [70,71]. Hence, m1A58 provides direct means of regulating the intracellular levels of tRNAiMet and thus the initiation of protein synthesis without affecting the elongation step of translation. The eraser ALKBH1 was recently shown to mediate the demethylation of m1A in tRNAs, providing evidence of reversible tRNA methylation [39]. Interestingly, glucose deprivation of HeLa cells resulted in increased expression of ALKBH1 which correlated with decreased levels of m1A and attenuated protein synthesis [39]. The levels of the m1A58 methyltransferase heterodimer Trmt6/Trmt61 showed no significant changes [39], suggesting the presence of an active demethylation pathway of m1A under glucose starvation. These studies provide the first evidence that reversible N1-adenosine methylation is involved in the control of protein synthesis in response to nutrient availability.

#### *4.2. Wobble tRNA Modifications Regulate Codon-Biased Translation of Stress Response Proteins*

Post-transcriptional modifications at the wobble position in the anticodon loop of transfer RNAs have direct means to influence the decoding of the genetic code by mediating codon–anticodon interactions. Depending on the type of chemical modification, the interaction with certain codons is preferred due to increased stabilization of specific base pairs. tRNA wobble modifications should be able to regulate the decoding rates of synonymous codons which differ by the third nucleoside, with dynamic changes in modification levels regulating translation. Using a unique bioanalytical platform, we have shown that stressors cause the reprogramming of dozens of modified ribonucleosides in tRNA that regulate the selective translation of codon-biased mRNAs of critical stress response proteins required for cell survival [11,72]. For example, exposure of yeast to the oxidizing agent hydrogen peroxide (H2O2) caused an increase in 5-methylcytosine (m5C) found in the wobble position of leucine

tRNA that reads UUG (1 of six leucine codons) (Figure 1). Codon reporter systems and proteomic studies have linked m5C to the enhanced translation of UUG-enriched mRNAs for oxidative stress response genes (Figure 3B). The role of m5C-based translation of UUG codons in cell survival was demonstrated when loss of the writer tRNA methyltransferase 4 (Trm4), catalyzing the formation of wobble m5C, rendered the cells sensitive to H2O2 exposure. Similar results have been shown for writers of mcm5U and mcm5Um modifications in yeast and mouse models, as deficiencies lead to sensitivity to alkylating and oxidizing agents, respectfully [12,73,74]. In both cases, codon specific reporter constructs, transcript- and protein-based studies, and increased levels of wobble uridine modifications in response to stress have supported the idea that stress promotes the translation of codon specific transcripts, which is coordinated by epitranscriptomic reprogramming. A similar phenomenon has been shown to occur in mycobacteria exposed to hypoxia [75]. This stress led to increased steady-state levels of proteins derived from ACG-enriched genes, with the increase dependent upon stress-induced conversion of mo5U to cmo5U at the wobble position of the UGU anticodon on the threonine tRNA that reads the ACG codon (Figure 3B).

**Figure 3.** Stress-induced changes in tRNA modification levels can regulate (**A**) translation initiation, (**B**) translation elongation, and (**C**) tRNA cleavage.

#### *4.3. tRNA Modifications Restrict Stress-Induced tRNA Cleavage*

A third role for modified nucleosides in tRNA involves regulation of tRNA degradation and cleavage of tRNAs into small regulatory RNA fragments (Figure 3C). The latter is illustrated by angiogenin-mediated endonucleolytic cleavage of tRNAs in the anticodon loop, which is a widely conserved oxidative stress response in eukaryotes [76–78]. One example of modification-dependent

angiogenin cleavage of tRNA under stress involves 5-methylcytidine (m5C). In nuclear-encoded eukaryotic tRNAs, m5C commonly occurs at six cytidine positions, namely, C34 and C38 in the anticodon loop; C48, C49, and C50 in the variable region; and C72 in the acceptor stem [79]. Cytosine-C5 methylation in mitochondrial encoded tRNAs is restricted to the variable loop and the acceptor stem and appears at C48, C49, and C72 [80]. Introduction of methyl groups to these sites is mediated by several members of a large protein family of conserved RNA m5C-methyltransferases, namely tRNA aspartic acid MTase 1 (TRDMT1, also known as DNMT2) and the NOP2/Sun domain proteins (NSUN) 2 and 6 [79,81]. While TRDMT1 specifically methylates position 38 in glycine, aspartic acid, and valine tRNAs [76,82], NSUN6 is responsible for the methylation of position 72 in threonine and cysteine tRNAs [81]. NSUN2 has wider substrate specificity and methylates the vast majority of tRNAs at positions 34, 48, 49, and 50 [83,84]. Studies by Blanco et al. revealed that the absence of NSUN2-dependent m5C sites in the variable loop leads to increased tRNA cleavage by the endonuclease angiogenin and the accumulation of 5 tRNA-derived small RNA fragments in mouse models and dermal fibroblasts obtained from patients with Dubowitz-like syndrome [85]. These 5 tRNA fragments induce cellular stress responses that lead to reduced protein translation rates, decreased cell size, and increased cell death in vitro and in vivo causing a syndromic disorder characterized by growth and neurodevelopmental deficiencies [85]. Similar to NSUN2-mediated methylation in the variable region, DNMT2-mediated cytosine-C5 methylation in the anticodon loop can also protect tRNAs from endonucleolytic cleavage by angiogenin under stress in *Drosophila* [76]. The prominent role of Dnmt2 in the stress response is further confirmed by the fact that *Drosophila* Dnmt2 mutants show significantly reduced viability under oxidative or heat stress [76]. While the presence of m5C limits the fragmentation of tRNA at various locations under stress, it is currently unclear how cytidine-C5 methylation modulates the activity of stress-induced endonucleases. The lack of m5C could result in a more flexible tRNA structure in which the anticodon loop becomes more exposed to tRNA cleavage enzymes. Alternatively, the modification could mask the sequence recognition motif of these enzymes.

m5C undergoes oxidative processing to hm5C [46–48]. It has been shown by quantitative isotope dilution-mass spectrometry that hm5C is enriched in tRNA fractions of HEK293T cells [48]. This provides evidence that m5C is dynamically controlled in tRNAs. While it is currently unclear whether hm5C also protects tRNAs against stress-induced cleavage, metabolic labeling studies have shown that hm5C is subject to enhanced turnover in RNA either due to specific tRNA degradation/cleavage or reversibility to unmodified C [48]. The oxidation of m5C to hm5C could provide means of dynamically regulating the cleavage potential of tRNAs under changing environmental conditions. Further studies should address the exact positions of hm5C and how this oxidative derivative of m5C affects angiogenin-mediated cleavage of tRNAs.

