DNA Day TOP ARTICLES SUPPLEMENT

CONTENTS RESEARCH ARTICLE: Misclassified exposure in epigenetic mediation analyses. Does DNA methylation mediate effects of smoking on birthweight? Epigenomics Vol. 9 Issue 3 REVIEW: Epigenetics and allergy: from basic mechanisms to clinical applications Epigenomics Vol. 9 Issue 4 RESEARCH ARTICLE: DNA methylation at diagnosis is associated with response to disease-modifying drugs in early rheumatoid arthritis Epigenomics Vol. 9 Issue 4 SPECIAL FOCUS y Effects of the in utero environment on the epigenome

Research Article

For reprint orders, please contact: [email protected]

9 Research Article 2017/02/28 Misclassified exposure in epigenetic mediation analyses. Does DNA methylation mediate effects of smoking on birthweight?

Epigenomics Aims: Assessing whether epigenetic alterations mediate associations between Linda Valeri*,1,2, Sarah L environmental exposures and health outcomes is increasingly popular. We investigate Reese3, Shanshan Zhao3, the impact of exposure misclassification in such investigations. Materials & methods: Christian M Page4, Wenche 4 5 We quantify bias and false-positive rates due to exposure misclassification in mediation Nystad , Brent A Coull & Stephanie J London3 analysis and assess the performance of the simulation extrapolation method (SIMEX). 1Department of Psychiatry, Harvard We evaluate whether DNA-methylation mediates smoking–birth weight relationship Medical School, Boston, MA 02115, USA in the Norwegian Mother and Child Study birth cohort. Results: Ignoring exposure 2Psychiatric Biostatistics Laboratory, misclassification increases type I error in mediation analysis. The direct effect is McLean Hospital, Belmont, MA 02478, underestimated and, when the mediator is a biomarker of the exposure, as is true for USA 3 smoking, the indirect effect is overestimated. Conclusion: Misclassification correction National Institute of Environmental Health Sciences, National Institutes of plus cautious interpretation are recommended for mediation analyses in the presence Health, Department of Health & Human of exposure misclassification. Services, Research Triangle Park, NC, USA 4National Institute of Public Health, Oslo, First draft submitted: 17 October 2016; Accepted for publication: 22 December 2016; Norway 5 Published online: 21 February 2017 Department of Biostatistics, Harvard T.H. Chan School of Public Health *Author for correspondence: Keywords: DNA methylation • mediation analysis • misclassification [email protected]

Evidence is accumulating that environmen- biased reporting occurs whereby some pro- tal exposures modify the epigenome. In portions of smokers falsely claim, on surveys, humans, the best-studied epigenetic modi- to be nonsmokers [12] . In addition, because of fication is methylation and the best-studied the well-publicized negative health impacts exposure is smoking. Smoking in adults has of maternal smoking during pregnancy been reproducibly associated with alterations on the developing fetus, pregnant women 3 in methylation at specific loci [1] . Similar under-report smoking more than nonpreg- effects have been seen in newborns whose nant smokers of reproductive age [13] . None- mothers smoked during pregnancy [2]. These theless, studies that address whether meth- smoking-methylation signals have been used ylation signatures from smoking mediate its 2016 to develop novel biomarkers of exposure [3– health outcomes have ignored the potential 6]. In addition to its value as an exposure role of measurement error in assessment of biomarker, there is great interest in the pos- smoking [7–10,14]. Given this measurement sibility that differential methylation at rel- error, evaluation of mediation may be com- evant loci mediates well-established associa- plicated by the fact that the proposed media- tions between smoking and disease, both for tors, DNA sites differentially methylated by adult [7,8] and in utero exposures [9,10]. smoking, are excellent biomarkers that may It is widely acknowledged that measure- better capture the exposure than ­self-reported ment of human environmental exposures, smoking [3–6]. including smoking, is prone to error [11] . In the field of mediation analysis, bias Random error exists for all exposures. How- introduced by measurement error in ever, for smoking in particular, differential the mediator variable has been investi- part of

10.2217/epi-2016-0145 © US Government Epigenomics (2016) 9(3), 253–265 ISSN 1750-1911 253 Research Article Valeri, Reese, Zhao et al.

gated [15–17] . However, misclassification of the expo- weight that is mediated by the DNA-methylation level, sure variable has not been well evaluated. Reduced relative to the direct effect of smoking on birthweight birth weight is a well-established sequelae of maternal through pathways independent of DNA methyla- smoking during pregnancy [18]. Given strong evidence tion (Figure 1B). Under the counterfactual framework of differential methylation in newborns in relation to for causal inference direct and indirect causal effects smoking by the mother [2], it has been of interest to have been rigorously defined [21,22] (section A1 of the consider whether these signals mediate the effects of Supplementary material). maternal smoking on birth weight. It has recently been To validly estimate direct and indirect effects, the reported that differential DNA methylation of a single following four confounding assumptions need to be CpG site in placenta mediates up to 36% of the effect satisfied. Conditioning on covariates C, there is no of smoking on lower birth weight [9]. In another study, unmeasured confounding of the exposure–outcome differential methylation in newborn’s blood at a single relationship, the mediator–outcome relationship, the CpG site in a different gene was reported to mediate exposure–mediator relationship and there are no medi- 19–46% of the relationship between smoking and ator–outcome confounders affected by the exposure. birthweight [10] . Because self-reported smoking status See [22,23] for further discussion of these assumptions. during pregnancy is prone to misclassification, we were Furthermore, models for the outcome and mediator interested in evaluating the sensitivity of mediation need to be correctly specified. For continuous outcome analysis with methylation data to exposure misclassi- and mediator (as in the current setting of outcome birth fication bias. For this purpose, we considered a pub- weight and mediator methylation), under the assump- lished scenario in perinatal epidemiology, for which we tion of no exposure–mediator interaction in the out- have relevant data [10], as an example. come model, typically made by published applications Our study makes several contributions. First, we of mediation analysis in environmental epigenetics, if study the impact of exposure misclassification on the we specify three linear regression models: estimation of direct and indirect causal effects and ++oo+o testing of the indirect effect in methylation studies EY( Aa= , C = cc) = ii01++i ' analytically. Second, we assess the impact of misclas- sification on estimation and testing via a simulation EY( Aa= , M ==ma,)Cc= ii01++i2 m + i'c study. Third, we evaluate the ability of the SIMEX approach to adjust for exposure misclassification in EM( Aa==,)Cc= b0 +b1 a + b'c this setting. Finally we use data from the Norwegian Mother and Child Cohort Study (MoBA) [19,20] to con- then the estimators of total effect (TE), direct effect duct a mediation analysis accounting for misclassifi- (NDE) and indirect effect (NIE) take the form [24,25]: cation of self-reported smoking status using SIMEX. +o Our study provides evidence that ignoring misclassifi- TE = i1 cation can bias results of mediation analyses and shows

the value of incorporating misclassification correction NDE = i1 in mediation analysis in the context of environmental +o epigenetic studies. NIE ==bi12 ii11-

Methods Under the assumption of no unmeasured confound- Mediation analysis in the absence of exposure ing and that models 1–3 are correctly specified, these misclassification estimators (which are equivalent to the ones proposed With reference to our example of mediation of the by [24] in the psychology literature) can be interpreted effect of maternal smoking during pregnancy on as causal direct and indirect causal effects [22]. Esti- newborn birth weight by smoking-related differential mators for direct and indirect effects in the presence methylation, let A denote the exposure, maternal smok- of exposure-mediator interaction are given in Section ing and M denote the mediator, DNA methylation. Let A1 of the Supplementary Material. A discussion on the Y denote the outcome, birth weight and C denote a comparison between traditional and causal inference vector of covariates representing potential confound- approaches to mediation analysis is given in [26]. ers. The directed acyclic graph in Figure 1 describes the The most popular test for indirect effects is based on setting of mediation analysis. Mediation analysis can the product method, also known as the Sobel test [27].

be employed to quantify how much of the total effect This is a Wald test for the null hypothesis H0: β1 θ2 = 0 of maternal smoking on birth weight (Figure 1A) is based on the delta method standard error 2 222 2 2 explained by the indirect effect of smoking on birth vvNIE = ib21bv1 + i2 where vi2 and vb1 are the

254 Epigenomics (2016) 9(3) future science group Exposure misclassification in mediation analysis Research Article

variances of the maximum likelihood estimates of θ2 and β , respectively. 1 + θ 1 Mediation analysis in the presence of exposure A Y misclassification Let A* denote a binary exposure, self-reported smoking θ1 status during pregnancy in our example, potentially misclassified and A the true smoking status. We express, without loss of generality, measurement error in an addi- tive form, A* = A + U. For a binary exposure, A* and A take values in {0, 1}, while the misclassification error U A M Y β θ takes values in {-1,0, 1}. Assume that U is independent 1 2 of the outcome, the mediator and the covariates, given true maternal smoking status A (i.e., misclassification error is nondifferential with respect to outcome, media- C tor and covariates). In this case the misclassification probabilities are characterized by sensitivity SN = P(A* Figure 1. (A) Directed acyclic graph for average causal = 1|A = 1) and specificity SP = P(A* = 0|A = 0) of the effect of sustained smoking during pregnancy (A) on + potentially misclassified exposure A*, yieldingbirth weight (Y) (TE = i1 from Equation 1); (B) Directed P(A*|A,M,Y,C) = P(A*|A). Under these assumptions acyclic graph for direct of sustained smoking during misclassification is dependent on the true latent expo- pregnancy (A) on birth weight (Y) and indirect effect of sustained smoking during pregnancy (A) on birth sure because Cov(U,A) ≠ 0 [28]. This misclassification weight (Y) through DNA-methylation (M) (NDE = θ + 1 mechanism is realistic for self-reported maternal smok- from Equation 3, NIE = ii11- = bi12) (C) denotes a ing during pregnancy and the results presented here can vector of confounders. be easily extended to the case in which the error U is dependent on covariates as well. We expect perfect spec- approach to correct for misclassification of expo- ificity (SP = 1) because it is reasonable to assume that if sure [29,30]. In the second stage the SIMEX coefficient the mother is a nonsmoker (A = 0), she will report cor- estimates are plugged into the formulas of NDE and rectly to be a nonsmoker (A* = 0). However, we expect NIE to obtain misclassification corrected estimates that some smoking mothers (A = 1) might incorrectly of the causal contrasts of interest with standard errors report being nonsmokers (A* = 0), leading to imperfect obtained via the bootstrap [29]. SIMEX has been shown % % sensitivity (SN ≠ 1). Let, NDE * and NIE * denote the to perform well in mediation analysis when the media- naive direct and indirect effect estimators, respectively, tor is measured with error both in linear and nonlinear when we fit the regression models inEquations 1–3 models [15] . For an assumed amount of measurement replacing the true exposure (A) with the self-reported error (or misclassification error in this case), SIMEX exposure (A*). Let i denote the naive outcome regres- simulates new datasets by additional error and calcu- V * sion parameter estimators when the true exposure (A) is lates estimates for each of these new datasets, yielding replaced with the self-reported exposure (A*) in data on the expected coefficient estimates as a function Equation 2. Let b denote the naive mediator regres- of the amount of measurement error. The procedure W * sion parameter estimators when the true exposure (A) is then fits a parametric model to this function, and then replaced with the self-reported exposure (A*) in extrapolates this function back to the no-measurement Equation 3. In the results section we will assess analyti- error case. We used a quadratic model for the measure- cally the bias of naive estimators of the natural direct ment error – coefficient estimate relationship because and indirect effects and the type I error of the Sobel test of its flexibility. Additional information on SIMEX when the assumptions given above hold and when and on its implementation, using the R package Simex ­exposure misclassification is ignored in the analysis. can be found in [31] . Our code can be found in the Supplementary Material Section A8. When the amount Correction approach of misclassification is not known from external valida- To correct for misclassification and obtain valid esti- tion data, as in our situation, we obtain SIMEX esti- mates of natural direct and indirect effects, we use a mates under a range of specificity and sensitivity values. two-stage approach introduced in [15,16]. In the first stage, assuming plausible values for SP and SN, media- Simulation study tor and outcome regression coefficients are estimated We conducted a simulation study to assess the perfor- using the SIMEX (simulation and extrapolation) mances of the naive mediation analysis that ignores

future science group www.futuremedicine.com 255 Research Article Valeri, Reese, Zhao et al.

exposure misclassification and the SIMEX correction Study (MoBa) pregnancy cohort. Naive analyses that approach. In particular, we investigated the bias ignore misclassification were conducted using the %% same approach employed in [10] and described in the (*NDEN--DE,*NIENIE), previous sections. We then applied the SIMEX correc- %%tion approach to assess the sensitivity to the results to relative bias ((bias NDEN*)/,DE bias(*NIEN)/ IE) exposure misclassification. MoBa is a large population- %% based pregnancy study targeting all women in Norway and variance ((varNDE *),(varNIE *)) in the who gave birth between 1999 and 2008 [19,20]. Illumina estimates of direct and indirect effects. Further, we HumanMethylation450K data from cord blood were

studied Type I error rates of the Sobel test for H0: NIE measured on a subcohort of MoBa participants born

= β1θ2 = 0 Note that the NIE will be zero if either (a) between 2002 and 2004 (n = 1068), along with ques- there is no effect of exposure on the mediator and no tionnaire data on smoking and potential confounders

effect of the mediator on the outcome (θ2 = 0 from at about weeks 17 and 30 of pregnancy and cotinine Equation 2 Equation 3 and β1 = 0 from ); or if (b) there measured in maternal plasma collected at about gesta- is no effect of exposure on the mediator, but there is an tional week 18 of pregnancy [32]. Data on birthweight

effect of the mediator on the outcome (i.e., β1 = 0, θ2 ≠ and gestational age were obtained from the Medical 0); or if (c) there is an effect of exposure on the media- Birth Registry of Norway. Gestational age at birth was tor, but there is no effect of the mediator on the out- based mainly on routine fetal ultrasonographic exami-

come (i.e., β1 ≠ 0, θ2 = 0). In considering the indirect nation at week 17–19, administered to more than 98% effect of smoking on birth weight through DNA meth- of Norwegian women. When ultrasound data were ylation, we are particularly concerned about falsely missing, gestational age was calculated using last rejecting the null hypothesis under this last scenario ­menstrual period [33].

(β1 ≠ 0, θ2 = 0). The data generating process for these The Regional Committee for Medical Research simulations, including sample size (n = 500) and distri- Ethics, Norway and the NIEHS Institutional Review butions of exposure, mediator, covariates and outcome, Board approved the study. was designed to mimic the recent study reporting that We first performed a naive mediation analysis in smoking-related differential methylation at specific MoBa, ignoring exposure misclassification, to quan- CpG sites mediates part of the effect of maternal smok- tify the amount of mediation of the smoking–birth ing exposure on birth weight [10] . Further, we specified weight association due to methylation at each of three mediator and outcome regression parameters accord- GFI1 CpG sites that were replicated in the published ing to the reported findings of this study. We assessed mediation analysis [10] . Gestational age was included as bias under the null hypothesis of no indirect effect and a linear variable as in the published mediation analy- under the alternative hypothesis, assuming DNA sis [10] as well as in recent epigenome wide methylation methylation is a strong biomarker of the exposure. analyses [34,35]. Women who reported smoking early Therefore, under the alternative hypothesis we assumed in pregnancy but who quit early on were not consid- Equation 3 Equation 2 β1 ≠ 0 in and θ2 ≠ 0 in . Under ered sustained smokers. We used this exposure vari-

the null hypothesis of no indirect effect we assumed β1 able because the smoking methylation associations Equation 3 Equation 2 ≠ 0 in and θ2 = 0 in . For simula- observed in previous studies are not seen for smoking tions of type I error rates under the null hypothesis of that ends early in pregnancy but rather require more no indirect effect, we considered the case in which the sustained exposure across the pregnancy [2,36]. indirect effect is null because of no effect of methyla- To better evaluate the impact of misclassification we

tion on birth weight (θ2 = 0) and no effect of smoking considered two definitions of sustained smoking. One

status on methylation (β1 = 0), but where smoking is based on self-report alone. We then enhanced self-

affects methylation (θ2 = 0, β1 ≠ 0). Full description of report by using cotinine measurements done at about the simulation scenarios is given in Section A3 of 18 weeks so mothers who reported being nonsmokers Supplementary Material. but had cotinine values compatible with current smok- ing status were reclassified as smokers. We fitted the Analysis of Norwegian Mother & Child Study data naive outcome and mediator regressions of the form of Finally, to assess the impact of misclassification of Equations 1–3 adjusting for sex, maternal age, maternal self-reported smoking status on estimates of media- education, gestational age, parity and maternal pre- tion of the effect of maternal smoking on birthweight pregnancy BMI as potential confounders for compa- by specific methylation signals, we analyzed the same rability with the analysis presented in [10] . There were exposure–mediator–outcome scenario presented in [10] 1022 individuals with data on maternal smoking, the using data from the Norwegian Mother and Child CpGs and all covariates available for these analyses.

256 Epigenomics (2016) 9(3) future science group Exposure misclassification in mediation analysis Research Article

We then ran mediation analysis taking misclassifica- overestimation of the indirect effect. In other words, tion into account using the SIMEX method [30] for when the biomarker mediator captures the variabil- binary smoking status (sustained smoking across the ity of true latent smoking exposure better than the pregnancy, yes or no). Applying SIMEX to outcome self-reported measure of smoking, some of the direct and mediator regressions, we obtained corrected esti- effect is incorrectly attributed to the mediator (the mates of the regression parameters and then used those indirect effect). Results on asymptotic bias in the estimates in the equations for the direct and indirect presence of exposure-mediator interaction are less effects. Because SIMEX requires specifying sensitiv- intuitive and for a full description of the results and ity and specificity, and a wide range of sensitivities has proofs, the reader can refer to ­sections A4–A7 of the been reported for self-reported smoking status [13,37], Supplementary Materials. we assessed the robustness of the naive results to a wide Another important issue when studying a potential range of plausible sensitivity (SN) parameter values mediator that is a strong biomarker for the exposure is (between 0.6 and 0.9). We assumed perfect specificity that under the null hypothesis of no indirect effect, % (SP = 1) because we do not expect pregnant women to NIE * will be biased. Under this setting, one of the falsely report smoking if they are nonsmokers. We used necessary conditions for the validity of type I error of the bootstrap to estimate ­standard errors of the direct the Sobel test for an indirect effect is not met. There- and indirect effects. fore, under the null hypothesis of no indirect effect, if In addition to sustained smoking, we evaluated the exposure is misclassified and the mediator is a bio- exposure to any smoking during the pregnancy. marker for the exposure, the Type I error rate will not Women who reported smoking on either pregnancy be preserved. In the scenario we consider [10], the expo- questionnaire were coded as yes for any smoking with- sure is related to the mediator. Therefore, in reasonable out regard to whether they reported quitting early in scenarios of mediation analysis in environmental epi- pregnancy. This variable was used to classify mater- genetic studies, the naive mediation analysis is likely nal smoking in the discovery cohort in the published biased and there is risk of reporting false-positive find- mediation analysis [10] . ings of mediated effects through DNA methylation whenever the exposure is imperfectly measured and Results DNA methylation is a biomarker of the exposure. Asymptotic bias & type I error In the absence of exposure-mediator interaction, Simulation study % NDE = i* and by directly applying results on the We now illustrate the bias of estimates of natural direct * X 1 impact of exposure misclassification in linear regres- and indirect effects and type I error of tests for mediation sion [28], the naive estimator of the natural direct effect in the presence of exposure misclassification. Under the * i i1 is shown to be biased toward the null ()Y 1 < . The simulation settings described in the previous section, in % bias of NDE * depends on the magnitude of the mis- the presence of misclassification of a binary exposure (Table 1 & Supplementary Figure 1) classification error and the true parameter θ1. We show due to misreport , in the Supplementary Material sections A5–A7 that the direct effect is underestimated and the indirect under the special case of no direct effect, the naive effect is overestimated under all simulation scenarios

­estimator of the direct effect is unbiased. (i.e., 1 the alternative hypothesis [β1 ≠ 0, θ2 ≠ 0] and

Exposure misclassification will bias the estima- 2 the null hypothesis of no indirect effect [β1 ≠ 0, θ2 tor of the exposure coefficient in the mediator model = 0]). The exposure–mediator association (β1) and

(β1) downward but will bias the estimator of the the exposure–outcome association (θ1) are underes- coefficient for the mediator in the outcome model timated and the mediator–outcome association (θ2) (Supplementary Figure 2) (θ2) upward. The indirect effect, in the absence of is overestimated . Applica- exposure–mediator interaction, is the product of the tion of the SIMEX approach significantly reduces the coefficient for the exposure in the mediator model bias of direct and indirect effect estimators, resulting and the coefficient of the mediator in the outcome in approximately unbiased estimates of direct and model (Equation 6 & Figure 1). Therefore, in theory, ­indirect effects (Supplementary Table 1). the bias of the naive indirect effect estimator can be We also note that the type I error of the Sobel test in either direction. However, when the mediator is a of the indirect effect is conservative in the absence of strong biomarker for the exposure (i.e., β1 ≠ 0), as is misclassification [38]. However, in the presence of expo- the case for smoking methylation signals, our ana- sure misclassification, the Type I error of the Sobel test lytic results and simulation studies below show that of the indirect effect is elevated above the nominal the bias of the total effect estimator is larger than the 5% when the mediator is a strong biomarker of the bias of the natural direct effect estimator, leading to ­exposure (Table 2).

future science group www.futuremedicine.com 257 Research Article Valeri, Reese, Zhao et al.

Mediation analysis in the MoBa study ylation) that is marginally statistically significant (NIE Küpers et al. [10] performed an epigenome-wide asso- = -30.3; 95% CI: -60.5–0.0), based on the bootstrap ciation study of the association between dichotomous CIs, which are recommended when sample size is small maternal smoking (129 exposed, 129 unexposed) and to moderate. The Sobel test yields stronger evidence of DNA methylation data in the Groningen Expert Cen- mediation (p = 0.021). The naive analyses estimate that ter for Kids with Obesity (GECKO) cohort and then 32% of the total effect of smoking on birthweight is analyzed the 35 top CpGs (those epigenome-wide mediated by this CpG. For the other two CpGs, which significant at FDR < 0.05) to assess whether methyl- are less strongly associated with smoking and birth- ation at these CpGs mediates the effect of maternal weight, the naive analyses provide weaker evidence of smoking on birth weight. Among eight CpG sites in mediation: the indirect effects are nonsignificant and the GFI1 gene that showed the most robust media- the proportions mediated are much lower (Table 4). tion in the GECKO cohort, three gave significant Correcting for potential misreporting of smoking Sobel p-values both in meta-analysis of two additional during pregnancy using the SIMEX approach weakens birth cohorts (Avon Longitudinal Study of Parents the evidence for mediation at cg09935388 (Table 4). and Children (ALSPAC) and Generation R) [39,40] and Under the assumption of fairly severe, but realistic [13] in meta-analysis across all three cohorts. The authors misclassification of smoking based on self-report in reported a significant indirect effect whereby differ- pregnant women (SN = 0.70), comparison of results ential methylation of each of these three GFI1 CpGs after application of the SIMEX approach suggests mediated 19–46% of the decrease in birth weight in that the direct effect of smoking on birth weight, not the GECKO discovery cohort and a smaller 12–19% through methylation of CpG cg09935388, is underesti-

in the three cohort meta-analysis. mated (NDESN = 0.7 = -73.8; 95% CI: -171.3–35.9), and

Analyses for the MoBa cohort were conducted for the indirect effect is overestimated (NIESN=0.7 = -28.4; the same three CpG loci in GFI1 using R software ver- 95% CI: -59.0–6.0; proportion mediated = 27%) by sion 3.1.3. Commented code of the analyses can be the naive analyses. Under all sensitivity values consid- found in the Supplementary Material Section A8. ered, SIMEX corrected analyses for CpGs cg12876356 Table 3 contains naive regression analyses of the and cg14179389 indicate severe under-estimation of MoBa cohort, using self-reported sustained smok- the direct effect of smoking in the naive analysis and ing as the exposure, birth weight as the outcome and weaker evidence of mediation by the CpGs. the three GFI1 CpGs (cg09935388, cg12876356, We repeated the mediation analyses (naive and cg14179389) that were found to significantly medi- SIMEX corrected) after enhancing the exposure vari- ate the smoking-birth weight association in [10] . Our able by incorporating cotinine, a short-term biomarker study population of 1022 individuals with nonmissing of smoking, measured in mid-pregnancy. We reclassi- data for all covariates included 117 women classified as fied as smokers 18 mothers who reported being non- ­sustained smokers by self-report. smokers, but had a cotinine level consistent with smok- In naive linear regression analyses, without any ing resulting in 135 sustained smokers. This enhanced potential CpG mediator in the model, (Table 3) mater- exposure variable should have less measurement error nal smoking is significantly related to birthweight; birth than smoking assessed by self-report alone. The reduc- weight was 93 g lower in newborns of smoking mothers. tion in birth weight for infants of smoking mothers However, controlling for the CpG mediators attenuates is greater for this enhanced variable (-116 g; 95% CI: the association between smoking and birth weight and -195 to -36) than for the self-report alone variable it ceases to be statistically significant (Table 3). Among (-93 g; 95% CI: -180 to -8) (Supplementary Table 2 the three CpGs, all previously reported at epigenome includes results for the other two CpGs). In naive anal- wide Bonferroni significance in relation to maternal yses (Table 5) for CpG cg0993538 when using this smoking in MoBa [32], adjustment for cg09935388 enhanced smoking variable, the direct effects are larger leads to the greatest reduction in the effect estimate (-89 vs -64 g) compared with the mediation analysis for smoking on birthweight (Table 3). Among the three using self-reported smoking in Table 4 (results for the CpGs, cg09935388 also had the strongest association other two CpGs in Supplementary Table 3). The indi- with smoking (β = 0.12 vs β = 0.07 for the other two). rect effects are smaller, have wider CIs and the pro- In our MoBa data, naive mediation analysis impli- portion mediated is correspondingly much smaller cates methylation at CpG cg09935388 as a potential (0.24 compared with 0.32) than in the analysis of self- mediator of the smoking–birth weight relationship reported smoking. Moreover, the Sobel test is margin- (Table 4). We estimate a nonsignificant natural direct ally statistically significant (p-value = 0.051). Applica- effect of smoking on birth weight (NDE = -64.1; tion of the SIMEX correction approach yields a more 95% CI: -148.1–30.6), an indirect effect (via meth- substantial reduction in the proportion mediated by

258 Epigenomics (2016) 9(3) future science group Exposure misclassification in mediation analysis Research Article

Table 1. Bias, Relative bias and variance of naive estimators of total effect, natural direct effect, natural indirect effect and proportion mediated for simulation scenario I assuming sensitivity = (0.70, 0.80, 0.90, 0.95), and specificity = 1. True SN = 0.70 SN = 0.80 SN = 0.90 SN = 0.95 Bias Rel. Var Bias Rel. Var Bias Rel. Var Bias Rel. Var bias bias bias bias

a b H1(β1 ≠ 0, θ2 ≠ 0) TE -194 77.6 -0.40 990 58.2 -0.30 1013 32.98 -0.17 1305 19.4 -0.10 1397 NDE -149 74.5 -0.50 1090 59.6 -0.40 1176 37.25 -0.25 1942 20.86 -0.14 1952 NIE -45 -1.80 0.04 149 -0.9 0.02 193 -2.7 0.06 270 -2.25 0.05 330 PM 0.24 0.16 0.67 0.04 0.12 0.50 0.02 0.072 0.30 0.02 0.04 0.17 0.01

a b H0 (β1 ≠ 0, θ2 = 0) TE -150 60.0 -0.40 990 45.0 -0.30 998 25.5 -0.17 1019 15.0 -0.10 1045 NDE -150 75.0 -0.50 1090 59.6 -0.40 1177 37.5 -0.25 1305 20.86 -0.14 1398 NIE 0 -16 -16/0 127 -14 -14/0 174 -11 -11/ 0 225 -6 -6/0 319 PM 0 0.21 0.21/0 0.02 0.15 0.15/ 0 0.03 0.10 0.10 / 0 0.03 0.05 0.05/0 0.37 †Exposure-mediator association from Equation 3. ‡Mediator-outcome association from Equation 2. NDE: Natural direct effect; NIE: Natural indirect effect; PM: Proportion mediated; SN: Sensitivity; SP: Specificity; TE: total effect; var: Variance.

CpG cg0993538 (from about 0.24 to 0.15) than in 0.59) with smaller reductions after adjustment for each the analysis with self-reported smoking (reduced from of the other two CpGs (Supplementary Table 4). In about 0.32 to 0.26). SIMEX also increased the size and naive mediation analyses there was a significant indi- the precision of the estimated natural direct effect of rect effect -23.3 (95% CI: -41.4 to -2.9, Sobel test p = smoking more than in the analysis of the self-reported 0.011) of smoking on birthweight through this CpG variable (at sensitivity of 0.6, NDE was -89.4 [95% but no significant direct effect of smoking on birth- CI: -179.5–2.6] in the naive and -106.9 [95% CI: weight and the proportion mediated was much larger -216.4–8.1] in the SIMEX for the enhanced exposure than in the sustained smoking analyses at 58%, even variable compared with -64.1 [95% CI: -148.1–30.6] higher than that the 46% observed in the GECKO naive and -76.3 [95% CI: -183.5–37.3] SIMEX for the study of the same exposure variable [10] (Table 6). After self-reported variable). SIMEX measurement error correction the proportion To evaluate in our MoBa data, the worst-case sce- mediated was reduced, although the reduction was nario for misclassification of exposure to maternal proportionally smaller than in the analyses of the two smoking, we also repeated the analysis using any sustained smoking variables (Table 6). Results for the smoking during pregnancy as the exposure variable. other two CpGs are reported in Supplementary Table 5. Women coded as yes to any smoking (N = 288) include the more than 50% of women who quit early in preg- Discussion nancy. In linear regression, the coefficient for any Mediation analysis is the primary tool for investigating smoking during pregnancy was -40.0 g birthweight the role of epigenenetic mechanisms in health effects (SE = 30.9; p = 0.20) and was greatly reduced after of environmental exposures. Its use is increasing along adjustment for cg0993538 to -17.5 g (SE = 32.5; p = with evidence for epigenetic impacts of smoking and

Table 2. Type I error of Sobel test for indirect effect for simulation scenario I assuming sensitivity = (0.70, 0.80, 0.90, 0.95), and specificity = 1. Truea SN = 0.70 SN = 0.8 SN = 0.9 SN = 0.95

b c H0 (β1 = 0, θ2 = 0) 0% 0% 0% 0% 0% b c H0 (β1 = 0, θ2 = 0) 4.9% 27% 18% 10% 6.5% aType I error of Sobel test when the true exposure is used in the regression analyses. bExposure–mediator association from Equation 3. cMediator–outcome association from Equation 2. SN: Sensitivity.

future science group www.futuremedicine.com 259 Research Article Valeri, Reese, Zhao et al.

