A reassessment of DNA-immunoprecipitation-based genomic profiling

Abstract

DNA immunoprecipitation followed by sequencing (DIP-seq) is a common enrichment method for profiling DNA modifications in mammalian genomes. However, the results of independent DIP-seq studies often show considerable variation between profiles of the same genome and between profiles obtained by alternative methods. Here we show that these differences are primarily due to the intrinsic affinity of IgG for short unmodified DNA repeats. This pervasive experimental error accounts for 50-99% of regions identified as 'enriched' for DNA modifications in DIP-seq data. Correction of this error profoundly altered DNA-modification profiles for numerous cell types, including mouse embryonic stem cells, and subsequently revealed novel associations among DNA modifications, chromatin modifications and biological processes. We conclude that both matched input and IgG controls are essential in order for the results of DIP-based assays to be interpreted correctly, and that complementary, non-antibody-based techniques should be used to validate DIP-based findings to avoid further misinterpretation of genome-wide profiling data.

Figure 1

Characterization of off-target antibody binding in DIP-seq. (a) Signal track in mESCs showing similar enrichment between 5modC and IgG DIP-seq samples over repetitive regions. WGBS, whole-genome bisulfite sequencing; STRs, short tandem repeats. (b) Signal track of 5mC, 5hmC and IgG DIP-seq in DNMT triple knockout (TKO) or wild-type (WT) mESCs over 5hmC-(left) or IgG enriched regions (right). (c) Mass spectrometry quantification of 5mC and 5hmC in TKO and WT mECSs for n = 3 biologically independent samples. Data shown as mean ±s.d. P-values calculated using two-tailed T-test. (d) DIP using a 5hmC antibody in wild-type (WT) (left) and DNMT^TKO (right) mESCs for DIP-qPCR n = 3 and DIP-seq n = 1 biologically independent samples. Data represented as in c. Correlation between mean DIP-qPCR and DIP-seq values calculated using two-tailed Spearman correlation. STRs, short tandem repeats. (e) 5hmC enrichment in mESCs with different profiling techniques over 5hmC n = 31265 enriched regions (left) or IgG n = 137557 enriched regions (right). (f) Consensus motif enrichment for raw IgG reads compared to Input of n = 3 biologically independent samples.

Figure 2

Characterization of similarities between 6mA and IgG DIP-seq in different species. (a) Signal track for Input and 6mA DIP-seq in mouse tissues and IgG DIP-seq in mESCs. STRs, short tandem repeats. (b) Enrichment over IgG enriched DIP-seq regions for 6mA DIP-seq n = 11 and Input n = 4 biologically independent samples. P-values calculated using two-tailed T-test. Boxplots represent median and first and third quartiles with whiskers extending 1.5 * inter-quartile range. (c) Fraction of DIP-seq enriched regions located in short tandem repeats (STRs) for 6mA n = 11, TKO n = 3, Input n = 4 and 5hmC n = 6 biologically independent samples. P-values calculated from biologically independent samples using pairwise two-tailed T-tests with Benjamini-Hochberg correction for multiple testing. Data represented as in b. (d) Motif enrichment for raw 6mA or IgG DIP-seq reads compared to Input in multiple species. Motif with highest correlation to IgG motifs shown for each cell type and antibody. (e) Fraction of motifs highly similar (r > 0.75) to mouse IgG motifs for M. musculus n = 11, D. rerio n = 2, X. laevis n = 8, C. elegans n = 1 and E. coli n = 2 biologically independent samples. Data represented as in b. (f) Proportion of CA-repeats in the genomes of model organisms.

Figure 3

Biological impact of IgG correction. (a) Estimated false positive rate of DIP-seq enriched regions using IgG or Input as control in mESCs for 5caC n = 2, 5fC n = 2, 5hmC n = 7 and 5mC n = 6 biologically independent samples. Data shown as mean ±s.d. (b) Overlap of 5hmC and 5mC regions using IgG or Input controls showing decreased overlap when using IgG controls. Venn diagram of 5mC and 5hmC overlap using IgG or Input controls (top) and paired line plot of 5mC and 5hmC overlap using IgG or Input controls for multiple studies (indicated by symbols, bottom). Data shown as mean and individual data points of n = 6 biologically independent samples. P-values calculated using two-tailed paired T-test. ▲ = ERP000570, ● = GSE31343, ■ = GSE24841, ▼ = GSE42250. (c) TET1 binding over IgG n = 137557 enriched regions or 5hmC n = 31265 enriched regions using IgG or Input controls. (d) GO term enrichment for top genes (n = 500) enriched for 5hmC using DIP-seq with either IgG or Input controls or 5hmC-Seal or anti-CMS techniques. P-values calculated using PANTHER biological processes. (e) Relative enrichment of ENCODE mESC histone ChIP-seq data for n = 26 biologically independent samples in regions enriched for IgG in DIP-seq or random regions of same size and chromosome. Boxplots represent median and first and third quartiles with whiskers extending 1.5 * inter-quartile range. P-values calculated using two-tailed T-test. (f) Enrichment of ENCODE mESC histone ChIP-seq data for 5hmC-(left) or 5mC (right) enriched regions using IgG or Input as controls. Data presented as mean (IgG) and bootstrapped mean (Input) of H3K27ac n = 2, H3K36me3 n = 4, H3K4me1 n = 6, H3K4me3 n = 4, H3k9ac n = 2, H3K27me3 n = 2, H39me3 n = 4 biologically independent samples, #P<1e-5, bootstrap resampling (n = 100,000).

Figure 4

Antibodies in DIP-seq experiments bind repetitive elements which are incorrectly identified as enriched regions when not controlled for IgG binding.