The overall safety and efficacy of retroviral vectors for clinical applications has been proven in many clinical trials (reviewed in (11, 12)), first in the gene therapy space and more recently in immune cell therapy. Still, several potential risks remain associated with the use of retroviral vectors in the clinics.

One obvious safety concern inherently linked to the use of viral vectors is that recombination of the vector with wild-type viruses may lead to the unintended generation of replication-competent viral particles, during vector or CAR T-cell manufacturing or in the recipient patients. Thanks to the long-standing preclinical and clinical experience with viral vector delivery in gene therapy, this risk has been greatly mitigated by optimizing the packaging system design. Indeed, in current third-generation retroviral packaging systems, the genes required for vector production are split across three different plasmids: one plasmid carries the engineered viral genome, containing the transgene(s) and only the viral sequences necessary for packaging and integration, whereas all the other components are supplied in trans by a second plasmid providing the structural gag gene and the regulatory proteins, and by a third plasmid providing the env protein. Such design ensures that, even in case of a recombination event between the engineered vectors and the genome of wild-type viruses, the resulting genetic material will never carry all the essential viral genes and therefore that no replication-competent viral particles can emerge. Safety can be further increased by introducing deletions in the U3 region of the LTR, which functions as promoter/enhancer for the viral genome, thus making the viral vector self-inactivating (SIN) upon integration (13). Most of the lentiviral vectors used in clinical applications are third-generation SIN vectors (14), ensuring a higher safety profile, although they pose alternative issues during manufacturing, as the LTR deletions tend to reduce the vector titers (15), and thus require further tweaking of the manufacturing conditions, such as optimization of the packaging cell lines, of the transduction efficiency and of the purification steps. At the regulatory level, the U.S. Food and Drug Administration (FDA) requires replication-competent retroviruses (RCRs) and lentiviruses (RCLs) detection assays for viral vectors and virally transduced cell products (16), and many CAR T-cell clinical trial protocols establish patient testing for preexisting viral infections as a requirement.

A second risk associated with CAR T-cell manufacturing with viral vectors is the unintentional introduction of the transgene in other cell types. Viral vectors are pseudotyped, i.e., the original env glycoprotein is replaced to improve the tropism for different primary cell types. Lentiviral vectors, in particular, commonly use the vesicular stomatitis virus glycoprotein (VSV-G), which recognizes the ubiquitously expressed low-density lipoprotein (LDL) receptor (14). Pseudotyping with VSV-G allows the transduction of a wide range of cells, but at the same time increases the chance of off-target transduction. Development of a resistant transgenic leukemia clone has already occurred in a patient treated with the CAR T-cell therapy Kymriah, manufactured with a VSV-G pseudotyped lentiviral vector, upon the accidental transduction of a leukemia B cell during production (17). Although rare, such events highlight the risk of using integrating vectors with broad tropism. As alternatives, the introduction of more stringent purification steps during manufacturing or the use of more specific env glycoproteins, such as a measles envelope-based chimeric protein capable of targeting the CD3 receptor only present on T-cells (18), have been proposed.

