CRISPR GENome and epigenome engineering improves loss-of-function genetic-screening approaches

Summary

CRISPR-Cas9 technology has revolutionized genotype-to-phenotype assignments through large-scale loss-of-function (LOF) screens. However, limitations like editing inefficiencies and unperturbed genes cause significant noise in data collection. To address this, we introduce CRISPR gene and epigenome engineering (CRISPRgenee), which uses two specific single guide RNAs (sgRNAs) to simultaneously repress and cleave the target gene within the same cell, increasing LOF efficiencies and reproducibility. CRISPRgenee outperforms conventional CRISPR knockout (CRISPRko), CRISPR interference (CRISPRi), and CRISPRoff systems in suppressing challenging targets and regulators of cell proliferation. Additionally, it efficiently suppresses modulators of epithelial-to-mesenchymal transition (EMT) and impairs neuronal differentiation in a human induced pluripotent stem cell (iPSC) model. CRISPRgenee exhibits improved depletion efficiency, reduced sgRNA performance variance, and accelerated gene depletion compared to individual CRISPRi or CRISPRko screens, ensuring consistency in phenotypic effects and identifying more significant gene hits. By combining CRISPRko and CRISPRi, CRISPRgenee increases LOF rates without increasing genotoxic stress, facilitating library size reduction for advanced LOF screens.

Graphical abstract

Highlights

• •CRISPRgenee combines simultaneous gene knockout and epigenetic silencing
• •CRISPRgenee significantly improves loss-of-function effects
• •CRISPRgenee improves sgRNA heterogeneity improving hit-calling quality
• •CRISPRgenee can be used for high-resolution small-library screening

Motivation

CRISPR technology, including CRISPRko and CRISPRi, enables specific depletion of genes with fewer off-target effects, facilitating precise investigations of gene function. Despite its benefits, CRISPR applications have limitations. Residual active protein expression mediated by in-frame DNA repair or alternative splicing and strong epigenetic regulation along with difficulties in sgRNA design targeting the transcription start site (TSS) hinder the full potential of loss-of-function studies using CRISPRko or CRISPRi. We aimed to achieve robust target gene reduction to improve the reproducibility of the CRISPR technology by integrating the widely used CRISPRko and CRISPRi approaches into a single application.

Stadager et al. present CRISPRgenee, a dual-action gene-editing system that improves the efficiency and reliability of gene loss-of-function studies by combining gene knockout and epigenetic repression in a single step, enabling faster and more reliable identification of genes controlling cell function and disease mechanisms.

Introduction

CRISPR represents the ideal genome engineering system for large-scale forward-genetic-screening approaches to systematically identify new factors involved in normal and pathological processes. Such screens have been employed in many studies,¹ but applications around CRISPR knockout (CRISPRko) can show unpredictable outcomes of non-homologous end joining (NHEJ), resulting in in-frame DNA repair or alternative splicing products^{2,3,4,5,6,7,8,9} leading to residual active protein expression. CRISPR interference (CRISPRi), without relying on error-prone DNA double-strand break (DSB) repair, exhibits a more homogeneous response and an improved gene-depletion efficiency without inducing genotoxic stress, which increases with each DSB leading to severe off-target effects in dual-CRISPRko screens.¹⁰ However, studies have shown that the binding position in the promoter region as well as the native epigenetic landscape play an important role in CRISPRi silencing efficiency, with genes harboring multiple TSSs complicating complete repression.^11,12,13,14 To compensate for possible inefficient CRISPRi/CRISPRko gene suppression, each gene is generally targeted by 5–20 single guide RNAs (sgRNAs),^13,15,16 which not only increases the overall variability of the investigated phenotype but also drastically elevates the costs of library synthesis as well as sequencing depth. This is not ideal for experimental designs in which cell numbers are limited. Nevertheless, such large-scale sgRNA libraries have been deployed to conduct systematic genetic screens to identify essential protein-coding and non-coding genes^{13,17,18,19,20,21} and to uncover gene regulatory networks and regulators of disease-associated states^13,22,23,24 among other things.

To date, several strategies have been employed to optimize current CRISPR LOF approaches to improve on-target and reduce off-target efficiency.^{13,15,25,26,27,28,29,30,31} Nonetheless, commonly used libraries target each gene with five or more sgRNAs^1,15 with the development of newer highly active LOF libraries reducing the number of sgRNAs to 2–4 sgRNAs per gene.^32,33,34 The generation of an ultra-compact (1–3 sgRNAs per gene), highly active dual CRISPRko or CRISPRi sgRNA library^27,35 was another recent approach aiming to reduce the constraints imposed by the large size of sgRNA libraries and challenges in generating cell models with consistent CRISPRi-mediated knockdown. However, the high representation of genes targeted only with a single construct complicates the robust identification of screen hits and increases the identification of false-positive hits due to off-target binding.¹ In some cases, the dual-guide approach was not capable of overwriting strong epigenetic marks,²⁷ and dual-guide libraries targeting only one gene with a single combination are largely ineffective in identifying potential off-target effects. The development of compact, highly active sgRNA libraries would enable CRISPR LOF screens in primary or stem-cell-derived models in vivo as well as in pooled CRISPR screens with spatial transcriptome or proteome resolution with high-content readout and other experimental designs where cell numbers are limiting. To address this problem and to improve the reproducibility of the technology in LOF studies, we have developed a CRISPR system, which enables robust target gene reduction through the combination of Cas9 nuclease-mediated DNA cleavage and repressive epigenome editing of the same target gene. Through the fusion of active Cas9 to a powerful transcriptional repressor and the delivery of two sgRNAs from a dual-expression construct, this approach makes use of the optimized sgRNAs developed previously for CRISPRko approaches, which already demonstrate high knockout (KO) effects³¹ and simultaneously induces the additional downregulation of residual target transcript expression. The system leads to an increase in the overall LOF of the entire cell population allowing for smaller libraries with much lower sgRNA variation per gene and replicate. The combinatorial CRISPR gene and epigenome engineering (CRISPRgenee) approach paired with reduced-scale sgRNA libraries enables systematic forward genetic screens to interrogate the depletion of genes essential for cell growth and the resulting phenotypes with high efficacy and reproducibility.

Results

Truncated sgRNAs prevent CRISPR nuclease activity while simultaneously silencing gene expression

To improve CRISPR LOF screens, our approach was focused on increasing the phenotypic effect by simultaneous gene and epigenome engineering (CRISPRgenee). Repression of target gene expression while introducing a DNA DSB in a shared exon appeared most effective (Figure 1A). For this purpose, we fused functional Cas9 to the KRAB domain of ZIM3 (ZIM3), which has previously been reported to show superior silencing efficiency among a high number of KRAB domains.^27,29 We confirmed these findings by comparing a fusion of dCas9 with either ZNF10-KRAB or ZIM3-KRAB in an NIH/3T3 reporter cell line expressing mCherry.³⁶ Promoter targeting revealed stronger silencing by ZIM3-KRAB than ZNF10-KRAB over 14 days (Figure S1A). To simultaneously achieve repression and DNA cleavage, the nuclease activity of Cas9 has to be controlled to maintain the recruitment of Cas9 at the promoter region. It was recently shown that the Cas9 DNA cleavage activity is impaired when sgRNAs are shortened from the 5′-end^37,38 and that targeting of dCas9 can be achieved with <20-nt sgRNAs to repress gene expression.³⁹ We tested whether truncated sgRNAs could recruit dCas9-ZIM3 and induce an efficient and continuous downregulation of mCherry reporter gene expression (Figure S1B). Recruitment of dCas9-ZIM3 to the synthetic promoter for 8 days did not result in a significant difference in reporter gene silencing when comparing the 20-nt sgRNA and various PAM-distal truncated sgRNAs (Figure S1B). We fused ZIM3-KRAB to active Cas9 (ZIM3-Cas9) in a conditional lentiviral expression vector allowing the timed induction of expression via doxycycline (Dox). We transduced either the ZIM3-Cas9 fusion or dCas9-ZIM3 into the erythroleukemia cell line TF-1 with sgRNAs targeting two genes encoding non-essential transmembrane receptor proteins. We targeted the TSS of CD13 and CD33 in TF-1 cells using 20- or 15-nt sgRNAs (Figure 1B). Fluorescent antibody staining showed that both ZIM3-Cas9 with a 15-nt sgRNA and dCas9-ZIM3 with a 20-nt sgRNA reduced CD13 and CD33 levels, with faster silencing by truncated sgRNAs at day 5 but comparable efficiency by day 14 (Figure 1B). As a next step, we wanted to investigate the DNA cleavage ability of ZIM3-Cas9 using 15-nt sgRNAs as well as 20-nt sgRNAs and compare this effect to Cas9 alone. ZIM3-Cas9 or Cas9 expression was induced through the addition of Dox for 14 days with both 20-nt sgRNAs targeting either CD13 (>1,000 bp distance to the TSS) or CD33 (<1,000 bp distance to the TSS) resulting in a strong reduction of CD13 and CD33 in ZIM3-Cas9 as well as in Cas9-expressing cells (Figure 1C). Both ZIM3-Cas9 as well as wild-type (WT) Cas9 had comparable efficiencies with the 20-nt sgRNA guide, whereas the 15-nt analog targeting the same genomic region within the gene body of CD13 failed to induce any loss of functional protein. However, we observed that the 15-nt sgRNA that targets the gene body of CD33 did induce a significant protein reduction in cells expressing ZIM3-Cas9 but not with WT Cas9 (Figure 1C). After Dox withdrawal, we tracked CD13/CD33 expression over 64 days (Figure S1C). No recovery of CD13 expression was observed in cells where ZIM3-Cas9 or Cas9 was targeted with the 20-nt sgRNA hinting at an irreversible DNA DSB induced by ZIM3-Cas9 or Cas9 cleavage (Figure S1C). The reduction of CD33 observed for the truncated sgRNA when using ZIM3-Cas9 was not stable after the removal of Dox, hinting at a reversible CRISPRi effect on the 800-bp distant promoter region. In contrast, the identical 20-nt-long sgRNA demonstrated an irreversible reduction of CD33, observed after removal of Dox in both ZIM3-Cas9 and Cas9 cells (Figure S1D). The striking difference observed between the two targets when using the truncated sgRNA could be explained by the distance between the targeted region and the TSS. For CD33, the targeting region was within 1,000 bp of the TSS, whereas, for CD13, the targeting region was outside this range (Figure 1C) suggesting that, in the case of CD33, binding of ZIM3-Cas9 was able to silence CD33 expression, which was recovered after Dox removal (Figure S1D). To confirm that CD13 and CD33 loss with 20-nt sgRNAs resulted from DNA cleavage, we performed a mismatch-cleavage assay, detecting cleavage products only with 20-nt guides, not with 15-nt sgRNAs. (Figure S1D). Additionally, we performed amplicon sequencing of the Cas9 cleavage site to quantify the amount of DSB breaks observed for the 15- and 20-nt guides (Figure 1D) and did not detect a significantly higher insertion or deletion (indel) frequency when using the 15-nt guide compared to the control, whereas, when using the 20-nt guide, a significant increase in indel frequency was observed (Table S4). This suggested that, in the case of CD33, for which ZIM3-Cas9 had a significantly stronger silencing effect compared to Cas9 when using the 20-nt sgRNA, we observed the simultaneous effect of genome editing and transcriptional interference, which was not anticipated, demonstrating that our envisioned approach could improve LOF approaches by simultaneous gene editing and epigenome interference.

