Smoking-Associated DNA Methylation and Offspring Caries Experience: Findings from the GUSTO Study

Abstract

Self-reported smoking during pregnancy is subject to measurement error, bringing into question previously reported associations with offspring caries experience. We evaluated whether a previously identified prenatal smoking-associated DNA methylation (DNAm) signature, and separately gestational smoking indexed through self-report and plasma cotinine, was associated with offspring caries. The Growing Up in Singapore Towards healthy Outcomes (GUSTO) cohort (n = 577 mother–child dyads, recruited from June 2009–September 2010) ascertained gestational smoking exposure (by questionnaires and plasma cotinine), DNAm (using umbilical cord tissue), and child dental caries (by clinical examinations at age 3 y). We used zero-inflated Poisson regression to evaluate whether a weighted smoking-associated DNAm risk score (wMRS_DNAm) was associated with the count of tooth surface caries, adjusted for maternal age, education, ethnicity, breastfeeding, preterm status, child sex, and toothbrushing frequency. Of the women, 53% were never smokers and 3% were heavy smokers; children had a mean (standard deviation) of 2.47 (5.20) decayed, missing, and filled surfaces. The mean wMRS_DNAm for the 16 CpGs included was −0.21 (0.02), with an interquartile range of 0.02. Each additional IQR of wMRS_DNAm was associated with a 20% higher adjusted caries experience risk, relative risk (RR) (95% confidence interval [CI]) = 1.20 (1.10, 1.31), and a lower adjusted odds of excess zeros, odds ratio (95% CI) = 0.91 (0.70, 1.17). Children of mothers who smoked during gestation (based on self-report/cotinine) also had higher adjusted caries experience risk: relative risk (RR) (95% CI) = 1.31 (1.11, 1.55) and RR (95% CI) = 1.12 (0.81, 1.54) for light and heavy smoking, respectively. These findings corroborate a link between prenatal smoking and offspring caries and provide novel information of a positive association between maternal smoking-associated DNAm risk scores and greater offspring caries experience. While replication in independent studies is warranted as these results are based on CpGs derived from umbilical cord DNA, these findings point to a need for greater smoking cessation support.

Introduction

The measurement of environmental exposures is prone to error (Rhomberg et al. 2011). For exposures such as tobacco smoking, there is biased reporting when some smokers falsely claim to be nonsmokers. This is particularly concerning for smoking during pregnancy because of the consequent negative maternal and child health impacts. Self-reported prenatal smoking is subject to an underreporting magnitude of 17% to 29% (Dietz et al. 2011) as compared with an average of 11% in the general population (Patrick et al. 1994). While attempts have been made to study the effects of prenatal smoking on offspring caries experience (Claudia et al. 2016; Akinkugbe et al. 2020) with mild to moderate effects, these studies relied on self-reported prenatal smoking. This method of assessing prenatal smoking (i.e., self-report) is subject to misreporting, recall bias, and bias in the magnitude of association with offspring caries (Kellesarian et al. 2017). Serum cotinine is a biomarker of smoking. While umbilical cord blood cotinine collected at delivery is highly correlated with maternal blood cotinine levels (Mattsson et al. 2016), there are some limitations of cotinine worth noting. As a metabolite of nicotine, cotinine is also elevated with the use of other nicotine-containing products besides cigarettes (Patrick et al. 1994); thus, it is not specific to tobacco smoking. Furthermore, cotinine levels do not provide a detailed history of smoking exposure; rather, they provide an indication of whether smoking occurred in the previous 30 h (Benowitz 1996). During pregnancy, cotinine metabolism is high (half-life is 9 h) (Dempsey et al. 2002), and pregnant women have lower cotinine concentrations for the same level of tobacco exposure and are more likely to be light smokers than their nonpregnant counterparts (Dietz et al. 2011). Therefore, for serum cotinine to accurately capture smoking status/intensity during pregnancy, frequent blood sampling would be required. Repeated blood sampling during pregnancy is often not feasible, underscoring the need for a biomarker measure that does not rely on maternal self-report and that can capture recent, cumulative exposure to tobacco smoke without the need for repeated measurement.

