CRISPR Screens: Analysis of Sequencing Data

Introduction to Functional Genomics and CRISPR-Cas9

  • Functional Genomics

    • Definition: The process of determining the functional link between a specific gene and the role it plays in cellular processes.

    • Context: Various techniques and applications exist to study gene function, with a current focus on the CRISPR-Cas9 system.

  • The CRISPR-Cas9 System

    • Full Name: Clustered Regularly Interspaced Short Palindromic Repeats/CRISPR-associated protein 9 system.

    • Cas9 Enzyme: Functions as a "pair of scissors" that cuts double-stranded DNA at a precisely defined location in the genome.

    • Guide RNA (gRNA): A short, predesigned RNA sequence embedded within a longer RNA scaffold. Its purpose is to guide the Cas9 enzyme to the intended genomic target.

    • Single Guide RNA (sgRNA): The combination of at least one guide RNA with one scaffold RNA.

  • Mechanism of CRISPR-Cas Action

    • Step 1: The Cas9 enzyme binds to the guide RNA.

    • Step 2: The guide RNA binds to the specific target DNA sequence.

    • Step 3: The Cas9 enzyme performs a double-strand break (cuts both strands of DNA).

CRISPR Screen Modalities and Mechanisms

  • CRISPR Knockout (KO) Screens

    • Principle: Systematically knocking out every gene that could be of importance, typically targeting only one gene per individual cell.

    • Timeline: Cells are screened after a designated period (e.g., a few days).

    • Molecular Outcome: Cas9-mediated DNA cutting leads to insertions or deletions (indels) following imprecise cellular repair mechanisms.

    • Functional Outcome: These indels result in nonfunctional proteins or nonsense mRNA, leading to a complete loss of gene function.

  • CRISPR Interference (CRISPRi) Screens

    • Objective: To repress the transcription of potentially important genes (one gene per cell) without necessarily cutting the DNA.

  • CRISPR Activator (CRISPRa) Screens

    • Objective: To increase the transcription levels of potentially important genes (one gene per cell).

Library Construction and Experimental Design

  • Library Types

    • Pooled Screens: A single population of cells is transduced with a mixed library containing all sgRNAs in one pool.

    • Arrayed Screens: Every gene is targeted separately across different wells of a multiwell plate.

  • Workflow of Pooled Library Construction

    • Libraries are designed with multiple sgRNAs per gene of interest (or targeting the whole genome).

    • The pool of sgRNAs is transferred into a viral plasmid.

    • The resulting virus is utilized to transduce the target cells.

    • Library Size: Determined by the number of sgRNAs per gene multiplied by the total number of targeted genes.

  • Multiplexed CRISPR Screens

    • Involves paired sgRNA libraries targeting pairs of genes to study genetic interactions.

Phenotype Screening and Selection Types

  • Relative Viability Assessment

    • After transduction, successful sgRNAs of interest are identified by monitoring their relative abundance in the population.

  • Negative Selection (Dropout Screens)

    • Identifies genes whose loss leads to cell death or slow growth.

    • Example: Screening for essential genes or drug efficacy in cancer cells.

    • Phenotype: Live cells are retained while dead (depleted) cells are removed.

  • Positive Selection (Nondropout Screens)

    • Identifies genes whose modification provides a growth advantage.

    • Example: Drug resistance screens or toxin resistance screens.

    • Phenotype: Differentiation screens where cells are sorted (e.g., Type A vs. Type B).

Sequencing and Computational Analysis Workflow

  • Sequencing (NGS)

    1. Next Generation Sequencing (NGS): Considered the standard approach for quantifying sgRNA abundance.

    2. Quality Control (QC): Similar protocols to mRNA-sequencing.

    3. Mapping: Similar to mRNA-sequencing, though no splicing logic is required.

    4. Summarization: Similar to mRNA-sequencing.

  • Computational Analysis Steps

    1. Identification of Genes: Calculate the relative enrichment of every sgRNA comparing treatment groups to control groups.

