Creating improved CAR T-cells using CRISPR gene editing

DSB induction permits gene disruption and/or transgene insertion into target loci. However, there are risks associated with such breakage when it occurs at off-target sites, potentially introducing unintended genome alterations. Thus, precise control of when and where DSBs occur is beneficial for reducing the frequencies of unintended indels and translocations [38] and preventing selection of p53 inactivated cells [39, 40]. It has been demonstrated that direct delivery of a pre-formed Cas9/gRNA ribonucleoprotein (RNP) complex into cells allows for Cas9 to be active immediately. The RNP is also quickly degraded once internalized, and therefore decreasing the amount of time Cas9 is present for potential off-target cleavage, but enough time to still maintain an optimal threshold of on-target editing efficiency [41–43].

Several Cas9 enzyme variants have also been developed to overcome the disadvantages of DNA DSB cleavage. For example, Nickase Cas9 (nCas9) variants contain mutations in either the RuvC or HNH domain which render Cas9 capable of cleaving only the targeted or non-targeted DNA strand, respectively [44] (Fig. 2b). The use of two pairs of nCas9/gRNA complexes flanking a targeted site allows for greater control of where a DSB can occur, as both complexes need to be at the same cut site for inducing the DSB, allowing for reduced off-target DSB generation [45]. Mutations in both catalytic domains create a catalytically dead Cas9 (dCas9) variant (Fig. 2c). dCas9 can easily be tethered to other functional enzymes to elicit a variety of site-specific modifications, independently of Cas9 catalytic activity [46].

In the process of base editing (Fig. 3b), introduction of DNA point mutations can create or repair single nucleotide variants, modify donor and acceptor splicing sites, alter codon composition to change an amino acid code or introduce a premature STOP codon [47–49]. Accordingly, base editors can introduce four transition mutations that substitute a pyrimidine for another pyrimidine or purine for another purine (CT, TC, GA, AG) [48]. The enzymes used to catalyze these reactions are cytidine and adenosine deaminases. Cytidine deaminase induces the deamination of cytosine bases to uracil. During DNA replication, DNA polymerase ‘reads’ uracil as thymine, as these bases have the same base-pairing properties, resulting in a functional cystine-to-thymine correction in the nascent DNA strand. Adenosine deaminase acts in a similar manor, catalyzing the deamination of adenosine to inosine (I), which is ‘read’ by DNA polymerase as a guanine [50]. Base editing acts within a small window of specificity when directed to a precise site via a guide RNA (i.e., 13–18 base pairs upstream of the PAM site) [48]. To avoid off-target cuts [51–55] while achieving maximal editing efficiency, several Cas9 and deaminase variants have been developed that exhibit different PAM specificities [56–58] or editing windows [59–61]. Initial base editors relied on dCas9, but more recent investigations demonstrated that nCas9 is preferred for higher editing efficiency, as it can nick the non-edited strand to encourage the C-U or AI modified strand to be used as the template during DNA repair, resulting in a fully preserved edit [50]. Cytidine deaminase base editors are commonly used to generate premature STOP codons during transcription [48]. Finally, adenosine deaminase base editors can be used to alter splice donor/acceptor sites.

A major limitation of base editing is that it cannot be applied to introduce transversion mutations (i.e., pyrimidine to purine and vice versa) as well as small deletions or insertions. To address this constraint, an alternative technology known as ‘prime editing’ was developed to facilitate all possible base-to-base conversions and targeted insertions or deletions without inducing DSBs [62]. In this strategy, nCas9 is tethered to an engineered reverse transcriptase (RT) enzyme and a prime editing guide RNA (pegRNA) directs this nCas9/RT complex to the desired genome region. In addition to the target recognition site, the pegRNA contains a binding region that associates with the nicked DNA strand serving as a primer, followed by an RNA sequence encoding the new template. RT synthesizes DNA from the pegRNA sequence onto the nicked DNA strand for further ligation. This creates a DNA mismatch, which can be corrected using a second nCas9/guideRNA complex (PE3 editing strategy). After nicking the unedited strand, the cellular DNA repair machinery corrects this lesion using the initial pegRNA-edited strand as a template, which leads to the full establishment of the desired gene edit at the target locus (Fig. 3b). Prime editors are capable of not only inducing base pair resolution substitutions, but also insertions of up to 44 base pairs and deletions as high as 80 base pairs [63]. Thus, like base editors, prime editors can be used to modify post-mitotic cells and demonstrate a reduced risk for genotoxicity that often accompanies the generation of multiple DSBs. Current strategies modulating the enzyme activity [64, 65] or the pegRNA structure have been undertaken to further improve the efficiency of prime editing [54, 66].