Another example of a link between tRNA modifications and stress-induced tRNA cleavage involves queuosine restriction of tRNA cleavage during oxidative stress. Queuosine (Q) is a hypermodified residue found at the wobble position of tRNAs with GUN anticodons, namely histidine, asparagine, tyrosine, and aspartic acid (Figure 1A) [86]. While bacteria can synthesize Q de novo by a complex biosynthetic pathway [87], eukaryotes lack its synthesis pathways and rely on the uptake of the micronutrient queuine from dietary sources and the gut microbes, for subsequent enzymatic incorporation into tRNA [88,89]. Since Q-deficient mice do not exhibit any pathological symptoms in a stress-less environment, it has been suggested that the role of Q may be to protect the organisms against stress. A link between Q and the oxidative stress response in mice was established when it was shown that exogenous administration of queuine to mice with Dalton's lymphoma ascites transplanted (DLAT) tumors improved the activities of antioxidant enzymes, such as catalase, superoxide dismutase, and glutathione peroxidase [90]. While this promotion of the antioxidant defense system could be evoked by queuine itself or by Q-modified tRNA, Wang et al. recently showed that Q-deficient HEK293T and HeLa cells produce significantly more tRNA halves from tRNAHis and tRNAAsn upon arsenite stress and angiogenin treatment (Figure 3), suggesting that Q directly protects its cognate tRNAs against

ribonuclease cleavage [91]. The total tRNA pool is not altered. In mammals, Q can be further modified by an unknown glycosyltransferase by addition of a mannosyl or galactosyl group to yield manQ or galQ, respectively [92]. The current lack of high-throughput methods for the simultaneous detection and quantification of Q, manQ ,and galQ from limited starting material, has hampered the studies of the Q-derivatives and thus their exact physiological roles remain poorly understood.

Modifications do not always restrict the cleavage of tRNAs. For instance, wobble mcm5s2U is a target for the eukaryotic γ-toxin secreted by Kluyveromyces lactis killer strains and promotes the cleavage of tRNAs, causing irreversible growth arrest of sensitive yeast cells [93]. The importance of modified ribonucleosides is not limited to the eukaryotic host defense response. Colicin E5 and PrrC are *E. coli* endoribonucleases that specifically cleave Q- and 5-methylaminomethyl-2-thiouridine (mnm5s2U)-modified tRNAs, respectively [94,95]. These examples demonstrate that tRNA modifications can be critical determinants in defending host cells from the invasion of viruses or biotic stresses.

#### *4.4. tRNA Modifications Affect tRNA Maturation during Stress*

Eukaryotic tRNAs are initially transcribed as larger precursors (pre-tRNAs) that require a variety of post-transcriptional alterations to become fully mature and functional tRNAs. These processing steps include the removal of the 5 -leader and 3 -trailer sequences, addition of the nucleotides CCA to the 3 -end, intron splicing, and introduction of a large number of ribonucleoside modifications [96,97]. Using Northern blot analysis and RNA sequencing, it was shown that tRNA maturation is differentially regulated during temperature and nonfermentable carbon source stress in yeast. Accumulation of aberrant tRNA precursors was observed upon shifting yeast to elevated temperatures and/or to glycerol-containing medium [98]. Interestingly, several tRNA modifications are added at the pre-tRNA stage [97,99], ensuring proper folding. For instance, studies in *Xenopus oocytes* showed sequential addition of base modifications during tRNA tyrosine maturation [100]. While in this particular organism, m1A, ψ, and m5C already occur in the pre-tRNA with immature 5 -leader and 3 -trailer sequences, m<sup>2</sup> 2G, m2G, and D are introduced after maturation of the 5 - and 3 -termini but before intron splicing [100]. The order and location of incorporation of some modifications can be species-specific. For instance, wobble inosine modifications are incorporated into pre-tRNAs in the nucleus in human tRNAs and into mature tRNAs in the cytosol in *Trypanosoma*, respectively [101,102]. These findings demonstrate that tRNA modifications are stringently coupled with tRNA processing and maturation and suggest that environmental cues can affect tRNA precursor forms by controlling tRNA modification levels in a tRNA- and condition-specific manner.

#### **5. RNA Modifications as Potential Biomarkers of Exposure and Disease Pathology**

Based on these diverse roles for tRNA modifications in the cell stress response, it is reasonable to propose that ribonucleoside modifications can serve as biomarkers of specific stresses and environmental changes. Support for this idea comes from the observed role of tRNA modifications as sensors for changes in environmental and intracellular conditions. For instance, mitochondrial t 6A is sensitive to intracellular CO2 [103]. The growth of HEK293T cells in sodium bicarbonate-free medium in the absence of CO2, caused a significant decrease in the frequency of t6A in mitochondrial tRNAs (mt-tRNAs), which could be rescued by the addition of sodium bicarbonate to the cell culture medium [103]. As a result, Lin et al. speculated that hypoxic conditions in solid tumors could affect t 6A formation as mitochondrial CO2 is predominantly provided by the TCA cycle. In support of this, they found hypomodification of t6A37 in mitochondrial tRNA serine isolated from solid tumor xenografts [103].

Similarly, agents and exposures that cause macromolecular damage lead to the predictable reprogramming of RNA modifications. The predictive power in tRNA modifications is illustrated with the response of yeast exposed to four oxidants and five alkylating agents [11]. tRNA modification patterns accurately distinguished between the two types of toxicant, with 14 modified ribonucleosides forming the basis for a data-driven model that predicted toxicant chemistry with >80% sensitivity and specificity [11]. tRNA modification subpatterns also distinguished among chemically similar toxicants such as SN1 and SN2 alkylating agents [11]. This distinction further linked to codon-biased translation: SN2-induced increases in m3C in tRNA led to selective translation of threonine-rich membrane proteins from genes enriched with ACC and ACT degenerate codons for threonine [11]. tRNA modifications thus serve as predictive biomarkers of exposure.

However, other types of stress response can also promote characteristic epitranscriptomic changes. Looking beyond tRNA, analysis of all cellular RNAs (including mRNA, rRNA, tRNA, and snoRNA) at the nucleoside level has been used to demonstrate that there is broad reprogramming of mRNA-, rRNAand, tRNA-based modifications in response to osmotic stress. For example, using yeast, it was observed that there were dramatic increases in monomethylated C, representing m3C, m5C, Cm, and m4C in mRNA, rRNA, tRNA, and snoRNA during the osmotic stress response [64]. Also, the tRNA-based wobble U34 modifications mcm5U and mcm5s2U, as well as i6A at position 37, were increased in response to osmotic stress. Mechanistically the reason for osmotic stress induced changes in the epitranscriptome could be to promote RNA stabilization, RNA localization as well as translational regulation. Regardless of the mechanistic details, the observation that there are stress-induced changes in modification levels specific to mRNA- and rRNA-based species supports the idea that global epitranscriptomic changes are ingrained in regulatory responses.

