An optimized genome-wide, virus-free CRISPR screen for mammalian cells

Summary Pooled CRISPR screens have been widely applied to mammalian and other organisms to elucidate the interplay between genes and phenotypes of interest. The most popular method for delivering the CRISPR components into mammalian cells is lentivirus based. However, because lentivirus is not always an option, virus-free protocols are starting to emerge. Here, we demonstrate an improved virus-free, genome-wide CRISPR screening platform for Chinese hamster ovary cells with 75,488 gRNAs targeting 15,028 genes. Each gRNA expression cassette in the library is precisely integrated into a genomic landing pad, resulting in a very high percentage of single gRNA insertions and minimal clonal variation. Using this platform, we perform a negative selection screen on cell proliferation that identifies 1,980 genes that affect proliferation and a positive selection screen on the toxic endoplasmic reticulum stress inducer, tunicamycin, that identifies 77 gene knockouts that improve survivability.


INTRODUCTION
CRISPR screens have been widely applied to decipher mammalian gene function at the genome scale, but most rely on lentiviral delivery methods (Adamson et al., 2016;Joung et al., 2017;Liu et al., 2018). However, lentivirus is not always an option. Working with lentivirus requires specialized facilities as well as trained personnel, which are not available in all laboratories. In some industrial and medical facilities, it is likewise considered an increased risk to work with live viruses. As we demonstrate in this paper, a recombinase-mediated cassette exchange (RMCE)-based, virus-free (VF) method can not only eliminate those concerns, but also result in less noisy data. RMCE is a method in which one can move a piece of DNA, for example, from a plasmid, into a pre-established genome landing pad without causing large-scale disruptions that can occur with CRISPR-mediated targeted insertion.
VF CRISPR screens are starting to emerge using various alternative strategies to perform the screen and detect hits. Examples are replacing an integrated dummy guide RNA (gRNA) with a pooled library of gRNAs by using homologous recombination (Rajagopal et al., 2016) or performing whole-genome sequencing to detect mutations caused by transient CRISPR expression . However, these strategies suffer from drawbacks. The method based on homologous recombination reports that the efficiency of the gRNA cassette integration is much lower than lentiviral integration efficiency, and it cannot be enriched by selection because of a risk of random integration of promoter-less gRNA that complicates ''hit detection.'' This makes it difficult to perform very large screens because of the MOTIVATION Although lentivirus-based delivery of genome-wide CRISPR screen components has proven successful, there are situations in, e.g., industry and hospitals where working with live viruses is difficult or simply not an option. For those situations we have developed an alternative to virus-based, genome-wide CRISPR screens that retains compatibility with the software tools developed for analyzing the results, takes a similar amount of time, and offers improved signal-to-noise ratio. very large number of cells needed (Rajagopal et al., 2016). The second method, applying whole-genome sequencing to detect CRISPR-caused mutations in a pool of cells, is in our opinion more promising, but reliably linking such mutations to phenotype is not trivial, especially when using genomically unstable cell lines such as HEK, HeLa, other cancer cell lines, or Chinese hamster ovary (CHO) cells and, in addition, full genome sequencing will need to be performed at a very high depth to reliably detect any effect less than ''extreme'' .
In this study, we present a new VF CRISPR screening platform and demonstrate it in CHO cells, the model of choice for producing pharmaceutical proteins (Walsh, 2018). We apply this platform to investigate key phenotypes in protein production contexts, namely, cell proliferation and ER stress resistance.
We obtain an in-cell library that covers 99.95% of our 75,488 gRNA large library with 92.5% of the cells containing exactly one gRNA. We show that this VF approach has low noise and works for both depletion and enrichment studies.

RESULTS
Design of the virus-free, genome-wide CRISPR pooled screening platform Using 1,558 RNA-sequencing (RNA-seq) samples from different CHO cell lines and culture conditions, we selected $15,000 expressed genes from the CHO genome . Then, we designed gRNAs against the coding sequences of these genes (Table S1). These gRNAs were used to build a genomewide VF library that contained 75,488 unique gRNAs, of which 72,149 target 15,028 genes, 2,218 target intergenic regions, 1,051 are non-targeting (NT) gRNAs, and finally 70 gRNAs target multiple genes ( Figure 1A, Table S1).