Figure 1.

Design and validation of the CRISPRgenee concept to improve standard CRISPR LOF approaches

(A) Schematic overview of the CRISPRgenee setup. An irreversible DNA DSB is introduced by targeting ZIM3-Cas9 to an exon of the gene of interest using a 20-nt guide RNA. This is combined with a simultaneous recruitment of ZIM3-Cas9 to the promoter region of the target gene using a truncated guide RNA inducing a stable repression of gene expression.

(B) Flow-cytometry analysis of the depletion of the non-essential proteins CD13 and CD33 using standard 20-nt-long sgRNAs or truncated 15-nt sgRNAs in TF-1 cells expressing the dCas9-ZIM3 fusion protein. The location of the sgRNA target region is highlighted in red. (n = 3, mean ± SD).

(C) Flow-cytometry analysis of the depletion of CD13 and CD33 using standard 20-nt sgRNAs or truncated 15-nt sgRNAs in TF-1 cells expressing either Cas9 or the ZIM3-Cas9 fusion protein. The location of the sgRNA target region is highlighted in black, indicating whether the sgRNA is within 1,000 bp of the nearest transcription start site (TSS) (n = 3, mean ± SEM).

(D) Indel frequency of the CD13 and CD33 locus in ZIM3-Cas9+ TF-1 cells treated with control, 15-nt, or 20-nt sgRNAs.

(E) Time-resolved quantification of CD13 and CD33 negative TF-1 cells expressing the indicated sgRNAs after induction of ZIM3-Cas9. A total of eight CRISPRko (ko) and seven CRISPRi (i) sgRNA designs were combined in 22 CRISPRgenee (g) constructs for comparison. Data are displayed as a single datapoint for each sgRNA or sgRNA combination and replicate summarized in a boxplot (n = 3, mean, box, and whiskers minimum to maximum); ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001; n.s., non-significant.

Significance was determined using a one-way ANOVA for (D) and a two-way ANOVA for (B), (C), and (E).

See also Figures S1 and S2.

CRISPRgenee significantly accelerates and increases CRISPR-mediated gene depletion

We next tested whether CRISPRgenee improves the time and efficiency of gene silencing compared to CRISPRi and CRISPRko. To this end, we performed a time-resolved reduction of CD13 and CD33 protein levels in TF-1 cells using the same inducible ZIM3-Cas9 construct. We used sgRNAs targeting promoters or gene bodies and tracked CD13/CD33 protein levels by flow cytometry (Figure 1E). In order to investigate a potential combinatorial effect when using CRISPRgenee, the CD13 and CD33 sgRNAs were chosen based on their predicted in silico depletion efficiencies⁴⁰ (low, medium, and high) and used in combination (CRISPRgenee) compared to CRISPRi or CRISPRko alone. Given that 15-nt sgRNAs silence as effectively but faster than 20-nt sgRNAs (Figure 1B), we used them as CRISPRi controls to isolate CRISPRgenee’s benefits to the combination of gene and epigenome editing. Initial CRISPRgenee combinations of sgRNAs with varying predicted efficiency significantly enhanced CD33 suppression (Figure S1E).

Building on these initial results, we tested a broader set of CRISPRko and CRISPRi sgRNAs (with varied in silico prediction scores)⁴⁰ and their CRISPRgenee combinations on CD13/CD33 expression over 14 days (Figure 1E). Overall, CRISPRgenee showed significantly stronger CD13/CD33 suppression after 7 days, consolidating over time. (Figure 1E). Strikingly, it also reduced variance in sgRNA performance, yielding more consistent silencing than CRISPRi/CRISPRko (Figure 1E). To assess whether the increased depletion efficiencies follow a combinatorial effect, we analyzed the amount of CD13/CD33 negative cells at day 9 and day 14 for all CRISPRi sgRNAs with their respective CRISPRko sgRNA combinations. Interestingly, we were not able to identify a clear trend for the combinatorial effect (Figure S2A). While combining two non-functional sgRNAs did not boost depletion, pairing a weak with a stronger sgRNA improved silencing. Thus, CRISPRgenee accelerates and standardizes gene depletion compared to single-guide LOF approaches (Figures 1E and S2A). We confirmed the combinatorial benefit by comparison to conventional dual-sgRNA setups and observed that CRISPRgenee led to faster, more efficient silencing (Figure S2B). Overall, these results demonstrate that the combination of CRISPRko and CRISPRi sgRNAs within the CRISPRgenee system leads to a beneficial LOF effect.

Truncated and normal-length sgRNAs exhibit similar off-target effects

We anticipated broader off-target effects from the 15-nt sgRNA due to its role in transcriptional repression, while 20-nt guides were expected to behave like standard Cas9 guides. Although recent literature highlights that only the first five bases in the seed region influence on- and off-target activity of dCas9,^41,42,43 and therefore the reduction of 5 bp in the PAM-distal part should not influence binding, we aimed to determine the transcriptional effects of two sgRNAs targeting CD33. To this end, we selected sgRNAs predicted to harbor several perfect off-targets in the genome. We transduced dCas9-ZIM3 TF-1 cells with 15- or 20-nt sgRNAs and analyzed transcriptome changes after 14 days by RNA sequencing (RNA-seq) (Table S5). Interestingly, the principal-component analysis (PCA) showed that truncated and full-length guides clustered closely, indicating minimal differences (Figure 2A). Log fold change (LFC) values (sgRNA to control) showed high Spearman correlation between matched 15- and 20-nt sgRNAs, suggesting the final five bases to have minimal impact on off-target effects (Figure 2B). Direct comparison revealed only minor differences in gene expression between 15- and 20-nt guides (Figure 2C). We predicted all off-targets with up to three mismatches (MMs) in silico and filtered for all off-targets in the range of −1,000 to 1,000 bp to the nearest TSS. No significant differences in LFCs were observed between sgRNA lengths, regardless of mismatch number (Figures 2D and S3A). Exemplarily, we identified a subset of off-targets with a variety of MMs to the sgRNA, demonstrating similar effects of the 15-nt sgRNA and 20-nt sgRNA (Figure 2E). Overall, recruitment of dCas9 to the DNA seems to tolerate MMs more than expected, to the extent that there are no significant differences when using a 15-nt sgRNA with two MMs to the target region or the corresponding 20-nt sgRNA with four or five MMs to the target region (Figure 2D).

Figure 2.

