Fast and easy creation of multiple guide RNAs for CRISPR-Cas9 gene editing

The CRISPR-Cas9 system has revolutionized the field of genome engineering by enabling targeted modification of specific sequences. This system is typically divided into two components: the nuclease protein (Cas9) and a guide RNA (gRNA) that directs Cas9 to its target region [1, 2]. The spacer sequence of the gRNA can be customized to target regions of interest [3]. The nuclease domain has been customized in a variety of ways for different applications, including gene knockouts/knock-ins through double-stranded breaks [4–6], transcriptional regulation (CRISPRa [7] and CRISPRi [8]), and rewriting the genome (base editors [9] and prime editors [10]) and epigenome (CRISPRoff and CRISPRon) [11]. Concurrent delivery of multiple gRNAs is essential to target multiple regions within a cell simultaneously [12]. For instance, Perturb-seq screens target multiple genes within a single cell to identify genetic interactions and reveal cellular pathways [13]; lineage tracing and event recording often edit multiple recorder modules simultaneously to record the history of a cell [14–16]; and targeting the same gene with multiple gRNAs can increase the effect of CRISPRi and CRISPRa in pooled CRISPR-screens [17].

Despite the numerous applications requiring gRNA multiplexing, concurrent cloning and delivery of multiple guides still faces several challenges. An ideal solution would ensure that a single construct includes all the gRNAs of interest to prevent different cells getting different amounts of each gRNA, as would happen with delivery of individually cloned gRNAs. For transfection, this may also be more efficient as, for the same mass of DNA, the concentration and thus the activity of each guide is higher when cloned on the same plasmid as opposed to transfecting individually pooled gRNA expressing plasmids. Integrating guides into the genome is often required for sustained expression and consistent dosage across cells; here too, having all guides on a single construct can facilitate their simultaneous integration because only a single selectable marker is needed to ensure successful delivery. Several approaches have been reported to build multiplexed gRNA systems [12] (Table 1). Polycistronic pre-gRNAs can be processed into functioning gRNAs by simultaneous expression of the Csy4 nuclease [19, 20, 24]. However, this approach requires co-expression of exogenous Cys4 ribonuclease and gRNAs positioned towards the end of the long RNA polymerase III-transcribed polycistronic array were not functional [20]. It is also possible to leverage endogenous tRNA processing by interspersing gRNA units with tRNA sequences in a polycistronic pre-gRNA [21, 22, 25, 26], but this can perturb endogenous tRNA pools. Another technique involves creating arrays of self-contained gRNA units using Gibson assembly, but this approach tends to be inefficient, especially for many gRNA units, resulting in incomplete assemblies. This approach also requires multiple long primers for every gRNA unit as the spacer sequences act as the homology arms to mediate Gibson assembly [23]. Another study used Golden Gate assembly of cloned gRNA fragments into a single vector, including up to 10 gRNAs total, but the initial gRNA cloning and sequence verification adds substantially to the total cloning time and cost [18].

A previous study [27] introduced a method to assemble multiple gRNAs into a single array using polymerase cycling assembly (PCA) [28] to generate the gRNA units, which are then assembled in a specific order and cloned into the vector using Golden Gate assembly [29] (Fig. 1). Using this approach, the authors assembled sets of five gRNAs into a single plasmid for multiplexed CRISPR-activation. Although this study demonstrated a modular way to assemble gRNA arrays, this method was not optimized or evaluated for assembly efficiency, error rates, or potential for scalable gRNA library generation. Building on this foundation, we optimized the approach into RAPID-DASH—Rapid Assembly of PCA-produced Individual DNAs and Directed Assembly through type IIS overHangs for assembling arrays of gRNAs that enables at least 10 gRNAs to be cloned into an array within a single day. Each gRNA unit is created by PCA of a dsDNA U6 promoter, an ssDNA gRNA spacer oligo, and dsDNA gRNA scaffold terminator, resulting in a single dsDNA gRNA unit. A unique primer set is used for each position of the ordered gRNA array, which, in addition to amplifying the assembled gRNA units, adds type IIS overhangs that enable ordered assembly of the gRNA units as an array (Fig. 1a, step 1). During the Golden Gate guide assembly process, BsaI, a type IIS restriction enzyme, digests the ends of each gRNA unit, exposing overhangs that are complementary only to the adjacent gRNA unit, facilitating correct orderly assembly by ligation (Fig. 1a step 2). The lacZ gene within the destination plasmid enables efficient identification of colonies with the gRNA array via blue-white colony screening [30] (Supplementary Fig. S1). In addition to cloning more gRNAs using RAPID-DASH, we provide a detailed characterization of the method, including measurement of assembly efficiency, quantitative error-rate profiling, and assessment of gRNA array library generation. These optimizations substantially enhance robustness, reduce synthesis costs, and establish RAPID-DASH as a versatile framework for constructing large, high-fidelity gRNA arrays suitable for multiplexed genome engineering applications.