dCas9 can be used to promote or attenuate transcriptional activity based on recruitment of transcriptional activators or repressors to specific sites in the processes of CRISPR activation (CRISPRa) and CRISPR interference (CRISPRi), respectively [67–70] (Fig. 3c). CRISPRi incorporates a tethered transcriptional repressor such as a Krüppel associated box (KRAB) domain to induce repression of transcription at the specified transcriptional start site [46, 71]. Alternatively, CRISPRa involves tethering of transcriptional activators such as VP64 to promote endogenous gene expression [46, 72]. In addition, tethering demethylating ten-eleven translocation (TET) methylcytosine dioxygenases or DNA methyltransferases (DNMTs) can permit locus and region-specific epigenetic modifications [73, 74] (Fig. 3c). This strategy has previously been used to target hypermethylated promoter regions of tumor-suppressor genes [75].

Several approaches have been taken to improve the survival and function of CAR T-cells, including optimization of costimulatory endodomains [77], prevention of tonic signaling [78], inhibition of NR4A transcription factors [79], TOX/TOX2 blockade [80] and overexpression of the canonical AP-1 factor, c-Jun [81]. Recently, global analysis of histone modifications in T-cells has elucidated marked differences in the chromatin structure of distinct subsets of exhausted T-cells and their transcriptional regulators (reviewed in [82]). Weber and colleagues [83] recently demonstrated that transient cessation of antigen receptor signaling through forced downregulation of the CAR protein or inhibition of proximal CAR signaling kinases restores functionality in exhausted T-cells through global and site-specific chromatin remodeling. Because intermittent CAR T-cell ‘rest’ mitigates exhaustion and enhances antitumor potency in association with epigenetic reprogramming, engineering loss-of-function or modulation of genetic and epigenetic targets that potentiate T-cell exhaustion using CRISPR-Cas9 technology has the potential to prevent and perhaps even reverse CAR T-cell dysfunction. Targeting inhibitory receptors, transcription factors and/or other mediators of CAR T-cell dysfunction through gene editing may induce reinvigoration of infused cell products. Removing negative regulators of T-cell persistence and effector function (e.g., PD-1, CTLA-4 and LAG-3) using CRISPR-Cas9 may indeed represent a first obvious tractable point of intervention [84–87]. In addition to inhibitory receptors, disruption of Fas receptor/Fas ligand interactions have been shown to reduce activation-induced cell death and potentiate increased in vivo CAR T-cell antitumor function [88]. Finally, CRISPR/Cas9-mediated targeting of diacylglycerol kinase (DGK) metabolism in CAR T-cells renders them resistant to immunosuppressive mediators in the TME, such as TGFβ [89]

Genetic strategies to modulate cytokine signaling during CAR T-cell activation and expansion have the potential to bolster antitumor activity, enhance T-cell persistence and/or reduced toxicity. CRISPR-Cas9 systems that incorporate transgene knock-in approaches have been exploited for this purpose. More specifically, CRISPR/Cas9-based gene editing combined with viral or non-viral DNA delivery permits simultaneous bi-allelic or sequential gene targeting to engineer T-cells with expression cassettes in a site-specific manner [90, 91] (Fig. 4b). Using this technology, cytokine-encoding DNA cassettes can be knocked into targeted genomic loci placing these genes under the control of specific promoters for temporal control of expression. For example, IL-15 has been knocked into the IL-13 gene locus, thus placing IL-15 expression under control of the endogenous IL-13 promoter, which is highly active upon T-cell activation. This creates an inducible T-cell specific IL-15 activation switch [92]. Additionally, removal of genes encoding cytokines that drive neurotoxicity and cytokine release syndrome (CRS) such as GM-CSF [93] and IL-6 [94] with CRISPR-Cas9 editing has the potential to produce an optimally potent and durably persistent cell product, while reducing adverse events associated with aberrant cytokine production. Accordingly, GM-CSF knockout CAR T-cells maintain normal functions and increased antitumor activity in vivo, and potentiate improved overall survival, compared to conventional CD19 CAR T-cells [93]. Genetic knockdown or ablation of the IL-6 gene also has the potential to ameliorate CRS-like toxicity in leukemia-bearing mice [95] (Fig. 4c).

