Genomic DNA was extracted using Qiagen Plant DNA Easy kit from 100 14 day-old seedlings and sheared to an average size of 1-3kb using a nebulizer. DNA was digested by McrBC and run side-by-side on a small 1% agarose gel, along with a mock-treated sample of the same DNA. DNA fractions larger than 800-1000bp were cut from the gel and DNA purified using Qiagen Qiaex II resin. In this way, the McrBC-treated sample represents enrichment of un-methylated portion of the genome.
Label
Cy5 and Cy3
Label protocol
Three micrograms of the DNA sample were labeled with Cy3 or Cy5 dyes by random priming and purified over Qiagen PCR cleanup columns according to the supplier’s protocol.
Genomic DNA was extracted using Qiagen Plant DNA Easy kit from 100 14 day-old seedlings and sheared to an average size of 1-3kb using a nebulizer. DNA was digested by McrBC and run side-by-side on a small 1% agarose gel, along with a mock-treated sample of the same DNA. DNA fractions larger than 800-1000bp were cut from the gel and DNA purified using Qiagen Qiaex II resin. In this way, the McrBC-treated sample represents enrichment of un-methylated portion of the genome.
Label
Cy5 and Cy3
Label protocol
Three micrograms of the DNA sample were labeled with Cy3 or Cy5 dyes by random priming and purified over Qiagen PCR cleanup columns according to the supplier’s protocol.
Hybridization protocol
Equal amounts of Col DNA labeled with Cy3 or Cy5 and Ler DNA labeled with Cy5 or Cy3, respectively, were mixed for hybridization to a single array in 4X SSC and 0.2% SDS. Arrays were hybridized at 65C for 16-18 hrs, followed by 4 washes (1X SSC 0.2% SDS for 5 minutes at 42C, followed by 1X SSC at room temperature for 5 minutes and two washes in 0.1X SSC at room temperature for 2 minutes).
Scan protocol
Arrays were immediately scanned using Axon 4000 array scanner and the scans were quantified with GenePix v4.0.
Description
A reference design including a dye reversal was used to examine DNA cytosine methylation on Arabidopsis chromsome 4 in the Col-o ecotype. Three technically replicated dye swaps were performed for a total of 6 hybridizations.
Data processing
The methylation intensity for each spot is computed by subtracting the background median intensity from the corresponding foreground median intensity for both r signals. If the foreground intensity is less than the background intensity, the methylation level is set as 1 (log-transformed intensity 0). To stabilize the variance, log2 transformation is applied to the background corrected intensities. MA plots revealed a general trend across all the slides for the Cy5 to yield higher measurements than Cy3. To correct this, a robust local regression (loess) was used to remove the unbalanced dye effect using all features on the array. After the normalization, the log-ratio, M, has a mean of approximately zero.
Methylation genomic tiling microarray data were analyzed using a linear model analysis of variance (ANOVA). Specifically, a linear model of the form
ln(Yijkmr)= µ + Ai +Di +Tk+Gm+AGim+DGjm+TGkm +εijkmr
was employed to partition the sources of variation. The natural logarithm transformed, raw, background corrected data are denoted as ln(Yijkmr). The overall mean effect is µ, and A, D, T, and G represent the array, dye, treatment, and gene (or feature) main effects, respectively. The interactions of the main effects are AG, DG, TG, and represent array by gene, dye by gene, and treatment by gene, respectively. The random error εijkmr is assumed to be normally distributed, with mean zero and constant variance. Once the sources of technical variation (e.g., global and feature-specific array and dye effects) and experimental variation (e.g., treatment and treatment by feature interaction) are estimated, differential fluorescence was tested using hypotheses that acknowledge both the average treatment effect and the treatment by feature interaction.
H0:Tk+TGkm=Tk’+TGk’m
Ha: Tk+TGkm ≠Tk’+TGk’m
Statistically significant differences were determined by evaluating signals at each tile using a one-sided t-test. 576 randomly selected tiles across chromosome 4 regions with no gene or repeat annotation were used as controls. These controls supplied an average unmethylated value against which other features on chromosome were tested using a one-sided t-test. The multiple testing issues that arise when testing 21,815 tiles across chromosome 4 the false discovery rate (FDR) were controlled at the 5% level using a Benjamini and Hochberg correction.
