Validated inference of smoking habits from blood with a finite DNA methylation marker set

Abstract

Inferring a person's smoking habit and history from blood is relevant for complementing or replacing self-reports in epidemiological and public health research, and for forensic applications. However, a finite DNA methylation marker set and a validated statistical model based on a large dataset are not yet available. Employing 14 epigenome-wide association studies for marker discovery, and using data from six population-based cohorts (N = 3764) for model building, we identified 13 CpGs most suitable for inferring smoking versus non-smoking status from blood with a cumulative Area Under the Curve (AUC) of 0.901. Internal fivefold cross-validation yielded an average AUC of 0.897 ± 0.137, while external model validation in an independent population-based cohort (N = 1608) achieved an AUC of 0.911. These 13 CpGs also provided accurate inference of current (average AUC_{crossvalidation} 0.925 ± 0.021, AUC_{externalvalidation}0.914), former (0.766 ± 0.023, 0.699) and never smoking (0.830 ± 0.019, 0.781) status, allowed inferring pack-years in current smokers (10 pack-years 0.800 ± 0.068, 0.796; 15 pack-years 0.767 ± 0.102, 0.752) and inferring smoking cessation time in former smokers (5 years 0.774 ± 0.024, 0.760; 10 years 0.766 ± 0.033, 0.764; 15 years 0.767 ± 0.020, 0.754). Model application to children revealed highly accurate inference of the true non-smoking status (6 years of age: accuracy 0.994, N = 355; 10 years: 0.994, N = 309), suggesting prenatal and passive smoking exposure having no impact on model applications in adults. The finite set of DNA methylation markers allow accurate inference of smoking habit, with comparable accuracy as plasma cotinine use, and smoking history from blood, which we envision becoming useful in epidemiology and public health research, and in medical and forensic applications.

Keywords: DNA methylation; Epidemiology; Epigenetics; Forensics; Smoking inference.

Fig. 1

DNA methylation β-value differences between smokers and never-smokers for the top 20 smoking-associated CpGs. Previously reported differences in β-values in mean or median (depending on availability per EWAS) between smokers and never-smokers (^¤ or non-smokers, when non-smoking data was available) for the selected 20 top-associated CpGs obtained from the 14 reviewed EWASs on smoking habits that did not include samples used here for model building

Fig. 2

Cumulative AUC profile for smoking habit inference from blood based on the top 20 CpGs. The 20 CpGs were selected from previous EWASs on smoking habits (see Fig. 1) and were tested in the model-building set (N = 3764). Presented is the cumulative contribution of each of the selected 20 CpGs to the model-based smoking habit inference, shown as the AUC plotted against the number of CpGs included in the binary logistic regression model. In the model selection process, first all CpGs were included, and using backward elimination procedures, those with the lowest z-statistic per model were removed one by one. After 13 CpGs, the AUC plateaus; therefore, and by considering the results from Chi squared testing, these 13 CpGs were used for further analyses

Fig. 3

Inferred probability of being a smoker versus the percentage of correctly inferred smoking habits. Histogram of predicted probabilities in our model building dataset (N = 3764), probabilities determined using the 13 CpGs included in the final prediction model. The y-axis presents the number of individuals for whom the predicted probability of being a smoker was within the given probability range (x-axis). The red dots present the percentage of individuals in each probability bin that were accurately inferred using a > 0.5 probability threshold for being a smoker