In the VF screening platform, we used Bxb1 RMCE to precisely integrate a single gRNA expression cassette. We made a CHO-S cell line with a Bxb1 RMCE landing pad that contains an EF1a promoter followed by a recombinase target site (attP), an mCherry expression unit, and an SV40 promoter driving expression of a hygromycin resistance gene (HygroR) followed by the second recombinase target site (mutant attP). The integration site of this landing pad was previously selected as an active site surrounded by highly expressed genes Pristov sek et al., 2019;Xiong et al., 2019). This cell line is termed CHO-attp-mCherry. In the donor gRNA library plasmid, the gRNA expression cassette for RMCE is flanked by recombinase target sites (attB and mutant attB). To allow for enrichment of the integrated gRNA while also preventing enrichment of randomly integrated gRNAs, we placed a promoter-less puromycin resistance gene (PuroR) within the recombination site so that the PuroR gene is expressed only after correct insertion into the landing pad with the EF1a promoter upstream. Lack of mCherry expression thus indicates a successful RMCE of the gRNA cassette. After transfection with the Bxb1 recombinase plasmid and gRNA library plasmids, $4% of the cells demonstrated successful RMCE ( Figure S1). The RMCE-positive cells were then further enriched by puromycin (10 mg/mL) selection for 14-20 days ( Figure 2). Twenty-five days after gRNA RMCE, the RMCE cell population was fully enriched, as evidenced by 99.78% of the cells being negative for mCherry ( Figure 1C).

Spurious gene editing is reduced by using transiently expressed Cas9
Spurious gene editing can occur by transient expression of gRNAs from plasmids after transfection of the gRNA library if Cas9 is simultaneously expressed ( Figure 3A). To overcome this issue, instead of using a stable Cas9-expressing cell line, we used transient Cas9 expression in the gRNA RMCE cell pool, introduced after cell-division-based dilution of non-integrated gRNA plasmids. To confirm that spurious gene editing was avoided using this strategy, we performed a test to compare the transient strategy with a cell line stably expressing Cas9. gRNA targeting Mgat1 (Mgat1-gRNA) and an NT-gRNA were mixed at the ratio 1:1. This mixture of gRNAs was co-transfected with Bxb1 into CHO-attp-mCherry cells. Ten days after transfection of the gRNA, the CHO-attp-mCherry cells were transiently transfected with Cas9, and meanwhile, the mixture of gRNAs was co-transfected with Bxb1 into CHO cells stably expressing Cas9 (CHO-attp-mCherry/Cas9). Two days later, both cell line populations were single-cell sorted into 96-well plates ( Figure 3B). We then tested the cells for Mgat1 gene editing and gRNA integration in the landing pad by Sanger sequencing. Spurious gene editing would be a case in which Mgat1 was edited but the NT-gRNA was integrated in the landing pad ( Figure 3C). The sequencing analysis confirmed that there is no spurious gene editing when using transient Cas9, whereas undesired spurious editing was found in the stable Cas9 cells, presumably because of gRNA transcribed from unintegrated gRNA plasmids ( Figure 3D). We thus successfully avoided such undesired and hard-to-detect spurious gene editing. Knowing this, in the following library establishment, we transfected Cas9 25 days after transfecting the gRNA RMCE system to allow sufficient time for dilution/degradation of unintegrated gRNA expression plasmids.
Precise integration of gRNA expression cassette results in lower clonal variance in cell population and achieves high library coverage While constructing and testing the VF library, we simultaneously ran experiments by using a previously established lentiviral library ( Figure 1B) to verify that our VF library gave similar results.
We hypothesized that the targeted integration of gRNA might lead to less clonal variation than the randomly integrated gRNA in the lentiviral library. To investigate the effect of gRNA integration methods on the phenotype of CHO cells, before Cas9 delivery, we analyzed the variance of cell proliferation in the lentiviral library and VF library by measuring the colony confluence 14 days after single-cell sorting into 96-well plates ( Figure 1D). The cell population resulting from lentiviral library transduction showed higher variation across cells than the VF library population. We hypothesize that this higher clone variance in the lentiviral library population is caused by random, and potentially multiple, genomic integrations of gRNA.