Epigenetics is the study of environmental influences on gene expression (Godfrey et al. 2015) and are inherited changes that affect gene function and phenotypic expression but not the underlying DNA base sequence (Dupont et al. 2009). DNA methylation (DNAm), the molecular process by which a methyl group is covalently attached to cytosines in the DNA sequence at the cytosine–phosphate–guanine (CpG) dinucleotide sites, is the most studied. Prior studies have reported associations between tobacco smoking and altered patterns of DNAm in children (Rauschert et al. 2019) and adults (Joehanes et al. 2016). Prenatal smoking results in global and CpG-specific DNAm (Breton et al. 2009) that persists across the life course (Lee et al. 2015). DNAm changes associated with smoking during pregnancy (Joubert et al. 2012) is an objective biological marker of prenatal smoking. Methylation signals detect smoking across pregnancy, and the extent of methylation at a site captures information about the amount and duration of smoking not captured by self-report or a single measure of cotinine.

The current study aims to investigate associations among prenatal smoking-associated DNAm changes at birth on the risk of offspring caries experience at age 3 y. This study also aims to confirm previously reported associations between prenatal smoking (using cotinine and self-report) and offspring caries experience.

Methods

Study Population

We used data on 577 mother–child dyads in the Growing Up in Singapore Toward healthy Outcomes (GUSTO) birth cohort. The cohort has been previously described (Soh et al. 2014). Briefly, pregnant women age ≥18 y were recruited in the first trimester of pregnancy from the 2 public hospitals in Singapore from June 2009 and September 2010. Follow-up for mother and offspring occurred at postnatal week 3, 3 mo, and every 3 mo thereafter until age 15 mo followed by further clinic visits for the offspring at 18, 24, and 36 mo. Umbilical cord tissue samples were obtained at birth. Examination to determine offspring oral health status occurred during clinic visits at 24 and 36 mo. The GUSTO study received ethical approval from the Institutional Review Boards of the participating hospitals.

Exposure Characterization

DNA was extracted from umbilical cord tissue stored at −80 °C. Tissue processing and DNA extraction steps have been previously described (Lin et al. 2017). DNA methylation profiling on an Infinium 450k array was performed following the manufacturer’s instruction. Briefly, 1 mg of the extracted genomic DNA from frozen umbilical cord specimens was bisulfite converted using the EZ-96 DNA Methylation™ Gold Kit. Successful conversion was confirmed via methylation-specific polymerase chain reaction before applying the Infinium assay protocol (Pan et al. 2012). Data preprocessing and quality control were performed after receiving the raw .idat files, following the recommended protocol and in-house quality control steps. The methylation level at each CpG site was calculated as beta (β), the ratio of the methylated probe intensity, and the overall intensity, that is, [β = intensity of the methylated allele (M)/(intensity of the unmethylated allele (U) + intensity of the methylated allele (M) + 100)] with a range of 0 (no cytosine methylation) to 1 (complete cytosine methylation) (Du et al. 2010; Bibikova et al. 2011). Probes were discarded if any sample had fewer than 3 beads in either methylated or unmethylated channel, detection P value >0.01, or on sex chromosomes. After filtering, 336,684 CpGs were available for analysis. We obtained prenatal smoking DNAm-specific CpGs from a previous epigenome-wide scan (EWAS) of cord blood that identified and independently replicated 26 CpGs related to maternal smoking during pregnancy (Joubert et al. 2012). Only 16 of these 26 CpGs were available in the GUSTO cohort. We created weighted (wMRS_DNAm) and unweighted (uMRS_DNAm) methylation risk scores. For the wMRS_DNAm, we multiplied each CpG site methylation intensity in GUSTO with the corresponding effect estimate with prenatal smoking for the same CpG site derived from the published EWAS study and summed the products.

$$wMR{S}_{DNAm}={\alpha }_1{x}_1+{\alpha }_2{x}_2+....{\alpha }_k{x}_k+{\alpha }_n{x}_n,$$

where α_k is the smoking-associated effect estimate for CpG site k, x_k is the CpG site methylation level, and n = total of CpG sites used in creating the methylation risk score, in this case n = 16. We also created a uMRS_DNAm by summing the methylation level across each CpG site in the GUSTO cohort. We created both risk scores for the purpose of dimensionality reduction. In other words, we combined individual CpGs into a single score to allow us to exploit the combined effect of the CpGs in a single variable, which gives us more power to detect effects, if present. We created the wMRS_DNAm to standardize the methylation intensity in GUSTO to the effects of each CpGs in the EWAS study from which they were derived, as opposed to giving each CpG equal weights, as was the case for uMRS_DNAm. Higher values on the risk scores indicates more methylation. The wMRS_DNAm and uMRS_DNAm were modeled continuously and as a distributional-based tertile for dose-response assessment.