A third outstanding point is whether CAR integration in the T-cell genome can affect the safety and efficacy of CAR T-cell therapy. Despite the integration of retroviral vectors being not targeted, it is not a random process. Rather, it has been defined as semi-random, as different genomic features are diversely susceptible (19, 20): vector integration preferentially occurs at fragile sites, transcriptionally active regions and those recurrently involved in translocation events (21–26). γ-retroviruses and lentiviruses exhibit a distinct pattern of integration within the host genome (27), which is maintained in the derivative gene delivery systems (28) and has been confirmed in CAR T-cells generated with these technologies (29). γ-retroviral vectors preferential insert near transcription start sites, CpG islands, enhancers, and promoters. Given the promiscuous nature of the γ-retroviral LTR enhancer/promoter, insertion close to regulatory elements can lead to induction of neighboring host genes. If proto-oncogenes are involved, this may result in oncogenic transformation: in two distinct clinical trials for the treatment of inherited immunodeficiencies, patients injected with γ-retrovirus-engineered hematopoietic stem cells developed leukemia as a result of insertional activation of proto-oncogenes LMO2, MDS1-EVI1, PRDM16 and SETBP1 (30, 31). Lentiviral vectors, on the other hand, integrate more frequently within transcription units, preferentially in introns of transcriptionally active genes. Such integration pattern, along with the frequent use of eukaryotic promoters in SIN vectors to replace the inactive LTR promoter/enhancer, reduces the risk of insertional oncogenesis for lentiviral vectors (32). Still, the site of CAR integration may affect gene expression at the transcriptional level, by leading to loss-of-function mutations, or at the post-transcriptional level, by impairing alternative splicing (29). These alterations can in turn impact CAR T-cell function and, as a consequence, the therapeutic outcome. In a genome-wide analysis of retroviral vector integration on pre-infusion CAR T-cell products, poor clinical response was associated with more integration events in genes involved in neutrophil activation (29), which may mediate immune suppression activity (33, 34). Likewise, in patients experiencing high-grade cytokine release syndrome (CRS) insertions were most commonly found in pathways involved in acetyltransferase/transmembrane transporter activity (29). In a leukemia patient, complete response was driven by the profound expansion of a single CD19 CAR T-cell clone (35). The unusual clonal expansion was linked to loss of TET2 activity: the patient carried a missense mutation in one allele of the TET2 locus, and the CAR transgene had integrated into the functional TET2 allele, thus abolishing TET2 function (35). Similarly, clonal expansion of a CD22 CAR T-cell upon integration of the transgene in the CBL locus was observed prior to eradication of residual disease in a patient with B-cell acute lymphoblastic leukemia (36). Although in the last two examples the insertional mutagenesis events resulted in the expansion of therapeutically effective clones, insertion in undesirable loci may stochastically lead to less favorable outcomes. As of December 31, 2023, the FDA has reports 22 cases of secondary T-cell cancers in patients treated with five of the six FDA-approved CAR T-cell products (37). Among the 14 cases for which adequate data were available, half have manifested within 1 year from administration (range: 1 to 19 months), and in three cases the CAR transgene has been detected in the malignant clone, suggesting a direct involvement of the CAR T-cell product in the development of the secondary malignancy. With more than 27,000 doses of administered only in the United States to date, the overall rate of CAR T-cell-related secondary cancers is still quite low. However, the diffusion of CAR T-cell therapies, particularly for indications outside oncology (38–42), calls for particular attention to the matter. Along this line, the FDA has recently issued a draft guidance recommending the long-term monitoring for adverse events, including cancer, of patients receiving CAR T-cell products engineered with integrating vectors (43).

Among non-viral gene delivery methods, transposon-based systems are the most advanced and promising in preclinical and clinical settings. Transposons are mobile genetic elements with the ability to reposition themselves within the genome (45, 46). Classically, transposons encode for a transposase gene flanked by inverted terminal repeats (ITRs). The transposase recognizes and binds sequences into the ITRs, catalyzes the excision of the transposon from its original position, and integrates it into another chromosomal locus, without the need for sequence homology. Transposon-based delivery systems have been designed by splitting the transposase function and the ITRs into two components, with the payload lying between the two ITRs in a transposon vector or minicircle, and the transposase supplied in trans. The most widely used transposon delivery systems are the Sleeping Beauty (SB), reconstructed from inactive transposon sequences isolated from fish genomes (47) and the first transposon shown capable of efficient transposition in vertebrate cells, and the PiggyBac (PB), originally identified in insect cell lines (48). The two systems have many common characteristics and advantages. They allow permanent genomic insertion of transgene cassettes, leading to sustained and efficient transgene expression. Opposed to retroviral vectors, that undergo a severe loss of vector titer for payloads above ~9 kb, they have less strict constrains on the cargo size: the PB system can accommodate up to ~14 kb, and the SB has been pushed to over 100 kb when in combination with bacterial artificial chromosome (BACs) (49). The transposon elements can be maintained and propagated as plasmid DNA, making them simple and inexpensive to manufacture, with estimated costs 5 to 10 times lower than the viral vector production (50). Moreover, transposon systems efficiently transfect resting and naïve primary T-cells, not requiring T-cell activation as a prerequisite for gene delivery during CAR T-cell manufacturing, and thus potentially leading to products with superior phenotypical characteristics (51). As such, transposon-based gene delivery systems maintain the favorable characteristics of integrating viral vectors (i.e., stable chromosomal integration and long-lasting transgene expression) while bypassing many of their shortcomings (15).

