edit close to 100% of the targeted alleles. They hypothesize that the binding of the most efficient (driver) gRNA (in RNP complex format) alters the local chromatin context, which in turn helps recruit a less efficient (helper) gRNA in its vicinity (Joberty et al. 2020).
For situations where a gRNA needs to be positioned to a particular genomic location, for example, to enable KIs, or to edit genomes other than human or mouse, custom gRNA design is still required. An abundance of freely available computational tools have been generated by academic institutions, although some, like the popular CHOPCHOP (Labun et al. 2019), are reserved for nonprofit and academic use only. All major commercial providers also offer custom gRNA design tools at their websites. To assist researchers in making better‐informed decisions, several recent publications have sought to benchmark gRNA design and make recommendations on the best tools available (Bradford and Perrin 2019a; Bradford and Perrin 2019b; Liu et al. 2020). Nevertheless, these studies advise caution in choosing the right tool as experimental datasets used to build models for predicting gRNA specificity or efficiency are disparate. Moreover, some tools are specific to a particular organism or genome build.
4.2.5 CRISPR Libraries: CRISPR KO, CRISPRa, CRISPRi
There are several critical decisions to consider when choosing reagents from commercial vendors for CRISPR screening applications. The execution of screens is logistically challenging, and the scale of resources required (cost, time, expertise, staffing) is substantially greater than single‐gene perturbation experiments. Key factors to consider when designing genome‐wide CRISPR screens are summarized by the following formula describing the multiplier effect of Model × Assay × Perturbation (M × A × P). With the biological question under investigation in mind, the correct combination of these factors is critical for the success of the screen. The following section focuses mainly on the key considerations for sourcing reagents for the “Perturbation” component of this formula.
The most fundamental question to be answered is which flavor of CRISPR technology is most appropriate; CRISPRko, CRISPRi, or CRISPRa? As a general rule of thumb, CRISPRko generates truly null alleles and will link loss of a target to a particular phenotype. It is applicable for the majority of screens. Conversely, CRISPRi and CRISPRa may be better suited to studying genes that manifest different phenotypes at different gene doses and essential genes that cannot tolerate complete knockout or to examine noncoding regions (Borys and Younger 2020; Doench 2018; le Sage et al. 2020). The selection of an appropriate fusion domain to dead Cas9 is an important consideration when using CRISPRa as there are several options (e.g. SAM or VP64) which may need to be tested for activity in your cellular model of interest (Doench 2018; Sanson et al. 2018).
Pooled genome‐wide libraries are available for all three applications. CRISPRko libraries are the most mature, as they have been most widely used by the community to date. CRISPRko libraries are still undergoing rapid evolution, with improvements in performance of the most recent KO libraries over their predecessors being equal to the level of improvement that was observed between the first CRISPR libraries when compared with RNAi screens (Sanson et al. 2018). The number of gRNAs in a pooled library targeting each gene is an important consideration. Initially, genome‐wide libraries were designed with 4–6 gRNAs targeting each gene (Doench et al. 2016; Tzelepis et al. 2016) with the Toronto Knock‐Out library composed of 12 gRNAs per gene (Hart et al. 2015). The total size of the library is extremely important as in order to generate high‐quality data, adequate “coverage” must be maintained at all times throughout the experimental process. “Coverage” refers to the number of cells that contain each individual gRNA and is generally denoted as 100× or 500× to represent that each gRNA is present in 100 or 500 cells within the pool. Based on the Poisson distribution, aiming for 30% transduction efficiency results in ~1 gRNA vector entering each cell. As an example, if your genome‐wide library is 100 000 vectors total (20 000 genes with 5 gRNA per gene) and you are aiming for 500× coverage with 30% transduction efficiency, you must transduce 165 million cells for each replicate of the experiment (multiple replicates are recommended). At each stage of the process, it is important to balance logistical considerations with generating the best‐quality data. For many cell models, it would not be possible to acquire a large number of cells (e.g. 3D organoids, primary cells), so a decision is necessary regarding decreasing coverage or using a smaller library. A smaller library can be obtained by designing a bespoke, focused gRNA set, but if genome‐wide scale is required, minimal libraries have been designed with only 2 gRNAs targeting each gene. In generating such libraries, two highly efficient gRNAs are selected for each gene based on experimental data, reducing the library size to <40 000 gRNAs resulting in a significant reduction of cell requirements (Goncalves et al. 2020).
Dual or multi‐gRNA libraries are commonly used in genetic screens to achieve KO of two genes simultaneously. Dual CRISPRi libraries have been successfully used to create genetic interaction maps (Horlbeck et al. 2018). There are several methods to deliver multiple gRNAs simultaneously to a cell including use of “Big Papi” (Najm et al. 2018), Cas12a (Sanson et al. 2020), dual promoter (Erard et al. 2017), and single promoter with tRNA separator strategies (Zhao et al. 2019). In addition, CRISPRa and CRISPRko have been successfully delivered simultaneously to cells allowing for incredible versatility for experimentally fine‐tuning genetic perturbations to determine effects on phenotypes of interest (Boettcher et al. 2018).
All technologies are also available in arrayed formats. Pooled libraries are generally applicable in relatively simple assay systems, for example, where gRNA depletion is used as a surrogate for cell death resulting from KO of the target gene (Behan et al. 2019), whereas a major advantage of arrayed screening approaches is that they are much more amenable to multiple read outs, such as multiplex bead‐based cytokine assays, flow cytometry, or high content imaging (de Groot et al. 2018; Metzakopian et al. 2017; Strezoska et al. 2017). Moreover, arrayed screens are more suitable to screening of primary cells, if a large‐enough number of cells can be cultured. While an arrayed approach has the potential to give more detailed insight into the phenotypic changes associated with the perturbation, execution of arrayed screens at scale requires specialized liquid handling equipment and the ability to manage, process, and analyze large amounts of data. Another disadvantage of arrayed screening is the substantial increase in cost relating to all aspects of the process, from infrastructure to reagents and consumables to data management. This increased cost limits the scale and throughput potential of arrayed screens as genome‐wide screens would be too large; focused, bespoke libraries are more suited to this format. The trade‐off required when choosing between arrayed and pooled screens may not be an issue for much longer as encouraging work emerging from the Blainey lab indicates that imaging‐based readouts of pooled screens is possible, however not yet at scale (Feldman et al. 2019).
4.3 In vivo CRISPR Screening
4.3.1 Pooled In vivo CRISPR Screening in Rodent Models
While most CRISPR screens to date have been conducted in cell‐based in vitro models, in vivo CRISPR screens have the advantage of directly interrogating gene functions and revealing molecular mechanisms in the native context of animals. In 2015, the first in vivo CRISPR screen to identify loss‐of‐function mutations that promote tumor growth and metastasis was published (Chen et al. 2015). In this study, the authors mutagenized a nonmetastatic mouse cancer cell line using a genome‐wide gRNA library and the resulting pooled cell library was subcutaneously transplanted into immunocompromised mice. By sequencing the enriched gRNAs in the late‐stage primary tumors and lung metastases, they discovered and subsequently validated a small set of genes whose disruptions drove tumor growth and metastasis in vivo. Similar in vivo CRISPR screening approaches using xenograft mouse models have been applied to identify tumor suppressors (Katigbak et al. 2016; Song et al. 2017; Takeda et al. 2019), oncogenes (Kodama et al. 2017), and synthetic lethal drug targets (Yau et al. 2017). CRISPR screening has also been applied in syngeneic models, which have full murine immunity and comprehensive stroma, providing a more relevant setting to assess gene functions in tumor immunity and immunotherapy response. In 2017, Manguso