Prenatal smoking (never, former, light, and heavy smoker)

Structured questionnaires were administered by an interviewer at 26 to 28 wk of gestation to obtain participants’ smoking and environmental tobacco smoke (ETS) exposure history. To determine smoking history, participants were asked if they had ever smoked regularly and from what age, whether they smoked in the 1 y prior to conception, and if they were still smoking at 26 to 28 wk of gestation. For ETS exposure, participants were asked to consider exposure just prior to pregnancy, during pregnancy, at home, and at work. The concentration of plasma cotinine was determined by liquid chromatography–tandem mass spectrometry from blood samples collected at gestational week 26. The limit of detection (LOD) was 0.17 ng/mL. Prenatal smoking was based on a combination of maternal plasma cotinine levels, self-reported smoking, and ETS exposure (Ng et al. 2019). When cotinine was below the LOD (<0.17 ng/mL), self-reported smoking and ETS exposure history were used to differentiate participants who had never smoked and had not been exposed to ETS (never smokers) from those who had been exposed to tobacco either through smoking periconceptually and in early pregnancy or exposed to ETS (former smokers). When the cotinine level was detectable, participants were stratified according to the concentration: 0.17 to 13.99 ng/mL, which indicates appreciable exposure due to ETS or light smoking (light smokers), and ≥14 ng/mL (heavy smokers), a level that had been used to identify active smokers with >95% sensitivity and specificity in nonpregnant populations (Jarvis et al. 1987).

Outcome Characterization

Oral examinations at the 2- and 3-y clinic visits were done by trained and calibrated dentists with interexaminer and intraexaminer reliability scores of 0.76 and 0.95, respectively. Teeth were assessed visually with the aid of plane surface mouth mirrors and by tactile inspection when necessary (Un Lam et al. 2017). Dental caries was determined using the International Caries Detection and Assessment System (ICDAS) criteria at the tooth surface level. However, due to a high proportion of missingness (34%) for caries at age 2 y, this study considered caries among 3-y-olds only, to preserve sample size. Tooth surfaces with ICDAS codes ≥3 indicated noncavitated, cavitated carious or filled lesions. For this study we considered as caries experience the count of decayed, missing (due to caries), or filled primary tooth surfaces (dmfs).

Covariates

Covariates were maternal age at delivery (years), child sex (boys, girls), preterm birth (preterm vs. term), maternal education (≤secondary, >secondary but <university, university), maternal ethnicity (Chinese, Malay, Indian), duration of any breastfeeding (<1 mo, 1 to 6 mo, 6 to <12 mo, ≥12 mo), and child daily toothbrushing frequency (<1 daily, once, and >1 daily).

Statistical Analysis

We used descriptive statistics to summarize continuous and categorical variables and calculated the Pearson correlations between prenatal smoking and the methylation risk scores. Next, we evaluated whether smoking-associated DNAm risk score was independently associated with dmfs count at age 3 y using zero-inflated Poisson regression (Preisser et al. 2012), which accounts for the excess zeros of caries data by modeling caries risk in the susceptible group in a Poisson model and the probability (Logit model) of excess zeros in the nonsusceptible subgroup. The risk scores were rescaled such that estimates are interpreted as a change in interquartile range value as opposed to a unit change and as tertile of weighted and unweighted risk scores. We rescaled the risk scores to be consistent with methylation intensity, which are in fractions of 1. Likelihood ratio tests indicated a better fit of the zero-inflated Poisson model. We also estimated the association between categorical (never, light, former, heavy) and binary (light/heavy vs. never/former) specifications of prenatal smoking and dmfs count at age 3 y to determine if previously reported associations hold in this specific birth cohort. Data analyses were conducted in SAS v. 9.4 (SAS Institute) and R (R version 4.3.2), and reporting adhered to the STROBE guidelines for observational studies.

