Natural killer (NK) cells, a type of innate lymphoid cell (ILC), are crucial for host defense by providing rapid cytotoxic responses through germline-encoded receptors rather than peptide–MHC recognition like CD8⁺ T cells (Chiossone et al., 2018). Originating from CD34⁺ hematopoietic stem cells in the bone marrow, NK cells mature in secondary lymphoid tissues and undergo a licensing process in the thymus (Freud and Caligiuri, 2006; Stokic-Trtica et al., 2020). Their migration and tissue distribution depend on chemokines, integrins, and selectins: CX3CR1 and CXCR4 regulate bone marrow exit, while CCR7-CCL19 directs organ-specific homing (Pesce et al., 2016; Ran et al., 2022). This enables NK cells to monitor multiple tissues, including the liver, lungs, intestine, uterus, and CNS (Moroso et al., 2011; Vacca et al., 2013).

NK cell maturation is marked by morphological and phenotypic changes regulated chiefly by transcription factors EOMES and T-bet, influencing cytotoxicity and tissue residency (Abel et al., 2018; Kiekens et al., 2021). Cytokines like IL-12, IL-15, and IL-18 promote maturation and memory-like traits with stronger responses upon restimulation (Romee et al., 2012). Single-cell transcriptomics show diverse NK phenotypes, including terminally differentiated, memory-associated subsets (Smith et al., 2020). NK cells are identified by CD56 expression. Functional subsets are classified mainly by CD56 and CD16 surface density: CD56^brightCD16^dim/⁻ cells secrete cytokines and reside in tissues, while mature, lytic CD56^dimCD16⁺ cells are predominant in blood (Hu et al., 2019; McErlean and McCarthy, 2024). Though considered less mature, CD56^bright NKs express NKG2 receptors and proliferate in response to IL-2 and IL-15. CD56^dim NKs that express KIRs (killer immunoglobulin-like receptors) are highly cytolytic. CD57 expression marks terminal differentiation and immunologic memory capability (Stabile et al., 2017; Liu et al., 2021).

NK cells, unlike classic lymphocytes, use somatically encoded receptors for activation and cytotoxicity rather than receptor recombination (Marin et al., 2024). They detect tumor cells with reduced MHC expression and act through cytokine secretion and direct killing mechanisms like ADCC (antibody dependent cell cytotoxicity), Fas-mediated lysis, and TRAIL-mediated apoptosis (Smyth et al., 2001). NK cells also produce cytokines such as IFN-γ (interferon-gamma) and TNF-α (tumor necrosis factor alpha), promoting inflammation and activating other immune cells (Vivier et al., 2011; Fauriat et al., 2010; Wang et al., 2012). CD16 enables recognition of IgG-coated targets, triggering ITAMs and intracellular signaling for ADCC (Biassoni and Malnati, 2018). Signaling leads to degranulation and apoptosis in target cells via perforin, granulysin, and granzymes (Prager and Watzl, 2019). The diversity of NK cell functions is linked to their maturation, phenotype, and location, offering insights into overcoming tumor immune evasion (Capuano et al., 2021).

NK-cell research has shifted from basic immunobiology to translational engineering, focusing on their pharmacological profile as living drugs. Understanding NK cell maturation, activation, and cytokine needs is crucial for optimizing design and manufacturing. This review links biological mechanisms to next-generation engineering decisions that drive CAR-NK clinical outcomes.

Pre-existing antigen-negative tumor subclones (e.g., CD19⁻ or CD22⁻ malignant B cells), or the loss of a target antigen under CAR-T selection pressure, can lead to therapeutic resistance. A retrospective analysis of 628 B-ALL patients found that approximately 17% harbored CD19-negative and 22% had CD22-negative leukemic clones prior to CAR-T therapy (Rosenthal et al., 2018; Lin et al., 2024). This indicates that antigen-loss variants can exist at baseline and, upon CAR-T treatment, these antigen-negative cells can outgrow the targeted tumor population, causing relapse via immune escape. CAR-NK cells, while also susceptible to this escape, however, retain additional innate cytotoxicity pathways that can kill tumor cells beyond CAR recognition. This means CAR-NK cells are capable of partially eliminating antigen-negative tumor cells via natural NK receptors, an advantage over conventional, antigen-restricted, CAR-T cells (Jørgensen et al., 2025). Nevertheless, complete loss of the target antigen poses a major challenge, and preventing the escape of antigen-negative clones remains a priority in CAR-NK engineering.

Another cause of CAR-T ineffectiveness is genetic alteration of the target antigen itself. Mutations or alternative mRNA splicing in the antigen’s gene can downregulate its expression or generate variant antigens, effectively “shielding” the tumor from CAR binding (Lin et al., 2024). In one study of 12 B-ALL patients who relapsed after anti-CD19 CAR-T therapy, each patient had a unique insertion or deletion in the CD19 gene (most frequently affecting exons 2–5). These mutations led to truncated CD19 proteins missing the CAR recognition epitope, thereby enabling immune evasion. Similarly, alternatively spliced CD19 transcripts that skip exons (for instance, loss of exon 2) result in a truncated protein or an isoform lacking the CAR-binding domain, rendering the CAR-T cells ineffective (Fischer et al., 2017; Orlando et al., 2018; Bagashev et al., 2018).

Defects in the intracellular processing or presentation of the target antigen can result in insufficient surface expression for effective CAR-T cells detection. For example, loss of critical chaperone proteins or disruptions in antigen trafficking may prevent the antigen from reaching the cell surface. Tumor cells may also evade CAR-T cells by internalizing the target receptor or sequestering it in subcellular compartments, effectively “hiding” the antigen from immune surveillance (Lin et al., 2024).

