Smoking has long been recognized as a leading risk factor for a wide range of chronic conditions, including cardiovascular diseases and cancers [15]. Current clinical practice incorporates patient self-reported measures of smoking exposure, typically by assessing smoking status (never, ex or current smoker). Limitations and potential inaccuracies of such self-reported data have been suggested. For example, there is likely to be a wide range of smoking-related risk among ex-smokers depending on their volume of exposure (pack years) and duration of cessation. Similarly, second-hand exposure is not readily quantified with such measures.

Epigenetic markers, such as DNA methylation, are emerging as ‘precision medicine’ biomarkers that can surpass the performance of such ‘conventional’ markers in risk prediction [16]. This is particularly true for smoking, for which numerous epigenome-wide association studies have demonstrated robust and reproducible DNA methylation changes correlated with smoking exposure, with the strongest association consistently being cg05575921, within the gene body of AHRR [1, 2]. Consequently, sensitive PCR methods have been developed to assess the methylation status of this specific CpG site [5, 17]. The use of cg05575921 (AHRR) assays have been suggested to have the potential to improve risk prediction utility by a range of effects including assessment of second-hand smoke exposure [18], correcting for false or inaccurate self-reported measures [19] and by virtue of strong morbidity and mortality associations, even after adjusting for self-reported smoking history [5].

In this study, we report a technically accurate restriction enzyme digestion ddPCR assay, which appears to have similar performance to that of previously reported ddPCR assays [3, 4] without the need for the chemically harsh bisulfite conversion process [6], which can be an issue particularly when working with samples containing limited amounts of input DNA. However, previous methylation-sensitive restriction enzyme-based techniques do provide some reason for caution regarding this approach. Bravo-Gutierrez and colleagues reported a HpaII restriction enzyme digest method, but using qPCR rather than ddPCR as the detection platform. Unfortunately, the methylation patterns reported for the Bravo-Gutierrez assay do not match those typically associated with bisulfite conversion PCR and Illumina array-based measures of this specific CpG site [7]. Specifically, there was a substantial overlap in cg05575921 values when comparing smoking and non-smoking groups, and many current smokers were reported as having 100% methylation. This is contrary to the well-established literature, which consistently reports cg05575921 methylation values of 50–60% in the vast majority of current smokers [2, 17]. Nevertheless, other groups have convincingly demonstrated that some RE-based ddPCR assays can be highly sensitive and specific [20]. In addition, RE-based approaches are substantially less time-consuming than that of bisulfite conversion (30–60 min versus 12–16 h).

In contrast to the qPCR-based observations of the Bravo-Gutierrez assay, the RE ddPCR assay developed in this study produced results that are consistent with the prior literature [2, 5, 17]. Regardless of which of the three (RE or BIS ddPCR or Illumina array) assessing techniques used, we observed that the hypomethylation of the cg05575921 DNA methylation site was statistically significantly, sensitively and specifically associated with smoking status. Interestingly, using a large cohort of 1227 individuals, we were also able to confirm that the RE ddPCR assay had significantly better predictive power for smoking status than measures derived from Illumina arrays. We do, however, note a slight potential difference with a previous report by Philibert et al. [4], who showed a strong and linear (R2 = 0.98) correlation between EPIC array and BIS ddPCR cg05575921 results (in 92 sample comparisons) which did not match the nonlinear correlation in our current RE ddPCR versus EPIC array comparison (Additional file 1: Sheet S3). The quantification of the BIS and RE ddPCR differed, with fractional abundance assessment being utilized in the (single amplicon) BIS ddPCR assay and, in contrast, the RE ddPCR method involved comparison of two separate amplicons (AHRR and KIT), therefore necessitating comparison via a simpler ratio of FAM/ HEX signal. Nevertheless, we still observed a strong linear correlation between the two (BIS and RE) ddPCR methods, and the reason for the nonlinear effect observed between the RE ddPCR and EPIC array paired comparisons remains unclear. One possible explanation that we considered was the effect of array normalization. This is an essential step in the analysis of array data that is needed to reduce the size of batch effects, correct for probe type bias and adjust for background noise and dye bias [12, 21]. It is worth noting that ddPCR assays do not require such correction, as they are internally calibrated, in the case of the restriction enzyme assay to the number of copies of a reference gene (KIT). While we did observe subtle differences due to the normalization method that was applied to the array data (Additional file 1: Sheet S2), the lower RE ddPCR values observed in current smokers (Additional file 1: Sheet S3B) was a much larger effect and an explanation for this particular observation therefore remains elusive. Nevertheless, an apparent consequence of this ‘skewing’ of the hypomethylated signal appeared to be associated with a more sensitive and specific smoking group classification performance for the RE ddPCR assay compared to array derived measures. Indeed, as shown in Additional file 1: Sheet S3A, we noted several (10 of 147, 6.8%) known current smokers that, while having RE ddPCR values (below 60%) consistent with their smoking status, had Illumina array values (over 80%) which were more consistent with never smoked status.

It has also been suggested that normalization, using a separate no RE reaction, may be required for RE ddPCR assays [20]. In a sub-analysis, we demonstrated that inclusion of a no enzyme normalization step did not significantly alter the high smoking status sensitivity and specificity of this particular DNA methylation ddPCR assay (Additional file 1: Sheet S4).

In conclusion, we report the development of a restriction enzyme digestion-based droplet digital PCR assay for cg05575921 (AHRR) that produces at least comparable results to that of previously reported ddPCR and Illumina array-based methods without the need for bisulfite conversion of the DNA sample. We further confirm that this RE ddPCR method is able to sensitively and specifically predict smoking status in a large clinically relevant elderly cohort. This bisulfite conversion-free ddPCR assay appears to have potential clinical utility as a precise molecular diagnostic method for the evaluation of smoking exposure.