Sensitivity Analysis

To account for bias from unmeasured confounders, we conducted sensitivity analysis using the E-value package in R (VanderWeele and Ding 2017; Mathur et al. 2018). The E-value provides an estimate of the minimum strength of association that the unmeasured confounder(s) must have with both the exposure and outcome (conditional on measured covariates) to fully explain away the observed associations.

Results

Among the mothers in this study, 56% were Chinese, 28% were Malay, and 16% were Indian. The mean (standard deviation; SD) maternal age at delivery was 31 (5) y, most mothers had a greater than secondary school education (69%), and 22% breastfed intermittently after 12 mo (Table 1). The overall mean (SD) wMRS_DNAm for the 16 CpGs included in this study was −0.21 (0.02), with an interquartile range (IQR) value of 0.02. The overall mean uMRS_DNAm was 8.12 (0.33), with an IQR of 0.47. Most of the women were either never (53%) or former smokers (30%), while 14% were light and 3% were heavy smokers (Table 1). The overall mean (SD) dmfs was 2.47 (5.20) surfaces. The mean dmfs was higher with greater intensity of the wMRS_DNAm ranging from 1.89 in the lowest tertile to 2.88 in the highest tertile. Children whose mothers were light or heavy smokers had higher mean dmfs (3.21 and 3.00 surfaces) than children whose mothers were former (2.61 surfaces) and never (2.19 surfaces) smokers. Boys had slightly higher mean dmfs than girls did (2.52 vs. 2.43 surfaces), and children of Malay mothers had slightly higher mean dmfs than children of Chinese mothers did (2.90 vs. 2.71 surfaces), whereas children of Indian mothers had the lowest mean dmfs at 0.94 surfaces. Children of mothers with less than or equal to a secondary education had higher mean dmfs than those whose mothers had a university education or higher (3.57 vs. 2.00 surfaces). Similarly, children of mothers who were breastfed for <1 mo or >12 mo had higher mean dmfs at 2.67 and 3.16 surfaces, respectively, than their counterparts who were breastfed between 1 and <12 mo did (Table 1).

Table 1.

Distribution of Maternal and Child Covariates according to Caries and Cord Tissue Weighted Methylation Risk Score Levels.

	N (%)	dmfs	wMRSDNAm
	606	Mean (SD)	Median (IQR)
Covariates		2.47 (5.20)	−0.206 (0.023)
Maternal age at delivery, y, mean (SD)	31 (5)
Maternal prenatal smoking
Never	323 (53)	2.19 (4.81)	−0.205 (0.023)
Former	184 (30)	2.61 (4.78)	−0.206 (0.023)
Light smoker	82 (14)	3.21 (7.01)	−0.208 (0.023)
Heavy smoker	16 (3)	3.00 (6.71)	−0.203 (0.015)
Missing	1
Maternal ethnicity
Chinese	341 (56)	2.71 (5.36)	−0.204 (0.023)
Malay	167 (28)	2.90 (6.00)	−0.207 (0.020)
Indian	98 (16)	0.94 (1.88)	−0.206 (0.022)
Maternal education
≤Secondary	187 (31)	3.57 (7.50)	−0.206 (0.023)
>Secondary to <university	209 (35)	2.00 (3.69)	−0.205 (0.021)
≥University	206 (34)	1.98 (3.63)	−0.206 (0.023)
Missing	4
Child sex
Girl	296 (49)	2.43 (4.77)	−0.199 (0.020)
Boy	310 (51)	2.52 (5.62)	−0.213 (0.021)
Child preterm status
Preterm	37 (6)	1.97 (4.78)	−0.211 (0.022)
Term	569 (94)	2.51 (5.23)	−0.205 (0.022)
Duration of any breastfeeding
<1 mo	156 (27)	2.67 (5.81)	−0.206 (0.022)
1 to <6 mo	209 (36)	2.13 (4.29)	−0.204 (0.022)
6 to <12 mo	87 (15)	1.93 (5.12)	−0.203 (0.026)
≥12 mo	129 (22)	3.16 (6.03)	−0.207 (0.021)
Missing	25
Child daily toothbrushing frequency
<1 daily	32 (5)	1.81 (2.67)	−0.207 (0.025)
Once daily	249 (41)	2.48 (5.45)	−0.206 (0.023)
>1 daily	323 (53)	2.54 (5.21)	−0.205 (0.021)
Missing	2
wMRSDNAm tertile
Low	192 (33)	1.89 (3.76)	−0.222 (0.016)
Middle	193 (33)	2.68 (5.50)	−0.206 (0.007)
High	192 (33)	2.88 (6.24)	−0.195 (0.013)