Lineage switching is a rare but striking mechanism of immune escape wherein the tumor changes its cellular lineage identity. This phenomenon, observed mostly in B-ALL treated with CD19-targeted CAR-T cells or monoclonal antibodies, involves leukemic cells shifting from a lymphoid lineage to myeloid lineage. Such a transition can occur via transcriptional reprogramming or the emergence of a distinct leukemic subclone (Kurzer and Weinberg, 2022; Coorens et al., 2023; Lin et al., 2024). As a result, CD19⁺ B-ALL can relapse as CD19⁻ acute myeloid leukemia (AML), escaping CAR-T recognition. In fact, several cases of lineage switch have been reported post-CAR therapy: chronic lymphocytic leukemia transforming into plasmablastic lymphoma (Evans et al., 2015), mantle cell lymphoma transforming into histiocytic sarcoma (Zhang et al., 2020), and T-ALL transforming into AML after CD7 CAR-T treatment (Aldoss et al., 2023). This extreme plasticity allows tumors to evade lineage-specific therapies (like anti-CD19 CAR-T) by evolving into different tumors.

Although uncommon, epitope masking is a particularly insidious mechanism of CAR-T failure. In a case reported by Ruella et al. (2018), an accidental transduction of a leukemic B cell with the CD19 CAR construct caused the tumor cell to express the CAR on its own surface. The CAR on the tumor bound to its own CD19 antigen, effectively concealing CD19 from therapeutic CAR-T cells. This self-masking of the target antigen rendered the CAR-T treatment ineffective. Consequently, strict quality control during CAR-T manufacturing is essential to prevent accidental tumor cell transduction. Moreover, the use of allogeneic “off-the-shelf” CAR-T or CAR-NK cell products (derived from healthy donors) could avert this complication entirely, since tumor cells would not be present in the engineered cell product (Ruella et al., 2018; Lin et al., 2024).

Trogocytosis, a process of membrane exchange between cells, can also drive antigen escape. CAR-T cells can strip target antigen from tumor cells by trogocytosis, reducing antigen density on the tumor cells and thereby building resistance to further CAR-T engagement. This phenomenon contributes to CAR-T cell exhaustion and cellular fratricide, as they may present acquired tumor antigens on their own surface. In some situations, trogocytosis can also, paradoxically, enhance tumor cell survival and migration by transferring certain immune molecules to tumor cells, allowing them to acquire immune-like features and better adapt within the tumor microenvironment (Hamieh et al., 2019; Lin et al., 2024). An in-depth analysis of trogocytosis-related effects is explored in subsequent sections of this review.

While the above escape mechanisms have been well-characterized in the context of CAR-T therapy, similar challenges may arise, to a lesser extent, with other adoptive cell therapies. CAR-NK cells share some vulnerabilities with CAR-T cells, especially when tumors evade by modulating the target antigen. Tumor-intrinsic resistance, coupled with limitations of CAR designs, can significantly impede the efficacy of these therapies. For example, the high genomic instability of cancer cells means that antigen downregulation could diminish CAR-NK cell effectiveness (Zhong and Liu, 2024).

Collectively, these immune evasion mechanisms not only drive resistance to CAR-T therapy but also provide critical insights for next-generation CAR-NK development. NK cells can recognize targets independently of MHC class I, making them less vulnerable to the loss of MHC-I or certain antigen-downregulation strategies. However, even CAR-NKs could be hindered by trogocytosis-mediated antigen loss, which reduces target density on tumor cells (Abel et al., 2018). Therefore, emerging CAR-NK design optimizations are being geared toward overcoming these escape mechanisms, as discussed next.

To address antigen loss and tumor escape, researchers are developing several CAR engineering strategies (many inspired by advancements in CAR-T cells). These strategies aim to prevent or circumvent tumor antigen escape and improve the durability of responses.

The ectodomain is made up of the single-chain fragment variant (scFv), which consists of a linker protein that unites a heavy and light chain and, also, of a hinge region, anchoring the ensemble to the transmembrane domain. This structure is designed specifically for cognate antigen recognition. Since different scFv are capable of binding different epitopes in the same protein, this domain can determine both the specificity and the function of the CAR-NK cell (Haso et al., 2013). Since the scFv is synthetic in nature, antigen specificity can be affected due to the changed connectivity of the VL (variable light chain) and VH (variable heavy chain) domains. Computational-assisted design of the scFv aids in configuring functional structures by assembling the amino acid sequence of the CDRs (complementary-determining regions) and making precision target engagement more effective (Thokala et al., 2016; Krokhotin et al., 2019).

The transmembrane (TM) region acts like an anchor for the CAR against the cell wall and further connects to the endodomain, the structure responsible for generating intracellular signaling (Gong et al., 2021). Therefore, the TM domain is a critical component due to its role in influencing the function and the activation potential of the CAR construct.

