Comparison of Bisulfite Pyrosequencing and Methylation-Specific qPCR for Methylation Assessment

Abstract

Different methodological approaches are available to assess DNA methylation biomarkers. In this study, we evaluated two sodium bisulfite conversion-dependent methods, namely pyrosequencing and methylation-specific qPCR (MS-qPCR), with the aim of measuring the closeness of agreement of methylation values between these two methods and its effect when setting a cut-off. Methylation of tumor suppressor gene p16/INK4A was evaluated in 80 lung cancer patients from which cytological lymph node samples were obtained. Cluster analyses were used to establish methylated and unmethylated groups for each method. Agreement and concordance between pyrosequencing and MS-qPCR was evaluated with Pearson's correlation, Bland-Altman, Cohen's kappa index and ROC curve analyses. Based on these analyses, cut-offs were derived for MS-qPCR. An acceptable correlation (Pearson's R2 = 0.738) was found between pyrosequencing (PYRmean) and MS-qPCR (NMP; normalized methylation percentage), providing similar clinical results when categorizing data as binary using cluster analysis. Compared to pyrosequencing, MS-qPCR tended to underestimate methylation for values between 0 and 15%, while for methylation >30% overestimation was observed. The estimated cut-off for MS-qPCR data based on cluster analysis, kappa-index agreement and ROC curve analysis were much lower than that derived from pyrosequencing. In conclusion, our results indicate that independently of the approach used for estimating the cut-off, the methylation percentage obtained through MS-qPCR is lower than that calculated for pyrosequencing. These differences in data and therefore in the cut-off should be examined when using methylation biomarkers in the clinical practice.

Keywords: DNA methylation; bisulfite pyrosequencing; cut-off; cytological lymph node samples; methylation biomarker; methylation-specific qPCR; non-small cell lung cancer.

Figure 1

Representation of the CpG island analyzed in p16/INK4a promoter. The 18 CpG sites analyzed by pyrosequencing and MS-qPCR are represented as grey bars and are numbered from 1 to 18 in the reverse strand. Above, the design used for the pyrosequencing approach, including PCR primers (Pyro F and Pyro R) and sequencing primers (S1 and S2). Below, the design of the nested-qPCR approach, including PCR outer primers (Outer F and Outer R), qPCR primers (MS-qPCR F and MS-qPCR R) and probe (MS-qPCR Probe).

Figure 2

Methylation analysis based on pyrosequencing. (A) Boxplots for the distribution of methylation percentage expressed as (log10 (Pyrosequencing + 1)) of each CpG site and the mean pyrosequencing methylation percentage (PYRmean). (B) Histogram and kernel density estimation of the distribution of PYRmean expressed as (log10(PYRmean + 1)) used to establish a natural cut-off of methylation percentage. On top of the graph, the corresponding classification graph is shown. Black lines: all samples; dark grey lines: unmethylated (U) group; light grey lines: methylated (M) group.

Figure 3

Methylation analysis using MS-qPCR. Histogram and kernel density estimation of the distribution of NMP expressed as (log10(NMP + 1)). The Classification graph based on the natural cut-off derived from cluster analysis is shown above the histogram. Black lines: all samples; dark grey lines: unmethylated (U) group; light grey lines: methylated (M) group.

Figure 4

Analysis for the comparison between pyrosequencing and MS-qPCR. (A) Scatterplot for the correlation between PYRmean and NMP. (B) Bland–Altman data plot used to analyze the agreement between PYRmean and NMP. Black line represents the mean difference between methods, and the discontinue lines represent the 95% limits of agreement. (C) Box-plot representing the performance of the NMP methylation percentages with relation to the methylated and unmethylated subgroups derived from the natural cut-off of PYRmean (12.54%). Horizontal lines show the natural cut-off according to the cluster analysis (6.86%; solid line), the cut-off according to the maximization of kappa (2.53%; dashed line) and the cut-off according to the ROC curve analysis (1.90%; dotted line). (D) Cohen’s kappa curves for possible MS-qPCR cut-off points to analyze the agreement with pyrosequencing; (E) ROC curve analysis of the MS-qPCR (NMP) data to predict the dichotomous classifier based on pyrosequencing (PYRmean).