Table 3. Linear regression of self-reported sustained maternal smoking during pregnancy in relation to infant birth weight before and after adjustment for effects of maternal smoking on methylation at three CpGs in the GFI1 gene in the Norwegian Mother and Child Cohort Study. Regression model specification Coeff† SE p-value No mediator (CpG) adjustment -93.22 43.55 0.03 Adjusting for cg09935388 -63.99 46.21 0.17 Adjusting for cg12876356 -77.01 45.11 0.09 Adjusting for cg14179389 -82.48 45.42 0.07 †Regression coefficient interpretable as difference in birth weight, in grams between offspring of smoking mothers relative to nonsmokers. Each separate linear regression model (only the specified CpG included) includes the following covariates: gestational age, child gender, maternal age, maternal education, parity, selection group and maternal prepregnancy BMI. SE: Standard error.

other environmental exposures and the desire to iden- suggested as an option to evaluate the reduced effect tify biologic and public health implications of these of measurement error in mediation analyses of meth- epigenetic signals. However, mediation analysis is sub- ylation signals [44,45] in epigenetic studies. In MR, if ject to various biases. It relies on stringent and untest- there are genetic variants robustly associated with the able assumptions of no-unmeasured confounding and exposure of interest, these can be used to help infer correct model specification. In observational studies, causality by serving as correctly measured instrumen- biases due to selection, missing data and measurement tal variables, which are not associated with various error further challenge the validity of mediation analy- confounders and are not directly influenced by the sis [41] . Sensitivity analyses for violation of the assump- outcome of interest. Richmond et al. recently sug- tions of no-unmeasured confounding and no selection gested that an earlier report that 30% of the associa- bias have been proposed [42,43]. Recently, Mendelian tion between adult smoking and lung cancer can be randomization (MR) estimation strategies have been explained by methylation at a single smoking related

Table 4. Estimates of natural direct and natural indirect effects of sustained maternal smoking, assessed by self-report, on birth weight and proportion mediated by three methylation cites (CpGs) in GFI1 in naive analyses and after SIMEX correction for measurement error in the Norwegian Mother and Child Study study. CpG SN NDE (95% CI) NIE (95% CI) PM cg09935388 Naive -64.1 (-148.1 to 30.6) -30.3 (-60.5 to 0.0) 0.32 0.6 -76.3 (-183.5 to 37.3) -27.8 (-60.4 to 9.2) 0.26 0.7 -73.8 (-171.3 to 35.9) -28.4 (-59.0 to 6.0) 0.27 0.8 -70.4 (-161.6 to 38.8) -29.6 (-59.8 to 3.1) 0.29 0.9 -66.3 (-153.5 to 34.4) -30.0 (-60.7 to 1.1) 0.31 cg12876356 Naive -78.0 (-158.9 to 11.7) -16.3 (-38.0 to 6.8) 0.17 0.6 -86.5 (-183.8 to 17.3) -13.9 (-39.6 to 12.5) 0.14 0.7 -85.5 (-178.9 to 22.1) -14.5 (-37.9 to 10.9) 0.15 0.8 -82.1 (-170.6 to 15.5) -15.3 (-38.7 to 9.6) 0.16 0.9 -79.5 (-164.7 to 15.1) -16.0 (-38.4 to 7.9) 0.17 cg14179389 Naive -81.4 (-168.0 to 12.4) -11.7 (-35.0 to 11.4) 0.13 0.6 -93.8 (-194.2 to 14.6) -8.0 (-36.9 to 20.6) 0.08 0.7 -91.3 (-184.2 to 14.0) -9.4 (-36.3 to 18.0) 0.10 0.8 -86.0 (-181.8 to 15.9) -9.9 (-36.2 to 15.8) 0.10 0.9 -85.0 (-173.7 to 15.4) -10.9 (-35.3 to 13.1) 0.11 The SIMEX corrected values are presented for four different values for sensitivity of the self-reported maternal smoking exposure variable: 0.6, 0.70, 0.80, 0.90 where specificity = 1. Median and 95% percentile CIs for the bootstrap estimates are in units of grams of birth weight. NDE: Natural direct effect; NIE: Natural indirect effect; SN: Sensitivity; PM: Proportion mediated.

260 Epigenomics (2016) 9(3) future science group Exposure misclassification in mediation analysis Research Article

CpG might be spurious and reflect measurement error outcome thus supplementing the measured exposure in self-reported smoking [44]. They suggested that MR, (self-reported smoking) itself. Thus mediation by the using genetic variants related to smoking, might be smoking methylation biomarker is overestimated. a way to evaluate that possibility. Assumptions and We based our analyses on a recently published application of MR analysis for the estimation of direct scenario whereby three CpGs in GFI1, differentially and indirect effects are further outlined in [45]. methylated in newborns in relation to self-reported Here we examine the potential impact of exposure maternal smoking, were reported to mediate 19–46% misclassification using the example of maternal smok- of the effect this in utero exposure on offspring birth ing during pregnancy as the exposure and smoking- weight [10] . These three CpGs are robust sites of dif- related methylation signals in the newborn as the ferential methylation from maternal smoking reported putative mediators of the well-established relationship in various studies including the MoBa dataset used for between smoking and reduced birth weight. the current analysis [32]. The Küpers et al. [10] study Our simulation studies and analyses of the MoBa was well conducted and included replication of the data show that in environmental epigenetic studies mediation finding in two additional high-quality on the mediating role of DNA methylation, when the birth cohorts [10] . Nonetheless, similar to other stud- methylation signal is a good biomarker of an exposure ies of smoking or other exposure-related methylation that is measured with error, there is a substantial risk signals as mediators of exposure-related health effects, of false positives and overestimation of the proportion the potential contribution of misclassification was not mediated. This is the case for maternal smoking during estimated. We would expect misclassification to act pregnancy and newborn methylation where the smok- the same way in the discovery and replication cohorts ing-related methylation signals are excellent biomarkers and thus replication of the apparent mediation does of exposure compared with self-report [3–6]. It is known not reduce possibility that exposure misclassification that some smokers self-report as nonsmokers and this contributes to the finding of mediation. Of inter- misreporting is more prominent during pregnancy est, the proportion mediated by methylation at GFI1 because of the well-publicized health effects of smoking cg09935388 was much higher in the GECKO dis- on the newborn and the attendant stigma to acknowl- covery cohort (0.46), where the smoking variable was edging this behavior [13] . The methylation signals detect any smoking during pregnancy, than in the combined smoking across pregnancy that is not reported by moth- two replication cohorts (0.16) where the smoking vari- ers. In addition, the degree of methylation difference able was sustained smoking across the pregnancy [10] . at a site captures information about the amount and The methylation signals in GFI1, like other top sites duration of smoking across the pregnancy which influ- in epigenome wide analyses, reflect sustained smoking ences birth weight (or other health effects of maternal during the pregnancy as opposed to smoking that ends smoking) but is not captured by yes or no self-report. early in pregnancy that is captured by the any smok- As a result, the indirect effect captures part of the direct ing variable [36]. Reduced birthweight is also more effect because DNA methylation is less subject to mea- strongly associated with sustained smoking than any surement error than self-reported smoking, identifies smoking [3]. The much stronger proportion mediated falsely reported nonsmokers and captures quantitative observed in the GECKO discovery cohort than for the information on duration and amount relevant to the combined replication cohorts may reflect the greater

Table 5. Estimates of natural direct and natural indirect effects of sustained maternal smoking, assessed by cotinine-enhanced self-report, on birth weight and proportion mediated by CpGs cg09935388 in GFI1 in both naive analyses and after SIMEX correction for measurement error in the Norwegian Mother and Child Study. CpG SN NDE (95% CI) NIE (95% CI) PM cg09935388 Naive -89.4 (-179.5 to 2.6) -27.8 (-59.7 to 5.1) 0.24 0.6 -106.9 (-216.4 to 8.1) -18.7 (-60.4 to 21.7) 0.15 0.7 -102.1 (-205.9 to 16.2) -21.5 (-60.8 to 17.6) 0.17 0.8 -97.9 (-195.5 to 4.2) -23.9 (-59.6 to 11.9) 0.20 0.9 -92.2 (-184.0 to -0.9) -25.7 (-59.1 to 8.3) 0.22 The SIMEX corrected values are presented for four different values for sensitivity of the self-reported maternal smoking exposure variable: 0.6, 0.70, 0.80, 0.90 where specificity = 1. Median and 95% percentile CIs for the bootstrap estimates are in units of grams of birth weight NDE: Natural direct effect; NIE: Natural indirect effect; SN: Sensitivity; PM: Proportion mediated.

future science group www.futuremedicine.com 261 Research Article Valeri, Reese, Zhao et al.

Table 6. Estimates of natural direct and natural indirect effects of any maternal smoking during pregnancy, assessed by self-report, on birth weight and proportion mediated by CpGs cg09935388 in GFI1 in both naive analyses and after SIMEX correction for measurement error in the Norwegian Mother and Child Study. CpG SN NDE (95% CI) NIE (95% CI) PM cg09935388 Naive -16.7 (-84.6 to 52.9) -23.3 (-41.4 to -2.9) 0.58 0.6 -21.1 (-112.1 to 64.6) -26.8 (-51.6 to -0.9) 0.55 0.7 -20.7 (-103.8 to 65.4) -25.8 (-48.0 to -1.7) 0.55 0.8 -18.7 (-96.0 to 63.5) -25.0 (-46.1 to -2.1) 0.57 0.9 -18.3 (-90.0 to 54.3) -24.1 (-43.4 to -2.8) 0.57 The SIMEX corrected values are presented for four different values for sensitivity of the self-reported maternal smoking exposure variable: 0.6, 0.70, 0.80, 0.90 where specificity = 1. Median and 95% percentile CIs for the bootstrap estimates are in units of grams of birth weight. NDE: Natural direct effect; NIE: Natural indirect effect; PM: Proportion mediated; SN: Sensitivity.

misclassification associated with any compared with methylation differences. A reference panel based on sustained smoking during pregnancy. Accordingly, cord blood has become available to estimate cell type when we analyzed the any smoking during pregnancy proportions for analyses of DNA extracted from whole variable in MoBa, we estimated a similarly greater cord blood [46]. When we add these estimated cell proportion of mediation by CpG methylation than types to the model shown in Table 3, there is no fur- obtained with sustained smoking. ther attenuation of the coefficient for birthweight in When we analyzed the cotinine-enhanced exposure relation to sustained maternal smoking beyond that variable, we found that the direct effect of smoking from adding cg09935388 to the model (β = -63.99 g was larger and more precisely estimated and the pro- before cell type adjustment vs -65.99 g after cell type portion mediated was much lower than in the analysis adjustment). While these cell-type correction methods of sustained smoking based on self-report alone. This have limitations, this result provides some reassur- finding has two implications. First, an enhanced mea- ance that the associations between maternal smoking, sure of exposure, with less measurement error, leads to cg09935388 methylation and birthweight evaluated lower estimates of the proportion of mediation by the by [10] and followed up in this paper are not simply smoking CpG signal, a biomarker of the exposure. The due to confounding by cell type. However, it should proportion mediated was reduced more and the direct be noted that neither mediation analysis nor measure- effect was both larger and more precisely estimated ment error correction address the potential influence after SIMEX correction. Given that misclassification of exposure related difference in cell composition on correction may be less widely used than it might be reported ­findings regarding effects of exposure on because of the perception that is generally leads to less methylation. precise effect estimates, this is an important result. Integration of measurement error strategies within These results complement a body of literature on the standard mediation analyses can reduce measurement impact of misclassification in mediation analysis under error bias. It confers the important added benefit of a regression framework. This previous work has focused realistically quantifying uncertainty around the medi- on instances where the mediator, rather than the expo- ation estimates. We illustrate a sensitivity analysis for sure, is measured with error or misclassified [15,16]. In our misclassification bias employing the SIMEX correc- study methylation might be measured with some error, tion approach [30], which can be implemented with however, the three smoking associated CpGs have been easy-to-use software. The SIMEX approach for mis- reported at genome wide significance in at multiple indi- classified categorical variables allows the misclassifica- vidual studies of varying sizes [32,10] and are among the tion mechanism to be dependent on the true exposure top eight findings in a meta-analysis of 13 cohorts [2]. status, as is typically the case for smoking where some Thus they probably have larger effect sizes and are likely smokers report themselves as nonsmokers on surveys. measured with lower error than less robust signals. SIMEX can also be adopted for continuous variables All studies of exposure and methylation in blood and in nonlinear models as well (e.g., in the presence are potentially confounded by effects of the exposure of exposure–mediator interaction) and if a categorical on cell composition. Thus this is a limitation in the outcome, mediator or covariates are misclassified as interpretation of the study of [10], our analyses of the well. MoBa data and any other analyses examining poten- We included gestational age as a linear term tial mediation of health effects by exposure related for comparability with the mediation analysis of

262 Epigenomics (2016) 9(3) future science group Exposure misclassification in mediation analysis Research Article

Kupers et al. [10] . Of note, in our MoBa data, meth- Because current misclassification correction methods ylation at cg09935388 was not significantly related to are even more effective when better exposure measures gestational age (p = 0.79). It is possible that in a study are used, both improved exposure assessment and novel of birthweight a different form of gestational age, such statistical developments in the field of mediation anal- as Z scores, might have been a better adjustment vari- ysis currently underway will improve both the validity able [47]. Nonetheless gestational age as a simple linear and power of mediation analyses to quantify the role of term has been robustly associated in recent genome DNA methylation or other epigenetic signals in medi- wide analyses with methylation at many individual ating the effects of environmental exposures effects on CpG sites [34,35]. Likewise, in both papers, there is human health across the life course. likely measurement error in gestational lag. However, possible misspecification or misclassification of gesta- Supplementary data tional age as an adjustment term in either study would To view the supplementary data that accompany this paper not change our conclusion regarding measurement please visit the journal website at: www.futuremedicine.com/ error in the self-reported smoking variable leading to doi/full/10.2217/epi-2016-0145 potential overestimation of mediation of the smoking birthweight association by methylation at cg09935388, Acknowledgements a strong biomarker of smoking. The authors are grateful to all the participating families in There are limitations to this study. The SIMEX Norway who take part in the MoBa study. The authors thank approach yields only approximately consistent estimates F Day of NIEHS and for expert computing assistance. in small samples and therefore residual measurement error bias is possible despite the seemingly large sam- Financial & competing interests disclosure ple size in our study. We, and no doubt others, plan to This work was supported in part by the Intramural Research address these issues in future work. In the meanwhile, Program of NIH, National Institute of Environmental Health residual effects of misclassification on estimates of Sciences (NIEHS). The Norwegian Mother and Child Cohort mediation are likely, given currently available exposure Study is supported by the Norwegian Ministry of Health and measures and measurement error correction methods. the Ministry of Education and Research, NIH/NIEHS (contract Thus, even when employing measurement error correc- number N01-ES-75558 and ZO1 ES-49019), NIH/NINDS (grant tion, great caution is warranted in the interpretation of number 1 UO1 NS 047537-01) and the Norwegian Research mediation analysis involving a potential exposure bio- Council/FUGE (grant number 151918/S10), and the present marker. While our example focuses on methylation, the project by the Norwegian Research Council/BIOBANK (grant results are applicable to other types of mediators that no 221097). The authors have no other relevant affiliations or may also capture some of the exposure under study. In financial involvement with any organization or entity with a this work we focused on evaluating the impact of mis- financial interest in or financial conflict with the subject mat- classification on quantification of the mediating role of ter or materials discussed in the manuscript apart from those DNA methylation CpG sites at GFI1 reported in the disclosed. literature. Application of misclassification correction No writing assistance was utilized in the production of this approaches in epigenome-wide analysis is an important manuscript. future direction. Ethical conduct of research Future perspective The authors state that they have obtained appropriate institu- Appreciation of the potential overestimation of media- tional review board approval or have followed the principles tion by epigenetic signals of exposure disease relation- outlined in the Declaration of Helsinki for all human or animal ships will lead to both adoption of measurement error experimental investigations. In addition, for investigations in- correction approaches and greater caution in interpret- volving human subjects, informed consent has been obtained ing apparent mediation in environmental epigenetics. from the participants involved.

Executive summary • Analytic results, an extensive simulation study, and analysis of real data show that ignoring exposure misclassification when evaluating mediation of exposure disease relationships, or similar biomarkers of the exposure, can lead to false or exaggerated conclusions regarding mediation. • Measurement error correction approaches that acknowledge potential exposure misclassification can improve validity of findings on potential epigenetic targets on the pathway between environmental exposures and health outcomes. However, even when using these correction approaches caution is warranted in the interpretation of apparent mediation by epigenetic signals are good exposure biomarkers.

future science group www.futuremedicine.com 263 Research Article Valeri, Reese, Zhao et al.

References relation to this exposure at CpGs in the GFI1 gene. In the Papers of special note have been highlighted as: • of interest; current paper, we evaluate this same exposure-methylation- •• of considerable interest birthweight scenario in data from the Norwegian Mother 1 Joehanes R, Just AC, Marioni RE et al. Epigenetic signatures and Child Cohort and assess effects of measurement error of cigarette smoking. Circ. Cardiovasc. Genet. 9(5), 436–447 correction. (2016). 11 Rhomberg LR, Chandalia JK, Long CM and Goodman JE. 2 Joubert BR, Felix JF, Yousefi Pet al. DNA methylation in Measurement error in environmental epidemiology and the newborns and maternal smoking in pregnancy: genome- shape of exposure-response curves. Crit. Rev. Toxicol. 41(8), wide consortium meta-analysis. Am. J. Hum Genet.98(4), 651–671 (2011). 680–696 (2016). 12 Rebagliato M. Validation of self reported smoking. • The large-scale meta-analysis across many birth cohorts J. Epidemiol. Commun. Health 56(3), 163–164 (2002). identifies numerous replicable loci involved in response to 13 Dietz PM, Homa D, England LJ. Estimates of nondisclosure maternal smoking in pregnancy with persistence into later of cigarette smoking among pregnant and nonpregnant childhood. women of reproductive age in the . Am. J. Epidemiolog. 173(3), 355–359 (2011). 3 Reese SE, Zhao S, Wu MC et al. DNA methylation score as a biomarker in newborns for sustained maternal smoking 14 Timms JA, Relton CL, Rankin J, Strathdee G and during pregnancy. Environ. Health Perspect. doi:10.1289/ McKay JA. DNA methylation as a potential mediator EHP333 (2016) (Epub ahead of print). of environmental risks in the development of childhood acute lymphoblastic leukemia. Epigenomics 8(4), 519–536 •• Develops a biomarker in newborns of sustained maternal (2016). smoking during pregnancy using CpGs differentially methylated in relation to this exposure in the Norwegian 15 Valeri L, Lin X and VanderWeele TJ. Mediation analysis when a continuous mediator is measured with error and Mother and Child Cohort Study. the outcome follows a generalized linear model. Stat. Med. 4 Zhang Y, Yang R, Burwinkel B, Breitling LP, Brenner H. 33(28)-4875–4890 (2014). F2RL3 methylation as a biomarker of current and lifetime • Studies bias of measurement error on a continuous smoking exposures. Environ. Health Perspect. 122(2), 131 (2014). mediator in mediation analysis and develops correction approaches using regression calibration, and SIMEX. 5 Zhang Y, Schottker B, Florath I et al. Smoking-associated DNA methylation biomarkers and their predictive value 16 Valeri L and Vanderweele TJ. The estimation of direct and for all-cause and cardiovascular mortality. Environ. Health indirect causal effects in the presence of misclassified binary Perspect. 124(1), 67–74 (2016). mediator. Biostatistics 15(3), 498–512 (2014). 6 Shenker NS, Ueland PM, Polidoro S et al. DNA methylation •• Studies bias of misclassification on a binary mediator in as a long-term biomarker of exposure to tobacco smoke. mediation analysis and develops correction approaches. Epidemiology 24(5), 712–716 (2013). 17 VanderWeele TJ, Valeri L and Ogburn EL. The role of 7 Zhang H, Zheng Y, Zhang Z et al. Estimating and testing measurement error and misclassification in mediation high-dimensional mediation effects in epigenetic studies. analysis. Epidemiology 23(4), 561 (2012). Bioinformatics 32(20), 3150–3154 (2016). 18 General S. The health consequences of smoking – 50 years of 8 Reynolds LM, Wan M, Ding J et al. DNA methylation of progress: a report of the surgeon general. US Department of the aryl hydrocarbon receptor repressor associations with Health and Human Services (2014). cigarette smoking and subclinical atherosclerosis. Circ. www.surgeongeneral.gov Cardiovasc. Genet. 8(5), 707–716 (2015). 19 Magnus P, Birke C, Vejrup K et al. Cohort Profile Update: 9 Morales E, Vilahur N, Salas LA et al. Genome-wide DNA The Norwegian Mother and Child Cohort Study (MoBa). methylation study in human placenta identifies novel loci Int. J. Epidemiol. 45(2), 382–388 (2016). associated with maternal smoking during pregnancy. Int. J. 20 Rønningen KS, Paltiel L, Meltzer HM et al. The Biobank of Epidemiol. 45(5), 1644–1655 (2016). the Norwegian Mother and Child Cohort Study: a resource • Identifies CpGs differentially methylated in placenta for the next 100 years. Eur. J. Epidemiol. 21(8), 619–625 in relation to maternal smoking in a birth cohort and (2006). applies mediation analysis to assess the proportion of the 21 Robins JM and Greenland S. Identifiability and effect of this exposure on birthweight mediated by one of exchangeability for direct and indirect effects. Epidemiology these CpG. 3(2), 143–155 (1992). 10 Küpers LK, Xiaojing X, Soesma AJ et al. DNA methylation 22 Pearl J. Direct and indirect effects. In: Proceedings of the mediates the effect of maternal smoking during pregnancy Seventeenth Conference on Uncertainty in Artificial Intelligence. on birth weight of the offspring. Int. J. Epidemiol. 44(4), Morgan Kaufmann Publishers Inc., CA, USA, 411–420 1224–1237 (2015). (2001). •• Applies mediation analysis to assess the proportion of the 23 Joffe M, Small D, Hsu CY. Defining and estimating effect of exposure to maternal smoking during pregnancy intervention effects for groups that will develop an auxiliary on birthweight explained by differential methylation in outcome. Stat. Sci. 22, 74–97 (2007).

264 Epigenomics (2016) 9(3) future science group Exposure misclassification in mediation analysis Research Article

24 Baron RM and Kenny DA. The moderator–mediator variable 36 Joubert BR, Håberg SE, Bell DA et al. Maternal smoking and distinction in social psychological research: conceptual, DNA methylation in newborns: in utero effect or epigenetic strategic, and statistical considerations. J. Pers. Soc. Psychol. inheritance? Cancer Epidemiol. Biomarkers Prev. 23(6), 51(6), 1173–1182 (1986). 1007–1017 (2014). 25 VanderWeele TJ and Vansteelandt S. Conceptual issues 37 Patrick DL, Cheadle A, Thompson DC, Diehr P, Koepsell concerning mediation, interventions and composition. Stat. T, Kinne S. The validity of self-reported smoking: a review Interface 2, 457–468 (2009). and meta-analysis. Am. J. Public Health 84(7), 1086–1093 26 Valeri L and VanderWeele TJ. Mediation analysis allowing (1994). for exposure–mediator interactions and causal interpretation: 38 MacKinnon DP, Warsi G, Dwyer JH. A simulation study of theoretical assumptions and implementation with SAS and mediated effect measures. Multivar. Behav. Res. 30(1), 41–62 SPSS macros. Psychol. Methods 18(2), 137 (2013). (1995). 27 Sobel ME. Asymptotic confidence intervals for indirect 39 Fraser A, Macdonald-Wallis C, Tilling K et al. Cohort effects in structural equation models. Sociol. Methodol. 13, profile: the Avon Longitudinal Study of Parents and 290–312 (1982). Children: ALSPAC mothers cohort. Int. J. Epidemiol. 42(1), 28 Aigner DJ. Regression with a binary independent variable 97–110 (2013). subject to errors of observation. J. Econometrics 1, 49–60 40 Jaddoe VW, van Duijn CM, Franco OH et al. The (1973). Generation R Study: design and cohort update 2012. Eur. J. 29 Carroll RJ, Ruppert D, Stefanski LA, Crainiceanu CM. Epidemiol. 27(9), 739–756 (2012). Measurement Error In Nonlinear Models: A Modern 41 VanderWeele TJ. Explanation In Causal Inference: Methods Perspective. CRC Press, FL, USA (2006). For Mediation And Interaction. Oxford University Press, UK 30 Küchenhoff H, Mwalili SM, and Lesaffre E. A general (2015). method for dealing with misclassification in regression: the 42 VanderWeele TJ. Bias formulas for sensitivity analysis for misclassification SIMEX.Biometrics 62(1), 85–96 (2006). direct and indirect effects. Epidemiology 21(4), 540 (2010). 31 Lederer W and Küchenhoff H. A short introduction to the 43 Valeri L, Coull BA. Estimating causal contrasts involving SIMEX and MCSIMEX. The Newsletter of the R Project 6(4), intermediate variables in the presence of selection bias. Stat. 26 (2006). Med. 35(26), 4779–4793 (2016). •• Provides theoretical explanation and practical description 44 Richmond RC, Hemani G, Tilling K, Smith GD, Relton of the SIMEX approach for measurement error and CL. Challenges and novel approaches for investigating misclassification correction using the R package ‘simex.’ molecular mediation. Hum. Mol. Gen. 25(R2), R149–R156 (2016). 32 Joubert BR, Håberg SE, Nilsen RM et al. 450K epigenome- wide scan identifies differential DNA methylation in 45 Burgess S, Daniel RM, Butterworth AS, Thompson newborns related to maternal smoking during pregnancy. SG, EPIC-InterAct Consortium. Network Mendelian Environ. Health Perspect. 120(10), 1425–1431. (2012). randomization: using genetic variants as instrumental variables to investigate mediation in causal pathways. Int. J. 33 Backe B. [Routine ultrasonography in obstetric care in Norway, Epidemiol. 44(2), 484–495 (2015). 1994]. Tidsskr. Nor Laegeforen. 117(16), 2314–2315 (1997). 46 Bakulski KM, Feinberg JI, Andrews SV et al. DNA 34 Bohlin J, Håberg SE, Magnus P et al. Prediction of methylation of cord blood cell types: applications for mixed gestational age based on genome-wide differentially cell birth studies. Epigenetics 11(5), 354–356 (2016). methylated regions. Genome Biol. 17(1), 207 (2016). 47 Fenton TR, Kim JH. A systematic review and meta-analysis 35 Knight AK, Craig JM, Theda C et al. An epigenetic clock to revise the Fenton growth chart for preterm infants. BMC for gestational age at birth based on blood methylation data. Pediatr. 13(1), 1 (2013). Genome Biol. 17(1), 206 (2016).

future science group www.futuremedicine.com 265 SPECIAL FOCUS y Epigenetics of the immune system & alterations in inflammation & autoimmunity Review

For reprint orders, please contact: [email protected]

9 Review 2017/03/30 Epigenetics and allergy: from basic mechanisms to clinical applications

Epigenomics Allergic diseases are on the rise in the Western world and well-known allergy- Daniel P Potaczek‡,1,2,3,4, Hani protecting and -driving factors such as microbial and dietary exposure, pollution and Harb‡,1,2,3, Sven Michel5, Bilal smoking mediate their influence through alterations of the epigenetic landscape. Alashkar Alhamwe1, Harald 1,2,3 ,6 Here, we review key facts on the involvement of epigenetic modifications in allergic Renz & Jörg Tost* 1Institute of Laboratory Medicine diseases and summarize and critically evaluate the lessons learned from epigenome- and Pathobiochemistry, Molecular wide association studies. We show the potential of epigenetic changes for various Diagnostics, Philipps-Universität clinical applications: as diagnostic tools, to assess tolerance following immunotherapy Marburg, Marburg, Germany or possibly predict the success of therapy at an early time point. Furthermore, new 2International Inflammation (in-FLAME) technological advances such as epigenome editing and DNAzymes will allow targeted Network, Worldwide Universities Network (WUN) alterations of the epigenome in the future and provide novel therapeutic tools. 3German Centre for Lung Research (DZL) 4John Paul II Hospital, Krakow, Poland First draft submitted: 21 November 2016; Accepted for publication: 30 January 2017; 5Secarna Pharmaceuticals GmbH & Co KG, Published online: 21 March 2017 Planegg, Germany 6Laboratory for Epigenetics & Keywords: allergy • asthma • cell-free DNA • DNA methylation • environment • epigenetic Environment, Centre National editing • EWAS • exposure • food allergy • FOXP3 • immunotherapy • Th cell lineages de Génotypage, CEA-Institut de Génomique, Evry, France *Author for correspondence: Allergic diseases display a broad spectrum of of the conditions may present simultaneously Tel.: +33 016 087 8423 Fax: +33 016 087 8485 clinical manifestations and conditions. The such as eczema together with food allergies or [email protected] major group of them is mediated by immuno­ eczema together with bronchial asthma [4,5]. ‡Authors contributed equally and should globulin E (IgE) and is commonly referred As this co-occurrence might arise even prior be considered as joint first authors to as atopic disorders. These include aller­ to the first actual ingestion of the food, the gic bronchial asthma (main manifestations early postnatal period seems to be of crucial in the lung and the lower airways), allergic importance for the development of IgE-medi­ 4 rhinitis (AR; hay fever; the upper respira­ ated allergies [6]. Although still under debate, tory tract), allergic conjunctivitis (the eyes), IgE-mediated food allergies seem to be (extrinsic) atopic dermatitis (AD; eczema; increasing at a high rate in the Western coun­ the skin) and food allergies (the upper and tries and concern now approximately 3–5% 2017 lower gastrointestinal tract) (Figure 1). of children and about 1–3% of adults, with Although each patient shows an individual substantial geographical and ethnical differ­ spectrum of these conditions during the ences [7–9]. Typical childhood associated food entire life span, certain patterns are well rec­ allergies, such as cow’s milk allergy (CMA), ognized. This includes the ‘atopic (allergic) also increasingly persist until later age. At march’, which describes the prototypical early age, there is a strong sex bias in favor sequential manifestation throughout life, of males in the development of asthma and starting with atopic eczema or food allergies, other allergic diseases including food allergy, shifting then later to respiratory conditions, which is inversed after puberty and through­ during which compromised barrier function out adulthood with a higher prevalence in at one surface will lead to other dysfunctional women [10–14] . Allergic disorders not associ­ part of epithelial barriers [1–3] . Furthermore, several ated with IgE do also exist such as allergic

