Cytotoxicity of activator expression in CRISPR-based transcriptional activation systems

Abstract

CRISPR-based transcriptional activation (CRISPRa) has extensive research and clinical potential. Here, we show that commonly used CRISPRa systems can exhibit pronounced cytotoxicity. We demonstrate the toxicity of CRISPRa vectors expressing the activation domains (ADs) of the transcription factors p65 and HSF1, components of the synergistic activation mediator (SAM) CRISPRa system. Based on our findings for the SAM system, we extended our studies to additional ADs and acetyltransferase core domains. We show that the expression of potent transcriptional activators in lentiviral producer cells can lead to low lentiviral titers, while their expression in the transduced target cells leads to cell death. Using inducible lentiviral vectors, we could not identify an activator expression window for effective SAM-based CRISPRa without measurable toxicity. The toxicity of current SAM-based CRISPRa systems hinders their wide adoption in biomedical research and introduces selection pressures that may confound genetic screens. Our results suggest that the further development of CRISPRa technology should consider both the efficiency of gene activation and activator toxicity.

CRISPR-based transcriptional activation (CRISPRa) is a powerful tool for controlling gene expression. Here, Liang et al. show that commonly used CRISPRa systems are surprisingly toxic to cells due to their potent activation domains, informing the application and development of this technology.

Introduction

The development of programmable Clustered Regularly Interspaced Short Palindromic Repeats (CRISPR)/CRISPR-associated protein (Cas)-based transcriptional activation (CRISPRa) tools is of high interest for research and clinical applications¹. In these approaches, transcriptional activators are recruited to specific sites in the genome, by fusion to endonuclease-inactivated Cas proteins (“dCas”, most commonly dCas9²) or through aptamers in the associated single guide RNA (sgRNA)^3–7. The appeal of CRISPRa over traditional cDNA expression approaches lies in its simplicity of sgRNA design, scalability, ability to multiplex, and ability to overexpress relevant gene isoforms and large transcripts from their endogenous loci. CRISPRa can furthermore target non-coding genes and regulatory loci for transcriptional activation. CRISPRa is most commonly achieved by the fusion of dCas9 with the transactivation domains (ADs) of transcription factors (TFs), which promote transcription by, in turn, recruiting transcriptional and epigenetic machinery, including general transcription factors, the Mediator complex, and chromatin-modifying enzymes. The first generation of CRISPRa vectors used several copies of an 11 amino acid peptide representing the minimal activation domain (AD) of the herpes simplex virus type 1 TF virion protein 16 (VP16)^3,5,7. More potent second-generation activators relied on recruiting additional ADs^4,6,8–10, either by fusion to dCas9 or through bacteriophage RNA-aptamers engineered into the scaffold portion of the sgRNA, or both. Among the most potent and commonly used CRISPRa approaches is the synergistic activation mediator (SAM) system^4,10 (Fig. 1a). In the SAM system, dCas9 is fused to four copies of the VP16 minimal AD (VP64) and loaded with an aptamer-modified sgRNA which in turn recruits the MS2 or PP7 bacteriophage coat protein (MCP/PCP)-fused ADs of p65 or HSF1. We will abbreviate MCP or PCP-fused p65^AD-HSF1^AD synthetic transcriptional activators as MPH and PPH here. The original SAM system consists of 3 lentiviral vectors (LVs)⁴, expressing dCas9-VP64, MPH, and the aptamer-modified sgRNA, respectively. A subsequent version of SAM aimed to improve the titer of the original MPH-encoding LV and reduce the number of required LVs to two vectors¹¹ by combining a PPH activator protein and an aptamer-modified sgRNA in a single vector (pXPR_502, Fig. 1b). Other systems rely on dCas9 or MCP fusions with several viral or cellular ADs, including VPR (composed of VP64, and p65 and Epstein-Barr virus RTA ADs), and the recently reported NFZ/NZF (composed of compact ADs from NCOA3, FOXO3, and ZNF473 in two different configurations)¹², MSN/NMS (composed of ADs from MRTF-A and STAT1, and an engineered AD from NRF2 called eNRF2, in two different configurations)¹³, and eN3x9 (including eNRF2 and three 9 amino acid ADs from MRTF-B and MYOCD)¹³. In a conceptually different approach¹⁴, with potentially distinct target preferences¹⁵, CRISPRa is achieved through dCas9- or sgRNA-mediated recruitment of histone or DNA-modifying enzymatic domains, such as the catalytic histone acetyltransferase (HAT) core domains of the human E1A-associated protein p300^16,17 or CBP^15,18,19.

Fig. 1. Published p65^AD-HSF1^AD-expressing LVs have low titers and result in lower-than-expected outgrowth after transduction.

a Schematic of the SAM CRISPRa system. sgRNA in orange, SL: stem-loop. b Schematic of pXPR_502, not to scale, showing Rous sarcoma virus (RSV)-human immunodeficiency virus hybrid 5′-long terminal repeat (RSV/5′LTR), packaging signal (Ψ), Rev response element (RRE), the U6 promoter, sgRNA location, central polypurine tract (cPPT), human PGK promoter (hPGK), Kozak sequence (K), nuclear localization signals (NLS), PCP, p65^AD and HSF1^AD coding sequences, separated from a puromycin resistance cassette (Puro^R) by a T2A ribosomal skipping peptide, self-inactivating 3′-LTR (SIN 3′LTR). The number at left indicates genome size in kb. c Schematic of pLC-ZsGreen-P2A-Puro, not to scale. Abbreviations as in (b), except for CMV promoter (CMV), ZsGreen coding sequence, and Woodchuck Hepatitis Virus posttranscriptional regulatory element (WPRE). The number at left indicates genome size in kb. d RNA titers of n = 3 independent lentiviral stocks from pLC-ZsGreen-P2A-Puro, or pXPR_502 expressing sgAAVS1 (AAVS1), Calabrese Set A (Cala-A), or sgCRBN-a1 (CRBN). All titers were significantly different from the ZsGreen control, determined using One-Way ANOVA with Tukey’s multiple comparison test. e Functional titers in parental BC-3 of the n = 3 independent LV stocks from (d). All titers were significantly different from the ZsGreen control, determined using One-Way ANOVA with Tukey’s multiple comparison test. f Growth curve analyses of pXPR_502-transduced BC-3/dCas9-VP64. Cumulative live cell counts relative to pLC-ZsGreen-P2A-Puro (ZsGreen)-control transduced cells. Cells were transduced at MOI 0.3, based on either functional (F) or RNA (R) titers. 3 independent repeats, using the 3-LV preps from (d, e). All values differed significantly from the ZsGreen control, determined using One-Way ANOVA with Tukey’s multiple comparison test. n = 3 independent repeats. g Fold increases over the previous passage on days 6, 9, and 12, from the experiments in (f), show that normally proliferating pXPR_502-transduced cells can be grown out 9 days after transduction. Values were normalized to the untransduced and unselected control samples (NT) and differed significantly from NT at each time point unless specified by ns, determined using One-Way ANOVA with Tukey’s multiple comparison test. h Western Blot analyses of CRBN, PPH, and GAPDH in representative lysates taken on day 3 or 12 after transduction from samples shown in (f, g). Endogenous HSF1 is marked “HSF1”. Molecular weight (kDa) markers are at left. For quantification over replicates, see (i). i Quantification of results as shown in (h). Data from 5 (day 3) or 2 (day 12) independent repeats. *** denotes adj. p < 0.001, determined using One-Way ANOVA with Tukey’s multiple comparison test. j As in (d), but using hygromycin-resistant control vector pLC-ZsGreen-P2A-Hyg²² and lenti MS2-P65-HSF1_Hygro⁴ or lentiMPH v2²⁵. Data are from three independent LV preparations. Values differed significantly from the ZsGreen control, determined using One-Way ANOVA with Tukey’s multiple comparison test. k As in (f), using LVs from (j). One parallel repeat with each stock, n = 3 overall. Results differed significantly from the control. Differences between the two MPH vectors were not significant. Significance was determined using One-Way ANOVA with Tukey’s multiple comparison test. Throughout, error bars represent SEM, and ns denotes “not significant”. Source data and adj. p-values are provided in the Source Data file.

We attempted to establish the SAM system for our work in primary effusion lymphoma (PEL) B cell lines, a cell model we have extensively used in CRISPR/Cas9 screens^20–22. During these experiments, we encountered difficulties with LVs encoding MPH or PPH fusion proteins, including apparently low lentiviral titers and an inability to obtain and maintain transduced cell pools at expected efficiencies in several different PEL cell lines. Here, we systematically investigated the technical barriers to adapting CRISPRa for our work. Our results suggest that vectors expressing activators used for CRISPRa exhibit pronounced cytotoxicity, leading to low lentiviral titers and target cell death. This toxicity could confound studies using CRISPRa and should be considered in the further development of this technology.