Off-target analysis of 15- and 20-nt-long sgRNAs does not reveal significant differences

(A) Principal-component analysis (PCA) of the RNA-seq data for two full-length (shades of blue) and truncated (shades of red) CD33-targeting sgRNAs, non-targeting control (black), and WT (gray).

(B) Correlation of the log2 fold change of sgRNA to control for each transcript between the truncated and full-length sgRNA was assessed using Spearman correlation analysis.

(C) Log2 fold change of 15- and 20-nt sgRNAs targeting the TSS of CD33. Off-targets toward TSSs of other genes were predicted in silico for up to three mismatches for the 15-nt sgRNA and are indicated by color.

(D) Log2 fold change of the 15- or 20-nt CD33-targeting sgRNA and the sgRNA control for all potential off-targets identified in silico.

(E) Log2 fold change of identified off-targets for the 15-nt sgRNA calculated for both sgRNA variations with the DNA base mismatches indicated in red.

See also Figure S3.

We validated this in the mCherry reporter system³⁶ using matched sgRNAs with up to two MMs (Figure S3B). Only two consecutive mismatches in the seed region significantly impacted the silencing of mCherry, whereas guide length had no effect (Figures S3B and S3C). These data imply that DNA cleavage-based off-target models for Cas9 do not reliably predict dCas9 binding activity, aligning with recent studies.^41,42,43

Assessment of phenotypic consequences following CRISPRgenee-mediated BUB1 depletion

Given CRISPRgenee’s improved depletion of CD13/CD33, we hypothesized it would cause stronger phenotypic effects than conventional methods. To evaluate this hypothesis, we applied CRISPRgenee to BUB1, a key spindle assembly checkpoint (SAC) kinase essential for cell-cycle progression and proliferation.⁴⁴ However, the essentiality of BUB1 was under debate^45,46,47,48 since KO cells often retain partial expression via alternative splicing.^44,47,48,49 Only the removal of the remaining BUB1 mRNA by RNAi or the complete removal of the BUB1 gene substantially affected the cell cycle.⁵⁰ Targeting distinct domains of BUB1 did not result in a loss of functionality, with alternative splicing rendering the generated indels useless.^49,51 Only the full removal of the gene by CRISPR was successful to result in a complete loss of BUB1 in haploid cells, making it an ideal target to benchmark CRISPRgenee.^51,52

Standard LOF screening approaches assess the essentiality of a specific gene by determining the phenotypic effects of 5–20 sgRNAs. To determine whether the increased efficiency observed on the level of non-essential proteins (Figure 1E) can be translated to a potential multiplexed LOF screening approach we aimed to assess whether targeting BUB1 would generate comparable phenotypic results, which was shown to express alternative variants after CRISPR-Cas knockout.^44,49 To this end, we selected the top-five predicted sgRNAs for CRISPRko and CRISPRi⁴⁰ targeting BUB1 and sequentially combined the top-three sgRNAs to be used in all possible combinations in the CRISPRgenee system (Figure S4A). TF-1 cells expressing conditional ZIM3-Cas9 were transduced with the respective BUB1 sgRNAs, and the fraction of sgRNA-positive cells was monitored for 21 days. Despite selecting top-ranked sgRNAs, most CRISPRi/CRISPRko guides showed weak and variable effects (Figure S4A). We integrated the mean effect of the top-three sgRNAs for CRISPRi, CRISPRko, and the respective CRISPRgenee combinations to simulate a screening setup in which three sgRNAs per gene would be used (Figure 3A). CRISPRgenee combinations caused early and strong depletion, with effects growing over time. Overall, the CRISPRgenee sgRNA combinations not only led to a stronger phenotypic effect after a shorter time frame but also displayed less variance, resulting in better statistical significance compared to the three sgRNAs used for CRISPRi and CRISPRko (Figure 3A), suggesting that CRISPRgenee achieves significant effects at earlier time points. With the top-three predicted CRISPRi sgRNAs targeting BUB1 producing a high heterogeneity in response to BUB1 depletion, we wanted to assess whether transcriptional silencing of BUB1 might be inefficient due to the essentiality of this gene. Therefore, we used the recently published CRISPRoff technology³⁰ in combination with three dual-sgRNA plasmids targeting the TSS of BUB1. CRISPRoff produced delayed and inconsistent effects, with significant depletion only by day 19 (Figures S4B and S4C). To investigate the biological impact of BUB1 depletion on cell-cycle progression, we selected BUB1 sgRNA combinations, which displayed a faster negative proliferative effect and investigated these in HEK293 cells expressing conditional ZIM3-Cas9 or CRISPRoff (Figure 3B). Additionally, we compared CRISPRgenee to conventional Cas9 using a dual-sgRNA strategy simultaneously targeting the first and last exon of BUB1, which was previously reported to achieve a full removal of the BUB1 gene in HAP1 cells but not in any other cell line.⁵¹ CRISPRgenee matched the benchmark dual-cutting Cas9 strategy in reducing proliferation (Figure 3B). The BUB1 depletion using CRISPRoff (dual sgRNA) resulted in a faster phenotypic effect compared to CRISPRi but reached similar endpoint values and was not able to increase the negative proliferative effect further (Figure 3B). CRISPRoff led to toxicity even with scrambled (scr) guides, likely due to nonspecific silencing by DNMT3A-3L, consistent with prior reports (Figures S4D and S4E).^28,50,53,54 CRISPRgenee reduced BUB1 mRNA to 30%, compared to ∼50% for CRISPRi/CRISPRoff Figure S4E). To determine whether the negative proliferative effect observed upon BUB1 suppression is a result of its essential function in cell-cycle regulation, which will result in DNA replication stress and subsequent apoptosis, we performed DNA content analysis to assess cell-cycle effects of BUB1 depletion (Figure S4F). Both CRISPRgenee and CRISPRoff in combination with an sgRNA targeting a non-essential gene did not result in significant differences in cell-cycle distribution compared to the HEK WT cells, whereas a significantly decreased G1 cell population was observed for all BUB1 depletion samples (CRISPRi, CRISPRko, CRISPRgenee, benchmark, and CRISPRoff) (Figure 3C). However, only CRISPRgenee and the benchmark resulted in a significant difference in other cell-cycle stages compared to HEK WT cells. The benchmark sgRNAs demonstrated a significant accumulation of cells in the S phase, whereas CRISPRgenee resulted in a significantly higher cell population in the G2/M phase of the cell cycle, which would be in conjunction with the proposed function of BUB1 in the M phase of mitosis.⁴⁴ Only CRISPRgenee increased the sub-G1 population, indicating apoptosis, which was absent in other samples (Figure 3C). These findings imply that the observed phenotype does not stem from higher genome toxicity but, rather, from the impaired cell cycle attributed to the improved reduction of BUB1.

Figure 3.

CRISPRgenee outperforms standard CRISPR LOF methods and validates the essential role of BUB1 in cell-cycle progression

(A) Competitive proliferation assays of TF-1 cells expressing the indicated sgRNAs (gray scale) targeting BUB1. The data show the mean relative fraction of GFP+/sgRNA+ cells for three CRISPRi (i) and CRISPRko (ko) sgRNAs, and the respective CRISPRgenee (g) sgRNA combination, relative to the initial measurement over a 21-day period (n = 3, mean ± SD).

(B) Competitive proliferation assays of HEK293 cells expressing ZIM3-Cas9 or the CRISPRoff construct and the indicated sgRNAs (dual- or single-sgRNA setup) targeting BUB1 to validate the improved CRISPRgenee effect observed in TF-1 cells (n = 3, mean ± SEM).

(C) Inferred distribution of cell-cycle phases of HEK cells harvested at day 5 of BUB1 depletion as indicated in (B). Percentages in each phase of the cell cycle were automatically assigned using FlowJo (n = 3, mean ± SEM).

(D) EMT was induced in MCF10A cells expressing either dCas9-ZIM3 or ZIM3-Cas9 and scr (ctrl) or SMAD2-targeting sgRNAs using TGFβ. The fold change of cells detected in the epithelial and mesenchymal population was calculated in SMAD2-depleted cells relative to the control (n = 3, mean ± SEM).

(E) Schematic overview of the experimental setup used to validate the tolerability of CRISPRgenee in iPSCs.

(F) Differentiation of iPSCs expressing CRISPRgenee and an ART1-targeting sgRNA was monitored for 3 days by assessing cell morphology and expression of Syn1.

(G) Relative TRA-1/60 signal (a marker for pluripotency) of ZIM3-Cas9-positive and ZIM3-Cas9-negative Dox-stimulated iPSCs either expressing an sgRNA targeting ART1 or the transgene Ngn2 responsible for neuronal differentiation.

∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001; n.s., non-significant. Significance was determined using a two-way ANOVA for (A)–(C) and a one-way ANOVA for (D).

See also Figures S4 and S5.

