Universal strategy to generate lentiviral vectors encoding lethal proteins

Abstract

Lentiviral vectors have emerged as a versatile and robust platform for gene therapy. However, their use for delivering toxic genes has been hindered by challenges related to packaging lentiviral genomes that encode inherently lethal transgenes. In this study, we introduce a novel multi-layered approach to overcome this obstacle. Our results show that combining a regulatable promoter, an inverted orientation of the lethal transgene expression cassette, and CRISPR-based inhibition to repress vector transgene transcription effectively reduces toxicity in producer cells, enabling the efficient production of high-titer lentiviral vectors expressing lethal proteins. Notably, these lentiviral vectors were able to deliver transgenes encoding pro-apoptotic proteins or toxins that induced cell death in cancer cells, providing a translational proof of concept for therapeutic payload delivery. Overall, this work establishes an innovative framework that expands the utility of the lentiviral platform for the controlled delivery of cytotoxic factors in gene therapy applications.

Graphical abstract

An integrated strategy, combining inducible control, inverted toxic-gene orientation, and CRISPR-mediated transcriptional repression, enables high-titer production of lentiviral vectors carrying lethal transgenes. This platform successfully delivers pro-apoptotic or toxic genes to cancer cells, restoring selective cell killing and expanding lentiviral gene therapy applications.

Introduction

Gene therapy has advanced rapidly and holds promise for treating cancer and various genetic diseases. Various viral vector systems are used to deliver therapeutic genetic material to target cells, with lentiviral vectors (LVs) standing out as one of the most exploited platform.¹

Suicide gene therapy relies on the expression of cytotoxic proteins, which can be classified as either extrinsically cytotoxic, requiring secondary activation, or intrinsically cytotoxic. The former group has been more extensively studied, likely due to its more manageable application and the challenges associated with producing recombinant viral particles that express inherently toxic factors.²

While extrinsically activated cytotoxic systems such as Herpes simplex virus type 1 thymidin kinase gene (TK), the bacterial nitroreductase gene, or the inducible human caspase 9 (iC9) have been widely developed due to their tolerability in producer cells,^3,4,5 the expression of intrinsically cytotoxic proteins within LVs can severely hinder the assembly of recombinant particles. Lentiviral genome expression can impair vector assembly, requiring specialized packaging cells or genetic modifications to counteract the lethal effects of the transgene.^6,7,8 Despite progress, the field lacks universal strategies to produce LVs carrying lethal payloads independent of their mode of action.⁹

Since stress-responsive promoters have been used to control the expression of extrinsically cytotoxic transgenes,¹⁰ we considered employing a similar approach to produce LVs capable of delivering inherently lethal transgenes. A physiologically stress-responsive promoter, consisting of six copies of the TRIB3 amino acid response elements (AAREs) linked to a minimal promoter and known as 2xAARE, allows regulatable gene expression through the activating transcription factor-4 (ATF4).¹¹ In mammalian cells, the ATF4 pathway is activated by the integrated stress response (ISR), triggered by endoplasmic reticulum (ER) stress, oxidative stress, double-strand RNA, or shortages of nutrients (such as amino acids) or oxygen. This activation occurs via the phosphorylation of PERK, GCN2, PKR, or HRI, leading to the subsequent phosphorylation of eIF2α, which blocks the translation of capped mRNAs but favors the translation of certain sequences containing UTRs, such as ATF4 and other genes involved in stress regulation. Once expressed, ATF4 regulates the expression of adaptive genes by binding to AAREs in their promoters, enabling cells to cope with stress.^12,13 Based on these principles, we applied the ATF4-dependent 2xAARE system to produce LVs that encode lethal proteins.

We engineered the 2xAARE system with an optimized TATA box¹⁴ and placed various toxic transgenes under its control. During vector packaging in HEK293T cells, the 2xAARE promoter did not restrict expression of toxic transgenes, resulting in reduced vector particle titers. Therefore, we reversed the orientation of the expression cassette within the lentiviral provirus and employed CRISPR-mediated inhibition of transgene transcription. This approach enabled high-titer manufacturing of LVs encoding lethal genes that, upon ER stress induction, efficiently triggered apoptosis in cancer cells, thereby establishing a platform for the controlled delivery of inherently cytotoxic transgenes.

Results

Regulatable toxic gene cassette yields LVs at low titers

To achieve efficient packaging of LVs encoding inherently lethal transgenes, we constructed lentiviral vector plasmids (pLVs) carrying a stress-responsive expression cassette dependent on ATF4 expression.¹¹ This cassette was driven by the 2xAARE promoter, which controlled the expression of the GFP alone or in combination with toxic transgenes (TTgs) via a P2A peptide (Figure 1A). In this system, each transgene was designed to be expressed only after activation of the integrated stress response (ISR) and subsequent translation of ATF4 in transfected or transduced cells. Five TTgs were included, encoding: (1) the human proapoptotic member of the BCL-2 family BAX, with S184 deletion, which induces cell death independently from the apoptosis signaling cascade¹⁵; (2) the constitutively lethal V266E-mutated caspase-3 (CASP3)¹⁶; (3) the mutant H37A influenza A virus-derived ion channel M2, which induces cell death upon membrane translocation¹⁷; the potent cytotoxic diphtheria toxin fragment A form Corinebacterium diphteriae (DTA), which is the catalytic domain of the toxin¹⁸; and its 10- to 100-fold attenuated H21A variant (aDTA).¹⁹ Additionally, inducible caspase-9 (iC9) was used as a control cytotoxic transgene, which requires a dimerizing prodrug for activation.²⁰ Using these constructs, recombinant LV particles were produced in HEK293T cells.

Figure 1.