Never smokers had a cotinine level of <0.17 ng/mL, had never smoked or been exposed to environmental tobacco smoke (ETS); former smokers had a cotinine level of <0.17 ng/mL but self-reported smoking periconceptually and in early pregnancy or were exposed to ETS; light smokers (including appreciable ETS exposure) had a cotinine level between 0.17 and 13.99 ng/mL, while heavy smokers had cotinine levels of ≥14 ng/mL. Preterm (born at <37 wk of gestation), term (born after 37 wk of gestation), N = 606, was the starting sample size with complete exposure and outcome data. The overall and covariate-specific negative median weighted risk score is interpreted as a reduced methylation across the 16 CpG sites included in this study when we standardized the individual methylation values in Growing Up in Singapore Towards healthy Outcomes (GUSTO) to the effects of the original epigenome-wide scan (EWAS) study.

dmfs, decayed, missing, and filled tooth surfaces; IQR, interquartile range; SD, standard deviation.

Figures 1 and 2 show histograms of dmfs count stratified by prenatal smoking groups and tertile of wMRS_DNAm. Thirty-nine percent of children whose mothers were never smokers had dmfs ≥ 1 as compared with 49%, 48%, and 50% of children whose mothers were former, light, and heavy smokers, respectively (Fig. 1). Similarly, 38%, 48%, and 45% of children in the low, middle, and high wMRS_DNAm tertile had dmfs ≥ 1, respectively (Fig. 2). Pearson correlations between the prenatal smoking variable and the methylation risk scores were weak (Pearson r = −0.004 and r = 0.015) for wMRS_DNAm and uMRS_DNAm respectively.

Figure 1.

Histogram of count of decayed, missing, and filled surfaces by smoking groups.

Figure 2.

Histogram of count of decayed, missing, and filled surfaces by weighted DNAm risk score tertile.

Methylation Risk Scores and Offspring Caries Experience

Table 2 shows the zero-inflated Poisson regression results for the association between smoking-associated DNAm risk scores and offspring caries experience. Each IQR (0.02) increase in wMRS_DNAm was associated with a 20% greater caries experience risk, adjusted relative risk (RR) (95% confidence interval [CI]) = 1.20 (1.10, 1.31), and a lower excess zero odds, odds ratio [OR] (95% CI) = 0.91 (0.70, 1.17). Likewise, the middle and high wMRS_DNAm tertile relative to the low tertile was associated with higher dmfs risk in a dose-dependent manner. The corresponding adjusted RRs (95% CI) were 1.16 (1.00, 1.34) and 1.40 (1.20, 1.62), respectively. The middle and high wMRS_DNAm tertiles were also associated with lower odds of excess zeros at 0.65 (0.42, 1.00) and 0.74 (0.47, 1.17) respectively.

Table 2.

Parameter Estimates and 95% CI for ZIP Models for the Association between Prenatal Smoking Associated DNAm Weighted Risk Score and Offspring Caries Experience at Age 3 y.

		Logit Model for Excess Zeros	Poisson Model	E-Values
		OR (95% CI)	RR (95% CI)	Eest	ECI
Prenatal smoking DNA methylation
wMRSDNAm (IQR change)	0.023 units change	0.91 (0.70, 1.17)	1.20 (1.10, 1.31)	1.69	1.43
wMRSDNAm tertile	Low	Ref.	Ref.
	Medium	0.65 (0.42, 1.00)	1.16 (1.00, 1.34)	1.59	1.00
	High	0.74 (0.47, 1.17)	1.40 (1.20, 1.62)	2.15	1.69
uMRSDNAm (IQR change)	0.467 units change	1.07 (0.79, 1.45)	1.16 (1.05, 1.27)	1.59	1.28
uMRSDNAm tertile	Low	Ref.	Ref.
	Medium	1.37 (0.88, 2.12)	1.15 (1.00, 1.33)	1.57	1.00
	High	0.77 (0.47, 1.28)	1.21 (1.04, 1.40)	1.71	1.24

Models (Logit and Poisson) adjusted for smoking group, age at delivery, child sex, preterm status, maternal education, breastfeeding duration, and child daily toothbrushing frequency. Complete case analysis was done on an analytic sample size of 577.