CRISPRgenee extends to non-tumorigenic and stem cell contexts

Next, we wanted to assess whether CRISPRgenee can be applied to cells sensitive to DNA damage and epigenetic silencing such as primary cells or stem cells. To test whether we could effectively perturb epithelial-to-mesenchymal transition (EMT) induced by transforming growth factor (TGF)-β, we transduced the non-tumorigenic epithelial MCF10A cells with dCas9-ZIM3 or ZIM3-Cas9 and observed normal growth behavior for either construct. We then transduced these cells with sgRNAs targeting SMAD2, a known mediator of TGF-β signaling as well as an scr sgRNA control. After 1 week, we induced EMT using TGF-β, stained for E-cadherin and vimentin and validated the loss of SMAD2 via qPCR (markers of EMT) (Figures 3D, S5A, and S5B). Cells expressing the scr sgRNA control showed reduced E-cadherin and increased vimentin, while SMAD2-depleted cells showed significantly reduced EMT using CRISPRgenee but not CRISPRi (Figure 3D). Importantly, the cells that did not react to TGF-β-induced EMT due to downregulation of SMAD2 by CRISPRgenee did not exhibit negative effects on viability due to genotoxic stress but instead remained in the epithelial cell population (Figure 3D).

We further validated CRISPRgenee in human induced pluripotent stem cells (iPSCs) engineered for neuronal differentiation upon Dox induction via mouse transgenes Neurogenin-2 (Ngn2) and Neurogenin-1 (Ngn1) (Figure 3E).⁵⁵ We designed CRISPRgenee sgRNA combinations, one targeting a non-essential gene expressed in neuronal cells (ART1) and another sgRNA combination targeting the transgene Ngn2, validated the loss of ART1 via qPCR (Figure S5C), and induced neuronal differentiation. iPSCs expressing the sgRNA targeting ART1 showed the anticipated morphological changes after 3 days (Figure 3F) and upregulation of SYN1 (Figure 3F), a gene that is activated in neuronal differentiation.⁵⁵ The cells were harvested and stained for a cell-surface pluripotency marker (TRA-1/60), which is reduced after differentiation to neuronal cells.⁵⁵ We measured a significant loss of TRA-1/60 compared to the uninduced WT, without finding any significant differences between the ZIM3-Cas9-negative and ZIM3-Cas9-positive cells expressing the sgRNA targeting ART1 (Figure 3G). This indicates that the iPSCs maintain normal phenotypic behavior and signaling, unaffected by potential DNA damage stress from CRISPRgenee expression. However, targeting Ngn2 significantly impaired differentiation, as shown by higher TRA-1/60 levels in ZIM3-Cas9-positive cells compared to negative controls (Figure 3G), confirming that CRISPRgenee effectively disrupts cellular differentiation.

CRISPRgenee improves LOF screen performance

Having demonstrated that CRISPRgenee improves depletion efficiency and phenotypic consistency with reduced variance (Figures 1E, 3A, and 3B), we hypothesized that its enhanced suppression and lower variance would enable faster, more robust hit calling at lower library coverage. We therefore decided to test the CRISPRgenee system using a multiplexed LOF phenotypic screening approach and collaborated with Twist Bioscience to design and clone dual-sgRNA oligos comprising a 15-nt sgRNA, tracr, promoter, and 20-nt sgRNA (Figure S6A). To synthesize dual sgRNAs on a <300-nt oligo, we used a minimal hybrid H1 promoter substituting all core elements except the Staf domain with 7SK promoter elements (minH1/7SK), which reportedly only shows Pol III activity.⁵⁶ Since the H1/7SK promoter substitutes the H1 promoter used for the CRISPRko sgRNA, we tested previous CRISPRko sgRNAs targeting CD33 observing no significant differences in CD33 depletion (Figure S6B). Notably, replacing H1 with minH1/7SK significantly enhanced hU6-driven truncated CRISPRi sgRNA activity, yielding faster CD13 reduction (Figure S6B). Furthermore, silencing effects using CRISPRi can be seen faster compared to when using CRISPRko, which might also explain why a difference could be detected for the CRISPRi but not the CRISPRko. Using the minH1/7SK promoter, we synthesized and cloned a CRISPRgenee oligonucleotide pool designed to target 1137 genes (including 10 internal controls) involved in chromatin regulation.^57,58 For each gene, three CRISPRgenee combinations were designed by pairing the top-three CRISPRi (15 nt) and CRISPRko sgRNAs selected using the latest prediction algorithms.^31,40 For genes with multiple annotated transcription start sites (TSSs) with a distance of >1,000 bp from each other, the same CRISPRko sgRNAs were combined with different 15-nt sgRNAs targeting each TSS (Figure S6C). The initial 270-nt oligo pool was synthesized and quality control (QC) checked by deep sequencing post cloning into the screening vector, and the overall distribution was calculated (Figure S6D). The CRISPRgenee library harboring 3686 sgRNAs in total (Table S6) was transduced in triplicate into three independent conditional ZIM3-Cas9 TF-1 cell clones. After selection with neomycin (NEO) for 7 days, ZIM3-Cas9 expression was induced by the addition of Dox (Figure 4A). Based on encouraging BUB1 suppression results (Figures 3A and 3B), showing earlier, more reproducible negative proliferative effects with CRISPRgenee, cells were passaged only seven times (14 days, 70-h doubling time).²⁷ The initial chimera rate in the oligo pool was 9.4%, which increased after library preparation from gDNA ranging from 14% to 20% per single-cell clone (Figure S7A), while library preparation of the oligo pool produced a 34.6%–38.0% chimera rate, likely due to unfavorable PCR conditions optimized for sgRNA amplification from gDNA. We filtered sequenced reads where the 15-nt and 20-nt sgRNAs matched the designed oligonucleotides. Overall, the CRISPRgenee screen showed remarkable reproducibility across replicates (Figures S8 and S9), and we assessed sgRNA and gene performance using the CRISPRBetaBinomial (CB2) algorithm.⁵⁹ Neutral-control sgRNAs targeted non-essential CD13/CD33 genes, whereas positive controls targeted RPA3, MYC, and CSF2RA, a receptor subunit of the granulocyte colony-stimulating factor (G-CSF), essential for TF-1 proliferation (Figure S7B). As expected, neutral-control sgRNAs did not show a strong effect on cell proliferation in contrast to positive sgRNAs (Figure 4B). Next, we investigated the effect of individual sgRNAs targeting a set of established common essential and non-essential genes.⁶⁰ Targeting essential genes led to negative LFC values contrary to sgRNAs targeting non-essential genes, which clustered around LFC values from 0 to 1. This effect was also visible on the gene LFC level (Figures 4C and S7C). We calculated the receiver operating characteristic (ROC) area under the curve (AUC) using all sgRNAs targeting gold-standard essential/non-essential genes (Figure S10A), demonstrating that CRISPRgenee, despite fewer sgRNAs per gene and passages, rivals many high-end published CRISPR screens (Table S7).^{13,15,21,27,31,32,33,35} Spearman correlations between replicates for read counts, sgRNA, and gene LFC showed high reproducibility (Figures 4D, S8, and S9). Our previous data suggested that CRISPRgenee reduces the overall noise within phenotypic effects among sgRNAs targeting the same gene (Figures 1 and 3). Therefore, we compared the ΔLFC (maximum LFC minus minimum LFC) between sgRNAs targeting the same gene to the identical genes in recently published LOF screens.^{13,15,21,27,31,32,33,35} Among essential gene targets, CRISPRgenee showed significantly lower ΔLFC across sgRNAs when three or more sgRNAs per gene were used (Figure 4E), which was also the case when filtering for essential genes, indicating an overall more consistent sgRNA performance (Figure 4E). A similar trend was observed when we investigated the variance of the LFC for all sgRNAs targeting the same gene (Figure S10B). To test whether reduced variance improved hit calling, we performed the same analysis on –log10 adjusted p values (false discovery rate [FDR]) for sgRNAs in published screens filtered for CRISPRgenee chromatin library genes, showcasing that sgRNAs exhibited a much higher reproducibility in significant phenotypic effects in the CRISPRgenee screen compared to the other analyzed public screens (Figures 4E and S11). Interestingly, high sgRNA-level significance in some published LOF screens did not transfer to the gene level, suggesting that sgRNA heterogeneity raises variance and reduces gene-level significance. With the CRISPRgenee approach, this loss in significance was not observed (Figure 4E), resulting in more consistent significant gene hits with the CRISPRgenee screening approach.

Figure 4.

CRISPRgenee outperforms published LOF dropout screens with improved sgRNA consistency

(A) Schematic overview of the screening setup used for the CRISPRgenee screen. A library composed of 3,686 sgRNAs targeting 1,137 chromatin-related genes was virally transduced into TF-1 erythroleukemia cells expressing ZIM3-Cas9. After antibiotic selection, cells were treated with Dox for 14 days.

(B) Performance of individual sgRNAs. The log fold change (LFC) of all sgRNAs targeting internal positive (red) and neutral controls (blue) as well as sgRNAs targeting known essential (red) and non-essential genes (blue) is depicted. Data are averaged across three individual replicates.