Results

SAM activation domain vectors are toxic under conditions used for screening

To investigate the reasons for our inability to implement robust protocols for SAM-based CRISPRa in PEL cell lines, we initially focused on a set of pXPR_502 vectors¹¹ (Fig. 1b), expressing the PPH activator fusion protein and either the genome-scale Calabrese sgRNA library set A¹¹ (Cala-A) or individual sgRNAs targeting the safe harbor locus adeno-associated virus integration site 1 (AAVS1) or the promoter of the non-essential gene cereblon (CRBN), which we have stably overexpressed in PEL cells using a lentiviral cDNA vector in an unrelated study²³. As a control, we used an LV expressing a ZsGreen-P2A-Puro^R cassette²² (Fig. 1c). All transfer vectors used in this study were well below the size limit for efficient packaging of HIV-based LVs (see Fig. 1b, c and below). We titered LV stocks by qRT-PCR and functionally in the PEL cell line BC-3²⁴. For functional titration, we counted the percentage of cells that survived puromycin selection for all vectors and additionally used flow cytometry for the ZsGreen vector (see “Methods” section). pXPR_502 vector preparations had lower qRT-PCR-based titers than the LV expressing ZsGreen (Fig. 1d), despite using an optimized amount of transfer vector during pXPR_502 packaging (see “Methods” section and compare relative titers in Fig. 1d to those in Fig. 2a, where we used a standard packaging protocol). The discrepancies in the calculated functional titers were even more evident than for the genomic RNA (LV-gRNA) content (Fig. 1e), which could result from a loss of transduced cells due to MPH toxicity soon after transduction, thereby confounding results from antibiotic selection.

Fig. 2. Toxicity results from the expression of strong ADs.

a RNA titers of lentiviral stocks used in (b), Figs. 4 (MPH and ΔPH) or 3 (all others) independent stocks. Titers of all pL2M-CMV vectors differed significantly from the CMV-Puro control, except for MPH-ΔPH, determined using One-Way ANOVA with Tukey’s multiple comparison test. For other comparisons, see the figure and Source Data File. b Relative live cell numbers of A375 cells following puromycin selection after transduction with pLC-ZsGreen-P2A-Puro at MOI 0.25 based on functional titration and other LVs based on LV-gRNA content relative to pLC-ZsGreen-P2A-Puro. 4 independent repeats, except for pXPR-sgAAVS1 (n = 3). Results from pL2M-CMV vectors differed significantly from the matched CMV-Puro control, unless indicated by ns, determined using One-Way ANOVA with Tukey’s multiple comparison test. For other comparisons, see the figure and Source Data File. c Schematics of the LV vectors used in this figure, not to scale, using abbreviations explained in Fig. 1b, c and in the text. Numbers at left indicate genome sizes in kb. d RNA titers for additional LVs, as in (a). Only the MPH vector had a significantly lower titer than the Puro control, while all AD-vectors had a lower titer than the ZsGreen control, determined using One-Way ANOVA with Tukey’s multiple comparison test. e RNA titers for additional LVs, as in (a). AD-vectors, except for eN3x9, had significantly lower titer than either control, determined using One-Way ANOVA with Tukey’s multiple comparison test. f, g Relative live cell numbers of A375 cells after transduction at MOI 0.25, as in (b). Live cell numbers of pL2M-CMV-AD transduced samples other than eN3x9 were significantly different from the Puro control, determined using One-Way ANOVA with Tukey’s multiple comparison test. Throughout, error bars represent SEM, **** denotes adj. p < 0.0001, *** adj. p < 0.001, ** adj. p < 0.01, * adj. p < 0.05, and ns “not significant”. Source data and adj. p-values are provided in the Source Data file.

To further quantify this effect, we transduced BC-3 expressing dCas9-VP64 or parental BC-3 at a multiplicity of infection (MOI) of ~0.3, either based on the calculated functional titer of each vector (“F”) or based on LV-gRNA copy numbers relative to the ZsGreen-expressing control vector (“R”). Similar MOIs are typically used in CRISPR screens to ensure delivery of one sgRNA per transduced cell. We selected the resulting cell pools using puromycin and performed growth curve analyses. For both cell lines, a close to expected fraction of the ZsGreen control vector-transduced cells survived puromycin selection (~39% compared to untransduced and unselected control cells on day 3 after transduction, ~6.5% range) and proliferated similarly to untransduced cells once puromycin selection was complete (Fig. 1f, g, Supplementary Fig. 1a, b). In contrast, dramatically fewer pXPR_502-transduced cells survived over time for either titration approach, likely indicating ongoing transgene toxicity (Fig. 1f, g, Supplementary Fig. 1a, b). This toxicity was independent of the specific sgRNA insert and the presence of dCas9-VP64 and associated gene activation, which we readily observed for sgCRBN (Fig. 1h, i). After continued passage, pXPR_502-transduced cell pools that proliferated similarly to the untransduced control pool were obtained by about day 9 after transduction, demonstrating that it is possible to obtain cells that have overcome PPH toxicity (Fig. 1g, Supplementary Fig. 1b). Western blot analyses show that these passaged cell pools had ~5-fold reduced expression levels of PPH (Fig. 1h, Supplementary Fig. 1c, d) and, therefore, either contain cells with low initial PPH expression or those that have undergone changes resulting in lower PPH expression. These cell pools had a reduced level of CRISPRa-based gene activation, which did not reach significance using sgCRBN (Fig. 1h, i). Low titer and severe toxicity in BC-3 cells were also evident with commonly used MPH-encoding vectors for the 3-LV SAM system^4,25, showing that our findings are not exclusive to pXPR_502 (Fig. 1j, k). In these experiments, differences between the original (lenti MS2-P65-HSF1_Hygro⁴) and an updated (lentiMPH v2²⁵) MPH vector did not reach statistical significance over three independent virus preparations and transductions.

Most MCP-fused AD-based CRISPRa activators are toxic across contexts

We next tested whether these observations were unique to the PEL model by repeating the experiment for pXPR_502-sgAAVS1 in the melanoma cell line A375, which was used in several published CRISPRa screens^4,11. Following transductions at ~MOI 0.25, based on LV-gRNA content and the functional titer of the ZsGreen-expressing positive control vector (bars 1 and 2 in Fig. 2a), pXPR_502-toxicity was also pronounced in A375 (bars 1 and 2 in Fig. 2b), suggesting that toxicity is not unique to the PEL model.

To map potentially cytotoxic components of the M/PCP-p65^AD-HSF1^AD fusion proteins, we constructed a lentiviral vector expressing only MCP-p65^AD-HSF1^AD (MPH), and versions lacking MCP (ΔM), p65^AD (ΔP), HSF1^AD (ΔH), or p65^AD-HSF1^AD (ΔPH). We focused on MCP fusion proteins because the MS2 aptamer is more commonly used. We constructed a matched control vector expressing only the puromycin resistance gene (CMV-Puro; for vector schematics, see Fig. 2c, and for expression controls, see Supplementary Fig. 2) and a vector expressing MCP-VP64. All fusion proteins were targeted to the nucleus using dual nuclear localization signals (NLS) flanking MCP, like in pXPR_502. After normalizing for LV-gRNA copies relative to the ZsGreen control, the CMV-Puro LV achieved the expected numbers of transduced cells, validating our titration strategy (Fig. 2b). In contrast, few cells survived transduction with the MPH-encoding LV. Deletion of each AD partially rescued vector toxicity, and deletion of both ADs eliminated toxicity (Fig. 2b and Supplementary Fig. 3a for BC-3). Interestingly, p65^AD-HSF1^AD deletion also rescued LV titers, suggesting that the low titers of MPH vectors likely result from AD toxicity in the producer cells (Fig. 2a). MCP-VP64-transduced cells did not experience toxicity (Fig. 2b), consistent with our ability to establish dCas9-VP64-expressing PEL cell lines. The lack of VP64 toxicity in A375 could be due to its weaker activator activity^4,10 and lower expression levels (Supplementary Fig. 2b, c). Using the same approach, we tested MCP fusions with additional AD-based CRISPRa activators, including VPR, and the recently reported NFZ/ NZF¹², MSN¹³, eN3x9¹³ AD cassettes (Fig. 2c–g). While NFZ and NZF were previously only tested following direct dCas-mediated recruitment¹², we tested them as MCP fusion proteins here, which could eventually improve AD potency during CRISPRa through multicopy recruitment through MS2 stem-loops, while mitigating MPH toxicity. NFZ/NZF fusion to MCP furthermore enables a direct comparison to other constructs in this figure. Similarly, VPR is typically used as a dCas9-fusion protein, but fused to MCP here. MSN and eN3x9 have been shown to be effective as MCP or dCas9-fusion proteins¹³. MCP-AD-encoding LVs other than NFZ/NZF and eN3x9 had significantly reduced viral titer (Fig. 2d, e). Except for eN3x9, MCP-AD fusions had measurable toxicity in A375 (Fig. 2f, g) and BC-3 cells (Supplementary Fig. 3b). eN3x9 is a recently developed activator¹³. eN3x9-encoding LVs have been shown to have higher titer and better tolerability in T cells than LVs expressing VPR. While the effectiveness of eN3x9 can approach that of MSN and SAM in assays with pooled guides¹³, this CRISPRa system has not yet been characterized comprehensively or used in genome-scale screens with only one guide per locus.