We tested RAPID-DASH to assemble an array of 10 gRNAs (see Methods section). Digestion of plasmids isolated from individual clones revealed that the assembled arrays are of expected size (4 kb) (Fig. 1b). To verify the sequence of the assembled gRNA arrays, we extracted plasmid from a single clone via lacZ screening and sequenced it using nanopore sequencing. We verified that it included all 10 gRNAs (Fig. 1c), and its gRNA sequences were correct (Supplementary Material S1) and assembled in the expected order. Whole plasmid sequencing of a pool of transformants (i.e. plasmids extracted from the recovered transformation mix grown in liquid LB + spectinomycin, without plating or lacZ screening) revealed that, on average, 81% (n = 3 replicates) of the assemblies have all 10 gRNA units (Fig. 1d; Supplementary Table S2). In order to estimate the error-rate within the assembly, we sequenced 28 gRNA array clones using nanopore sequencing. Error rates were estimated at the gRNA unit level, breaking it down into mutations that are presumably inconsequential as they are only used for gRNA unit amplification during PCA, and potentially deleterious mutations impacting the gRNA (U6 promoter, gRNA spacer, and scaffold), finding that 7.5% of the total 280 gRNA units had one or more potentially deleterious mutation (Supplementary Fig. S2). Given these results, screening of four colonies has an 84% chance of recovering a full-length array with no putatively deleterious mutations within any gRNA unit (see Methods section). Pre-screening the colonies by restriction digest can help eliminate incomplete assemblies (Fig. 1b). Furthermore, while we did not purify our oligos (standard desalting only), this would likely substantially reduce the error rate, although at a high cost per oligo. Depending on a laboratory's individual circumstances, these may or may not be more economical than just directly screening colonies.

To validate the functionality of gRNA units within the array, we used a cell line that can report successful CRISPR base editing by activating a GFP gene [32]. Here, the GFP gene is initially defective because its start codon has been mutated to GTG. By targeting the Target-AID base editor [34] to the start codon with a gRNA, the start codon is repaired to an ATG, enabling gRNA activity to be assayed easily via cells turning green (Fig. 2a). We cloned the GFP-targeting gRNA within each position of the gRNA arrays independently and observed that the 10 gRNA positions had similar efficacy, activating GFP in 20.17%–26.53% of cells (Fig. 2b), which was far more than the non-targeting gRNA control (0.48% GFP positive; Welch's t-test P < .01 for all gRNAs; Supplementary Table S3). The efficacy of gRNAs in arrays was slightly less than that of a plasmid containing only a single GFP-targeting gRNA (30.56% GFP positive; .0218 ≤ P ≤ .419 for all gRNAs; Supplementary Table S3), which was expected due to the higher transfection efficiency of the smaller GFP-only construct and dilution effect caused by other GFP non-targeting guides competing with limited Cas9 to form ribonucleoproteins. These results illustrate that these gRNA arrays are suitable for multiplexed gene editing as all the positions within the array are functional and have similar editing efficiencies.