10.2217/epi-2016-0162 © Tost et al. Epigenomics (2017) 9(4), 539–571 ISSN 1750-1911 539 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

contact dermatitis, non-IgE-mediated food allergy or are characterized by a polygenetic signature. A great intrinsic AD [15,16]. Furthermore, different subtypes of variety of genes encoding molecules regulating and/ asthma can arise unrelated to IgE (non-atopic asthma) or participating in T cell activation, B cell develop­ and although these might represent different disease ment and allergen presentation, cytokines and cyto­ entities, they share a number of similarities includ­ kine receptors, chemokines, growth factors, proteins ing the below described cytokine shift and resulting involved in remodeling, wound healing and repair, inflammation as well as mast cell activation [17,18]. As epithelial signatures, members of metabolic pathways many publications studying epigenetic modifications and others have been identified as susceptibility loci in asthma do not provide extensive information on the for allergic diseases and related intermediate pheno­ type of asthma in the recruited individuals, it should types or quantitative traits [21,26–29]. Despite these be kept in mind that this review describes results from advances, the polygenetic signature cannot explain the studies including probably a variety of asthma phe­ dramatic increase in incidence and prevalence of atopic notypes, which might be a confounding factor for diseases throughout the past 70 years. It is now well ­epigenetic changes. accepted that environmental factors and (Western) Common to the atopic syndrome is the presence lifestyle conditions play an important role in initiating of chronic inflammation at body–environment inter­ and maintaining type-2 inflammation inpredisposed ­ faces [19,20]. This inflammation covers the muco­ individuals­ [19,22,23,30–32]. sal membranes and the skin, and reflects an active, Based on this paradigm, allergic diseases are now chronic misguided immune response directed against considered prototypic examples of conditions deter­ otherwise harmless environmental antigens. Such anti­ mined by gene × environment interactions [33,34]. A few gens are termed ‘allergens.’ This imbalanced immune decades ago, research efforts focused on the identifi­ response is characterized by allergic inflammation cation and characterization of risk factors, which may including the development of T-helper (Th)-2 (Th2) favor the development and/or maintenance of allergic immune responses as the consequence of allergen diseases. Key risk factors include smoking (active or presentation by innate immune cells termed antigen- passive), allergen exposure levels, ozone, diesel exhaust

presenting cells (e.g., dendritic cells, macrophages) to particles (DEPs), SO2 and NO2 [32,35]. Furthermore, naive T cells. Following the exchange of a set of sig­ viruses play an important role in disease exacerba­ naling events, these naive T cells develop into Th2 tion or maybe even development, especially in case of effector cells, which are defined and characterized by asthma [36–38]. Bacterial superinfections, particularly the production of a unique set of so-called ‘type-2’ with endotoxin-positive Staphylococcus aureus strains (see section on ‘T cells’) cytokines including interleu­ and other microbes represent important contribu­ kin (IL)-4, IL-5, IL-13 and IL-9, the latter considered tors to the pathophysiology of AD and other allergic now to by synthesized predominantly by Th9 cells ­diseases [39–41] . (Figure 1) [21–24]. These cytokines direct the effector Since not everybody with a genetic predisposition phase of the allergic response [21–23]. IgE molecules eventually develops an atopic disease, recent research bind to high- (FcεRI) and low-affinity (FcεRII, CD23) focuses on potential protective factors. Particularly, IgE receptors. FcεRI is expressed particularly on mast the insufficient exposure to (infectious) microorgan­ cells and basophils. Allergen binding to IgE molecules isms (the hygiene hypothesis) or dietary and bacte­ occupying FcεRI on the surface of those cells results rial metabolites, and frequent use of antibiotics and in cross-linking of these receptors and subsequent cell antipyretics have gained broad attention [30,42]. An degranulation and mediator release, leading to the important example for this concept is the traditional development of allergic symptoms typical for type I farming environment, particularly present in small (immediate) hypersensitivity reactions (Figure 1) [21,25]. Alpine villages, which confers a strong protection to Recently, it was shown that not only Th2 cells secrete allergy. This has been attributed to the direct exposure type-2 cytokines but also innate lymphoid cells (ILCs) to farm animals, consumption of non-homogenized type 2 (ILC2) are important producers of these media­ and non-pasteurized milk, a high prevalence of breast- tors. The orchestration of the inflammatory response feeding, intra-uterine exposure to airborne microbial by the above-mentioned cellular components together factors and the inhalation of microbes and microbial with the cytokine mediators is now termed ‘type-2 compounds throughout early childhood [19,33–35,43]. inflammation’(Figure 1) [22,23]. This concept is now further expanding and offers great The development of the atopic syndrome requires a opportunities for the development of novel allergy and genetic predisposition. Great efforts have been under­ asthma protective strategies. taken to unravel the genetic basis of allergic diseases In order to advance in this field, it is critical to bet­ and it is now accepted that allergic/atopic diseases ter understand the cellular and molecular mechanisms

540 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

B cells Allergens IgE Naive T cells Th2 cells MHC II ILL-4 TCR ILIL-13

APCs ILIL--99 IL--44 Basophils IL-9- IL-13 Mast cells FCεRI IL-4 IL-5 Degranulation: IL-13 IL-4 release of histamine IL-5 IL-13 and other mediators

Mucus hyperproduction

Eosinophilia Impaired epithelial barrier Increased permeability (leakage)

Smooth muscle contractility

SharSha ed pathophysiology • High IgE production/secretion • Eosinophilia • Epithelial hyperplasia • Basal membrane thickening • Barrier disruption • Inflammatory infiltration

Allergic asthma Allergic rhinitis Atopic dermatitis (eczema) Food allergy (local) • Smooth muscle • Rhinorrhea • Inflamed skin • Nausea contractility/ – Excessive fluid secretion • Skin lesions • Vomiting bronchoconstriction – Mucus hyperproduction • Itching • Diarrhea • Airway hyper-responsiveness • Nasal polyps • Impaired skin barrier • Stomach cramps • Mucus hyperproduction • Chronic sinusitis • Transepidermal water loss • Abdominal pain • Wheezing/dyspnea/cough

Figure 1. Allergic (type 2) inflammation: basic mechanisms and pathophysiology, and selected clinical consequences. Th2 cytokines can be synthesized also by innate lymphoid cells type 2 (ILC2), which is not shown. Contribution of Th9 cells, major IL-9 producers, is also not presented. MHC-II denotes MHC class II molecules; TCR, T cell receptor. For remaining abbreviations, detailed description of the figure content and additional information, please, refer to the main text. This figure was inspired by several previously published images, especially by the one by Gandhi et al. [23]. through which environmental factors influence the environmental factors, microbial and non-microbial initiation and/or maintenance of the disease. In this components on disease development. Epigenetic path­ regard, epigenetic mechanisms may contribute signifi­ ways could thus mediate the gene × environment inter­ cantly to the answer [19,31,32,44]. Epigenetics may explain actions. This field is also gaining increasing attention the high degree of plasticity among immune responses due to the potential use of epigenetic signatures as bio­ observed throughout life, and the impact of external markers or interfering therapeutically with the epigen­

future science group www.futuremedicine.com 541 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

etic code. The knowledge gained over the last few years groups added [32,56–62]. DNA methylation and histone in this emerging field of research will be summarized modifications mutually interact [63]. As a more detailed in this article. description is beyond the scope of this review, the inter­ ested reader is referred to more detailed review articles, Basics of epigenetics for example, those referenced in this subchapter. DNA methylation & histone modifications Epigenetic modifications are biochemical changes MicroRNA of the chromatin, in other words, DNA or histones, In addition to the ‘classical’ epigenetic mechanisms that are functionally relevant, but do not affect the including DNA methylation and histone modifica­ nucleotide sequence of the genome. Although they tions, also some post-transcriptional control elements are thought to contribute to some other processes, for such as microRNAs (miRNAs) have become widely instance response to damage and DNA repair, epigen­ recognized as important epigenetic regulators of gene etic modifications are best known for their effects on expression [64,65]. These approximately 22-nt noncod­ the accessibility of certain genomic loci to transcrip­ ing RNA molecules are highly abundant, with more tion enzymes and thus their expression. DNA methyla­ than 2500 mature miRNA molecules characterized in tion, a covalent addition of a methyl group, occurs at humans. Canonical miRNAs are encoded in humans the cytosine nucleotide belonging to CpG dinucleotide mostly (but not solely) by introns of both non-cod­ (called ‘CpG site’), which is a DNA sequence where a ing or coding transcripts. They are transcribed from cytosine nucleotide (C) is directly followed by a gua­ dsDNA usually by RNA polymerase II. The tran­ nine nucleotide (G). CpG sites frequently cluster to scripts are further processed by two RNase III-type form ‘CpG islands’, typically located in the elements of enzymes, first in the nucleus (Drosha) and then in a gene regulatory element with impact on its transcrip­ the cytoplasm (Dicer). To exert its function, mature tion, for example, promoters or enhancers [45–47]. Low miRNAs become incorporated into the RNA-induced DNA methylation levels in a promoter region are often, silencing complex. The RNA-induced silencing com­ but not necessarily, associated with a higher transcrip­ plex is in turn guided by miRNAs to specifically target tional activity (or at least the potential of the gene to mRNAs. This leads to the cleavage or degradation of become expressed), while high DNA methylation levels the bound mRNA molecule or suppression of its trans­ in the CpG island of a promoter are usually associated lation by reducing the speed of the ribosomal machin­ with lower gene expression up to full gene silencing. ery. The magnitude of the silencing effect depends on The reaction of DNA methylation is catalyzed by DNA the level of complementarity between the miRNA and methyltransferases (DNMTs), including DNMT1 the targeted mRNA [64–69]. It has been proposed that (target: hemimethylated dsDNA; role: maintenance of the post-transcriptional regulation of gene expression DNA methylation patterns after DNA replication) and facilitated by miRNAs plays an important role in the DNMT3A and DNMT3B (target: hemimethylated fine-tuning of transcriptional programs in the context and unmethylated dsDNA; role: de novo DNA methyl­ of bigger regulatory networks and in buffering fluc­ ation) [48–51] . The best-characterized post-translational tuations in gene expression resulting from random histone modifications include phosphorylation, ubiqui­ internal cellular modulation and/or environmental tination, acetylation and methylation, the last two of influences [70]. Considering their biological impor­ which are the most extensively studied [52,53]. Histone tance, miRNAs have been involved in multiple human acetylation occurs at the lysine residues and it is cata­ pathologies [66]. These include also allergic diseases, in lyzed by histone acetyltransferases, while the opposite which the role of miRNAs has been rather extensively reaction by histone deacetylases (HDACs). Higher his­ studied [71–76]. Not surprisingly, analyzing miRNAs tone acetylation levels lead to less tight wrapping of the might become an important diagnostic tool in aller­ proteins around the DNA increasing its accessibility to gies [77–79]. It is also worth mentioning that the mecha­ the transcriptional machinery, whereas decreased his­ nism of RNA-mediated silencing of gene expression tone acetylation has the opposite effect. Histone acety­ has been utilized in biomedical research as a powerful lation independent of the position of the lysine amino laboratory tool [80] and in therapeutic applications as acid generally correlates with potentially active genes or one of the possible antisense approaches [81] . However, gene regulatory elements [54,55]. Histones can become due to space restrictions, we will in this review focus methylated at lysine or arginine residues and this reac­ on the ‘classical’ epigenetic mechanisms, DNA meth­ tion is catalyzed by histone methyltransferases. How ylation and histone modifications. For more informa­ histone methylation influences the chromatin conden­ tion on miRNAs, readers are referred to other articles sation and thus gene expression depends on the location including those referenced in this short subsection. of the amino acid residue and on the number of methyl Connections between miRNAs and DNA methyla­

542 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review tion/histone modifications have also been described in naive CD4+ T cells under the influence of IL-4 and detail elsewhere [64,66,68,69,82]. transforming growth factor-β (TGF-β) but they can possibly differentiate also from TGF-β stimulated Th2 T cells: importance in allergy & epigenetic cells [83,93]. Although several TFs such as interferon regulation of differentiation regulatory factor 4 (IRF4; gene IRF4), PU.1 (gene: Many types of immune cells contribute to the devel­ SPI1) and others are involved in the development of opment of atopic predisposition and to the effector the Th9 phenotype, none of them can be considered phase of allergic inflammation (Figure 1), with several a master TF here (in the same way as e.g., GATA3 for subpopulations of Th cells differentiating from naive Th2 or TBX21 for Th1). On the other hand, some CD4+ T cells playing a pivotal role in those processes believe that PU.1 is the most important Th9 regula­ and their modulation (Figure 2). The direction of naive tor since it is the only TF able to convert another Th + CD4 T cell maturation determines their subsequent linage into Th9 (Figure 2) [24,93–95]. TGF-β and IL-6 effects, including those on allergy. T cell differen­ together stimulate naive CD4+ T cells to express their tiation is strictly regulated, with changes in epigen­ master TF, RAR related orphan receptor C isoform etic marks at main lineage-determining loci playing a 2 (RORC2, traditionally called also RORγT), which pivotal role. In this review, only the most important is one of the two isoforms expressed from the same humoral and selected epigenetic mechanisms modu­ encoding locus (gene: RORC), and thus to differen­ lating naive T cell fate along with the major allergy- tiate toward Th17 cells [85,96–98]. As indicated by the related effects of mature Th subpopulations, such as name, Th17 cells produce IL-17A (traditionally called Th1, Th2 (and Th9), regulatory T cells (Treg cells) also IL-17; gene IL17A) and IL17F (gene: IL17F ), but and Th17, are shortly outlined (Figure 2). also IL-21 and -22. Allergy-related effects of those cells Differentiation toward each of the above-mentioned comprise airway hyper-responsiveness as well as pro­ T cell types is driven by a precisely defined cytokine motion of neutrophils and therefore their participation milieu (one major cytokine or a set of most impor­ in the forms of asthma, in which neutrophils predomi­ + tant cytokines) influencing expression of lineage- nate [91,96,98,99]. Stimulation of naive CD4 T cells with specific transcription factors (TFs), so-called master only TGF-β results in the expression of the forkhead TFs or master regulators (Figure 2) [31,83–87]. Th1 cells box protein 3 (FOXP3; gene FOXP3) TF and their develop from naive CD4+ T cells under the influ­ maturation toward Treg cells, which secrete IL-10 and ence of interferon-γ (IFN-γ; gene: IFNG) and IL-12. TGF-β [84,98,100,101]. Major effects of these cytokines in Stimulation with those cytokines leads to the expres­ allergy include suppression of inflammation (TGF-β sion of T-box 21 (TBX21, traditionally called also and IL-10), and promotion of airway ­remodeling T-bet; gene: TBX21), a master TF of the Th1 lin­ (TGF-β) (Figure 2) [91,98]. eage and secretion of IFN-γ, a major Th1-produced Differentiation and synthesis of specific cytokines cytokine [88–90]. The most important allergy-related by the mature T cells populations is strictly con­ effect of Th1 cells is the inhibition of Th2 lineage dif­ trolled, with a prominent role of epigenetics (Figure 2). ferentiation and thus type-2 inflammation [91]. The In general, important regulatory regions of the genes maturation of naive CD4+ T cells toward Th2 cells encoding master (or very important) TFs and cyto­ occurs under the influence of IL-4, which stimulates kines characteristic for a ‘target’ T cell lineage undergo the expression of GATA binding protein 3 (GATA3; changes of the epigenetic landscape favoring transcrip­ gene: GATA3) [31,83,92], a Th2 master TF. Th2 cells tion. Simultaneously, regulatory elements in respective produce IL-4 (gene: IL4), IL-13 (gene: IL13), IL-5 genes of TFs or cytokines specific for the other T cell and IL-9 (gene: IL9), the latter considered recently types are subjected to modifications having silencing to by synthesized predominantly by a separate T cells properties, especially if their expression would oppose subpopulation, namely Th9 cells (see further) [22]. the development of the ‘target’ phenotype [31,83,102–104]. The importance of Th2 cytokines for allergic (type 2) DNA methylation of the genes specific for Th1 cells inflammation is actually fundamental as they simply is a very good example here. In Th1 cells, loci charac­ mediate its development [21–24,91] . In more detail, major teristic for this lineage such as TBX21 and IFNG are effects of those cytokines include differentiation of fur­ demethylated during the maturation, whereas those ther Th2 cells and B cell class switching toward IgE essential for Th2 (IL4, IL13) or Th17 (RORC, IL17A) (IL-4), activation and survival of eosinophils (IL-5), cells remain methylated. DNA methylation of IFNG airway hyper-responsiveness, hyperplasia of goblet cells is in turn preserved in Th2 and Th17 cells. In addi­ and mucus production (IL-13) and proliferation/sur­ tion, IL4 and IL13 loci are hypomethylated in Th2 vival of mast and ILC2 cells and mucus production cells [31,83,88,104–108]. RORC remains methylated and (IL-9) [21–24,91] (Figures 1 & 2). Th9 cells develop from FOXP3 undergoes demethylation in developing Treg

future science group www.futuremedicine.com 543 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

Th2 • Th2 cell differentiation, B cell class switching towards IgE (IL-4) • Activation and survival of eosinophils (IL-5) • AHR, goblet cell hyperplasia, mucus production (IL-13)

IL-4 GATA3 Th1 IL-5 Inhibition of Th2 cell differentiation IL-13 IFNG CpG Th17 IL4 CpG IL13 CpG IFN-γ TGF-β Airway neutrophilia, AHR TBX21 IFNG H3ac TNF-α IFNG H3K4me2 IFNG H3K9me2 IL-17 RORC IFNG H3K27me2/3 IL-21 IL4 H3K4me3 IL-22 IFNG CpG IL17A CpG IL-4IL-4 TBX21 CpG IFNG CpG TGF-β RORC CpG IL17A CpG IFN-IFN-γ IL-6IL-6 IL4 CpG TBX21 CpG IL-12IL-12 IL13 CpG RORC CpG IFNG H3ac FOXP3 CpG IFNG H3K4me2/3 IL17A/IL17F H3ac IFNG H3K9me2 IL17A/IL17F H3K4me3 IFNG H3K27me2/3 RORC H3K4me3 Naive T cell

RORC CpG FOXP3 CpG TGFGF-β TTGFGF-β RORC H3K4me3 IL-4-4 FOXP3 H3K9/14ac FOXP3 H3K4me3 IFNG CpG IL4 CpG PU.1 IL-9 IL13 CpG IRF4 IL-10 IL17A CpG RORC CpG IL-10 FOXP3 CpG FOXP3 TGF-β

Th9 Mast cell proliferation and activation, IL9 H3ac Treg ILC2 survival, mucus production IL9 H3K9ac • Suppression of in ammation; promotion of IL9 H3K18ac IL9 H3K27me3 airway remodeling (TGF-β) IL9 H4ac • Suppression of in ammation (IL-10) SPI1 H3ac : H3K27me3 SPI1 H3K4me3 : H3K27me3

Figure 2. Major types of Th cells, their differentiation and its epigenetic regulation, and their crucial allergy-related functions. DNA methylation (CpG) status of the loci pivotal for Th lineages is shown. Levels of epigenetic histone marks at loci specific for Th subpopulations is depicted (selection). For more details and abbreviations, please, refer to the main text. This figure was inspired by several previously published images, especially by the one by Suarez-Alvarez et al. [83].

cells, while the situation in Th17 cells is completely in the literature focusing on Th9 cells [24,94,95,103] as opposite, with DNA demethylation of RORC and sus­ well as in the papers describing mechanisms of CD4+ tained methylation of the FOXP3 locus. DNA demeth­ T cell differentiation in a wider context [87,91,93,102,104]. ylation of IL17A and TBX21 is also observed in Th17 However, other types of epigenetic modifications cells (Figure 2) [31,83,106,108–112]. such as histone acetylation or methylation have been Our knowledge on the epigenetic regulation of reported to contribute to Th9 lineage differentiation Th9 cells differentiation is comparatively limited, (Figure 2) [95,104]. The levels of total H3 (H3ac) and which may result from the relative novelty of this only H4 (H4ac) as well as specific H3K9 [H3K9ac; K cor­ recently identified Th cell lineage. Not much is known responds to a lysine residue which can be further modi­ about Th9-specific DNA methylation changes; at least fied with acetyl or methyl (here: acetyl) groups] and our search failed to identify any relevant information H3K18 (H3K18ac) histone acetylation, all having a

544 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review permissive character, have been found to be the high­ human. Finally, although Th cells play a crucial role in est, while trimethylation of H3K27 (H3K27me3), the development of allergy and in the mechanisms of a repressive histone modification [62], to be lowest at allergic inflammation, many other cells whose matura­ the IL9 locus in naive T cells cultured under Th9 dif­ tion is also epigenetically regulated contribute as well, ferentiating conditions when compared with Th cells for instance B cells (Figure 1) [83]. cultured toward the Th1, Th2 or Th17 lineage [95,113]. In addition, the negative ratio of permissive to repres­ Epigenetic modifications mediating the sive histone modifications (H3ac/H3K27me3 and influence of environmental factors in allergic H3K4me3/H3K27me3) at the SPI1 promoter char­ diseases acteristic for naive T cells has been reported to revert The genetic susceptibility to allergic disorders is under Th9-polarizing conditions leading to the induc­ known to have a polygenic character [21,26–29,43,124,125]. tion of PU.1 expression [104,114]. Histone modifications Since it had not been possible to explain the increase play an important role also in the development of the in incidence and prevalence of allergy and related dis­ other Th lineages. For example, the IFNG locus in Th1 orders throughout several decades solely by changes in cells is characterized by high H4ac, high H3K4me2 the genetic background, the contribution of modify­ (di-methylation; a permissive mark), high H3K9me2 ing factors of environmental character became evident, (usually a repressive mark; hence its role in this context which gained further support from epidemiological requires further investigation) and low H3K27me2/3 observations. For instance, although the genetic pro­ levels. At the same time, an opposite pattern has been file of the population remained unchanged, the inci­ identified in Th2 cells, with the exception of transient dence of allergies increased dramatically in the former increases in H3K4me2 and H3K9me2 during the German Democratic Republic (East Germany) in a early phases of the differentiation [88,115]. Moreover, period of only 20 years since the reunification of Ger­ high levels of H3K4me3 were observed at IFNG in many [126–128]. Furthermore, epigenetic mechanisms, Th1, IL4 in Th2, IL17A and IL17F in Th17, RORC in known to convey genomic adaptation to the external Th17 and a subgroup of Treg cells, and FOXP3 in Treg circumstances, became a natural candidate to explain cells [116] . Simultaneously, IFNG, IL17A, IL17F and these effects of the environment on the development of RORC, and to some extent also FOXP3 and IL4, were allergy [31,129]. characterized by high levels of H3K27me3 in Th lin­ The unfavorable change in the incidence of allergic eages listed in the previous sentence, except for those disorders could be explained by an increased contribu­ in whom H3K4me3 levels have been high [116] . It has tion of negative (risk or risk-increasing) environmental been also demonstrated that promoters of IL17A and modifiers, decreased exposition to positive (protec­ IL17F are in Th17 cells characterized by high levels tive or risk-reducing) environmental factors or both. of not only H3K4me3 but also H3ac [117] . In addition Although this review does not aim to profoundly to H3K4me3, increased levels of several other histone discuss the mechanisms of inheritance, it is worth marks, especially H3K9/14ac (both permissive modi­ mentioning that, theoretically, environmental influ­ fications), are present at theFOXP3 locus in Treg cells ences could not only modify genetic predisposition, (Figure 2) [118]. Inhibition of histone deacetylation, i.e., change it to a certain degree, but also completely especially targeting HDAC9, increased the number eliminate it. For example, a person with genetic suscep­ and suppressive potential of Treg cells concomitant tibility to bee allergy living in an island not inhabited with a demethylation of the FOXP3 regulatory regions by Hymenoptera will never develop this type of atopic and decreased inflammation in inflammatory bowel sensitization and thus related symptoms of allergic disease [119] . reaction. Naturally, we cannot address here the whole com­ According to the hygiene hypothesis, early-life plexity of Th cell differentiation and its epigenetic reg­ microbial exposure correlating with the number of ulation. As such, only major cell lineages and related children in a family offers protection against the devel­ cytokines are described. Gene expression mechanisms opment of allergic conditions. Considering changes with more complex epigenetic aspects, for example in the life-style, such as declined family size but also the role of the Th2 locus control region or conserved improvements in public health and hygiene, it should noncoding sequence (CNS) elements present in several not be surprising that the incidence of allergic disor­ loci, for example in FOXP3 (Treg-specific demethyl­ ders has substantially increased over the last several ated region – TSDR; also known as CNS2) [21,86,93,120– decades [19,30,33–35,42,43,126]. However, exposure to 123], are not discussed due to space restrictions. We also microorganisms or their structural parts present in do not distinguish between human and mice, and the various indoor or outdoor environments can occur also simplified genetic nomenclature used always refers to in the absence of infection. Such microbial substances,

future science group www.futuremedicine.com 545 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

recognized by the innate immune system, can also intrauterine regulation of CD14 to be involved in the stimulate chronic inflammation leading in turn to the protective effects of farming against allergy develop­ development of the protection against inflammatory ment [148] . Prenatal exposures not only to microorgan­ disorders [30,33,35,126,130–133]. This has been demonstrated isms but also to other environmental factors such as also for atopic diseases, especially in the investigations pollutants or nutrients (and resulting changes in the on the effects of farming on allergic predisposition. epigenetic status) are in general getting more and more Those studies have shown that children exposed early recognition as important if not crucial players in the in life to microbiota present in farming/rural envi­ development of allergic predisposition [32,129,150–152] and ronments develop less allergic conditions (than their are discussed in more detail in other sections of this inner-city counterparts) [30,33,35,126,131,132,134]. review. It is also worth mentioning that another major Substantial efforts have been undertaken to clarify factor involved in direct or transmaternal anti-allergic the mechanisms of this protective phenomenon, exem­ effects of farming is consumption of unprocessed cow’s plified by the studies onAcinetobacter lwoffii (A. lwoffii) milk obtained directly from a farm, not necessarily due F78, a strain, which has been identified in the farming to its microbial contamination/content [153–155]. In any environment along with Lactococcus lactis G121, Staphy- case, FOXP3 demethylation and activation of Treg cells lococcus sciuri W620 and a number of other bacteria not seem to be involved in protective effects of farm milk listed here [135,136]. All three bacterial isolates have been against allergies (see also later) [144,153]. However, lon­ shown to reduce allergic reactions in mice and to induce gitudinal studies following children from in utero and Th1-polarizing program in human dendritic cells in during childhood with repeated biological samplings vitro [135,137]. A subsequent study in mice demonstrated will be required to assess if the observed ­epigenetic protective effects of maternal intranasal exposure to A. changes precede the onset of allergic diseases. lwoffii against the development of atopic asthma-mim­ In contrast, viruses and viral infections do not pro­ icking allergic airway inflammation in the progeny [138] . tect against atopy, but rather result in exacerbations or Finally, this transmaternal protection was shown to be maybe even development of allergic disorders, with a IFN-γ-dependent, which was, at least partly, medi­ remarkable example of airborne viruses and (atopic) ated by the A. lwoffii-induced preservation of histone asthma [36–38,156]. Among airborne viruses, human H4 acetylation at the IFNG promoter of CD4+ T cells rhinoviruses (HRVs) and respiratory syncytial virus ­isolated from spleens of the mice offspring [139] . (RSV) seem especially important here. While RSV Transmaternal anti-allergic protection induced by is frequently detected in infants with bronchiolitis A. lwoffii is in line with some more recent concepts. It and subsequent wheezing, HRV is the most com­ used to be commonly accepted that the first contact of mon (viral) asthma exacerbation trigger at any age fetus/neonate with microorganisms occurs during the after [36,157]. Furthermore, it is thought that repeated delivery, and results from the exposure to maternal vag­ early life viral infections, especially those involving the inal and fecal bacterial microflora [30,126,131,140]. Indeed, lower respiratory tract, can lead to the development of children born by Caesarian section have been reported atopic asthma that can persist through childhood and to be at higher risk of allergies when compared with into adulthood [36,37,157,158]. This effect is especially those delivered by vaginal birth, although the data are strong if repeated viral infections are combined with conflicting [131,140–143]. However, it has become quite atopic predisposition, and mechanisms linking the evident that microorganisms can stimulate the devel­ contribution of innate antiviral and atopic immunoin­ opment of anti-allergic protection even earlier, in the flammatory pathways have also been proposed [37,159– prenatal period [144,145] either through a direct in utero 161] . FcεRI is thought to play an important role in this contact with fetus, as suggested by the studies dem­ mechanism [37,159,160], which seems to get some indirect onstrating the presence of bacterial DNA in placenta support from genetic associations observed for single- or meconium of preterm babies [146,147], or indirectly, nucleotide polymorphisms in FCER1A encoding the with the mother influencing the neonatal immune sys­ FcεRI α-subunit [162,163]. We are not aware of any data tem development in response to contact with micro­ showing pro-allergic effects of airborne viruses to be biota. Interestingly, differential expression (mRNA epigenetically mediated but it seems the most reason­ level) of a number of immune-related genes in placenta able possibility [164] . It is important to mention that the in respect to lifestyle (anthroposophy and living on location of airway respiratory infection may be critical a farm) and parental allergen sensitization has been for its effect on allergic predisposition [158] . In addition, reported [148,149]. A follow-up study on one of those inverse associations between infections with some non- loci, namely CD14, revealed a relation between DNA airborne viruses such as hepatitis A virus and allergies methylation of the CD14 promoter region with the have been reported, although inconsistently between level of CD14 mRNA expression, suggesting epigenetic different populations/studies [156,165].

