edgeR: a versatile tool for the analysis of shRNA-seq and CRISPR-Cas9 genetic screens

Introduction

Pooled library sequencing screens couple gene knock-down/editing technology with second generation sequencing to allow researchers to elucidate gene function in an unbiased, high-throughput manner^1,2. Several recent high impact studies have exploited this approach to discover novel genes involved in processes including cell fate decisions of normal and cancer cells, drug resistance, and to generate genetic interaction maps in mammalian cells using RNA interference (RNAi)^3–5 and sgRNAs with the clustered regularly interspaced palindromic repeats-Cas9 (CRISPR-Cas9) genome editing system^2,6.

Pooled screening relies on the stable genomic integration (often by viral transduction) of a library of uniquely identifiable expression constructs within a population of cells. Each construct expresses an RNA transcript that targets nuclease machinery to a specific nucleotide sequence. This is currently achieved in two main ways: shRNAs can be designed to target specific mRNA transcripts for degradation via the DICER/RISC pathway⁷ or sgRNAs can be designed to target a co-expressed Cas9 nuclease to a specific sequence in the genome⁸. By targeting constitutive exons at the 5’ region of a gene, Cas9-mediated double-stranded breaks are repeatedly repaired by nonhomologous end joining until a mutation is introduced that renders the site unrecognisable by the sgRNA. Such mutations typically comprise an insertion or deletion and can give rise to altered coding sequences, disrupted splice sites, frame shifts and/or premature stop codons in the target gene⁸.

Depending on the biological question of interest, typically two or more cell populations are compared either in the presence or absence of a selective pressure, or as a time-course before and after a selective pressure is applied. Gain of shRNA/sgRNA representation within a pool suggests that disrupting target gene function confers some sort of advantage to a cell. Similarly, genes whose knockdown/knockout is disadvantageous may be identified through loss of shRNA/sgRNA representation. Screening requires a library of constructs in a lentiviral or retroviral vector backbone that is used to generate a pool of virus for transducing cells of interest. The relative abundance of these constructs in transduced cells is then quantified by PCR amplification of proviral integrants from genomic DNA using primers designed to amplify all cassettes (shRNA/sgRNA) equally, followed by second-generation amplicon sequencing (Figure 1A). Sample-specific primer indexing allows many different conditions to be analyzed in parallel.

Figure 1. Summary of the raw data, workflow and diagnostic plots from edgeR.

(A) Structure of the amplicons sequenced in a typical shRNA-seq screen. Each amplicon will contain sample and hairpin specific sequences at predetermined locations. In sgRNA-seq screens, the amplicon sequences have a similar structure, with the sgRNA sequence replacing the hairpin. After sequencing, the raw data is available in a fastq file. (B) The main steps and functions used in an analysis of shRNA/sgRNA-seq screen data in edgeR are shown. (C) Example of a multidimensional scaling (MDS) plot showing the relationships between replicate dimethyl sulfoxide (DMSO) and Nutlin treated samples (data from Sullivan et al. (2012)⁴). MDS plots provide a quick display of overall variability in the screen and can highlight inconsistent samples. (D) Plot of log₂-fold-change versus hairpin abundance (log₂CPM) for the same data. Hairpins with a false discovery rate < 0.05 from an exact test analysis in edgeR (highlighted in red) may be prioritized for further validation.

As the popularity of these approaches grows, there is a need to develop suitable analysis pipelines to handle the large volumes of raw data that each screen generates. The major steps in an analysis involve processing the raw sequence reads, assessing the data quality and determining representational differences in the screen in a statistically rigorous way.

Two pipelines are currently available for this task that have been tailored for data from shRNA-seq screens. The shALIGN program⁹ is a custom Perl script that trims the sequence reads to the pre-defined base positions and then matches these to a library of hairpin sequences. Mismatch bases are permitted, and any ambiguous matches are ignored from the final hairpin count. Statistical analysis of the data is then performed using the shRNAseq R package⁹, which calculates log-ratios of the counts from each screen replicate, normalizes these values and ranks hairpins by their median, mean or t-statistic. Another solution is the BiNGS!SL-seq program¹⁰ that uses Bowtie to perform sequence mapping followed by statistical analysis in edgeR¹¹.

In this article, we describe a complete analysis solution for shRNA/sgRNA-seq screens accessible from within the edgeR package available from Bioconductor¹².

Implementation

A summary of the main steps in a typical shRNA/sgRNA-seq analysis alongside the functions in edgeR that perform each task is given in Figure 1B.

Case studies and further extensions

We provide example data sets and a complete analysis script that demonstrate how to use the edgeR package to prioritize data from four different shRNA-seq screens and two sgRNA-seq screens¹⁹. These examples were chosen to showcase edgeR’s ability to deal with experiments of varying size (from tens to thousands of genes) and complexity, from two-group situations, to settings with four groups, or a time-course design, where a GLM with a slope and intercept term is most appropriate. We have also developed a Galaxy tool^20–22 that implements this workflow as a point-and-click application to improve accessibility for researchers who are unfamiliar with the R programming environment (Figure 2).

Figure 2. Screenshots of the Galaxy tool for analyzing pooled genetic sequencing screens using edgeR.

(A) From the main screen, the user selects the appropriate input files and analysis options. (B) The results of an analysis are summarized in an HTML page that includes various diagnostic plots. (C) Output also includes a table of ranked results at the hairpin/guide and gene-level (where appropriate) as well as barcode plots (D) that highlight the ranks of hairpins/guides targeting a specific gene relative to all other hairpins/guides in the data set.

Discussion

Although the major functionality of edgeR has been developed with RNA-seq data in mind, the analysis of numerous in-house data sets¹⁹ and the results of others⁴ have demonstrated its utility for count data derived from pooled amplicon sequencing screens. edgeR provides users with a unique tool for the analysis of data from this emerging application of second generation sequencing technology, that is capable of handling both the biological variability and experimental complexity inherent in these screens. Provision of a Galaxy module puts these powerful statistical methods within reach of experimentalists. Future work will be focused on the use of a suitable control data set to compare this analysis pipeline with other approaches such as shRNAseq⁹.

Software availability