For the endodomain, recent CAR models utilize the CD3ζ chain signaling domain, which has 3 ITAMs (immunoreceptor tyrosine activation motifs) per CAR. These ITAMs, in turn, activate the Syk or ZAP70 tyrosine kinases and PI3-kinase signaling (Orr and Lanier, 2010). In other instances, signaling domains derived from NK-specific activating receptors (CD28, CD16, NKp44, NKp46, NKG2D, DNAM-1 and 2B4) have been used in NK-92 cell lines to enhance toxicity and optimize signaling pathways (Li et al., 2018). 2B4 is a co-stimulatory domain that is known for its role in NK cell anti-tumor effects through improvement of cytotoxic activity and cytokine release when compared to the typical 4-1BB counterpart (Xu et al., 2019). A signaling adaptor molecule, DAP12 (DNAX-activation protein 12), was also associated with greater anti-cancer roles compared to the more traditional CD3ζ (Imai et al., 2005). A study conducted by Ye Li et al. analyzed CAR constructs that aim to specifically enhance NK cell potency in vivo and in vitro. Their conclusions established that sCFv-NKG2D-2B4-CD3ζ (NK-CAR4) boasted superior cytotoxicity, expansion and persistence capabilities than their CAR-T homologue with the scFv-CD28-4-1BB-CD3ζ (T-CAR structure) (Li et al., 2018). Moreover, DAP10 and DAP12 molecules, signaling adaptor proteins involved in activation of Syk-vav1-Erk and NF-kB pathways, demonstrated superior roles in NKG2C and NKG2D receptor activation compared to CD3ζ (Topfer et al., 2015; Wang et al., 2020). New 3^rd and 4^th generation CAR-NK constructs aim to enhance the tumor penetration and function of these cells in the immunosuppressive microenvironment.

More recent work has addressed NK specific signaling strategies. One approach adapts DNAM-1 (CD226)-based chimeric receptors, which can stabilize the immune synapse and enhance cytotoxic function when used as CAR signaling modules (Wu et al., 2015; Cifaldi et al., 2023; Focaccetti et al., 2022). Other approaches focus on unique activating receptors, such as NKp46 or NKG2D, and their appropriate adaptor pathways (DAP10/DAP12 or FcRγ), to better mirror native NK activation and limit tonic signaling (Wang et al., 2022; Zhang et al., 2024; Peng et al., 2024). Recent work has further advanced NK-specific CAR design through composite signaling domains that better recapitulate native NK activation circuitry. DNAM-1-CD3ζ chimeric receptors in human NK cells boost recognition and killing of PVR/Nectin-2⁺ solid tumors, an effect enhanced by Nutlin-3a–induced upregulation of DNAM-1 ligands (Focaccetti et al., 2022; Cifaldi et al., 2023; Cifaldi et al., 2024). In parallel, other CAR platforms are built to utilize NCR such as NKp30 and NKp46 fused to adaptor modules (CD3ζ, DAP10 or DAP12), which in preclinical models increase cytokine production and tumor lysis against ALL, ovarian carcinoma, osteosarcoma, prostate carcinoma and rhabdomyosarcoma and more closely mimic native NK signaling (Cifaldi et al., 2024; Hu et al., 2018; Yu et al., 2022; Wang et al., 2024). Together with newer multi-component endodomains that integrate NKG2D/DAP10, 2B4 and CD3ζ signaling to enhance activation (Hu et al., 2018; Yu et al., 2022; Yi et al., 2025), these architectures exemplify the rapid evolution of NK-tailored CAR backbones over the past few years.

Collectively, these data support designing CARs that signal through NK-preferential pathways rather than directly importing T cell backbones, in order to boost efficiency. We illustrated how the structural principles and CAR backbones evolved when adapted to T-cell versus NK-cell biology in a schematic comparison of CAR-T and CAR-NK architectures and their respective effector mechanisms (Figure 1).

Out of all available sources, peripheral blood (PB) derived NK cells are the easiest to obtain, although their use is restricted by their low transduction potential and inefficient expansion (Szmania et al., 2015).

UCB-derived NK cells have been proven to generate greater proliferative capabilities (Della Chiesa et al., 2012). However, given their immature nature, their cytotoxic potential is significantly diminished compared to other sources. Compared to PB-NKs, UCB-NK cells express CD56^bright, high density of NCR receptors and NKG2D, which are mostly involved in cytokine secretion. By contrast, adhesion molecules and receptors associated with cytotoxic killing, such as CD16, KIRs, DNAM-1, NKG2C, IL-2R and CD57 are expressed at lower levels (Tanaka et al., 2003; Luevano et al., 2012).

HSCs allow large numbers of NKs to be collected and are more permissive to engineering enhancements (Zeng et al., 2017). CD34⁺ HSCs from the bone marrow (BM) and UCB can be used to generate NK cells, showing remarkable functionality, similar to PB-NKs in regard to cytokine generation, cytotoxic capabilities and activating potential (Luevano et al., 2014).

There are multiple possibilities of harnessing NK cells for “off the shelf” use in CAR manufacturing, such as utilizing various NK cell lines or by apheresis of iPSC-NKs (induced pluripotent stem cell-derived NK cells). In a comprehensive review, Wei et al. proposes that iPSC-derived NK cells can be generated with high clonal uniformity, a standardized manufacturing approach and enhanced genetic engineering to bolster cytotoxicity and ADCC potential. Progress in this direction was made possible by introducing 3D embryoid-body/spin-EB methods for generating CD34⁺ hematopoietic progenitors and then mature NK cells. However, challenges for successful iPSC-NK translation remain, as improving differentiation efficiency and reproducibility, gene-editing safety (off-target risks), and enhancing in-vivo trafficking are still elements that solicit finer tuning (Wei et al., 2025).