Although the greater cargo capacity remains one of the main advantages of transposons over viral delivery systems, an inverse correlation between the size of the payload and the efficiency of the transposition has been observed, and overall transposons display lower transfection efficiencies than viruses (52, 53). This is partially linked to the technology most widely used to deliver the transposon components into the target cells. Indeed, the primary method of non-viral gene delivery is through electroporation, i.e. the application of electrical fields to cells to generate pores in the cell membrane, allowing the entrance of the exogenous material. This technique, however, can cause high stress to the recipient cells, resulting in decreased viability and cell loss (54–56). As one of the main factors determining the magnitude of the damage is the amount of DNA delivered, the electroporation toxicity can be mitigated by reducing the size of the transposon vector, e.g. by using minicircles (57), and by delivering the transposase in other forms than plasmid DNA, e.g. mRNA or protein (58). These alternatives also grant a higher level of safety over the delivery as DNA, due to the transient expression of the enzyme and the impossibility of integration of the transposase-coding sequence into the host genome, thus preventing the repeated and unintentional mobilization of the transposon. In parallel, transposases have also been extensively optimized, increasing their transposition efficiency (59–62), to make CAR T-cell production through this method more scalable (60). In particular, the current benchmark SB transposase, SB100X (59), allows for viral-vector-like efficiency of gene integration, thanks to a 100-fold increase in transposition activity in comparison to the first-generation enzyme. Such improvement comes from the combination of molecular evolution and rational, crystal structure-driven optimization of the DNA binding domain through point mutation (61). Through a similar approach, the SB100X transposase has been further engineered to overcome the stability, solubility, and aggregation issues that limited its use as recombinant protein in CAR T-cell manufacturing (58). The improved biochemical properties of this high solubility SB (hsSB) enable purification of biologically active recombinant protein, electroporation of the protein into human cells, and freeze-thaw cycles without compromising transposase activity. As a result, anti-CD19 CAR T-cells could be efficiently generated via hsSB electroporation and displayed antitumor potency in xenograft mice comparable to approved viral vector-based commercial products (58).

In the last decade, an increasing number of clinical trials has been launched using CAR T-cells generated via transposon-based delivery systems (extensively reviewed in (15, 63, 64)), with efficacy results in line with trials based on retroviral-vector generated CAR T-cells. Unfortunately, however, the same kind of safety issues have also emerged. In the CARTELL trial (ACTRN12617001579381), a phase-I study investigating the efficacy and safety in relapsed and refractory B-cell malignancies of an anti-CD19 CAR T-cell product generated with the PB technology, 2 out of 10 patients developed CAR T-cell-originated lymphoma, resulting in one fatality (65). Post-hoc analysis revealed an unusually high vector copy number (VCN) of the transgene (24 in one patient, compared to the FDA-recommended threshold of 5) in the malignant CAR T-cells, associated with significant copy number gains and losses of multiple chromosomes and transcriptional readthrough from the transgene promoter, although the other patient only showed a VCN of 4 (66). Interestingly, intronic insertion into the BACH2 gene, with consequent downregulation of the gene expression, was observed in both patients. BACH2 is a DNA-binding transcriptional regulator with a putative tumor suppressor role and has already been associated with cutaneous T-cells lymphomas (67, 68). Although no causal correlation could be established between these events and the CAR T-cell-originated malignancies in exam, it is conceivable that a high VCN and the integration into or in the vicinity of proto-oncogenes increase the probability of insertional oncogenesis (64).

Conversely, no adverse events were observed in clinical trials where CAR T-cells were generated via SB technology. The reason may reside in the different integration profiles of the two transposon-derived vectors. Indeed, PB displays a γ-retrovirus-like integration pattern, with a higher frequency of insertion into transcriptional start sites of genes, CpG islands and DNase I hypersensitive sites (69, 70), and it is more prone than SB to associate with oncogenes (71). Conversely, SB integration occurs in a close-to-random manner (69, 72), so that SB has a higher probability to land in safe harbor sites compared to PB and retroviral vectors (69).

Another safety liability of the PB system may be the recent discovery in the human genome of an active DNA transposase with high homology to the PB enzyme, namely piggyBac transposable element derived 5 (PGBD5) (73, 74). PGBD5 catalytic activity has already been mechanistically linked to site-specific DNA rearrangements associated with several childhood solid tumor types (75). PGBD5 can possibly mediate the remobilization of PB transposons in PB-engineered human cells, although some observations suggest that PGBD5 may not be able to efficiently bind, excise or integrate the PB transposon, due to species-restricted recognition of the cognate ITRs (76).

Based on the available information, despite both transposon-based systems overall represent a valid alternative to viral vectors for CAR T-cell manufacturing, SB may have an edge over PB and viral vectors themselves thanks to its more favorable safety profile. Still, it is important to remark that any technology based on the integration of genetic material into the host genome presents an inherent risk of causing adverse events due to gene dysregulation or insertional mutagenesis.