The preceding examples with cultured cells and specific stresses illustrate the potential for tRNA modifications to serve as biomarkers of disease and pathology. Defects in RNA modification or their corresponding writers have been linked to cancer, neurodegenerative and neurological defects, and diabetes [4–6]. For example, defects in Q levels have been observed in ovarian and lung tumors [104,105]. Defects in the mitochondrial and nuclear writer of dimethylguanosine (m2,2G), tRNA methyltransferase 1 (TRMT1), have been shown to promote intellectual disabilities, with this epitranscriptomic system linked to redox metabolism [106]. Type II diabetes has also been linked to defects in tRNA modifications, with sequence variants and a mouse model defective in the 2-methylthio-N6-threonylcarbamoyladenosine (ms2t 6A) writer CDKAL1 demonstrating how modification of tRNA for lysine plays an important role in maintaining pancreatic islet function and controlling glucose levels [107,108]. Expression of RNA modification systems have also been linked to cancer survival, as the wobble U modification writer enzyme ALKBH8 has been shown to be required for growth of bladder cancers [109], while the ALKBH3 eraser of m6A in RNA is required for survival of non-small cell lung cancer and other cancers [110,111].

#### **6. Conclusions and Perspectives**

The dynamic regulation of RNA modifications plays an important role in the response of cells to environmental fluctuations and xenobiotic exposures. On average there are 13 modifications in each tRNA and many distinct tRNA isoacceptors in each cell. Global analyses have demonstrated that there are coordinated changes in tRNA modifications, for example, the alkylation-induced increases in mcm5U and mcm5s2U occurring as there are increases in m3C and m7G [11]. These coordinated changes in epitranscriptomic marks suggest that a program of translational regulation specific to many codon–anticodon pairs is driving the translational response to stress. While almost all tRNA modifications occur in multiple tRNAs at several positions, toxicant-induced changes are affecting very specific positions in individual tRNAs. It is currently not understood, how multiple tRNAs bearing the same modification can be differentially regulated when the sites are targeted by a single enzyme. RNA modification enzymes may rely on the presence of other modifications for the introduction of a modification, as observed for the Dnmt2-mediated introduction of m5C, which is stimulated by the presence of Q at position 34 [112]. The investigation whether such crosstalk between modifications is a general phenomenon requires the development of novel techniques allowing the single-base resolution mapping of individual modifications, such as bisulfite sequencing [113]. The presence of certain modifications could lead to structural changes in the three-dimensional structure of tRNA, making

other sites more accessible for modification enzymes that were previously hidden. Furthermore, it will not only be essential to unveil how tRNA modifications are coordinately regulated, but also to start integrating the tRNA modification landscape with other epitranscriptomic marks on messenger RNA, such as m6A (Figure 4). So far, tRNA, mRNA, and modifications of other RNA types have all been considered as separate entities and their interplay remains completely unstudied. However, some enzymes, like NSUN2, target multiple RNA species for modification, pointing towards the existence of interaction systems between multiple RNA modification systems. It is also important to place translational regulation in the context of a larger program of stress-induced epitranscriptomic changes, which should be regulating tRNA stability, translation initiation, and microRNA-based regulation of transcripts. The observation that there are stress-induced changes in modification levels specific to mRNA- and rRNA-based species supports the idea that global epitranscriptomic changes are ingrained in regulatory responses [64]. Linking tRNA modification-based translational regulation in the context of mRNA modification-based regulation should be an active area of research in the future. There is abundant data to show that the most prevalent mRNA modification—m6A—plays dynamic roles in regulating fertility and development, and we suggest that tRNA-based and other epitranscriptomic marks should be important for human development. Further, the identification of m6A erasers in the form of demethylase enzymes highlights the dynamic potential of epitranscriptomic marks and suggests that a wide array of tRNA-specific erasers should be present in human cells.

**Figure 4.** Coordinated changes in epitranscriptomic marks on tRNA, mRNA, rRNA, and snRNA are theorized to drive the translational response to stress, and other physiological responses.

**Author Contributions:** The manuscript was written by S.M.H, P.C.D, and T.J.B. The figures and legends were produced and written by A.L. All authors have given approval to the final version of the manuscript.

**Funding:** This research was funded by grants from the National Institutes of Health, grant numbers R01ES026856 and R01ES024615, and by a grant from the National Science Foundation of Singapore to support the Singapore-MIT Alliance for Research and Technology Antimicrobial Resistance IRG. S.M.H was supported by an Early Postdoc.Mobility Fellowship from the Swiss National Science Foundation (P2SKP\_174681).

**Acknowledgments:** We would like to thank members of the Dedon and Begley labs for constructive comments.

**Conflicts of Interest:** The authors declare no conflicts of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Determinants of Hair Manganese, Lead, Cadmium and Arsenic Levels in Environmentally Exposed Children**

#### **Thomas Jursa 1, Cheryl R. Stein <sup>2</sup> and Donald R. Smith 1,\***


Received: 5 February 2018; Accepted: 20 March 2018; Published: 22 March 2018

**Abstract:** Biomarkers of environmental metal exposure in children are important for elucidating exposure and health risk. While exposure biomarkers for As, Cd, and Pb are relatively well defined, there are not yet well-validated biomarkers of Mn exposure. Here, we measured hair Mn, Pb, Cd, and As levels in children from the Mid-Ohio Valley to determine within and between-subject predictors of hair metal levels. Occipital scalp hair was collected in 2009–2010 from 222 children aged 6–12 years (169 female, 53 male) participating in a study of chemical exposure and neurodevelopment in an industrial region of the Mid-Ohio Valley. Hair samples from females were divided into three two centimeter segments, while males provided a single segment. Hair was cleaned and processed in a trace metal clean laboratory, and analyzed for As, Cd, Mn, and Pb by magnetic sector inductively coupled plasma mass spectrometry. Hair Mn and Pb levels were comparable (median 0.11 and 0.15 μg/g, respectively) and were ~10-fold higher than hair Cd and As levels (0.007 and 0.018 μg/g, respectively). Hair metal levels were higher in males compared to females, and varied by ~100–1000-fold between all subjects, and substantially less (<40–70%) between segments within female subjects. Hair Mn, Pb, and Cd, but not As levels systematically increased by ~40–70% from the proximal to distal hair segments of females. There was a significant effect of season of hair sample collection on hair Mn, Pb, and Cd, but not As levels. Finally, hair metal levels reported here are ~2 to >10-fold lower than levels reported in other studies in children, most likely because of more rigorous hair cleaning methodology used in the present study, leading to lower levels of unresolved exogenous metal contamination of hair.