There is still the possibility of utilizing cell lines, such as YT or NK-92, to engineer CAR-NKs with the desired characteristics. These cell lines represent a potentially readily available and abundant source of NK cells for immunotherapy, especially due to their ability to retain cytotoxic potential during the transduction process (Maki et al., 2001; Cheng et al., 2012). However, there are still important safety challenges when generating CAR-NKs from tumor cell lines and, often, irradiation is mandatory, which significantly affects their persistence in the host (Klingemann et al., 1996; Tang et al., 2018). Feeder cell lines can be used to expand NK cells ex vivo. For instance, K562 is a cell line that is MHC negative and is engineered with the purpose of generating IL-15 and IL-21 cytokines to expand and mature NK cultures (Tonn et al., 2013).

Importantly, while each NK cell source differs in cytotoxic potency, expansion kinetics, and suitability for “off-the-shelf” production, the efficiency and stability of CAR expression are determined by the gene delivery and manufacturing platform rather than the cellular origin itself (Morgan et al., 2021).

The translation of CAR-NK therapies from experimental stages to widespread clinical use requires rigorous adherence to Good Manufacturing Practice (GMP) standards governing cell handling and genetic modification to expansion, formulation, and cryopreservation. Unlike autologous CAR-T modalities, which require patient-specific manufacturing, CAR-NK therapies benefit from the feasibility of batch-based production from universal donor sources or iPSC-derived NK populations. This model enables a single manufacturing run to generate large numbers of clinical doses, with lower patient variability, shorter production time, and improves cost efficiency (Rezvani et al., 2017).

Viral vectors (lentiviruses and retroviruses) remain widely used for CAR gene integration due to their stable expression, but they come with manufacturing complexity and cost due to extensive biosafety measures needed to mitigate the risk of insertional mutagenesis (Milone and O’Doherty, 2018). Therefore, interest in alternative engineering methods has grown, particularly as NK cells present unique challenges in viral transduction.

GMP expansion of NK cells depends on regulated cytokine support, such as IL-2 or IL-15, delivered alongside feeder-cell systems. Automated and closed-system bioreactors have markedly improved scalability and manufacturing reproducibility, enabling the generation sufficient NK cells per batch for allogeneic CAR-NK platforms (Jahan et al., 2024).

Fourth-generation armored CAR-NK cells are engineered to co-express self-sustaining survival signals, such as IL-15, so that they maintain proliferation, function and persistence without requiring exogenous cytokine administration.

Distinctions between IL-15–armored CAR-NK and IL-15–engineered CAR-T cells are important to note. In CAR-NK products, IL-15 primarily enhances NK-cell survival, metabolic fitness, and serial cytotoxicity through IL-15Rβ/γ signaling without provoking excessive cytokine-release toxicity, reflecting the innate regulatory checkpoints of NK cells (Christodoulou et al., 2021). By contrast, IL-15-expressing CAR-T products activate potent autocrine IL-2/IL-15 receptor pathways that can drive uncontrolled proliferation, high systemic cytokine levels, and increased risks of CRS, often requiring additional safety circuits or switch-based architecture (Hurton et al., 2016; Li et al., 2025).

Early proof-of-concept studies demonstrated that expression of membrane-bound IL-15 supports autonomous NK expansion and enhances cytotoxicity, providing a strong argument for incorporating IL-15 into efficient CAR-NK designs (Imamura et al., 2014; Liu et al., 2018). From a pharmacological perspective, these constructs reduce the need for systemic cytokine dosing, which, in turn, limits off-target immune activation and repeated infusions. IL-15 withdrawal leads to rapid NK-cell apoptosis, therefore, IL-15–armored CAR-NK cells typically achieve prolonged persistence with a low incidence of secondary toxicity (Vivier et al., 2011).

The key ability to destroy tumor cells without MHC restriction makes NK cells extremely versatile immune effectors and perfectly suited for adoptive cell therapies (Ruggeri et al., 2002; Miller et al., 2005). Furthermore, CAR-NKs lyse tumor cells, promoting apoptosis, through FasL, TRAIL, perforins/granzymes pathways and cooperation with T cells, macrophages and dendritic cells (Screpanti et al., 2001; Vivier et al., 2011). These multiple killing mechanisms complimenting the CAR interaction support broad antitumor activity and lower the risk antigen escape mechanisms (Farag and Caligiuri, 2006; Sotillo et al., 2015) and contribute to the generally favorable safety profile of NK cell therapies (Simonetta et al., 2017).

CAR engineering is continuously advancing to enhance receptor affinity while improving tumor selectivity, aiming to increase efficacy, as well as reduce off-target effects and antigen escape (Rodriguez-Garcia et al., 2020). One of the components that can be selectively improved is the scFv extracellular fragment. Researchers are developing dual or multi-targeting scFVs, which aim to enhance CAR antigen binding by either engaging two different epitopes on the same target, or by recognizing multiple antigens on the tumor cell (Tahmasebi et al., 2021).

Moreover, unconventional scFv fragments are currently being developed, such as nanobody derived single domain variable heavy chain (Safarzadeh Kozani et al., 2022) or fully human heavy-chain-only variable domain (FHVH), which would allow superior expression, stability and safety of smaller CAR constructs with considerably less immunogenicity than standard constructs (Lam et al., 2020). Another notion that is being developed is that of switchable CARs, where instead of directly binding tumor antigens, they target intermediary molecules—such as antibody fragments or adaptor proteins like zipFv—that, in turn, recognize the tumor. This personalized “switch” allows for precise, tunable control of the CARs specificity and function (Raj et al., 2019; Qi et al., 2020; Cao et al., 2021).