LVs carrying 2xAARE-controlled toxic transgene are produced at low titers in HEK293T cells

(A) Schemes of lentiviral vector genomes (LV), control LV-2xAARE-GFP (upper) and suicide LV-2xAARE-GFP-P2A-TTg (lower, TTg: Toxic Transgene), carried by the plasmid pLV used for lentiviral particles production. Lentiviral genome in pLV harbors HIV-1 cis-regulating sequences: the 5′ LTR, the Ψ (psi) packaging signal, the rev responsive element (RRE) for RNA vector genome nuclear export, the central polypurine tract (cPPT) and central termination sequence (CTS), and a 3′ self-inactivated (SIN) LTR with truncated U3 sequence. (B) Comparison of titers of recombinant LV particles expressed in transducing units (TU) per mL of supernatant from HEK293T cells transiently transfected with helper plasmids and pLVs conveying the control transgene GFP or the GFP-P2A-TTg, under the control of the 2xAARE promoter. Results are mean values ±SD from 5 independent productions; statistics, one-way ANOVA, and Dunnett’s multiple comparisons test; ∗∗∗∗p < 0.0001; nd, not detected. (C) Viability assay performed 24 h after transient transfection of HEK293T cells with control GFP or GFP-P2A-TTg-carrying pLVs; results are presented as mean percentage ±SD from 6 independent experiments as compared to non-transfected HEK293T cells; statistics, one-way ANOVA, and Dunnett’s multiple comparisons test; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001. (D) Cytometry quantification of GFP expression in HEK293T cells 24 h after pLV transfection with pLV-2xAARE-GFP and the six pLV-2xAARE-GFP-P2A-TTg; histograms show the detection and mean fluorescence intensity of GFP expression (values); a progressive reduction of GFP expression is observed for pLV-2xAARE-GFP-, iC9, CASP3, BAX, and M2 and its complete abolition with aDTA and DTA.

To generate LV recombinant particles carrying these transgenes, HEK293T cells were transfected with the trans-complementing plasmids p8.9 and pVSV, along with pLVs carrying the 2xAARE-GFP-BAX, -iC9, -CASP3, -M2, -aDTA, or -DTA internal cassettes, or with control pLV-2xAARE-GFP. All pLV-2xAARE-TTg constructs produced significantly lower yields of recombinant particles compared to the control pLV-2xAARE-GFP. Particle production was reduced by 9-fold for BAX, 40-fold for iC9, and 220- and 250-fold for CASP3 and M2, respectively, while the pLV-2xAARE-GFP-aDTA or -DTA constructs failed to produce detectable recombinant particles (Figure 1B). Transfected HEK293T cells exhibited decreased viability, correlating with the reduction in vector titers for LV-BAX, -iC9, -CASP3, and -M2 (Figure 1C). However, transfection with pLVs carrying aDTA or DTA, despite yielding no detectable particles, only mildly affected cell viability. This is likely due to the mode of action of DTA, which inhibits mRNA translation in HEK293T cells, as evidenced by the absence of GFP detection 24 h after transfection with pLV carrying aDTA or DTA (Figure 1D). Thus, the expression of all six TTgs exerts varying degrees of toxicity on packaging cells during LV production, leading to reduced yields of LV. Interestingly, this effect was also observed for iC9 in the absence of AP1903, indicating that iC9 is inherently toxic to HEK293T cells even without dimerization.

To determine whether ATF4 is responsible for 2xAARE promoter transactivation and the leakage of TTg expression from transfected pLV-2xAARE-GFP-TTg during packaging, we generated ATF4-knockout HEK293T cells (HEK^ATF4−/−) (Figure S1A). In the absence of trans-complementation of the 5′ long terminal repeat (LTR) with HIV-TransActivator of Transcription (Tat), transfection of the plasmid pLV-2xAARE-GFP into both HEK293T and HEK^ATF4−/− cells induced similar levels of GFP expression (Figure S1B). This indicates that ATF4 is not required for GFP expression from a transfected pLV-2xAARE-GFP plasmid in HEK293T cells.

In contrast, transduction of HEK293T and HEK^ATF4−/− cells with LV-2xAARE-GFP revealed that only wild-type cells, and not ATF4-deficient cells, expressed GFP under amino acid (leucine) starvation, which induces ATF4 expression (Figure S1C). This establishes that the transactivation of the 2xAARE promoter is strictly ATF4 dependent when the vector is integrated into the cell genome. However, an ATF4-independent mechanism allows transgene expression after pLV transfection in HEK293T and HEK^ATF4−/− cells. Collectively, these results suggest that TTg expression in packaging cells during particle production is not linked to the concurrent ATF4 expression. They also imply that the internal 2xAARE promoter does not restrict transgene expression during LV packaging, pointing to the presence of an alternative promoter within the pLV that facilitates constitutive expression of the transgene in both wild-type and ATF4-knockout cells following plasmid transfection.

Inverting the internal cassette halts 5′ LTR overriding activity

We then tested the aforementioned hypothesis to decipher the mechanisms underlying the leakiness and toxicity of the 2xAARE-driven TTgs during LV production. The closest cis-regulatory sequence upstream of the 2xAARE promoter is the U3 region in the HIV 5′ LTR, which is crucial for lentiviral genome transcription and encapsidation and is enhanced by the HIV transactivator Tat (Figure 2A). Therefore, we assessed whether the Tat-mediated transactivation of the HIV 5′ LTR would upregulate GFP expression from transfected pLV-2xAARE-GFP in HEK293T cells. The presence of Tat increased both the percentage of GFP-expressing cells and the fluorescence intensity per cell (Figure 2B). Transfecting a promoterless pLV-GFP in HEK293T cells in the presence of Tat also resulted in a 3-fold increase in GFP expression compared to that measured with pLV-2xAARE-GFP (Figures S2A and S2B). Although indirect, these results suggest that, in the absence of an internal promoter, and despite the presence of the stress-responsive 2xAARE promoter within the lentiviral genome, transgene expression in a transfected pLV is primarily driven by the 5′ LTR.