CI, confidence interval; DNAm, DNA methylation; E_CI, E-value that would shift observed estimate to include the null value; E_est, estimated E-value that unmeasured confounders need to have with exposure and outcome to explain away the observed effects; OR, odds ratio; ZIP, zero-inflated Poisson.

For the uMRS_DNAm, each IQR (0.47) increase was associated with a 16% greater caries experience risk; adjusted RR (95% CI) = 1.16 (1.05, 1.27). The middle and high uMRS_DNAm tertiles relative to the low tertile were associated with higher dmfs risk in a dose-dependent manner: the adjusted RRs (95% CI) were 1.15 (1.00, 1.33) and 1.21 (1.04, 1.40), respectively. However, the estimates of the uMRS_DNAm for the excess zero model were inconsistent.

Prenatal Smoking and Offspring Caries Experience

The adjusted estimates comparing caries experience risk in children of mothers who were former, light, and heavy smokers to children of never-smoking mothers were RR = 1.11 (95% CI: 0.98, 1.26), RR = 1.31 (95% CI: 1.11, 1.55), and RR = 1.12 (95% CI: 0.81, 1.54), respectively. When the prenatal smoking groups was dichotomized into any (light/heavy) versus (never/former), the adjusted estimate for offspring caries experience was RR = 1.20 (95% CI: 1.04, 1.39).

Sensitivity Analysis for Unmeasured Confounders

With an observed RR of 1.20 for the association between wMRS_DNAm and offspring caries at age 3 y, an E-value estimate of 1.69 suggests that unmeasured confounder(s) that were associated with both offspring caries and the wMRS_DNAm by a risk ratio of 1.69-fold each, above and beyond the measured confounders, could explain away the estimate, but weaker confounding could not. Unmeasured confounder(s) that were associated with both offspring caries and the wMRS_DNAm by a risk ratio of 1.43-fold each, above and beyond the measured confounders, could shift the CI to include the null, but weaker confounding could not (Table 2).

Discussion

In this multiethnic Asian mother–offspring cohort, we found evidence that tobacco smoking during pregnancy is harmful to offspring oral health. Specifically, we found a significant positive association between prenatal smoking and childhood caries experience at age 3 y, greater caries experience risk with higher DNAm intensity of both the unweighted and weighted smoking-associated DNAm risk scores, but weak correlations between prenatal smoking and the prenatal smoking-associated MRS.

Some explanation for the weak correlation might be due to measurement error (Valeri et al. 2017) in our measure of prenatal smoking that was created from a combination of self-reported maternal smoking, exposure to ETS before and during pregnancy, and plasma cotinine levels. Furthermore, cotinine levels consistent with ETS exposure during pregnancy were not associated with the CpGs investigated in this study. The CpGs we studied were most strongly related to prenatal smoking past 18 wk of gestation (i.e., sustained smoking) (Joubert et al. 2016). While data on prenatal smoking and cotinine were collected in the GUSTO cohort at gestational week 26, information on the duration of smoking was not collected, so we are unable to more fully assess if smoking in GUSTO was sustained. If smoking was not sustained due to low smoking exposure in GUSTO compared with other cohorts, it may also explain the low correlation between prenatal smoking and the DNAm scores. In addition, cotinine is also elevated with the use of other nicotine-containing products besides cigarettes, thus offering another explanation for the weak correlation between smoking and DNAm scores, especially if the GUSTO mothers with detectable cotinine levels used nicotine-containing products besides cigarettes. Prenatal smoking DNAm differs from ETS-associated DNAm (Fuemmeler et al. 2021); therefore, our findings of a weak correlation confirm this and suggest that the association with offspring caries is likely robust, given that the CpGs we studied came from a study that confirmed their associations with prenatal smoking.

Prenatal smoking is hypothesized to disrupt primary tooth mineralization, a process that begins at about the 14th week of gestation, thereby affecting the quality of the mineralizing tooth (Caufield et al. 2012) and the timing of tooth eruption (Aligne et al. 2003; Ntani et al. 2015). Poor-quality (i.e., hypomineralized) and prematurely erupted teeth are more susceptible to colonization by caries bacteria given their longer exposure to the oral environment. Furthermore, prenatal smoking is associated with low birth weight and prematurity (Jaddoe et al. 2008), conditions that increase the risk for developmental enamel defects (Vello et al. 2010), and these enamel defects are highly susceptible to cariogenic bacteria (Salanitri and Seow 2013).