Integrating viral vectors remain a mainstay for CAR insertion in NK cells. Third-generation lentiviral and retroviral vectors support stable genomic integration and long-term CAR expression in both primary NK cells and NK-92 cell lines, building on decades of experience in gene therapy and CAR-T manufacturing (Tomanin and Scarpa, 2004; Milone and O’Doherty, 2018).

Although these platforms are thoroughly characterized from a manufacturing perspective and comply with GMP workflows in CAR-Ts, they pose unique challenges for NK cells. Primary NK cells are notably refractory to viral transduction. Compared with T cells, they express higher levels of PRRs (pattern-recognition receptors) and antiviral sensors, including TLR3, RIG-I, MDA5 and downstream MAVS-dependent pathways, which detect viral RNA and vector components (Littwitz et al., 2013; Schmidt et al., 2021; Robbins et al., 2021; Afolabi et al., 2019). Activation of these pathways reduces transduction efficiency and can trigger apoptosis or functional impairment. To further elevate transduction efficiency, PDK1 inhibitors against RIG-I-like and Toll-like receptors were used in some instances (Clark et al., 2009; Sutlu et al., 2012), even though multiple transduction processes were required. Optimization of activation status and culture conditions, vector pseudotyping, transient modulation of innate signaling are strategies that have improved NK transduction in specific settings, but more universally efficient protocols still remain in development (Sutlu et al., 2012; Schmidt et al., 2022). Cost and regulatory complexity are additional constraints. Lentiviral vector manufacture is a major cost driver in CAR-T programs, and similar considerations apply to CAR-NK models. Large-scale vector production requires specialized facilities, extensive testing and long lead times, all of which can overturn the need for cost-effective deployment of allogeneic products (Milone & O’Doherty, 2018).

Although viral vectors remain indispensable for many current products, their limitations have sparked rising interest in non-viral engineering platforms for future CAR-NK products.

Non-viral engineering platforms are promising for CAR-NK manufacturing, especially for scalable, cost-effective, and safe allogeneic applications. Transposon systems such as Sleeping Beauty and piggyBac enable stable genomic integration of large genetic material using plasmid DNA, avoiding variability and insertional mutagenesis risks associated with high-titer viral vectors (Rostovskaya et al., 2012; Matosevic, 2018; Robbins et al., 2021). piggyBac-engineered CAR-NK cells have demonstrated efficient gene transfer, robust cytotoxicity, and favorable manufacturing characteristics in preclinical models, including CAR NK-92 cells expressing NKG2D and primary CAR-NK products targeting CD73⁺ solid tumors (Wang et al., 2018; Li et al., 2020; Du et al., 2021).

Electroporation-based approaches provide a flexible, virus-free route to deliver DNA, mRNA, or ribonucleoprotein complexes. Ingegnere et al. developed a procedure that can efficiently introduce CAR and CCR7 genes via an electroporation-based plasmid DNA transfection in both human NK cell and in NK-92 lines, demonstrating that optimized plasmid DNA electroporation protocols now achieve superior transfection efficiencies in IL-2–expanded NK cells. However, short-term viability is still a concern, reportedly reaching only 50%–60% after transfection in carefully tuned systems (Ingegnere et al., 2019; Heintz and Gong, 2020; Schmidt et al., 2021; Wang et al., 2022). Multi-gene constructs delivered through DNA insertion methods allow enhanced expansion and persistence compared to mRNA-based strategies. However, harsher nuclear delivery conditions and DNA-sensing pathway activation can still affect product consistency and potency.

mRNA electroporation achieves transient CAR expression without genomic integration, making it particularly useful for early-phase, dose-escalation studies and for indications where reversible activity is desirable. Several groups have demonstrated efficient CAR-mRNA delivery into both primary NK cells and NK-92, with expression lasting a few days and maintaining potent short-term cytotoxicity (Schmidt et al., 2022; Laskowski et al., 2022). Transient mRNA CAR-NK products are more suitable to function as a built-in safety layer, enabling careful titration of exposure while platforms are optimized.

Because “off-the-shelf” CAR-NK products are allogeneic, they are inherently susceptible to host-versus-graft immune rejection driven by residual recipient T cells, NK cells, and macrophages, which can limit their persistence. In addition to gene delivery and construct stability, non-viral platforms are progressively integrating immune-editing approaches to improve the persistence of allogeneic CAR-NK cells. Recent studies show that selectively knocking down classical HLA-A/B/C and adding PD-L1 and/or HLA-E expression can prevent rejection by host T and NK cells while maintaining self-tolerance, allowing for prolonged engraftment and antitumor effects (Liu et al., 2025). A 2025 review paper highlights complementary non-viral approaches such as β2-microglobulin knockout, HLA-E overexpression, and CD47 upregulation to inhibit phagocytosis, all of which can be implemented through non-viral editing strategies (Kim, 2025).

Viral and non-viral vectors vary in efficiency, durability, and safety. Lentiviral and retroviral vectors achieve high CAR insertion rates in NK cells (∼50%–90%) and stable long-term expression, but they pose insertional mutagenesis risks and increase manufacturing complexity and costs (Schmidt et al., 2021; Dan and Kang-Zheng, 2025). Non-viral platforms like transposons (piggyBac, Sleeping Beauty) and mRNA electroporation yield variable engineering efficiencies (about 30%–70%, depending on NK source, activation, and electroporation conditions), but provide lower genotoxicity, faster production, and better scalability for allogeneic products (Ingegnere et al., 2019; Du et al., 2021; Bexte et al., 2024; McErlean and McCarthy, 2024). In comparison to viral integration, mRNA-based approaches result in transient CAR expression lasting several days, which improves safety profiles but restricts cell persistence. By contrast, transposon systems enable stable gene integration and offer lower vector-related complexity and cost (Boissel et al., 2009; Schmidt et al., 2022; Maia et al., 2024). Together, these distinctions inform platform selection for next-generation CAR-NK manufacturing and highlight complementary trade-offs in efficiency, safety, and clinical applicability.