**Keywords:** manganese; lead; cadmium; arsenic; hair; children; environment

#### **1. Introduction**

Exposure biomarkers play an important role in estimating the internal dose of a person exposed to an environmental contaminant, and they are often essential in determining exposure–health effect relationships in epidemiological studies [1–9]. For example, lead (Pb) levels in blood and bone are accepted as well-validated Pb exposure biomarkers, and they helped establish the association between Pb exposure and health risk in children and adults [7,10,11]. Similarly, blood cadmium (Cd) levels have been shown to reflect Cd exposure from environmental sources [8,9]. However, for metals such as manganese (Mn), studies are mixed on whether blood Mn levels are an indicator of exposure and risk of health effects, presenting a need to develop and validate alternative exposure biomarkers [1–3]. Recent studies have suggested that hair Mn levels may help fill this need [2,3,12–14].

A number of studies have reported associations between levels of Mn in hair and Mn exposure and associated health effects in children and adolescents [12–15], including a recent review showing that hair was the most consistent and valid biomarker of Mn-associated health effects in children [3]. Similarly, levels of some other metals in hair, including mercury and arsenic, have been reported as both meaningful exposure biomarkers and indicators of health risks from exposure [16,17]. Hair may provide some practical advantages over other tissues as an exposure biomarker; scalp hair grows at a rate of roughly one centimeter per month, providing a possible indicator of exposure integrated over periods of one to five months or more, depending on the length of collected hair [18,19]. Moreover, analysis of sequential sections of hair may provide useful information on the temporal variability of exposure, although few studies have investigated whether sequential hair segments are useful for retrospective exposure assessment over the duration of hair growth [17].

However, the potential utility of hair as an exposure/effect biomarker is not without some challenges. Most notably, hair is susceptible to contamination from exogenous sources such as dust, water, and use of hair products [14,17,20]. Studies reporting hair metal levels as a biomarker of exposure vary widely in the type of method used to remove exogenous contamination from the hair, with methods varying from a simple water rinse to detergent and acid sonication [2,12,14,17,21–25]. Not surprisingly, reported hair metal levels vary widely by study, though it is unclear if this reflects differences in exposure versus differences in effectiveness of cleaning exogenous contamination. Studies have shown that metal levels in hair are derived largely from exogenous contamination, with the rigor of hair cleaning prior to analyses affecting the contribution of exogenous contamination to the measured hair metal levels [14,20,26].

Here, we determined levels of Mn, Pb, Cd and arsenic (As) in scalp hair samples from 222 male and female children aged 6–12 years living in the Mid-Ohio Valley. For 169 female subjects, hair samples were cut into sequential segments to determine the reproducibility of hair metal levels within the same subject, and the variation in hair metal levels over different seasons of growth. Hair samples were cleaned using a rigorous cleaning method shown to effectively remove exogenous metal contamination [14] and processed for analyses by inductively coupled plasma–mass spectrometry (ICP-MS).

#### **2. Methods**

#### *2.1. Subjects*

Hair samples were collected and processed from 222 subjects age 6–12 years (169 female, 53 male) recruited through the C8 Health Project Neurobehavioral Development Follow-up, which was investigating the neurodevelopmental health effects of perfluoroocatnoate (PFOA) exposure in southeastern Ohio and northwestern West Virginia. A detailed description of subject recruitment, as well as information on subject demographics, residence, and medical histories collected via maternal report at the time of the neurodevelopment follow-up study in 2009–2010 is provided elsewhere [27]. Mothers provided informed consent and children provided verbal assent; child and mother each received \$50 for participation. The Mount Sinai Program for the Protection of Human Subjects and the Battelle Centers for Public Health Research & Evaluation Institutional Review Board approved all study procedures. Investigations were carried out following the rules of the Declaration of Helsinki of 1975 (https://www.wma.net/what-we-do/medical-ethics/declaration-of-helsinki/), revised in 2008. Relevant to the present study, this Ohio River Valley region also hosts the longest operating ferromanganese refinery in North America in Marietta, OH (Eramet Marietta, Inc., Marietta, OH, USA), and studies by others have reported elevated Mn exposures and associated health effects in children in the region [2,28].

Hair samples were collected proximal to the occipital lobe scalp and stored in zip top plastic bags at room temperature until analysis. Male subjects provided a single 2 cm segment of hair proximal to the scalp, while female subjects provided longer hair samples that were cut into sequential 2 cm segments (0–2, 2–4, 4–6 cm from the scalp), based on the overall length of the sample. A single 2 cm hair sample was analyzed for all male subjects (*n* = 53), since most males had hair too short to provide multiple segments. For females (*n* = 169), three sequential 2 cm hair segments were available for 159 subjects (referred to as proximal, medial, and distal 2 cm segments, relative to the scalp), two hair segments were available for *n* = 4 females (*n* = 3 for proximal and medial, *n* = 1 for proximal and distal), and one hair segment was available for *n* = 6 females (all proximal).

#### *2.2. Experimental*

All cleaning and processing of hair samples was conducted in a HEPA filtered-air trace metal clean room, using acid-cleaned labware and ultrapure trace metal grade reagents. Individual hair segments/samples weighing 5–30 mg each were cleaned of exogenous metal contamination as described previously [14]. Briefly, samples were placed in acid-cleaned 5 mL polypropylene syringe tubes and sonicated (20 min) in 0.5% Triton, rinsed five-times with ultrapure Milli-Q water, sonicated (10 min) in 1 N trace metal grade nitric acid (Fisher Scientific, Santa Clara, CA, USA), rinsed with 1 N nitric acid, and rinsed five-times with Milli-Q water, and then dried at 65 ◦C for 48 h. Subsequently, samples were digested in 0.5 mL 15.7 N quartz-distilled nitric acid (Fisher Scientific, optima grade) at 80 ◦C for 6 h in a Class-100 HEPA filtered-air fume hood, and then diluted with 5 mL Milli-Q water. For analyses, 0.25 mL of digestate was transferred to an acid-cleaned polyethylene microfuge tube, diluted with 0.25 mL Milli-Q water, and centrifuged at 13,000× *g* for analysis. Rhodium and thallium were added to samples as internal standards, and samples analyzed for Mn, Pb, Cd, and As by magnetic sector inductively coupled plasma mass spectrometry (Thermo Element XR ICP-MS, Waltham, MA, USA), as described elsewhere [1,14]. Methane was added to the argon (Ar) carrier gas to minimize ArCl formation. 208Pb, 111Cd, and 113Cd were measured in low resolution, while 55Mn and 75Ar were measured in medium resolution. Typical analytical limits of detection (LOD's) over five analytical runs were 0.0077, 0.0038, 0.0004, and 0.0018 ng/mL for Mn, Pb, Cd, and As, respectively. For metal levels below the analytical LOD, the LOD was multiplied by 0.5 and adjusted using the sample dilution factor and sample weight of processed hair to derive a value for half the procedural detection limit, and that value included in the data set for statistical analyses. Standard reference material (SRM) NIES 13 (human hair) was used to assess analytical accuracy; mean SRM recoveries (% recovery ± % RSD) based on 17 replicates over five analytical runs averaged 101 ± 10 for Pb and 98 ± 6 for Cd (both certified values), and 75 ± 14 for Mn and 98 ± 9 for As (both non-certified reference values).