(C) Scatterplot depicting all genes ranked by the average LFC of all sgRNAs per gene across all three replicates. Internal positive (red) and neutral (blue) control genes (top), as well as essential (red) and non-essential genes (blue) are highlighted (bottom).

(D) Spearman correlation r was calculated for the replicates of the CRISPRgenee screen and replicates of published screen datasets on read count, sgRNA LFC, and gene LFC level.

(E) Violin plots comparing the CRISPRgenee system with a set of published CRISPR screening approaches. Left: comparison of the ΔLFC (maximum LFC – minimum LFC) of sgRNAs targeting the same gene depicted for essential genes and all genes investigated in the screen. Right: comparison of the −log 10 adjusted p-value distribution at the sgRNA and gene level. The black vertical lines depict the median for each screen and the red dashed line is the median of the CRISPRgenee screen (∗p ≤ 0.05, ∗∗∗p ≤ 0.001, ∗∗∗∗p ≤ 0.0001; one-way ANOVA with a Dunnett post hoc test).

See also Figures S6–S11.

CRISPRgenee generates improved dropout effects through a combinatorial effect of CRISPRi and CRISPRko

To test whether CRISPRgenee’s improved performance (Figure 4) resulted from combining CRISPRko and CRISPRi sgRNAs on one construct, as suggested by earlier experiments (Figures 1 and 3), we compared dropout effects in genes with two targeted TSSs. We identified genes showing strong LFC differences between two TSSs (Figure S12A), independent of their distance (Figure S12B). Additionally, we did not identify any correlation between LFC and distance of the targeting KO sgRNA to the TSS for essential genes (Figure S12C), suggesting no epigenetic silencing over long distances through retention of ZIM3-Cas9 after DNA cleavage. We further classified TSSs as functional (F-TSS, where CRISPRi enhances LOF) or non-functional (NF-TSS, no CRISPRi effect), revealing a significant gene-depletion difference when targeting different TSSs (Figures 5A and S12D). For instance, CRISPRgenee sgRNAs targeting the myeloid selective genes ADAR and ERG (DepMap)^61,62 resulted in a relatively moderate negative LFC when sgRNAs were directed against the NF-TSS. However, combining the same CRISPRko sgRNAs with sgRNAs targeting the F-TSS led to a drastic LFC decrease for all three CRISPRko sgRNAs (Figure 5B). Since the CRISPRi sgRNAs used to assemble the CRISPRgenee constructs were adapted from the CRISPRi-v2 library,¹³ we were able to compare the effects of targeting individual TSSs of the essential gene DDX46⁶⁰ in both CRISPRi-v2 as well as CRISPRgenee screens. The effect observed with sgRNAs for the F-TSS resulted in a strong negative LFC in both screens; however, only the CRISPRgenee approach was able to produce a negative LFC when targeting the NF-TSS (Figure 5C). This improved performance was also visible when comparing false-positive rates determined by non-essential genes and plotted against the true-positive rate for both screens (Figure 5D) suggesting that combining CRISPRi with CRISPRko leads to a significantly improved gene-dropout effect in multiplexed screening approaches (Figure S12E).

Figure 5.

Combination of CRISPRi and CRISPRko has a highly additive effect in a multiplexed LOF screening setup

(A) Differential depletion effect when targeting the functional (F) and non-functional (NF) TSS in the CRISPRgenee screening approach to the effect observed in a published screening dataset using the same sgRNAs.²⁷

(B) Performance of individual sgRNA combinations demonstrating that CRISPRgenee improves the KO effect observed when targeting the NF-TSS compared to when targeting the F-TSS of essential genes.

(C) CRISPRgenee rescues nonfunctioning CRISPRi indicated by the depletion observed when the NF-TSS is targeted compared to a published CRISPRi screen using the same sgRNAs.¹³

(D) ROC sensitivity curve of the CRISPRgenee screen compared to a published CRISPRi screen that utilized the same CRISPRi sgRNAs¹³ based on sgRNAs targeting essential and non-essential genes.

(E) Individual sgRNA performance comparison of CRISPRgenee with a published screening approach employing a dual-CRISPRi sgRNA strategy targeting the same TSS.²⁷ The CRISPRi sgRNAs identical in both screens are highlighted (blue) as well as the CRISPRi sgRNA solely used in the CRISPRgenee screen (red).

(F) Boxplot depicting the ΔLFC difference when targeting the F and NF-TSS comparing CRISPRgenee with a published dual-CRISPRi screening dataset²⁷ (∗p ≤ 0.05; non-parametric Wilcoxon test).

See also Figure S12.

To verify the combinatorial phenotypic effect of our CRISPRgenee library, we conducted the same analysis for a dual-CRISPRi screen utilizing the same sgRNAs^13,27 and observed a significantly higher ΔLFC between the sgRNAs targeting individual TSSs of the same gene in the CRISPRgenee approach compared to the dual-CRISPRi screen (Figure 5E). Again, we observed the same improvement when combining a CRISPRi with a CRISPRko in the dropout effect (Figure 5F) as illustrated for SETD8 and ADAR (DepMap common essential genes) (Figure 5F). These data suggest that the additional CRISPRko effect has a greater influence on the LFC than targeting the same TSS with two CRISPRi sgRNAs, as seen when comparing sgRNA performance on essential genes in CRISPRgenee TF-1 screens versus CRISPRi-v2 and dual-CRISPRi LOF screens (Figure S12E).

Hit validation illustrates specific CRISPRgenee benefits

Finally, we benchmarked the CRISPRgenee screen against DepMap dependency scores (https://depmap.org/portal/) for the same cell line, first comparing the essential genes targeted in the CRISPRgenee library and observing similar effects between the CRISPRgenee screen and DepMap annotation (Figure S12F).^61,62 After k-means clustering, we observed four distinct groups (Figure 6A). Group 3 had strong negative DepMap scores but positive/slightly negative CRISPRgenee effects, whereas group 4 showed strong negative CRISPRgenee LFCs but positive/slightly negative DepMap scores (Figure 6A). From the 147 genes in group 4, we focused on TDRD12, KDM1A, GFI1B, and DPF1, which showed minor DepMap depletion but strong CRISPRgenee phenotypic effects (Figure 6B). Interestingly, KDM1A (LSD1) was recently shown to retain residual mRNA and protein in CRISPRko approaches due to alternative splice sites rescuing mutation effects,^{2,3,4,5,6,7,8,9} potentially explaining the weak phenotypic effects in previous screens,^61,62 although TF-1 cells are sensitive to LSD1 inhibition.⁶³ In addition, GFI1B was shown to play an important role in LSD1 biology⁶⁴ and scored as the top-20 hit in our screen (Figure 6B). We assessed the performance of all sgRNAs targeting the respective genes, confirming that all single CRISPRgenee sgRNAs resulted in a negative LFC (Figure 6C) and validated the CRISPRgenee benefits, comparing its phenotypic effects to individual sgRNAs used to design the CRISPRgenee construct (CRISPRi-v2 and Vienna Bioactivity CRISPR score)^13,31 as well as the Brunello/AVANA sgRNAs¹⁵ used to derive the dependency score from DepMap. In all cases, CRISPRgenee outperformed single sgRNAs (Figure 6D). So far, the central role of the LSD1 and GFI1B interaction in leukemia^63,64 has never been observed in a multiplexed LOF screening approach, although the LSD1-GFI1B regulatory axis is critical for cell proliferation.^63,64 Again, we identified rescue effects, with one sgRNA of the CRISPRgenee construct contributing more to the negative proliferative effect in case the CRISPRi (DPF1, TDRD12) or CRISPRko (GFI1B, KDM1A) did not induce significant effects. These data further highlight the beneficial effect achieved by the combination of CRISPRi and CRISPRko, demonstrating the relevance of the CRISPRgenee system for future LOF approaches.

Figure 6.

CRISPRgenee can identify novel dependencies in multiplexed LOF screening approaches

(A) Heatmap showing the LFC of genes targeted in the CRISPRgenee screen compared to the dependency score of TF-1 cells from DepMap and clustered into four distinct groups using k-means clustering.

(B) Scatterplot showing the CRISPRgenee gene-level depletion compared to the dependency score from the DepMap portal (https://depmap.org/portal/) for TF-1 cells. Cluster 4 is depicted in red and the differential genes selected for validation of the improved CRISPRgenee effect are highlighted with a black circle.

(C) Individual sgRNA performance of sgRNAs targeting the indicated differential genes.

(D) Competitive proliferation assays in TF-1 cells expressing ZIM3-Cas9 and the indicated sgRNAs. For the validation, the highest-performing CRISPRgenee combination and the resulting single sgRNA controls as well as the best in silico predicted CRISPRko sgRNA from the Avana library¹⁶ used to contribute to the DepMap dependency score and a CRISPRgenee combination using this KO sgRNA was used (n = 3, mean ± SEM).

See also Figure S12.