Figure 2.

Inversion of the internal LV expression cassette prevents its transcription after transfection

(A) Scheme representing a plasmid encoding the LV genome (pLV as represented in Figure 1A, carrying an internal cassette of expression with 2xAARE promoter and a transgene. Arrows indicate putative transcription start sites in the pLV, in the 5′ LTR, in the SV40 promoter, a cryptic TATA box in the pMB1 origin of replication (Ori), in the TATA box of the 2xAARE promoter, and possibly in the 1,451 bp sequence of the human TBC1D5 exon 8 and flanking intron cloned downstream of the 3′ LTR. (B) Effect of the presence (+) or absence (−) of Tat on the transcription of GFP by a pLV-2xAARE-GFP plasmid transfected in HEK293T cells. Left: the percentage of GFP-expressing cells (% GFP-positive), right: the mean fluorescence intensity of GFP per cell. Results are mean values ±SD from 3 independent experiments, statistics, unpaired t test, ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001. (C) Schemes of lentiviral vector (LV) genomes expressing the luciferase transgene (Luc), with an expression cassette in the same orientation of transcription as the vector genome, LV-2xAARE-Luc (upper), or in an inverted orientation, LVi-2xAARE-Luc (lower). (D) Luciferase expression in HEK293T cells transfected with pLV-2xAARE-Luc or pLVi-2xAARE-Luc in the presence of Tat. Results are expressed as mean values ±SD from 3 independent experiments, statistics, unpaired t test, ∗∗∗∗p < 0.0001.

Next, we designed pLVi constructs with the transgene cassette cloned in the reverse orientation relative to the 5′ LTR. Using luciferase as a reporter, which allows a wider dynamic range for transgene measurement, we compared the performance of the forward (pLV)- and inverted (pLVi)-oriented cassettes (Figure 2C). Transfection of the pLVi-2xAARE-Luc into HEK293T cells in the presence of Tat reduced luciferase expression by 6,237-fold compared to pLV-2xAARE-Luc (Figure 2D). This strongly suggests that the 5′ LTR overdrives internal transgene expression during pLV transfection, which can be substantially reduced by inverting the orientation of the internal cassette relative to the 5′ LTR.

Expression cassette orientation is critical for transgene expression and vector titers

The aforementioned findings prompted an investigation into whether inverting the transgene expression cassette could enhance the titers of LVi carrying TTg. Effectively, the pLVi-2xAARE-GFP-P2A-TTg constructs achieved elevated vector titers for LVi-2xAARE-BAX, -iC9, -CASP3, -M2, and notably -aDTA but not DTA, which remained undetectable, likely due to its much higher toxic propensity compared with aDTA (Figure 3A). Titers for LVi carrying BAX, iC9, M2, and CASP3 increased by approximately 2-, 23-, 29-, and 88-fold, respectively, compared to LV (Figure 3B). Cell viability remained close to control levels in HEK293T cells transfected with all pLVi-2xAARE-TTg constructs except DTA (Figure 3C). Thus, inverting the regulatable toxic gene cassette consistently reduced toxicity to packaging cells and increased the titers of most of our stress-responsive lethal LVis. Notably, the LVi-2xAARE-GFP vector titer was 4.4-fold lower (p < 0.0001) than that of the control LV-2xAARE-GFP particles (Figure 3A). In addition, transfection with pLVi-2xAARE-GFP appeared to reduce HEK293T cell viability compared to transfection with pLV-2xAARE-GFP (Figure 3D). Thus, despite the efficiency of the pLVi constructs, they likely experience some degree of leaky expression from the inverted 2xAARE promoter.

Figure 3.

Inverting the expression cassette enhances the titers of LVi carrying a TTg

(A) Titers of recombinant LV (gray columns) and LVi (white columns) particles expressed in transducing units (TU) per mL of supernatants obtained from HEK293T cells transfected with helper plasmids and pLVs or pLVis conveying the transgene GFP alone or associated to the TTgs, BAX, iC9, CASP3, M2, aDTA, or DTA, under the control of 2xAARE promoter. Results are presented as mean values ±SD from 5 independent experiments; statistics, multiple unpaired t tests, ns p > 0.05; ∗∗p < 0.01. (B) Fold increase of LVi titers as compared to LV conveying the transgene GFP alone or associated to the TTg, BAX, iC9, CASP3, or M2, under the control of 2xAARE promoter. Results are mean values ±SD from 5 independent experiments. (C) Viability assays 24 h after transfection of HEK293T cells with pLVi-2xAARE-GFP or -GFP-p2a-TTg. Results are presented as mean values ±SD from 5 independent experiments; statistics, one-way ANOVA, and Dunnett’s multiple comparisons test; ns, not significant; ∗p < 0.05. (D) Viability assays 24 h after transfection of HEK293T cells with pLV-2xAARE-GFP (pLV) or pLVi-2xAARE-GFP (pLVi); results are presented as mean percentage ±SD from 6 independent experiments as compared to non-transfected HEK293T cells; statistics, unpaired t test; ∗∗p < 0.01.

Mechanistically, this leakiness may arise from the generation of double-stranded RNA, triggering activation of the stress response machinery, as reported in previous studies.²¹ Indeed, we observed increased activation of the integrated stress response (ISR), as evidenced by eIF2α phosphorylation, in HEK293T cells transfected with pLVi-SV40-GFP and pLVi-2xAARE-GFP compared to pLV-SV40-GFP (Figures S2C and S2D). Accordingly, while the reversed expression cassette improves LVi titers, the 2xAARE promoter still exhibits some leakiness, which activates the ISR and limits the full efficiency of LVi particle yield.