#### *2.3. Data Analyses*

Summary data are expressed as median or mean ± standard error (SE), or mean ± standard deviation (SD). If necessary, data were square root- or log-transformed to achieve normality and variance equality. To examine within and between subject variation in hair metal levels, data were analyzed using mixed models with metal concentration as dependent variable, hair segment as independent variable, and a repeated statement identifying subject (within subject referring to separate segments of the same strand of hair). Models were adjusted for subject age (continuous) and season of collection, with four seasonal periods (August 2009–October 2009; November 2009–February 2010; March 2010–May 2010; June 2010–August 2010, selected to reflect different climatological seasons that were balanced by the number of subjects). A within-subject effect of hair segment on metal concentration was identified by a significant Type 3 effect. To determine whether the within-subject variation differed by tertile of hair metal concentration, we stratified models by tertile of hair metal concentrations and calculated least square means with a Tukey adjustment. We qualitatively examined the upper and lower confidence bounds of the difference in hair metal levels across tertiles. If the difference and bounds were comparable across tertiles then we concluded that the segment effect on hair metal levels did not differ by tertile. Lastly, to assess whether the segment effect differed by season we added a segment–season interaction term to the unstratified models and identified differences by a significant Type 3 effect. A *p*-value ≤ 0.05 for the various outcomes was considered statistically significant. All data were analyzed using SAS (Version 14.1) or JMP (Version 13.0) software (SAS Institute Inc., Cary, NC, USA, 2016).

#### **3. Results**

#### *3.1. Hair Metal Levels in Children Vary by Several Orders of Magnitude between Subjects and Were Highly Correlated*

In this study of 222 subjects (53 male, 169 female), median levels of Mn and Pb in the proximal segment of children's hair were comparable at 0.109 μg/g (range 0.005–4.10 μg/g) and 0.152 μg/g (range 0.008–7.73 μg/g), respectively (Table 1, Figure 1). Hair levels of Cd and As were both about 10-fold lower than Mn and Pb levels, with median Cd levels of 0.007 μg/g (range 0.0005–0.463) and median As levels of 0.018 μg/g (range 0.004–0.438). Notably, for all four metals there was a ~>100-fold range in proximal segment metal levels between subjects, suggesting substantial differences in the exposure levels and/or incorporation of metal levels into hair between subjects. Moreover, levels of all four metals were higher in the proximal segment of males compared to the proximal segment of females (*p*'s < 0.0001), based on mixed model analysis on log10 transformed data with sex (fixed) and subject (random) factors (*n* = 157 female, 45–53 male, depending on metal). The large majority of hair samples possessed metal levels that were above the analytical detection limits (>97% for Mn, Pb, and Cd, 92% for As).



\* The reported '*n*' and the ranges in metal levels are for measured values above the limit of detection (see text).

Associations between the levels of different metals in hair could suggest similar exposure sources. Spearman's correlation analysis of metal levels in all hair samples of male (proximal segment) and female (proximal, medial, and distal segments) subjects shows that all four metals are highly correlated (*p* < 0.0001), with the correlation between Cd and Pb being strongest (Spearman's *ρ* = 0.5780), followed by the correlation between Mn and Cd (*ρ* = 0.4475), and Mn and Pb (*ρ* = 0.3513). Correlations between As and the other three metals were weaker (*ρ* < 0.21) (Table 2).

**Mn Pb** 0.3513 (530) **Pb Cd** 0.4475 (532) 0.5780 (529) **Cd As** 0.1872 (526) 0.1814 (523) 0.2081 (526)

**Table 2.** Spearman's correlation (*n* = number of hair samples) between hair metal levels in all hair samples from males (proximal segment) and females (proximal, medial, and distal segments); all Spearman's *ρ* values are highly significant, *p* < 0.0001).

**Figure 1.** Hair metal concentrations (μg/g, note log scale) in the proximal 2 cm hair segment for male (*n* = 53) and female (*n* = 169) subjects. The horizontal line within the box represents the median, while the upper and lower margins of the boxes represent the 75th and 25th percentiles; whiskers are drawn to the furthest data point within 1.5-times the interquartile range. *N* = 214–222; hair metal values below the limit of detection are excluded.

#### *3.2. Variance in Hair Metal Levels between-Subjects is Much Greater than within-Subjects, and Hair Mn, Pb, and Cd, but Not As Concentrations Increase from Proximal to Distal Segments*

To determine whether metal levels in hair segments grown over a period of one to two months were more variable between subjects than between adjacent hair segments within a subject, hair samples from female subjects, who typically provided hair samples sufficiently long for segmentation, were cut into sequential two centimeter segments (designated proximal, medial, and distal segments relative to the scalp) for analyses; each two centimeter segment is assumed to reflect roughly two months of hair growth and metal exposure [18,19]. Variance component analysis was used to determine the contribution of subject and hair segment (as variance components) to the variation in hair metal levels between and within subjects. The subject variance component accounted for 65–73% of the variance in hair metal levels, and was substantially greater than the within-subject (i.e., hair segment) variance component, which accounted for 0.1–4% of the variance in hair Mn, Pb, and As, and 12% of the variance in Cd (Table 3). The remainder of the variance in hair metal levels (i.e., 17–35%) was accounted for by the subject–hair segment interaction (Table 3).


**Table 3.** Percent of variance in hair metal levels attributed to between-subject, within-subject (i.e., hair segment), and subject–segment interaction components according to interclass correlation analysis.

Given the modest contribution of the hair segment (i.e., within-subject) factor to the variability in hair metal levels in female subjects (i.e., 4–12% for Mn, Pb, and Cd), data for the proximal, medial, and distal two centimeter hair segments of female subjects were analyzed to determine if the metal concentrations *systematically* varied between segments of hair within subjects. To facilitate this, since hair metal concentrations varied substantially between subjects, hair metal concentrations for each segment were expressed as a percentage of the average of the three segments for each female subject, and the geometric mean across subjects of the normalized (%) value for each segment was then calculated for the subjects. Results show that Mn, Pb, and Cd, but not As hair metal concentrations systematically *increased* from the proximal to distal segment (Figure 2a). The relative increase in metal levels from proximal to medial to distal segments was comparable for Mn (72%, 94%, 111% of the three-segment average, respectively) and Pb (74%, 90%, 114% of the three-segment average), and slightly greater for Cd (57%, 92%, 133% of the three-segment average). In contrast, As concentrations were relatively invariant across the proximal, medial, and distal hair segments (98%, 91%, 96% of the three-segment average, respectively) (Figure 2a). Consistent with this, in mixed model analysis with a repeated statement identifying subject, a Type 3 test for fixed effects revealed a significant effect of hair segment on Mn, Pb, and Cd levels (*p* < 0.001, *p* = 0.0015, and *p* < 0.001, respectively), but no significant variation between hair segments in As concentration (*p* = 0.658).