Discussion

Pooled LOF genetic screens are highly effective in investigating specific phenotypes like proliferation or cell survival. Efforts to reduce library size to reduce cost and increase scalability often rely on fewer sgRNAs per gene,^{13,15,16,31,65,66} but this can lead to the discovery of false-positive hits (Figure S12E). Meanwhile, standard CRISPRi and CRISPRko screens face limitations in suppression efficiency, particularly with essential or tightly regulated genes.^{17,27,29,67,68} This hinders a broader application of pooled CRISPR LOF screening approaches with complex molecular readouts or when cell numbers are limited.

To address these problems, we developed CRISPRgenee, a dual-sgRNA approach combining a 20-nt sgRNA for Cas9-mediated cleavage and a truncated 15-nt sgRNA for epigenetic repression at the TSS (Figure 1). We confirmed that 15-nt sgRNAs do not induce cleavage and provide faster initial silencing likely due to lower ΔG for duplex unwinding and RNA secondary-structure disruption.⁶⁹ This combination significantly improved phenotypic effects, even when single guides were ineffective (Figures 1, 3, 5, and 6), and performed reliably in non-tumorigenic cells like MCF10A and iPSCs without affecting proliferation or differentiation (Figure 3). Finally, we confirmed that CRISPRgenee can also be used as an LOF tool for genetic-dropout screens (Figure 4) performing similarly to comparable pooled LOF screens (Figures 5 and 6).

Truncated guides theoretically raise concerns about off-target binding, but our transcriptomic analysis found no significant differences between 15- and 20-nt sgRNAs (Figures 2 and S3). Consistent with literature, off-target effects are largely governed by the seed region, not PAM-distal bases.^37,70,71 A large portion of Cas9 off-target analysis involves DNA-cleavage assays rather than recruitment assays of Cas9 to DNA. More recent studies investigating dCas9 off-target binding also confirm that mismatches within the seed region lead to severely reduced dCas9 association rates,^41,72 demonstrating that the binding of Cas9 mainly depends on five bases within the seed region.^42,43

By using three CRISPRgenee sgRNA combinations per gene and requiring two to show consistent effects, we maintain statistical robustness even with reduced library size (Figure S12G).⁴¹

To streamline dual-sgRNA cloning, we used a compact H1/7SK promoter system under 300 bp, enabling accurate synthesis and low chimera rates (∼5%) compared to ∼30% in other systems²⁷ (Figure S7A). While alternative cloning strategies exist, our design minimizes recombination and supports paired-end sequencing for sgRNA pair validation. However, several sophisticated cloning strategies for the generation of dual-sgRNA libraries have already been published, which can be employed.^{26,27,73,74,75}

Overall, our initial findings that the combination of gene and epigenome engineering improves individual CRISPRko and CRISPRi LOF approaches were confirmed in a multiplexed screening approach, ranking as comparable to or higher than other published screening approaches in terms of consistency in performance across sgRNAs targeting the same gene, thereby increasing the significance of hit calling.

Beyond standard dropout screens, CRISPRgenee can be applied to dual-CRISPRi or CRISPRko formats and adapted to orthogonal screening, including in vivo mouse or patient-derived xenograft (PDX) models and primary patient contexts. When combined with emerging technologies such as single-cell and time-resolved transcriptomics,^76,77,78 orthogonal screening approaches,⁷⁹ or organoid model screening,⁸⁰ CRISPRgenee offers a powerful tool to dissect genetic dependencies at scale and resolution previously unattainable.

Limitations of the study

As a hybrid of CRISPRi and CRISPRko, CRISPRgenee cannot fully compensate when both sgRNAs are ineffective. This limitation can be mitigated by using ≥3 sgRNAs per gene and advanced design algorithms. We also observed high chimera rates during library amplification, likely due to template switching and recombination, which can be reduced through optimized dual-sgRNA cloning strategies.

Although we did not observe adverse effects in our models, Cas9-induced DSBs could potentially trigger DNA damage responses or transient cell-cycle arrest, possibly leading to false positives in essential gene screens.⁸¹ However, prior work indicates single cuts may not activate γH2AX signaling.⁸²

The use of truncated sgRNAs for dCas9-ZIM3 targeting increases the theoretical risk of perfect off-target binding. While our transcriptomic analyses did not reveal significant effects, the full extent of off-targets remains to be determined. Until more precise models emerge, off-target effects should be controlled using conservative thresholds (e.g., requiring consistent LFC reductions from multiple sgRNAs, as applied in our screen).

Looking forward, CRISPRgenee shows strong potential as a go-to approach for LOF studies and high-throughput screening. While its advantages were evident in TF-1 cells, broader applications across diverse cell types, in vivo models, and disease contexts are needed to fully assess its generalizability. Continued functional validation will help establish CRISPRgenee’s robustness, precision, and therapeutic potential. Addressing current limitations will further refine the system and unlock its full utility in genetic and biomedical research.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Philipp Rathert (Philipp.rathert@ibtb.uni-stuttgart.de).

Materials availability

Plasmids generated in this study are available from Addgene (# 239603, 239604, 239605, 239608, 239609, and 239610).

Data and code availability

• •The raw CRISPRgenee screening files, the off-target RNA-seq, as well as amplicon sequencing of the truncated and non-truncated sgRNAs are available at GEO (GEO: GSE238225). All other data are available from the corresponding author upon request.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank all laboratory members for constructive discussions and Regina Philipp and Ama Amoateng for technical support. We are grateful to A. Jeltsch for advice throughout this study. We thank the Twist Bioscience team, in particular Tavneet Gill, Xianan Liu, Caitlin Hoeber, and Carlo Antonio Bilbao, for cloning the CRISPRgenee library. pCAG-Eco was a gift from Arthur Nienhuis and Patrick Salmon. pCMVR8.74 was a gift from Didier Trono.

We thank Johannes Zuber (The Research Institute of Molecular Pathology [IMP], Vienna, Austria) for sharing reagents and all members of the Rathert and Jeltsch labs for reagents, protocols, and discussions.

The work described here was supported by the Wilhelm Sander Foundation (2016.082.1 and 2020.055.1) and the German Cancer Aid (70113426). The work of S.A.J. and Z.N. was funded by the Robert Bosch Stiftung. The work of C.B. and F.R.T. was funded by CRC1309 (grant no. 325871075, project C08).

Author contributions

Conceptualization, J.J. and P.R.; data curation, J.S.; formal analysis, J.S. and P.R.; funding acquisition, P.R., S.A.J., S.L., and F.R.T.; investigation, J.S., H.M., J.W., C.B., L.H., and Z.N.; project administration, P.R.; resources, C.B., L.H., M.K., S.L., S.A.J., and F.R.T; supervision, S.L., F.R.T., and P.R.; visualization, J.S. and P.R.; writing – original draft preparation, J.S. and P.R.; writing – review & editing, J.S., J.J., and P.R.

All authors have read and agreed to the published version of the manuscript.

Declaration of interests

J.J. is an employee of Twist Bioscience.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-CD90.1(Thy1.1)-APC	Thermo Fisher	AB_469420
Anti-CD90.1(Thy1.1)-APC	BioLegend	AB_1595470
PE anti-human CD33	BioLegend	AB_314347
PE anti-human CD13	BioLegend	AB_314179
anti-CD324 (E-Cadherin) - clone 4A2 monoclonal mouse	Cell signaling	AB_2728770
anti-Vimentin - clone D21H3 - monoclonal rabbit	Cell signaling	AB_10695149
Alexa Fluor 647 goat anti-mouse IgG (H + L)	Thermo Fisher	AB_2536165
Alexa Fluor 488 goat anti-rabbit IgG (H + L)	Thermo Fisher	AB_143165
TRA-1-60 Antibody, anti-human, REAfinity™	Miltenyi Biotec	AB_2654228
Bacterial and virus strains
E. coli Stbl3™	Invitrogen	C737303
Chemicals, peptides, and recombinant proteins
Recombinant human GM-CSF	Miltenyi Biotec	130-095-372
hHOLO-Transferrin	Merck Millipore	616424
hrInsulin	Sigma-Aldrich	I9278
hFGF-2-IS	Miltenyi Biotec	130-104-921
hrTGF-b1	Miltenyi Biotec	130-095-067
Deposited data
CRISPRgate screen, RNA-seq and amplicon seq	This paper	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE238225
Experimental models: Cell lines
NIH3T3	ATCC	CRL-1658
LentiX293T	Takara Bio	632180
HEK293T	DSMZ	ACC 635
TF-1	ATCC	CRL-2003
iNGNs	Collaboration partner	N/A
MCF10A	ATCC	CRL-10317
Oligonucleotides
Primers for cloning, amplification cleavage site, qPCR and illumina sequencing see Table S1	This paper	N/A
sgRNAs used in this study see Table S2	This paper	N/A
CRISPRgate library see Table S6	This paper	N/A
Recombinant DNA
Plasmids generated in this study are deposited at addgene see Table S3	N/A	N/A
Software and algorithms
Graphpad Prism 5	Graphpad Software, Inc	N/A
Illustrator CS6	Adobe	N/A
Excel 2016	Microsoft	N/A
Tableau 2023.2	Tableau	N/A
R	The R Project	N/A
R-studio	Posit	N/A
Other
Normalized screen count of all analyzed screens in this publication see Table S7	This paper	N/A
Normalized RNA-seq count and DE-seq analysis see Table S5	This paper	N/A
Amplicon sequencing count see Table S4	This paper	N/A