CRISPR inhibition prevents leakiness and improves LVi titers

Given the results described earlier, we aimed to block transcription from the internal 2xAARE promoter during LVi production. To accomplish this, we first assessed LVi packaging in HEK^ATF4−/− cells. The absence of ATF4 did not have a consistent impact on LVi production. The titers of control LVi-2xAARE-GFP were reduced in the absence of ATF4, while the titers of LVi-2xAARE-GFP-P2A-TTg varied divergently among vectors (Figure S3A). This suggests that ATF4 is required for LVi packaging but may also play a protective or bolstering role depending on the toxic gene being expressed. Therefore, to prevent transcription from the 2xAARE promoter on pLVis during LVi production in HEK293T cells, we employed clustered regularly interspaced short palindromic repeat interference (CRISPRi)²² (Figure 4A). We used an inactivated dCas9 fused to Krüpel-associated box (KRAB) inhibitory domain of transcription, designed to target the 2xAARE promoter or the TATA box (Figure S3B). After screening three guide RNAs (gRNAs) (Table S1), we selected KRAB1 (K1), which most efficiently blocked transcription from the activated 2xAARE promoter under leucine starvation in the presence of KRAB-dCas9 (Figure S3C). We then screened six HEK293T clones, named A–F, constitutively expressing KRAB-dCas9 and the K1 gRNA (HEK^2xAAREi), and amplified clone F, which successfully reduced luciferase expression from a transfected pLVi-2xAARE-Luc construct to levels similar to those observed in HEK^ATF4−/− cells (Figure S3D). Transfection of pLV-2xAARE-Luc with CRISPR inhibition (Ci) in HEK^2xAAREi cells (Ci-pLV) resulted in a 1.6-fold reduction of luciferase expression as compared to HEK293T cells (WT-pLV) (Figure 4B), whereas transfection of pLVi in HEK^2xAAREi cells (Ci-pLVi) resulted in nearly 9-fold reduction compared to HEK293T (WT-pLVi) (Figure 4C). Thus, CRISPRi blocks transgene transcription more efficiently on pLVi than on pLV.

Figure 4.

CRISPRi prevents transgene expression during packaging increasing LVi titers

(A) Schematic representation of CRISPR inhibition (CRISPRi) on sequences overlapping the ATF4-binding domain of 2xAARE promoter; in this setting, a dead spCas9 fused to the KRAB inhibitory domain of transcription is used in association to a gRNA with sequence homology to the six ATF4-binding domains of 2xAARE promoter. (B) Levels of luciferase expression by pLV-2xAARE-Luc transfected in HEK293T (WT-pLV) or in HEK^2xAAREi cells, HEK293T cells expressing the protein KRAB-dCAS9 and the gRNA-K1 (Ci-pLV), 24 h after transfection in the presence of Tat. Results are presented as mean percentage ±SD from 3 independent experiments, statistics, unpaired t test, ∗∗p < 0.01. (C) Levels of luciferase expression by pLVi-2xAARE-Luc transfected in HEK293T (WT-pLVi) or in HEK^2xAAREi (Ci-pLVi) cells, 24 h after transfection in the presence of Tat. Results are presented as mean percentage ±SD from 3 independent experiments, statistics, unpaired t test, ∗∗∗∗p < 0.0001. (D) Titers of recombinant LV and LVi particles expressed in transducing units (TU) per mL of supernatant obtained from packaging HEK293T (gray columns) or HEK^2xAAREi (white columns) cells transfected with packaging and envelope plasmids and the vector plasmids pLVs or pLVis conveying the control transgene GFP alone or associated to the TTgs, BAX, iC9, CASP3, M2, aDTA, or DTA, under the control of 2xAARE promoter. Results are presented as mean percentage ±SD from 5 independent experiments, statistics, multiple unpaired t test, ∗∗p < 0.01; ∗∗∗p < 0.001. (E) Fold increase of titers of LVi conveying the transgenes GFP, BAX, iC9, CASP3, M2, or aDTA under the control of 2xAARE promoter, produced in packaging HEK^2xAAREi cells as compared to that obtained in HEK293T cells. Results are mean values ±SD from 5 independent experiments.

Indeed, pLV transfection in HEK293T cells represented the least efficient scenario with the highest leakage from the 2xAARE promoter. In contrast, transfection of pLVi in HEK^2xAAREi cells led to a >50,000-fold reduction in transgene expression, representing the most effective combination for restricting leaky transgene expression from the 2xAARE promoter. We then generated LVi-2xAARE-GFP-P2A-TTg particles in HEK^2xAAREi cells and observed significantly higher titers compared to the same LVi particles produced in HEK293T cells. Notably, HEK^2xAAREi, but not HEK293T cells, allowed the packaging of pLVi-2xAARE-DTA (Figure 4D). The titers of our six toxic LVi produced in HEK^2xAAREi cells were increased by approximately 3- to 6-fold as compared to those obtained in HEK293T cells (Figure 4E).

Taken together, these results demonstrate that combining an inverted expression cassette with CRISPRi targeting of the internal pLVi promoter enables the efficient production of a variety of cytotoxic LVs.

Regulatable cytotoxic LVis impact the growth of melanoma cells

To evaluate the functionality of recombinant LVi particles, we assessed their impact on the fate of EST-109 melanoma cells as a model. xCELLigence real-time cell analysis, a label-free, impedance-based method, was employed to monitor cell expansion by calculating the “cell index.” This index reflects the collective cell surface, attachment mode, and membrane composition. It increases as new adherent cells form, suggesting cell growth, and decreases as cells detach, suggesting cell death.