To explore whether this systematic relative increase in hair Mn, Pb, and Cd levels from proximal to distal segments was comparable for subjects with low versus high hair metal levels, we stratified subjects into tertiles based on their three-segment average hair metal concentrations and performed mixed model analysis with a repeated statement identifying subject, as above. For perspective, the systematic ~40–70% relative increase in hair Mn, Pb, and Cd concentrations from the proximal to distal hair segments noted above, while significant, is small compared to the ~100–1000-fold difference in hair metal levels between subjects (Table 1), or the ≥5-fold difference in mean hair metal levels of the lowest and highest tertiles of hair metal levels (Figure 2b). Interestingly, mixed model results show that the relative increase in metal levels from proximal to distal hair segments does not vary by tertile of hair metal concentration for any of the metals (*p*'s = 0.2–>0.9). To visualize this, we stratified subjects into tertiles by their three-segment average hair metal concentrations as above, and normalized the metal concentration for each segment to the three-segment average per subject (expressed as a percent), and then calculated the geometric mean for these normalized (%) values for each tertile of hair metal concentrations. These data reflect the mixed model null results noted above by showing that the *relative* increase in hair Mn, Pb, and Cd from proximal to distal hair segments within female subjects does not differ by tertile of hair metal levels (Figure 2c).

**Figure 2.** (**A**) Normalized hair Mn, Pb, and Cd, but not As levels systematically increase from proximal to distal two centimeter hair segments from female subjects. Normalized hair metal levels (%) for each segment were calculated by dividing the hair segment metal concentration by the average of all three segments (proximal, medial, distal) for each individual subject. Data are geometric mean (±SE) for all female subjects (*n* = 153–155 per segment and metal); (**B**) Hair Mn, Pb, Cd, and As concentrations differ by ≥5-fold between the lowest and highest tertiles of hair metal levels. Data are mean (±SE) three-segment average of female subjects segregated into tertiles (*n* = 56–57 per tertile and metal); (**C**) The relative increase in hair Mn, Pb, and Cd levels from proximal to distal segments is comparable for subjects with lower (first tertile) versus higher (third tertile) hair metal levels (see text for details).

#### *3.3. Hair Metal Levels Vary Seasonally*

To explore whether the season of hair sample collection, as a surrogate of possible seasonal differences in exposure or residual exogenous contamination, could explain the systematic increase in hair Mn, Pb, and Cd concentrations from the proximal to distal hair segments, we performed analysis in which a season factor and a segment–season interaction were added to the mixed-model noted above. The four season intervals of August–October 2009, November 2009–February 2010, March–May 2010, and June–August 2010 were selected to align with climate seasons in the Mid-Ohio Valley and to achieve reasonable balance in the number of female subjects across the four season intervals. Results show a significant effect of season on hair Mn, Pb, and Cd levels (*p*'s < 0.001, 0.023, and <0.001, respectively), but not As (*p* = 0.60). There was no season–segment interaction for Mn or Pb (*p*'s > 0.70), although for Cd the interaction was trending towards significance (*p* = 0.073).

This effect of season on hair Mn, Pb, and Cd levels is illustrated by generally lower metal concentrations in samples collected in the late fall to spring seasons, and higher concentrations in hair samples collected in summer to early fall seasons (Figure 3). For example, the lowest hair Mn, Pb, and Cd concentrations (seasonal medians of 0.081, 0.146, and 0.0059 μg/g, respectively; all three segments per subject combined) were for samples collected in the November–February (Mn and Pb) or March–May (Cd) seasons. In contrast, the highest hair Mn, Pb, and Cd levels (seasonal medians of 0.157, 0.235, and 0.0142 μg/g, respectively) were for samples collected in the June–August (Mn) or August–October (Pb, Cd) seasons. Across subjects, median levels of hair Mn increased by ~90% between the two seasons, whereas Pb increased by ~60% and Cd by ~240%.

**Figure 3.** Hair segment Mn (**A**), Pb (**B**), Cd (**C**), and As (**D**) levels in females (μg/g, note log scale) vary with season of collection. Hair metal levels are plotted by three to four months season of collection intervals. The horizontal line within the box represents the median, while the upper and lower margins of the boxes represent the 75th and 25th percentiles; the whiskers are drawn to the furthest data point within 1.5 times the interquartile range. Only female subjects with proximal (P), medial (M), and distal (D) hair segments are shown (*n* = 153–155 subjects per metal).

We similarly performed mixed model analysis restricted to metal levels in the proximal segments of males and females, with sex and season of collection as fixed effects and subject as a random effect. Results show a significant effect of sex (*p*'s < 0.0001) and season (*p*'s ≤ 0.004) on proximal segment Mn, Pb, and Cd levels, with higher metal levels in males and higher levels in hair collected in summer/early fall versus winter/spring. For proximal segment As levels, there was a significant effect of sex (*p* < 0.0001, males higher), but no effect of season of hair collection (*p* = 0.137). There was no sex–season interaction for any of the metals (*p*'s > 0.43), indicating that the season of collection did not differently affect male and female hair metal levels.

#### **4. Discussion**

Hair offers several advantages over other biological tissues/matrices as an exposure biomarker, most notably the potential to retrospectively reconstruct exposures over sequential integrated periods of weeks to months, depending on the length of hair. Here we report Mn, Pb, Cd, and As levels in hair samples from male and female children/adolescents age 6–12 years living in the Mid-Ohio Valley, a region noted for its industrial activity, including the longest operating ferromanganese refinery in North America in Marietta, OH (Eramet Marietta, Inc., Marietta, OH, USA) [2]. This study is unique in that hair samples from females (*n* = 159) were divided into sequential two centimeter segments, and all samples were cleaned prior to analyses using a rigorous cleaning method previously shown to effectively remove exogenous metal contamination [14].

#### *4.1. Correlations between Metals Suggests Some Shared Exposure Sources/Pathways*

The significant correlations among hair levels of all four metals, with correlations of hair Mn, Pb, and Cd being strongest (Spearman's *ρ*'s ~0.35–0.58 for all hair samples, Table 2) suggests some shared environmental sources/pathways for the incorporation of these metals into hair, and possibly also similar chemistries of interaction of several of these metals with hair keratin. The Mid-Ohio valley region has a history of industrial activity, including ferromanganese alloy and perfluoroocatnoate (PFOA) chemical manufacturing, and air monitoring in 2007–2008 measured levels of Cd, As, and particularly Mn that exceeded ATSDR and EPA health-based comparison values [29]. Further, human hair is a complex biological matrix composed predominantly of proteins (65–95%), water (up to 32% by weight depending on its moisture content), lipids, pigment, and trace elements that are coordinated with the functional groups of protein amino acids or with fatty-acid groups of lipids [19]. Though the protein composition of hair may vary across individuals, it is generally rich in polar and charged amino acids, including hydroxyls, amides, acidic and basic amino acids, and disulfides, and these are the components of hair that may readily coordinate with endogenously and exogenously incorporated metals [19,30]. Thus, the stronger correlations between hair Pb and Cd could also reflect similar chemistries of interaction of these two metals with hair keratin; Cd and Pb have similar affinities to sulfur and nitrogen ligands, while Mn coordinates strongly with oxygen ligands [19,30]. Since As is assumed to exist mainly as oxyanion species, it likely chemically coordinates with different functional groups in hair keratin fibers than the cationic metals.