546 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

Pollen exposure (pollen season) is an obvious and were observed as well [174] . It has been also found that well definable environmental factor influencing aller­ intermediate or high levels of blood DNA methylation gic disorders. An interesting study on seasonal AR at a CpG island in the 5′-untranslated region of ADRB2, (SAR) has been conducted, which combined the the gene whose polymorphic variants are known to usage of DNA methylation and gene expression arrays. affect the response to therapy with β2-adrenoreceptor Methylation patterns observed in isolated and in vitro agonists [26,175], have been found to be associated with + cultured CD4 T cells (but not mRNA expression pro­ severe childhood asthma [176] . Moreover, a relationship files) made it possible to clearly distinguish between between indoor exposure to NO2 and severe asthma in samples obtained from SAR patients and controls [166] . children with high (but not low) ADRB2 methylation An additional nice feature of the study was that the levels has been observed [176] . samples were collected from each participant within Smoking, either as prenatal and/or postnatal tobacco and outside the pollen season and thus also the envi­ smoke exposure or in its active form, is another extrin­ ronmental influence on the DNA methylation could sic factor known to increase the risk of allergic disor­ be directly investigated. Moreover, the methylation ders, especially asthma [177–181], and its effects on the profiles were found to be significantly associated with epigenetic status have been clearly demonstrated [182– disease severity in patients during the season [166] . 185]. For example, active smoking was shown to influ­ Furthermore, pollutants, especially air pollutants ence methylation of DNA isolated from peripheral increase the risk of developing allergic disorders and blood [184], and in utero tobacco smoke exposure was these effects are at least partly epigenetically mediated found to affect methylation patterns of DNA obtained and are shortly exemplified below. Greater average from buccal cells or whole blood (of the offspring), polycyclic aromatic hydrocarbons exposure has been fetal lungs and placenta [182,183,185]. Moreover, an exten­ shown to be associated with higher DNA methylation sive study on the effects of prenatal smoking involv­ at several CpG sites within the FOXP3 locus of periph­ ing not only array but also whole genome sequencing eral blood mononuclear cells (PBMCs) derived Treg DNA methylation data, additionally supplemented by cells, with the effect being more pronounced in asth­ the results of genome-wide histone modification and matic than in non-asthmatic children [167]. In addition, gene expression analyses, was recently published [186] . maternal exposure to polycyclic aromatic hydrocar­ Whole blood samples obtained from mothers during bons has been associated with DNA hypermethylation gestation and from children and mothers up to several of IFNG promoter in cord white blood cells (from the years after the birth of a child and cord blood (CB) offspring) [168]. Loss of DNA methylation at a single specimens were analyzed. Maternal smoking-related CpG site located in the promoter of the TET1 gene differences in DNA methylation were found to be has been demonstrated in a comparison of affected and enriched in enhancer elements and to persist over years non affected siblings to correlate in airway epithelial of life. Interestingly, differential DNA methylation in cells with asthma. At the same time, increased DNA enhancers was found to be more often functionally rel­ methylation in response to traffic-related air pollu­ evant compared with other genomic regions, based on tion exposure at participants’ current homes was also the combined analysis involving also histone modifica­ observed [169] . The family of TET enzymes is involved tion and RNA expression data [186] . Finally, in a fol­ in the successive oxidation of methylated cytosines to low-up functional study including also animal experi­ 5-hydroxymethylcytosine and further to 5-formyl- and ments, epigenetic deregulation of the enhancer element 5-carboxylcytosine [170,171] and preliminary data in a targeting MAPK9 (gene: MAPK9), previously known limited number of subjects have shown slightly altered as c-Jun N-terminal kinase 2 (JNK2), was linked hydroxymethylation levels in allergic individuals [172] . to impaired lung function in early childhood [186] . Coming back to FOXP3, an increase in DNA meth­ Another study investigated a possible relationship ylation levels in its 5′ region has been shown in saliva between methylation of DNA isolated from cord white samples to be associated with higher DEPs exposure. blood cells, prenatal smoking exposure ascertained by Moreover, children with higher FOXP3 DNA meth­ measuring CB cotinine levels and the development ylation have been demonstrated to be at a higher of AD in offspring (assessed at 2 years of age) [187]. It risk of asthma, persistent wheezing or early transient identified DNA methylation of thymic stromal lym­ wheezing [173] . Another study showed fine particle air phopoietin (TSLP; gene: TSLP) promoter to be associ­ pollutant exposure to correlate with buccal-brushing ated with prenatal smoke exposure and AD. Addition­ DNA methylation of several CpG sites in nitric oxide ally, the degree of TSLP methylation was shown to be synthase genes (NOS1, NOS2 and NOS3) in children; inversely correlated with CB plasma TSLP protein lev­ some current wheezing- or asthma medication-related els [187]. Moreover, the differences in TSLP promoter differences in DNA methylation status of those genes DNA methylation and TSLP mRNA or TSLP protein

future science group www.futuremedicine.com 547 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

levels were observed between inflamed skin lesion Epigenetic modification as mediators of samples obtained from children with AD and normal genetic susceptibility skin tissue sections [188]. Effects of exposure might Considering that the development of allergies is deter­ not need even to be direct, as experiments in rodent mined by gene × environment interactions [33,34], it is models showed that exposure to nicotine altered the not surprising that genetic and epigenetic ­analyses histone modifications profiles in the two subsequent are frequently conducted together. Although the offspring generations correlating with a reduced lung assessment of epigenetic effects was frequently a function [189] . Interestingly, modulatory effects of ciga­ major research target, potential influences of genetic rette smoke on the epithelial innate immune response variability were also taken into account. For exam­ to HRV infection were observed in vitro [190,191], sug­ ple, both DNA methylation and genotyping data gesting the possibility of the interaction­ also on the were used in the studies expanding our knowledge epidemiological level. on the functionality of the 17q21 locus [195,196], the Very recently, epigenetic effects of smoking have most widely replicated (childhood) asthma suscepti­ been subjected to two large meta-analyses [192,193]. The bility locus ever identified by genome-wide associa­ first study focused on the influences of maternal smok­ tion studies (GWAS) [26,125,197]. The association with ing during pregnancy on newborn blood DNA meth­ asthma is stronger in boys compared with girls and ylation. Thirteen epigenome-wide association studies sex-specific DNA methylation changes were found in (EWAS 450K) cohorts with almost 6700 newborns the proximal promoter of the ZPBP2 gene with lower were included [192] . More than 6000 CpGs demon­ methylation levels in boys compared with girls that strated differential methylation in relation to maternal diminish with aging, suggesting that DNA methyla­ smoking at the genome-wide statistical significance tion might mediate the sex- and age-specific associa­ level. Out of those, almost 3000 CpGs belonged to tions [196] . DNA methylation analysis was also used more than 2000 genes not previously related to smok­ in a study investigating the functional background ing. For a number of differentially methylated CpGs, of the association between total serum IgE levels associations with gene expression were seen. Finally, a and polymorphisms in RAD50 [120], a gene encod­ substantial persistence of the epigenetic signatures of ing an important DNA repair molecule [21] . Further prenatal maternal smoking observed in newborns into examples of the genes for which combined genetic– later childhood was observed [192] . Another large meta- epigenetic analysis has been conducted in the con­ analysis comprehensively determined the association text of allergy, specifically asthma and/or related between cigarette smoking and blood-derived DNA phenotypes, include NPSR1 [198], ALOX12 [199] and methylation. Sixteen EWAS (450K) cohorts compris­ ADCYAP1R1 [200]. In addition, a very recent family- ing together over 15,900 individuals (including more based study employing in parallel genome- (linkage than 2400 current, over 6500 former and approxi­ scan) and epigenome-wide approaches identified a mately 7000 never smokers) were subjected to a meta- polymorphism located upstream of the MTRN1A analysis [193] . Comparison between current and never locus to be associated with asthma and AR comor­ smokers identified with genome-wide statistical signifi­ bidity and the effect was mediated by a differentially cance more than 2600 differentially methylated CpGs methylated intronic CpG site of MTRN1A [201] . A annotated to over 1400 genes. Integration of transcrip­ large GWAS in food allergy (milk, egg, peanut) ana­ tome data revealed associations with gene expression lyzing children and their parents showed that the at many differentially methylated CpGs. At a number genetic variants at the major risk loci (HLA-DBR1 of loci, the effects of smoking on blood-derived DNA and HLA-DQB1) correlated with differential DNA levels were demonstrated to persist for many years as methylation and epigenetic changes might there­ evidenced by a comparison between former and never fore mediate the risk of genetic factors to peanut smokers [193] . allergy [202]. Given the ratio of reviews to original Besides all these environmental changes, the simple papers observed in the existing literature on the role presence of allergic diseases is sufficient at least in genome × environment (epigenetic code) interactions rodent models to substantially increase the risk for the in the development of allergic disorders, many fur­ transmission of the allergy independent of the genetic ther studies in the field are to be expected. These background. Offspring of peanut-allergic mothers investigations are facilitated by the development of showed a Th2 dominated response at low levels of sen­ novel bioinformatics tools, such as the Gene, Envi­ sitization with increased IgE and IL-4 levels compared ronment and Methylation software package created with control mice and the inflammatory response cor­ to handle EWAS data, integrate genotype informa­ related with decreased DNA methylation at the IL4 tion and model interactions between genotype and promoter [194] . the environment [203].

548 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review [221] [210] [303] [207] [206] Ref. ­ ­ performed Comments Study shows a tenfold greater capacity of DNA methylation to explain variability in serum IgE levels compared with all so far performed GWAS Only 5 CpGs with diff. methylation >5%; reasonable replication rate in blood, not in CB; potential functional importance as in gene regulatory regions such as CTCF sites No correction for heterogeneity,cellular small sample sizes DNA methylation sig out IgE and skin prick testing for oral food challenge outcome → potential for novel diagnostic assays Longitudinal study with sample at birth and 1 year showing presence of epigenetic alterations predicting later food allergy; measurement of gene expression Yes; 10 loci/2Yes; 10 cohorts: 5 CMA cases/20 controls whole blood, white children and 8 CMA cases/132 controls CB black children Validation cohorts:Two n (extreme= 149 IgE levels) and n asthmatic= 160 families) bisulfite Amplicon sequencing of four genes (n = specified)not Yes; 48 samples from an independent cohortFA (79% accuracy of the methylation signature) No , ,

, , , EVL , LPCAT2 , RPS6K2 , SLC25A33 , STAT4, IL4 , STAT4, , IL5RA 36 loci validated in all three cohorts with high confidence; a number of CpGs with correlation between methylation and IgE levels including IL4 (all> 5%), ILIRL1 IL5RA CCL18 ZNF22 Main findings DMPs575 (568 hypo/7 hyper) including CpGs in or close to NDFIP2 TRAPPC9 Hypermethylation in CMA DNA methylation comprised signature of 96 CpGs predicted clinical outcome 179 CpGs179 associated atwith 1 year, FA CpGs136 at birth, 92 common to both time points T cells + PBMCs and eosinophils/yes Tissue used/Tissue correction for cellular heterogeneity blood/Whole yes blood/Whole no CD4 PBMCs/yes from CBMCs and PBMCs/NA

year) 355 adults355 and children including asthmatic 113 children Sample size of primary screen (cases/controls) 106/76 children 20/23 children and including boys10 that outgrew CMA children12 with FA/ controls;12 2 time points (birth and 1 58 food- sensitized children (50% clinically reactive)/13 controls 27K/450K Technology Technology used 450K 450K 450K 450K Asthma with high and low IgE levels; physician- diagnosed asthma for replication Asthma definition - - - - Serum IgE levels CMA IgE- Various mediated FA (mainly hen’s egg) Multiple FA Allergy- trait related CMA

al. al. al. rhinitis.

al. al. Table 1. Overview 1. Table of epigenome-wide association studies performed in allergic diseases. Liang et Liang 1 summarizesTable a selection of EWAS analyzing at least 500 genes simultaneously in allergic diseases in humans. Studies in animal models are450K: not listed. Infinium Human Methylation450 Bead Array (485,000 CpGs covering all genes); Infinium 27K: HumanMethylation27 CpG covering BeadChip >14,000 (27,578 genes);AEC: Airway 5 hmC: 5-Hydroxymethylcytosine; epithelial cell; CMA: milk Cow’s allergy; CB: Cord blood; CBMC: mononuclear CB cell; DDE: Dichlorodiphenyldichloroethylene;FEV1: Forced Diff.: expiratory Differentially; volume DMP: in one second; Differentially GoldenGate: methylated Illumina position; GoldenGate FA: Food BeadArray allergy; CpG sites (1505 – 807 genes); PBMCs: Peripheralallergic blood mononuclear cells; PUFA: Polyunsaturated fatty acid; SAR: Seasonal Study Hong et Petrus et Martino et Martino et

future science group www.futuremedicine.com 549 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost [199] [166] [209] [304] Ref. 10% of ∼ Comments Methylation in newborns is not associated with season, suggesting a postnatal occurrence; longitudinal data available Identification of diff. methylated CpGs between PBMCs and AECs in each subject category ( CpGs); analyzed expression analysis, but in a different data set Epigenetic variation determined to a large part by genetic polymorphisms Demonstrates the potential of DNA methylation based signatures; but very small sample size, and small differences per CpG site; expression profiles do not separate patients and controls; no single CpG differentiates patients and controls stimulation Validation 207 children (age 8 years) replicates four of the CpGs No Independent cohort of 236 children Technical Technical validation of 3 CpGs in 4 SAR patients and 4 controls; similar affects seen by in vitro of PBMCs associated associated Main findings CpGs92 associated with season of birth, 20 CpGs nominally significant with allergy No diff. methylated CpGs in PBMCs, eight in AECs (atopy vs asthma) including STAT5A Hypomethylation of ALOX12 with persistent wheezing, partly related to DDE exposure DNA methylation separate signatures patients and controls even outside the pollen (357 genes diff. methylated, including mainly CpGs in gene bodies of HSPD1, TGFBR2, (TNFRSF4,CD134 OX40), CD45 (PTPRC), SPN (CD43), IL2RA, IL13RA1) T cells/NA + CD4 Whole blood/Whole yes PBMCs and AECs/No Whole blood/partly (eosinophil counts) Tissue used/Tissue correction for cellular heterogeneity 8 cases/8 controls; sampled in and outside the season pollen 367 individuals individuals 367 (age 18) children 25 children 122 Sample size of primary screen (cases/controls) 450K 450K GoldenGate GoldenGate Technology Technology used - - IgE skin prick test, but combined analysis of asthmatics (independent of atopy status) - Asthma definition SAR Impact of season of birth on allergy development Atopy with/ or without asthma Wheezing before age of 6 versus controls Allergy- trait related al. rhinitis.

al. al. al. Table 1. Overview 1. Table of epigenome-wide association studies performed in allergic diseases (cont.). Nestor et Study Lockett et Stefanowicz et Morales et 1 summarizesTable a selection of EWAS analyzing at least 500 genes simultaneously in allergic diseases in humans. Studies in animal models are450K: not listed. Infinium Human Methylation450 Bead Array (485,000 CpGs covering all genes); Infinium 27K: HumanMethylation27 CpG covering BeadChip >14,000 (27,578 genes);AEC: Airway 5 hmC: 5-Hydroxymethylcytosine; epithelial cell; CMA: milk Cow’s allergy; CB: Cord blood; CBMC: mononuclear CB cell; DDE: Dichlorodiphenyldichloroethylene;FEV1: Forced Diff.: expiratory Differentially; volume DMP: in one second; Differentially GoldenGate: methylated Illumina position; GoldenGate FA: Food BeadArray allergy; CpG sites (1505 – 807 genes); PBMCs: Peripheralallergic blood mononuclear cells; PUFA: Polyunsaturated fatty acid; SAR: Seasonal

550 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review Ref. [183] [305] [306] [307] [192] [308] Analyses show some loci to be replicated in different tissues; promoter-focused design limits the regions that can be analyzed Atopy to dust mite Comments assessed Smoking maternal through cotinine levels Longitudinal design allows to identify CpGs that reverse methylation and where methylation persists Large-scale EWAS identifying a huge number of new loci associated with maternal smoking supporting the need for large-scale analyses Smoking Smoking through assessed questionnaire; some overlap with previous EWAS validation [305] 2 CpGs replicated independent in cohorts (CB and adult whole blood) No Validation 21/26 CpGs validated in a small cohort (n = 36) In silico of ten gene regions using data from No Replication in older children (n = 3187) 19 CpGs19 altered in function of prenatal smoke tobacco exposure Differential methylation in 52 genes in atopic asthma versus asthma,most of them hypomethylated Main findings CpGs26 genes in 10 altered in function of prenatal tobacco exposure smoke CpGs185 in 100 gene regions CpGs15 sites associated maternal with smoking; dose – response relationship in function of smoking intensity and duration for some CpGs >6000 CpGs differentially methylated,including 2000 new ones Whole blood/Whole yes Bronchial mucosal tissue/ precisednot CB/yes Whole blood shortly after delivery/yes CB/yes Various/yes Tissue used/Tissue correction for cellular heterogeneity 527 children (age years) 5–12 Atopic asthma (n = 7) versus nonatopic asthma (n = 7) versus controls (n = 10) 1062 children 1062 889 children 800 children: three time points per child (birth, age 7 years)and 17 6685 children cohorts) (13 Sample size of primary screen (cases/controls) 27K 27K 450K 450K 450K 450K Technology Technology used ­ Symptoms of asthma or use of asthma medication Different iation of IgE- associated and non IgE-mediated asthma - - - - Asthma definition Asthmatics/ prenatal tobacco smoke exposure Atopic asthma Prenatal tobacco smoke exposure Prenatal tobacco smoke exposure Prenatal tobacco smoke exposure Prenatal tobacco smoke exposure Allergy- trait related al. al. rhinitis.

al. al. al. al. Table 1. Overview 1. Table of epigenome-wide association studies performed in allergic diseases (cont.). Breton et 1 summarizesTable a selection of EWAS analyzing at least 500 genes simultaneously in allergic diseases in humans. Studies in animal models are450K: not listed. Infinium Human Methylation450 Bead Array (485,000 CpGs covering all genes); Infinium 27K: HumanMethylation27 CpG covering BeadChip >14,000 (27,578 genes);AEC: Airway 5 hmC: 5-Hydroxymethylcytosine; epithelial cell; CMA: milk Cow’s allergy; CB: Cord blood; CBMC: mononuclear CB cell; DDE: Dichlorodiphenyldichloroethylene;FEV1: Forced Diff.: expiratory Differentially; volume DMP: in one second; Differentially GoldenGate: methylated Illumina position; GoldenGate FA: Food BeadArray allergy; CpG sites (1505 – 807 genes); PBMCs: Peripheralallergic blood mononuclear cells; PUFA: Polyunsaturated fatty acid; SAR: Seasonal Study Joubert et Markunas et Richmond et Joubert et Kim et

future science group www.futuremedicine.com 551 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost Ref. [278] [309] [169] [310] hypomethylation hypomethylation expression and Very large number of diff. methylated CpGs might point to unknown confounders (e.g., blood cell composition) Comments In addition, methylation changes in repetitive elements were observed; smallhowever, sample size, no validation, replication, no inclusion of healthy controls TET1 correlated with higher TET1 higher global 5 hmC levels; increased methylation in function of traffic pollution; EWAS with robust multitissue validation Exposure to filtered air or diesel exhaust followed by for bronchoscopy delivery; allergen CpGs differ in function of the order of allergen/diesel exposure; small cohort size No Validation No 35 additional siblings pairs with saliva, PBMCs, lung bronchial epithelial cells available; independent cohort n = 186 1 CpG by pyrosequencing in the same samples promoter promoter No methylation differences for asthma; 10K CpG sites diff. methylated between highly polluted and control regions Main findings 2827 CpG sites, mainly hypomethylated, enriched in protein kinase and NF-kB signaling Demethylation at a single CpG site in the TET1 associated with asthma Altered methylation at 7 CpG sites 48 hours after exposure; if allergen and diesel exhaust were separated by 1 month, 500 diff. methylated (mainly hypomethylated) CpGs were identified suggesting a priming mechanism Whole blood/Whole no Tissue used/Tissue correction for cellular heterogeneity PBMCs/yes Nasal epithelial cells/no Bronchial epithelial cells/ precisednot 200 children years)(age 7–15 Sample size of primary screen (cases/controls) asthmatics16 randomized to filtered air or diesel exhaust 12/12 African– American siblings adult 17 patients (47% asthmatics) 27K Technology Technology used 450K 450K 450K Not precised Not Asthma definition Physician- diagnosed asthma Physician- diagnosed IgE-related diseases including asthma Asthma/ pollution Allergy- trait related Asthma/ diesel exhaust exposure Asthma/ pollution Allergen/ diesel exhaust exposure al. rhinitis.

al. al. al. Table 1. Overview 1. Table of epigenome-wide association studies performed in allergic diseases (cont.). Rossnerova et 1 summarizesTable a selection of EWAS analyzing at least 500 genes simultaneously in allergic diseases in humans. Studies in animal models are450K: not listed. Infinium Human Methylation450 Bead Array (485,000 CpGs covering all genes); Infinium 27K: HumanMethylation27 CpG covering BeadChip >14,000 (27,578 genes);AEC: Airway 5 hmC: 5-Hydroxymethylcytosine; epithelial cell; CMA: milk Cow’s allergy; CB: Cord blood; CBMC: mononuclear CB cell; DDE: Dichlorodiphenyldichloroethylene;FEV1: Forced Diff.: expiratory Differentially; volume DMP: in one second; Differentially GoldenGate: methylated Illumina position; GoldenGate FA: Food BeadArray allergy; CpG sites (1505 – 807 genes); PBMCs: Peripheralallergic blood mononuclear cells; PUFA: Polyunsaturated fatty acid; SAR: Seasonal Study Jiang et Somineni et Clifford et

552 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review Ref. [279] [311] [257] [208] levels; 1 Few, but some DMRs related to IgE levels percentage and predicted FEV Comments T cell signaling and macrophage activation pathways selectively hypomethylated in obese asthmatics Demonstrates the need for tissue- specific DNA methylation analyses; complementary analyses expression which show some overlap predominantly in genes implicated epidermal in differentiation and innate immune response integration of gene expression and DNA methylation data Some CpGs showed nonetheless a correlation to PUFA levels No Validation No No No , IL13 ) , TIGIT RUNX3 81 DMRs81 between asthmatics and controls with hypomethylation of several immune related genes ( Main findings Diff. methylated CpGs in immune related genes including CCL5, IL2RA, TBX21, FCER2, TGFB1 A large number of methylation between differences patients and controls are only found in the skin No statistically significant CpGs at the genome-wide level in relation to treatment or PUFA levels T cells + PBMCs/yes Tissue used/Tissue correction for cellular heterogeneity PBMCs/no CD4 Whole, blood, T cells, B cells, epidermis skin from CBMCs ­ 97 patients97 versus 97 controls Sample size of primary screen (cases/controls) 32 preadolescent children (8 per group asthmatics and controls with obesity not) or 70 mother– infant pairs; fish oil suppl ementation (n = 36), control group (n = 34) 28 patients with atopic dermatitis, 29 controls 450K Technology Technology used 1.8–2 M CpG sites (sequencing based) 450K 27K IgE-related asthma Asthma definition precised Not - - ­ Asthma Allergy- trait related Asthma and obesity Allergy protective effects of fish oil suppl ementation during pregnancy Atopic dermatitis al. al. ­ rhinitis.

al. al. Table 1. Overview 1. Table of epigenome-wide association studies performed in allergic diseases (cont.). Yang et 1 summarizesTable a selection of EWAS analyzing at least 500 genes simultaneously in allergic diseases in humans. Studies in animal models are450K: not listed. Infinium Human Methylation450 Bead Array (485,000 CpGs covering all genes); Infinium 27K: HumanMethylation27 CpG covering BeadChip >14,000 (27,578 genes);AEC: Airway 5 hmC: 5-Hydroxymethylcytosine; epithelial cell; CMA: milk Cow’s allergy; CB: Cord blood; CBMC: mononuclear CB cell; DDE: Dichlorodiphenyldichloroethylene;FEV1: Forced Diff.: expiratory Differentially; volume DMP: in one second; Differentially GoldenGate: methylated Illumina position; GoldenGate FA: Food BeadArray allergy; CpG sites (1505 – 807 genes); PBMCs: Peripheralallergic blood mononuclear cells; PUFA: Polyunsaturated fatty acid; SAR: Seasonal Study Rastogi et Amara sekara et Rodriguez et

future science group www.futuremedicine.com 553 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

Understanding allergic diseases through for example observed in a study on AD, where only EWAS skin samples showed differential methylation between Despite numerous large-scale studies, genetic poly­ patients and controls, but neither purified blood cell morphisms confer only a low to moderate level of pre­ populations nor PBMCs [208]. Similarly, although air­ disposition, which cannot explain the recent rise in way epithelial cells and PBMCs shared most of their prevalence of IgE-mediated allergic syndromes. Most methylation patterns, asthma specific methylation studies have so far focused on the analysis of one (or differences were only detectable in the lung epithelial several) candidate gene(s). Technological advances in cells [209]. While in a recent large and well-controlled sequencing and microarray technology do now allow EWAS no DNA methylation changes were found to be for the genome-wide analysis of DNA methylation pat­ associated with asthma, this study found 36 loci (34 terns with specific phenotypes and a number of large genes) at which DNA methylation levels associated cohorts have been profiled for their DNA methyla­ with serum IgE levels and most of these genes were tion patterns [204]. Although not without a number of particularly important for gene regulation in eosino­ potential pitfalls requiring a careful design and analy­ phils [210]. This study also demonstrated the power of sis of the data [205], the analyses have identified some well-conducted EWAS with a tenfold greater capac­ CpGs associated with disease phenotype and early-life ity of the genome-wide DNA methylation patterns to exposure in different diseases ranging from metabolic explain the observed variability in IgE concentrations to psychiatric diseases, from cancer to cardiovascular compared with genetic variation. complications as well as immune-related disorders. While there is definitely room for improvement in Few EWAS have so far been performed in allergic dis­ sample size, replication efforts, but also phenotypic eases with the exception of asthma and most have been characterization of the analyzed samples, the avail­ analyzing a limited number of individuals and find­ able studies provide first encouraging steps toward ings have often not been replicated (Table 1). To draw deciphering the role of epigenetics in allergic diseases general conclusions on the studies involving asthmatic and have clearly demonstrated the power of integrating subjects is further complicated by the fact that studies epigenetic analyses in the endeavor for understanding report rarely the specific sub- or endotypes and rely on allergic diseases. a phenotypic diagnosis (Table 1), thereby introducing New generations of DNA methylation microarrays probably some molecular heterogeneity in the studied interrogating an ever increasing number of CpG sites population, which renders the detection of asthma- such as the EPIC BeadChip analyzing DNA meth­ type specific changes very difficult even at the broad ylation levels at more than 850,000 CpGs and which level of atopic versus nonatopic asthma. increasingly focus on gene-regulatory elements outside EWAS in allergic diseases have notably shown genes and promoters have recently been devised [211,212], that DNA methylation signatures can separate aller­ together with the decreasing cost of whole-genome gic patients from normal controls and show increased bisulfite sequencing [213], will allow a more comprehen­ discriminatory power compared with gene expression sive assessment of the DNA methylation landscape and based signatures [166] . However, due to small differ­ provide novel insights in the development and course ences and significant interindividual variability of of allergic diseases. DNA methylation patterns, no single CpG showed sufficient discriminatory power to distinguish between Clinical applications of epigenetics the two groups. Signatures of differentially methylated in allergies: current status & future positions are in some cases already present at birth and perspectives predict future disease onset of – in this case – food Epigenetic modifications as biomarkers for allergy [206]. EWAS have also shown that allergy- allergy or response to treatment related autosomal DNA methylation changes differ Current molecular biomarkers used in clinical practice substantially between boys and girls when compared for the management of allergic patients are mainly based with sex-matched controls [207]. Although the results on the measurement of circulating antigen-specific IgE were obtained in a cohort of limited size, they provide levels as well as basophil activation tests, but lack suf­ further evidence that DNA methylation alterations ficient sensitivity to predict onset of atopic diseases or might underlie (or at least correlate with) the observed early efficacy during therapeutic interventions [214]. age- and sex-specific differences in the prevalence of While detection and quantification of disease-specific allergic diseases [11–13] . As also discussed in more detail cell subsets using multilevel fluorescent activated cell below, the optimal tissue is yet an unsolved issue and sorting (FACS) or mass spectrometry (MS) do provide the choice of the tissue will have a major impact on alternatives, it might be difficult to implement them in the identified differentially methylated positions, as a routine clinical setting. Epigenetic modifications and

554 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review in particular DNA methylation represent a ‘molecular ers such as the measurement of allergen-specific IgE memory’ and might mediate the above described gene levels or skin-prick tests and have thus the potential × environment interactions. DNA methylation signa­ to replace the food challenge, a procedure potentially tures have in other complex diseases such as cancer and associated with the risk of serious complications such autoimmune disorders shown their value as biomarkers as ­anaphylaxis [221] . of diagnosis and/or therapeutic response [215]. As a first Few studies have so far addressed the potential of application, DNA methylation changes might thus be epigenetic biomarkers as a surrogate for the efficacy of used as molecular biomarkers to quantify the different the treatment of allergic diseases, with the exception allergy enhancing or protective exposures. of methylation of FOXP3. FOXP3, expressed only by Altered methylation patterns at genes implicated in Treg cells, plays a crucial role in allergic diseases, with the balance of Th1/Th2 populations such as IL4, IL5, smaller numbers of Treg cells in allergic children. Sev­ IL10 and IFNG have frequently been observed with eral presumably allergy-protective stimuli such as con­ generally higher methylation levels at Th1 key genes sumption of raw milk [153] have been shown to increase and lower methylation levels at Th2 genes in PBMCs the number of Treg cells correlating well with the or purified CD4+ T cells after exposure to potentially demethylation of the FOXP3 TSDR. Demethylation allergenic substances such as DEPs [216] or dust mite of the TSDR was observed in children with reduced allergens [217]. Changes in the DNA methylation pat­ atopic sensitization and asthma [153], as well as in chil­ terns have also been associated with acquisition of tol­ dren outgrowing IgE-mediated CMA after dietary erance or better-termed sustained unresponsiveness in intervention [222], children receiving oral immuno­ children with CMA [218] and with in utero exposition to therapy (OIT) to peanut allergy [223], and patients the allergy-protective farm environment [219]. Epicuta­ receiving sublingual immunotherapy to timothy grass neous immunotherapy against milk proteins followed and house dust mite [224]. Of note, in children who by adoptive Treg cells transfer prevented sensitization remained unresponsive 12 weeks after the end of OIT to a new allergen (peanuts or house dust mite) through to peanuts the FOXP3 methylation level remained low, a mechanism including increased methylation of the while in those who showed loss of tolerance the FOXP3 promoter of the key TF GATA3 in splenic cells of the methylation increased to baseline raising the possibil­ recipient mice and persisted until at least 2 months after ity to potentially use methylation levels of the FOXP3 the end of the sensitization [220]. It remains nonetheless TSDR as a biomarker for sustained unresponsive­ to be shown if the observed effects are due to changes ness [223]. Similarly, the above-described DNA meth­ in the cell composition of the analyzed blood or tissue ylation changes correlating with serum IgE concentra­ samples or due to altered epigenetic profiles of a spe­ tion [210] might ultimately be useful to select patients cific cell-type as, for example, IFNG DNA methylation that are most likely to profit from treatment with anti- levels have been shown to vary between cell types. The IgE antibodies such as omalizumab, an expensive med­ increasing number of EWAS will yield additional and ication that is currently only applicable to and effective perhaps more specific and sensitive diagnostic mark­ in a subset of patients [225,226]. ers for allergic diseases, which will allow to identify It remains unknown for the moment if these and allergic subjects at an early stage, but also allergy-prone other observed DNA methylation changes are sim­ individuals prior to the onset of disease. Examples ply reflecting the cellular defects or if the epigenetic include specific DNA methylation profiles in CD4+ changes precede the disease symptoms and set the T cells of patients with SAR both in but also outside stage for disease development. Furthermore, signatures the pollen season compared with healthy controls [166] . composed on a number of epigenetic markers and/or As mentioned above, the simultaneously performed combined with genetic or biochemical markers will be genome-wide expression analyses did not allow to dif­ necessary to improve the sensitivity and specificity of ferentiate between the two groups showing the supe­ clinically useful biomarkers. rior potential of DNA methylation as a diagnostic bio­ Recently, a number of studies have shown the poten­ marker. In addition, differential DNA methylation in tial of the analysis of DNA methylation patterns in cir­ genes of the MAPK pathway were detected in children culating cell-free (ccf) DNA (ccfDNA), the so-called developing food allergies later on [206], which provides liquid biopsy, for the detection and monitoring of com­ new opportunities for prevention or early therapeutic plex diseases other than cancer including autoimmune intervention if signatures are sufficiently validated. or metabolic diseases [227–231] . In all examples, the tis­ Furthermore, diagnostic DNA methylation signatures sue of origin had a high rate of apoptosis facilitating the composed of 96 CpGs have been devised predict­ detection of the DNA molecules from relevant cells. ing oral food challenge outcome with an accuracy of Due to the tissue-specificity of DNA methylation pat­ 79%, which outperformed conventional biomark­ terns, analysis of ccfDNA allows to identify the tissue

future science group www.futuremedicine.com 555 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

or even cell type of origin [232,233]. Blood cell type-spe­ occurrence of allergic conditions. More importantly, cific DNA methylation markers are also used to decon­ as described in this review, the altered T cell polariza­ volute the confounding effects of cellular composition tion is largely driven by epigenetic modifications at key in EWAS [234] or can be used to accurately quantify TFs, and cytokines as well as CNS. Epigenetic editing the number of immune cell types and subtypes [235]. provides potentially the tools to, for example, alter the Potential applications in allergic diseases are numer­ balance between different T cell populations by mak­ ous including monitoring changes in T cell balance ing use of two vectors where the expression of the key or eosinophilic infiltration in less accessible tissues, components of the editing complex is driven by two such as lung or epithelial tissue in the gastrointestinal cell-type-specific promoters to achieve expression of tract, or accurately and rapidly quantify low abundant the editing complex for example only in T cells [244]. ­cellular subtypes. Epidemiological evidence from asthma/allergy pro­ tective environments points to the prenatal and the Treatments influencing directly or indirectly early postnatal periods as critical windows for estab­ epigenetic marks in allergy lishing an allergy susceptible or protective phenotype. DNA methylation inhibitors such as 5-azacytidine This time window coincides with the establishment of have been little investigated in allergic diseases and the gut microbiota, maturation of the immune system data are currently conflicting with some studies find­ and epicutaneous allergen sensitization, which are all ing beneficial actions through notably an induction of important factors for the development of future aller­ Treg cells [236] or demethylation of IFNG [237], while in gies. The hope is that intervention in these processes other studies the use led to a worsening of the allergic might be sufficient to avoid or delay allergic reactions conditions [188]. As most studies investigated only the in many cases. effect on a single target gene, it is difficult to draw any Treatment with probiotics such as the farm-derived conclusions on the potential use of these genome-wide Acinetobacter lwoffii F78 during pregnancy has been epigenetic modifiers, but caution is warranted. Simi­ shown to confer protective effects in a mouse model larly the use of HDAC inhibitors has yielded conflicting of ovalbumin-induced sensitization through epigen­ result, with some studies showing anti-inflammatory etic modulation of Th1/Th2 balance genes (see also effects and others pointing to enhanced inflammation earlier) [139] and addition of Lactobacillus rhamnosus in requiring thus further investigation of the use of this a small nutritional intervention study including chil­ treatment [73,238]. Little data are currently available on dren with CMA, facilitated outgrow of the allergy as treatments targeting other histone modifications such evidenced again through epigenetic modification at as histone methylation, but overall the use of epigenetic Th1/Th2 regulatory regions [222]. Overall, modifica­ drugs altering the genome-wide epigenetic landscape tion of the gut microbiota and increasing the abun­ might be associated in allergic diseases with a number dance of bacteria such as Clostridia or Roseburia, of drawbacks. which produce epigenetically active short chain fatty Recently, major technological advances using the acids including butyrate, seems to be beneficial for CRISPR/dCas9 system allow the targeted engineer­ decreasing inflammation and re-establishing barrier ing of the epigenome [239–241]. These approaches allow function notably through induction of Treg cells and/ for the first time to functionally investigate and vali­ or activation of ILCs and IL-22 production [245–248]. date the importance of epigenetic modifications at any However, evidence is still controversial with several locus in the genome and might provide novel alterna­ large-scale studies do not providing solid evidence tives to modulate the epigenome in allergic diseases. for the prevention of allergies through postnatal The combination of CRISPR/dCas9, which does not administration of probiotics [249,250]. A few studies introduce double strand breaks in the genome through have shown that already at birth epigenetic changes the use of a nuclease-deficient Cas9 enzyme, allows to associated with an asthma-protective status or later guide epigenetic enzymes to specific loci in the genome development of allergies were present [206,219] suggest­ where they can specifically (de)methylate DNA or (de) ing a very early altered (epigenetic) programming of methylate/(de)acetylate histones [242,243]. While only future immune responses and raising the question on recently devised, this technology bears in the future the utility of in utero interventions. Supplementation great promise for the treatment of diseases without with methyl donors has been shown in mouse models clearly-defined underlying mutations. As epigenetic to actually exacerbate asthmatic conditions in the off­ modifications are in general reversible, the identifica­ spring [251] . However, dietary intervention programs tion of key molecular changes induced by environmen­ during pregnancy with polyunsaturated fatty acids tal exposure might provide new treatment alternatives present, for example, in high concentrations in fish in which these changes might be reversed prior to the oil have been shown to reduce the risk of developing