To verify the desired single-copy integration of gRNA cassettes, we used qPCR to determine the copy number of genome-integrated gRNA cassettes in the VF RMCE cells and compared with the lentiviral library. Using VF RMCE, 92.5% of the cells showed single integration of gRNA cassettes, whereas the number of integration events in the lentiviral library followed a Poisson distribution ( Figure 1E). Our previous study of the A genome-wide VF library was designed, containing 74,617 gRNAs targeting 18,353 expressed genes in CHO-S cells with 1,000 non-targeting control gRNAs. The gRNA expression cassette was precisely integrated into the working cell line by Bxb1 RMCE. A promoter-less puromycin-resistance gene (PuroR) in the RMCE donor DNA enables the precise integration of a single gRNA expression cassette. The lentiviral library contained 15,645 gRNAs targeting 2,599 genes with 1,000 non-targeting control gRNAs (lentiviral library). (C) In the VF library, the cell pool containing the gRNA library was enriched after 25 days of puromycin selection. The mCherry cassette was finally replaced by the gRNA expression cassette. (D) Before Cas9 was delivered, the cell clone variance in the lentiviral library and VF library was compared by measuring the colony confluence 14 days after single-cell sorting into 96-well plates. CV indicates coefficient of variation. (E and F) The (E) gRNA copy number in the lentiviral library and VF library was further detected by qPCR. gRNA coverage of the VF library was calculated in (F). lentiviral library also demonstrated that it is common to have multiple different types of gRNAs in the cells that have multiple gRNA integrations. In summary, the VF integration results in low cloneto-clone variation in the cell pool before performing the screen.
The coverage of the gRNA library after integration was as high (99.95%) in the cells as in the plasmid pool (99.98%), and showed an even distribution of gRNAs (skew ratio [Konermann et al., 2015;Joung et al., 2017] = 1.85 and 2.84 in the plasmidand cell-based libraries, respectively, Figures 1F and 1G).
Enriching for Cas9-transfected cells results in a high gene editing rate To identify an efficient method to deliver Cas9, we tested a vector expressing Cas9 together with a blasticidin resistance gene (BSD) in CHO cells followed by treatment with blasticidin to kill the cells without transient Cas9 expression. Four days of blasticidin treatment was sufficient to kill untransfected cells (Figure S2). We verified this method via next-generation sequencing (NGS) by using the Mgat1 gene as a model. In the blasticidinselected Cas9-transfected CHO cells with RMCE of gRNA targeting Mgat1 (Mgat1-gRNA), Mgat1 gene editing efficiency was $70%, which is higher than the efficiency via transfection of 1 or 2 rounds of Cas9-BSD without blasticidin selection. Also, the efficiency was higher than transient transfection of the gRNA into a CHO cell line with constitutive Cas9 expression (CHO-Cas9, Figure 1H). With the protocol for enriching Cas9transfected cells now validated, we transfected the Cas9-BSD into the cells with the integrated gRNA library, and we subsequently followed with blasticidin selection to establish the final knockout (KO) CHO cell pool ( Figures S2C and S2D).
Genes essential for cell proliferation are identified in the VF CRISPR screen Nine and eighteen days after Cas9 transfection, the genomic DNA from the cell pool was harvested for NGS analysis to measure the gRNA depletion. The percentage of depleted gRNAs increased after Cas9 transfection into the cell-based gRNA library ( Figure 4A). Eighteen days after Cas9 transfection in the cell-based gRNA library, gene set enrichment analysis revealed significantly depleted genes in fundamental cellular processes: genes involved in DNA replication and repair, gene expression, RNA processing and CHO-attp-mCherry cells with RMCE landing pads were co-transfected with Bxb1 recombinase and the VF gRNA library. Three days after transfection, cells were selected with 10 mg/mL puromycin for around 20 days. Cells transfected with GFP (GFP + puro) or without transfection (None + puro) were set as controls for puromycin selection. Cells transfected with only the VF gRNA library and without Bxb1 recombinase (VF-gRNA-lib + puro) confirm the limited random integration of gRNA cassettes. Nontreated cells (None) were used as a control.
splicing, and protein translation (Mi et al., 2019a(Mi et al., , 2019b) ( Figure 4B, Table S3). The significant depletion of gRNAs targeting 1,980 genes was shown in three biologically independent samples compared with NT-gRNAs ( Figure 4C and Table S4), indicating that these genes are likely essential to cell viability and/or proliferation under our culture conditions.