Experimental model and study participant details

Cell culture

All media were supplemented with 10% Fetal Bovine Serum, 4 mM L-Glutamine, 10 mM HEPES, 1 mM Sodium pyruvate solution, 100 U/mL Penicillin and 100 μg/mL Streptomycin. NIH/3T3 (male), Lenti-X293T (female) and HEK293T (female) cells were cultivated in DMEM high glucose media (Sigma-Aldrich) and TF-1 (male) cells were cultivated in RPMI 1640 supplemented with 2 ng/mL of recombinant human GM-CSF (130-095-372 Miltenyi Biotec). MCF10A cells were seeded in in DMEM/F12 medium (Thermo Fisher, #21331020) supplemented with 100 ng/mL cholera toxin (Sigma, #C8052), 20 ng/mL epidermal growth factor (EGF) (Preprotech, #AF-100-15), 10 μg/mL insulin (Sigma, #I9278), 500 ng/mL hydrocortisone (Sigma, #H0888), GlutaMax (Thermo Fisher, #35050038), 5% horse serum (Thermo Fisher, #16050122), 1× penicillin/streptomycin (Thermo Fisher, #15140122). iNGNs⁵⁵ were grown on Geltrex (ThermoScientific #A1413202)-coated tissue plates and cultivated at 37°C in water-saturated, CO₂-enriched (5%) atmosphere. Uninduced iNGNs were cultured in hiPSCs-medium (1:1 DMEM:F-12, GlutaMAX-Supplement (Gibco #10565018), supplemented with 0.2 mM L-ascorbic acid 2-phosphate, 77.6 nM sodium selenite, 10.90 mM NaCl, 0.1 mM nicotinamide, 10 μg/mL hHOLO-Transferrin (Merck Millipore #616424), 20 μg/mL hrInsulin (Sigma-Aldrich #I9278), 20 ng/mL hFGF-2-IS (Miltenyi Biotec #130-104-921), 2.0 ng/mL hrTGF-b1 (Miltenyi Biotec #130-095-067).

Method details

Plasmids

The DNA sequence encoding ZIM3-KRAB (ZIM3), without the stop codon, was synthesized and cloned in front of Cas9 in a Dox inducible pRRL-TRE3G-Cas9-P2A-GFP plasmid kindly provided by the Zuber lab²² using standard cloning methods. ZIM3-Cas9 expression was coupled to GFP via a P2A element. Additionally, antibiotic resistance against Blasticidin driven by a PGK promoter was added to select for positively transduced cells (TRE3G-ZIM3-Cas9-NLS-P2A-GFP-PGK-BlastR). For the constitutive CRISPRgenee vector, the ZIM3 element without the stop codon was cloned in front of Cas9 of an EF1as-Cas9-P2A-GFP plasmid using standard cloning methods. Additionally, a Blasticidin resistance driven by a PGK promoter was added. For dual sgRNA expression, a dual filler plasmid was cloned adding an additional tracr and promoter to a lentiviral single sgRNA expression vector kindly provided by the Zuber lab²² using an hU6 promoter to express the first sgRNA and either the H1 or minimal H17SK to express the second plasmid. Successful transduction of the dual sgRNA expressing plasmid was monitored via an outer membrane protein (Thy1.1) driven by an Ef1a short promoter and coupled to Neomycin resistance via a P2A element (hU6-filler1-tracr-H1/minH17SK-filler2-tracr-Ef1as-Thy1.1-P2A-Neo).

sgRNA and library cloning

The CRISPRi and CRISPRko sgRNAs were ordered as complementary oligos harboring overhangs fitting to the BsmBI restriction site of the respective fillers. The complementary sgRNA oligos were phosphorylated, annealed and afterward cloned into the dual-sgRNA filler plasmid using golden-gate assembly.

For the pooled sgRNA library we designed 270 bp long oligo fragments consisting of the 15 nt CRISPRi sgRNA, a tracr, the minimal H1/7SK promoter and the 20 nt CRISPRko sgRNA. TWIST Biosciences pool synthesized the oligo fragments and cloned these into the filler of a PRRL lentiviral sgRNA vector (PRRL-PBS-hU6-filler-tracr-Ef1as-Thy1.1-P2A-Neo) harboring a primer binding site for PCR amplification for Illumina sequencing kindly provided by the Zuber lab.²² TWIST performed library quality control and deep sequencing, identifying an initial chimera rate of 9.78%.

Cell culture, lentiviral transduction, generation of Tet-on competent cells and single-cell clones

Generation of Tet-on competent cells was performed as previously described.⁵⁷ For lentiviral packaging of pRRL-vectors, plasmids were mixed with helper plasmids pCMVR8.74 (Addgene plasmid #22036) and pCAG-Eco (Addgene plasmid #35617) and 3× (w/w) excess of polyethyleneimine 25K in DMEM. The mix was added dropwise to LentiX cells at 70–80% confluency. Media was exchanged after 12 and 24 h. The virus particles were harvested 48h after transfection with an optional second and third harvest after 56 to 72 h. To prevent contamination of the target cells with LentiX, the virus was filtered using a 45 μm filter. pRRL vectors for expression of ZIM3-Cas9 were introduced into the Tet-on competent target cells by transduction at a transduction efficiency <20% to ensure single plasmid integration. For selection, TF-1 cells were treated with 4 μg/mL Blasticidin. The dual sgRNA expression vector was transduced and the cells were selected with 500 μg/mL G418 solution for 7 days. Afterward, expression of ZIM3-Cas9 was induced by the addition of 1 μg/mL Doxycycline (Dox). Successful integration and expression of the desired plasmids was analyzed two days after transduction and monitored throughout selection and flow cytometry using a MACSQuant Vyb flow cytometer. To generate single-cell clones for the CRISPRgenee screen, expression of ZIM3-Cas9 was induced in TF-1 cells and single-cell clones were sorted based on the GFP fluorescence using the Sony SH800S FACS. The single-cell clones were monitored based on the GFP signal, keeping the cells that were able to reversibly induce ZIM3-Cas9 expression by addition and removal of Dox. We then validated the single-cell clones by transducing these with sgRNAs targeting CD13 and CD33, induced the expression of ZIM3-Cas9 using Dox and monitored the depletion of CD13/CD33. We chose three distinct sgRNAs which all showed similar CD13/CD33 depletion efficiencies and were able to specifically respond to Dox induction and removal.

Immunodetection of CD13 and CD33 depletion

sgRNAs targeting CD13 and CD33 (Table S2) were stably integrated using lentiviral transduction. ZIM3-Cas9 expression was induced by the addition of 1 μg/mL Doxycycline (Dox) and the loss of functional CD13 and CD33 was monitored every two days for 14 days through AB staining of CD13 and CD33 and detection by flow cytometry.

RNA-seq for off-target detection

Two sgRNAs targeting CD33, were selected based on their in-silico predicted off-target activity (minimum of one perfect off-target for the 15mer) and cloned as standard 20 nt sgRNA or the respective 15 nt truncation. The sgRNAs targeting CD33 as well as a scrambled sgRNA (scr) were transduced into dCas9-tagBFP-ZIM3 expressing cells and selected for one week. After two weeks of sgRNA expression, the cells were sorted based on the dCas9-ZIM3 and sgRNA expression, harvested and RNA was isolated using the RNeasy Plus Mini Kit (QIAGEN, Hilden, Germany). RNA concentration and quality were assessed using the 260/280 and 260/230 ratios obtained at the NanoDrop. We enriched for mRNA using the NEBNext Poly(A) mRNA Magnetic Isolation Module (NEB) and used this as input for library generation using the NEBNext Ultra II RNA Library Prep (NEB). The quality and concentration of the cDNA library was assessed using the BioAnalyzer 2100 high sensitivity DNA kit. In case adapter fragments were detected, size exclusion DNA purification using NEBNext Sample Purification Beads (NEB) was performed according to the manufacturer’s instructions. The samples were sent for deep sequencing and the obtained reads were filtered using Trimmomatic⁸³ removing adapter contamination and filtering reads based on sequencing quality, keeping high-quality reads for further analysis. Next, we aligned both forward and reverse reads to the human genome (hg38) using HISAT2⁸⁴ and obtained the gene count files, which were used for DESeq2 analysis⁸⁵ comparing both sgRNAs (15 vs. 20) with each other as well as comparing the individual sgRNAs and the scr control (Table S5).

