Complementary information derived from CRISPR Cas9 mediated gene deletion and suppression

CRISPR-Cas9 provides the means to perform genome editing and facilitates loss-of-function screens. However, we and others demonstrated that expression of the Cas9 endonuclease induces a gene-independent response that correlates with the number of target sequences in the genome. An alternative approach to suppressing gene expression is to block transcription using a catalytically inactive Cas9 (dCas9). Here we directly compare genome editing by CRISPR-Cas9 (cutting, CRISPRc) and gene suppression using KRAB-dCas9 (CRISPRi) in loss-of-function screens to identify cell essential genes. CRISPRc identified 98% of previously defined cell essential genes. After optimizing library construction by analysing transcriptional start sites (TSS), CRISRPi identified 92% of core cell essential genes and did not show a bias to regions involved in copy number alterations. However, bidirectional promoters scored as false positives in CRISRPi. We conclude that CRISPRc and CRISPRi have different off-target effects and combining these approaches provides complementary information in loss-of-function genetic screens.

1. All experiments were performed on the large-scale in a pooled manner. The authors relied on previous knowledge that "a gene is an essential gene". To validate this claim and confirm "the gene is indeed an essential gene", they should validate some of their sgRNA hits for CRISPRc and CRISPRi in a non-pooled manner. 2. The bi-directional off-target results are interesting. The authors should propose how many of these genes in the genome. Since the human genome is sparse with only 2% coding proteins, is this a major limitation for CRISPRi? Also, is there a way to design sgRNAs to avoid such off-target effects? The authors suggested targeting within 100bp downstream of TSS to be most effective. Following this rule, is it possible to avoid bi-directional off-target effects? 3. There are several conclusions comparing CRISPRc and CRISPRi, with each having some pros and cons. For readers to better understand these points, this reviewer suggests the authors to include a  Figure 3B). 5. In page 7, the authors indicated that they re-evaluated CAGE-seq promoter annotations and found 4.7% of protein-coding genes share bidirectional promoters (Fig. S3D). However, this figure is missing in the supplementary information. 6. There are a number of typos in the manuscript and mistakes in the figures, and please carefully revise the manuscript and correct them. In this study, Rosenbluh et al demonstrated that CRISPRi screens can provide complementary phenotypic information to CRISPRc (CRISPR-cutting) screens. The authors show that both CRISPRi (after optimization of library design) and CRISPRc screens can recover most known essential genes, indicating low rates of false negatives. However, they found that both CRISPRc and CRISPRi have significant numbers of false positives. Importantly, they show that the source of false-positives with these two loss-of-function screening modalities is generated by distinct (nonoverlapping) mechanisms; copy number effects for CRISPRc and bidirectional promoters for CRISPRi. Therefore, both types of screens serve to address the shortcomings of the other, and when combined can provide a more complete and accurate picture of gene function. The findings of this study significantly increase our understanding of the types of artifacts that can arise with CRISPRi compared to CRISPRc, and therefore will be of strong interest to the broader research community. The concerns outlined below should be addressed before publication: 1. The authors demonstrated that TSS-targeting guides exhibited very similar phenotypes regardless of whether Cas9 or dCas9-KRAB were used for many genes, even for untranslated regions (Figs 1B, 1D, S1B). This is quite a surprising finding, given that the principal mechanism of Cas9 gene inactivation is thought to be by inducing frameshift mutations in protein coding regions. Upon further inspection, I noticed that many of the TSS-targeting guides exhibiting the strongest lethal phenotypes with Cas9 contain low complexity repeats (eg. GCCGCCGCCGCCGCC). This is likely due to how the tiling library guides were chosen with no filtering for multi-cutters or low complexity sequences. The authors should address whether these TSS regions are inherently less complex and therefore more likely to cause non-specific lethality due to multi-site cutting (as described in the prior work cited by the authors). I would strongly recommend to either remove (or at least clearly label in the graphs) all potential multi-cutters and low complexity sequences before performing downstream analysis, as it may be biasing results.

Detailed responses to the reviewers
Reviewer #1: 1. All experiments were performed on the large-scale in a pooled manner. The authors relied on previous knowledge that "a gene is an essential gene". To validate this claim and confirm "the gene is indeed an essential gene", they should validate some of their sgRNA hits for CRISPRc and CRISPRi in a non-pooled manner.