Using the essential gene database (www.essentialgene.org) (Luo et al., 2014), we compared our results to previous findings ( Figures 4D and 4E, Table S5). Using all 20 human datasets from the essential gene database, we created a consensus set of essential genes consisting of all genes present in at least five of those studies. Compared with this consensus set, we found that 1,322 genes ($47%) are overlapping ( Figure 4D). In addition, we analyzed previous pooled CRISPR KO screens in two human cell lines and found that K562 cells (Wang et al., 2017b) and HeLa cells (Hart et al., 2015) have similar amounts of overlap with each other as they do with our CHO data ($37% between K562 and HeLa, $35% between K562 and CHO, $33% between HeLa and CHO) ( Figure 4E, Table S5). From this we conclude that our VF CRISPR screen method can identify essential genes via gRNA depletion analysis as well as other libraries.
We also wanted to see how our results compared with a similar screen performed on our smaller lentivirus-based library. The overall fold change of targeted genes in both screens showed high correlation after Cas9 induction ( Figure S3). The VF library and the lentiviral library have an overlap of 1,553 targeted genes, and of these, 344 were found to be significantly enriched in the lentiviral library and 286 were found in the VF library ( Figure 4F). Of those genes, 228 (57%) were identical, much higher than the overlap between the two human cell lines K562 and HeLa ($37%).
In summary, we identified 1,980 genes that have a negative impact on cell proliferation and/or viability. The list includes genes coding for key catalytic enzymes such as fumarate hydratase, NADH dehydrogenase, and fatty acid synthase (Tables S3  and S4). This information can be used to further understand cell metabolism and proliferation.
Genes sensitive to induced ER stress are identified in the VF CRISPR screen Although many proteins, most notably antibodies, are readily produced in CHO cells, some proteins are currently difficult or impossible to produce. Such proteins are termed ''difficult-toexpress" (DTE) or "difficult-to-produce" (DTP). Some of these proteins result in high levels of unfolded proteins, which leads to the activation of the unfolded protein response (UPR) and ultimately leads to apoptosis (Prashad and Mehra, 2015). Some investigations indicate that reducing the UPR-mediated ER stress in CHO cells can improve protein productivity (Tigges and Fussenegger, 2006;Borth et al., 2008;Mohan and Lee, 2010;Le Fourn et al., 2014). As a simple demonstration of an enrichment screen, we used the ER-stress-inducing compound tunicamycin (TM) to simulate a DTE/DTP production situation. CHO cells grown in TM stop proliferating and ultimately die. By applying the VF CRISPR screen platform, we identified genes that mitigate the effect of TM on CHO cells.
The CRISPR KO CHO cell pool (60 mL, starting density 7 3 10 5 cells/mL) was treated with TM (20 ng/mL) ( Figure 5A). After 4 days of TM treatment, cells recovered for 5 days by being cultured in the medium without selection chemicals ( Figure S4). The fold change in gRNAs in TM-treated and control cells was evaluated by NGS and computational analysis (Spahn et al., 2017), and the different fold-change patterns were observed in TM-treated and control groups ( Figure 5B). We hypothesized that Mfsd2a (major facilitator domain containing 2A) would be a positive control, given that MFSD2A has been identified as an essential plasma membrane transporter of TM in a human cell line (Reiling et al., 2011) ( Figure 5C). As expected, the gRNAs targeting Mfsd2a were significantly enriched after TM treatment ( Figure 5D). In total, gRNAs targeting 77 genes enriched in this screen ( Figure 5D, combined p < 0.01 in Table S6). Network topology-based analysis (Liao et al., 2019) demonstrated that 36 of these candidate genes showed a high degree of protein-protein interactions (PPIs) between candidate genes ( Figure 5E). Further analysis of PPIs by using the STRING database (Szklarczyk et al., 2019) demonstrated that these candidates were significantly clustered (p = 7.42 3 10 À8 ; Figure S5).