Competitive proliferation assay

TF-1 cells were transduced with the indicated sgRNAs (Table S2) and ZIM3-Cas9 expression was induced using Dox. The amount of Cas9 and sgRNA positive cells were measured after two days and set as reference. The proliferation of ZIM3-Cas9 and sgRNA-positive cells were monitored for the indicated time points. HEK293T cells were transduced with the indicated sgRNAs and selected for one week. Afterward, 80% of sgRNA-positive cells were mixed with 20% WT cells, and expression of ZIM3-Cas9 was induced using Dox. Two days after, the amount of sgRNA and ZIM3-Cas9 positive cells were measured and set as reference, and the proliferation was monitored for the indicated time points.

Cell cycle analysis

200 000 HEK293T cells were harvested after five days of BUB1 depletion and fixed in 70% ice-cold ethanol for 2h. Afterward, ethanol was removed by centrifugation and the pellet was resuspended in 200 μL cell cycle staining solution (100 μg/mL RNAseA, 50 μg/mL PI, 0.1% Triton X- in PBS) and incubated for 30 to 60 min at room temperature in the dark. Directly after incubation, the DNA content was measured by flow cytometry and the cell cycle tool (model Dean-Jett-Fox) of FlowJo was used to automatically detect the cell cycle phase distribution of the different samples.

Gene expression analysis

Cells were harvested after five days of BUB1 depletion, and mRNA was extracted by using the RNeasy Plus Mini Kit (QIAGEN, Hilden, Germany). RNA concentration and quality was assessed using the 260/280 and 260/230 ratios obtained at the NanoDrop. Reverse transcription was performed with 500 ng of purified RNA using oligo(dT)₁₈ primers for the Multiscribe Reverse Transcriptase (Invitrogen) according to the manufacturer’s instructions. Quantitative PCR reactions were carried out using the CFX Connect Real-Time System from Bio Rad using the ORA SEE qPCR Green ROX H Mix (highQu) and the human BUB1 and ACTIN primer set (Table S1). The following cycling conditions were performed: 95°C for 3 min, 39 cycles of 95°c for 5 s then 60°C for 30s with the cycling conditions for the melt curve performed afterward.

Amplicon sequencing and T7 endonuclease assay

TF-1 cells were transduced with 20 nt or 15 nt long sgRNAs targeting either CD13 or CD33 as well as with a scr control sgRNA and selected for seven days. ZIM3-Cas9 expression was turned on using Dox and the cells were harvested after 14 days. Primers were designed to amplify the sgRNA targeting region either for amplicon sequencing or mismatch cleavage assay (Table S1). Genomic DNA of >1 Mio sgRNA positive and WT cells was amplified and extracted from an agarose gel and the DNA concentration was determined. For amplicon sequencing, Illumina adapters were added to the PCR product using a second PCR step, the resulting PCR amplicons were then extracted from an agarose gel. The size, concentration, and purity of the DNA amplicon was assessed using the BioAnalyzer 2100 high sensitivity DNA kit. All amplicons were pooled and sent for Illumina sequencing. For analysis, the reads were split based on the internal barcode, the site of cleavage was identified and the percentage of InDels was calculated for the scr control, 15 nt and 20 nt sgRNAs (Table S4). For the T7 assay, 200 ng of PCR product (WT and sgRNA) was stepwise annealed and T7 endonuclease was added. Cleaved fragments were determined using agarose gel electrophoresis.

TGFβ induced EMT

MCF10A cells stably expressing dCas9-ZIM3 or ZIM3-Cas9 were transduced with sgRNAs targeting SMAD2, the cells were stained using PE-labeled antibodies against Thy1.1 (receptor co-expressed with the sgRNA) and sgRNA-positive cells were sorted two days after transduction. After one week of sgRNA expression cells were supplemented with 1:1000 Doxycycline to induce the expression of ZIM3-Cas9. One day after initial seeding, MCF10A cells were treated for 8 days with 100 pM TGFβ1 (Prepotech, #100-21C) with re-stimulation every 48 h, and harvested by TrypLETM (ThermoFisher, #12604013). For cell fixation and permeabilization, 100.000 cells per condition were treated with 4% PFA in DPBS for 15 min, and 0.1% Triton in FACS media (DPBS +5% FHS) for 15 min at 4°C. Cells were then incubated with anti-CD324 (E-Cadherin) antibody (1:150) and anti-Vimentin antibody (1:150) for 1 h. After blocking in FACS media for 10 min, cells were incubated with secondary antibodies Alexa Fluor 647 goat anti-mouse IgG (H + L), (1:1000) and Alexa Fluor 488 goat anti-rabbit IgG (H + L) (1:250) for 45 min. All antibody incubations were performed in FACS media. Cells were washed between fixation with DPBS and during antibody staining twice with FACS media and afterward analyzed by flow cytometry. For validation of the SMAD2 depletion, a subset of cells was harvested before TGFβ stimulation, and the RNA was isolated and reverse-transcribed into cDNA. Using specific primers for SMAD2 (Table S1) the relative expression compared to ACTIN was measured using qPCR as described in “Gene expression analysis”.

Neuronal stem cell differentiation and analysis of Tra-1/60

iNGNs were passaged in the uninduced state when they reached about 70–80% confluency. For passaging, 2 μM (final concentration) of thiazovivin (Merck Millipore 420220) was added to the hiPSCs medium for 24 h, afterward medium was changed to hiPSCs-medium without thiazovivin. iNGNs were stably transduced with EF1a-ZIM3-Cas9-P2A-GFP-PGK-BlastR and selected for one week using 2μg/mL blasticidin. Successful selection and expression of ZIM3-Cas9 was monitored by measuring the expression of GFP at the MacsQuant. Afterward, sgRNAs targeting ART1 or Ngn2 were stably transduced and cells were selected for one week using 100–200 μg/mL G418 solution. For induction of neuronal differentiation, the medium was changed to induction medium (hiPSCs-medium without FGF2-IS and TGF-β1), and Doxycycline (Sigma-Aldrich #D9891) was added to a final concentration of 0.5 μg/mL. Neuronal differentiation was induced for three days before harvesting for qPCR and Tra-1/60 antibody staining.

Multiplexed CRISPRgenee LOF screen

A CRISPRgenee library designed to target 1137 target genes involved in chromatin regulation using a set of 3686 sgRNAs (Table S6) was transduced into three independent TF-1 cell clones with comparable ZIM3-Cas9 expression levels of with an sgRNA representation of 1 000 ×. After 7 days of antibiotic selection, ZIM3-Cas9 expression was induced using Dox and cells were cultivated for 14 days (7 passages, 70 h doubling time). During cultivation, sgRNA representation and ZIM3-Cas9 expression was monitored using flow cytometry. After 14 days (7 passages), genomic DNA for the three single-cell clones was isolated using phenol-extraction using PhaseLock tubes, followed by ethanol precipitation. Multiple parallel 50 μL PCR reactions, each containing 1 μL gDNA template adding up to 40 μg gDNA and 300 ng for the library pool, using the AmpliTaq Gold Polymerase (Life Technologies) were performed to maintain sgRNA representation. In a first round of PCRs, random barcodes and sample barcodes were added to the sgRNA sequences using the following cycling parameters: 95°C for 10 min; 28 cycles of (95°C for 30 s, 54°C for 45 s and 72°C for 60 s); 72°C for 7 min. PCR products for each single cell clone were combined and purified using the NucleoSpin Gel and PCR clean-up kit (Macherey-Nagel). Afterward using a second round of PCR, using similar cycling conditions with the exception of 10 ng input as template and 7 total cycles the standard Illumina P7 and P5 adaptors were added. All primers used for the library preparation are listed in Table S1. The final libraries were cleaned up from a 2% agarose gel, pooled, and analyzed on a P2 flow cell (400 mio reads) using the Illumina NextSeq 2000 with a 35% PhiX spike-in (75 bp paired-end), using standard Illumina primers. Sequence processing was performed using a custom Galaxy workflow (www.usegalaxy.eu). sgRNA count data, as well as data used for comparison from other screens, are provided in Table S7. Forward and reverse reads were combined and non-mapped reads as well as sgRNA chimeras were filtered out. The read count for each sgRNA combination was normalized as count per million using the r-studio tool “CB2”.⁵⁹ Using the same tool fold depletion of individual sgRNA combinations as well as for the targeted genes was calculated.

Quantification and statistical analysis

All details regarding statistical analysis are provided in the respective figure legends and figures, including numbers of replicates for each experiment, statistical tests used and the obtained p values. Results are presented as means ± standard error of the mean [S.E.M.]. If not stated otherwise, statistical significance was calculated by two-way ANOVA with a post-hoc test indicated for each experiment with p ≤ 0.05 considered statistically significant. Statistical significance levels are denoted as follows: ∗∗∗∗p ≤ 0.0001; ∗∗∗p ≤ 0.001; ∗∗p ≤ 0.01; ∗p ≤ 0.05; n.s. = non-significant.

Published: June 10, 2025