Toxicity of HAT-based CRISPRa activators

The core HAT domains of the lysine acetyl transferases p300 (p300^Core) or CBP (CBP^Core) have been developed for targeted histone acetylation, which can result in CRISPRa when directed to promoters or enhancer elements. Since the ADs of TFs often recruit p300 and CBP alongside other effectors, their high expression could result in competition for HATs with endogenous TF complexes and thereby impair the viability of AD-expressing cells. We reasoned that the targeted recruitment of HAT domains could overcome the toxicity associated with such competition effects. While the HAT core domains are typically fused to dCas9, MCP-fused HAT domains have previously been used with dCas9-VP64²⁶. We constructed MCP fusions with previously described human p300^Core and CBP^Core domains here^16,18 (Fig. 3a), to directly compare MCP-HAT^Core toxicity with that of the MCP-AD-fusion proteins above. For p300^Core, we included a mutant with inactivated acetyltransferase activity (p300^Core/D1399Y¹⁶) and furthermore tested the emerging p300^Core/I1417N mutant¹⁷, which alters the p300^Core interactome, but not its acetyl transferase activity or its ability to deposit the histone 3 lysine 27 acetylation mark (H3K27ac), a key activating histone modification. I1417N mutation has furthermore been proposed to reduce the toxicity associated with p300^Core expression and allow for higher lentiviral titers and improved recovery of transduced cells¹⁷. In our experiments, wildtype MCP-p300^Core and -CBP^Core constructs produced LVs at very low titers, while the titers of the p300^Core-D1399Y and I1417N mutants did not differ significantly from the Puro control LV (Fig. 3b). Both wildtype MCP-HAT^Core proteins were toxic in A375 and BC-3 (Fig. 3c and Supplementary Fig. 4), while both p300 mutations increased the number of surviving cells to levels comparable to control vector transductions. Wildtype MCP-HAT^Core-transductions at single copy resulted in a readily detectable increase in lysine acetylation events, including autoacetylation of the MCP-HAT^Core fusion proteins (Fig. 3d), suggesting an sgRNA-independent off-target activity of these fusion proteins, as reported elsewhere^15,17. The p300^Core-D1399Y HAT inactivating mutation eliminated off-target acetylation and autoacetylation of MCP-p300^Core after transduction of A375 or transfection of 293T (Fig. 3d and Supplementary Fig. 5a). p300^Core-I1417N mutation reduced off-target acetylation and autoacetylation to undetectable levels upon transduction at a single copy, while it strongly and selectively reduced most off-target acetylation events after transfection of 293T cells. Wt and I1417N mutant MCP-p300^Core resulted in similar increases of H3K27ac upon transfection in 293T cells, consistent with the original report¹⁷ (Supplementary Fig. 5b). Wildtype MCP-p300^Core-induced lysine acetylation was furthermore partially reversed by treatment with increasing concentrations of A-485 (Fig. 3e), an inhibitor of the catalytic activity of p300/CBP²⁷. A-485 treatment also significantly, albeit partially, rescued the survival of MCP-p300^Core-transduced cells (Fig. 3f, Supplementary Fig. 5c for additional controls). While the toxicity of A-485 precluded testing higher concentrations, the substantial rescue of MCP-p300^Core toxicity by partial HAT inhibition further validates our titration approach and confirms that the toxicity of this vector is due to p300 acetyltransferase activity. The absence of detectable toxicity of p300^Core-I1417N is intriguing, since it suggests that the toxicity of p300^Core is separable from its ability to acetylate H3K27, as proposed¹⁷. This mutant has been shown to be effective for CRISPRa, although with reduced efficiency at some loci compared to wildtype. The further development and characterization of p300^Core-I1417N for CRISPRa is therefore of high interest.

Fig. 3. Toxicity of MCP-HAT^Core domain fusion proteins.

a Schematics of the inserts used in this figure, as in Fig. 2c. b RNA titers of 3 (p300^Core/I1417N and CBP) or 6 (all others) independent LV stocks used in (c). Differences from the Puro control were significant, unless indicated by ns, determined using One-Way ANOVA with Tukey’s multiple comparison test. c Relative live cell numbers of A375 cells following puromycin selection after transduction with pLC-ZsGreen-P2A-Puro at MOI 0.25 based on functional titration and other LVs based on LV-gRNA content relative to pLC-ZsGreen-P2A-Puro. 4 (p300^Core/I1417N and CBP) or 7 (all others) independent repeats. Significance was determined using One-Way ANOVA with Tukey’s multiple comparison test. d Western Blot analyses in A375 two days after transduction at MOI 0.25, without selection, but otherwise as in (c). Due to the low MOI, off-target acetylation is detected against the background of a majority of untransduced cells. Bands marked by an asterisk likely represent autoacetylated MCP-HAT^Core proteins. Molecular weight (kDa) markers are at left. Representative of 3 independent repeats in this exact configuration. e As in (d) but including treatment with the HAT inhibitor A-485 at the indicated concentrations. Molecular weight (kDa) markers are at left. Representative of 3 independent repeats. f The relative live cell number of MCP-p300^Core-transduced cells, assayed as in (c), including treatment with DMSO (0 μM A-485) or indicated concentrations of A-485. Survival compared to the DMSO-treated control was significantly increased, except for the lowest A-485 concentration, as indicated by ns, determined using One-Way ANOVA with Tukey’s multiple comparison test. See Supplementary Fig. 5c for an extended version of this panel. 3 independent repeats. Throughout, error bars represent SEM, **** denotes adj. p < 0.0001, and ns “not significant. Source data and adj. p-values are provided in the Source Data file.

Inducible MPH expression is toxic in various cell lines

In the experiments shown above, activator toxicity occurred already during the antibiotic selection of the transduced cells, making it difficult to distinguish low titer from transgene toxicity. Our finding that cells that survive PPH expression have reduced activator expression (Fig. 1g–i, Supplementary Fig. 1b–d) suggests that high levels of activator expression are toxic, while lower expression levels might be tolerated. To uncouple antibiotic selection from transgene toxicity and allow for tunable activator expression, we constructed a doxycycline-inducible expression vector for MPH and established cell lines based on BC-3, BC-3/dCas9-VP64, A375, the T cell line Jurkat, and 293T by LV transduction at a single copy. We observed dose-dependent doxycycline-induced toxicity upon MPH induction in each cell line (Fig. 4, Supplementary Fig. 6), showing that toxicity can be uncoupled from LV transduction and suggesting that the activation domains of at least p65^AD and HSF1^AD are toxic across an expanded set of cellular contexts.

Fig. 4. Inducible MPH expression is toxic in various cell lines.

a Schematic of the Dox-inducible lentiviral vector pTO-Zeo-ΔWPRE, not to scale. Abbreviations are as in Fig. 1b, c, except for a constitutive transport element and polyA signal (CTE-pA), Dox-inducible promoter (TO), coding sequence for a Dox-binding transactivator protein (Tet-On^3G), the Simian Virus 40 promoter (SV40), and a Zeocin resistance cassette (Zeo^R). The number at left indicates genome size in kb. b Relative live cell numbers of untransduced (NT) BC-3, or BC-3 expressing Dox-inducible hrGFP2 or MPH, 50 h into treatment with the indicated concentrations of Dox. Three independent repeats. Significance was determined using One-Way ANOVA with Tukey’s multiple comparison test. c–e As in (b), but in A375, Jurkat, and 293T, respectively. Significance was determined using One-Way ANOVA with Tukey’s multiple comparison test. Throughout, error bars represent SEM, **** denotes adj. p < 0.0001, *** adj. p < 0.001, ** adj. p < 0.01, * adj. p < 0.05, and ns “not significant”, n = 3–4 as indicated. Source data and adj. p-values are provided in the Source Data file.