We compared our VF screen results with those obtained by using the smaller lentiviral library. Here the overlap was less impressive than for the proliferation screen. When looking only at genes targeted by both libraries, 11 genes were found to be significant in the lentiviral library and 11 in the VF library, but only 1 overlapping gene, Degs1, was identified.
Validation of candidate genes Degs1 (Delta 4-Desaturase, Sphingolipid 1) catalyzes the final step in the de novo biosynthesis of ceramides, and KO has been previously demonstrated to have an antiapoptotic effect  (Breen et al., 2013;Siddique et al., 2013;Rodriguez-Cuenca et al., 2015;Zhu et al., 2016). In the screens we performed, gRNAs targeting Degs1 were enriched after ER stress in both the VF screen and the lentivirus screen ( Figure 6A). To confirm the effects caused by Degs1 KO, we designed two additional gRNAs, also targeting Degs1, in addition to the five original gRNAs in the VF library ( Figure 6B). Cas9-CHO cells with stable Cas9 expression were transfected with these two gRNAs, independently. Seven days after transfection, we treated the cell pools with TM (20 ng/mL) for an additional 7 days, and the cell density in each group was measured ( Figure 6C). The cell growth curves indicated that the Cas9-CHO cells transfected with either gRNA targeting Degs1 survived better than the cells in the control groups. We further established two cell lines with isogenic Degs1 KO by co-transfection of Cas9 and gRNA-Degs1 in wild-type CHO-S cells followed by single-cell sorting and PCR detection and confirmed by Sanger sequencing ( Figure 6D). Two clones were banked with +1 bp insertion near the gRNA tar-geting sequence (clone 1 was generated by gRNA1 and clone 2 was generated by gRNA2), and both of these Degs1 KO CHO cell lines demonstrated resistance to TM-induced apoptosis compared with wild-type CHO-S cells ( Figure 6E).
In addition, we selected Bag6 (BCL2-Associated Athanogene) and Zfx (Zinc-Finger Protein X-Linked) for further validation, given that all the gRNAs targeting them significantly increased after TM treatment ( Figure 5D). Two new gRNAs were designed targeting Bag6 and Zfx, respectively ( Figure 7A). Seven days after transfection of the new gRNA for each gene into CHO-Cas9 cells with stable Cas9 expression, the different cell pools were treated with TM (20 ng/mL) for an additional 4 days. Viability and viable cell density (VCD) improved for the CHO-Cas9 cells transfected with gRNAs targeting Bag6 or Zfx compared with the control cells, which were transfected with NT-gRNA or GFP. For example, the Bag6 mutant recovered 88% of viability by day 4 after TM treatment ( Figure 7B), and the VCD doubled compared with the control groups ( Figure 7C). Surprisingly, (C-E) Shown are the (C) fold change (median of log 2 ) and the standard deviation of gRNAs among three biological screening replicates. The significantly depleted gRNAs were classified as gRNAs targeting essential genes (highlighted in red). Venn diagrams of candidate essential genes in CHO cells and the reported essential genes (D) in human and (E) in K562 and HeLa cells. (F) Overlap of essential genes between VF growth screen and lentiviral growth screen.
6 Cell Reports Methods 1, 100062, August 23, 2021 Article ll OPEN ACCESS almost full resistance to TM was obtained in CHO cells after disruption of Zfx, wherein the cell viability after Zfx disruption remained above 90% ( Figure 7D) and the VCD was almost 5 times more than that of the control groups 4 days after TM treatment ( Figure 7E). Furthermore, the expression levels of the CHOP gene, a well-known ER stress marker, significantly increased after TM treatment in control cells but not in the cells with Zfx disruption (Figure 7F), indicating blockade of the ER stress pathway after Zfx KO. However, growth in TM-free culture medium showed that none these KOs gave an advantage to cell viability or VCD under normal conditions ( Figure S6). We further investigated the effects of Zfx KO in isogenic CHO cell lines. We established three Zfx KO CHO-S cell lines by single-cell sorting and confirming the Zfx KO with Sanger sequencing. Resistance to TM was likewise observed in these three Zfx KO cell lines (Figure 7G). TM resistance obtained from Zfx KO was also observed in two established Enbrel-producing CHO cell lines (Sergeeva et al., 2020) (dual-RMCE ETN2_T9) with Zfx KO ( Figure 7H). Thus, the VF CRISPR screen can successfully identify candidate genes via gRNA enrichment analysis.