Approaches that do not require the stable integration of the CAR-encoding transgene may overcome the risks associated with integration. Potential alternatives under preclinical investigation are non-viral episomal DNA nanovectors, such as the nano-S/MARt (nS/MARt) (77), and integration-deficient viral vectors (78, 79). Despite proof of concept for the generation of clinical-grade CAR T-cells using these technologies has been obtained, the therapeutic efficacy of such products still awaits clinical validation. Another strategy to eliminate the risk of oncogenic insertion is to transiently express the CAR from an mRNA template (80). The CAR-encoding mRNA, typically delivered into T-cells by electroporation, ensures transgene expression for approximately one week, with expression levels declining over time (80, 81). Such transient expression may also be a valuable means to reduce potential toxicity, particularly when the CAR target is also present in healthy tissues. The available clinical data suggest that mRNA-generated CAR T-cells have a good safety profile and exhibit short-term anti-tumor efficacy (82, 83), although by means of multiple injections, but their ability to achieve durable responses has yet to be proven.

VSV-G pseudotyped lentiviral vectors have a broad tropism, achieving high transduction efficiencies on different human cell types. A major risk for in vivo delivery is the unintentional engineering of non-target cells, as no selection and purification process of the recipient cells is possible. Attempts to alter receptor usage of the VSV-G protein have been made, e.g. by fusing to the VSV-G antibody single-chain variable fragments (scFv) specific for surface markers expressed on the target cells.Lentiviral vectors have been successfully targeted towards CD30 and epidermal growth factor receptor (EGFR) (86), but the engineering of VSV-G remains challenging, as it mediates both receptor binding and membrane fusion (87). A more viable strategy relies on the substitution of VSV-G with glycoproteins from alpha- and paramyxoviruses, which have separate envelope proteins for binding and fusion and thus allow for alterations of the tropism without interfering with the fusion process (88, 89). Such lentiviral vectors have been targeted against T-cell markers (CD3, CD4, CD8) using both scFvs and DARPins (reviewed in (85), reaching an on-target selectivity in human peripheral blood mononuclear cells (PBMCs) of up to 99% (90).

Despite their long history in gene therapy, AAVs suffer several limitations for in vivo CAR T-cell generation. Their single-stranded DNA genome allows for limited cargo capacity (~4 kb) (91) and for mostly transient transgene delivery, especially in actively proliferating cells such as activated lymphocytes (92). Moreover, due to the lack of an envelope, AAVs need to rely on clathrin-mediated endocytosis for binding and internalization, with the involvement of several capsid and cellular proteins in the process. This adds a layer of complexity to the tweaking of AAV specificity, requiring either the engineering of the capsid proteins or the functionalization of the capsid itself with the binders. Nonetheless, highly selective modification of CD8+ within primary human splenocytes was achieved though DARPin-targeted AAVs (DART-AAVs) (92). Furthermore, AAVs were successfully used for in vivo CAR T-cell generation in humanized mouse models (93) The obtained CAR T-cells showed potent, dose-dependent antitumor activity ex vivo, and in vivo functionality and efficiency comparable to that of conventionally manufactured CAR T-cells (93).

The effort imbued in the development of SARS-CoV-2 vaccines led to a substantial advancement in non-viral vector technologies. These delivery systems rely on the chemical and physical properties of the payload and carrier combination, rather than on the sophisticated viral machinery. In LNPs, the nucleic acid is encapsulated in a lipid particle through electrostatic interaction, whereas in NCs its negative charges are exploited to complex it with positively charged polymers. They can host DNA or mRNA payloads and better preserve T-cell viability compared to other non-viral delivery technologies, such as electroporation (94). Furthermore, the possibility to be easily functionalized with targeting molecules and their low immunogenicity makes them highly suitable for in vivo CAR T-cell generation. In a head-to-head comparison between electroporation and LNPs for the ex vivo generation of CAR T-cells via mRNA delivery, LNPs led to prolonged mRNA persistence and CAR surface expression. The obtained CAR T-cells also showed a less exhausted phenotype, likely due to the reduced stress compared to electroporation (94). Both LNPs and NCs proved capable of delivering plasmid DNA and in vitro-transcribed RNA cargos to T-cells in vivo (95–97), and antibody-conjugated LNPs were specifically targeted toward PECAM-1 (98), CD4 (99) and CD5-positive (96) cell populations.

