Errors in the bisulfite conversion of DNA: modulating inappropriate- and failed-conversion frequencies

Abstract

Bisulfite treatment can be used to ascertain the methylation states of individual cytosines in DNA. Ideally, bisulfite treatment deaminates unmethylated cytosines to uracils, and leaves 5-methylcytosines unchanged. Two types of bisulfite-conversion error occur: inappropriate conversion of 5-methylcytosine to thymine, and failure to convert unmethylated cytosine to uracil. Conventional bisulfite treatment requires hours of exposure to low-molarity, low-temperature bisulfite (‘LowMT’) and, sometimes, thermal denaturation. An alternate, high-molarity, high-temperature (‘HighMT’) protocol has been reported to accelerate conversion and to reduce inappropriate conversion. We used molecular encoding to obtain validated, individual-molecule data on failed- and inappropriate-conversion frequencies for LowMT and HighMT treatments of both single-stranded and hairpin-linked oligonucleotides. After accounting for bisulfite-independent error, we found that: (i) inappropriate-conversion events accrue predominantly on molecules exposed to bisulfite after they have attained complete or near-complete conversion; (ii) the HighMT treatment is preferable because it yields greater homogeneity among sites and among molecules in conversion rates, and thus yields more reliable data; (iii) different durations of bisulfite treatment will yield data appropriate to address different experimental questions; and (iv) conversion errors can be used to assess the validity of methylation data collected without the benefit of molecular encoding.

Untitled

“…The posterior distribution for scaled variance g c is concentrated on small values, suggesting that the error rate c does not vary greatly across CpG sites (Figure 4A), in accord with experimental findings (Genereux et al, 2008). …”

“…Our estimate of this rate is lower than that obtained by Genereux et al (2008) using synthetic oligonucleotides (average rate 0.035; 95% confidence interval: 0.027–0.049). This difference may derive, in part, from the different lengths of the DNAs used in the two experiments (Genereux et al, 2008). …”

“…The estimation precision will improve as new experimental protocols yield data with lower measurement error rates (Genereux et al, 2008) and lead to better estimates of de novo methylation rates. With our statistical methods one can investigate differences among fidelity rates in different genomic regions.…”

“…However, due to limitations of current experimental technologies, these data also present challenges. For example, it is difficult to obtain this type of data in large quantities, some key biological variables are unobservable, certain parameters of interest may be confounded, and the data are subject to measurement error at perhaps a non-trivial rate (Genereux et al, 2008). These characteristics put a premium on efficient inference methods that make full use of the data while dealing with complexities intrinsic to the data and to the biological problem.…”

Strategies for analyzing bisulfite sequencing data

“…A simple solution for this would be to disregard the affected positions in the sequenced reads (Bock, 2012;Hansen et al, 2012). Furthermore, incomplete conversion can occur during bisulfite treatment, where not all unmethylated Cs are converted to Ts (Genereux et al, 2008). Incomplete conversion causes false positive results due to interpretation of the unconverted unmethylated cytosines as methylated.…”