The SAM system is unlikely to allow efficient CRISPRa without measurable toxicity

Leveraging the inducible MPH expression vector, we tested whether there is a window of MPH expression that allows for CRISPRa without measurable toxicity or prior adaptation to MPH expression. Based on our result that toxicity was measurable 2 days after induction with 8 ng/ml Dox, we treated BC-3/dCas9-VP64/Tet-ON-MPH cells with concentrations between 2 and 8 ng/ml Dox. This resulted in a window of MPH expression with a ~45-fold range two days into induction (Fig. 5a, b, Supplementary Fig. 7a). The increased MPH expression was accompanied by increasing toxicity (Fig. 5c, Supplementary Fig. 7b). Based on these data, we additionally transduced the cells with vectors expressing sgAAVS1 or sgCRBN-a1 (see Supplementary Fig. 7c for a schematic). This approach resulted in a modest CRISPRa-mediated overexpression of CRBN upon induction of MPH, but not of hrGFP2, despite the expression of VP64 (Fig. 5d, e). The overexpression of CRBN reached significance at 5 ng/ml Dox, a concentration that killed 40% of the culture within 6 days of induction. This result suggests that the SAM system is unlikely to allow for robust CRISPRa without any measurable toxicity.

Fig. 5. The SAM system is unlikely to allow efficient CRISPRa without measurable toxicity.

a Western Blot analysis of MPH expression, using anti-HSF1, 2 days into Dox-induction. Molecular weight (kDa) markers are at left. Representative of n = 3, quantified in (b). b Quantification of results from (a) over n = 3 independent repeats; MPH expression was sequentially normalized to α-tubulin and its normalized intensity for 8 ng/ml Dox. Colors are as in (c). c Growth curve analysis of BC-3/Tet-ON-MPH after treatment with Dox at the indicated concentrations. Toxicity reached significance with 6 ng/ml Dox on day 2 and 5 or 3 ng/ml on days 4 and 6, respectively, determined using One-Way ANOVA with Tukey’s multiple comparison test. n = 4 independent repeats. d Western Blot analyses of CRBN and α-tubulin expression in representative lysates taken on day 2 after induction of BC-3/dCas9-VP64/Tet-ON-MPH or -hrGFP2 that were additionally transduced with sgAAVS1 or sgCRBN-a1. Molecular weight (kDa) markers are at left. Quantified over 4 replicates in (e). e Quantification of results shown in Fig. 4d over 4 independent repeats. Significance was determined using One-Way ANOVA with Tukey’s multiple comparison test. Throughout, error bars represent SEM, **** denotes adj. p < 0.0001, *** adj. p < 0.001, ** adj. p < 0.01, * adj. p < 0.05, and ns “not significant”. Source data and adj. p-values are provided in the Source Data file.

MPH causes cell cycle arrest, apoptosis, and reduced survival gene expression

We finally used Dox-inducible MPH expression in BC-3 cells to gain insights into the consequences of MPH expression. Within 24 h of induction, MPH-expressing cells accumulated in the G1 phase of the cell cycle, with concomitantly reduced percentages of cells in the S and G2 phases, consistent with a G1 cell cycle arrest (Fig. 6a, b, Supplementary Fig. 8a). This cell cycle arrest was accompanied by an increase in apoptotic cells, measured by Annexin V staining (Fig. 6c, d, Supplementary Fig. 8b). To define the underlying gene expression changes, we performed mRNA sequencing (mRNA-seq) 21 h into Dox treatment, after bead-based removal of apoptotic and dead cells to avoid artifacts (see Supplementary Fig. 9 for a principal component analysis and Supplementary Data 1 for analyzed data). These data revealed a striking downregulation of PEL hallmark dependency genes²¹, including cyclin D2 (CCND2), interferon regulatory factor 4 (IRF4), and MYC (Fig. 6e). IRF4 is the transcriptional master regulator that coordinates PEL-specific super-enhancer (SE)-linked gene expression^28–30. The downregulation of SE-linked genes, including IRF4 itself, cyclin D2, and MYC, likely explains the prominent G1 arrest after MPH induction. Gene set enrichment analysis (GSEA) confirmed the downregulation of MYC and its downstream targets (Fig. 6f, Supplementary Data 2). We furthermore observed downregulation of MCL1, which has a candidate PEL-SE. MCL1 is an anti-apoptotic Bcl2 family protein whose downregulation triggers rapid apoptosis in PEL cells²¹. Together, these findings suggest that MPH disrupts the super-enhancer-dependent transcriptional identity of PEL cells, leading to rapid cell cycle arrest and apoptosis. These findings are consistent with competition by MPH for critical transcriptional regulators at sites that are particularly sensitive to such disruption, including super-enhancers.

Fig. 6. Functional consequences of MPH expression.

a Representative FACS-based cell cycle analysis. Cell cycle phases are marked G1, S, and G2, and percentages of cells in each cell cycle phase are indicated by numbers. Quantified over 3 replicates in (b). b Quantification of results as in (a) over three independent experiments. Significance was determined using One-Way ANOVA with Tukey’s multiple comparison test. Error bars represent SEM, **** denotes adj. p < 0.0001, *** adj. p < 0.001, ** adj. p < 0.01, * adj. p < 0.05, and ns “not significant”. Source data and adj. p-values are provided in the Source Data file. c Representative FACS-based analysis of apoptotic cells, percentages of cells in each quadrant are indicated by numbers. Quantified over 3 replicates in (d). d Quantification of results as in (c) over three experiments. Significance was determined using One-Way ANOVA with Tukey’s multiple comparison test. Error bars represent SEM, **** denotes adj. p < 0.0001, *** adj. p < 0.001, ** adj. p < 0.01, * adj. p < 0.05, and ns “not significant”. Source data and adj. p-values are provided in the Source Data file. e Volcano Plot visualizing mRNA-seq data comparing MPH and hrGFP2-expressing cells 21 h after induction. f Visualization of the top ten pathways identified by gene set enrichment analyses of genes that were differentially expressed in comparisons between MPH and hrGFP2-expressing cells. NES: normalized enrichment score.

Discussion

Our data show that ectopic expression of transcriptional activator cassettes that are commonly used for CRISPRa is cytotoxic. For the SAM system, toxicity is pronounced even when published lentiviral vectors are delivered at a single copy into cell types that were previously used for CRISPRa screens, such as A375. Activator domain toxicity is likely responsible for the low titers of these LVs, since AD deletions or p300^Core mutations that rescued toxicity after transduction also rescued RNA-based LV titers. While titers for LVs may be confounded by additional phenomena, including LV genome size and unintended splicing and polyadenylation events, the idea that toxicity is responsible for low LV titers is supported by our finding that inducible MPH expression is toxic in 293T cells (Fig. 4e). p300^Core point mutants furthermore abolish toxicity and rescue titers of LVs with equal genome size, representing the largest activator-expressing LV genomes we have tested in Figs. 2 and 3.

While surviving MPH-transduced cells can be grown out, the observed vector toxicity represents a strong perturbation, which may affect CRISPRa screening results, particularly when MPH is delivered together with the sgRNA. Activator toxicity can additionally cause functional titrations to substantially underestimate LV titer (Fig. 1d, e), which could result in the unintentional delivery of several sgRNA per cell when the sgRNA is delivered together with MPH, thereby further confounding results. Although the literature reports the low titer of LVs used for CRISPRa^11,25, there are relatively few mentions of sgRNA-independent CRISPRa toxicity^{12,13,15,17,31–34}, perhaps because remaining surviving cells can often be grown out after the loss of most transduced cells. The most straightforward explanation for the ability to derive stable cell pools is that cells with a tolerable level of activator expression survive, while those with high expression undergo cell cycle arrest and/or cell death. In this case, surviving cells would be expected to resemble the parental population. Alternatively, the surviving cells may have undergone adaptations conferring resistance to activator expression, including selection for mutations or specific epigenetic features. Strategies including re-exposure to high MPH expression, CRISPR screening to probe for genetic dependencies, and exome sequencing could distinguish between these possibilities. We did not pursue MPH re-expression experiments with the Dox-inducible cell pools, since we have observed that a minor fraction of cells does not induce the hrGFP2 reporter cassette under our conditions. Thus, selecting surviving cells after MPH induction would likely enrich cells that cannot efficiently induce MPH expression. However, re-transduction of cells that survive initial constitutive MPH expression with another MPH-expressing LV would be feasible in future work. In flies, the ubiquitous expression of the SAM system is lethal³⁴ and fusion of either MCP or dCas9 to the catalytic domain of the HAT CBP was reported to result in male sterility and off-target lysine acetylation¹⁵, similar to Fig. 3d, e. The expression of either dCas9-VPR or dCas9-VP64 is detrimental for zebrafish development³³, and a recent report furthermore mentions an inability to establish several cell lines expressing dCas proteins fused to VPR¹². Similar sgRNA-independent off-target toxicity was also observed in cytidine and adenine base editors^35–37, due to random deaminase activity across the genome. This issue has prompted efforts on screening and engineering for deaminases with increased fidelity and reduced off-target activity.