DISCUSSION
Here we present a new and improved VF CRISPR screen approach with low noise that works with standard CRISPR screen analysis tools. In our VF system, 92.5% of cells demonstrated single integration of the gRNA expression cassette, thus decreasing the biases in screens caused by no or multiple integrations of gRNA ( Figure 1E). Targeted integration also decreases clone-to-clone variance on productivity compared with random insertion . Similarly, we see in our VF platform significantly reduced clonal variation in cell proliferation compared with the random insertions in lentiviral methods ( Figure 1D). The gRNA coverage in our cell-based genome-wide gRNA library was 99.95% ( Figure 1G). This is comparable to that achieved by a lentiviral strategy. Almost no gRNA expression cassettes were lost during the long-term cell culture ( Figures 1F and  1G), indicating the stability of our platform. Therefore, our VF strategy improves on the genetic instability seen in lentiviral-transduced cells (Schambach et al., 2013;Rothe et al., 2014). In addition, the time cost for establishing the VF CRISPR screen platform is similar to that for establishing a lentiviral CRISPR screen platform ( Figure S7).
Our platform also solved the problem that gRNA donor-integrated cells in the pool cannot be enriched by resistance-gene selection when random integration occurs. By applying Bxb1 RMCE to precisely integrate a donor-DNA-containing gRNA library with the promoter-less selection marker, the integrated cells could be enriched by puromycin selection in the final population, in which 99.78% of cells demonstrated successful integration of the gRNA expression cassette ( Figure 1C). This Article ll OPEN ACCESS efficiency is much higher than that achieved previously by a VF CRISPR screen approach based on homologous recombination, which showed an efficiency of around 30% (Rajagopal et al., 2016).
For the application of the VF CRISPR screen in our study, Bag6 and Zfx are suggested for KO to reduce TM-induced apoptosis. Bag6 has previously been demonstrated to mediate apoptosis induced by ER stress (Desmots et al., 2008) and DNA damage (Sasaki et al., 2007;Krenciute et al., 2013). Similarly, a former study showed that Bag6-null mouse cells are resistant to cell death induced by thapsigargin, which is another chemical that induces ER stress (Desmots et al., 2008). Our study finds for the first time that Zfx KO provides absolute resistance to TM-induced apoptosis in CHO cells (Figure 7). It has previously been reported that TM treatment reduces the N-glycosylation of Zfx and that the non-glycosylated form of Zfx accumulates in the cell, possibly being the cause of ER stress and/or apoptosis (Zhu et al., 2013). That hypothesis would explain why Zfx KO provides resistance to TM treatment. However, several previous studies showed that downregulating the Zfx gene generally results Seven days after transfection of gRNA, the transfected Cas9-CHO cells were treated with 20 ng/mL TM, and the viability and VCD were recorded for 4 days afterward. Statistical analysis of data at day 4 was performed (*p < 0.05, **p < 0.01, ****p < 0.0001 compared with ''None'' group). (F-H) As shown in (F), the ER stress marker CHOP expression was measured at day 4 after TM treatment in (D) and (E), or at day 4 after cell seeding in Figure S8. Article ll in increased apoptosis in human A549 cells (Zha et al., 2013), malignant glioma cells (Zhou et al., 2011), B lymphocytes (Arenzana et al., 2009), and mouse embryonic stem cells and hematopoietic stem cells (Cellot and Sauvageau, 2007;Galan-Caridad et al., 2007), suggesting an incomplete understanding of the role of Zfx and the toxicity of unglycosylated Zfx.
