Fig 1.
Schematic of experimental design to measure stability of microarray experiments. (A) Schematic of experimental design. Briefly, for each human subject, DNA was isolated from eight replicate blood aliquots, and then independently bisulfite converted, amplified, and frag-mented. An aliquot of DNA from each of the eight replicates was combined into a pooled DNA sample. Both independently isolated and pooled tech-nical replicates were measured across using Infinium MethylationEPIC microarrays. (B) Layout of pooled and independently prepared samples from each subject (Letters) with regard to chamber number/Sentrix posi-tion (numbers) within each slide/Sentrix Barcode (Roman numerals). Samples were used for independent statistical assessments as indicated.
Fig 2.
Multivariate Analysis and Statistical Comparison of Positional Effects on CpGs in the Illumina MethylationEPIC platform (A) Hierarchical clustering dendrogram representing the clustering of subjects based on the first 50 principal components from methylation beta values after preprocessing with SeSAMe. The vertical axis represents the dissimilarity between subjects, while the horizontal axis shows the subjects’ labels. (B) Principal component analysis (PCA) plot, representing the first principal component (PC1) against the second principal component (PC2) for each sample. Each data point represents an individual technical replicate, with colors indicating different subjects. (C-D) PCA plots showing the first two principal com-ponents for each sample, with colors indicating different chamber numbers (C) or different slides (D). (E) Hierarchical clustering dendrogram represent-ing the clustering of subjects based on the first 50 principal components from adjusted beta values after preprocessing with SeSAMe and ComBat adjust-ments for chamber number. (F-H) PCA plots showing clustering of subjects after ComBat adjustment for chamber number, colored by subject (F), cham-ber number (G), or array (H). (I-J) Box plots presenting F statistics obtained from analysis of variance (ANOVA) performed on beta values preprocessed either with SeSAMe alone (I) or with SeSAMe + ComBat using chamber number as batch (J). The box represents the interquartile range (IQR), with the median indicated by a line inside the box. Whiskers extend to the minimum and maximum values within 1.5 times the IQR. Data points beyond 1.5 times the IQR are plotted individually.
Fig 3.
Differential methylation analysis of alternatively preprocessed methylation beta values from technical replicates. (A) Ratio of averaged standard deviation within subjects to standard deviation across all subjects for 614,831 CpGs shared across all preprocessing pipelines shown. (B-C) Representation of samples used for differential methylation testing in comparison group 1. Experimental subjects B and D were each represented by n = 4 technical replicates, with each replicate located on a different array for comparison group 1 (B) and each replicate on the same array for group 2 (C). (D-K) Histograms depicting BH adjusted p-values for all CpGs after differential methylation testing of beta values using the specified preprocessing pipeline for group 1 (in red box) and group 2 samples (in light blue box). Number of CpGs with BH-adjusted p-values < 0.05 listed on upper right of each plot. (L-O) Venn diagrams representing the number of CpGs detected as significantly different between subjects B and D in Group 1 (red circle), Group 2 (blue circle), or in both sets of samples (overlap), using the specified preprocessing pipeline. (P) Representation of samples used for differential methylation testing to screen for the introduction of false positives due to positional effects. (Q) Upset plot relating the number of false positives within and between preprocessing pipelines (limited to 9-most abundant comparisons). (R-U) Histograms depicting BH adjusted p-values for all CpGs after differential methylation testing of beta values using the specified preprocessing pipeline. Number of CpGs with BH-adjusted p-values < 0.05 listed on upper right of each plot.
Fig 4.
Fluorescence values compared in technical replicates measured in different chamber numbers.
(A) Configuration of samples within slides/chambers. The average of four within-subject centered and scaled fluorescence intensity values for each CpG/chamber number were extracted from slides, as depicted. (B-C) Distribution histograms illustrating within-subject centered and scaled fluorescence intensities (FI) for all Type II probes on the MethylationEPIC array in the red (B) and green (C) channels. (D-E) Mean and standard deviation of fluorescence intensities (not centered or scaled) from all type 2 probes within in each chamber number. (F-G) Boxplots comparing the coefficients of variation from bead-level measurements (standard devia-tion/mean) for unprocessed FIs. Data represent all type 2 probes across each of the eight chamber numbers in the Red (F) and Green channels (G). Values for each chamber were calculated from measurements of four subjects, apart from chamber three which had one outlier sample removed based on quality concerns. Outlier values (>1.96 IQR) were hidden from boxplots due to the large size of the dataset.
Fig 5.
Beta values compared in technical replicates measured in different chamber numbers.
(A) Configuration of samples within slides/chambers. The average of four within-subject centered and scaled beta values for each CpG/chamber number were extracted from slides, as de-picted. (B-D) Histograms representing the distributions of within-subject centered and scaled beta values for all probes on the MethylationEPIC array. The histograms correspond to different preprocessing steps: (B) raw beta values, (C) beta values preprocessed using SeSaMe’s recommended settings, and (D) SeSaMe-preprocessed beta values with adjustment for chamber number batch effects using ComBat.
Fig 6.
Relative brightness in technical replicates related to probe intensity.
Plots depicting Type 2 probes binned according to median within-subject fluorescence intensity (FI). The y-axis represents differences from the median value for each bin. Each point shown represents one percentile bin from one of the four subjects in the study. Differences between median intensity are represented for the Red FI, Green FI, and beta values in Raw (A-D), SeSAMe-normalized (E-H), or ComBat-adjusted data (I-L). Axes for figure panels were intentionally limited for the purposes of visualization, resulting in the omission of some data points which had very high variability.
Fig 7.
Variability in FI for epigenetic clock CpGs.
(A-B) Scatter plots displaying the mean vs range of beta values for CpGs included in the Horvath Skin and Blood clock. Each point represents summary statistics for a single CpG, calculated across technical replicates from one of the four subjects in the study. The size of each point is proportional to the coefficient value in the clock model. (C) Summary statistics for low-level measurements of cg03184882 within each sample in the study. The outlier sample is highlighted in red.
Fig 8.
Variability in DNA methylation clock age predictions from technical replicate samples.
Scatter plots depict age predictions from 6 different DNAm clocks on technical replicates. Colors of dots and lines indicate predicted age and actual age of subjects, respectively. Red point indicates outlier sample highlighted in panel C. (A) Age predictions generated using raw beta values. (B) Age predictions generated using beta values processed with the ‘recommended’ SeSaMe preprocessing settings. (C) Age predictions generated using SeSaMe-preprocessed beta values, with chamber number batch effect correction applied using ComBat.