Many current CAR T-cell manufacturing protocols involve ex vivo autologous T-cell expansions followed by transduction with a viral vector containing the chimeric receptor sequence [96]. While lentiviral transduction and integration is stable and considered generally safe for clinical trials and FDA-approved treatments [97], there are possible risks for malignant transformation of engineered CAR T-cells via insertional mutagenesis of tumor suppressor genes or oncogenes [98]. Additionally, since lentiviruses integrate semi-randomly in the genome, the CAR transgene can insert in sites with high or low relative transcriptional activity, leading to variable cell-surface CAR expression, generating a sub-optimal therapeutic product [99]. Placing the CAR transgene under control of a strong exogenous promoter may also lead to high constitutive receptor expression. Elevated surface expression and interaction with other CAR receptors can generate ligand-independent tonic signaling in the absence of exogenous signal. This can induce both systemic production of cytokines as well as a cell profile that drives rapid transition to poor effector function and T-cell exhaustion [100, 101]. To address these caveats associated with lentiviral transduction, CRISPR-Cas9 can be used to deliver a CAR-encoding DNA cassette to a specific genomic location, allowing for targeted knock-in of the CAR into desired sites. For example, an anti-CD19 CAR can be directed to the T-cell receptor α constant (TRAC) locus, resulting in uniform CAR expression, reduced tonic signaling, decreased exhaustion and increased antitumor efficacy [90] (Fig. 4b) Targeting TRAC also gives the added benefit of producing a potential universal product.

The primary challenge to overcome in the setting of allogenic products is the induction of graft-versus-host-disease (GvHD), where the endogenous donor T-cell receptor recognizes ‘non-self’ surface human leukocyte antigen (HLA) molecules in the patient, eliciting an immune reaction. Healthy donor allogeneic CAR T-cells can be derived from patients’ previous HLA-matched hematopoietic stem cell transplant (HSCT) donor or from gene-edited cells which have been modified to allow them to be given to non-HLA matched subjects. There is increasing enthusiasm for the use of so-called ‘third party CAR T-cells’ (i.e., non-autologous T-cells). An off-the-shelf product would allow one to start with T-cells from healthy donors to create large quantities of universal tumor-specific T-cells that could be used in any patient without the need for HLA matching. This strategy would reduce costs, speed drug administration, and make the T-cell products accessible to lymphopenic and critically ill cancer patients that often do not have sufficient numbers of healthy T-cells for treatment. As proof-of-concept for this approach, CAR T-cells derived from healthy unrelated donors have conferred anti-leukemic efficacy in children and adults with relapsed ALL [102].

Ideally, universal T-cell products should have 3 essential characteristics: 1) resistance to recognition/rejection by host natural killer (NK) cells and T-cells, 2) optimized for potency, persistence and safety, and 3) available as on-demand therapeutics. Without the ability to incorporate such features, large-scale and reproducible low-cost production of high quality, safely engineered CAR T-cells cannot be achieved and the promise of cell therapies will remain elusive to vast populations of cancer patients around the world.

The T-cell receptor α constant (TRAC) locus has been extensively studied as an ideal target for both gene knockout and CAR knock-in [90, 91]. Placing the CAR transgene under control of the endogenous TRAC promotor will drive CAR expression in a stable and robust manner parallel to physiological TCR expression. This will also simultaneously knock out the endogenous TCR, which will eliminate GvHD concerns and allow for an allogenic T-cell product to be generated. Combining the CAR T-cell manufacturing process with simultaneous TCR knockout allows for a streamlined approach to universal CAR T-cell generation (Fig. 4b, c).

Successful multiplex CRISPR-Cas9 editing of CAR T-cells has been done with targeting inhibitory genes (e.g., PDCD1, PD-1) and those encoding death receptors such as CD95/Fas to prevent TME-mediated inhibition or apoptosis [88]. TRAC, PDCD1 and B2M triple knockout CAR T-cells have demonstrated robust anti-tumor function in in vitro and in vivo models incorporating CD19 and prostate stem cell antigen (PSCA)-directed CARs [86, 87]. In terms of the safety and clinical feasibility of such an approach, we conducted the first-in-human trial of multiplex CRISPR-Cas9 edited (i.e., TRAC, TRBC and PDCD1 knockout) T-cells with a transgenic TCR specific for tumor-associated antigen NY-ESO-1 in patients with myeloma and sarcoma [3]. Transfer of gene-edited TCR-engineered T-cells into patients resulted in durable engraftment with edits at all three genomic loci [3]. Although not involving multiplex gene editing, generation of PD-1 deficient T-cells via CRISPR-Cas9 editing of PDCD1 was also demonstrated to be feasible in the setting of clinical product scalability, and this product proved to be safe in a phase I trial for advanced non-small-cell lung cancer patients [4]. The results of the above investigations clearly demonstrate the feasibility and short-term safety of treating patients with CRISPR-edited human T-cells. Favorable outcomes from additional studies conducted with larger numbers of patients should hopefully justify more advanced Phase II and Phase III trials in the future (Fig. 4c).