    2. Aggregation: Combine sgRNA-level effects into gene-level statistics.

    3. Downstream Analysis: Includes visualization, Gene Set Enrichment Analysis (GSEA), and pathway analysis.

Statistical Challenges in Enrichment and Depletion Analysis

  • General Challenges

    • High-dimensional Data: Estimating data variation is difficult. Information is shared across sgRNAs via empirical Bayes methods.

    • Sample Differences: Unwanted noise must be removed while preserving biological relevance through normalization, offsets, or specific statistical model parameters.

  • CRISPR-Specific Challenges

    • Off-target Effects: Managing False Discovery Rate (FDR) control.

    • sgRNA Efficiency: Different sgRNAs have varying levels of effectiveness; this is mitigated when aggregating to the gene level.

    • Baseline Differences: Accounting for differences at $t_0$ (initial time point) requires measuring relative changes and accounting for correlations in paired data.

  • Count-Data Challenges

    • Dispersion: Data may exhibit overdispersion or underdispersion. Models must be selected to account for these specific distributions.

    • Zero-inflation: The presence of many zero counts.

      • Solutions: Filtering or using zero-inflated models like ZIP (Zero-Inflated Poisson) or ZINB (Zero-Inflated Negative Binomial).

Computational Methods and Tools

  • Available Methods Overview

    • RSA, ScreenBEAM, HiTSelect, MAGeCK, BAGEL, CRISPhieRmix, PBNPA, drugZ, CRISPRBetaBinomial, and ShrinkCrispr.

    • Note: No single method addresses all challenges; all rely on specific assumptions or shortcuts.

  • Simple Methods (Test-Statistic Based)

    • drugZ

      • Designed for drug-CRISPR knockout screens.

      • Operates on rescaled $\log_{2}(\text{Fold Change})$.

      • Assumes a normal distribution.

      • fcr,i=log⁡2(Treat,rt)−log⁡2(Control,rt)fcr,i = \log_2(\text{Treat},rt) - \log_2(\text{Control},rt)

      • eb.stdfcr=1N∑iN(fcrj−0)2eb.stdfcr = \sqrt{\frac{1}{N} \sum_{i}^{N} (fcr_j - 0)^2}

      • Zfcr,i=fcr,ieb.stdfcrjZfcr,i = \frac{fcr,i}{eb.stdfcrj}

      • normZgene=∑Zfcr,innormZ_{gene} = \frac{\sum Zfcr,i}{\sqrt{n}}

    • CB2

      • Based on methods proposed for SAGE data.

      • Uses proportions of sgRNAs in samples.

      • Calculates a t-statistic with weighted proportions based on a beta-binomial distribution.

      • Aggregates p-values using Fisher’s method.

    • PBNPA

      • Non-parametric method normalization using a scaling factor.

      • Utilizes a permutation-based test.

      • Calculates gene-level p-values based on the median sgRNA effect.

    • MAGeCK

      • Models overdispersion using the negative binomial distribution.

      • Employs a modified Robust Rank Aggregation (RRA) to determine gene-level information.

  • Model-Based Methods (Hierarchical)

    • ScreenBEAM

      • Originally developed for RNAi screens.

      • Utilizes a linear mixed-effects model to manage correlations between sgRNAs targeting the same gene.

      • Yij=α+aj+(β+bj)Xij+ϵijY_{ij} = \alpha + a_j + (\beta + b_j)X_{ij} + \epsilon_{ij}

      • ϵij∼N(0,σ2)\epsilon_{ij} \sim N(0, \sigma^2)

      • (aj,bj)′∣μ,Σ∼N(μ,Σ)(a_j, b_j)' | \mu, \Sigma \sim N(\mu, \Sigma)

Comparison with mRNA-seq Methods

  • Standard mRNA-seq tools include edgeR, DESeq2, and voom-limma.

  • While these address related issues in differential gene expression, they are generally less powerful than methods specifically engineered to handle the unique properties of CRISPR screen data.