In general, several factors need to be considered for the clinical translation of in vivo strategies. The vast majority of preclinical studies are performed in humanized mouse models, which provide only on-target cells of human origin, making the prediction of the vector biodistribution and of the off-targets in the human body difficult. Likewise, the kinetics of in vivo-generated CAR T-cells are necessarily different from their ex-vivo-produced counterparts: for the latter, high numbers of effector cells are instantly available upon administration, whereas for the former the vector injection leads to a limited pool of engineered cells, that will have to expand in vivo over time. How this difference may affect the behavior and the efficacy of the CAR T-cells in the clinical setting has yet to be elucidated. In all non-integrating technologies, the transgene expression is eventually lost, possibly requiring re-dosing to maintain CAR T-cell levels that ensure long-lasting tumor management. However, the immunogenicity of most vectors for in vivo use may be an obstacle for their repeated administration. Neutralizing antibodies have been shown to rise already after the first systemic injection for both LVs and AAVs (9, 100), and pre-existing neutralizing antibodies may be present in a relevant fraction of the patients due to vaccinations or previous infections (101). Lastly, despite the eased manufacturing of in vivo-generated CAR T-cells, the availability of sufficient quantities of GMP-grade vectors will be a key factor in the clinical deployment of in vivo strategies.

A first risk factor while using nuclease-based genome editing approaches is the possibility of accidentally introducing off-target cleavages. Indeed, Cas9:gRNA complexes can recognize and bind genetic loci with as little as 5-nt of homology with their RNA guide (122), leading to unintended, irreversible off-target genetic alterations. A variety of strategies has been proposed to mitigate such risk, i.e., more sophisticated approaches to gRNA design (124, 125) that rely on a deeper understanding of the parameters leading to off-target binding (126, 127), the use of single-strand nickases along with paired sets of gRNAs (128–130), or the delivery of a pre-formed Cas9:gRNA ribonucleoprotein (RNP) complexes into cells. The latter strategy allows for Cas9 to be active immediately, but the RNP is also quickly degraded once internalized, therefore decreasing the amount of time Cas9 is present for potential off-target cleavage (131, 132).

Field evidences clearly point at the necessity of multiple genetic interventions to overcome the current CAR T-cell limitations, and gene editing technologies have already been explored for the simultaneous targeting of multiple genomic sites in T-cells (133–137). Still, their use for multiplexed gene editing poses relevant biological and technical challenges. Some evidence show that the targeting of Cas9 to different genes simultaneously in human cells could in principle mediate genetic disruptions for each target at efficiencies similar to those achieved by targeting each locus individually (138). However, most often the co-occurrence of all the desired edits within the same cell wanes for high-grade multiplexing (122), posing a serious limit to the scalability and manufacturability of multi-edited products. The competition for a dwindling pool of endonucleases as the number of gRNAs scales up (139) may contribute to this phenomenon, giving raise to unpredictable patterns of genetic modification. To limit cross talk between components, different Cas nucleases have been used in combination, e.g. Cas9 and Cas12, each recognizing a slightly different gRNA structure, thus enabling orthogonal assembly of the ribonucleoprotein complexes (140). The downside of this approach, however, is that a much larger payload needs to be accommodated in the delivery vector, due to the combined presence of the different Cas nucleases.

If the off-target activity and the complexity of multiplexing might be mitigated in the future by advancements in gene-editing technologies, dealing with mutational events and chromosomal rearrangements represents a bigger challenge. Indeed, the biological mechanism behind gene editing itself, which relies on DSBs and on the error-prone DSB-repair machinery, is intrinsically at risk of introducing undesired mutations and chromosomal aberrations. CRISPR/Cas9 cleavage has been linked to gross chromosomal aberrations such as large deletions in early mouse (141, 142) and human embryos (143), as well as in embryonic stem cells and induced pluripotent stem cells (144, 145). More specifically in the CAR T-cell field, CRISPR-based editing of primary human T-cells at the TRAC locus (Ch14q11.2) was shown to lead to chromosome 14 truncation at the cleavage site in 5.3% of the cells (146). Given the acrocentric nature of chromosome 14, this is functionally equivalent to chromosomal loss. Even more strikingly, when the TRAC and TCRB (Ch7q34) loci were co-edited, the percentage of cells with chromosome 14 truncation raised to 9%, along with chromosome 7 truncation at the TCRB locus in 9.9% of the cells (146). This body of evidence is especially relevant for the immune cell therapy field, as TCR disruption is an obliged step for the generation of allogeneic CAR T-cell products. Lastly, the intrinsic modus operandi of the repair machinery, that acts stochastically on different alleles in the resolution of DSBs, may lead to intra-allelic mosaicism and, eventually, to batch-scale chimerism even in clonal genetically edited cell products (137, 147, 148).