#### *4.2. Hair Metal Levels Vary Substantially More between Subjects than within Subjects, and Hair Mn, Pb, and Cd, but Not As Concentrations Increase from Proximal to Distal Segments*

We found that hair Mn, Pb, Cd, and As levels were higher in males compared to females, and that levels varied by ~100–1000-fold between all subjects (Table 1, Figure 1), but varied comparatively little within subjects, with relative changes of ~40–60% in hair Mn, Pb, and Cd, and <10% in As levels across the proximal, medial, and distal hair segments of female subjects (Figure 3). Consistent with this, variance component analysis showed that the between subject factor accounted for the majority (65–73%) of variance in hair metal levels, while the within subject component (i.e., variation between hair segments within female subjects) accounted for ~4% of the variance in Mn and Pb, 12% in Cd, but very little of the within subject variance in As (0.1%). Together, these findings suggest that hair metal

levels reflect important between subject differences in metal exposure and incorporation of metals into hair.

Notably, concentrations of hair Mn, Pb, and Cd, but not As systematically increased from the proximal to distal two centimeter hair segment in female subjects. Since hair samples were collected from subjects throughout the Mid-Ohio Valley over a 13 months period, it is unlikely that this increase in metal concentrations from proximal to distal hair segments reflects temporal differences in metal exposure common to all subjects. Alternatively, we considered whether residual exogenous metal contamination that remained after rigorous cleaning could account for the increase in hair metal levels across segments. Given that subjects likely inhabited environments with inherently different environmental exposure burdens, as suggested by the 100–1000-fold difference in hair metal levels between subjects, we reasoned that subjects in higher metal exposure environments would have experienced both higher endogenous metal exposures, leading to greater metal incorporation into growing hair, and higher exogenous metal contamination of hair compared to subjects living in lower metal burden environments. Moreover, we considered that the older, distal hair segments likely acquired more exogenous metal contamination than the younger proximal segments, because the older distal segments were in contact with the environment roughly four months longer than the proximal segments, and thus were exposed to a greater cumulative environmental exposure burden. Following this logic with the assumption that environmental exposures from water, air, dust, etc. are likely the primary exposure source(s) for both the endogenous and exogenous components of hair metal levels, the *relative* contribution of residual exogenous contamination to total hair metal levels would scale with (i) the magnitude of environmental metal contamination and the duration of time the hair was in contact with the environment, and (ii) the endogenous (metabolically incorporated) component of hair metal levels. In this case the *relative* (percent) increase in hair Mn, Pb, and Cd levels from the proximal to distal hair segments would be comparable for subjects in the lowest and highest tertiles of hair metal levels, which is consistent with our findings (Figure 2c). Collectively, these findings further suggest that the proximal segment hair metal levels reflect predominantly endogenously-incorporated metals, while distal segment metal levels reflect both endogenously-incorporated metals and a relatively small but notable exogenously-added (contamination) component of hair metals (i.e., Mn, Pb, and Cd), the latter in spite of the rigorous hair cleaning methodology used here [14].

Skröder et al. [17] similarly reported systematic increases in hair Mn, Pb, and Cd, but not As in sequential hair segments over eight-centimeter of hair length in a small number of Bangladeshi children (*n* = 19). In that study the relative increase in hair metal levels with distance from the scalp was 4.6-fold for Mn, and roughly two to three-fold for Pb and Cd-relative increases that are much greater than the ~40–70% relative increase from proximal to distal hair segments observed in the present study. Skröder et al. interpreted their findings as evidence of exogenous metal contamination that was most pronounced for hair Mn levels. In light of (i) the very elevated groundwater Mn levels in the Bangladeshi subjects' environment; and (ii) our studies showing that hair is readily and significantly contaminated from direct contact with Mn-contaminated water and that exogenous Mn contamination from water is incompletely removed even with rigorous cleaning [14], it is likely that the hair Mn levels reported by Skröder et al. are dominated by unresolved exogenous contamination.

#### *4.3. Hair Metal Levels Vary by Season of Collection*

We found that the season of hair collection was associated with hair levels of Mn, Pb, and Cd, but not As, and that there was no season–sex (proximal segments only) or season–hair segment (females only) interaction in the mixed model analyses. This suggests that the main effect of season of hair collection may reflect a contribution of seasonal differences in residual exogenous metal contamination that slightly but measurably contributed to hair metal levels. Since hair samples were collected over a 13 month period, the proximal and distal hair segments of females would have grown over different seasons depending on the season of hair collection [18,19]. Male and female subjects whose hair was collected in late summer/early fall, when outdoor activity and some routes of exposure might be greatest, had higher levels of Mn, Cd, and Pb in proximal hair segments (males and females) and across all three segments of females, compared to hair samples collected in winter and spring. If hair contained only endogenously-incorporated metals, then seasonal differences in exposure would be associated with the season of hair growth, not the season of hair collection as observed.

#### *4.4. Hair Metal Concentrations Reported Here Are Generally Lower than other Studies in Children*

To facilitate inter-study comparison of reported hair metal levels in children, we summarized reported findings from 15 studies of similarly aged children (Table 4). Median or mean hair Mn levels differ by ~120-fold across studies, while hair levels of Pb, Cd, and As differ across studies by ~15-fold, 4-fold, and 180-fold, respectively. These differences between studies may reflect, at least in part, differences in endogenous metal exposure and incorporation of metals from the circulation into hair. However, because of the susceptibility of hair to environmental metal contamination, differences in hair metal levels between studies likely also reflect differences in unresolved exogenous contamination. The listed studies used a variety of different hair cleaning methodologies, from no cleaning to multi-stage cleaning procedures employing various combinations of detergents and/or solvents, weak acid, and sonication (Table 4). For example, Skröder et al. [17] reported median hair Mn levels of 5.0 μg/g in Bangladeshi children exposed to elevated Mn in drinking water, while Hernandez-Bonilla et al. [22] and Menezes-Filho et al. [21] reported hair Mn levels greater than 10 μg/g in Mexican and Brazilian children, respectively, living in the vicinity of ferromanganese alloy plants. These hair Mn levels are nearly two orders of magnitude or more higher than levels reported in the present study, or levels reported by Torrente et al. [31] and Lucas et al. [12] for Spanish and Italian children, respectively, living in areas impacted by industrial emissions. Skröder et al. [17] and Hernandez-Bonilla et al. [22] reported cleaning hair prior to analysis with a Triton detergent wash, while Menezes-Filho et al. [21] and Torrente et al. [31] used a Triton wash with ultrasound sonication. The present study and Lucas et al. [12] used Triton sonication followed by sonication in a 1 N nitric acid solution. While it is difficult to separate the influence of environmental exposure from the efficacy of hair cleaning methods to reduce exogenous hair contamination, these data suggest that studies that used more rigorous cleaning procedures reported lower hair Mn (and generally other metals) concentrations, consistent with studies showing that the rigor of hair cleaning prior to analyses significantly influences hair metal levels from exogenous contamination [14,20,26].