For applications requiring a randomized combinatorial genome editing or perturbation, it may be desirable to create libraries of gRNA arrays, each with gRNAs targeting distinct genomic loci. For example, in a large-scale CRISPRa/CRISPRi functional genomics screen, one might aim to interrogate the regulatory effect of ~ 100 000 genomic elements (e.g. promoters or enhancers) by randomly activating or repressing 10 elements per cell. This enables combinatorial perturbation, where different cells receive different combinations of targets, allowing higher-order interaction effects to be observed. By using pools of gRNA sequences during the PCA step, such libraries can be created with our approach. As a proof of concept, we tested this by including 10 gRNA sequences for each of the gRNA unit PCAs, and then assembled them as before into a library, where each library member gets an array of 10 gRNAs, each randomly sampled from the 10 possibilities. For most applications, one would want to sample from a larger pool of gRNAs, but for this demonstration we limited it to 10 so that we could obtain better sequencing coverage, although this strategy nonetheless could produce 10 billion (10¹⁰) unique correctly assembled arrays. Nanopore sequencing of these pooled arrays showed all 10 gRNAs were present in all 10 positions (Fig. 2c), albeit at a slightly reduced frequency of full-length assembly (69% versus 81.2%; Supplementary Fig. S5). Combined with the relatively low cost per oligo when ordering pooled oligos, this approach can be scaled up to include many more gRNA sequences to create libraries of gRNA arrays efficiently, enabling low-cost multiplexed perturbation assays. Of note, because the PCA of each gRNA unit is performed separately, one could limit the possible gRNAs in each position, ensuring that only certain combinations of gRNAs are created. For instance, this could be used to generate a gRNA array where each position targets a subset of genes, with different subsets targeted in each gRNA position, ensuring every array will target exactly one gene within each subset.

Using RAPID-DASH, we can generate and validate gRNA arrays with only a single day of hands-on time (Fig. 1e). RAPID-DASH offers substantial cost savings because only a short (~60 nt) spacer sequence needs to be ordered for each new gRNA. Researchers can also skip the PCA step if they have gRNA units synthesized (433 bp; e.g. gBlocks); however, this will increase the cost substantially as the PCA-based synthesis requires only a single 59 nt spacer oligo for each new gRNA unit. Furthermore, the high-efficiency of our approach saves additional time and labour, reducing the need for layers of screening and enabling one to use libraries of gRNA arrays directly in downstream applications. The major bottlenecks of RAPID-DASH reflect oligo synthesis and sequencing.

Although we only scale to 10 gRNA units, which should satisfy most applications for the time being, we anticipate that RAPID-DASH can be scaled further by using appropriate type II restriction enzyme overhangs to assemble up to 52 gRNA units, although this would result in a plasmid of substantial size (~24 kb) and likely reduce the efficiency of the assembly and stability of the assembled arrays. RAPID-DASH enables faster and more efficient construction and more robust delivery of gRNA arrays, facilitating rapid experimental iteration at scale (Table 1). In order to facilitate adoption of RAPID-DASH, we developed a Shiny app (https://deboerlab.shinyapps.io/OligoDesigner/↗) that enables design of spacer sequence oligos for assembly of up to 20 gRNA units within a single array. Arrays assembled using RAPID-DASH can be transfected directly, used to create virus-like particles containing CRISPR ribonucleoproteins [35], or cloned into PiggyBac vectors furthering the scale to perform multiplexed CRISPR screens [36]. Although lentivirus is commonly used for CRISPR screens, long and repetitive gRNA arrays, such as those assembled using RAPID-DASH are poorly compatible with lentiviral delivery as they reduce viral titres and undergo reverse-transcription–mediated recombination that leads to gRNA dropout within the array. Readouts of CRISPR screens vary by experiment, but we anticipate that RAPID-DASH will be most useful for approaches that read out the mutations directly [37], incorporate a barcode that can be used to infer the guide array [38], or via directly capturing and sequencing gRNAs (10X CRISPR Guide Capture) [17], where imputation could be used to infer gRNA presence even when unobserved if it was known to be on the same array.