Response:
Over the past several years, several groups have identified genes that are essential in human cells (Wang et al. Science, 2015, Blomen et al. Science, 2015. For this manuscript, we used a gene set that has been previously validated in several different tissue types Similar to what others have observed, we found that suppression or deletion of genes in this essential gene set inhibits proliferation ( Fig. 2A, C and D). However, we agree with the reviewer that we did not describe how we defined these essential genes in the original manuscript. We have added a more detailed description of how we selected core essential genes (Supplementary methods Pooled sgRNA library 2) as well as specific citations of these reports (Hart et al. and Doench et al.) to the revised manuscript.
2. The bi-directional off-target results are interesting. The authors should propose how many of these genes in the genome. Since the human genome is sparse with only 2% coding proteins, is this a major limitation for CRISPRi? Also, is there a way to design sgRNAs to avoid such off-target effects? The authors suggested targeting within 100bp downstream of TSS to be most effective. Following this rule, is it possible to avoid bi-directional offtarget effects?

Response:
The reviewer asked two questions. a) What is the frequency of bidirectional promoters? Although protein coding genes occupy only 2% of the human genome, such genes are not uniformly distributed but instead are clustered, resulting in an increased occurrence of bidirectional promoters over what might be expected if protein coding genes were randomly distributed. In the original manuscript, we used a conservative cutoff for defining bidirectional promoters (less the 100 bp apart) and found that 4.7% of RefSeq genes contain a bidirectional promoter. To further address the reviewer's question, we have now extended our analysis by calculating the percentage of genes that that are located near a core essential gene and are identified as cell essential in CRISPRi (Fig. S2F). This new analysis, which likely underestimates the occurrence of bidirectional promoters, demonstrates that targeting promoters that are located within 1000 bp from each other can be affected by this bidirectional promoter effect. Based on this new analysis, we reanalyzed the prevalence of bidirectional promoters in the human genome and found that 7.13% of RefSeq genes are within 200 bp from a TSS of another gene transcribed in a different direction. When we increased the threshold to 1000 bp, 12.95% of RefSeq genes are associated with bi-directional promoters. We have added a new figure to the revised manuscript that describes these findings (Fig. S2F).
b) Can one design libraries to avoid bidirectional promoters? Our analysis of bidirectional promoters relied on information derived using a genome scale CRISPRi library, which targets sequences within the first 100 bps of the TSS. Furthermore, our new analyses show that this effect is observed even if promoters are found within 1000 bp (Fig. S2G). It is clear that further work will be necessary to develop libraries that avoid these issues. In the short term, suggest the use of combined CRISPRi and CRISPRc as an approach to identify this effect. We have added text to the revised manuscript to discuss these issues (p. 8 in the revised manuscript).
3. There are several conclusions comparing CRISPRc and CRISPRi, with each having some pros and cons. For readers to better understand these points, this reviewer suggests the authors to include a Table in the manuscript that summarizes and compares the features of CRISPRc and CRISPRi.

Response
We thank the reviewer for that helpful suggestion and have added a table that describes the pros and cons of CRISPRc and CRISPRi to the revised manuscript (  Figure 3B).

Response
Since the submission of the original manuscript, we analyzed PELO expression following CRISPRi mediated suppression of TSS_1 or TSS_2 using quantitative PCR. Unexpectedly, we found targeting of TSS_1 inhibited PELO expression in both HT29 and MIAPACA2 cells while ITGA1 expression was increased in HT29. These observations indicate that predicting the specific effect of CRISPRi on bidirectional promoters likely requires one to test many sgRNA within the region between the two promoters. Since it is clear that further work will be necessary to understand these effects, which may be loci specific, we have removed supplemental figure S3 from the revised manuscript but have added text to describe this complexity (p. 8).
5. In page 7, the authors indicated that they re-evaluated CAGE-seq promoter annotations and found 4.7% of protein-coding genes share bidirectional promoters (Fig. 3D). However, this figure is missing in the supplementary information.

Response
We thank the reviewer for noticing and apologize for the typo we corrected the figure to Fig. 3C. In addition, we have added a more detailed description of bidirectional promoters to the revised manuscript and have also added these new figures (Fig S2C-G).
6. There are a number of typos in the manuscript and mistakes in the figures, and please carefully revise the manuscript and correct them. For example, Supplementary Figure legends 3A and 3B are reversed; Fig 1B might have wrong labels for "Cas9" and "KRAB-dCas9".

Response
We thank the reviewer for noticing these typos. We have carefully reviewed the manuscript and have corrected these typos.