While our experiments have not addressed the mechanisms of CRISPRa vector toxicity directly, ADs of viral or cellular TFs may compete with endogenous transcription factor complexes for cofactors that are limiting, in a process reminiscent of “cofactor squelching,” for example, by VP16^38–40. The observed MPH-induced downregulation of cell-type-defining essential genes that are under the control of super-enhancer elements in PEL cells, including IRF4, MYC, and cyclin D2, is consistent with this model of AD toxicity. The phenotypic consequences of MPH expression in PEL cells, i.e., cell cycle arrest and apoptosis, therefore likely reflect the biology of this cancer model, which depends on the transcription factors IRF4 and MYC to coordinate its gene expression profile, the high expression of cyclin D2 for G1/S transition, and the anti-apoptotic MCL1 for survival²¹. Future studies should directly test whether cofactor squelching contributes to AD-fusion protein toxicity, for example, by establishing which transcriptional regulators are bound by each AD-fusion protein and assessing whether this results in their redistribution from endogenous binding partners and changes in cellular transcription rates.

An alternative explanation for AD toxicity could be that AD-fusion proteins are recruited to and interfere with the function of endogenous TF complexes. Since the weak activator VP64 was well tolerated after transduction, it appears likely that activator strength correlates with toxicity, although there may be cell-type-specific differences, as proposed by others¹⁸ and suggested by the imperfect correlation between titers of 293T-derived vector stocks (Fig. 2a) and toxicity in A375 (Fig. 2b).

While CRISPRa by recruitment of enzymatic domains could potentially overcome toxicity due to cofactor competition by highly expressed ADs, LVs expressing fusion proteins of MCP with the p300 or CBP HAT core domains were also toxic, at least partially due to unintended acetylation events. Rescue from toxicity upon AD deletion or p300^Core-HAT mutation suggests toxicity is unlikely due to competition for the nuclear import machinery since all constructs contained two NLS motifs in each fusion protein. Further studies of the mechanisms underlying CRISPRa toxicity might identify strategies to overcome the limitations of current CRISPRa systems and inform our understanding of basic concepts of transcriptional regulation, including physiological competition for limiting cofactors. Our results also underscore the importance of characterizing the recruited factors for each AD and developing additional ADs and HAT variants for CRISPRa^12,13,41,42. In our experiments, the recently described eN3x9 quadruple AD-fusion protein¹³ and the p300^Core/I1417N¹⁷-mutant did not show toxicity, and their further characterization and development are therefore of great interest.

We speculate that difficulties implementing CRISPRa, including the commonly used SAM system, in the broader community may be limiting the wide adaptation of this technology. The development of CRISPRa technology should therefore include assessing the toxicity of CRISPRa vectors and testing strategies to limit this toxicity. Approaches to assess CRISPRa activator toxicity include those we have used here, i.e., measuring discrepancies between expected and observed survival following lentiviral transduction, relative viability at each passage after antibiotic selection is complete in comparison to untransduced cells, monitoring decreased viability in inducible systems, or directly measuring undesirable consequences on the cell cycle, apoptosis, or gene expression in the absence of an sgRNA.

Since transduced cells that grow out after passage had strongly reduced p65^AD-HSF1^AD expression, one strategy to overcome or manage CRISPRa toxicity could be to reduce the expression of AD-fusion proteins, for example by avoiding unnecessary codon optimization, using weaker or inducible promoters, or omitting sequences that boost gene expression from lentiviral vectors, such as the Woodchuck Hepatitis Virus posttranscriptional regulatory element (WPRE). Our results with inducible MPH vectors, however, suggest that it might be difficult to identify tolerated expression levels that allow for efficient CRISPRa in the absence of toxicity or selecting surviving cells, as we have done in Fig. 1. The transfection of relevant dCas9-ribonucleoprotein (RNP) complexes or lentiviral sgRNA delivery followed by transfection of dCas9-transactivator-encoding mRNAs represent alternatives to exclusively LV-mediated delivery that could offer opportunities for fine-tuning the expression of CRISPRa activators and can reportedly be well-tolerated^43,44. However, these approaches only enable transient CRISPRa activity and are incompatible with pooled screening. In a similar approach, the transfection of lower amounts of mRNAs encoding dCas9-VPR did not overcome the toxicity associated with this CRISPRa system³³.

In principle, inducible dCas9-triple or -quadruple activator fusions, including those with VPR, VP64-p65^AD-HSF1^AD, NFZ/NZF, MSN, or eN3x9, could result in more efficient complex assembly due to the requirement for only two complex components (dCas9-AD-fusion protein and sgRNA) compared to the assembly of three components in the SAM system. dCas9-fusion proteins may also be less well-expressed and therefore less toxic than smaller MCP-AD fusions. Aptamer-mediated recruitment, in contrast, offers the potential for multicopy recruitment of MCP-AD fusions at a lower expression level. Regardless of the approach, monitoring and controlling for AD toxicity after transduction is likely easier in CRISPRa systems where AD-expressing cell lines are established and validated first, followed by delivery of only the sgRNA during screening. Experimentally evolving the CRISPRa machinery for more efficient dCas9-AD-sgRNA-target complex assembly, as recently proposed for the sgRNA⁴⁵, or reducing toxicity, as recently reported for p300^Core/I1417N, represents a final strategy. For studies using CRISPRa on a single locus, even dCas9-VP64 alone, perhaps in conjunction with several guide RNAs, could give robust results.

Until CRISPRa systems with less pronounced toxicity have been developed for screening at genome scale, best practices for performing CRISPRa experiments and screens we recommend include (i) performing titrations of constitutive AD-expressing LVs by qRT-PCR relative to a non-toxic vector, like in Figs. 1 and 2, or by including a fluorescent marker that can be analyzed on day 1 after transduction, before the loss of transduced cells; (ii) establishing and validating activator-expressing cell lines before delivery of sgRNAs, while monitoring vector toxicity, and (iii) validating any screening results in unmodified cell lines using orthogonal methods, such as cDNA expression. It is finally worth considering that constitutive CRISPRa may not be necessary for all approaches. Applications for transient CRISPRa could include processes that are under positive feedback regulation, such as the herpesviral lytic switch, where initial activation initiates a unidirectional gene expression cascade or reprogramming event. While these approaches may help control for the toxicity of CRISPRa in a laboratory setting, it could be more difficult to control and overcome the toxicity of CRISPRa in model organisms and clinical applications.

In sum, while CRISPRa remains a conceptually appealing approach, our work reveals the importance of measuring activator domain toxicity of CRISPRa systems. Our results also underscore the importance of understanding mechanisms of AD action and CRISPRa off-target toxicity, further developing additional CRISPRa activators, and designing approaches that overcome toxicity.

Methods

Cell culture

293T/17 (“293T”) (ATCC, CRL-11268) were grown in Dulbecco’s Modified Eagle’s Medium (DMEM, Corning, 10-017-CV) containing 10% Serum Plus™ II Medium Supplement (Sigma-Aldrich 14009C-500ML, Batch Number: 21C421) and 10 μg/ml gentamicin (Gibco, 15710072). A375 (ATCC, CRL-1619) were grown in DMEM containing 10% fetal bovine serum (FBS, Corning, 35-010-CV) and 10 μg/ml gentamicin. BC-3 (ATCC, CRL-2277) were grown in RPMI 1640 (Corning, 10-040-CV), containing 20% FBS and 10 μg/ml gentamicin. Jurkat E6-1 (ATCC, TIB-152) were originally obtained from the University of North Carolina Tissue Culture facility and grown in RPMI 1640, containing 10% FBS and 10 μg/ml gentamicin.