556 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

allergies of the offspring [252] through modulation of year and maybe lifelong application periods it will the neonatal immune response [253,254], which is at be essential to define molecular signatures that will least partly mediated by changes in the DNA meth­ predict the success of an immunotherapy as early as ylation patterns [255]. In addition, the above-described possible enabling a switch to alternative therapeutic allergy-protective effects of unprocessed farm milk are approaches (if available) without further delay. Simi­ at least partially attributable to its increased content larly, when a state of unresponsiveness is achieved, of polyunsaturated fatty acids [154] . While first candi­ signatures based on or including epigenetic markers date studies analyzing the DNA methylation patters might help to distinguish between patients that will of genes in T-lineage differentiation in CB of neonates show sustained unresponsiveness or even tolerance were only showing evidence for altered DNA methyla­ without requiring further therapy and those who tion patterns in children with mothers continuing to will be in need of continued administration of ther­ smoke during pregnancy [256], a recent genome-wide apy. Preliminary evidence in peanut allergic subjects study in CB CD4+ T cells identified some CpGs show­ shows that that about 30% of the initial patient cohort ing a dose-response effect in function of the polyun­ remained unresponsive after a 4-week interruption fol­ saturated fatty acid levels as well as differentially meth­ lowing 5 years of OIT [266] and response rate might be ylated CpGs [257]. However, the observed changes were increased through concomitant administration of the small raising questions on the biological significance above-described probiotics [267] suggesting the contri­ and did not reach statistical significance after correc­ bution of epigenetic mechanisms in achieving the state tion for multiple testing requiring thus confirmatory of unresponsiveness. As described above, methylation analyses in larger cohorts or other cell types. Other of FOXP3 has been shown to be lower in patients possible protective treatments could include the inges­ with increased unresponsiveness including those with tion of small amounts of an allergen during pregnancy peanut allergy at both the end of an OIT administra­ and lactation. This strategy has been applied to pea­ tion and 12 weeks after the completion of the treat­ nut sensitivity and shown in a mouse model to reduce ment [223] as well as in patients receiving dual sublin­ severe allergic reactions in the offspring when exposed gual immunotherapy against timothy grass and house to peanuts [258] and is supported by epidemiological evi­ dust mites [224]. dence [259]. There are, however, no data on ­epigenetic Another option to modulate the epigenome toward modifications­following this approach available. a lower risk of atopic diseases or at least toward the development of less pronounced allergic pheno­ Strategies modifying the epigenome in allergy types would be a direct influence on the differentia­ Although no US FDA approved specific therapy for tion of the cells playing the major regulatory in the allergy exists and the long-term benefits of different immune system, in other words, T cells, for example immunotherapies require further investigation, pre­ by reducing the expression of the TFs promoting liminary data suggest that immunotherapy allows Th2 responses such as GATA3/GATA3 (see also desensitizing patients to an allergen and leads to at least earlier). This targeted knock-out or at least knock- some degree of sustained unresponsiveness [260–263]. down could possibly be achieved using antisense Immunotherapy might provide life-changing alterna­ strategies. In principle, through specific binding and tives to subjects suffering from food allergy, where the subsequent degradation of mRNAs, antisense mol­ standard care is strict food avoidance and accidental ecules prevent translation of those mRNAs to pro­ ingestion leads to a high number of emergency room teins contributing to the pathogenesis of a certain visits [260,261]. Immunotherapies including mainly oral disease [81] . Such pathogenic proteins could be either (OIT), sublingual (SLIT), subcutaneous or epicutane­ normal (wild-type) translation products that are only ous (EPIT) administration of the allergen are thought too abundantly expressed (e.g., GATA3 in atopic to correct the shift from a largely Th2- to a more Th1- asthma, α4-integrin in relapsing-remitting multiple dominated response, as well as inducing immunosup­ sclerosis or coagulation factor XI in prevention of pressive Treg cells, and these changes in the polariza­ venous thromboembolism) [268–270] or they can be tion of T cells are considered critical for the efficacy synthesized based on the incorrect (mutant) genetic of immunotherapy [264,265]. Since epigenetic modifica­ background (e.g., superoxide dismutase-1 in familial tions are critically involved in the T cell differentiation amyotrophic lateral sclerosis or type II keratin-6A, in as outlined above, these therapies will necessarily act pachyonychia congenita) [271,272] or otherwise inap­ on the epigenome. propriately spliced mRNA (e.g., acetylcholinesterase There are currently no epigenetic biomarkers and read-through ­transcript in ­myasthenia gravis) [273]. few biomarkers at all predicting the success of the dif­ The possibility that antisense approaches could be ferent immunotherapies. As immunotherapies require used to edit the genome of developing T cells, is sup­

future science group www.futuremedicine.com 557 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

Abbreviations ported by recent studies on hgd40/SB010, a DNA­ A. lwoffii Acinetobacter lwoffii zyme-type antisense molecule against the above-men­ tioned GATA3. DNAzymes are antisense molecules AD Atopic dermatitis possessing inbuilt, internal catalytic activity, making AECs Airway epithelial cells it possible for them to deactivate targeted mRNAs AHR Airway hyper-responsiveness upon specific binding without a need of accessory APCs Antigen-presenting cells molecules possessing enzymatic activity, which are AR Allergic rhinitis additionally characterized by good stability and low toxicity [81] . First proofs for preventive and therapeu­ CB Cord blood tic effectiveness of DNAzymes targeting GATA3 were CBMCs CB mononuclear cells obtained in mouse models of allergic airway inflam­ ccfDNA Circulating cell-free DNA mation mimicking human atopic asthma [268]. After CMA Cow’s milk allergy successful preclinical testing [274], hgd40/SB010 CNS Conserved non-coding sequence (e.g., CNS2) entered Phase I clinical trials which demonstrated its safety and tolerability in humans [275]. Finally, a ran­ CpG (site) DNA sequence where a cytosine nucleotide (C) is domized, double-blinded, placebo-controlled, mul­ directly followed by a guanine nucleotide (G) ticenter Phase II clinical trial showed hgd40/SB010 DDE Dichlorodiphenyldichloroethylene to significantly attenuate late and early clinical asth­ DEPs Diesel exhaust particles matic responses after allergen provocation in patients Diff. Differentially with atopic asthma. In addition, the analysis of bio­ DMPs Diff. methylated positions markers showed also a weakening of Th2-regulated inflammatory responses in patients receiving hgd40/ DNMTs DNA methyltransferases (e.g. DNMT1, DNMT3A) SB010 [276]. EPIT Epicutaneous immunotherapy EWAS Epigenome-wide association study Unsolved issues: are easily accessible cells FA Food allergy suitable for the study of allergic diseases? FACS Fluorescent activated cell sorting Although there has been increasing interest in the FcεRI High-affinity IgE receptor analysis of epigenetic modifications in allergy, a few issues have been raised among which are interroga­ FcεRII (CD23) Low-affinity IgE receptor tions of the biological significance of the commonly FOXP3 Forkhead box protein 3 observed relatively small changes in DNA methyla­ GATA3 GATA binding protein 3 tion patterns as well as the relevance of the biologi­ GEM (software) Gene, Environment and Methylation (software) cal material used for epigenetic analyses, especially GoldenGate Illumina GoldenGate BeadArray (1,505 CpG sites - as cell lines display highly divergent DNA methyla­ 807 genes) tion patterns compared with the tissue or cell-type it was derived from [277]. Epigenetic modifications are GWAS Genome-wide association study in human studies analyzed in multiple types of bio­ H3/H4ac Histone H3 or H4 acetylation (e.g. H3K9ac, materials, among which blood is probably most com­ H3K18ac) mon. Whole peripheral blood [183,199,278] or its cellular H3K(X)me(x) Histone H3K (X = e.g. K4, K27) di- (x = 2) or tri- (x isolates, such as PBMCs [169,196,222,279] or subpopula­ = 3) methylation (e.g. H3K4me2, H3K27me3) tions of cells further sorted from PBMCs [167,280] are HATs Histone acetyltransferases frequently used. Comparative analyses of the results HDACs Histone deacetylases (e.g. HDAC9) of DNA methylation studies obtained from differ­ HDACi HDAC inhibitors ent types of blood specimens (sorted cells, PBMCs or whole blood) should be conducted with great caution. HMTs Histone methyltransferases In addition, interpreting the epigenetic data obtained HRVs Human rhinoviruses from whole blood or PBMCs may sometimes be diffi­ IFN-γ Interferon-γ cult, as these samples comprise varying proportions of IgE Immunoglobulin E diverse cell populations with different biological char­ IL Interleukin (e.g., IL-4, IL-5) acteristics. The comparative analysis of DNA meth­ ylation profiles from whole blood, PBMCs, granulo­ ILCs Innate lymphoid cells (e.g., ILC2) cytes and seven selected purified cell lineages collected IRF4 Interferon regulatory factor 4 from healthy male donors showed clear differences in JNK2 c-Jun N-terminal kinase 2 DNA methylation profiles between myeloid cells and

558 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

lympho­cytes as well as the distinct methylation pat­ Abbreviations tern of B cells [281] . Comparison between PMBCs and LCR Locus control region granulocytes revealed that at least one probe was dif­ MAPK9 Mitogen-activated protein kinase 9 ferentially methylated for 85% of the analyzed genes. On the individual gene level, in spite of apparently MEDALL Mechanisms of the Development of Allergy similar general methylation profiles present in main (consortium) (consortium) types of the blood cells, patterns observed for some MHC-II Major histocompatibility complex class II individual CpGs might be opposite (hypo- vs hyper­ miRNA MicroRNA methylation) [281] . To avoid potential confounding by NOS Nitric oxide synthase (e.g. NOS1, NOS2) cellular heterogeneity in the analyzed samples in stud­ OIT Oral immunotherapy ies which are not performed using sorted blood cell populations, it is important to estimate blood cell pop­ PACE Pregnancy And Childhood Epigenetics (consortium) (consortium) ulations in the analyzed samples [282]. If no cell count data are available, a number of bioinformatic tools have PAHs Polycyclic aromatic hydrocarbons been developed that can estimate and correct for cell PBMCs Peripheral blood mononuclear cells type heterogeneity in the analyzed samples based on PUFAs Polyunsaturated fatty acids either reference datasets or reference-free statistical RISC RNA-induced silencing complex algorithms [283–288]. While reference based approaches RORC2 RAR related orphan receptor C isoform 2 are commonly used for blood and CB-based studies, (traditionally called also RORγT) reference-free methods can be applied to any biological material for which no reference datasets are presently RSV Respiratory syncytial virus available including saliva (see also later, [289]) and stud­ SAR Seasonal allergic rhinitis ies in young children, for which current reference data SNPs Single-nucleotide polymorphisms sets are not adequate [290]. The choice of the reference SLIT Sublingual immunotherapy dataset can have a major impact on the detection of TBX21 T-box 21 (traditionally called also T-bet) differentially methylated positions and therefore only an appropriate reference dataset should be used for TCR T-cell receptor deconvolution [289]. Otherwise reference-free methods TET1 Tet methylcytosine dioxygenase 1 should be used. TFs Transcription factors Other types of relatively easily accessible biomateri­ TGF-β Transforming growth factor-β als include saliva [169,173,291] or buccal brushings [173], Th (cells) T-helper (cells; e.g. Th2, Th9) conventional sources of DNA for genetic testing that have also been used in studies on DNA meth­ Treg cells Regulatory T-cells ylation. A study comparatively analyzing epigenome- TSDR Treg-specific demethylated region wide DNA methylation profiles in saliva and PBMC TSLP Thymic stromal lymphopoietin samples from the same subjects (either patients with 27K Infinium HumanMethylation27 BeadChip (27,578 respiratory allergies or healthy controls) showed that, CpG covering >14,000 genes) indeed, saliva can be successfully used for this type of 450K Infinium HumanMethylation450 BeadArray analysis [289,292]. Although samples such as blood or (485,000 CpGs covering all genes) saliva do not represent local tissues directly affected 5hmC 5-hydroxymethylcytosine by allergic inflammation such as lungs, gut or skin, epigenetic analyses conducted in them, especially in sues directly affected by allergic inflammation, such peripheral blood cells, are very informative since a as skin [188,296], nasal epithelium [169,297] or bronchial/ huge part of allergic disease-related pathophysiologi­ lung epithelium [169,209,298], seems very important to cal processes take place not (only) in the end-organs verify if the results obtained in them correspond to but also systemically, in circulation, and their pres­ those deriving from the analyses of easily accessible ence there frequently precedes the establishment tissues such as blood. Taken together, all these might of a clinical diagnosis [37,159,293–295]. Hence, being be of huge importance not only for research but also easily accessible, peripheral blood samples can pos­ for diagnostic purposes, especially in case of bron­ sibly provide the results and might possibly be suf­ chi/lungs or the gastrointestinal tract, locations not ficiently good proxies of epigenetic changes in end- as easily accessible as skin or nose (see also earlier). organs, although as detailed above some results not Although as mentioned above it is at the moment really supporting this statement have also been pub­ difficult to assess the (potential) usefulness of liquid lished [208,209]. Thus, epigenetic testing of local tis­ biopsy in allergic diseases, one might speculate that

future science group www.futuremedicine.com 559 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

methylation analysis of these DNA samples could pro­ ment efficacy for allergic diseases. Following exten­ vide valuable information on the epigenetic status of sive validation and refinement, these signatures might end-organ tissues directly affected by allergic inflam­ in the future complement, and perhaps even replace, mation [227,299,300]. current diagnostic tests and might predict the success Finally, considering that epigenetic modifications of different treatment protocols using various forms in the prenatal period seem to play a very important, of immunotherapy that are currently developed for if not crucial, role in the development of allergic dis­ the treatment of allergic diseases. Finally, new tools orders (see also earlier) [43,150,151], CB [120,199,219], CB allow now in a cell-type-specific manner to interfere mononuclear cells [301] and further cell subpopulations with the epigenetic code enabling the assessment of isolated form CB mononuclear cells [302] as well as pla­ the functional relevance of the epigenetic alterations cental tissues [148,182] may represent valuable biological in experimental systems. These tools might also in the materials for epigenetic testing. future provide new therapeutic modalities to correct the epigenetically defined imbalance in Th cell sub­ Future perspective populations or activate/repress regions that are aber­ Epigenetic mechanisms play a key role in immune rantly altered in allergic diseases. Epigenetic changes regulation and are influenced by or might mediate a in allergic diseases will be a major subject of research variety of environmental exposures leading to persis­ in the next few years and the information gained will tent molecular alteration of the gene regulatory land­ probably have a major impact on clinical practices in scape. Although the field of epigenetics in allergic dis­ the near future. eases has only recently gained momentum and studies on the epigenetic components of allergy have so far Financial & competing interests disclosure been limited, analyzing often only a small number This work was supported by funds from the German Centre for of individuals and little attempts have been made to Lung Research (DZL; 82DZL00502/A2; DP Potaczek, H Harb, confirm the findings in replication studies or assess H Renz), the Universities Giessen and Marburg Lung Centre the functional relevance of the observed changes, the (UGMLC; H Renz) and the Von Behring-Röntgen-Foundation importance of epigenetic mechanisms in the develop­ (Von Behring-Röntgen-Stiftung; H Harb, H Renz). H Harb, DP ment of allergic diseases has now clearly been demon­ Potaczek and H Renz are the members of the International strated. Large-scale projects with a focus on epigenetic Inflammation (in-FLAME) Network, Worldwide Universities changes such as the Pregnancy And Childhood Epi­ Network (WUN) and DZL. B Alashkar Alhamwe is a German (PACE) in the US and the EU-funded Mech­ Academic Exchange Service (DAAD) fellow (personal refer- anisms of the Development of Allergy (MEDALL) ence number: 91559386). Work in the laboratory of J Tost consortia are currently underway and are combining is supported by grants from the ANR (ANR-13-EPIG-0003-05 data from multiple large cohorts to investigate the and ANR-13-CESA-0011-05), Aviesan/INSERM (EPIG2014- impact of environmental factors on the epigenome 01 and EPlG2014-18) and INCa (PRT-K14-049), a Sirius re- and their role in the initiation of allergic diseases dur­ search award (UCB Pharma S.A.), a Passerelle research award ing childhood. Moreover, further studies are required (Pfizer), iCARE (MSD Avenir) and the institutional budget of to better characterize the mechanisms underlying the CNG. Work on food allergy in the laboratory of J Tost is different forms of (allergic) asthma and their corre­ partly funded by DBV Technologies and J Tost had travel fees lation with clinical characteristics. With a number refunded by DBV technologies. S Michel is an employee of of studies demonstrating a transgenerational inheri­ Secarna Pharmaceuticals. Funding organizations as well as the tance of disease-associated phenotypic traits, which management of Secarna had no influence on the content of are recurrently accompanied by alterations in DNA the manuscript. The authors have no other relevant affiliations methylation patterns in the different generations, the or financial involvement with any organization or entity with a study of epigenetic modifications might also provide financial interest in or financial conflict with the subject mat- some clues on the transgenerational effects of envi­ ter or materials discussed in the manuscript apart from those ronmental exposure. Technological advances allow disclosed. the detailed analysis of DNA methylation changes No writing assistance was utilized in the production of this in a truly genome-wide manner using highly purified manuscript. disease-relevant cells or even single cells, particularly in distant gene-regulatory features such an enhanc­ Open access ers where subtle changes might lead to a major reor­ This work is licensed under the Attribution-NonCommercial- ganization of the chromatin landscape. First studies No Derivatives 4.0 Unported License. To view a copy of this support the potential of DNA methylation changes as license, visit http://creativecommons.org/licenses/by-nc- biomarkers for the diagnosis or assessment of treat­ nd/4.0/

560 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

Executive summary • The prevalence and persistence of allergic diseases including asthma and food allergies has recently experienced a major increase in the Western World. • Epigenetic modifications play a key role in the differentiation of T cell lineages and influence the balance between distinct Th cell populations. • Epigenetic modifications mediate the effect of (or are influenced by) various environmental exposures, both allergy-protecting and conferring susceptibility to allergic diseases. • DNA methylation-based signatures in the relevant tissue differentiate patients with allergic diseases from controls. • Changes in the DNA methylation patterns might predict susceptibility to disease prior to its onset and correlate with sustained unresponsiveness to allergens. • Novel technological approaches such as epigenetic editing and DNAzymes provide new avenues for the treatment of allergic diseases.

References 13 Nwaru BI, Hickstein L, Panesar SS et al. The epidemiology Papers of special note have been highlighted as: • of interest; of food allergy in Europe: a systematic review and meta- •• of considerable interest analysis. Allergy 69(1), 62–75 (2014). 1 Shaker M. New insights into the allergic march. Curr. Opin. 14 Schieck M, Schouten JP, Michel S et al. Doublesex and Pediatr. 26(4), 516–520 (2014). mab-3 related transcription factor 1 (DMRT1) is a sex- specific genetic determinant of childhood-onset asthma 2 Bantz SK, Zhu Z, Zheng T. The atopic march: progression and is expressed in testis and macrophages. J. Allergy Clin. from atopic dermatitis to allergic rhinitis and asthma. J. Clin. Immunol. 138(2), 421–431 (2016). Cell. Immunol. 5(2), pii: 202 (2014). 15 Caubet JC, Szajewska H, Shamir R, Nowak-Wegrzyn A. 3 Alduraywish SA, Standl M, Lodge CJ et al. Is there a Non IgE-mediated gastrointestinal food allergies in children. march from early food sensitization to later childhood Pediatr. Allergy Immunol. 28(1), 6–17 (2016). allergic airway disease? Results from two prospective birth cohort studies. Pediatr. Allergy. Immunol. 28(1), 30–37 16 Schmid-Grendelmeier P, Simon D, Simon HU, Akdis (2016). CA, Wuthrich B. Epidemiology, clinical features, and immunology of the “intrinsic” (non-IgE-mediated) type of 4 Dave ND, Xiang L, Rehm KE, Marshall GD Jr. Stress and atopic dermatitis (constitutional dermatitis). Allergy 56(9), allergic diseases. Immunol. Allergy Clin. North Am. 31(1), 841–849 (2001). 55–68 (2011). 17 Barnes PJ. Intrinsic asthma: not so different from allergic 5 Clausen M, Agner T, Thomsen SF. Skin barrier dysfunction asthma but driven by superantigens? Clin. Exp. Allergy 39(8), and the atopic march. Curr. Treat. Options Allergy 2, 218 1145–1151 (2009). (2015). 18 Robinson D, Humbert M, Buhl R et al. Revisiting Type 6 Palmer DJ, Metcalfe J, Makrides M et al. Early regular egg 2-high and Type 2-low airway inflammation in asthma: exposure in infants with eczema: a randomized controlled current knowledge and therapeutic implications. Clin. Exp. trial. J. Allergy Clin. Immunol. 132(2), 387–392 e381 (2013). Allergy 47(2), 161–175 (2016). 7 Savage J, Sicherer S, Wood R. The natural history of food 19 Holgate ST, Wenzel S, Postma DS, Weiss ST, Renz H, Sly allergy. J. Allergy Clin. Immunol. Pract. 4(2), 196–203; quiz PD. Asthma. Nat. Rev. Dis. Primers. 1, 15025 (2015). 204 (2016). 20 Bieber T. Atopic dermatitis 2.0: from the clinical phenotype 8 Ashley S, Dang T, Koplin J, Martino D, Prescott S. Food to the molecular taxonomy and stratified medicine.Allergy for thought: progress in understanding the causes and 67(12), 1475–1482 (2012). mechanisms of food allergy. Curr. Opin. Allergy Clin. Immunol. 15(3), 237–242 (2015). 21 Potaczek DP, Kabesch M. Current concepts of IgE regulation and impact of genetic determinants. Clin. Exp. Allergy 42(6), 9 Valenta R, Hochwallner H, Linhart B, Pahr S. Food 852–871 (2012). allergies: the basics. Gastroenterology 148(6), 1120–1131 e1124 (2015). 22 Lambrecht BN, Hammad H. The immunology of asthma. Nat. Immunol. 16(1), 45–56 (2015). 10 Jensen-Jarolim E, Untersmayr E. Gender-medicine aspects in allergology. Allergy 63(5), 610–615 (2008). • Summarizes current insights into the cellular and 11 Chen W, Mempel M, Schober W, Behrendt H, Ring J. molecular mechanisms underlying asthma. Gender difference, sex hormones, and immediate type 23 Gandhi NA, Bennett BL, Graham NM, Pirozzi G, Stahl N, hypersensitivity reactions. Allergy 63(11), 1418–1427 Yancopoulos GD. Targeting key proximal drivers of type 2 (2008). inflammation in disease.Nat. Rev. Drug Discov. 15(1), 35–50 12 Kotz D, Simpson CR, Sheikh A. Incidence, prevalence, and (2016). trends of general practitioner-recorded diagnosis of peanut •• Provides a recent update on the cellular and molecular allergy in England, 2001 to 2005. J. Allergy Clin. Immunol. mechanisms driving type 2 (allergic) inflammation and 127(3), 623–630, e621 (2011). describes its pathophysiological consequences.

future science group www.futuremedicine.com 561 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

24 Goswami R, Kaplan MH. A brief history of IL-9. 42 Strachan DP. Family size, infection and atopy: the first J. Immunol. 186(6), 3283–3288 (2011). decade of the “hygiene hypothesis”. Thorax 55(Suppl. 1), 25 Potaczek DP, Nishiyama C, Sanak M, Szczeklik A, Okumura S2–S10 (2000). K. Genetic variability of the high-affinity IgE receptor alpha- 43 Kabesch M. Early origins of asthma (and allergy). Mol. Cell. subunit (FcepsilonRIalpha). Immunol. Res. 45(1), 75–84 Pediatr. 3(1), 31 (2016). (2009). 44 Nielsen HM, Tost J. Epigenetic changes in inflammatory and 26 Binia A, Kabesch M. Respiratory medicine – genetic base for autoimmune diseases. Subcell. Biochem. 61, 455–478 (2013). allergy and asthma. Swiss Med. Wkly 142, w13612 (2012). 45 Tost J. DNA methylation: an introduction to the biology 27 Tamari M, Tanaka S, Hirota T. Genome-wide and the disease-associated changes of a promising biomarker. association studies of allergic diseases. Allergol. Int. 62(1), Mol. Biotechnol. 44(1), 71–81 (2010). 21–28 (2013). 46 Bird A. DNA methylation patterns and epigenetic memory. 28 Akhabir L, Sandford AJ. Genome-wide association studies Genes Dev. 16(1), 6–21 (2002). for discovery of genes involved in asthma. Respirology 16(3), 47 Schubeler D. Function and information content of DNA 396–406 (2011). methylation. Nature 517(7534), 321–326 (2015). 29 Tamari M, Hirota T. Genome-wide association studies of 48 Moore LD, Le T, Fan G. DNA methylation and its basic atopic dermatitis. J. Dermatol. 41(3), 213–220 (2014). function. Neuropsychopharmacology 38(1), 23–38 (2013). 30 Pfefferle PI, Renz H. Microbial exposure and onset of allergic 49 Smith ZD, Meissner A. DNA methylation: roles in diseases – potential prevention strategies? Allergol. Int. 63(1), mammalian development. Nat. Rev. Genet. 14(3), 204–220 3–10 (2014). (2013). 31 Kabesch M, Michel S, Tost J. Epigenetic mechanisms and 50 Robertson KD. DNA methylation, methyltransferases, and the relationship to childhood asthma. Eur. Respir. J. 36(4), cancer. Oncogene 20(24), 3139–3155 (2001). 950–961 (2010). 51 Jurkowska RZ, Jeltsch A. Mechanisms and biological roles of 32 Harb H, Alashkar Alhamwe B, Garn H, Renz H, Potaczek DNA methyltransferases and DNA methylation: from past DP. Recent developments in epigenetics of pediatric asthma. achievements to future challenges. Adv. Exp. Med. Biol. 945, Curr. Opin. Pediatr. 28(6), 754–763 (2016). 1–17 (2016). •• Reports the most recent advances in the role of epigenetic 52 Kouzarides T. Chromatin modifications and their function. mechanisms in the development and pathophysiology of Cell 128(4), 693–705 (2007). pediatric asthma. 53 Rothbart SB, Strahl BD. Interpreting the language of histone 33 von Mutius E. Gene-environment interactions in asthma. and DNA modifications.Biochim. Biophys. Acta 1839(8), J. Allergy Clin. Immunol. 123(1), 3–11; quiz 12–13 (2009). 627–643 (2014). 34 Renz H, Autenrieth IB, Brandtzaeg P et al. Gene- 54 Ernst J, Kellis M. ChromHMM: automating chromatin-state environment interaction in chronic disease: a European discovery and characterization. Nat. Methods 9(3), 215–216 Science Foundation Forward Look. J. Allergy Clin. Immunol. (2012). 128(Suppl. 6), S27–S49 (2011). 55 Hoffman MM, Ernst J, Wilder SP et al. Integrative 35 von Mutius E. The environmental predictors of allergic annotation of chromatin elements from ENCODE data. disease. J. Allergy Clin. Immunol. 105(1 Pt 1), 9–19 (2000). Nucleic Acids Res. 41(2), 827–841 (2013). 36 Jackson DJ, Johnston SL. The role of viruses in acute 56 Zentner GE, Henikoff S. Regulation of nucleosome exacerbations of asthma. J. Allergy Clin. Immunol. 125(6), dynamics by histone modifications.Nat. Struct. Mol. Biol. 1178–1187; quiz 1188–1179 (2010). 20(3), 259–266 (2013). 37 Holt PG, Sly PD. Viral infections and atopy in asthma 57 Wapenaar H, Dekker FJ. Histone acetyltransferases: pathogenesis: new rationales for asthma prevention and challenges in targeting bi-substrate enzymes. Clin. Epigenetics treatment. Nat. Med. 18(5), 726–735 (2012). 8, 59 (2016). 38 Ong PY, Leung DY. Bacterial and viral infections in atopic 58 Delcuve GP, Khan DH, Davie JR. Roles of histone dermatitis: a comprehensive review. Clin. Rev. Allergy deacetylases in epigenetic regulation: emerging paradigms Immunol. 51(3), 329–337 (2016). from studies with inhibitors. Clin. Epigenetics 4(1), 5 (2012). 39 Huvenne W, Hellings PW, Bachert C. Role of staphylococcal 59 Yi X, Jiang XJ, Li XY, Jiang DS. Histone methyltransferases: superantigens in airway disease. Int. Arch. Allergy Immunol. novel targets for tumor and developmental defects. Am. J. 161(4), 304–314 (2013). Transl. Res. 7(11), 2159–2175 (2015). 40 Bardana EJ Jr. Immunoglobulin E- (IgE) and non-IgE- 60 Song Y, Wu F, Wu J. Targeting histone methylation for mediated reactions in the pathogenesis of atopic eczema/ cancer therapy: enzymes, inhibitors, biological activity and dermatitis syndrome (AEDS). Allergy. 59(Suppl. 78), 25–29 perspectives. J. Hematol. Oncol. 9(1), 49 (2016). (2004). 61 Rossetto D, Avvakumov N, Cote J. Histone phosphorylation: 41 Sorensen M, Wickman M, Sollid JU, Furberg AS, a chromatin modification involved in diverse nuclear events. Klingenberg C. Allergic disease and Staphylococcus aureus Epigenetics 7(10), 1098–1108 (2012). carriage in adolescents in the Arctic region of Norway. 62 Harb H, Renz H. Update on epigenetics in allergic disease. Pediatr. Allergy Immunol. 27(7), 728–735 (2016). J. Allergy Clin. Immunol. 135(1), 15–24 (2015).