To fully realize the promise of allogeneic CAR-NK products, advances in genetic engineering must be matched by high-end manufacturing technologies. Closed, automated bioreactor systems now enable optimal expansions of PB, UCB, or iPSC-derived NK cells in clinical settings, while maintaining phenotypic stability and cytotoxic function (Zhang et al., 2022; Moscarelli et al., 2022). The automation process reduces operator variability, lowers contamination risk and integrates well with gene-modification workflows, making them well suited for large quantity allogeneic products. Current protocols can generate around 10⁹–10¹¹ NK cells per batch, sufficient to supply multiple patients from a single run, particularly when combined with optimized cryopreservation strategies (Ma et al., 2025).

Switchable, universal and adapter-based CAR architectures allow real-time modulation of NK cytotoxicity and targeting. Peer-reviewed studies in CAR-T systems (e.g., SUPRA CARs; inducible ON-switch CARs) demonstrate feasibility, and preclinical CAR-NK adaptations are emerging. First, switchable CAR systems provide external control over NK-cell activity: “ON-switch” CARs activate only when a harmless small molecule is present, while “OFF-switch” circuits (e.g., iCasp9) allow rapid shutdown of the therapy in case of toxicity. This reversible control enhances safety and reduces off-tumor effects (Zhao et al., 2023). Second, universal CAR platforms expand targeting flexibility by decoupling antigen recognition from NK-cell activation. These systems use adapter molecules or modular components (anti-FITC CARs, SUPRA CARs) that separate recognition and signaling domains. By exchanging adapters or modules, a single CAR-NK product can be redirected to multiple tumor antigens or tuned in activity (Zhao et al., 2023; Amoozgar et al., 2025).

Future development should prioritize Boolean-logic gates and adapter systems that specifically counter antigen heterogeneity and antigen-loss mechanisms as highlighted in Sections 3.1–3.3.

Dual-input AND-gate circuits, OR-gated multispecific receptors, and inhibitory NOT-gate systems show early clinical and preclinical success in enhancing selectivity and reducing off-tumor toxicity. Boolean logic–gated designs that integrate multiple antigens improve tumor specificity and reduce off-target toxicity. These CAR-NK cells can be programmed with AND, OR, and NOT gates, enabling activation only under precise antigen combinations (Zhao et al., 2023). SENTI-202 is an innovative CAR-NK therapy targeting AML. It uses two activating CARs (CD33 and FLT3) as an OR-gate to destroy leukemia cells with either antigen, countering tumor diversity. An inhibitory CAR detects EMCN, a marker on healthy stem cells, acting as a NOT-gate to prevent NK activation and protect normal bone marrow (Dos Reis et al., 2025).

Beyond SENTI-202, other types of synNotch receptors enable sequential AND-gates. These multi-input circuits requiring antigens A AND B, or A AND NOT B allow far more refined discrimination between malignant and healthy tissues than single-antigen CARs (Zhao et al., 2023; Dos Reis et al., 2025). Boolean logic gating represents a major step toward next-generation, highly selective CAR-NK immunotherapies that are particularly well-suited to counter lineage switching (Section 3.4) and mixed-antigen expression in AML and solid tumors.

iPSC-derived NK cells are emerging as a transformative platform for CAR-NK therapy, offering advantages that overcome the limitations of donor-derived NK sources. By expanding indefinitely and differentiating into NK cells at scale, iPSC-NKs provide a renewable, uniform, and batch-manufacturable source of effector cells (Amoozgar et al., 2025). Early clinical studies support their feasibility, as shown by FT596, an iPSC-derived CAR-NK engineered with IL-15 support, which demonstrated antitumor activity and a favorable safety profile in CD19⁺ lymphoma (Kim, 2025). Large-scale, cryopreservable manufacturing, clonal uniformity that reduces patient variability and the ability to generate “armored” or immunologically “stealthy” CAR-NK cells makes iPSC-NKs ideal for multiplex editing to counter host rejection (Section 3.7) and to install CARs resistant to immune evasion (Amoozgar et al., 2025; Kim, 2025).

iPSC-derived CAR-NKs represent a successful design that combines manufacturing scalability with ingenious engineering, accelerating development of next-generation CAR-NK therapies.

Multiplex genome editing is becoming central to building CAR-NK cells that can persist in allogeneic hosts and resist immune rejection. One strategy is knocking out β2M (β-2 microglobulin), which removes polymorphic HLA-A/B/C to prevent host T-cell recognition. To avoid triggering host NK “missing-self” responses, HLA-E or HLA-G are re-expressed, which subsequently engages NK inhibitory receptors and protects the infused cells (Kim, 2025). In addition to this mechanism, CD47 overexpression blocks macrophage phagocytosis, checkpoint deletions (PD-1, TIGIT) prevent tumor-mediated suppression, and CISH knockout boosts cytokine signaling and persistence in the tumor microenvironment (Amoozgar et al., 2025; Kim, 2025). These multiplexed edits can be combined to produce immune-evasive, long-lived CAR-NK cells.