Published constructs

The Human Calabrese CRISPR activation pooled library set A¹¹ was a gift from David Root and John Doench (Addgene #92379). pXPR_502¹¹ was a gift from John Doench & David Root (Addgene plasmid # 96923; http://n2t.net/addgene:96923; RRID:Addgene_96923). In pXPR_502, the sgRNA tetraloop is modified by two PP7 and two MS2 aptamers. pLX-sgRNA⁴⁶ was a gift from Eric Lander & David Sabatini (Addgene plasmid # 50662; http://n2t.net/addgene:50662; RRID:Addgene_50662). lentiGuide-Puro⁴⁷ was a gift from Feng Zhang (Addgene plasmid # 52963). pcDNA-dCas9-p300 Core¹⁶ was a gift from Charles Gersbach (Addgene plasmid # 61357; http://n2t.net/addgene:61357; RRID:Addgene_61357). MCP-MSN¹³ (Addgene plasmid #210704, http://n2t.net/addgene:210704; RRID:Addgene_210704), MCP-eN3x9¹³ (Addgene plasmid #210706, http://n2t.net/addgene:210706; RRID:Addgene_210706), and GCN5 FL-dCas9-CBPcore¹⁸ (Addgene plasmid # 179556; http://n2t.net/addgene:179556; RRID:Addgene_179556) were gifts from Isaac Hilton. Lenti dCAS-VP64_Blast⁴ was a gift from Feng Zhang (Addgene plasmid # 61425; http://n2t.net/addgene:61425; RRID:Addgene_61425). lenti MS2-P65-HSF1_Hygro⁴ was a gift from Feng Zhang (Addgene plasmid # 61426; http://n2t.net/addgene:61426; RRID:Addgene_61426). lentiMPH v2²⁵ was a gift from Feng Zhang (Addgene plasmid # 89308; http://n2t.net/addgene:89308; RRID:Addgene_89308). pMD2.G was a gift from Didier Trono (Addgene plasmid # 12259; http://n2t.net/addgene:12259; RRID:Addgene_12259). psPAX2 was a gift from Didier Trono (Addgene plasmid # 12260; http://n2t.net/addgene:12260; RRID:Addgene_12260). pLC-ZsGreen-P2A-Puro and pLC-ZsGreen-P2A-Hygro are available as Addgene plasmids #124302 and #124301²².

Vector construction

All primers and gBLOCKs were from IDT; for sequences, see Supplementary Data 3. All inserts and immediate vector context were confirmed by Sanger sequencing (ACGT), except when noted otherwise.

To clone pLC-dCas9-VP64-T2A-eGFP, dCas9-VP64 was amplified from lenti dCAS-VP64_Blast⁴ using primers 4396 and 4397, and EGFP was amplified from pLCE⁴⁸ using primers 4398 and 4395. Products were used for Gibson Assembly with the NheI-EcoRI vector fragment of pLCE.

To clone pXPR_502-sgAAVS1 and pXPR_502-sgCRBN-a1, pXPR_502 was cut using Esp3I (BsmBI, Thermo Fisher Scientific, ER0452) and subjected to T4 DNA ligation with annealed oligos 2692/2693 (AAVS1) or 4684/4685 (CRBN sg-a1, which was picked from the Calabrese library set A).

For Figs. 2 and 3, we initially constructed an “empty” lentiviral vector pLenti-2xMCS (pL2M) for flexible insertion of promoter-transgene cassettes upstream and downstream of a central polypurine tract (cPPT). For the vector backbone, we cut pLX-sgRNA⁴⁶ with NotI-HF and NheI-HF to remove a fragment beginning upstream of the Rev response element (RRE) and ending just upstream of the woodchuck hepatitis virus posttranscriptional regulatory element (WPRE). We then re-inserted PCR-amplified fragments containing the RRE (primers 5151 and 5152) and the cPPT (primers 5153 and 5154) using Gibson assembly, resulting in a vector with the following features: RSV/R-U5-Psi-RRE-(MluI-EcoRI)-cPPT-(XhoI-NheI-AgeI-SalI) -WPRE- SalI -SIN3′LTR. pL2M shares its backbone and LTR sequences with the commonly used sgRNA plasmids pLX-sgRNA⁴⁶, lentiGuide-Puro⁴⁷, and pXPR_502¹¹.

To insert a CMV promoter into the MCS 3′ to the cPPT, L2M was cut using XhoI and NheI-HF. The CMV promoter was amplified from pLCE⁴⁸ using primers 5157 and 5164. The resulting fragment was digested with XhoI and NheI-HF and ligated into the cut vector using T4 DNA ligase. We named the resulting vector pL2M-CMV. We next inserted a puromycin resistance gene under CMV control in pL2M-CMV to clone pL2M-CMV-Puro. First, pL2M-CMV was cut using NheI-HF and AgeI-HF. The puromycin resistance gene was amplified from lentiGuide-Puro⁴⁷ using primers 5163/5160, 5183/5184, and 5183/5160. Resulting PCR products were pooled, digested with NheI-HF and AgeI-HF, and ligated into the vector using T4 DNA ligase.

pL2M-CMV-MPH-P2A-Puro and deletion mutants: To insert the MPH-P2A-PuroR fusion proteins under CMV control, pL2M-CMV was cut using NheI-HF and AgeI-HF. A fragment containing a portion of the CMV promoter, and NLS-MCP-linker-(SV40-NLS) sequences was ordered as gBlock 5169 (IDT). We PCR-amplified a fragment containing codon-altered murine p65^AD and unaltered human HSF1^AD from an unpublished version of lenti MS2-P65-HSF1_Hygro⁴ that was modified for blasticidin resistance, using primers 5170/5171. We PCR-amplified a fragment containing P2A-puroR from pZIP-P2A-Puro²³ using primers 5172/5160. Fragments were joined by Gibson Assembly. In the context of the resulting vector, pL2M-CMV-MPH-P2A-Puro, we deleted the MCP coat protein using primers 5225/5226, the p65^AD using primers 5224/5221, the HSF1^AD using primers 5220/5249, and p65^AD-HSF1^AD using primers 5220/5221 and the Q5® Site-Directed Mutagenesis Kit (NEB, #E0552S). Resulting mutants were confirmed by full plasmid sequencing (Plasmidsaurus).

To clone pL2M-CMV-MCP-VP64-P2A-Puro, pL2M-CMV-MPH-P2A-Puro was cut using NheI-HF and BamHI-HF. A fragment containing VP64 was PCR amplified from pLC-dCas9-VP64-T2A-eGFP using primers 5370/5371. The vector, the PCR-amplified fragment, and gBlock 5169 were joined by Gibson Assembly. To clone pL2M-CMV-MCP-NFZ/NZF-P2A-Puro, we cut pL2M-CMV-MPH-P2A-Puro using NheI-HF and BamHI-HF and used the resulting vector for Gibson Assembly with synthesized fragments 5533 (NFZ) or 5534 (NZF) (Twist Bioscience). To clone pL2M-CMV-MCP-MSN-P2A-Puro, pL2M-CMV-MPH-P2A-Puro was cut using NheI-HF and BamHI-HF, and the resulting vector backbone was used for Gibson Assembly with gBlock 5169 (see above) and a fragment containing MSN that was PCR-amplified from MCP-MSN (Addgene # 210704) using primers 6047/6048. To clone pL2M-CMV-MCP-eN3x9-P2A-Puro, pL2M-CMV-MPH-P2A-Puro was cut using NheI-HF and BamHI-HF, and the resulting vector backbone was used for Gibson Assembly with gBlock 5169 (see above) and a fragment containing eN3x9 that was PCR-amplified from MCP-eN3x9 (Addgene # 210706) using primers 6049/6050. To clone pL2M-CMV-MCP-VPR-P2A-Puro (for an annotated sequence see Supplementary Fig. 10), pL2M-CMV-MPH-P2A-Puro was cut using NheI-HF and BamHI-HF, and the resulting vector backbone was used for Gibson Assembly with a fragment containing MCP-VPR that was PCR-amplified from an unpublished vector using primers 6053/6055. We note that this construct does not include an additional NLS between VP64 and p65^AD, as is typical for VPR, to enable comparisons with other activators here. We furthermore used a minimal RTA activation domain (RTA 125-190aa), as reported elsewhere⁴⁹.