There are several studies from different geographical regions that used the same cleaning method, as well as studies from the same geographical region that used different cleaning methods that can be readily compared to estimate the extent that differences in hair metal levels between studies reflect differences in exposure versus differences in exogenous contamination due to different hair cleaning methods. For example, prior studies from our lab in Italian children exposed to environmental metals from industrial ferroalloy emissions [12,14] used the same Triton sonication followed by dilute nitric acid sonication hair cleaning method as the present study. Hair Mn and Pb levels are very comparable between these studies (Table 4), allowing us to conclude that the subjects in the present study had ~10–30% higher Mn exposure and ~10% lower Pb exposure levels compared to the Italian subjects, based on hair metal levels. We can also compare hair Mn levels between two studies from the same region that used different hair cleaning methods. Haynes et al. [2] used a Triton detergent hair cleaning method (without sonication) and reported geometric mean hair Mn levels of 0.417 μg/g from children in the same Ohio Valley region as the present study, which are greater than three-fold higher than levels in the present study (geometric mean Mn of 0.119 μg/g, median 0.109 μg/g, Table 4). This difference in hair Mn levels between the two studies may be due to differences in Mn exposure between cohorts. However, it may also be that they are due to differences in hair cleaning methods, given that our prior study [14] found that a Triton sonication hair cleaning method similar to that used by Haynes et al. [2] yielded hair Mn and Pb levels that were ~2.5–4-fold higher than hair metal levels following the Triton + weak nitric acid sonication method used in the present study. Collectively, these findings suggest that the extent that hair metal levels reflect endogenous exposure will vary substantially depending on

the hair cleaning method and the extent that cleaning reduces exogenous hair contamination, the latter of which may also vary between subjects and hair type [14,26].

**Table 4.** Comparison of hair metal concentrations in children from the present study with levels reported in children (ages 4–14) in other studies that used a variety of different hair cleaning methods prior to analysis.


\* Study citation number and data type; <sup>a</sup> = data are median (SD); <sup>b</sup> = data are geometric mean (SD, geometric SD, or 95% CI); <sup>c</sup> = data are mean (SD); <sup>d</sup> = data are median (5th–95th percentile). NA = not analyzed or reported. # Hair cleaning methods, T = Triton, N = nitric acid, /S = sonicated, E = ethanol, A = acetone, U = uncleaned. & Study sub-population by gender and study sub-site location when reported (all = both males and females across study sub-sites).

#### *4.5. Hair Metal Levels as a Biomarker of Exposure and Associated Health Effects*

Studies have reported mixed results regarding the extent that hair metal levels are associated with metal levels in environmental media (e.g., water, soil, dust, and airborne particles, etc.), or biomarkers of endogenous metal exposure (blood, urine, nails). Lucas et al. [12] reported low but statistically significant correlations between children's hair Mn and Mn levels in household dust (*ρ*'s ~ 0.27, *p* < 0.001) and airborne particles (*ρ* = 0.126, *p* < 0.05). Bouchard et al. [13] and Oulhote et al. [23] reported that hair Mn levels were higher in Canadian children exposed to Mn-contaminated water compared to children living in homes with a private well with lower water Mn levels, while Skröder et al. [17] reported no correlation between Mn levels in water and hair Mn levels in Bangladeshi children.

Studies have also reported mixed results on the associations between metal levels in hair and other exposure biomarkers. Hair As levels have been shown to reflect the internal body burden of As, based on strong correlations between hair As levels with As levels in erythrocytes (*ρ* = 0.73, *p* < 0.001) and urine (*ρ* = 0.66, *p* < 0.001) [17], while hair Cd and Pb levels are not generally recognized as reliable predictors of exposure and internal dose [7,11,17]. A number of studies have reported no association between hair Mn and Mn levels in blood [2,12,38] or erythrocytes [17], while others reported low but significant correlations between hair and blood Mn levels [22,39], and associations between hair Mn and Mn levels in fingernails (*ρ* = 0.247, *p* < 0.001) [12]. In their recent study, Skröder et al. [17] concluded that levels of Mn in hair do not reflect the actual internal Mn dose in Bangladeshi children, but the authors acknowledged that their findings strongly pointed to significant external contamination of hair from Mn-contaminated water, which would preclude the ability to even test whether hair Mn reflects the internal Mn burden. Finally, there is substantial evidence showing the hair Mn levels are associated with a number of neurodevelopmental health effects, including reduced IQ, learning, memory, and perceptual reasoning, and greater hyperactive and oppositional behaviors [2,13,23,38–41], leading Coetzee et al. [3] to conclude in their recent review that hair was the most consistent and valid biomarker of manganese exposure and associated neurodevelopmental health effects in children.

The present study had several limitations. First, the parent C8 Health Project Neurobehavioral Development Follow-up study, which was investigating the neurodevelopmental health effects of perfluoroocatnoate (PFOA) exposure in southeastern Ohio and northwestern West Virginia [27], did not assess metal exposures in the subjects' environment (e.g., air, dust, water) or in other biomarker tissues (e.g., blood), thereby limiting our ability to interpret the hair metal levels reported here as a biomarker of environmental metal exposures or internalized body burden. Second, sequential hair segments were available only for female subjects, and not males, precluding assessment of a sex–hair segment interaction in our statistical models.

**Acknowledgments:** We thank Patricia Leung for analytical assistance, and Ashley M. Pajak for assistance with the statistical analysis plan. This study was supported in part by grants R21 ES019643 and R01 ES019222 from the National Institute of Environmental Health Sciences.

**Author Contributions:** T.J. performed hair metal analyses and assisted in data analysis and interpretation, and drafting the manuscript. C.R.S. conceived of and supervised the parent study on chemical exposure and neurodevelopment, including hair sample collection, and participated in data analysis and interpretation for the present study. D.R.S. participated in statistical analyses, data interpretation, and drafting the manuscript.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
