Comparative analysis of affinity-based 5-hydroxymethylation enrichment techniques

The epigenetic modification of 5-hydroxymethylcytosine (5hmC) is receiving great attention due to its potential role in DNA methylation reprogramming and as a cell state identifier. Given this interest, it is important to identify reliable and cost-effective methods for the enrichment of 5hmC marked DNA for downstream analysis. We tested three commonly used affinity-based enrichment techniques; (i) antibody, (ii) chemical capture and (iii) protein affinity enrichment and assessed their ability to accurately and reproducibly report 5hmC profiles in mouse tissues containing high (brain) and lower (liver) levels of 5hmC. The protein-affinity technique is a poor reporter of 5hmC profiles, delivering 5hmC patterns that are incompatible with other methods. Both antibody and chemical capture-based techniques generate highly similar genome-wide patterns for 5hmC, which are independently validated by standard quantitative PCR (qPCR) and glucosyl-sensitive restriction enzyme digestion (gRES-qPCR). Both antibody and chemical capture generated profiles reproducibly link to unique chromatin modification profiles associated with 5hmC. However, there appears to be a slight bias of the antibody to bind to regions of DNA rich in simple repeats. Ultimately, the increased specificity observed with chemical capture-based approaches makes this an attractive method for the analysis of locus-specific or genome-wide patterns of 5hmC.


S1.
Comparison of hmeDIP 5hmC profiles following whole genome sequencing approaches (Upper track -Shen et al 2013 hmeDIP-seq on mouse ES cells) to our high density tiled microarray profiling in the mouse brain. Although the patterns of 5hmC distribution appear largely similar at both the chromosomal level between ES cells and brain tissues (i & ii) genes which are unique in their transcriptional activities (such as the ES transcribed HoxA cluster) harbour distinct 5hmC profiles (iii).
Many gene remain similarly expressed (such as at the microtubule associated gene Mtap1s -iv) and as such reveals similar patterns of 5hmC distribution. At such loci both hmeDIP-seq and hmeDIParray report very similar patterns, indicating that using high resolution microarrays are viable is a cost effective method of accurately reporting on the distribution of 5hmC marked DNA. Microarray data is plotted on log2 scale from -2 to +2 whilst sequencing data plotted on the number of reads from 0 to 20. Refseq genes are annotated below (blue bars).Scale bars shown to the top right.
S2. JBP-1 affinity purification of 5hmC results in lower DNA yield than alternative techniques but does return similar patterns of 5hmC enrichment. (a) Cycle threshold plots of individual qPCR amplified samples over 3 loci (5hmC negative: Gapdh, 5hmC positive: H19 and Tex19.1 promoters) in the mouse liver. Triplicates are plotted to show technical replicates of amplification whilst for JBP-1 a separate enrichment was carried out to ensure the low levels of DNA returned was not due to a kit malfunction. JBP-1 purification 1 = yellow, JBP-1 purification 2 = red, hmeDIP = purple, hMe-seal = blue. (b) Non-normalised data following qPCR amplification of JBP-1 enriched DNA. Non-specific background was removed by subtracting the amplification values of DNA which had not been glucosylated from that which did. Plots are shown relative to the amplification of input DNA (% purified / input each following background subtraction). Although the patterns look encouraging the absolute levels of enrichment are extremely low (all <0.04% of the input).

S3.
Microarray patterns of 5hmC enrichment following JBP-1 purification reveal high levels of background noise. Data plotted on log2 scale from +3 to -3. Coordinates of the three loci are displayed on the right. JBP-1 enrichment = orange, hmeDIP enrichment = purple, hMe-seal enrichment = blue, depletion = grey, gene structures = blue bars.

S4.
Plot of total number of 5hmC enriched "peak probes" found following hmeDIP, hMe-seal or JBP-1 affinity purification in brain and liver tissues. The vast majority of the genes did not differ by more than log2 0.5 fold across the two techniques.
Genes with greater than log2 0.5 fold changes in either direction were selected for further analysis.

(b)
Genome browser visualisation reveals that many of the hmeDIP enriched genes are found over regions of low 5hmC, indicating higher levels of noise is introduced by the antibody based approach.
Genes denoted below are blue bars. (c) Examples of hMe-seal specific enriched genes. Genome browser visualisation of brain microarray datasets. Coordinates are given above each region. All plots are on log2 scale between +3 and -3. Purple bars = hmeDIP enrichment, blue= hMe-seal enrichment, grey = 5hmC depletion, green= hmeDIP specific 5hmC bias, red= hMe-seal specific bias. Genes denoted below are blue bars.

S8.
Regions of 5hmC enrichment are associated with select histone modifications in the mouse liver.

S11.
The strong region of technique dependant enrichment between H19 and Igf2r corresponds to several repetitive elements as well as a region of TC tandem repeats. Purple bars = hmeDIP enrichment, blue= hMe-seal enrichment, grey = 5hmC depletion, green= hmeDIP specific 5hmC bias, red= hMe-seal specific bias, SINEs= blue boxes, LTRs = red boxes, DNA repeat elements = green box, TC simple tandem repeats = black box.

S12.
Profiles of 5mC enriched liver DNA over the H19/Igf2r locus reveals similarly high levels of the modification between the two genes. Unlike 5hmC, this elevation in 5hmC can be independently validated by gRES-qPCR (boxed region below). Purple tracks = hmeDIP enrichment, red= MeDIP enrichment, grey = 5hmC or 5mC depletion. Genes are displayed as blue bars below with simple repeats shown as black bars. Boxed highlighted area denotes regions tested by gRES-qPCR. Boxed bar plots represent the percentage of each modification at a single CpG in the sequence CCGG following normalisation (purple; 5hmC, red; 5mC, green; C). Error bars display the standard error of the biological replicates.