To clone pL2M-CMV-MCP-p300^Core-HA-P2A-Puro, pL2M-CMV-MPH-P2A-Puro was cut using NheI-HF and BamHI-HF, and the resulting vector backbone was used for Gibson Assembly with gBlock 5169 (see above) and a fragment containing p300^Core-HA tag that was PCR amplified from pcDNA-dCas9-p300 (Addgene plasmid # 61357) using primers 5372/5373. We used the Q5® Site-Directed Mutagenesis Kit to introduce the D1399Y (primers 5535/6) and I1417N (primers 6051/2) mutations. The resulting mutants were confirmed by full plasmid sequencing (Plasmidsaurus). To clone pL2M-CMV-MCP-CBPcore-P2A-Puro, pL2M-CMV-MPH-P2A-Puro was cut using NheI-HF and BamHI-HF, and the resulting vector backbone was used for Gibson Assembly with a fragment containing MCP-CBP-core that was PCR-amplified from an unpublished vector using primers 6053/6054. The CBPcore domain was originally cloned from GCN5 FL-dCas9-CBPcore (Addgene plasmid # 179556)

For the sgRNA vectors used in Fig. 5, we excised the pXPR_502 PPH-2A-Puro cassette with BamHI and MluI. We used Gibson assembly to insert a puromycin resistance cassette we PCR amplified using primers 5554/5555 and pL2M-CMV-Puro as a template. The resulting vectors are pXPR-Puro-sgAAVS and pXPR-Puro-sgCRBN-a1 (see Supplementary Fig. 7c for a schematic).

To clone the Tet-ON LV pTO-Zeo-ΔWPRE, we cut pLVX-TetOne-Zeo⁵⁰ with MluI and NheI and performed Gibson Assembly with a PCR product (primers 5530/2) containing the self-inactivating LTR sequence amplified from pL2M. We next inserted MPH or hrGFP2 cassettes between the EcoRI/AgeI sites of this vector to clone pTO-Zeo-ΔWPRE-MPH or -hrGFP2.

Production of lentiviral vectors

To produce lentiviral vectors, we co-transfected each transfer vector with pMD2.G and psPAX2 into 293 T/17 cells using a 0.624 mg/ml PEI MAX (Polysciences, Catalog # 24765) stock solution (pH 7.4, adjusted using NaOH) and 0.4 pmol DNA/6 well, 3 pmol DNA/10 cm dish, or 7 pmol DNA/15 cm dish, at a ratio of 3.5 μl PEI MAX per 1 μg DNA. A molar ratio of 45% transfer vector, 35% psPax2, and 20% MD2.G was used, except for Fig. 1 and Supplementary Fig. 1, where all toxic vectors were packaged using 15% transfer vector, ~54% psPax2, ~31% pMD2.G, since we found that reducing the amount of transfer vector increases pXPR_502-based lentivirus titers (compare RNA-based titers in Fig. 1d, where 15% transfer vector was used, to those in Fig.2a, where 45% transfer vector was used, p = 0.04 from a 2-tailed unpaired t test, ~1.8× improved titer with 15% transfer vector). This observation is likely explained by reduced vector toxicity in LV-producing cells when a lower amount is transfected. Culture media were changed ~4 h after transfection. Approximately 72 h after transfection, we filtered supernatants through 450 nm pore size filters and froze aliquots at −80 °C. For Fig. 1, lentivirus was first concentrated by ultracentrifugation (Beckman SW 32 Ti rotor, 80,000 × g, 1 h, 4 °C), pellets were incubated with Opti-MEM (Gibco, 31985070) on a platform shaker for at least one hour at 4 °C, resuspended by pipetting up and down 20–25 times, and then frozen in aliquots. Reported lentiviral RNA genome sizes were measured starting from the HIV transcription start site to the end of the polyA signal, did not consider the cleavage site or the polyA tail, and were rounded up to the next 0.1 kb.

Lentiviral titration

For lentivirus titration by qRT-PCR, we used the LentiX qRT-PCR kit (Takara) according to the manufacturer’s instructions. For functional titration by flow cytometry (FACS) or cell counting, serial dilutions of lentiviral stocks were used to transduce target cells in the presence of 5 μg/ml polybrene. For A375, we plated 15,500 cells/cm² the afternoon before transduction. For BC-3 or BC-3 dCas9-VP64, we split cultures to 3–5 × 105 cells/ml the day before transduction, to ensure robustly proliferating cultures. The next day, BC-3 or BC-3 dCas9-VP64 were adjusted to 3 × 105 cells/ml and ~0.208 ml/cm². For GFP-based titration, FACS was performed on a BD FACS Canto II, two days after titration with (A375) or without (BC-3) changing media the day after transduction. For functional titration in A375, we changed the culture medium ~24 h after transduction to medium containing 1 μg/ml puromycin, maintaining unselected and selected untransduced controls. For functional titration in BC-3, we added 1 μg/ml puromycin without changing the medium, maintaining unselected and selected untransduced controls. 24–30 h later, when no viable cells remained in the selected untransduced control well, we counted the surviving cells using trypan blue exclusion assay and flow cytometry (ZsGreen controls) and calculated the percentage of live or GFP-positive relative to the untransduced and unselected control. Functional titers were calculated from 4 to 20% of surviving or GFP-positive cells, assuming a single transduction event per cell.

Lentiviral transductions and CellTiter-Glo 2.0 assays

For all transductions, cell numbers and media volumes were scaled approximately by surface area. LVs were added at the indicated MOIs. 24 h after transduction, A375 were split 1:2 and at the same time selected using 1 μg/ml puromycin. BC-3 or BC-3/dCas9-VP64 were collected by low-speed centrifugation and resuspended in new medium containing 1 μg/ml puromycin or 300 μg/ml hygromycin. CellTiter-Glo 2.0 (Promega) was used as instructed at the time points indicated in the manuscript, upon completion of selection, determined using a selected untransduced control sample. For A-485 (MCE, HY-107455) treatment, drug or DMSO was added at the time of transduction in 24 format. 24 h after transduction, cells were split as above, and the drug was re-added. 2 days into selection and drug treatment, cells were processed for CellTiter-Glo 2.0 measurements. Results for p300^Core-transduced cells are shown relative to the Puro control in Fig. 3f. Extended and unnormalized results, including those for untransduced cells and untreated (no DMSO or A-485) cells, can be found in Supplementary Fig. 5c. We note that untransduced/unselected controls in Supplementary Fig. 5c were split 1:8 to account for ~4x greater cell numbers compared to MOI 0.25-transduced, selected cells. For growth curve analyses involving Dox treatment, we spun the cells at every passage and replaced the growth medium before redosing with Dox.

Establishment of BC-3-dCas9-VP64

pLC-dCas9-VP64-T2A-eGFP LV was produced as described above and used to transduce BC-3 cells at ~MOI 0.6, resulting in ~45% GFP-positive cells. We sorted the top 20% GFP expressors using a FACSAria system, obtaining 4.4 × 10⁵ live cells that were grown out into the cell pool that was used here (BC-3/dCas9-VP64).

Growth curve analyses in BC-3

BC-3 or BC-3/dCas9-VP64 were transduced and selected as outlined above. The unselected and untransduced control cells were typically counted and split the next day, while other samples were first analyzed and passaged 3 (puromycin) or 4 (hygromycin) days after transduction, when selection was complete. After the first passage after selection, all samples were centrifuged, and cells were resuspended in medium without puromycin or hygromycin. From this time point onwards, untransduced and unselected control cells were split together with transduced cell pools. Growth curve analysis was done using CellTiter-Glo 2.0, and the resulting values were normalized to cell counts obtained by trypan blue exclusion assay and manual counting of control samples at each passage. At each passage, all samples were adjusted to 3 × 10⁵ cells/ml by either diluting or concentrating samples. For cumulative growth curve analyses in Fig. 1f and Supplementary Fig. 1a, cell counts at each passage were multiplied by all previous dilution factors, prior to normalization to numbers from the control cell pool.