EST-109 cells were transduced with LVi-2xAARE-GFP or each of the six LVi-2xAARE-GFP-P2A-TTg. While vectors expressing GFP, BAX, or CASP3 did not induce cell detachment with this system (not shown), LVi-2xAARE-GFP-P2A-iC9, LVi-2xAARE-GFP-P2A-aDTA, and LVi-2xAARE-GFP-P2A-M2 particles induced EST-109 cell death following stress induction with tunicamycin (Tm) but not with DMSO, used as a control vehicle. Melanoma EST-109 cells transduced with LVi-2xAARE-GFP-P2A-iC9 maintained their growth and the formation of new adherent cells upon Tm treatment but exhibited a significant decrease in cell index when treated with a combination of Tm and AP1903, the chemical that activates the iC9 suicide protein (Figure 5A). EST-109 cells transduced with LVi-2xAARE-GFP-P2A-aDTA showed significant decrease in cell index after treatment with Tm, suggesting cell death (Figure 5B). In cells transduced with LVi-2xAARE-GFP-P2A-M2, Tm induction appeared to induce growth arrest, a phenomenon not observed in control cells treated with Tm and amantadine (Amt), an inhibitor of the transmembrane domain of the viral M2 protein (Figure 5C). Together, these results indicate that the different LVi particles encoding TTg had distinct impacts on cell growth, highlighting their potential to modulate cell survival upon 2xAARE induction.

Figure 5.

Effect of regulatable toxic LVi on attachment and growth of melanoma cells in culture

(A) Detachment of EST-109 cells transduced with LVi-2xAARE-GFP-iC9 occurs upon treatment with Tm and AP1903 but not in cells treated with Tm or with DMSO. (B) Detachment of EST-109 cells transduced with LVi-2xAARE-aDTA and treated with Tm. (C) Growth arrest of EST-109 cells transduced with LVi-2xAARE-M2 and treated with Tm, but not in cells treated with Tm and amantadine (Amt) or treated with DMSO. Statistics, two-way ANOVA and Šídák’s multiple comparisons test (A and B), or unpaired t test (C), p < 0.001, ∗∗∗∗p < 0.0001.

Engineered LVi particles effectively deliver cytotoxic genes to melanoma spheroids

The results described earlier suggest that the toxic LVi particles affect EST-109 melanoma cells, with distinct constructs leading to different modulations of the cell index. To further demonstrate the potential of our TTg system in modulating cell fate, we tested the effect of the toxic LVi particles in a spheroid 3D culture system, a reliable solid tumor model.

EST-109 melanoma spheroids were transduced once with either the control LVi-2xAARE-GFP or one of the six LVi-2xAARE-GFP-P2A-TTg vectors at a multiplicity of infection (MOI) of 10. After treatment with either DMSO or Tm, the spheroids were monitored for growth over 78 h. EST-109 spheroids transduced with the control LVi-2xAARE-GFP and treated with Tm showed a 4- to 6-fold increase in GFP expression compared to control treatment with DMSO, demonstrating the functionality of the 2xAARE promoter in these cells. In addition, comparing GFP expression in Tm-treated spheroids to that of vehicle-treated ones showed that both control LVi-2xAARE-GFP spheroids and those carrying the GFP-P2A-TTg constructs, -BAX, -CASP3, and -M2, were reasonably transduced. Under non-induced conditions, the control vector GFP or that expressing BAX displayed higher basal GFP expression than LVis expressing M2 or CASP3, likely due to the selective pressure imposed by the toxicity of the transgenes. Spheroids transduced with LVi encoding aDTA or DTA displayed negligible GFP expression after Tm treatment (Figure 6A). As observed in Figure 1D, this is likely due to effective disruption of GFP mRNA translation by these toxins. Induction of all TTg with Tm for 3 days, including aDTA and DTA, resulted in significant growth shrinkage of the spheroids, while control spheroids expressing GFP were unaffected, indicating a substantial reduction in spheroid growth due to the expression of cytotoxic transgenes (Figure 6B). Among these TTgs, BAX induction had the lowest impact, reducing EST-109 cell growth by 33%, whereas CASP3 induction had the most significant effect, reducing melanoma cell growth by 70%. Constructs such as aDTA and DTA reduced spheroid growth by 40%–50% (Figures 6A and 6B). Spheroids expressing M2 exhibited >50% size reduction and a distinct pattern of disaggregation, likely reflecting a particular form of cell death. In spheroids transduced with LVi-2xAARE-iC9, Tm alone had no effect, but when combined with AP1903, it led to a 45% reduction in spheroid growth (Figures 6D and 6E). Importantly, quantification of cell death by 7-AAD fluorescence intensity at 78 h under conditions of induction and activation showed that the expression of BAX, M2, CASP3, DTA, aDTA, and iC9 induces a 2-, 2.8-, 3.5-, 5-, 4-, and 7-fold increase in cell death compared to non-induced conditions, with no significant death observed in control spheroids treated with DMSO or Tm alone in the case of iC9 (Figures 6C and 6F).

Figure 6.

