Epigenetic Biomarker: Improving Estimates of Fetal Exposure to Cigarette Smoke

Short abstract

The well-known cotinine test captures recent smoking, and survey responses are not always accurate. Now researchers propose a measure of DNA methylation in placental tissue that may be better than either.

Maternal cigarette smoking during pregnancy carries well-known fetal and childhood health risks.¹ Although its prevalence has declined over time, half of pregnant smokers in the United States continue to smoke after learning they are pregnant. An average of 4.6% of women reported smoking cigarettes during pregnancy in 2021; the figure was as high as 18% in some states.^2,3

Like other environmental exposures, cigarette smoke may cause epigenetic alterations, which do not change DNA sequences but persist over time and may affect gene expression.⁴ Prenatal smoking has a substantial and well-replicated effect on the most commonly studied epigenetic mark: patterns of DNA methylation, which refers to the addition of a methyl group to cytosine–guanine dinucleotides (CpG sites). These patterns have been found in multiple tissue types collected both at birth and later in life.^5,6

Researchers isolated epigenetic changes in placental blood that, if validated, could serve as a more accurate biomarker of prenatal exposure to cigarette smoking than currently available. Image: © UA-pro/Shutterstock.com.

A study recently published in Environmental Health Perspectives⁷ reports a new DNA methylation biomarker of prenatal smoking in the placenta, a unique organ that protects and nourishes the fetus⁸ and is readily sampled at birth. The authors suggest the new biomarker may capture sustained exposure to smoking better than either maternal blood-based biomarkers⁹ or self-reported information, which is prone to being underreported.^10,11

“Unlike blood, which contains a snapshot of circulating cells at the time of collection, the placenta filters out some exposures before they reach the fetus, suggesting that its methylation landscape may provide a more [accurate] cumulative exposure record from conception to birth,” says first author Lyndsey Shorey-Kendrick, a computational biologist at Oregon Health & Science University.

To develop their placental biomarker, the researchers used a training dataset of 72 pregnant smokers and 24 never-smokers whose placentas were collected at delivery as part of a randomized clinical trial called Vitamin C to Decrease the Effects of Smoking in Pregnancy on Infant Lung Function.¹² Placental cell methylation data for each participant were available at 715,000 genome-wide CpG sites measured on the Illumina MethylationEPIC platform.

The researchers applied machine learning methods⁷ in three separate training models to identify the most predictive subset of CpG sites, meaning those sites that best distinguished between smokers and nonsmokers. The first model started with the full EPIC array of CpG sites. The second model included the subset of CpG sites measured by a different platform, the Illumina 450K array, to identify a biomarker for studies that used either platform. The third model included 443 placental CpG sites previously associated with prenatal smoking in a meta-analysis by the Pregnancy and Childhood Epigenetics (PACE) consortium.¹³

The researchers determined biomarker values based on the methylation values of CpG sites selected in each of the three final models. They compared these values, called placental smoking index (PSI) scores, with maternal plasma cotinine, a common biomarker of recent smoking.¹⁴ Pregnancy reduces the half-life of cotinine from 16.6 to 8.8 hours,¹⁵ which may explain the observed moderate correlation between cotinine and PSI scores in the 72 smokers. Indeed, correlations between PSI scores and late-gestation cotinine concentrations were twice as high as with early gestation levels, supporting the hypothesis that changes in DNA methylation reflect sustained smoking.

Next, the researchers tested the ability of their three models to predict smoking status in two external datasets: the Extremely Low Gestational Age Newborn (ELGAN) study¹⁶ (426 participants, 11% smokers) and the Rhode Island Child Health Study¹⁷ (RICHS; 237 participants, 15% smokers). The ELGAN study used the EPIC platform and RICHS used the 450K platform to obtain methylation data. These validation datasets differed greatly from the training dataset not only in the proportion of smokers but also in ethnicity, socioeconomic factors, and gestational stage (ELGAN involved premature infants, compared with full-term infants for the other two studies).

The prediction accuracy was lowest for the PACE model. Between the other two models (EPIC and Illumina 450), prediction accuracy was highest when the validation dataset platform matched that used in model training: PSI scores calculated using the EPIC model correctly predicted smoking status in 60% of ELGAN study participants; scores from the 450K model correctly predicted smoking status in 72% of RICHS participants. After the team applied a process to infer missing data, these figures increased to 86% and 75%, respectively.

Given the substantial differences between the validation and training datasets, Freida Blostein was encouraged by these results. “Using the prediction models in very different contexts is one of the study’s strengths since biomarkers that work well across a range of populations may be more useful and generalizable,” says Blostein, a postdoctoral research fellow in genomic medicine at Vanderbilt University who was not involved in the study.

Research applications of the new biomarker include studies of exposures that are correlated with prenatal smoking, such as alcohol consumption. “Due to the large effect of prenatal smoking on fetal and childhood health outcomes, adjusting for it [before estimating] other associations is very important,” says Blostein. Using an epigenetic biomarker boosts the value of existing samples that lack maternal smoking data and reduces misclassification resulting from inaccurate self-reported information.¹¹

The three final models had five CpG sites in common, one of which (cg27402634) replicated previous findings in independent datasets^13,18 and is a top candidate for follow-up analyses. The authors showed that an inexpensive assay¹⁹ can accurately measure this CpG site.

Jaclyn Goodrich, a research associate professor of environmental health sciences at the University of Michigan who was not involved in the study, suggests that researchers should test how well cg27402634 alone predicts smoking status in other large cohorts. The inexpensive assay, she adds, would greatly reduce the testing cost, and if validated, it could also be applied to clinical settings. For example, children at increased risk of asthma from prenatal smoking exposure²⁰ could be closely monitored for earlier interventions.

“This study illustrates the potential of epigenetics to infer past exposures for precision environmental health,” says Goodrich. “It gives me hope that we can soon develop similar methods for other exposures where accurate measurements are difficult to obtain.”

Biography

Silke Schmidt, PhD, writes about science, health, and the environment from Madison, Wisconsin.

Refers to https://doi.org/10.1289/EHP13838