562 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

63 Jin B, Li Y, Robertson KD. DNA methylation: superior or 81 Potaczek DP, Garn H, Unger SD, Renz H. Antisense subordinate in the epigenetic hierarchy? Genes Cancer 2(6), molecules: a new class of drugs. J. Allergy Clin. Immunol. 607–617 (2011). 137(5), 1334–1346 (2016). 64 Saetrom P, Snove O Jr., Rossi JJ. Epigenetics and • Describes the most recent status of clinical studies testing microRNAs. Pediatr. Res. 61(5 Pt 2), R17–R23 (2007). antisense approach-based drugs. 65 Ha M, Kim VN. Regulation of microRNA biogenesis. Nat. 82 Sato F, Tsuchiya S, Meltzer SJ, Shimizu K. MicroRNAs and Rev. Mol. Cell. Biol. 15(8), 509–524 (2014). epigenetics. FEBS J. 278(10), 1598–1609 (2011). 66 Piletic K, Kunej T. MicroRNA epigenetic signatures in 83 Suarez-Alvarez B, Rodriguez RM, Fraga MF, Lopez-Larrea C. human disease. Arch. Toxicol. 90(10), 2405–2419 (2016). DNA methylation: a promising landscape for immune system- 67 Eulalio A, Mano M. MicroRNA screening and the quest related diseases. Trends Genet. 28(10), 506–514 (2012). for biologically relevant targets. J. Biomol. Screen. 20(8), •• Describes the role of DNA methylation in the 1003–1017 (2015). differentiation of several types of immune cells important 68 Kala R, Peek GW, Hardy TM, Tollefsbol TO. MicroRNAs: for atopic predisposition and allergic inflammation. an emerging science in cancer epigenetics. J. Clin. Bioinform. 84 Fontenot JD, Gavin MA, Rudensky AY. Foxp3 programs the 3(1), 6 (2013). development and function of CD4+CD25+ regulatory T cells. 69 Chuang JC, Jones PA. Epigenetics and microRNAs. Pediatr. Nat. Immunol. 4(4), 330–336 (2003). Res. 61(5 Pt 2), R24–R29 (2007). 85 Ivanov II, McKenzie BS, Zhou L et al. The orphan nuclear 70 Vidigal JA, Ventura A. The biological functions of miRNAs: receptor RORgammat directs the differentiation program lessons from in vivo studies. Trends Cell. Biol. 25(3), 137–147 of proinflammatory IL-17+ T helper cells. Cell 126(6), (2015). 1121–1133 (2006). 71 Liang Y, Chang C, Lu Q. The genetics and epigenetics of 86 Hirahara K, Nakayama T. CD4+ T-cell subsets in atopic dermatitis-filaggrin and other polymorphisms.Clin. inflammatory diseases: beyond the Th1/Th2 paradigm.Int. Rev. Allergy Immunol. 51(3), 315–328 (2016). Immunol. 28(4), 163–171 (2016). 72 Bin L, Leung DY. Genetic and epigenetic studies of 87 Hirahara K, Vahedi G, Ghoreschi K et al. Helper T-cell atopic dermatitis. Allergy Asthma Clin. Immunol. 12, 52 differentiation and plasticity: insights from epigenetics. (2016). Immunology 134(3), 235–245 (2011). 73 Brook PO, Perry MM, Adcock IM, Durham AL. 88 Aune TM, Collins PL, Chang S. Epigenetics and T helper 1 Epigenome-modifying tools in asthma. Epigenomics 7(6), differentiation. Immunology 126(3), 299–305 (2009). 1017–1032 (2015). 89 Placek K, Coffre M, Maiella S, Bianchi E, Rogge L. 74 Moheimani F, Hsu AC, Reid AT et al. The genetic and Genetic and epigenetic networks controlling T helper 1 cell epigenetic landscapes of the epithelium in asthma. Respir. differentiation. Immunology 127(2), 155–162 (2009). Res. 17(1), 119 (2016). 90 Athie-Morales V, Smits HH, Cantrell DA, Hilkens CM. 75 Perry MM, Adcock IM, Chung KF. Role of microRNAs in Sustained IL-12 signaling is required for Th1 development. allergic asthma: present and future. Curr. Opin. Allergy Clin. J. Immunol. 172(1), 61–69 (2004). Immunol. 15(2), 156 –162 (2015). 91 Ling MF, Luster AD. Allergen-specific CD4(+) T cells • Summarizes and discusses recent developments in the role in human asthma. Ann. Am. Thorac. Soc. 13(Suppl. 1), of miRNAs in asthma but also other allergic disorders. S25–S30 (2016). 76 Rebane A. microRNA and allergy. Adv. Exp. Med. Biol. 888, 92 Tindemans I, Serafini N, Di Santo JP, Hendriks RW. GATA- 331–352 (2015). 3 function in innate and adaptive immunity. Immunity 41(2), 191–206 (2014). •• Comprehensively summarizes and extensively discusses our knowledge on the role of miRNAs in allergic disorders. 93 Geginat J, Paroni M, Maglie S et al. Plasticity of human CD4 T cell subsets. Front. Immunol. 5, 630 (2014). 77 Alipoor SD, Adcock IM, Garssen J et al. The roles of miRNAs as potential biomarkers in lung diseases. Eur. J. 94 Kaplan MH, Hufford MM, Olson MR. The development Pharmacol. 791, 395–404 (2016). and in vivo function of T helper 9 cells. Nat. Rev. Immunol. 15(5), 295–307 (2015). 78 Alipoor SD, Mortaz E, Garssen J, Movassaghi M, Mirsaeidi M, Adcock IM. Exosomes and exosomal miRNA in 95 Kaplan MH. Th9 cells: differentiation and disease. Immunol. respiratory diseases. Mediators Inflamm. 2016, 5628404 Rev. 252(1), 104–115 (2013). (2016). 96 Cosmi L, Liotta F, Maggi E, Romagnani S, Annunziato 79 Pattarayan D, Thimmulappa RK, Ravikumar V, F. Th17 cells: new players in asthma pathogenesis. Allergy Rajasekaran S. Diagnostic potential of extracellular 66(8), 989–998 (2011). microRNA in respiratory diseases. Clin. Rev. Allergy 97 Wang C, Yosef N, Gaublomme J et al. CD5L/AIM regulates Immunol. doi:10.1007/s12016–016–8589–9 (2016) (Epub lipid biosynthesis and restrains Th17 cell pathogenicity. Cell ahead of print). 163(6), 1413–1427 (2015). 80 Miller JA, Witherow DS, Carson S. A laboratory-intensive 98 Figueiredo AS, Schumacher A. The T helper type 17/ course on RNA interference and model organisms. CBE Life regulatory T cell paradigm in pregnancy. Immunology 148(1), Sci. Educ. 8(4), 316–325 (2009). 13–21 (2016).

future science group www.futuremedicine.com 563 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

99 Diller ML, Kudchadkar RR, Delman KA, Lawson DH, Ford 115 Chang S, Aune TM. Dynamic changes in histone- ML. Balancing inflammation: the link between Th17 and methylation ‘marks’ across the locus encoding interferon- regulatory T cells. Mediators Inflamm. 2016, 6309219 (2016). gamma during the differentiation of T helper type 2 cells. 100 Galgani M, De Rosa V, La Cava A, Matarese G. Role of Nat. Immunol. 8(7), 723–731 (2007). metabolism in the immunobiology of regulatory T cells. 116 Wei G, Wei L, Zhu J et al. Global mapping of H3K4me3 J. Immunol. 197(7), 2567–2575 (2016). and H3K27me3 reveals specificity and plasticity in lineage + 101 Devaud C, Darcy PK, Kershaw MH. Foxp3 expression in fate determination of differentiating CD4 T cells. Immunity T regulatory cells and other cell lineages. Cancer Immunol. 30(1), 155–167 (2009). Immunother. 63(9), 869–876 (2014). 117 Akimzhanov AM, Yang XO, Dong C. Chromatin 102 Tripathi SK, Lahesmaa R. Transcriptional and epigenetic remodeling of interleukin-17 (IL-17)-IL-17F cytokine gene regulation of T-helper lineage specification.Immunol. Rev. locus during inflammatory helper T cell differentiation. 261(1), 62–83 (2014). J. Biol. Chem. 282(9), 5969–5972 (2007). •• Describes transcriptional and epigenetic regulation of 118 Zheng Y, Josefowicz S, Chaudhry A, Peng XP, Forbush K, Rudensky AY. Role of conserved non-coding DNA CD4+ T cells differentiation. elements in the Foxp3 gene in regulatory T-cell fate. Nature 103 Perumal NB, Kaplan MH. Regulating Il9 transcription in 463(7282), 808–812 (2010). T helper cells. Trends Immunol. 32(4), 146–150 (2011). 119 Tao R, de Zoeten EF, Ozkaynak E et al. Deacetylase 104 Rodriguez RM, Lopez-Larrea C, Suarez-Alvarez B. inhibition promotes the generation and function of + Epigenetic dynamics during CD4( ) T cells lineage regulatory T cells. Nat. Med. 13(11), 1299–1307 (2007). commitment. Int. J. Biochem. Cell. Biol. 67, 75–85 (2015). 120 Schieck M, Sharma V, Michel S et al. A polymorphism in the 105 Schoenborn JR, Dorschner MO, Sekimata M et al. TH 2 locus control region is associated with changes in DNA Comprehensive epigenetic profiling identifies multiple distal methylation and gene expression. Allergy 69(9), 1171–1180 regulatory elements directing transcription of the gene encoding (2014). interferon-gamma. Nat. Immunol. 8(7), 732–742 (2007). 121 Wilson CB, Rowell E, Sekimata M. Epigenetic control 106 Thomas RM, Gamper CJ, Ladle BH, Powell JD, Wells AD. of T-helper-cell differentiation. Nat. Rev. Immunol. 9(2), De novo DNA methylation is required to restrict T helper 91–105 (2009). lineage plasticity. J. Biol. Chem. 287(27), 22900–22909 122 Lee GR, Fields PE, Griffin TJ, Flavell RA. Regulation of (2012). the Th2 cytokine locus by a locus control region. Immunity 107 Santangelo S, Cousins DJ, Winkelmann N, 19(1), 145–153 (2003). Triantaphyllopoulos K, Staynov DZ. Chromatin structure 123 Fields PE, Lee GR, Kim ST, Bartsevich VV, Flavell RA. and DNA methylation of the IL-4 gene in human T(H)2 Th2-specific chromatin remodeling and enhancer activity cells. Chromosome Res. 17(4), 485–496 (2009). in the Th2 cytokine locus control region. Immunity 21(6), 108 Cohen CJ, Crome SQ, MacDonald KG, Dai EL, Mager DL, 865–876 (2004). Levings MK. Human Th1 and Th17 cells exhibit epigenetic 124 Potaczek DP. Links between allergy and cardiovascular or stability at signature cytokine and transcription factor loci. hemostatic system. Int. J. Cardiol. 170(3), 278–285 (2014). J. Immunol. 187(11), 5615–5626 (2011). 125 March ME, Sleiman PM, Hakonarson H. The genetics of 109 Janson PC, Winerdal ME, Marits P, Thorn M, Ohlsson R, asthma and allergic disorders. Discov. Med. 11(56), 35–45 Winqvist O. FOXP3 promoter demethylation reveals the (2011). committed Treg population in humans. PLoS ONE 3(2), e1612 (2008). 126 Renz H, Conrad M, Brand S, Teich R, Garn H, Pfefferle PI. Allergic diseases, gene–environment interactions. Allergy 110 Bending D, Newland S, Krejci A, Phillips JM, Bray S, Cooke 66(Suppl. 95), 10–12 (2011). A. Epigenetic changes at Il12rb2 and Tbx21 in relation to plasticity behavior of Th17 cells. J. Immunol. 186(6), 127 Kramer U, Oppermann H, Ranft U, Schafer T, Ring J, 3373–3382 (2011). Behrendt H. Differences in allergy trends between East and West Germany and possible explanations. Clin. Exp. Allergy. 111 Lal G, Zhang N, van der Touw W et al. Epigenetic 40(2), 289–298 (2010). regulation of Foxp3 expression in regulatory T cells by DNA methylation. J. Immunol. 182(1), 259–273 (2009). 128 Kramer U, Schmitz R, Ring J, Behrendt H. What can reunification of East and West Germany tell us about the 112 Polansky JK, Kretschmer K, Freyer J et al. DNA methylation cause of the allergy epidemic? Clin. Exp. Allergy 45(1), controls Foxp3 gene expression. Eur. J. Immunol. 38(6), 94–107 (2015). 1654–1663 (2008). 129 Feil R, Fraga MF. Epigenetics and the environment: 113 Chang HC, Sehra S, Goswami R et al. The transcription emerging patterns and implications. Nat. Rev. Genet. 13(2), factor PU.1 is required for the development of IL-9- 97–109 (2012). producing T cells and allergic inflammation.Nat. Immunol. 11(6), 527–534 (2010). • Describes the role of epigenetic mechanisms in mediating 114 Ramming A, Druzd D, Leipe J, Schulze-Koops H, Skapenko the effects of environmental influences. A. Maturation-related histone modifications in the PU.1 130 Rachid R, Chatila TA. The role of the gut microbiota in food promoter regulate Th9-cell development. Blood 119(20), allergy. Curr. Opin. Pediatr. 28(6), 748–753 (2016). 4665–4674 (2012).

564 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

131 Noval Rivas M, Crother TR, Arditi M. The microbiome in 147 Moles L, Gomez M, Heilig H et al. Bacterial diversity in asthma. Curr. Opin. Pediatr. 28(6), 764–771 (2016). meconium of preterm neonates and evolution of their fecal • Outlines the role of microbiota in allergic disorders, microbiota during the first month of life.PLoS ONE 8(6), e66986 (2013). especially asthma. 148 Slaats GG, Reinius LE, Alm J, Kere J, Scheynius A, Joerink 132 Braun-Fahrlander C, Riedler J, Herz U et al. M. DNA methylation levels within the CD14 promoter Environmental exposure to endotoxin and its relation to region are lower in placentas of mothers living on a farm. asthma in school-age children. N. Engl. J. Med. 347(12), Allergy 67(7), 895–903 (2012). 869–877 (2002). 149 Joerink M, Oortveld MA, Stenius F, Rindsjo E, Alm J, 133 McDade TW. Early environments and the ecology of Scheynius A. Lifestyle and parental allergen sensitization are inflammation.Proc. Natl Acad. Sci. USA 109(Suppl. 2), reflected in the intrauterine environment at gene expression 17281–17288 (2012). level. Allergy 65(10), 1282–1289 (2010). 134 von Mutius E, Radon K. Living on a farm: impact on asthma 150 Martino D, Kesper DA, Amarasekera M, Harb H, Renz induction and clinical course. Immunol. Allergy Clin. North H, Prescott S. Epigenetics in immune development and Am. 28(3), 631–647, ix–x (2008). in allergic and autoimmune diseases. J. Reprod. Immunol. 135 Debarry J, Garn H, Hanuszkiewicz A et al. Acinetobacter 104–105, 43–48 (2014). lwoffii andLactococcus lactis strains isolated from farm 151 Khan TK, Palmer DJ, Prescott SL. In-utero exposures and cowsheds possess strong allergy-protective properties. the evolving epidemiology of paediatric allergy. Curr. Opin. J. Allergy Clin. Immunol. 119(6), 1514–1521 (2007). Allergy Clin. Immunol. 15(5), 402–408 (2015). 136 Ege MJ, Mayer M, Schwaiger K et al. Environmental bacteria 152 Hinz D, Bauer M, Roder S et al. Cord blood Tregs with and childhood asthma. Allergy 67(12), 1565–1571 (2012). stable FOXP3 expression are influenced by prenatal 137 Hagner S, Harb H, Zhao M et al. Farm-derived Gram- environment and associated with atopic dermatitis at the age positive bacterium Staphylococcus sciuri W620 prevents of one year. Allergy 67(3), 380–389 (2012). asthma phenotype in HDM- and OVA-exposed mice. Allergy 153 Lluis A, Depner M, Gaugler B et al. Increased regulatory 68(3), 322–329 (2013). T-cell numbers are associated with farm milk exposure and 138 Conrad ML, Ferstl R, Teich R et al. Maternal TLR lower atopic sensitization and asthma in childhood. J. Allergy signaling is required for prenatal asthma protection by the Clin. Immunol. 133(2), 551–559 (2014). nonpathogenic microbe Acinetobacter lwoffii F78. J. Exp. 154 Brick T, Schober Y, Bocking C et al. omega-3 fatty acids Med. 206(13), 2869–2877 (2009). contribute to the asthma-protective effect of unprocessed 139 Brand S, Teich R, Dicke T et al. Epigenetic regulation in cow’s milk. J. Allergy Clin. Immunol. 137(6), 1699–1706, murine offspring as a novel mechanism for transmaternal e1613 (2016). asthma protection induced by microbes. J. Allergy Clin. 155 von Mutius E. Maternal farm exposure/ingestion of Immunol. 128(3), 618–625, e611–e617 (2011). unpasteurized cow’s milk and allergic disease. Curr. Opin. 140 Pistiner M, Gold DR, Abdulkerim H, Hoffman E, Celedon Gastroenterol. 28(6), 570–576 (2012). JC. Birth by cesarean section, allergic rhinitis, and allergic 156 Matricardi PM. 99th Dahlem conference on infection, sensitization among children with a parental history of atopy. inflammation and chronic inflammatory disorders: J. Allergy Clin. Immunol. 122(2), 274–279 (2008). controversial aspects of the ‘hygiene hypothesis’. Clin. Exp. 141 Renz-Polster H, David MR, Buist AS et al. Caesarean section Immunol. 160(1), 98–105 (2010). delivery and the risk of allergic disorders in childhood. Clin. 157 Jartti T, Gern JE. Rhinovirus-associated wheeze during Exp. Allergy. 35(11), 1466–1472 (2005). infancy and asthma development. Curr. Respir. Med. Rev. 142 Bager P, Melbye M, Rostgaard K, Benn CS, Westergaard 7(3), 160–166 (2011). T. Mode of delivery and risk of allergic rhinitis and asthma. 158 Sly PD, Kusel M, Holt PG. Do early-life viral infections J. Allergy Clin. Immunol. 111(1), 51–56 (2003). cause asthma? J. Allergy Clin. Immunol. 125(6), 1202–1205 143 Maitra A, Sherriff A, Strachan D, Henderson J, Team AS. (2010). Mode of delivery is not associated with asthma or atopy in 159 Subrata LS, Bizzintino J, Mamessier E et al. childhood. Clin. Exp. Allergy 34(9), 1349–1355 (2004). Interactions between innate antiviral and atopic 144 Schaub B, Liu J, Hoppler S et al. Maternal farm exposure immunoinflammatory pathways precipitate and sustain modulates neonatal immune mechanisms through regulatory asthma exacerbations in children. J. Immunol. 183(4), T cells. J. Allergy Clin. Immunol. 123(4), 774–782, e775 2793–2800 (2009). (2009). 160 Holt PG, Sly PD. Interaction between adaptive and innate 145 Ege MJ, Bieli C, Frei R et al. Prenatal farm exposure is immune pathways in the pathogenesis of atopic asthma: related to the expression of receptors of the innate immunity operation of a lung/bone marrow axis. Chest 139(5), and to atopic sensitization in school-age children. J. Allergy 1165–1171 (2011). Clin. Immunol. 117(4), 817–823 (2006). 161 Kusel MM, de Klerk NH, Kebadze T et al. Early-life 146 Aagaard K, Ma J, Antony KM, Ganu R, Petrosino J, respiratory viral infections, atopic sensitization, and risk of Versalovic J. The placenta harbors a unique microbiome. Sci. subsequent development of persistent asthma. J. Allergy Clin. Transl. Med. 6(237), 237ra265 (2014). Immunol. 119(5), 1105–1110 (2007).

future science group www.futuremedicine.com 565 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

162 Potaczek DP, Michel S, Sharma V et al. Different FCER1A 176 Fu A, Leaderer BP, Gent JF, Leaderer D, Zhu Y. An polymorphisms influence IgE levels in asthmatics and environmental epigenetic study of ADRB2 5′-UTR non-asthmatics. Pediatr. Allergy Immunol. 24(5), 441–449 methylation and childhood asthma severity. Clin. Exp. (2013). Allergy 42(11), 1575–1581 (2012). 163 Sharma V, Michel S, Gaertner V et al. Fine-mapping of IgE- 177 Cook DG, Strachan DP. Health effects of passive smoking. associated loci 1q23, 5q31, and 12q13 using 1000 Genomes 3. Parental smoking and prevalence of respiratory symptoms Project data. Allergy 69(8), 1077–1084 (2014). and asthma in school age children. Thorax 52(12), 164 Cicchini L, Westrich JA, Xu T et al. Suppression of 1081–1094 (1997). antitumor immune responses by human papillomavirus 178 Wang IJ, Hsieh WS, Wu KY et al. Effect of gestational through epigenetic downregulation of CXCL14. MBio 7(3) smoke exposure on atopic dermatitis in the offspring. Pediatr. (2016). Allergy Immunol. 19(7), 580–586 (2008). 165 Okada H, Kuhn C, Feillet H, Bach JF. The ‘hygiene 179 Vardavas CI, Hohmann C, Patelarou E et al. The hypothesis’ for autoimmune and allergic diseases: an update. independent role of prenatal and postnatal exposure to active Clin. Exp. Immunol. 160(1), 1–9 (2010). and passive smoking on the development of early wheeze in 166 Nestor CE, Barrenas F, Wang H et al. DNA methylation children. Eur. Respir. J. 48(1), 115–124 (2016). changes separate allergic patients from healthy controls and 180 Tanaka K, Miyake Y, Furukawa S, Arakawa M. Pre- and may reflect altered CD4+ T-cell population structure. PLoS postnatal smoking exposure and risk of atopic eczema in Genet. 10(1), e1004059 (2014). young Japanese children: a prospective pre-birth cohort •• Shows that allergic patients can be separated from controls study. Nicotine Tob. Res. pii:ntw299 (2016) (Epub ahead of print). based on DNA methylation signatures, but not gene expression signatures. 181 Fuchs O, Bahmer T, Rabe KF, von Mutius E. Asthma transition from childhood into adulthood. Lancet Respir. 167 Hew KM, Walker AI, Kohli A et al. Childhood exposure Med. doi:10.1016/S2213-2600(16)30187-4 (2016) (Epub to ambient polycyclic aromatic hydrocarbons is linked to ahead of print). epigenetic modifications and impaired systemic immunity in T cells. Clin. Exp. Allergy 45(1), 238–248 (2015). 182 Chhabra D, Sharma S, Kho AT et al. Fetal lung and placental methylation is associated with in utero nicotine exposure. 168 Tang WY, Levin L, Talaska G et al. Maternal exposure to Epigenetics 9(11), 1473–1484 (2014). polycyclic aromatic hydrocarbons and 5′-CpG methylation of interferon-gamma in cord white blood cells. Environ. Health 183 Breton CV, Siegmund KD, Joubert BR et al. Prenatal tobacco Perspect. 120(8), 1195–1200 (2012). smoke exposure is associated with childhood DNA CpG methylation. PLoS ONE 9(6), e99716 (2014). 169 Somineni HK, Zhang X, Biagini Myers JM et al. Ten-eleven translocation 1 (TET1) methylation is associated with 184 Breitling LP, Yang R, Korn B, Burwinkel B, Brenner H. childhood asthma and traffic-related air pollution. J. Allergy Tobacco-smoking-related differential DNA methylation: Clin. Immunol. 137(3), 797–805, e795 (2016). 27K discovery and replication. Am. J. Hum. Genet. 88(4), 450–457 (2011). 170 Tahiliani M, Koh KP, Shen Y et al. Conversion of 5-methylcytosine to 5-hydroxymethylcytosine in mammalian 185 Breton CV, Byun HM, Wenten M, Pan F, Yang A, Gilliland DNA by MLL partner TET1. Science 324(5929), 930–935 FD. Prenatal tobacco smoke exposure affects global and (2009). gene-specific DNA methylation.Am. J. Respir. Crit. Care Med. 180(5), 462–467 (2009). 171 Kriaucionis S, Heintz N. The nuclear DNA base 5-hydroxymethylcytosine is present in Purkinje neurons and 186 Bauer T, Trump S, Ishaque N et al. Environment-induced the brain. Science 324(5929), 929–930 (2009). epigenetic reprogramming in genomic regulatory elements in smoking mothers and their children. Mol. Syst. Biol. 12(3), 172 Godderis L, Schouteden C, Tabish A et al. Global 861 (2016). methylation and hydroxymethylation in DNA from blood and saliva in healthy volunteers. Biomed. Res. Int. 2015, • A very recent study investigating the epigenetic effects 845041 (2015). of maternal smoking using a combination of whole 173 Brunst KJ, Leung YK, Ryan PH et al. Forkhead box protein genome approaches assessing DNA methylation, 3 (FOXP3) hypermethylation is associated with diesel histone modifications, gene expression and genotypes of exhaust exposure and risk for childhood asthma. J. Allergy polymorphisms. Clin. Immunol. 131(2), 592–594, e591–e593 (2013). 187 Wang IJ, Chen SL, Lu TP, Chuang EY, Chen PC. Prenatal 174 Breton CV, Salam MT, Wang X, Byun HM, smoke exposure, DNA methylation, and childhood atopic Siegmund KD, Gilliland FD. Particulate matter, DNA dermatitis. Clin. Exp. Allergy 43(5), 535–543 (2013). methylation in nitric oxide synthase, and childhood 188 Luo Y, Zhou B, Zhao M, Tang J, Lu Q. Promoter respiratory disease. Environ. Health Perspect. 120(9), demethylation contributes to TSLP overexpression in 1320–1326 (2012). skin lesions of patients with atopic dermatitis. Clin. Exp. 175 Reinartz MT, Wetzke M, Happle C et al. Neutrophilic Dermatol. 39(1), 48–53 (2014). superoxide production can assess pharmacological and 189 Rehan VK, Liu J, Naeem E et al. Perinatal nicotine exposure pharmacogenetic beta-adrenoreceptor effects. Allergy 71(8), induces asthma in second generation offspring. BMC Med. 1223–1227 (2016). 10, 129 (2012).

566 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

190 Eddleston J, Lee RU, Doerner AM, Herschbach J, Zuraw BL. • Describes the development of a novel bioinformatics tool, Cigarette smoke decreases innate responses of epithelial cells the gene, environment and methylation (GEM) package, to rhinovirus infection. Am. J. Respir. Cell. Mol. Biol. 44(1), which handles epigenome-wide association studies, 118–126 (2011). integrates genotype information and models interactions 191 Hudy MH, Traves SL, Proud D. Transcriptional and between genotype and the environment. epigenetic modulation of human rhinovirus-induced 204 Flanagan JM. Epigenome-wide association studies (EWAS): CXCL10 production by cigarette smoke. Am. J. Respir. Cell. past, present, and future. Methods Mol. Biol. 1238, 51–63 Mol. Biol. 50(3), 571–582 (2014). (2015). 192 Joubert BR, Felix JF, Yousefi Pet al. DNA Methylation in 205 Heijmans BT, Mill J. Commentary: the seven plagues of newborns and maternal smoking in pregnancy: genome- epigenetic epidemiology. Int. J. Epidemiol. 41(1), 74–78 wide consortium meta-analysis. Am. J. Hum. Genet. 98(4), (2012). 680–696 (2016). 206 Martino D, Joo JE, Sexton-Oates A et al. Epigenome- 193 Joehanes R, Just AC, Marioni RE et al. Epigenetic signatures wide association study reveals longitudinally stable DNA of cigarette smoking. Circ. Cardiovasc. Genet. 9(5), 436–447 methylation differences in CD4+ T cells from children (2016). with IgE-mediated food allergy. Epigenetics 9(7), 998–1006 194 Song Y, Liu C, Hui Y et al. Maternal allergy increases (2014). susceptibility to offspring allergy in association with TH2- 207 Petrus NC, Henneman P, Venema A et al. Cow’s milk allergy biased epigenetic alterations in a mouse model of peanut in Dutch children: an epigenetic pilot survey. Clin. Transl. allergy. J. Allergy Clin. Immunol. 134(6), 1339–1345, e1337 Allergy 6, 16 (2016). (2014). 208 Rodriguez E, Baurecht H, Wahn AF et al. An integrated 195 Acevedo N, Reinius LE, Greco D et al. Risk of childhood epigenetic and transcriptomic analysis reveals distinct asthma is associated with CpG-site polymorphisms, regional tissue-specific patterns of DNA methylation associated with DNA methylation and mRNA levels at the GSDMB/ atopic dermatitis. J. Invest. Dermatol. 134(7), 1873–1883 ORMDL3 locus. Hum. Mol. Genet. 24(3), 875–890 (2015). (2014). 196 Naumova AK, Al Tuwaijri A, Morin A et al. Sex- and 209 Stefanowicz D, Hackett TL, Garmaroudi FS et al. DNA age-dependent DNA methylation at the 17q12-q21 locus methylation profiles of airway epithelial cells and PBMCs associated with childhood asthma. Hum. Genet. 132(7), from healthy, atopic and asthmatic children. PLoS ONE 7(9), 811–822 (2013). e44213 (2012). 197 Toncheva AA, Potaczek DP, Schedel M et al. Childhood 210 Liang L, Willis-Owen SA, Laprise C et al. An epigenome- asthma is associated with mutations and gene expression wide association study of total serum immunoglobulin E differences of ORMDL genes that can interact. Allergy concentration. Nature 520(7549), 670–674 (2015). 70(10), 1288–1299 (2015). •• A landmark EWAS demonstrating that DNA methylation 198 Reinius LE, Gref A, Saaf A et al. DNA methylation in the levels at key immune genes vary in function of IgE levels Neuropeptide S Receptor 1 (NPSR1) promoter in relation to and that the variability of serum IgE levels can be better asthma and environmental factors. PLoS ONE 8(1), e53877 explained by DNA methylation patterns than genetic (2013). variation. 199 Morales E, Bustamante M, Vilahur N et al. DNA 211 Moran S, Arribas C, Esteller M. Validation of a DNA hypomethylation at ALOX12 is associated with persistent methylation microarray for 850,000 CpG sites of the human wheezing in childhood. Am. J. Respir. Crit. Care Med. genome enriched in enhancer sequences. Epigenomics 8(3), 185(9), 937–943 (2012). 389–399 (2016). 200 Chen W, Boutaoui N, Brehm JM et al. ADCYAP1R1 and 212 Pidsley R, Zotenko E, Peters TJ et al. Critical evaluation of asthma in Puerto Rican children. Am. J. Respir. Crit. Care the Illumina MethylationEPIC BeadChip microarray for Med. 187(6), 584–588 (2013). whole-genome DNA methylation profiling.Genome Biol. 201 Sarnowski C, Laprise C, Malerba G et al. DNA methylation 17(1), 208 (2016). within melatonin receptor 1A (MTNR1A) mediates 213 Tost J. Current and emerging technologies for the analysis paternally transmitted genetic variant effect on asthma plus of the genome-wide and locus-specific DNA methylation rhinitis. J. Allergy Clin. Immunol. 138(3), 748–753 (2016). patterns. Adv. Exp. Med. Biol. 945, 343–430 (2016). 202 Hong X, Hao K, Ladd-Acosta C et al. Genome-wide 214 Odegard JM, Nepom GT, Wambre E. Biomarkers for association study identifies peanut allergy-specific loci antigen immunotherapy in allergy and type 1 diabetes. Clin. and evidence of epigenetic mediation in US children. Nat. Immunol. 161(1), 44–50 (2015). Commun. 6, 6304 (2015). 215 How Kit A, Nielsen HM, Tost J. DNA methylation based 203 Pan H, Holbrook JD, Karnani N, Kwoh CK. Gene, biomarkers: practical considerations and applications. environment and methylation (GEM): a tool suite to Biochimie 94(11), 2314–2337 (2012). efficiently navigate large scale epigenome wide association studies and integrate genotype and interaction between 216 Liu J, Ballaney M, Al-alem U et al. Combined inhaled diesel genotype and environment. BMC Bioinformatics 17, 299 exhaust particles and allergen exposure alter methylation (2016). of T helper genes and IgE production in vivo. Toxicol. Sci. 102(1), 76–81 (2008).

future science group www.futuremedicine.com 567 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