Liu et al. (2025) developed NK cells with simultaneous HLA modulation, CAR expression, and checkpoint targeting, resulting in resistance to T- and NK-cell clearance, longer persistence, and high cytotoxicity with lower inflammatory cytokines. Early clinical trials of multiplex-edited CAR-NKs show promising safety without CRS or ICANS, but extensive edits may cause toxicity, signaling the need for careful engineering.

Overall, multiplex editing can enable universal, durable, and immune-evasive CAR-NK therapies designed for broad off-the-shelf application and could be a future staple in CAR design.

Trogocytosis remains a major driver of antigen-density loss in CAR therapies, leading to reduced target availability, impaired effector function, and CAR cell fratricide. Studies in CAR-T cells have demonstrated that high-affinity scFvs promote excessive antigen extraction, leading to excessive loss of surface targets and enabling rapid immune escape (Olson et al., 2022). Conversely, lower-affinity CAR constructs preserve antigen density, reduce fratricidal interactions, and maintain durable cytotoxicity while achieving efficient synapse formation (Olson et al., 2022). Although these data stem from CAR-T systems, emerging NK-specific CAR platforms are beginning to adopt similar affinity-tuning strategies. Lower-affinity CARs may therefore allow NK cells to retain natural cytotoxic pathways, reduce tonic signaling, and limit trogocytosis-mediated resistance described in Section 3.7. Future work should optimize affinity ranges specifically for NK cells, considering NK-synapse kinetics, serial-killing behavior, and the balance between CAR-mediated and innate-mediated activation.

The TME induces metabolic and cytokine-mediated suppression that dampens NK cell activity, driving the rationale for developing “armored” CAR-NK designs. IL-15 armoring boosts NK survival, proliferation, mitochondrial fitness, and serial killing without the toxicities seen in IL-15–enhanced CAR-T cells (Christodoulou et al., 2021). Chemokine receptor engineering (CXCR1, CXCR4, CCR7) improves trafficking and tumor infiltration both in hematological and solid tumors (Ng et al., 2019; Andreou et al., 2025). Inducing resistance to TGF-β, a major NK-suppressive cytokine, can be achieved through receptors or gene knockouts, preserving NKG2D expression, cytotoxicity, and proliferation (Christodoulou et al., 2021; Valeri et al., 2022). Other mechanisms involve suppressing the metabolic pathways, such as removing A2A receptor signaling or enhancing glycolysis to counter hypoxia and adenosine build-up (Valeri et al., 2022).

In practice, the most effective CAR-NK cells will likely combine IL-15 support, optimized chemokine homing and resistance to TGF-β or metabolic inhibition, ideally integrated at the iPSC stage for stable, uniform expression and scalable production (Andreou et al., 2025; Valeri et al., 2022).

This review has identified and summarized the primary advantages and disadvantages of these immunotherapies, which are outlined in Table 2.

With their unique ability to locate and destroy target cells, NK cells make ideal candidates for CAR augmentation. Notably, CAR-expressing NK cells possess a safer therapeutic profile than CAR-T cells in clinical settings, and multiple clinical trials have demonstrated that NK cell immunotherapy is a viable alternative to CAR-T therapy (Becker et al., 2016). For instance, phase I/II trials have shown that allogeneic NK cell administration is well tolerated and does not induce GVHD or other severe adverse events, highlighting NK cells as general CAR drivers independent of autologous cells (Sakamoto et al., 2015; Ciurea et al., 2017). Additionally, newer constructs incorporate iCasp9 (inducible caspase-9) “safety switch” (Figure 1), which triggers programmed death of CAR-NK cells in the event that severe toxicity occurs (Nowakowska et al., 2018; Liu et al., 2020). This safety switch is a necessary control mechanism as even stronger “armored” CAR-NK cells are being developed, which could also incur greater side effects (Włodarczyk and Pyrzynska, 2022).

A major drawback of CAR-T therapy is the persistent on-target/off-tumor effects, such as CD19 CAR-T cells causing prolonged B-cell aplasia. In contrast, the limited lifespan of CAR-NK cells in circulation reduces such risks. Another possible complication is trogocytosis. This is a process in which immune cells express membrane proteins “nibbled” from target cells, which further leads to depletion of target cells antigens and, also, to cellular fratricide between CAR-T products (Miao et al., 2021). In CAR-NK cells, however, there is limited available data on this phenomenon and extensive research is already being carried out. It was observed that trogocytosis-mediated signaling can induce different types of behavior in NK cells (Vanherberghen et al., 2004; Reed et al., 2021). In some cases, NK functionality can be enhanced through trogocytosis-mediated acquisition of chemokine receptors such as CCR5, CXCR4 or CCR7 (Marcenaro et al., 2009; Somanchi et al., 2012; Vo et al., 2022) which improves homing to lymph nodes (Somanchi et al., 2012) or TYRO3, which boosts proliferation properties and effector functions (Lu et al., 2021). Potential negative effects of trogocytosis include acquisition of immunosuppressive proteins such as PD-1 (Gonzalez et al., 2021; Hasim et al., 2022), which significantly reduce their cytotoxic potential. To counter these effects, antibodies blocking CD9 were used in vitro with varying degrees of success in restoring antitumor efficacy (Gonzalez et al., 2021).