Base editing consists in the sgRNA-directed exchange of single nucleotides mediated by modified forms of Cas9, Cas9 nickases (nCas9), which lack the capacity to cleave DNA, but instead are fused to bacterial deaminases to substitute single nucleotides on just one strand. These point mutations are subsequently resolved during DNA replication, ultimately altering the codon sequence, e.g., to insert premature STOP codons (149, 150). CRISPR-free base editors have also been recently proposed (151). Base editors allow the correction of pathogenic allele variants, holding great promise for the therapy of monogenetic disorders, including sickle cell disease (152), β-hemoglobinopathies (153, 154) and heterozygous familial hypercholesterolemia (155). The first applications in cellular immunotherapy are also emerging, e.g. to engineer fratricide-resistant CAR T-cells by targeting the pan-T lineage antigens CD3 and CD7 (156) and to generate allogeneic CAR T-cells through simultaneous KO of TRAC, CD52, CD7 and PD-1 (157). Compared to CRISPR-edited CAR T-cells, base-edited CAR T-cells showed improved proliferation, lower DNA damage response pathway activation, and no karyotypic abnormalities following multiplexed editing while proving efficacious against T-cell acute lymphoblastic leukemia (T-ALL) both in vitro and in preclinical models (157, 158).

Base editing allows for transition mutations (purine-for-purine and pyrimidine-for-pyrimidine substitutions), but not for transversion mutations (swapping purines for pyrimidines and vice versa). Prime editing overcomes this limitation by linking nCas9 to a reverse transcriptase that allows for single base exchanges as well as insertions and deletions of synthetic DNA sequences (159). This technology, however, still suffers from limitations in editing efficiency and in the maximum size of the insert. Efforts are ongoing to improve these aspects, e.g. by designing more efficient guide RNAs for the nCas9 (160) and by building new iterations of the platform with increased payload capacity, such as paired-guide prime editing, which currently allows for insertions of up to ~5 kb (159).

Similarly to standard nuclease-based editors, base and prime editors are at risk of modifying off-target genomic sites. A general strategy to mitigate this risk has consisted in limiting cellular exposure to the editors beyond the duration necessary to achieve the desired on-target editing levels. The delivery of editors and guides as RNP complexes rather than as DNA sequences or the use of small-molecule-controlled editors greatly reduced off-target base editing (161–163), exploiting the faster rate of on-target compared to off-target editing.

Targeting the transgene integration to selected loci has been a long-standing goal in gene editing, as it would overcome many of the safety risks associated with insertional mutagenesis. Unfortunately, none of the current clinical-stage engineering technologies fulfills this goal: viral and transposon-based delivery systems do not allow any targeting, whereas nuclease-based, HDR-dependent technologies are limited to actively dividing cells, strongly constraining their applicability. However, novel technologies with the potential of site-directed integration of large payloads are quickly emerging. Phage-derived site-specific recombinases can catalyze the exchange of two dsDNA sequences by recognizing a “landing pad” DNA sequence at the site of insertion. This characteristic has been exploited to mediate the recombination of large cargo sequences in the desired genomic locus pre-installed with the unique landing pads (164). Using this approach, incorporation of very large DNA payloads (>100 kb) with very high efficiency (90%) after selection steps was shown in human iPSCs (4). Site-specific recombinases have remarkable advantages: they do not leave exposed DSBs, there is virtually no upper limit to the size of the cargo, and they are not prone to off-target effects, provided the pre-installed landing pad is inserted precisely on target. Still, their application is rather complex, consisting in multiple steps: the CRISPR-Cas-mediated pre-installment of the landing pads, the delivery of the recombinase-expressing vector and of the DNA payload, the removal of unnecessary auxiliary sequences (e.g. the vector backbones). Each of the steps requires selection and enrichment of the intermediate product cells to cope with the otherwise very low recombination rate (less than 1% in cells containing the landing pad), making this approach not yet scalable nor applicable to most primary cells (4). However, novel evolutions of this technology may overcome these constraints in the near future. Programmable addition via site-specific targeting elements (PASTE) technology, consisting of a Cas9 nickase fused to RT and a serine integrase, can integrate large sequences in human cell lines, primary T cells, and non-dividing primary human hepatocytes with efficiencies between 5% and 60% (165). Lastly, CRISPR-guided transposon systems (CAST), combining the site-specificity of a catalytically inactive Cas with the ability of the transposon machinery to integrate the cargo in the host genome, have been used to deliver payloads of up to 10 kb (166–168), although their applications has been so far limited to gene editing in bacteria.

