No more mixed signals: Improved ChIP-seq data analysis with greenscreen

Activation and repression of the genome are key steps in plant development and responses to the environment. During activation, factors bind to the DNA, the chromatin opens and transcription initiates. Chromatin immunoprecipitation followed by sequencing (ChIP-seq) is used to identify factors that bind to genomic DNA and chromatin modifications. ChIP-seq consists of crosslinking factors to chromatin, shearing of the genomic DNA, immunoprecipitating factor- or modification-associated DNA fragments, reverse crosslinking, and sequencing (Kaufmann et al., 2010). Input controls consist of genomic DNA that undergo all steps above, except no immunoprecipitation is performed. Despite the use of experimental controls, one of the bottlenecks in ChIP-seq experiments is the high false peak detection rate caused by amplified artifactual signals. These artifactual signals, tabulated on the so-called blacklists, should be removed from the dataset to facilitate the estimation of sample quality, replicate concordance, and identification of factor binding sites (Amemiya et al., 2019). Blacklists exists for human, mouse, and fruit flies; however, they have not yet been developed for many model and nonmodel species.

In this issue of The Plant Cell, Samantha Klasfeld and colleagues (Klasfeld et al., 2022) present a new tool, called greenscreen, to remove problematic signals from ChIP-seq datasets from multiple organisms. First, the authors used 20 high-quality input datasets from the model plant Arabidopsis thaliana that have been collected in different tissues by different laboratories and the recently published ENCODE blacklist (Amemiya et al., 2019) to generate a blacklist for Arabidopsis. They then removed reads that overlapped with the blacklist and employed the routinely used MACS2 tool (Zhang et al., 2008) for ChIP-seq peak identification. For a more facile and versatile approach to artifactual signal removal, they employed MACS2 to call “peaks” in the input control samples using the entire Arabidopsis genome as background (Figure). By optimizing the peak calling significance threshold, merging regions that contained artifactual peaks within a certain distance, and restricting the filtering masks to regions where significant peaks were called in for at least half of the input samples, the authors developed a greenscreen filter that minimized the number of genomic regions and genes masked out. Then they compared greenscreen and blacklist artifactual region outputs and found that greenscreen fully overlapped with the blacklist, regions in both filters were significantly more likely to be located near assembly gaps, and both tools were equally effective at removing artifactual signals. However, the biggest difference between the known blacklist and their greenscreen tool was that the blacklist identified broader regions and may over-mask signals, potentially leading to more false negatives.

Figure.

ChIP-seq data analysis including the greenscreen workflow. Adapted from Klasfeld et al. (2022), Figure 1.

The authors also assessed the effect of greenscreen on ChIP-seq replicate reproducibility by comparing ChIP-seq data from the same proteins that were produced in different laboratories. They used Pearson correlation and Principal Component Analysis and found that without artifactual signal removal the samples did not cluster according to the biological expectation. When greenscreen or blacklist masking were applied the true correlation structure between samples emerged. By testing greenscreen performance with fewer inputs, the authors found that it can be applied to remove artifactual signals from a single ChIP-seq experiment with at least two inputs. After that, the authors used samples from ChIP-seq experiments performed by different laboratories using large and/or repetitive genomes, such as of human cell lines and of rice (Oryza sativa), and found that greenscreen was also effective in removing artifactual signal from these datasets.

Finally, Klasfeld et al. (2022) present a versatile new ChIP-seq analysis pipeline that incorporates greenscreen (Figure) and that is not only effective in removing false positives but also enriches true binding peaks. Greenscreen allows the identification of additional biologically relevant target genes and accurate detection of changes in factor occupancy, making ChIP-seq data analysis more thorough and reliable for model and nonmodel species.