Abstract
Chromatin immunoprecipitation followed by sequencing (ChIP-seq) is widely used to detect genome-wide interactions between a protein of interest and DNA in vivo. Loci showing strong enrichment over adjacent background regions are typically considered to be sites of binding. Insufficient attention has been given to systematic artifacts inherent to the ChIP-seq procedure that might generate a misleading picture of protein binding to certain loci. We show here that unrelated transcription factors appear to consistently bind to the gene bodies of highly transcribed genes in yeast. Strikingly, several types of negative control experiments, including a protein that is not expected to bind chromatin, also showed similar patterns of strong binding within gene bodies. These false positive signals were evident across sequencing platforms and immunoprecipitation protocols, as well as in previously published datasets from other labs. We show that these false positive signals derive from high rates of transcription, and are inherent to the ChIP procedure, although they are exacerbated by sequencing library construction procedures. This expression bias is strong enough that a known transcriptional repressor like Tup1 can erroneously appear to be an activator. Another type of background bias stems from the inherent nucleosomal structure of chromatin, and can potentially make it seem like certain factors bind nucleosomes even when they don't. Our analysis suggests that a mock ChIP sample offers a better normalization control for the expression bias, whereas the ChIP input is more appropriate for the nucleosomal periodicity bias. While these controls alleviate the effect of the biases to some extent, they are unable to eliminate it completely. Caution is therefore warranted regarding the interpretation of data that seemingly show the association of various transcription and chromatin factors with highly transcribed genes in yeast.