In the model to determine the main effects of prenatal smoking on offspring caries experience, we found that the association between prenatal smoking and offspring caries was consistent with previous reports of moderate positive associations (Claudia et al. 2016; Akinkugbe et al. 2020). Estimates of light smoking alone were stronger than when light and heavy smoking were combined. Indeed, the binary specification of prenatal smoking (light/heavy vs. former/never) might have diluted the association, although it still existed with smaller magnitude. Pregnant women are more likely than nonpregnant women to be light smokers (Dietz et al. 2011); therefore, estimates of light smoking likely represent the actual association of prenatal smoking on offspring caries experience in this predominantly Asian cohort.

Information on DNAm in GUSTO came from cord tissue as opposed to cord blood and may be responsible for the fewer CpGs identified, even though the methylation panel was similar with the study that identified the 26 prenatal smoking CpGs. Absent CpGs may also be due to quality control filtering, as an approximate 59% of CpGs from cord tissue and cord blood showed no interindividual variation while 41% did (Lin et al. 2017). Bisulfite sequencing reports methylation only at cytosines that are part of CpG dinucleotides; thus, methylation at non-CpG sites is not captured. Therefore, we are likely missing potential DNAm in noncovered genome sites. Reliance on DNA extracted from cord tissue or blood places further constraints on inferences that can be made about tissue-specific DNAm patterns (Novakovic et al. 2014), given that the methylation profile of other tissues, say, buccal epithelium, may be different and more relevant to the study of dental caries.

Although plasma cotinine is easier to obtain than DNAm, GUSTO measured plasma cotinine at 1 time point during pregnancy, reflecting a snapshot, and does not accurately capture exposure over the course of pregnancy. Nonetheless, data on prenatal smoking were ascertained before cord tissue collection and offspring caries; thus, our findings are unlikely to be due to reverse causation. The CpGs we used in creating the risk scores came from a predominantly European ancestry cohort and may not reflect to the extent possible the DNAm-related changes associated with prenatal smoking in an Asian ancestry population. To our knowledge, this is the first study to report the associations of prenatal smoking-associated DNAm and offspring caries experience at age 3 y; thus, validation in independent cohorts is warranted.

Conclusions

While DNAm may be an objective marker of exposure to prenatal smoking, it was weakly correlated with our measure of prenatal smoking, likely due to the imprecision of our prenatal smoking variable. Nevertheless, the smoking DNAm risk scores were independently associated with worse offspring caries experience. Therefore, future studies should perform an EWAS to identify study-specific smoking CpGs with an objective measure of prenatal smoking (repeated cotinine measures) and use these CpGs to determine associations with offspring caries. The prenatal smoking variable in this study included ETS exposure, which highlights the detrimental effects of ETS on offspring oral health. While postnatal exposure to ETS at home (Hanioka et al. 2011; Tanaka et al. 2015) is associated with poor childhood caries outcomes, our findings show that prenatal ETS exposure adversely affects offspring caries outcomes. These findings support policies restricting not only smoking but also ETS exposure during pregnancy and the provision of greater smoking cessation support as the interplay of maternal smoking, fetal DNAm, and poor dental outcomes could help our understanding of the effects of other early life environmental exposures on the developing fetus and craniofacial development.

Author Contributions

A.A. Akinkugbe, contributed to conception, design, data analysis and interpretation, drafted and critically revised the manuscript; C.-Y.S. Hsu, contributed to conception, design, data acquisition, critically revised the manuscript; C. Lesseur, V. Midya, contributed to data analysis and interpretation, critically revised the manuscript; K.H. Tan, K.Y. Jerry Chan, J.G. Eriksson, S.-Y. Chan, Y.S. Chong, Y.S. Lee, D. Wang, J. Huang, N. Karnani, K.M. Godfrey, A.L. Teh, contributed to data acquisition and interpretation, critically revised the manuscript; R.O. Wright, R.J. Wright, contributed to conception, design, data interpretation, critically revised the manuscript. All authors gave final approval and agree to be accountable for all aspects of the work.