In summary, this work establishes a genome-wide VF CRISPR screening strategy that we believe will be of interest in cases in which working with viruses for whatever reason is unfeasible. In addition, for CHO cells, we present an expressed genomewide library that can be used for depletion or enrichment screening of phenotypes of interest in pharmaceutical protein production. Although applied here to CHO cells as an example, this approach can also be applied to other cell-based models. Thus, this approach presents a valuable method for precise VF genome-wide screens.

Limitations of the study
The presented method does have some limitations. For one, it requires the host cell to be pre-engineered with a landing pad for the recombinase system to work. This makes it more difficult to work with primary cells and likewise cells that are difficult to transfect. One can also speculate that picking one specific genomic integration spot for the gRNA might specifically perturb phenotypes that are somehow affected by that particular genome location.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following: in the method of copy number analysis. PCR products were purified by Macherey-Nagel NucleoSpin Gel & PCR purification kit. The amplicons were added with Illumina NGS index with adapters using DNA ligation. The resulting library was quantified with Qubit using the dsDNA HS Assay Kit (Thermo Fisher Scientific) and the fragment size was determined using a 2100 Bioanalyzer Instrument (Agilent) before running the samples on a NextSeq 500 sequencer (Illumina).

Clone variance analysis
Single cells were sorted from the LV-lib pool (before Cas9 transfection) and VF library pool (CHO-S cells infected with lentiviral gRNA library but without Cas9) and deposited into a 96-well plate by BD FACSJazz cell sorter (BD Biosciences). 14 days after single cell sorting, the confluence of the emerging colonies in each well was measured by Celigo Cell Imaging Cytometer (Nexcelom Bioscience). Colonies with confluence lower than 10% were filtered out. The clones of 4 plates from VF library and 4 plates from LV-lib was calculated.

Gene disruption validation
One new gRNA for each gene was designed to target Bag6 and Zfx gene, respectively. The target sequences of gRNAs targeting Bag6, Zfx are GGCATTCCGGTCATGAACAGAGG (Bag6-gRNA) and TTCTTCTGAAACAACATCGGCGG (Zfx-gRNA), respectively. Detailed materials and methods for gRNA design and the gRNA vector construction are elaborated in our previous study (Xiong et al., 2019). The CHO-Cas9 cells with stable expression of Cas9 was transfected with gRNAs. 7 days after transfection, the recovered cells were cultured in the medium described as above with 20ng/mL tunicamycin (TM) for 4 days as indicated in legends. The CHO-Cas9 cells transfected with GFP or NT-gRNA were treated with 20ng/mL TM as control. The medium was never changed during these 4 days of TM treatment. The cell viability and the VCD were measured using the NucleoCounter NC-200 Cell Counter (Chemo-Metec, Denmark) during this TM treatment process. A non-targeting gRNA was used as control as described previously (Karottki et al., 2020).

Generation of CHO-Cas9 cell line and CHO-attP-mCherry/Cas9 cell line
To establish a cell platform for stably expressing Cas9, a CHO-Cas9 cell line was generated from a parent cell line containing a RMCE landing pad. The parent cell line was made by CRISPR-mediated homology directed targeted integration of CHO-S cells (Life Technologies) as previously described Petersen et al., 2018;Pristov sek et al., 2019). The parent cell line contains a mCherry coding sequence flanked by a loxP sequence at the 5' end and a lox2272 sequence at the 3' end (pEF1-loxP-mCherry-lox2272-BGHpA), and the 5' and 3' homology arms target a transcriptionally active site within a noncoding region (See plasmid map and sequence). Promoter-less RMCE vectors were constructed by assembly of PCR fragments containing the Cas9 region flanked by loxP and lox2272 sequences. The parent cell line was transfected with RMCE Cas9 donor plasmids (See plasmid map and sequence) and Cre-recombinase vector in 3:1 ratio to exchange the mCherry coding sequence with Cas9. For Cre recombinase expression, PSF-CMV-CRE recombinase vector (OGS591, Sigma-Aldrich) was used. Transfected cell pool was passaged twice after transfection. After 7 days, mCherry-negative single cell sorting was performed by FACS using a BD FACSJazz cell sorter (BD Biosciences). After 14 days, the monoclonal CHO-Cas9 cell line was established after verification of proper integration and confirmation of single copy of Cas9. The methods and the materials for generating this CHO-Cas9 cell line were described before in our study (Hart et al., 2015). By using CHO-Cas9 cell line as the parent cell line, attP RMCE landing pad was inserted into the T2 site as the protocol described above to generate CHO-attP-mCherry/Cas9 cell line.
