During cancer initiation and progression, a plethora of DNA mutations and epigenomic dysregulations unlock and drive the acquisition of increasingly aggressive phenotypes and the inherent plasticity¹. These divergent genome profiles also condition the maintenance of tumor fitness on genes that are often neutral or dispensable for normal cells. Discovery of these genes, described as cancer vulnerabilities, is leveraged to identify therapeutic targets. To build a causal map of cancer vulnerabilities, gene editing techniques such as CRISPR-Cas9 are used to impair the function of a given gene and observe the effect on cancer cell fitness. Current large-scale mapping efforts have focused on the knockout of single genes. However, simultaneously knocking out gene pairs will enable the identification of cooperating loss-of-function events (paralogs, drug combinations and genetic epistasis). Among such combinations, paralog genes are of notable interest since the redundant function of paralogs is not queried by single gene knockouts. Importantly, many cancer therapeutics (e.g., CDK4/6 and MEK1/2 inhibitors) target closely related paralogous gene pairs.

The potential of combinatorial CRISPR screens to interrogate functional redundancy has led to the development of CRISPR Cas systems enabling digenic knockouts. These include multiplexing guides for S. pyogenes Cas9, using two different Cas9 enzymes (S. aureus: saCas9 and S. pyogenes: spCas9), using direct repeats to express multiple Cas12a guides, and a hybrid Cas9-Cas12 platform^2–4. Nonetheless, establishing an efficient and robust digenic knockout system remains challenging. Initially, single spCas9 multiplexed designs suffered from high rates of recombination due to repetitive elements in the promoter and spCas9 tracrRNA^5,6. To overcome this limitation the BigPapi approach used two different promoters and two Cas9 enzymes (spCas9 and saCas9), to enact combinatorial knockouts⁷. However, in utilizing this system to study paralog redundancy, we observed highly variable saCas9 guide performance². Given the availability of large-scale spCas9 screening data is not surprising that more optimized guide design and selection might favor spCas9. These considerations suggest that a two-guide system using a single enzyme (spCas9) but directing the sgRNAs with two different promoters and using divergent tracrRNA sequences might be more optimal. Finally, an enhanced version of Cas12a (enCas12a) has been elaborated having the advantages of reduced library size, simplified cloning, and a single promoter³.

In our paper “Comparative optimization of combinatorial CRISPR screens”, we sought to compare these digenic CRISPR screens methods. To do this, we evaluated three CRISPR systems (spCas9-saCas9, enCas12a, and 8 combinations of alternative spCas9 tracrRNA sequences) (Figure 1).

We designed the digenic libraries to target positive and negative controls at the single- and double-knockout levels to enable a robust determination of true-positive and false-positive rates. Selected sets of essential (n=52) and non-essential genes (n=94) identified in previous single knockout CRISPR screens⁸ were used as controls for single-gene knockouts. For double knockouts, we identified essential paralog pairs from the genome-wide single-gene knockout studies where lethal interactions arise from the genetic inactivation of one member of a pair of paralogous genes leading to dependence on the remaining paralog² (n=21). Nonessential pairs (n=111) were selected from non-expressed paralog pairs based on Cancer Cell Line Encyclopedia expression data⁹. We selected an identical set of crRNAs compatible with spCas9 that were used for all eight spCas9 tracrRNA combinations. crRNAs for enCas12a were selected using the GPP sgRNA designer.  For comparison we used previous spCas9-saCas9 data that we generated from a larger paralog study² with different choices of crRNAs for the spCas9 guides (Figure 1).

We first screened all libraries in IPC298 cells and investigated the efficacy of ten distinct combinatorial library systems for single-gene knockout efficacy. For each single gene knockout, we paired each targeting sgRNA with an sgRNA targeting the AAVS1 locus (negative control guide). We computed two comparative metrics: receiver operating characteristic (ROC)—area under the curve (AUC) and the null-normalized mean difference (NNMD) to benchmark the performance of those ten combinatorial systems in single-gene knockouts. Here, we identified two top performing alternative tracrRNA spCas9 systems (VCR1-WCR3 and WCR2-WCR3) and then rescreened them with the enCas12a system in two additional cell lines (MELJUSO and PK1). From screens in these three cell lines, we found that the alternative spCas9 VCR1-WCR3 tracrRNAs consistently outperformed other variant spCas9 tracrRNA designs, as well as the enCas12a and orthologous Cas9 systems. Additionally, the VCR1-WCR3 library exhibited stronger depletion of pan-essential genes than observed in the Avana library¹⁰ used in the DepMap screens.