Although these novel approaches do not yet show efficiencies required for clinical scalability and manufacturing and have not been sufficiently tested in primary cells, the extremely rapid evolution of the field may soon bring them closer to be applied in clinical settings.

Branching from the standard CRISPR/Cas9 gene editing approach, catalytically disabled Cas9 (dCas9) have been fused to transcriptional modulators. The dCas9 lacks its catalytic activity, but retains its sequence-specific binding ability, thus allowing the transcriptional modulators to alter the target gene expression at the epigenetic or transcriptional level, without the need of DSBs (169). This approach can be used to enhance or repress the transcription of the target gene. CRISPR activation (CRISPRa) exploits effectors such as VP64, VPR, Suntag, p300, and Synergistic activation mediator (SAM) to induce epigenetic changes, e.g. histone acetylation, that lead to enhanced gene expression (169, 170). CRISPR inhibition/interference (CRISPRi) uses epigenetic regulators of DNA methylation, histone acetylation, or histone methylation, e.g. Krüppel-associated box (KRAB), DNMT3A, and HDAC, to downregulate gene expression (169, 170). Recently, a tool for programmable epigenetic memory based on DNA methylation, named CRISPRoff, has also been described, that can make such gene inhibition heritable (171). CRISPRa and CRISPRi have mainly been used in functional genomics screenings thus far. A genome-wide CRISPRi/CRISPRa screening in primary human T-cells was employed to identify therapeutically relevant T cell states which may prove useful in the design of novel T cell-based immunotherapies (172). Likewise, a CRISPRa gain-of-function screening in murine CAR T-cells highlighted proline metabolism as a driver of CAR T-cell fitness and function (173). Still, the first applications in cell engineering are already emerging, notably also in contexts such as iPSCs (170) and primary T-cells (172, 174). Specifically, multiplex gene modulation of up to four genes (ITGA3, THY1, IL3RA, and NGFR) was achieved in human CD34+ hematopoietic stem and progenitor cells (HSPCs) and human CD3+ T-cells, without any discernible detrimental effects on the HSPCs capacity to engraft long-term in immunodeficient mice and on T-cell expansion (174). CRISPRa and CRISPRi have also been exploited to tweak primary T-cell responses, e.g. cytokine secretion, by impinging on their regulatory pathways (172). Interestingly, both the magnitude and the duration of the CRIPSRa/CRISPRi effect could be tuned by careful sgRNA selection, and further tuning may also be possible by adapting other sgRNA properties, such as the poly(A) tail length, codon usage, and incorporation of modified nucleotides (174). In terms of safety, several studies have found CRISPRa and CRISPRi to be highly specific (175–177). In contrast to gene editing with nuclease-active Cas9, the combination with epigenetic effectors makes CRISPRa and CRISPRi only active within a short distance from the targeted transcription start sites. In addition, the extent of the potential adverse effects is mitigated by the reversible nature of the modifications and can be further reduced by temporally restraining the activity of the editors via their transient delivery as RNP complexes.

A radically different approach for non-gene editing technologies consists in targeting the mRNA of the gene of interest, rather than its genomic sequence, thus inducing defined cellular phenotypes without introducing genomic alterations. In this light, the recent discovery of CRISPR/Cas13, a Cas-based system that uses CRISPR RNA guides to target RNA, spurred considerable interest (178, 179). This Cas allows for direct transcriptome engineering via RNA editing and KD, without the requirement for permanent genetic manipulations (180–182). In early reports CRISPR/Cas13 mediated potent and specific target RNA downregulation in eukaryotic cells, outperforming CRISPRi, apparently with minimal off-target transcriptome changes (178, 180, 183). The system has since been adapted for use in a variety of contexts, from yeast (184) to Drosophila (185) to mice (186–189), and has been applied to neutralize viral infections in animal models (190, 191). However, the high specificity of CRISPR/Cas13 in eukaryotic cells observed in early studies represents a paradox. Indeed, due to its molecular architecture, with the catalytic site located on the outside of the protein, facing away from the guide RNA-target RNA complex, Cas13 is prone to indiscriminately cleave any bystander RNA. This effect, termed collateral cleavage activity, has been observed both in vitro and in bacteria (192). Coherently, several recent publications report Cas13-mediated toxicity and collateral RNA cleavage in eukaryotes (reviewed in (192)), although the reason for the discrepancies with earlier studies is still matter of investigation.