Western blots

Cells were washed with cold PBS and lysed with RIPA (50 mM Tris [pH 8.0], 150 mM NaCl 1% IGEPAL CA-630, 0.5% sodium deoxycholate, 0.1% SDS) containing Protease Inhibitor Cocktail Set III, EDTA-Free (Calbiochem, catalog number 539134)²¹. For A375, cells were washed with cold PBS and scraped into RIPA buffer without prior detachment, 48 h after transduction. For BC-3, cells were collected by low-speed centrifugation. For day 3 BC-3 lysates, cultures were subjected to a dead-cell removal step using the Miltenyi Biotec Dead Cell Removal Kit (Order no. 130-090-101), MS columns (Miltenyi Biotec), and an OctoMACS separator (Miltenyi Biotec), all as instructed, since cultures contained large numbers of dead cells right after low MOI transduction and selection. Day 12 lysates did not contain a substantial number of dead cells and were collected directly. After 15 min lysis on ice, lysates were sonicated for 6 cycles (30 s on, 30–40 s off), cleared by centrifugation, and quantified using the BCA assay (Pierce™ BCA Protein Assay Kit, ThermoScientific, catalog number 23225). Equal amounts of total protein and PageRuler™ Prestained Protein Ladder (ThermoScientific, catalog number 26617) were separated on Bolt™ Bis-Tris Plus Mini Protein Gels (4-12%, Invitrogen), using 1x MOPS running buffer (50 mM MOPS. 50 mM Tris, 0.1% SDS, 1 mM EDTA [pH 7.7]) and transferred to nitrocellulose membranes. Membranes were blocked for one hour or overnight in TBS containing 5% non-fat milk powder. Primary antibodies were used at 1:1000 dilutions in TBS containing 0.1% Tween (TBST) and 5% non-fat milk powder. Membranes were washed 3 × 15 min in TBST, incubated with IRDye-800-conjugated secondary antibodies (LI-COR) at 1:10,000 in TBST containing 5% non-fat milk powder. The following primary antibodies were used: rabbit anti-HSF1 (Cell Signaling Technology 4356, lot 2), rabbit anti acetylated-Lysine (Cell Signaling Technology 9441, lot 16), rabbit anti-Enterobacteriophage MS2 Coat Protein (Millipore-Sigma, ABE76-I, lot 3764751), rabbit anti-CRBN (Sigma-Aldrich, HPA045910, lot BE106946, Fig. 1i), rabbit anti-CRBN (clone F4I7F, Cell Signaling Technology 60312, lot 1, Fig. 4d), mouse anti-α-tubulin (Cell Signaling Technology 3873, lot 19), rabbit anti Histone 3 (Cell Signaling Technology 4499, lot 20), and rabbit anti-H3K27ac (Cell Signaling Technology 8173, lot 8). Western Blots were imaged on LI-COR Odyssey FC or LI-COR M Imagers and quantified using Image Studio version 5.2.5 or Empiria Studio 2.

Cell cycle and apoptosis assays

Parental BC-3 cells, transduced with pTO-Zeo-ΔWPRE-MPH or -hrGFP2, were seeded at ~4 × 105 cells per ml in 6-well plates and cultured for 24 h ± doxycycline (Dox, 100 ng/ml; Sigma-Aldrich). The cell cycle distribution was quantified by 5-ethynyl-2´-deoxyuridine (EdU) pulse-labeling using the Click-iT™ Plus EdU Alexa Fluor™ 647 Flow Cytometry Assay (Invitrogen, C10634). EdU (10 µM, 1 h) was added, cells were pelleted (360 × g, 5 min, RT) and washed with 1% BSA in PBS. Fixation and permeabilization steps were performed for 15 min, while the click chemistry reaction was performed for 30 min at room temperature as recommended by the manufacturer, followed by total-DNA staining with FxCycle™ Violet (1 µl per 500 µl; Invitrogen, F10347) for 30 min at room temperature in the dark. Apoptosis and viability were assessed in parallel wells by dual Annexin V-Pacific Blue/7-AAD staining. Briefly, ~10⁶ cells were washed twice in ice-cold PBS, resuspended in 100 µl Annexin V Binding Buffer (10 mM HEPES, 140 mM NaCl, 2.5 mM CaCl₂, pH 7.4; Thermo Fisher, V13246) and incubated for 15 min at room temperature in the dark with 5 µl Annexin V-Pacific Blue (Thermo Fisher, A35122) and 1 µl 7-AAD (1 mg/mL stock in DMSO; Thermo Fisher, A1310). Samples were diluted with 400 µl Annexin V Binding Buffer, kept on ice, and acquired within 1 h. Results for both assays were acquired on a BD FACSymphony™ A1 cytometer and analyzed using FlowJo v10.10 (BD).

mRNA-seq

Parental BC-3 cells (not expressing dCas9-VP64), transduced with pTO-Zeo-ΔWPRE-MPH or -hrGFP2, were seeded at ~4 × 105 cells per ml, and treated with doxycycline (Dox, 100 ng/ml; Sigma-Aldrich) or not, in technical triplicates. 21 h later, we performed dead-cell removal as described above, cells were pelleted, and cell pellets were lysed in TRIzol (Invitrogen, 15596026). Total RNA was prepared as instructed and shipped to Genewiz (Azenta Life Sciences), who performed additional quality control, polyA selection, and stranded paired-end Illumina sequencing. Specifically, total RNA samples were quantified using Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) and RNA integrity was checked with a 4200 TapeStation (Agilent Technologies, Palo Alto, CA, USA). Strand-specific RNA sequencing libraries were prepared by using the NEBNext Ultra II Directional RNA Library Prep Kit for Illumina following the manufacturer’s instructions (NEB, Ipswich, MA, USA). Briefly, the enriched RNAs were fragmented for 8 min at 94 °C. First-strand and second-strand cDNA were subsequently synthesized. The second strand of the cDNAs was marked by incorporating dUTP during the synthesis. cDNA fragments were adenylated at 3′ends, and an indexed adapter was ligated to the cDNA fragments. Limited-cycle PCR was used for library enrichment. The incorporated dUTP in the second-strand cDNA quenched the amplification of the second strand, which helped to preserve the strand specificity. The sequencing library was validated on the Agilent TapeStation (Agilent Technologies, Palo Alto, CA, USA) and quantified by using Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Waltham, MA, USA) as well as by quantitative PCR (KAPA Biosystems, Wilmington, MA, USA). The sequencing libraries were clustered on the flowcell. After clustering, the flowcell was loaded on the Illumina instrument according to the manufacturer’s instructions. The samples were sequenced using a 2 × 150 bp Paired End (PE) configuration. Image analysis and base calling were conducted by the Control software. Raw sequence data (.bcl files) generated by the sequencer were converted into fastq files and de-multiplexed using Illumina’s bcl2fastq 2.20 software. One mismatch was allowed for index sequence identification.

Resulting data were processed and analyzed using the nf-core/rnaseq pipeline (v3.18.0) and GRCh38.d1.vd1, downloaded from https://gdc.cancer.gov/about-data/gdc-data-processing/gdc-reference-files. Gene annotations were derived from Gencode v36, but we added limited KSHV gene annotations from Genbank to an accompanying gene annotation gtf file and removed rRNA genes. Normalized read counts were imported into RStudio 2023.12.1 (R v4.3) and volcano plots were generated with ggplot2 v3.5.2 and ggrepel v0.9.6, coloring genes with an absolute log2 fold-change greater than 1 and Benjamini–Hochberg false-discovery-rate (FDR-adjusted p < 0.05) as up- or downregulated genes in red or blue, respectively. For pathway analysis, protein-coding genes were ranked by log₂(fold-change) and tested against the MSigDB Hallmark collection (release 2023.1) using fgsea v1.20.0 with gene sets retrieved via msigdbr v24.1.0. Pathways with FDR-adjusted p < 0.05 were considered enriched; the ten most enriched pathways were visualized as dot plots in ggplot2 and were all related to MPH-downregulated genes.

Statistics and reproducibility

Graphs were plotted and statistical analyses were done in GraphPad Prism 10. We used One-Way ANOVA with Tukey’s multiple comparison tests and considered adjusted p < 0.05 as significant, unless indicated otherwise for specific analyses. All error bars represent the standard error of the mean (SEM). Throughout, **** denotes adj. p < 0.0001, *** adj. p < 0.001, ** adj. p < 0.01, * adj. p < 0.05, and ns “not significant”. Numbers of independent repeats are indicated for each experiment, in each panel or legend, and complete results are provided in the Source Data file. In some cases, protein lysates from independent experiments were probed together on the same membrane (see Source Data file). No statistical method was used to predetermine sample size. No data points were excluded from the analyses. The experiments were not randomized, and the investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

A.M., S.J., and E.G. conceptualized and planned the study. Z.L., A.M., J.A.O., C.B.M., S.J., J.T., A.P., M.M., E.T.B., and E.G. cloned vectors and developed the methodology. Z.L., A.M., J.A.O., C.B.M., E.T.B., and E.G. performed the experiments and analyses shown. Unpublished experiments performed by S.J., A.P., M.M., S.R., and E.G. informed the design of this study. Z.L., A.M., J.A.O., C.B.M., and E.G. prepared figures. E.G. supervised lab members and wrote the paper. All authors provided feedback on the manuscript.

Peer review

Peer review information

Nature Communications thanks Samuele Ferrari and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data are available in the main manuscript or the supplementary materials. mRNA-seq data generated in this study have been deposited in the GEO database under accession code GSE299931. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-63570-4.