The aforementioned Cas9 variants have shown promise in the pre-clinical setting. For example, primary human T-cells have been modified using base editors for disruption of TRAC, B2M and CIITA to reduce expression of the endogenous TCR, together with MHC class I and II machinery as well as PDCD1 to prevent additional inhibition of CAR T-cell effector function [103, 104]. Base editors have also shown promising therapeutic potential for enhancing CAR T-cells for treatment of T-cell acute lymphoblastic leukemia (T-ALL). The CAR antigens for T-ALL are typical pan T-cell markers such as CD7 and CD3, and anti-CD7 CAR T-cells for T-ALL have already been proven safe and efficacious in Phase I clinical trials [105]. Introduction of premature STOP codons into the CD7 and CD3 loci of anti-CD7 and anti-CD3 CAR T-cells prevents the induction of CAR-mediated fratricide during the manufacturing and infusion process [106]. dCas9 tethered to transcriptional repressor KRAB targeting the PDCD1 gene is effective in generating anti-HER2 CAR T-cells with low PD-1 expression to confer checkpoint inhibition resistance [107]. Prime editors, while still in their infancy, can efficiently edit the genomes of human cell lines and human primary T-cells [108]. The search-and-replace nature of prime editing also provides a potentially more adaptable method to simultaneously introducing additional complex gene edits into T-cells (e.g., certain gain of function mutations) that can enhance CAR T-cell therapy, but are challenging to achieve with endonuclease-based gene or base editing approaches (Fig. 4c).

The challenges of producing a universal CAR T-cell product from finite numbers of mature, fully differentiated healthy donor T-cells are compounded by the anticipated engineering demands, particularly for solid tumors in which multiple CRISPR-Cas9-based gene edits are needed to be combined with CAR-based recognition to achieve safety and efficacy. For example, in the setting of allogeneic T-cell therapies involving checkpoint blockade, expression of a transgene encoding a synthetic antigen receptor is required, in addition to knockout of several inhibitory molecules (e.g., PD-1, CTLA-4, TIM-3, LAG-3) as well as T-cell receptor (TCR) alpha and beta constant regions (TRAC and TRBC). As these therapies require several genetic edits within mature T-cells, inefficiencies and combinatorial stochasticity in production result in a final product that typically contains many different populations of cells with variable editing combination efficiencies that need to be evaluated and optimized to increase efficacy [103].

To specifically address cell source material to generate allogeneic products, CAR T-cell therapeutics can be derived from inducible pluripotent stem cells (iPSCs) amenable to genetic modification and differentiation into mature tumor-targeted T-cells. iPSCs are obtained from adult somatic cells by inducing expression of a combination of transcription factors (c-MYC, OCT3/4, SOX2 and KLF4) that transform cells into a pluripotent state [109, 110] (Fig. 4c). Pluripotent stem cells can give rise to the three embryonic germ layers, rendering them capable of differentiating into any specified cell type, with lymphoid lineages stemming from mesoderm progenitors [111, 112]. Because iPSCs are stem-like cells, they can be cultured unlimitedly and indefinitely in vitro under proper conditions which allow for a continuous production of T-cells [111]. Proof-of-concept studies have demonstrated successful generation of CAR T-cells from T lymphocyte-derived iPSCs [113]. However, the use of ‘feeder-free’ culture systems will likely be necessary for clinical-grade scalability and production [114]. Ideal conditions to achieve differentiation of mature T-lymphocytes involve incorporation of thymic niche promoting Notch/TCR signaling, as well as soluble factors such as IL-7, FMS-like tyrosine kinase 3 ligand (FLT3), stem cell factor (SCF), and the chemokines CXCL12 and CCL25 [111, 114].

Because downstream differentiated T-cells are derived from the same iPSC clone, CRISPR-Cas9 editing of iPSCs permits stable and consistent knockdown of factors such as the endogenous TCR and major histocompatibility complex (MHC) genes as well as genes encoding inhibitory molecules such as PD-1 to create unique pools of allogenic T-cell products with superior antitumor functionality, as has been previously seen in edited primary T-cells directed against glioblastoma and acute lymphoblastic leukemia in vivo models [90, 115]. Ideally, this will lead to a bank of selected iPSC lines with a consistent and optimally potent downstream differentiated final product for off-the-shelf therapies (Fig. 4c). The combination of precision iPSC programming using CRISPR-Cas9 and CAR technology sets the stage for the emergence of a new class of renewable and highly controllable off-the-shelf T-cell therapeutics.