Effect of toxic transgene induction on melanoma spheroid growth and cell death

Video microscopy measurement of EST-109 melanoma cell spheroid size and fluorescence detection for GFP expression (green) and 7-AAD labeling (red) over time, following transduction with LVi-2xAARE-GFP or LVi-2xAARE-GFP-TTg. (A) Bright-field and fluorescence images of spheroids transduced with LVi-2xAARE-GFP, LVi-2xAARE-GFP-BAX, -CASP3, -M2, -aDTA, or -DTA, treated with DMSO or Tm at the endpoint of the experiment. (B) Size of spheroids transduced with LVi-2xAARE-GFP, LVi-2xAARE-GFP-Bax or -Casp3 (top), LVi-2xAARE-GFP or LVi-2xAARE-GFP-M2 (middle), or LVi-2xAARE-GFP, LVi-2xAARE-GFP-aDTA, or -DTA (bottom), over time after treatment with Tm. (C) Measurement of 7-AAD red fluorescence in spheroids transduced with LVi-2xAARE-GFP or LVi-2xAARE-GFP-TTg BAX, CASP3, M2, aDTA, or DTA) at 72 h after treatment with DMSO (−) or Tm (+). (D) Bright-field and fluorescence images of spheroids of EST-109 cells transduced with LVi-2xAARE-GFP-iC9 treated with DMSO, Tm, or Tm plus AP1903 at the endpoint of the experiment. (E) Size evolution of spheroids transduced with LVi-2xAARE-GFP or LVi-2xAARE-GFP-iC9 after induction with Tm, and co-treatment with AP1903 as indicated in the legend. For each measurement, the size of the control spheroid is represented by dotted line. (F) Measurement of 7-AAD labeling in spheroids transduced with LVi-2xAARE-GFP-iC9 and treated with DMSO, Tm, or Tm and AP1903. Results are from at least 2 independent experiments, statistics, two-way ANOVA of repeated measures of size, mixed model analysis of 7-AAD labeling, and Dunnett’s multiple comparison test (for iC9); ns, p > 0. 05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

Together, these results demonstrate that our engineered LVi particles can effectively transduce cells and express cytotoxic proteins capable of inducing cell death upon transgene induction. This underscores our approach as a promising strategy for producing LVs encoding inherently toxic factors for targeted therapy.

Discussion

To suppress the detrimental effects of toxic transgene leakage during LV production, which reduces particle yield, we developed a novel universal multi-pronged strategy. The approach combines (1) a stress-responsive promoter to minimize background transgene expression, (2) inversion of the internal LV expression cassette to prevent transcriptional control by the 5′ LTR, and (3) incorporation of a KRAB-dCas9 fusion protein with a gRNA complex in packaging cells to further prevent leakage of transgene transcription. These stepping stones enable efficient production of LVi particles carrying highly cytotoxic transgenes. As proof of translational potential, our LVi-based vectors successfully delivered toxic genes, such as BAX, CASP3, M2, or DTA, into melanoma spheroids, reducing growth and inducing cell death upon promoter induction. This highlights our approach as a promising strategy for advancing LV-based targeted therapies for cancer and beyond.

In lentiviral proviral DNA, the strong 5′ LTR promoter drives the transcription of the RNA vector genome, which often contains an internal promoter and transgene in the same 5′ to 3′ orientation. During production, however, 5′ LTR activity can override the internal promoter, causing unintended transgene expression. Antisense RNA strategies often used in DNA virus vectors are unsuitable for LVs, as they would degrade the proviral genome.^23,24 By inverting the transgene orientation, we effectively disrupt transcription from the 5′ LTR, providing a safer and more efficient LV design.

Despite its efficacy, an inverted cassette can reduce titers by generating double-stranded RNA during transient transfection, activating the ISR and inhibiting general mRNA translation.^21,25 Strategies such as knocking out ISR-related genes or expressing anti-ISR viral factors can restore titers,^21,26 but they do not suppress transgene expression. Silencing the promoter in packaging cells, therefore, provides a more suitable approach to prevent double-stranded RNA and TTg production during particle assembly. Alternative strategies have also been explored. For example, knocking out the endogenous DPH1 gene or complementing with a DTA-resistant EEF2 protein confers resistance to diphtheria toxin in HEK293T cells,^6,7 expressing that the antiapoptotic baculoviral p35 protein protects packaging cells from caspases.⁸ Procedures using bacterial tryptophan RNA-binding attenuation proteins have also been used to prevent transgene translation during LV production, improving titers for certain constructs, though their effectiveness with strong toxic transgenes has not yet been fully demonstrated.²⁷

Inverting the expression cassette alone was insufficient to fully restore titers for all TTgs, correlating with TTg leakage and toxicity as shown by a variable viability of packaging cells depending on the nature of the TTg. Residual transcription from the inverted transgene cassette likely generates double-strand RNA production, activating the ISR. Therefore, transcriptional repression became essential for producing LVi particles carrying highly toxic transgenes, to prevent leakage from the inverted internal stress-responsive promoter. While tetracycline-inducible promoters have been used for AVV TTgs,²⁸ CRISPR interference (CRISPRi) offers a versatile solution, easily adapted by modifying the gRNA. Though primarily used to suppress toxic transgene expression from plasmids in prokaryotic systems,²⁹ our work demonstrates CRISPRi’s effectiveness in repressing transgene expression from plasmids transiently transfected into mammalian cells.

By combining stress-responsive promoters, inverted cassettes, and CRISPRi, we present a multi-layered approach that surpasses current methods for LVs carrying inherently toxic transgenes. Our results demonstrate that engineered LVi particles can deliver lethal transgenes to melanoma spheroids, inducing cell death upon transgene transcriptional induction. This positions LVi-based therapies as promising tools for targeted cancer treatment, including intratumoral injection or targeted tumor transduction using engineered envelops. The differential effects of the TTg constructs on spheroid growth and cell death underscore the versatility of these LVis in targeted gene therapy.