217 Kwon NH, Kim JS, Lee JY, Oh MJ, Choi DC. DNA circulating DNA. Proc. Natl Acad. Sci. USA 108(47), methylation and the expression of IL-4 and IFN-gamma 19018–19023 (2011). promoter genes in patients with bronchial asthma. J. Clin. 229 Lehmann-Werman R, Neiman D, Zemmour H et al. Immunol. 28(2), 139–146 (2008). Identification of tissue-specific cell death using methylation 218 Berni Canani R, Paparo L, Nocerino R et al. Differences in patterns of circulating DNA. Proc. Natl Acad. Sci. USA DNA methylation profile of Th1 and Th2 cytokine genes are 113(13), E1826–E1834 (2016). associated with tolerance acquisition in children with IgE- 230 Hardy T, Zeybel M, Day CP et al. Plasma DNA methylation: mediated cow’s milk allergy. Clin. Epigenetics 7, 38 (2015). a potential biomarker for stratification of liver fibrosis 219 Michel S, Busato F, Genuneit J et al. Farm exposure and time in non-alcoholic fatty liver disease. Gut doi:10.1136/ trends in early childhood may influence DNA methylation in gutjnl-2016-311526 (2016) (Epub ahead of print). genes related to asthma and allergy. Allergy 68(3), 355–364 231 Chan RW, Jiang P, Peng X et al. Plasma DNA aberrations (2013). in systemic lupus erythematosus revealed by genomic and 220 Mondoulet L, Dioszeghy V, Puteaux E et al. Specific methylomic sequencing. Proc. Natl Acad. Sci. USA 111(49), epicutaneous immunotherapy prevents sensitization to E5302–E5311 (2014). new allergens in a murine model. J. Allergy Clin. Immunol. 232 Sun K, Jiang P, Chan KC et al. Plasma DNA tissue mapping 135(6), 1546 –1557, e1544 (2015). by genome-wide methylation sequencing for noninvasive 221 Martino D, Dang T, Sexton-Oates A et al. Blood DNA prenatal, cancer, and transplantation assessments. Proc. Natl methylation biomarkers predict clinical reactivity in Acad. Sci. USA 112(40), E5503–E5512 (2015). food-sensitized infants. J. Allergy Clin. Immunol. 135(5), 233 Moran S, Martinez-Cardus A, Sayols S et al. Epigenetic 1319–1328, e1311–1312 (2015). profiling to classify cancer of unknown primary: a •• An important study, which shows that a DNA methylation multicentre, retrospective analysis. Lancet Oncol. 17(10), signature composed of 96 CpGs, can predict the reactivity 1386–1395 (2016). of a child to multiple food allergens. The signature 234 Houseman EA, Accomando WP, Koestler DC et al. DNA outperformed classical measures and provides first evidence methylation arrays as surrogate measures of cell mixture to replace clinically challenging tests such as oral food distribution. BMC Bioinformatics 13, 86 (2012). challenge by molecular analyses. 235 Kleen TO, Yuan J. Quantitative real-time PCR assisted 222 Paparo L, Nocerino R, Cosenza L et al. Epigenetic features of cell counting (qPACC) for epigenetic – based immune cell FoxP3 in children with cow’s milk allergy. Clin. Epigenetics quantification in blood and tissue.J. Immunother. Cancer 3, 8, 86 (2016). 46 (2015). 223 Syed A, Garcia MA, Lyu SC et al. Peanut oral 236 Wu CJ, Yang CY, Chen YH, Chen CM, Chen LC, Kuo ML. immunotherapy results in increased antigen-induced The DNA methylation inhibitor 5-azacytidine increases regulatory T-cell function and hypomethylation of forkhead regulatory T cells and alleviates airway inflammation in box protein 3 (FOXP3). J. Allergy Clin. Immunol. 133(2), ovalbumin-sensitized mice. Int. Arch. Allergy Immunol. 500–510 (2014). 160(4), 356–364 (2013). • A clinical study showing the DNA methylation status of 237 Brand S, Kesper DA, Teich R et al. DNA methylation of FOXP3 correlating with response to oral immunotherapy TH1/TH2 cytokine genes affects sensitization and progress against peanut allergy and predict sustained of experimental asthma. J. Allergy Clin. Immunol. 129(6), unresponsiveness after the end of the therapy. 1602–1610, e1606 (2012). 224 Swamy RS, Reshamwala N, Hunter T et al. Epigenetic 238 Royce SG, Karagiannis TC. Histone deacetylases and modifications and improved regulatory -T cell function their inhibitors: new implications for asthma and chronic in subjects undergoing dual sublingual immunotherapy. respiratory conditions. Curr. Opin. Allergy Clin. Immunol. J. Allergy Clin. Immunol. 130(1), 215–224, e217 (2012). 14(1), 44–48 (2014). 225 Busse WW, Morgan WJ, Gergen PJ et al. Randomized trial 239 Thakore PI, D’Ippolito AM, Song L et al. Highly specific of omalizumab (anti-IgE) for asthma in inner-city children. epigenome editing by CRISPR-Cas9 repressors for silencing N. Engl. J. Med. 364(11), 1005–1015 (2011). of distal regulatory elements. Nat. Methods 12(12), 1143–1149 (2015). 226 Sorkness CA, Wildfire JJ, Calatroni Aet al. Reassessment of omalizumab-dosing strategies and pharmacodynamics in 240 Thakore PI, Black JB, Hilton IB, Gersbach CA. Editing inner-city children and adolescents. J. Allergy Clin. Immunol. the epigenome: technologies for programmable transcription Pract. 1(2), 163–171 (2013). and epigenetic modulation. Nat. Methods 13(2), 127–137 (2016). 227 Tost J. Follow the trace of death: methylation analysis of cell- free DNA for clinical applications in non-cancerous diseases. 241 Tost J. Engineering of the epigenome: synthetic biology to Epigenomics 8(9), 1169–1172 (2016). define functional causality and develop innovative therapies. Epigenomics 8(2), 153–156 (2016). •• Describes methylation analysis of cell-free DNA (liquid 242 Hilton IB, D’Ippolito AM, Vockley CM et al. Epigenome biopsy) as a useful tool in the future diagnostics of editing by a CRISPR-Cas9-based acetyltransferase activates noncancerous diseases. genes from promoters and enhancers. Nat. Biotechnol. 33(5), 228 Akirav EM, Lebastchi J, Galvan EM et al. Detection of 510–517 (2015). beta cell death in diabetes using differentially methylated

568 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

243 Kearns NA, Pham H, Tabak B et al. Functional annotation lactation reduces peanut allergy risk in offspring. J. Allergy of native enhancers with a Cas9-histone demethylase fusion. Clin. Immunol. 124(5), 1039–1046 (2009). Nat. Methods 12(5), 401–403 (2015). 259 Frazier AL, Camargo CA Jr., Malspeis S, Willett WC, Young 244 Liu Y, Zeng Y, Liu L et al. Synthesizing AND gate genetic MC. Prospective study of peripregnancy consumption of circuits based on CRISPR-Cas9 for identification of bladder peanuts or tree nuts by mothers and the risk of peanut or cancer cells. Nat. Commun. 5, 5393 (2014). tree nut allergy in their offspring. JAMA Pediatr. 168(2), 245 Arpaia N, Campbell C, Fan X et al. Metabolites produced 156 –162 (2014). by commensal bacteria promote peripheral regulatory T-cell 260 Pesek RD, Jones SM. Current and Emerging Therapies for generation. Nature 504(7480), 451–455 (2013). IgE-Mediated Food Allergy. Curr. Allergy Asthma Rep. 16(4), 246 Furusawa Y, Obata Y, Fukuda S et al. Commensal microbe- 28 (2016). derived butyrate induces the differentiation of colonic 261 Jones SM, Burks AW, Dupont C. State of the art on food regulatory T cells. Nature 504(7480), 446–450 (2013). allergen immunotherapy: oral, sublingual, and epicutaneous. 247 Stefka AT, Feehley T, Tripathi P et al. Commensal bacteria J. Allergy Clin. Immunol. 133(2), 318–323 (2014). protect against food allergen sensitization. Proc. Natl Acad. 262 Pajno GB, Nadeau KC, Passalacqua G et al. The evolution of Sci. USA 111(36), 13145–13150 (2014). allergen and non-specific immunotherapy: past achievements, 248 Fofanova TY, Petrosino JF, Kellermayer R. Microbiome- current applications and future outlook. Expert Rev. Clin. epigenome interactions and the environmental origins of Immunol. 11(1), 141–154 (2015). inflammatory bowel diseases.J. Pediatr. Gastroenterol. Nutr. 263 Vadlamudi A, Shaker M. New developments in allergen 62(2), 208–219 (2016). immunotherapy. Curr. Opin. Pediatr. 27(5), 649–655 249 Bertelsen RJ, Brantsaeter AL, Magnus MC et al. Probiotic (2015). milk consumption in pregnancy and infancy and subsequent 264 Akdis M, Akdis CA. Mechanisms of allergen-specific childhood allergic diseases. J. Allergy Clin. Immunol. 133(1), immunotherapy: multiple suppressor factors at work in 165–171, e161–e168 (2014). immune tolerance to allergens. J. Allergy Clin. Immunol. 250 Nermes M, Salminen S, Isolauri E. Is there a role for 133(3), 621–631 (2014). probiotics in the prevention or treatment of food allergy? 265 Moingeon P. Update on immune mechanisms associated Curr. Allergy Asthma Rep. 13(6), 622–630 (2013). with sublingual immunotherapy: practical implications for 251 Hollingsworth JW, Maruoka S, Boon K et al. In utero the clinician. J. Allergy Clin. Immunol. Pract. 1(3), 228–241 supplementation with methyl donors enhances allergic airway (2013). disease in mice. J. Clin. Invest. 118(10), 3462–3469 (2008). 266 Vickery BP, Scurlock AM, Kulis M et al. Sustained 252 Klemens CM, Berman DR, Mozurkewich EL. The unresponsiveness to peanut in subjects who have completed effect of perinatal omega-3 fatty acid supplementation on peanut oral immunotherapy. J. Allergy Clin. Immunol. inflammatory markers and allergic diseases: a systematic 133(2), 468–475 (2014). review. BJOG 118(8), 916–925 (2011). 267 Tang ML, Ponsonby AL, Orsini F et al. Administration of 253 Dunstan JA, Mori TA, Barden A et al. Fish oil a probiotic with peanut oral immunotherapy: a randomized supplementation in pregnancy modifies neonatal allergen- trial. J. Allergy Clin. Immunol. 135(3), 737–744.e8 (2015). specific immune responses and clinical outcomes in infants 268 Sel S, Wegmann M, Dicke T et al. Effective prevention and at high risk of atopy: a randomized, controlled trial. J. Allergy therapy of experimental allergic asthma using a GATA- Clin. Immunol. 112(6), 1178–1184 (2003). 3-specific DNAzyme.J. Allergy Clin. Immunol. 121(4), 254 Prescott SL, Irvine J, Dunstan JA, Hii C, Ferrante A. Protein 910–916.e5 (2008). kinase Czeta: a novel protective neonatal T-cell marker that 269 Limmroth V, Barkhof F, Desem N, Diamond MP, Tachas can be upregulated by allergy prevention strategies. J. Allergy G, Group ATLS. CD49d antisense drug ATL1102 reduces Clin. Immunol. 120(1), 200–206 (2007). disease activity in patients with relapsing-remitting MS. 255 Aslibekyan S, Wiener HW, Havel PJ et al. DNA methylation Neurology 83(20), 1780–1788 (2014). patterns are associated with n-3 fatty acid intake in Yup’ik 270 Buller HR, Bethune C, Bhanot S et al. Factor XI antisense people. J. Nutr. 144(4), 425–430 (2014). oligonucleotide for prevention of venous thrombosis. N. Engl. 256 Lee HS, Barraza-Villarreal A, Hernandez-Vargas H et al. J. Med. 372(3), 232–240 (2015). Modulation of DNA methylation states and infant immune 271 Miller TM, Pestronk A, David W et al. An antisense system by dietary supplementation with omega-3 PUFA oligonucleotide against SOD1 delivered intrathecally for during pregnancy in an intervention study. Am. J. Clin. Nutr. patients with SOD1 familial amyotrophic lateral sclerosis: 98(2), 480–487 (2013). a Phase I, randomised, first-in-man study. Lancet Neurol. 257 Amarasekera M, Noakes P, Strickland D, Saffery R, Martino 12(5), 435–442 (2013). DJ, Prescott SL. Epigenome-wide analysis of neonatal 272 Leachman SA, Hickerson RP, Schwartz ME et al. First- CD4(+) T-cell DNA methylation sites potentially affected in-human mutation-targeted siRNA Phase Ib trial of an by maternal fish oil supplementation.Epigenetics 9(12), inherited skin disorder. Mol. Ther. 18(2), 442–446 (2010). 1570 –1576 (2014). 273 Sussman J, Argov Z, Wirguin Y, Apolski S, Milic-Rasic V, 258 Lopez-Exposito I, Song Y, Jarvinen KM, Srivastava K, Li Soreq H. Further developments with antisense treatment for XM. Maternal peanut exposure during pregnancy and myasthenia gravis. Ann. N. Y. Acad. Sci. 1275, 13–16 (2012).

future science group www.futuremedicine.com 569 Review Potaczek, Harb, Michel, Alashkar Alhamwe, Renz & Tost

274 Turowska A, Librizzi D, Baumgartl N et al. Biodistribution 290 Yousefi P, Huen K, Quach Het al. Estimation of blood of the GATA-3-specific DNAzyme hgd40 after inhalative cellular heterogeneity in newborns and children for exposure in mice, rats and dogs. Toxicol. Appl. Pharmacol. epigenome-wide association studies. Environ. Mol. Mutagen. 272(2), 365–372 (2013). 56(9), 751–758 (2015). 275 Homburg U, Renz H, Timmer W et al. Safety and 291 Gaffin JM, Raby BA, Petty CRet al. beta-2 adrenergic tolerability of a novel inhaled GATA3 mRNA targeting receptor gene methylation is associated with decreased DNAzyme in patients with TH2-driven asthma. J. Allergy asthma severity in inner-city schoolchildren: asthma and Clin. Immunol. 136(3), 797–800 (2015). rhinitis. Clin. Exp. Allergy 44(5), 681–689 (2014). 276 Krug N, Hohlfeld JM, Kirsten AM et al. Allergen-induced 292 Langie SA, Szarc Vel Szic K, Declerck K et al. Whole-genome asthmatic responses modified by a GATA3-specific saliva and blood DNA methylation profiling in individuals DNAzyme. N. Engl. J. Med. 372(21), 1987–1995 (2015). with a respiratory allergy. PLoS ONE 11(3), e0151109 (2016). 277 Fernandez AF, Assenov Y, Martin-Subero JI et al. A DNA 293 Vercelli D. Does epigenetics play a role in human asthma? methylation fingerprint of 1628 human samples.Genome Res. Allergol. Int. 65(2), 123–126 (2016). 22(2), 407–419 (2012). 294 DeVries A, Vercelli D. Epigenetics in allergic diseases. Curr. 278 Rossnerova A, Tulupova E, Tabashidze N et al. Factors Opin. Pediatr. 27(6), 719–723 (2015). affecting the 27K DNA methylation pattern in asthmatic 295 Liang Y, Wang P, Zhao M et al. Demethylation of the and healthy children from locations with various FCER1G promoter leads to FcepsilonRI overexpression on environments. Mutat. Res. 741–742, 18–26 (2013). monocytes of patients with atopic dermatitis. Allergy 67(3), 279 Yang IV, Pedersen BS, Liu A et al. DNA methylation and 424–430 (2012). childhood asthma in the inner city. J. Allergy Clin. Immunol. 296 Kim JE, Kim JS, Cho DH, Park HJ. Molecular mechanisms 136(1), 69–80 (2015). of cutaneous inflammatory disorder: atopic dermatitis.Int. J. 280 Harb H, Raedler D, Ballenberger N et al. Childhood allergic Mol. Sci. 17(8), E1234 (2016). asthma is associated with increased IL-13 and FOXP3 297 Xiao C, Biagini Myers JM, Ji H et al. Vanin-1 expression and histone acetylation. J. Allergy Clin. Immunol. 136(1), methylation discriminate pediatric asthma corticosteroid 200–202 (2015). treatment response. J. Allergy Clin. Immunol. 136(4), 281 Reinius LE, Acevedo N, Joerink M et al. Differential DNA 923–931.e3 (2015). methylation in purified human blood cells: implications for 298 Stefanowicz D, Lee JY, Lee K et al. Elevated H3K18 cell lineage and studies on disease susceptibility. PLoS ONE acetylation in airway epithelial cells of asthmatic subjects. 7(7), e41361 (2012). Respir. Res. 16, 95 (2015). 282 Jaffe AE, Irizarry RA. Accounting for cellular heterogeneity 299 Heitzer E, Ulz P, Geigl JB. Circulating tumor DNA as a is critical in epigenome-wide association studies. Genome liquid biopsy for cancer. Clin. Chem. 61(1), 112–123 (2015). Biol. 15(2), R31 (2014). 300 Diaz LA Jr., Bardelli A. Liquid biopsies: genotyping 283 Houseman EA, Kile ML, Christiani DC, Ince TA, Kelsey circulating tumor DNA. J. Clin. Oncol. 32(6), 579–586 KT, Marsit CJ. Reference-free deconvolution of DNA (2014). methylation data and mediation by cell composition effects. 301 Curtin JA, Simpson A, Belgrave D, Semic-Jusufagic A, BMC Bioinformatics 17, 259 (2016). Custovic A, Martinez FD. Methylation of IL-2 promoter 284 Rahmani E, Zaitlen N, Baran Y et al. Sparse PCA corrects at birth alters the risk of asthma exacerbations during for cell type heterogeneity in epigenome-wide association childhood. Clin. Exp. Allergy 43(3), 304–311 (2013). studies. Nat. Methods 13(5), 443–445 (2016). 302 Harb H, Amarasekera M, Ashley S et al. Epigenetic 285 Zou J, Lippert C, Heckerman D, Aryee M, Listgarten J. regulation in early childhood: a miniaturized and validated Epigenome-wide association studies without the need for method to assess histone acetylation. Int. Arch. Allergy cell-type composition. Nat. Methods 11(3), 309–311 (2014). Immunol. 168(3), 173–181 (2015). 286 Koestler DC, Jones MJ, Usset J et al. Improving cell mixture 303 Hong X, Ladd-Acosta C, Hao K et al. Epigenome-wide deconvolution by identifying optimal DNA methylation association study links site-specific DNA methylation libraries (IDOL). BMC Bioinformatics 17, 120 (2016). changes with cow’s milk allergy. J. Allergy Clin. Immunol. 287 Heiss JA, Breitling LP, Lehne B, Kooner JS, Chambers 138(3), 908–911.e9 (2016). JC, Brenner H. Training a model for estimating leukocyte 304 Lockett GA, Soto-Ramirez N, Ray MA et al. Association of composition using whole-blood DNA methylation and cell season of birth with DNA methylation and allergic disease. counts as reference. Epigenomics 9(1), 13–20 (2017). Allergy 71(9), 1314–1324 (2016). 288 Gervin K, Page CM, Aass HC et al. Cell type specific DNA 305 Joubert BR, Haberg SE, Nilsen RM et al. 450K epigenome- methylation in cord blood: a 450K-reference data set and cell wide scan identifies differential DNA methylation in count-based validation of estimated cell type composition. newborns related to maternal smoking during pregnancy. Epigenetics 11(9), 690–698 (2016). Environ. Health Perspect. 120(10), 1425–1431 (2012). 289 Langie SA, Moisse M, Declerck K et al. Salivary DNA 306 Markunas CA, Xu Z, Harlid S et al. Identification of DNA methylation profiling: aspects to consider for biomarker methylation changes in newborns related to maternal identification.Basic Clin. Pharmacol. Toxicol. doi:10.1111/ smoking during pregnancy. Environ. Health Perspect. bcpt.12721 (2016) (Epub ahead of print). 122(10), 1147–1153 (2014).

570 Epigenomics (2017) 9(4) future science group Epigenetics & allergy: from basic mechanisms to clinical applications Review

307 Richmond RC, Simpkin AJ, Woodward G et al. Prenatal 310 Clifford RL, Jones MJ, MacIsaac JL et al. Inhalation of diesel exposure to maternal smoking and offspring DNA exhaust and allergen alters human bronchial epithelium methylation across the lifecourse: findings from the Avon DNA methylation. J. Allergy Clin. Immunol. 139(1), 112–121 Longitudinal Study of Parents and Children (ALSPAC). (2017). Hum. Mol. Genet. 24(8), 2201–2217 (2015). •• A very recent interesting interventional study showing 308 Kim YJ, Park SW, Kim TH et al. Genome-wide methylation that pollution and intrabronchial allergen delivery can profiling of the bronchial mucosa of asthmatics: relationship have a major impact on the methylome of the bronchial to atopy. BMC Med. Genet. 14, 39 (2013). tissue depending on time between allergic insults and 309 Jiang R, Jones MJ, Sava F, Kobor MS, Carlsten C. Short-term order. diesel exhaust inhalation in a controlled human crossover 311 Rastogi D, Suzuki M, Greally JM. Differential epigenome- study is associated with changes in DNA methylation of wide DNA methylation patterns in childhood obesity- circulating mononuclear cells in asthmatics. Part. Fibre associated asthma. Sci. Rep. 3, 2164 (2013). Toxicol. 11, 71 (2014).

future science group www.futuremedicine.com 571 SPECIAL FOCUS y Epigenetics of the immune system & alterations in inflammation & autoimmunity Research Article

For reprint orders, please contact: [email protected]

9 Research Article 2017/03/30 DNA methylation at diagnosis is associated with response to disease-modifying drugs in early rheumatoid arthritis

Epigenomics Aim: A proof-of-concept study to explore whether DNA methylation at first John R Glossop1,2, Nicola B diagnosis is associated with response to disease-modifying antirheumatic drugs Nixon2, Richard D Emes3,4, (DMARDs) in patients with early rheumatoid arthritis (RA). Patients & methods: Julius Sim5, Jon C Packham1,2, 1,2 DNA methylation was quantified in T-lymphocytes from 46 treatment-naive patients Derek L Mattey , William E Farrell1 & Anthony A Fryer*,1 using HumanMethylation450 BeadChips. Treatment response was determined in 6 1Guy Hilton Research Centre, Institute months using the European League Against Rheumatism (EULAR) response criteria. for Applied Clinical Sciences, Keele Results: Initial filtering identified 21 cytosine-phosphate-guanines (CpGs) that were University, Thornburrow Drive, Hartshill, differentially methylated between responders and nonresponders. After conservative Stoke-on-Trent, Staffordshire, ST4 7QB, adjustment for multiple testing, six sites remained statistically significant, of which UK 2 four showed high sensitivity and/or specificity (≥75%) for response to treatment. Haywood Rheumatology Centre, Haywood Hospital, High Lane, Burslem, Moreover, methylation at two sites in combination was the strongest factor associated Stoke-on-Trent, Staffordshire, ST6 7AG, with response (80.0% sensitivity, 90.9% specificity, AUC 0.85). Conclusion: DNA UK methylation at diagnosis is associated with disease-modifying antirheumatic drug 3School of Veterinary Medicine & treatment response in early RA. Science, University of Nottingham, Sutton Bonington Campus, Sutton Bonington, Leicestershire, LE12 5RD, UK First draft submitted: 14 April 2016; Accepted for publication: 12 October 2016; 4Advanced Data Analysis Centre, Published online: 25 November 2016 University of Nottingham, Sutton Bonington Campus, Sutton Bonington, Keywords: disease activity score with 28 joint counts (DAS28) • disease-modifying Leicestershire, LE12 5RD, UK ­antirheumatic drugs (DMARDs) • DNA methylation • early rheumatoid arthritis • Illumina 5School of Health & Rehabilitation, Keele 450K array • T-lymphocyte • treatment response University, Staffordshire, ST5 5BG, UK *Author for correspondence: Tel.: +44 1782 674245 Rheumatoid arthritis (RA) is a chronic Evidence also suggests that disease modify- Fax: +44 1782 747319 [email protected] inflammatory joint disease of autoimmune ing agents such as methotrexate may influ- 4 origin that affects 0.5–1.0% of the adult pop- ence DNA methylation [9,10]. Moreover, ulation [1,2]. Treatment of patients with centers methylation ­status as a potential biomarker on the use of a variety of synthetic disease- associated with response to therapy has been modifying antirheumatic drugs (DMARDs). demonstrated in other conditions [11] and 2017 Methotrexate is the first-line DMARD of proposed for use in RA by several investi- choice for the treatment and management of gators [12,13]. DNA is methylated through RA, prescribed as monotherapy or in combi- enzymatic conversion of cytosine to meth- nation with other DMARDs. Although these ylcytosine; this occurring almost invariably agents are efficacious for the treatment of at cytosine-phosphate-guanine (CpG) sites. RA [3–5], clinically meaningful responses are In the context of promoter-associated sites, not observed in all patients and a significant methylation is associated with transcriptional proportion remain refractory to treatment. repression and gene silencing [14] . In RA, A substantial body of literature supports alterations to the DNA methylome are appar- an important role for epigenetic dysregu- ent in multiple cell types important in the lation, including of DNA methylation, in disease process, including peripheral blood- part of the pathogenesis of RA (reviewed in [6–8]). derived mononuclear cells, lymphocytes

epi-2016-0144 © 2017 Future Medicine Ltd Epigenomics (2017) 9(4), 419–428 ISSN 1750-1911 419/10.2217 Research Article Glossop, Nixon, Emes et al.

Table 1. Demographic and clinical characteristics at baseline inflammatory and other mediators [18–20] . However, for the cohort of 46 treatment-naive patients with early no single factor or combination of factors have thus rheumatoid arthritis. far proven to be accurate and reliable in determining which patients will respond to DMARD therapy. Number 46 Our aim therefore, in this proof-of-concept study, Male/female, number (%) 16/30 (34.8/65.2) was to determine whether genome-wide DNA meth- Age, mean ± SD (years) 57.7 ± 13.9 ylation profiles at first diagnosis are associated with RF positive, number (%)†§ 23 (52.3) response to treatment with conventional DMARDs (as ACPA positive, number (%)‡§ 22 (48.9) determined by improvement in disease activity using the validated European League Against Rheumatism DAS28, mean ± SD 5.29 ± 1.4 [EULAR] response criteria) in a typical population ESR, mean ± SD 30.1 ± 23.7 of newly-diagnosed, treatment-naive patients with Corticosteroids, number (%) 45 (97.8) RA. As in our previous work, we examined methyla- Starting DMARD, number (%)¶ tion in purified T-lymphocyte populations, cells that Methotrexate (MTX) 43 (93.5) are instrumental in the disease process and chronic inflammation [21], and for which relationships with Hydroxychloroquine (HCQ) 29 (63.0) disease activity have recently been described [22–24]. Sulphasalazine (SSZ) 23 (50.0) Starting treatment regimens, number (%) Patients & methods Monotherapy (MTX)# 15 (32.6) Study population Triple therapy (MTX + HCQ + SSZ) 20 (43.5) A prospective cohort of 46 Caucasian patients attending the early synovitis clinic at the Haywood Rheumatol- Dual therapy (two of MTX, HCQ and SSZ) 10 (21.7) ogy Centre in Stoke-on-Trent (UK) and presenting with †Of 44 patients (data unavailable for two patients). ‡Of 45 patients (data unavailable for one patient). symptomatic inflammatory arthritis suspected to be §26/45 (57.8%) patients were positive for ACPA/RF (data unavailable for one patient). RA was recruited. All patients were subsequently classi- ¶The total number of patients starting treatment with a given DMARD, whether received as monotherapy or in combination with other DMARDs. fied as having RA, according to the 2010 ACR/EULAR #One further patient started monotherapy with hydroxychloroquine. classification criteria, by a consultant rheumatolo- ACPA: Anticitrullinated peptide antibody; DAS28: Disease activity score with 28-joint count; DMARD: Disease-modifying antirheumatic drug; ESR: Erythrocyte sedimentation gist [25]. No patients had been treated with DMARDs rate; RA: Rheumatoid arthritis; RF: Rheumatoid factor. or biological agents at the time of recruitment. Clini- cal data collected at baseline included disease activity, and joint-derived fibroblasts. Recently, we were the erythrocyte sedimentation rate, rheumatoid factor and first to define disease-associated methylation changes anticitrullinated peptide antibodies. Demographic and that were distinct to individual T- and B-lymphocyte clinical characteristics are presented in Table 1. At diag- populations [15] . Moreover, we reported methylation nosis with RA, all patients began treatment with one differences in these lymphocyte populations in treat- or more DMARDs (methotrexate, hydroxychloroquine ment-naive patients at first RA diagnosis [16] . While and sulphasalazine) and the majority received paren- providing evidence for a role in the development of the teral corticosteroids, solely for the clinical management disease, our findings support DNA methylation profil- of RA and as directed by a consultant rheumatologist. ing at diagnosis as a potential source of biomarkers for Patients were followed for 6-months and remained on response to treatment in RA. treatment throughout. The study was approved by the It is clear that the ability to identify which patients East Midlands (Derby) Research Ethics Committee. will respond to treatment offers considerable benefits All patients provided written informed consent. for the management of RA. For example, it would facil- Disease activity was determined at recruitment itate rapid dose-escalation and reduce time to effective (prior to initiation of DMARD therapy) and after 3 response in those likely to be poor responders to tra- and 6 months of treatment using the disease activ- ditional regimens, and avoid unwanted side-effects ity score with 28-joint counts (DAS28) [26], though in those likely to show an effective response to lower data at 3 months was excluded from further analysis doses or monotherapy. These benefits are all the more due to the known short-term effect of corticosteroid important given evidence that response to first treat- treatment on DAS28 scores. DAS28 scores range from ment with disease-modifying agents is strongly asso- 0–10: a score >5.1 indicates high disease activity while ciated with long-term outcome in these patients [17] . one of ≤3.2 denotes low disease activity. Response to The search for biomarkers associated with response has treatment was determined in 6 months according to encompassed demographic and clinical factors as well the DAS28-based EULAR response criteria [26–28], as genetic associations and expression profiling of pro- which evaluate response in patients with RA based

420 Epigenomics (2017) 9(4) future science group DNA methylation and treatment response in early RA Research Article on a composite categorization incorporating both Data analysis change in DAS28 from baseline (ΔDAS28) and final Array data (idat files) were processed and analyzed using absolute DAS28 score. Specifically, these criteria clas- the Bioconductor package Minfi [34]. We removed from sify response as ‘good’ (ΔDAS28 >1.2, current DAS28 analysis all CpGs with a detection p-value >0.01 in any ≤3.2), ‘moderate’ (ΔDAS28 >1.2, current DAS28 >3.2, one or more of the 46 samples and all probes targeting or ΔDAS28 >0.6–1.2, current DAS28 ≤5.1) and ‘no’ sites on the X and Y chromosomes (a total of 12,295 (ΔDAS28 ≤0.6, or ΔDAS28 >0.6–1.2, current DAS28 CpGs). Data were normalized by Subset-quantile >5.1) [28]. According to these criteria, responders were Within Array Normalization (SWAN), as described defined as patients with a ‘good’ or ‘moderate’ response by Maksimovic et al. [35], and multidimensional scaling to treatment, and nonresponders as patients with ‘no’ plots were examined to confirm appropriate adjustment response to treatment. for potential confounding due to batch effects (process- ing date, array position and slide). Isolation of T-lymphocytes To identify methylation differences associated Fresh peripheral blood samples (35 ml, EDTA) were with treatment response, patients were stratified into collected from each patient at baseline, prior to the responders and nonresponders. CpGs showing altered initiation of treatment. CD3+ T-lymphocytes were methylation between the two groups were identified isolated from mononuclear cell preparations using using the ‘dmpFinder’ function in Minfi. This func- positive selection with magnetic microbeads (MACS® tion performs an F-test to compare groups and was Separation System; Miltenyi Biotec, Surrey, UK). used with logit-transformed β-values (M-values), as We have previously shown this method to yield high- recommended by Du et al. [36]. p-values <0.05 were purity T-lymphocyte populations (mean ≥99%) in RA considered statistically significant and, together with patients [15] . Genomic DNA was extracted using an a mean β-value difference ≥0.1 between the groups, AllPrep DNA/RNA/miRNA Universal kit (Qiagen, were used as an initial screening tool to identify sites Manchester, UK) and stored at -20°C prior to use. displaying differential methylation. Two further filter- ing steps were subsequently applied to identify differ- Genome-wide DNA methylation profiling entially methylated CpGs as those sites where: at least DNA methylation was quantified at >480,000 CpG sites two-thirds of nonresponders showed a β-value differ- using the HumanMethylation450 BeadChip (Illumina, ence ≥0.1 relative to the responder mean; and at least Inc., Essex, UK; hereafter referred to as ‘array’). Details two-thirds of responders displayed a β-value equal to of array design and coverage have been described else- or in excess of the responder mean. Filtering criteria where [29]. Genomic DNA samples (n = 46) were treated are summarized in Figure 1. We then applied a Bon- with sodium bisulfite using an EZ DNA Methylation ferroni adjustment at stage 5, based on comparisons Kit (Zymo Research via Cambridge Bioscience, Cam- ­conducted using the final 21 CpGs identified. bridgeshire, UK) and subsequently were hybridized to The McNemar test was used to examine the inci- arrays according to manufacturer recommended pro- dence of patients with moderate/high disease activity tocols, as previously described (performed by Hologic between baseline and 6 months. The association of Tepnel Pharma Services, Lancashire, UK) [30]. All baseline methylation status with treatment response samples passed stringent internal array quality control, was determined by calculating sensitivity, specificity, including sample-independent (e.g., staining, hybrid- positive-predictive value (PPV) and negative-predic- ization) and sample-dependent (e.g., bisulfite conver- tive value (NPV), and by examining receiver operat- sion) controls. Methylation at individual CpG sites is ing characteristic (ROC) area under the curve (AUC) reported as a β-value ranging from 0 to 1 (­unmethylated plots. ROC curves were constructed based on logistic to fully methylated, respectively) [29]. regression analysis with response to treatment catego- rized as no response versus moderate/good response as Sodium bisulfite pyrosequencing described above. Analyses were performed using Stata Array candidates were independently validated by 12.0 (Intercooled; Stata Corporation, TX, USA) and bisulfite pyrosequencing using a PyroMark Q24 considering p-values <0.05 as statistically significant. instrument and analysis software (Qiagen), as we have previously described [15,30]. Briefly, fresh genomic DNA Results aliquots were sodium bisulfite-converted and ampli- Characteristics of the patients fied using whole genome amplification [30,31] . Thereaf- Table 1 summarizes the demographic and clinical ter, Touchdown PCR [32,33] was used to prepare PCR characteristics for the RA patients at recruitment. amplicons containing CpGs of interest. Assay details Most patients (43/46, 93.5%) started treatment with are provided in Supplementary Table 1. MTX, either as monotherapy or in combination with

future science group www.futuremedicine.com 421 Research Article Glossop, Nixon, Emes et al.

other DMARDs. The majority of patients (33/46, ment, 28/46 (60.9%) patients had moderate/high dis- 71.7%) remained on their indicated starting DMARD ease activity (p < 0.001 vs baseline, McNemar test), regimen throughout the course of the study. Of the with approximately two-thirds (63.0%) achieving an remaining patients, all but two introduced or discon- improvement in DAS28 ≥1.2. Classifying response by tinued a single DMARD on one occasion during the the EULAR response criteria, the number of patients 6-month follow-up period. achieving a good, moderate and no response to treat- ment in 6 months was 16 (34.8%), 19 (41.3%) and 11 Disease activity & treatment response (23.9%), respectively. On this basis, 76.1% (35/46) of Moderate or high disease activity (DAS28 >3.2) was patients were classified as responders and the remain- present in 43/46 (93.5%) patients at recruitment der as nonresponders. Details of baseline character- (three patients had low disease activity, with DAS28 istics and 6-month treatment regimens for the two scores of 2.27, 2.66 and 3.18). After 6 months of treat- groups are presented in Supplementary Table 2.

