To mock or not: a comprehensive comparison of mock IP and DNA input for ChIP-seq

Jinrui Xu, Michelle Kudron, Alec Victorsen, Jiahao Gao, Haneen Ammouri, Fábio C. P. Navarro, Louis Gevirtzman, R Waterston, Kevin P. White, V Reinke, Mark GersteinbioRxiv 2019 · DOI

TL;DR

ChIP-seq is a powerful technique for mapping protein-DNA interactions, but it's plagued by spurious binding sites. This paper shows that mock IP controls are superior to DNA input controls for removing these spurious sites, especially in complex samples like tissues and whole organisms. They also propose a novel method combining both controls for improved binding site detection.

What Problem Does This Paper Address?

Chromatin immunoprecipitation followed by sequencing (ChIP-seq) is the gold standard for identifying where proteins bind to DNA across an entire genome. This is crucial for understanding gene regulation and cellular function. However, ChIP-seq experiments are prone to generating "spurious" or false positive binding sites. These arise from two main sources: uneven shearing of DNA during sonication (which can preferentially fragment open chromatin regions) and non-specific binding of the antibody to proteins other than the target.

Traditionally, two types of controls are used to correct for these biases:

  1. DNA input control: This involves sequencing DNA that has undergone sonication but not the antibody-based enrichment step. It primarily corrects for sonication bias.
  2. Mock IP control: This uses the same antibody but either without the target protein present or with an antibody that doesn't bind the target (e.g., IgG). This control is designed to capture both sonication and non-specific antibody binding.

While mock IP theoretically addresses both sources of bias, it's often more susceptible to technical noise. Consequently, DNA input controls have become the de facto standard, despite their limitations in removing non-specific binding. This paper investigates whether mock IP controls are indeed superior and explores how to best leverage both control types.

Methods

The authors employed a comprehensive comparison of ChIP-seq data generated with DNA input controls versus mock IP controls across various sample types (human cell lines, worm, and fly). They developed computational experiments to quantify spurious sites and proposed a novel Bayesian framework to integrate signals from both DNA input and mock IP controls for improved binding site detection. This framework uses a probability score to rank potential binding sites, which is then further refined using the Irreproducible Discovery Rate (IDR) tool. They also analyzed motif enrichment within predicted binding sites as a measure of their biological relevance.

Datasets

The study utilized publicly available ChIP-seq datasets from the ENCODE portal for human cell lines, tissues, and organs. They also incorporated data generated by their own modERN consortium for worm and fly developmental stages. These datasets included paired IP, DNA input, and mock IP experiments. RNA-seq data was used to assess transcriptome activity, and DNase-seq and ATAC-seq data were used to measure genome accessibility.

Key Insights and Ideas

  • Mock IP is better at removing spurious sites: The paper strongly demonstrates that mock IP controls are significantly more effective than DNA input controls at removing spurious binding sites, particularly in complex biological samples like whole organisms.
  • Spurious sites are linked to sample properties: The abundance of spurious sites is not random but is tightly correlated with intrinsic sample properties like genome accessibility and transcriptome activity. More accessible and transcriptionally active genomes (e.g., in tissues, organs, or whole organisms) tend to have more spurious sites.
  • DNA input alone is insufficient for complex samples: While DNA input controls are informative, they fail to adequately address the non-specific binding that contributes to spurious sites, especially when dealing with samples that have high genome accessibility.
  • Combining controls offers the best of both worlds: A novel Bayesian framework that integrates both DNA input and mock IP controls, along with multiple scoring metrics, provides a more robust and accurate method for binding site detection.

Key Results

  • Human cell lines, on average, had 9 spurious sites per 100 million base pairs (Mb), while worm and fly samples showed significantly higher numbers (551 and 3,931 per 100 Mb, respectively).
  • Spurious site abundance was highly correlated with transcriptome activity (r² = 0.92) and genome accessibility (r² = 0.79 for human, r² = 0.99 for worm).
  • Using mock IP controls reduced the number of detected binding sites compared to DNA input controls, with substantial reductions observed in worm and fly samples (18-37% improvement in motif enrichment for top binding sites).
  • The proposed combined control method improved binding site detection quality by 8-21% compared to using mock IP alone.
  • Activating TFs, particularly in fly embryos, showed a tendency to have more binding sites detected by mock IP controls and fewer spurious sites when using DNA input controls, suggesting a difference in binding behavior.

Limitations and Future Work

  • Reference sites for bona fide identification: The authors acknowledge that estimating the prevalence of persisting spurious sites in IP experiments is difficult because the true "bona fide" sites are unknown, and their estimation relies on the accuracy of reference sites.
  • Technical noise in mock IP: While mock IP is effective, it can be more susceptible to technical noise than DNA input due to lower DNA yields.
  • Future work: The paper introduces a novel framework that combines mock IP and DNA input controls with multiple scoring metrics. This framework is presented as a valuable asset for improving binding site detection and enabling better understanding of TF roles. Further development and application of this method to large-scale projects like modERN are implied. The authors also suggest that the observed differences in spurious site abundance across species are due to biological properties, which warrants further investigation.

Summary updated

Abstract

ABSTRACT Chromatin immunoprecipitation (IP) followed by sequencing (ChIP-seq) is the gold standard to detect genome-wide DNA-protein binding. The binding sites of transcription factors facilitate many biological studies. Of emerging concern is the abundance of spurious sites in ChIP-seq, which are mainly caused by uneven genomic sonication and nonspecific interactions between chromatin and antibody. A “mock” IP is designed to correct for both factors, whereas a DNA input control corrects only for uneven sonication. However, a mock IP is more susceptible to technical noise than a DNA input, and empirically, these two controls perform similarly for ChIP-seq. Therefore, DNA input is currently being used almost exclusively. With a large dataset, we demonstrate that using a DNA input control results in a definable set of spurious sites, and their abundance is tightly associated with the intrinsic properties of the ChIP-seq sample. For example, compared to human cell lines, samples such as human tissues and whole worm and fly have more accessible genomes, and thus have more spurious sites. The large and varying abundance of spurious sites may impede comparative studies across multiple samples. In contrast, using a mock IP as control substantially removes these spurious sites, resulting in high-quality binding sites and facilitating their comparability across samples. Although outperformed by mock IP, DNA input is still informative and has unique advantages. Therefore, we have developed a method to use both controls in combination to further improve binding site detection.
BibTeX
@misc{xu2019mock,
  author = {Jinrui Xu and Michelle Kudron and Alec Victorsen and Jiahao Gao and Haneen Ammouri and Fábio C. P. Navarro and Louis Gevirtzman and R Waterston and Kevin P. White and V Reinke and Mark Gerstein},
  title = {{To mock or not: a comprehensive comparison of mock IP and DNA input for ChIP-seq}},
  year = {2019},
  doi = {10.1101/2019.12.17.880013},
  howpublished = {bioRxiv},
}

Loading related content...

Figure