Although our findings represent a significant advancement, limitations remain. While our methodology enabled the production of detectable levels of lentiviral particles encoding DTA, the extreme cytotoxicity of this toxin³⁰ hindered our ability to achieve titers comparable to those of control vectors. Validation in other packaging cell types and 3^rd generation LVs, which relies on a strong Rous Sarcoma Virus (RSV) or cytomegalovirus (CMV) promoter rather than on the Tat-dependent 5′ LTR, would broaden applicability. Additionally, the 2xAARE promoter can be leaky in certain transduced cells, which may require further promoter engineering to fully restrict toxic transgene expression.

In conclusion, this study provides a robust and broadly applicable strategy for producing LVs encoding inherently cytotoxic proteins, independent of their mechanism of action, thereby overcoming a major bottleneck in the field. By combining cassette inversion with targeted promoter inhibition, we offer a robust tool for safer and more controllable gene therapies. Our approach paves the way for more efficient vector production systems and expands the therapeutic potential of lentiviruses, particularly for high-unmet-need clinical applications.

Materials and methods

Cell lines

HEK293T (ATTC-CRL-11268), HEK^ATF4−/− and HEK^2xAAREi were cultured in DMEM high glucose, GlutaMAX, and pyruvate (Gibco, 10569010) supplemented with 10% fetal bovine serum (FBS) and 1× penicillin streptomycin (Gibco). ESTDAB-109 melanoma cells (UKRV-Mel-2, referred to as EST-109), a kind gift from Federico Garrido Torres-Puchol (Granada University),³¹ were maintained in Roswell Park Memorial Institute medium (RPMI) (Gibco) supplemented with 10% FBS, 1× GlutaMAX (Gibco), and 1× penicillin streptomycin (Gibco). EST-109 cells were transduced with an LVi at an MOI of 10 in their medium for at least 2 h before changing the medium. To induce ATF4 expression in HEK cells, we used DMEM for SILAC (Thermo Fisher Scientific, A33822) without leucine, or alternatively, we treated the cells (HEK and the EST-109) with tunicamycin at 4 μg/mL (Merck-Sigma, T7765). The HEK^ATF4−/− cell line was generated by knocking out ATF4 in HEK293T. Cells were transfected with two plasmids, each providing resistance to puromycin and constitutively expressing a Cas9 and a gRNA targeting exon 1 of ATF4 (tgatgtcccccttcgaccag and ggccaagcacttcaaacctc). After 24 h, the transfected cells were cultured with selective DMEM containing 4 μg/mL of puromycin (Merck-Sigma, P7255) for 72 h. Resistant clones were then amplified, and eventually cultured in DMEM lacking leucine for 16 h, and lysed for eIF2α phosphorylation and ATF4 expression analysis with immunoblot. One clone displaying phosphorylated eIF2α but lacking the ATF4 band was isolated and named HEK^ATF4−/−. HEK^2xAAREi cells were obtained from HEK293T cell line constitutively expressing KRAB-dCas9 and efficiently blocking transcription at the 2xAARE promoter, with one (K1) out of three screened gRNA targeting this promoter (Table S1; Figure S3). Briefly, a LV encoding KRAB-dCas9-p2a-BSD (blasticidin resistance gene) under the EF1α promoter (a gift from Jorge Ferrer, Addgene plasmid # 118154) and expressing the K1 gRNA (ggtttgcatcacccggaccg) under the U6 promoter was obtained and used to transduce HEK293T cells. Transduced cells were selected with 1 μg/mL blasticidin (Merck-Sigma, 203350), and several clones were isolated, transfected with a plasmid carrying a 2xAARE-Luciferase-polyA expression cassette and cultured in SILAC lacking leucine for 16 h. Luciferase activity was quantifyed, and the clone which most efficiently blocked luciferase expression was selected and called HEK^2xAAREi. Viability of transfected cells was assessed with PrestoBlue (Thermo Fisher Scientific P50200) according to manufacturer protocol and measured through fluorescence dosage in the cell’s supernatant with a SpectraMax i3x; it was referred as percentage of viability as compared to no transfected cells.

Lentiviral constructs and vector production

Recombinant self-inactivated (SIN) pLVs with cPPT-CTS sequences (pLV)³² were cloned using NEBuilder HiFi DNA Assembly (New England Biolabs) insertion of the synthesized (Twist Bioscience) or PCR amplified DNA fragment cassettes into destination lentiviral genome. The cDNA sequences of BAX (S184 del) (Uniprot: Q07812), CASP3 (V266E) (Uniprot: P42574), M2 (H37A) (Uniprot: M2 H37A), DTA (Uniprot: P00587), DTA(H21A) and iC9, were cloned under the regulatable 2xAARE-YB TATA promoter, composed of two TRIB3 AAREs¹¹ and the YB_TATA synthetic minimal promoter for increased induction fold as compared to the HSV TK TATA box.¹⁴ In pLVs, the 2xAARE-YB TATA promoter and the transgenes GFP or GFP-p2a-TTG were cloned between the cPPT and the 3′ LTR using NEBuilder HiFi DNA assembly (New England Biolabs). In pLVis, the 2xAARE-YB TATA promoter, the transgenes GFP or GFP-p2a-TTG and SV40 poly A were cloned in inverted orientation between the 3′ LTR and the cPPT and using NEBuilder HiFi DNA assembly (New England Biolabs) (Figure S4). Recombinant LV and LVi particles were obtained as previously described.³² Briefly, lentiviral particles (LV) or lentiviral particles with inverted cassette (LVi) were produced in HEK293T, HEK^ATF4−/− or HEK^2XAAREi cells. Cells were co-transfected with the pLV or pLVi and trans-complementation plasmids p8.91 and pVSVg envelope expression. Supernatants were collected 48 h post transfection and ultracentrifuged at 60,000 × g and 4°C for 1.5 h. Particles were concentrated 1,000 times in PBS and stored at −80°C until use. Recombinant particles were titrated by measurement of p24 viral protein with ELISA (Kit ref. 0801008, Zeptometrix). In all experiments, we arbitrarily considered that 1 pg of p24 is equivalent to 100 transducing units (TU), as previously established by others.³³

Flow cytometry

GFP expression and Viability with Live-or-Dye (405/452 Biotium, 32003) in HEK293T were assessed by flow cytometry using a MACSQuant Analyser 10 (Miltenyi Biotec). Then, downstream analysis was performed using FlowJo v.10.9 software.