Additionally, depletion of target antigens due to trogocytosis has been observed in both CAR-T and CAR-NK therapies (Hamieh et al., 2019; Li et al., 2022; Schoutrop et al., 2022; Camviel et al., 2022). Antigen density is critical for CAR cell function, therefore potential downregulation or loss through trogocytosis can be detrimental to the effectiveness of the therapy and facilitates tumor evasion. Strategies to mitigate this complication involve adjusting the affinity of the CAR for its cognate antigen. Recent papers highlight that lower affinity CAR constructs might be able to reduce the incidence of trogocytosis without compromising efficiency (Olson et al., 2022) and further optimizing this approach might contribute significantly to limiting antigen loss and, therefore, their overall performance and persistence (Singh et al., 2020).

With our current understanding of NK cell behavior, we know that they are less likely to show long term persistence as a shortened lifespan can affect overall efficiency in clinical trials (Bachanova et al., 2010; Shaffer et al., 2016; Nguyen et al., 2019). Much of the progress that has been made recently in improving NK therapy performance has been related to their cytotoxic potential. However, efforts are underway to increase NK persistence as well by engineering cells with immunostimulatory cytokines (Liu et al., 2018; Liu et al., 2020; Christodoulou et al., 2021; Du et al., 2021; Teng et al., 2022; Cichocki et al., 2022a; Cichocki et al., 2022b). Liu et al. introduced, in a clinical study, IL-15 engineered CD19-directed CAR-NK cells in relapsed/refractory hematological malignancies with moderate success (Liu et al., 2020). It was also demonstrated that exposure to IL-12, IL-15, IL-18 leads to promotion of memory NK cells, which are more persistent and efficient against tumor cells (Gang et al., 2020). This technique has proved itself in CAR-T therapies, where addition of immunomodulatory cytokines improved response, persistence and created armored CARs models (Yeku and Brentjens, 2016; Deng et al., 2020; Zhang et al., 2021). Recent evidence also suggests that chronic engagement of inhibitory KIRs contributes to functional exhaustion and reduced persistence of CAR-NK cells in vivo. This novel strategy proposes that inhibitory CAR (iCAR) targeting KIRs can convert an inhibitory signal into an activating one, thereby restoring metabolic fitness, enhancing proliferation, and improving tumor control in preclinical models (Li et al., 2022). Furthermore, addition of inducible promoters, which become active after recognizing tumor antigens or specific cellular signaling pathways can help increase functionality and safety of CAR therapies.

One of the most important factors that can alter NK cell persistence in patients undergoing treatment is the lymphodepleting regimen administered before cell infusion. To date, insights into the role of lymphodepletion in CAR-NK therapy have been based on CAR-T research, where this process proved its therapeutic efficacy (Amini et al., 2022). Typical regimens associate fludarabine and cyclophosphamide before CAR-NK infusion to prevent NK cell rejection and modulate the immunosuppressive tumor microenvironment (Xie et al., 2020). Additionally, a strategy utilizing a CD52-targeting monoclonal antibody in combination with a CD52-knockout CAR-T cell product has been explored and can be administered alongside cellular infusion (Caldwell et al., 2021), but, whether this strategy is applicable to CAR-NK cells, remains unknown so far. Currently, there are no clinical trials that directly compare lymphodepleting regimens in CAR-NK therapies, making this an area requiring further investigation. This need for research is even more pronounced when considering CAR-NK applications in solid tumors, where the value of lymphodepletion remains a topic of debate due to concerns that immune suppression may negatively impact endogenous anti-tumor immune responses.

Despite ongoing progress, CAR-NK therapies face several challenges that currently limit their durability and potential. First, in vivo persistence of CAR-NK cells is typically shorter than that observed with CAR-T cells, with detectable CAR-NK cells often declining within weeks in the absence of cytokine support (Liu et al., 2020; Romee et al., 2016). This is indicative of the intrinsic lifespan and homeostatic demands of NK cells, which may consequently limit the persistence of antitumor responses. Second, host immune clearance, including rejection by residual T cells, NK cells, and macrophages, can eliminate infused CAR-NK cells, particularly in allogeneic settings without adequate lymphodepletion or immunomodulation (Laskowski et al., 2022; Jørgensen et al., 2025). Third, NK cells show variable trafficking and infiltration into solid tumors, where dense stroma, hypoxia, and immunosuppressive cytokines such as TGF-β limit NK cytotoxicity (Andreou et al., 2025). Finally, large-scale, standardized manufacturing pipelines remain under development, and differences among NK sources contribute to variability in phenotype and functional potency (Cichocki et al., 2022a; Cichocki et al., 2022b).

Approaches to improve CAR-NK persistence include IL-15 armoring and induction of memory-like NK phenotypes described in preclinical models (Ma et al., 2022). To mitigate host-mediated clearance, current strategies under investigation in NK-cell therapy include the engineering of “do not eat me” signals such as CD47, overexpression of HLA-E, and gene editing to address mismatched HLA alleles (Laskowski et al., 2022). Enhancing solid tumor penetration may be achieved either by chemokine receptor engineering (CXCR4, CXCR1) to improve homing, or by altering the tumor microenvironment for improved immune infiltration (Andreou et al., 2025). Also, iPSC-derived CAR-NK platforms support reproducible, scalable, and standardized manufacturing for off-the-shelf application, helping to address variability in NK-cell source and to simplify production (Jørgensen et al., 2025).

To substantiate differences in clinical performance, we summarize commonly reported metrics from representative CAR-T and CAR-NK trials in Table 3, including overall survival, adverse event rates, expansion/persistence kinetics, and manufacturing costs. Values reflect early-phase CAR-NK studies and late-phase/approved CAR-T programs in B-cell malignancies; individual trials may deviate from these ranges.