482,421 CpGs

Remove CpGs: 1 1) With a detection p-value ≥ 0.01; 2) Located on the XY chromosomes

Data normalization 470,126

Retain CpGs with a signicant (p < 0.05) 2 difference in mean β-value between nonresponders and responders

67,672

Retain CpGs with a difference in mean 3 β-value ≥ 0.1 between nonresponders and responders

269

Retain CpGs where ≥7 out of 11 (64%) 4 nonresponders show a β-value difference ≥0.1 relative to the responder mean

83

Retain CpGs where ≥23 out of 35 (66%) 5 responders have a β-value equal to or in excess of the responder group mean

21

Figure 1. Filtering criteria for identification of cytosine-phosphate-guanines differentially methylated at baseline (pretreatment) between treatment responders and nonresponders in patients with early rheumatoid arthritis. The starting number of cytosine-phosphate-guanines (CpGs) indicated (482,421) is the total number of CpGs on the methylation array platform. Following initial processing (step 1), data were normalized using SWAN [35], implemented in the Bioconductor package Minfi [34]. Numbers in the figure indicate the number of CpGs remaining at each successive step. CpG: Cytosine-phosphate-guanine; SWAN: Subset-quantile within array normalization.

422 Epigenomics (2017) 9(4) future science group DNA methylation and treatment response in early RA Research Article

Relationship between DNA methylation & were responders (14 good and 14 moderate response; treatment response right chart, Figure 3). In contrast, all four patients fail- Use of the robust filtering steps described in the Meth- ing to satisfy either cut-off were nonresponders (left ods section and shown in Figure 1 identified 269 CpGs chart, Figure 3). The strength of the association of the with a statistically significant difference in mean meth- CpG-2 + CpG-3 combination with response was also ylation β-value ≥0.1 between responders and nonre- reflected in a ROC AUC of 0.85 (95% CI: 0.71–0.94). sponders. Moreover, for a subset of 21 sites, methyla- tion differences were present in at least two-thirds of Discussion the individual patients within each group (full anno- This is the first study to examine the link between tation for these 21 sites is provided in Supplementary DNA methylation and first-line treatment response in Table 3). The majority of these sites were hypermethyl- RA. Using a prospective cohort of patients recruited ated in responders (16/21, 76.2%), were linked with a at first diagnosis and prior to the initiation of treat- gene (15/21, 71.4%) and were associated with a CpG ment, our data indicate that baseline DNA methyla- island and/or the surrounding shores/shelves (13/21, tion levels for a discrete subset of sites are significantly 61.9%). associated with response to treatment with disease- To refine these sites further, we applied a conserva- modifying agents. The methylation status at two spe- tive Bonferroni adjustment for multiple testing, based cific sites assessed in combination, and which indepen- on the 21 comparisons undertaken. This revealed six dently were associated with response, proved to be the CpGs for which the methylation differences between ­strongest factor associated with treatment response. responders and nonresponders remained statistically Since early, effective intervention in RA reduces significant (padj <0.05; Supplementary Table 3). For disease activity and inflammation, and improves long- each of these six CpGs, we plotted methylation against term outcome [37–40], identification of baseline factors treatment response to determine a percentage meth- associated with treatment response has been a prior- ylation cut-off that in each case provided the greatest ity. However, examination of a broad range of clinical, discrimination between patients that responded to molecular and genetic factors has not produced defini- treatment and those that did not. Examples of two dif- tive biomarkers [18,19]. Our findings now provide the ferentially methylated CpGs are presented in Figure 2. first evidence that epigenetic profiling, in this case of We also calculated the corresponding sensitivity and DNA methylation, may have significant value in iden- specificity for each site to assess the association of tifying which patients with RA may respond to first- methylation status with response. Using this approach, line DMARD treatment. Furthermore, DNA meth- and as shown in Table 2, four sites were identified with ylation is an attractive biomarker since it is typically a sensitivity and/or specificity ≥75% for discrimina- stable over time, is minimally affected by short-term tion between responders and nonresponders. Most stimuli and is readily measured [12] . The potential util- notably, hypermethylation of CpG-2 and hypometh- ity of methylation profiling is further supported by a ylation of CpG-3 (shown in Figure 2 and validated very recently reported association between differen- by Pyrosequencing in Supplementary Figure 1) each tial DNA methylation and response to second-line demonstrated a sensitivity and PPV of approximately ­anti-TNF therapy in RA [41] . 90%, although the corresponding specificity and NPV We were unable to formally examine the indepen- were lower (63.6 and 70.0%, and 63.6 and 63.6%, dence of the CpG-2 + CpG-3 association with treat- for CpG-2 and CpG-3, respectively). Using ROC ment response in this proof-of-concept study. How- curve analysis to further evaluate the association with ever, a preliminary assessment using our data suggested response, CpG-2 and CpG-3 also demonstrated the that it was independent of baseline clinical variables highest AUC values (0.78 and 0.76, respectively). including disease activity, autoantibodies and systemic inflammatory markers, which individually did not Combinations of CpGs associated with appear to be associated with response. This would be treatment response in agreement with the main body of literature, which Focusing on the four sites identified above, we next indicates that erythrocyte sedimentation rate, rheu- examined the ability to discriminate between respond- matoid factor and anticitrullinated peptide antibod- ers and nonresponders for each of the six possible ies are not independently associated with response to pairs of sites. The combination of hypermethylation methotrexate and/or other DMARDs [reviewed in of CpG-2 and hypomethylation of CpG-3 demon- 18]. Although not reported by all studies [42], evidence strated the best overall performance with a sensitivity does indicate that male sex is associated with a bet- of 80.0% and specificity of 90.9% (Table 2). As shown ter response to methotrexate [43–45]. Our data suggest in Figure 3, 28 out of 29 patients with this combination a possible trend towards better response in males (p <

future science group www.futuremedicine.com 423 Research Article Glossop, Nixon, Emes et al.

1.0 1.0 0.9 0.9 0.8 0.8 0.7 0.7 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 Methylation β -value Methylation β -value 0.2 0.2 0.1 0.1

No Moderate Good No Moderate Good Response Response

Figure 2. Pretreatment methylation status discriminates responders and nonresponders in patients with early rheumatoid arthritis. In (A) CpG-2 (cg03018489) and (B) CpG-3 (cg14345882), nonresponders (n = 11) and responders (n = 35) are depicted by open circles and filled triangles, respectively, and where responders are divided into those showing a moderate (center, n = 19) and good (right, n = 16) response to treatment. Good, moderate and no response categories are defined in the EULAR response criteria [23–25]. The horizontal dashed line indicates the methylation cut-off for distinguishing between responders and nonresponders, and the short horizontal bar in each group indicates the mean value. CpG: Cytosine-phosphate-guanine; EULAR: European League Against Rheumatism.

0.1), which may reflect treatment with methotrexate ADAMTSL2 (CpG-2), which encodes a disintegrin for over 90% of the patients studied. and metalloproteinase with thrombospondin motif- The CpG-2 + CpG-3 combination, which we like protein, and in BTN3A2 (CpG-3), which encodes identified as the strongest independent factor asso- a butyrophilin family member. Although the function ciated with treatment response, comprises sites in of the ADAMTSL2-encoded protein has not been fully

Table 2. Association of baseline methylation status with treatment response in patients with early rheumatoid arthritis.† CpG ID Methylation Sensitivity Specificity PPV (%) NPV (%) ROC AUC (95% CI) in responders: (%) (%) hyper/hypo Individual sites CpG-1 (cg07225509) Hyper 77.1 72.7 90.0 50.0 0.75 (0.59–0.86) CpG-2 (cg03018489) Hyper 91.4 63.6 88.9 70.0 0.78 (0.64–0.89) CpG-3 (cg14345882) Hypo 88.6 63.6 88.6 63.6 0.76 (0.61–0.87) CpG-4 (cg23974730) Hypo 82.9 63.6 87.9 53.9 0.73 (0.59–0.86) Combinations CpG-1 + CpG-2 Hyper/hyper 71.4 90.9 96.2 50.0 0.81 (0.66–0.91) CpG-1 + CpG-3 Hyper/hypo 65.7 81.8 92.0 42.9 0.74 (0.59–0.86) CpG-1 + CpG-4 Hyper/hypo 60.0 90.9 95.5 41.7 0.75 (0.61–0.87) CpG-2 + CpG-3 Hyper/hypo 80.0 90.9 96.6 58.8 0.85 (0.71–0.94) CpG-2 + CpG-4 Hyper/hypo 77.1 72.7 90.0 50.0 0.75 (0.59–0.86) CpG-3 + CpG-4 Hypo/hypo 74.3 90.9 96.3 52.6 0.83 (0.69–0.92) †Of the six CpGs identified as significantly differentially methylated between responders and nonresponders (see main text), shown are the four CpGs with a sensitivity and/or specificity ≥75% and that showed most promise for discriminating between responders and nonresponders. Also shown are the six possible CpG pairs derived from these four sites. All individual sites and combinations shown were significantly associated with treatment response (p < 0.05, Fisher’s exact test). The CpG-2 + CpG-3 combination displayed the best overall performance (p < 0.001; bold font). CpG: Cytosine-phosphate-guanine; Hyper: Hypermethylated; Hypo: Hypomethylated; NPV: Negative-predictive value; PPV: Positive-predictive value; ROC AUC: Receiver operating characteristic area under the curve.

424 Epigenomics (2017) 9(4) future science group DNA methylation and treatment response in early RA Research Article

Response No Moderate Good

Hypo/hyper Hypo/hypo Hyper/hypo (n = 4) Hyper/hyper (n = 29) (n = 13) CpG-2 + CpG-3 methylation

Figure 3. Pretreatment methylation status at two cytosine-phosphate-guanine sites in combination is associated with response to treatment in patients with early rheumatoid arthritis. For CpG-2 (cg03018489) and CpG-3 (cg14345882) methylation status was defined as hypermethylated (above) or hypomethylated (below) relative to a cut-off of 60% and 20%, respectively. Shown on the x-axis are the four possible methylation combinations, with methylation status of CpG-2 given first and of CpG-3 given second, as indicated (the two combinations in which only one CpG satisfied the cut-off value are grouped together (center chart)). Each chart depicts the proportion of patients achieving a good (white), moderate (striped) and no response (dark gray) to treatment, stratified by methylation status for the CpG-2/CpG-3 combination. CpG: Cytosine-phosphate-guanine; Hyper: Hypermethylated; Hypo: Hypomethylated. determined, evidence supports a role in the regulation which on methylation has been suggested by several of TGF-β [46]. TGF-β is a pleiotropic cytokine with groups [9,10,53,54]. important immunoregulatory functions [47,48], which Although our proof-of-concept study is the first of its is implicated in RA synovial pathology [49]. Butyro- kind in RA, a limitation of our work was the relatively philins are transmembrane proteins that share struc- small number of patients that we were able to recruit. tural similarities with B7 co-stimulatory molecules In an attempt to address this, we used a number of and are emerging as novel regulators of T-lymphocyte sequential filtering steps to identify sites differentially function and immune responses [50,51] . methylated between responders and nonresponders to We focused on DNA methylation factors associated treatment. Furthermore, for the two CpGs comprising with response in the context of DMARD treatment the strongest biomarker associated with response, we strategies that reflected standard clinical practice. validated the array data by also quantifying methyla- Both responder and nonresponder groups included tion using an independent method (Pyrosequencing). patients receiving methotrexate monotherapy and This significantly reduces the risk of type I errors asso- patients receiving combination therapy, the propor- ciated with genome-wide approaches. However, we rec- tions of which were not significantly different either ognise that an important next step will be to confirm at baseline or at the 6-month follow-up assessment our findings and determine the true predictive value of (Supplementary Table 2). Importantly, methylation this biomarker in larger, independent patient cohorts. at two CpGs in combination was strongly associated with treatment response despite the limited variation Conclusion in treatment regimens, supporting its potential utility In conclusion, we report the identification of a novel as a marker of response at diagnosis in a real-world DNA methylation combination that is associated with clinical setting. Furthermore, we purposefully used response to treatment with conventional disease-mod- the EULAR criteria as the response measure in this ifying drugs in newly diagnosed patients with RA. study as these are universally accepted and encom- While our findings will require verification in larger, pass both improvement in disease activity over time independent early RA cohorts, they provide the first and end-point disease activity. Reassuringly, the evidence to support epigenetic profiling as a novel proportion of responders in this study is consistent approach to identifying biomarkers associated with with previous reports using these criteria [44,52]. By response to DMARD therapy. Ultimately, this has the quantifying methylation at baseline, we are also able potential to inform clinical management and patient to exclude potential confounding associated with care, towards the goal of a stratified, personalized DMARDs, including methotrexate, an impact of ­medicine approach to treatment.

future science group www.futuremedicine.com 425 Research Article Glossop, Nixon, Emes et al.

Acknowledgements matism Research and Development Foundation. The authors The authors would like to thank the patients who partici- have no other relevant affiliations or financial involvement pated in the study. They also thank J Turner, C Thwaites and with any organization or entity with a financial interest in or fi- M ­Dishman for assistance with the collection of clinical data. nancial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed. Financial & competing interests disclosure No writing assistance was utilized in the production of this This work was supported by funding from the Haywood Rheu- manuscript.

Summary points Background • Newly diagnosed patients with rheumatoid arthritis (RA) demonstrate variability of response to treatment with disease-modifying antirheumatic drugs (DMARDs). • To date, no definitive biomarkers associated with response have been identified. • This proof-of-concept study explored whether DNA methylation at first diagnosis is associated with response to treatment with DMARDs in patients with treatment-naive early RA. Patients & methods • HumanMethylation450 BeadChips were used to quantify genome-wide DNA methylation at diagnosis in T-lymphocytes from 46 treatment-naive patients with early RA. • Response to DMARD treatment was determined at 6 months using the DAS28-based EULAR response criteria. Sensitivity, specificity and receiver operating characteristic AUC data were used to assess associations of baseline methylation with treatment response. Results • At 6 months, the numbers of patients achieving a good/moderate/no response to treatment were 16/19/11 (35/41/24%), respectively. • Array analysis identified 21 CpGs displaying methylation differences between responders and nonresponders,

of which four statistically significant sites (padj <0.05, Bonferroni) showed high sensitivity and/or specificity ≥75% for treatment response. • Methylation at two individual sites in combination (cg0301849 and cg14345882) was the strongest factor associated with response, with 80.0% sensitivity and 90.9% specificity (AUC 0.85). Twenty-eight of 29 patients with this combination were responders. Conclusion • DNA methylation of a novel CpG combination is associated with treatment response at first diagnosis in early RA patients prior to commencing treatment with DMARDs. • These findings provide the first evidence to support epigenetic profiling as a novel approach to identifying biomarkers associated with DMARD treatment response in RA. This may ultimately have the potential to inform clinical management and patient care.

placebo-controlled trial. Arthritis Rheum. 46(5), 1164–1170 References (2002). Papers of special note have been highlighted as: • of interest; •• of considerable interest 5 Nurmohamed MT, Dijkmans BA. Efficacy, tolerability and cost effectiveness of disease-modifying antirheumatic drugs 1 Symmons D, Turner G, Webb R et al. The prevalence and biologic agents in rheumatoid arthritis. Drugs 65(5), of rheumatoid arthritis in the United Kingdom: new 661–694 (2005). estimates for a new century. Rheumatology 41(7), 793–800 (2002). 6 Bottini N, Firestein GS. Epigenetics in rheumatoid arthritis: a primer for rheumatologists. Curr. Rheumatol. Rep. 15(11), 2 Myasoedova E, Crowson CS, Kremers HM, Therneau TM, 372 (2013). Gabriel SE. Is the incidence of rheumatoid arthritis rising? Results from Olmsted County, Minnesota, 1955–2007. 7 Glant TT, Mikecz K, Rauch TA. Epigenetics in the Arthritis Rheum. 62(6), 1576–1582 (2010). pathogenesis of rheumatoid arthritis. BMC Med. 12, 35 (2014). 3 O’Dell JR, Haire CE, Erikson N et al. Treatment of rheumatoid arthritis with methotrexate alone, sulfasalazine 8 Klein K, Gay S. Epigenetic modifications in rheumatoid and hydroxychloroquine, or a combination of all three arthritis, a review. Curr. Opin. Pharmacol. 13(3), 420–425 medications. N. Engl. J. Med. 334(20), 1287–1291 (1996). (2013). 4 O’Dell JR, Leff R, Paulsen G et al. Treatment of rheumatoid 9 Kim YI, Logan JW, Mason JB, Roubenoff R. DNA arthritis with methotrexate and hydroxychloroquine, hypomethylation in inflammatory arthritis: reversal methotrexate and sulfasalazine, or a combination of the three with methotrexate. J. Lab. Clin. Med. 128(2), 165–172 medications: results of a two-year, randomized, double-blind, (1996).

426 Epigenomics (2017) 9(4) future science group DNA methylation and treatment response in early RA Research Article

10 de Andres MC, Perez-Pampin E, Calaza M et al. 23 Carvalheiro H, Duarte C, Silva-Cardoso S, da Silva JA, Assessment of global DNA methylation in peripheral Souto-Carneiro MM. CD8+ T cell profiles in patients with blood cell subpopulations of early rheumatoid arthritis rheumatoid arthritis and their relationship to disease activity. before and after methotrexate. Arthritis Res. Ther. 17, 233 Arthritis Rheumatol. 67(2), 363–371 (2015). (2015). 24 Kuuliala K, Kuuliala A, Koivuniemi R et al. Constitutive • First rheumatoid arthritis (RA) study of global DNA STAT3 phosphorylation in circulating CD4+ T lymphocytes methylation levels in distinct blood cell subpopulations and associates with disease activity and treatment response apparent reversal with methotrexate treatment. in recent-onset rheumatoid arthritis. PLoS ONE 10(9), e0137385 (2015). 11 Esteller M, Garcia-Foncillas J, Andion E et al. Inactivation of the DNA-repair gene MGMT and the clinical response 25 Aletaha D, Neogi T, Silman AJ et al. 2010 Rheumatoid of gliomas to alkylating agents. N. Engl. J. Med. 343(19), arthritis classification criteria: an American College of 1350–1354 (2000). Rheumatology/European League Against Rheumatism collaborative initiative. Ann. Rheum. Dis. 69(9), 1580 –1588 12 Plant D, Wilson AG, Barton A. Genetic and epigenetic (2010). predictors of responsiveness to treatment in RA. Nat. Rev. Rheumatol. 10(6), 329–337 (2014). 26 Prevoo ML, van ‘t Hof MA, Kuper HH, van Leeuwen MA, van de Putte LB, van Riel PL. Modified disease activity 13 Cribbs A, Feldmann M, Oppermann U. Towards an scores that include twenty-eight-joint counts. Development understanding of the role of DNA methylation in rheumatoid and validation in a prospective longitudinal study of patients arthritis: therapeutic and diagnostic implications. Ther. Adv. with rheumatoid arthritis. Arthritis Rheum. 38(1), 44–48 Musculoskelet. Dis. 7(5), 206–219 (2015). (1995). 14 Bird A. DNA methylation patterns and epigenetic memory. 27 van Gestel AM, Prevoo ML, van ‘t Hof MA, van Rijswijk Genes Dev. 16(1), 6–21 (2002). MH, van de Putte LB, van Riel PL. Development and 15 Glossop JR, Emes RD, Nixon NB et al. Genome-wide DNA validation of the European League Against Rheumatism methylation profiling in rheumatoid arthritis identifies response criteria for rheumatoid arthritis. Comparison with disease-associated methylation changes that are distinct to the preliminary American College of Rheumatology and the individual T- and B-lymphocyte populations. Epigenetics World Health Organization/International League Against 9(9), 1228–1237 (2014). Rheumatism Criteria. Arthritis Rheum. 39(1), 34–40 (1996). 16 Glossop JR, Emes RD, Nixon NB et al. Genome-wide 28 van Gestel AM, Haagsma CJ, van Riel PL. Validation of profiling in treatment-naive early rheumatoid arthritis rheumatoid arthritis improvement criteria that include reveals DNA methylome changes in T- and B-lymphocytes. simplified joint counts.Arthritis Rheum. 41(10), 1845–1850 Epigenomics 8(2), 209–224 (2016). (1998). • First DNA methylation array study in newly diagnosed 29 Bibikova M, Barnes B, Tsan C et al. High density DNA RA patients naive for treatment with disease-modifying methylation array with single CpG resolution. Genomics antirheumatic drugs. 98(4), 288–295 (2011). 17 Farragher TM, Lunt M, Fu B, Bunn D, Symmons DP. Early 30 Glossop JR, Nixon NB, Emes RD et al. Epigenome-wide treatment with, and time receiving, first disease-modifying profiling identifies significant differences in DNA methylation antirheumatic drug predicts long-term function in patients between matched-pairs of T- and B-lymphocytes from healthy with inflammatory polyarthritis.Ann. Rheum. Dis. 69(4), individuals. Epigenetics 8(11), 1188–1197 (2013). 689–695 (2010). 31 Mill J, Yazdanpanah S, Guckel E, Ziegler S, Kaminsky Z, 18 Romao VC, Canhao H, Fonseca JE. Old drugs, old Petronis A. Whole genome amplification of sodium bisulfite- problems: where do we stand in the prediction of rheumatoid treated DNA allows the accurate estimate of methylated arthritis responsiveness to methotrexate and other synthetic cytosine density in limited DNA resources. Biotechniques DMARDs? BMC Med. 11, 17 (2013). 41(5), 603–607 (2006). • Discusses current evidence for predicting response to 32 Don RH, Cox PT, Wainwright BJ, Baker K, Mattick JS. disease-modifying drugs in RA. ‘Touchdown’ PCR to circumvent spurious priming during 19 Smith SL, Plant D, Eyre S, Barton A. The potential use of gene amplification.Nucleic Acids Res. 19(14), 4008 (1991). expression profiling: implications for predicting treatment 33 Haworth KE, Farrell WE, Emes RD et al. Combined response in rheumatoid arthritis. Ann. Rheum. Dis. 72(7), influence of gene-specific cord blood methylation and 1118–1124 (2013). maternal smoking habit on birth weight. Epigenomics 5(1), 20 Grabiec AM, Reedquist KA. The ascent of acetylation in 37–49 (2013). the epigenetics of rheumatoid arthritis. Nat. Rev. Rheumatol. 34 Aryee MJ, Jaffe AE, Corrada-Bravo H et al. Minfi: a flexible 9(5), 311–318 (2013). and comprehensive Bioconductor package for the analysis 21 Cope AP. T cells in rheumatoid arthritis. Arthritis Res. Ther. of Infinium DNA methylation microarrays.Bioinformatics 10(Suppl. 1), S1 (2008). 30(10), 1363–1369 (2014). 22 Peres RS, Liew FY, Talbot J et al. Low expression of CD39 35 Maksimovic J, Gordon L, Oshlack A. SWAN: Subset- on regulatory T cells as a biomarker for resistance to quantile within array normalization for illumina infinium methotrexate therapy in rheumatoid arthritis. Proc. Natl HumanMethylation450 BeadChips. Genome Biol. 13(6), Acad. Sci. USA 112(8), 2509–2514 (2015). R44 (2012).

future science group www.futuremedicine.com 427 Research Article Glossop, Nixon, Emes et al.

36 Du P, Zhang X, Huang CC et al. Comparison of Beta-value of the SWEFOT trial. Ann. Rheum. Dis. 70(3), 469–475 and M-value methods for quantifying methylation levels by (2011). microarray analysis. BMC Bioinformatics 11, 587 (2010). 45 Ma MH, Ibrahim F, Walker D et al. Remission in early 37 Finckh A, Liang MH, van Herckenrode CM, de Pablo rheumatoid arthritis: predicting treatment response. P. Long-term impact of early treatment on radiographic J. Rheumatol. 39(3), 470–475 (2012). progression in rheumatoid arthritis: a meta-analysis. Arthritis 46 Le Goff C, Morice-Picard F, Dagoneau N et al. ADAMTSL2 Rheum. 55(6), 864–872 (2006). mutations in geleophysic dysplasia demonstrate a role for 38 de Vries-Bouwstra JK, Goekoop-Ruiterman YP, Verpoort ADAMTS-like proteins in TGF-β bioavailability regulation. KN et al. Progression of joint damage in early rheumatoid Nat. Genet. 40(9), 1119–1123 (2008). arthritis: association with HLA-DRB1, rheumatoid factor, 47 Schmidt-Weber CB, Blaser K. Regulation and role of and anti-citrullinated protein antibodies in relation to transforming growth factor-beta in immune tolerance different treatment strategies. Arthritis Rheum. 58(5), induction and inflammation.Curr. Opin. Immunol. 16(6), 1293–1298 (2008). 709–716 (2004). 39 Klareskog L, Catrina AI, Paget S. Rheumatoid arthritis. 48 Li MO, Flavell RA. TGF-β: a master of all T cell trades. Cell Lancet 373(9664), 659–672 (2009). 134(3), 392–404 (2008). 40 Scirè CA, Verstappen SM, Mirjafari H et al. Reduction of 49 Pohlers D, Beyer A, Koczan D, Wilhelm T, Thiesen HJ, long-term disability in inflammatory polyarthritis by early Kinne RW. Constitutive upregulation of the transforming and persistent suppression of joint inflammation: results from growth factor-β pathway in rheumatoid arthritis synovial the Norfolk Arthritis Register. Arthritis Care Res. (Hoboken) fibroblasts.Arthritis Res. Ther. 9(3), R59 (2007). 63(7), 945–952 (2011). 50 Afrache H, Gouret P, Ainouche S, Pontarotti P, Olive D. 41 Plant D, Webster A, Nair N et al. Differential methylation The butyrophilin (BTN) gene family: from milk fat to the as a biomarker of response to etanercept in patients with regulation of the immune response. Immunogenetics 64(11), rheumatoid arthritis. Arthritis Rheumatol. 68(6), 1353–1360 781–794 (2012). (2016). 51 Abeler-Dorner L, Swamy M, Williams G, Hayday AC, Bas • Association between differential DNA methylation and A. Butyrophilins: an emerging family of immune regulators. response to second-line anti-TNF therapy in RA. Trends Immunol. 33(1), 34–41 (2012). 42 Hider SL, Silman AJ, Thomson W, Lunt M, Bunn D, 52 Svensson B, Schaufelberger C, Teleman A, Theander J. Symmons DP. Can clinical factors at presentation be used to Remission and response to early treatment of RA assessed predict outcome of treatment with methotrexate in patients by the disease activity score. Rheumatology (Oxford) 39(9), with early inflammatory polyarthritis?Ann. Rheum. Dis. 1031–1036 (2009). 68(1), 57–62 (2009). 53 Ellis JA, Munro JE, Chavez RA et al. Genome-scale case- 43 Wessels JA, van der Kooij SM, le Cessie S et al. A control analysis of CD4+ T-cell DNA methylation in juvenile clinical pharmacogenetic model to predict the efficacy of idiopathic arthritis reveals potential targets involved in methotrexate monotherapy in recent-onset rheumatoid disease. Clin. Epigenetics 4(1), 20 (2012). arthritis. Arthritis Rheum. 56(6), 1765–1775 (2007). 54 Cribbs AP, Kennedy A, Penn H et al. Methotrexate restores 44 Saevarsdottir S, Wallin H, Seddighzadeh M et al. Predictors regulatory T cell function through demethylation of the of response to methotrexate in early DMARD naive foxp3 upstream enhancer in patients with rheumatoid rheumatoid arthritis: results from the initial open-label phase arthritis. Arthritis Rheumatol. 67(5), 1182–1192 (2015).

428 Epigenomics (2017) 9(4) future science group