We further addressed a challenge specific to combinatorial CRISPR screens, the possibility of imbalance in knockout efficiency due to the variation between either the two different spCas9 tracrRNA sequences or the effects of driving the sgRNAs from two different promoters (U6 and H1 used in our system). To address this problem, we generated libraries where the two tracrRNAs were flipped (VCR1-WCR3 compared to WCR3-VCR1) while maintaining the same crRNA sequences and promoters. Further, we included identical sgRNA (crRNA+tracrRNA) sequences for 10 genes under both the U6 and H1 promoters to enable a direct comparison of the promoters by constraining the impact of crRNA coupled with tracrRNA. These comparisons showed little variation resulting from the use of two different tracrRNAs and that the slight observed bias is likely driven by promoter differences (Figure 2). As these differences were small, it seems unlikely that changing promoters would result in a more balanced performance and instead one can computationally correct for this bias if desired.

Interestingly, we observed stronger depletion by the sgRNAs utilizing the WCR3 tracrRNAs in the VCR1-WCR3 library compared to the same sgRNAs using the WCR3 tracrRNA in the WCR2-WCR3 library despite having the same promoter, crRNA sequence, and tracrRNA composition. This raised the possibility that the combination with either the VCR1 or WCR2 tracers altered the performance of WCR3. We thought it likely that this might arise due to higher rates of recombination between WCR2 and WCR3 compared to VCR1 and WCR3. Typically, guide sequencing only extends into the crRNA and does not across the tracrRNAs. Thus, to directly measure recombination, we sequenced our sgRNA construct from the representative screens with longer (150bp) paired-ended NGS reads. Here, we found higher rates of recombination between the WCR2-WCR3 tracrRNA (due to homology of tracrRNA sequences) (89%) when compared to VCR1-WCR3 (77%). Thus, recombination between tracrRNAs significantly contributed to decreased knockout performance.

Finally, we evaluated the efficacy of the different library systems in generating digenic knockouts. Again, we determined the ability to separate essential paralog pairs (true positives) from nonessential ones (true negatives). VCR1-WCR3 again outperformed the rest of the libraries based on separation metrics, AUC-ROC, the magnitude and percent of pan-essential pairs exhibiting robust depletion, and the number of synergistic dependencies observed. We also demonstrated, using examples in MAPK signaling pathway, that both pan-lethal gene pairs, as well as selective essential gene pairs are robustly found in these screens.

In a nutshell, we assessed the robustness of ten pooled dual-knockout systems and found that a combination of alternative spCas9 tracrRNAs (VCR1-WCR3) is associated with superior performance in both the single-gene and combinatorial dropout screens. The improvements can be attributed to several aspects including sgRNA design and the more divergent tracrRNA sequences between the two sgRNAs. It is possible that the sub-optimal performance of multiplexed enCas12a system in our screen might be contributed by the choices of sgRNAs, where there is significantly less available data upon which to either select guides or build optimized design algorithms. On the other hand, for Cas12 we observed a significant drop off in performance among guides not using the canonical Cas12 PAM site. Restricting guide designs to this specific PAM site would have the drawback of limiting the number and location of guides that can be selected.

The VCR1-WCR3 system represents a robust and powerful methodology to examine genetic interactions between multiple gene pairs or to reduce genome-scale pooled libraries. Our lentiviral construct and library design have been further applied in other CRISPR-based perturbations studies including in murine systems and have showed comparable performance to our original study.

Please read the full article at here.

References

1. Hanahan, D. Hallmarks of cancer: New dimensions. Cancer Discovery 12, 31–46 (2022).
2. Ito, T. et al. Paralog knockout profiling identifies DUSP4 and DUSP6 as a digenic dependence in MAPK pathway-driven cancers. Nature genetics 53, 1664–1672 (2021).
3. DeWeirdt, P. C. et al. Optimization of AsCas12a for combinatorial genetic screens in human cells. Nature biotechnology 39, 94–104 (2021).
4. Gonatopoulos-Pournatzis, T. et al. Genetic interaction mapping and exon-resolution functional genomics with a hybrid Cas9–Cas12a platform. Nature biotechnology 38, 638–648 (2020).
5. Wong, A. S. et al. Multiplexed barcoded CRISPR-Cas9 screening enabled by CombiGEM. Proceedings of the National Academy of Sciences 113, 2544–2549 (2016).
6. Gasperini, M. et al. CRISPR/Cas9-mediated scanning for regulatory elements required for HPRT1 expression via thousands of large, programmed genomic deletions. The American Journal of Human Genetics 101, 192–205 (2017).
7. Najm, F. J. et al. Orthologous CRISPR–Cas9 enzymes for combinatorial genetic screens. Nature biotechnology 36, 179 (2018).
8. Hart, T. et al. High-resolution CRISPR screens reveal fitness genes and genotype-specific cancer liabilities. Cell 163, 1515–1526 (2015).
9. Ghandi, M. et al. Next-generation characterization of the cancer cell line encyclopedia. Nature 569, 503–508 (2019).
10. Pacini, C. et al. Integrated cross-study datasets of genetic dependencies in cancer. Nature communications 12, 1–14 (2021).