gRNA design for Mgat1 KO and Mgat1 editing analysis A gRNA targeting Mgat1 (gRNA-Mgat1) was designed to target GTGGAGTTGGAGCGGCAGCGGGG. gRNA-Mgat1 was cloned into the attB vector to achieve RMCE of gRNA-Mgat1 showed in Figures 1H and S2. gRNA-Mgat1 was cloned into identical gRNA vector as described in our previous study for Mgat1 KO by transient expression in Figure 1H. A pair of primers with the sequence 5 0 -TTCTGGACACGCCCAGC-3 0 (forward) and 5 0 -GCCACGGTGGGCACTTT-3 0 was applied to detect Mgat1 editing by Sanger sequencing ( Figure S2) or by NGS analysis (Figure1H). A non-targeting gRNA was used as control for confirming the Mgat1 KO ( Figure S2).
Establishment of Zfx KO cell lines CHO-S WT cells or dual-RMCE ETN2_T9 CHO cells were transfected with Zfx-gRNA described above. The pools were single cell sorted by FACS into 96-well cell culture plates. After 14 days, the single cell derived populations were tested for double KO of Zfx using sanger sequencing of amplicons generated with the following primers: forward: 5 0 -TACATGTGGCTGACGTTGGT-3 0 and reverse: 5 0 -GGAAATCATAAGGTAGTCCTCACA-3 0 . CHO-S WT cells and dual-RMCE ETN2_T9 cells transfected with NT-gRNA and otherwise processed identically as above were used as control ( Figures 7G and 7H).

Plasmid maps and sequences
Sequences and maps of all plasmids can be found in the zenodo data repository https://doi.org/10.5281/zenodo.4494299

NGS data analysis
The gRNA coverage is calculated by the percentage of gRNA with at least 1 read in the library and skew ratio analysis was calculated by the ratio of reads at cumulative percentage 90% to 10%, following suggestions in former CRISPR GeCKO library screens protocol from Feng Zhang 0 s lab (Konermann et al., 2015;Joung et al., 2017). For gRNA fold change analysis, raw FASTQ files were analyzed using default parameters in PinAPL-PY (Spahn et al., 2017) (http://pinapl-py.ucsd.edu/) along with a file containing the sequences for all gRNAs contained in the library. Top candidates for enriched and for depleted gRNAs were ranked by the adjusted robust rank aggregation (aRRA) method (Li et al., 2014) and filtered for significance, compared between the replicates and used for verification of the screen. A gene was labelled ''significant'' if it was both statistically significant on the gene levels as determined by pinapl-py and statistically significant on the gRNA level for at least 3 gRNAs and at least half the gRNAs targeting the gene. Biological process depletion was performed in the PANTHER classification system (Mi et al., 2019a(Mi et al., , 2019b (http://www.pantherdb.org). The list of gRNA fold change was tested against the PANTHER GO-Slim Biological Process annotation set. Network topology-based analysis based on protein-protein-interaction (PPI) of candidate genes was performed in WEB-based GEne SeT AnaLysis Toolkit (Wang et al., 2017a;Liao et al., 2019). Network Retrieval & Prioritization method was applied in this NTA analysis (http://www.webgestalt.org). The PPI analysis was confirmed in STRING database (string-db.org) (Szklarczyk et al., 2019). Essential genes in human and mouse were obtained from the DEG database (Luo et al., 2014) for essential genes (http://tubic.tju.edu.cn/deg). Essential gene lists of K562 cell and Hela cell were obtained from previous CRISPR screen studies (Hart et al., 2015;Wang et al., 2015) collected in the essential gene database (http://www.essentialgene.org/). Raw data, PinAPL-PY output, and data analysis scripts can be downloaded from https://doi.org/10.5281/zenodo.4494299 Details of statistical tests can be found in relevant figure legends.