Luciferase activity

Cells were lysed in Ripa buffer (Thermo Fisher Scientific, 89901), and total protein was quantified using the Pierce BCA Protein Assay kit (Thermo Fisher Scientific, 23227). Firefly Luciferase activity was quantified in lysate using the Luciferase Assay System kit (Promega, E1500). Luminescence (Luc assay) and absorbance (BCA assay) were measured on Spectramax i3X (Molecular Devices). Relative light units (RLUs) were normalized to the mass of protein per μL of lysate.

Immunoblotting

HEK cells were lysed with 1× RIPA buffer (Thermo Fisher Scientific, 89901), and protein contents were determined using Pierce BCA Protein Assay kit (Thermo Fisher Scientific, 23227). Proteins from each sample were then separated on 10% SDS-PAGE gel under reducing conditions and transferred onto a nitrocellulose membrane (using Invitrogen iBlot 2). After the transfer, membranes were blocked for 1 h with blocking solution (1× tris-buffered saline, 0.1% Tween [TBST], supplemented with 2% BSA) at room temperature and then incubated with blocking solution containing the following primary antibodies: rabbit anti-ATF4 antibody (clone D4B8, Cell Signaling Technology, 11815), rabbit anti-p-eIF2α (ABCAM, 32157), or rabbit anti-RPL13 antibody (Atlas Antibodies, HPA038751), each at 1/1,000 dilution in TBST overnight at 4°C. The next day, membranes were washed in TBST and incubated for 1 h with a goat anti-rabbit HRP- conjugated secondary antibody (Bio-Rad, STAR 124P) in TBST before being washed again with TBST. Membranes were then incubated for 5 min with SuperSignal West Pico PLUS Chemiluminescent Substrate reagents (Thermo Fisher scientific) before being analyzed using the ImageQuant LAS 4000/4010 (GE Healthcare).

Spheroid formation

Five thousand EST-109 melanoma cells were transduced with the LVi of interest at an MOI of 10, seeded in 150 μL of EST-109 spheroid medium (RPMI supplemented with 5% FBS [Gibco], 1× GlutaMAX [Gibco], 1× penicillin streptomycin [Gibco], and 2.5% Matrigel basement membrane matrix [Corning]) in ultra-low attachment round-bottomed 96-well plates (PHCBI, MS9096UZ), and were then spinoculated at 200g for 3 min. Spheroids were grown for 4 days before being used for further experiments.

3D Incucyte cytotoxicity assay

EST-109 spheroid media were supplemented with 50 μL of medium containing DMSO (1/10,000) or Tm (4 μg/mL at final concentration in the well). The red fluorescent intercalating agent 7-AAD was added in the culture medium at 0.1%, 24 h before endpoint experiment according to data sheets (Miltenyi Biotec, 130-111-568). Red (7-AAD) and green (GFP) fluorescence were assessed every 2 h using a 4× objective on the Incucyte system (Essen Bioscience), located within a cell culture incubator set at 37°C and 5% CO₂, capturing a whole well image at each time point; images were captured using a 4× objective with the S3/SX1 G/R optical module over a well area of 29.4 mm², with acquisition time of 300 ms in the green channel and 400 ms in the red channel. The size of the largest GFP object was determined using Incucyte software (Essen Bioscience). For analysis of spheroid growth, at each time point, each spheroid size was normalized with the average size of the control spheroids expressing GFP, and spheroid sizes were expressed as percent of relative size change compared to respective controls.

Statistics

Statistical analyses were conducted with GraphPad Prism 10. Details regarding sample size and plot descriptions can be found in figure legend for each analysis. To assess statistical variances between two groups, an unpaired Student’s t test was employed. Statistical differences between three or more groups were performed using analysis of variance (one-way or two-way ANOVA) with the appropriate multiple comparison tests. In plots, significance is denoted by p values less than 0.05; not significant (ns), p > 0.05; ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001; ∗∗∗∗p < 0.0001.

Data and code availability

The source data and materials that support the findings of the study are available upon request to che.serguera@asfaliabiologics.fr.

Acknowledgments

This work was funded Asfalia Biologics. The authors thank Dr. Niclas Setterblad, Christelle Doliger, and Claire Maillard from Technology platform of Saint-Louis Research Institute (Paris University, Paris, France) and Blandine Bonnamy and Sofian Laoues from the Vectorology platform of Institut du Cerveau (ICM).

Author contributions

B.V., R.A.-D., and C.S. conceived the study and designed the experiments. B.V. and E.M. performed the experiments. B.V. and A.S. produced vectors. R.A.-D., C.S., and B.V. analyzed the data. J.M., Y.C., D.C., and P.R. contributed to the conception and critical review of the manuscript. Y.C., J.M., and C.S. acquired funding. B.V., R.A.-D., and C.S. wrote the manuscript. R.A.-D. and C.S. critically reviewed and approved the manuscript.

Declaration of interests

A.S., Y.C., D.C., J.M., and C.S. serve for Asfalia Biologics. B.V. was a PhD student at Asfalia Biologics and serves for Vector Builder. The authors declare that patent applications related to this work have been filled.