A more established approach for transcriptional regulation, RNA interference (RNAi), has also been tweaked and adapted in the last few years for the engineering of CAR T-cells, with interesting results. RNAi is a bundle of technologies, namely small interfering RNAs (siRNAs), short hairpin RNAs (shRNAs), and microRNAs (miRNAs), all based on small non-coding RNAs that regulate gene expression post-transcriptionally. They do not require a direct intervention on the target’s genomic sequence and thus they do not incur the risk of causing genomic alterations.

siRNAs are delivered in their mature, functionally active form, with no possibility to self-replenish the initial siRNA pool. Hence, siRNAs are not suitable for the development of engineered immune cells. Indeed, the rapid dilution of the siRNAs in the fast-dividing activated T cells makes this approach only applicable to obtain transient biological effect. On the other hand, shRNAs and miRNAs are continuously transcribed as precursors in the recipient cells, ensuring the long-term gene downregulation required for effective CAR T-cell engineering. shRNAs and miRNAs share the same molecular machinery for their maturation: following transcription, the immature hairpin structure is processed by RNase III enzymes (first Drosha, then DICER in the case of miRNAs, only DICER in the case of shRNAs), leading to the formation of a mature RNA duplex, which is in turn incorporated into the RNA-induced silencing complex (RISC). The accessory passenger RNA strand is then released and the RISC-guide strand riboprotein mediates target mRNA recognition and downregulation (193–195). Two important differences exist in the biogenesis of shRNAs and miRNAs. First, shRNAs are transcribed by RNA polymerase III, whereas miRNAs are driven by RNA polymerase II. Hence, miRNAs are usually expressed at lower, more tolerable levels than shRNAs and can be easily embedded in polycistronic transcriptional units that facilitate CAR T-cell engineering (196). Second, in addition to the very high expression levels, shRNA bypass Drosha during their maturation process, possibly overloading the cytoplasm with double-stranded RNA which may obstruct the natural miRNA pathway and thus lead to toxicity (197, 198). Synthetic miRNAs, in which the guide sequence has been swapped for an shRNA-based one directed against the gene of interest, can overcome this issue, as they still closely exploit the natural miRNA pathway (199), standing out as the most valid RNAi technology for immune cell engineering. Clinical validation of the miRNA-based RNAi approach in CAR T-cells has been obtained in two phase I clinical trials (NCT04613557, NCT03466320). These studies show both the functional efficiency and the high safety profile of the technology.

Beyond the aforementioned examples, RNAi has been used to engineer various features of CAR T-cells, with promising results in preclinical models, and some of these strategies are currently under evaluation in clinical trials (NCT06051695, NCT05617755, NCT06245915, NCT04649112, NCT04825496, NCT05028933). Notable applications are the downregulation of IFNγ or GM-CSF to reduce the risk of cytokine release syndrome (200, 201), the targeting of immune checkpoints to inhibit CAR T-cell exhaustion and enhance functionality (136, 202), the silencing of factors, such as adenosine 2A receptor (A2aR), to improve resistance to the immunosuppressive tumor microenvironment in the context of solid tumors (203), and the simultaneous knock-down of TCR and HLA class I for the generation of allogeneic CAR T-cells (196). These applications elucidate one of the greatest advantages of RNAi over other engineering technologies, i.e. the possibility to modulate the expression of the target gene to levels compatible with its biology, reaching functionally relevant levels of inhibition without the detrimental effects that a full KO may cause.

Another valuable characteristic of miRNAs is their frequent co-occurrence within the same transcriptional unit within the genome. About 50% of conserved vertebrate miRNAs are organized in clusters (194, 204), making them an ideal engineering strategy for multiplex applications. miRNA systems for multiple gene targeting have already been used successfully against HIV-1 and HCV (205–207), although their clinical applicability in immune cell therapy has been hampered by the relatively weak knock-down efficiencies obtained thus far (208). More recently, however, miRNA-based multiplex platforms able to effectively target up to four genes have been proposed for the engineering of CAR T-cells (196, 209) and optimized to reach efficiencies compatible with the use in clinical products (196). Overall, multiplex RNAi technologies allow for an easy, safe, efficient, and tunable modulation of several genes simultaneously.