While CAR-T cell therapy has demonstrated unprecedented response rates, not all patients benefit from it, and a significant percentage experience relapses (43). Antigen escape stands out as the most common cause of relapse in CD19-positive B-cell malignancies following CAR-T cell therapy. Published data indicate that CD19-negative relapse occurs in 7-25% of B-cell acute lymphoblastic leukemia (B-ALL) cases and approximately 30% of large B-cell lymphoma (LBCL) cases in patients treated with CD19 CAR therapy (4, 44–46). Mechanisms underlying CD19 loss have been extensively investigated, including CD19 gene mutations or splice variants, abnormal processing or trafficking of CD19 due to CD81 expression deficiency, and lineage marker switching from the lymphoid to the myeloid lineage (46, 47). Zah et al. developed a CD19-CD20 CAR and demonstrated its efficacy in preventing the spontaneous emergence of CD19-negative tumor cell variants in immune-deficient mice. In contrast to CD19, loss of B cell maturation antigen (BCMA) appears to be infrequent following anti-BCMA CAR-T cell therapy and has only been reported in a few studies (48). For instance, only 4% (3 out of 71) of patients in a phase II clinical trial using Idecabtagene Vicleucel for multiple myeloma (MM) treatment developed BCMA expression loss (12). Deletion and mutation of the biallelic TNFRSF17 gene encoding BCMA have been identified as the main mechanisms causing BCMA loss (49–51). To mitigate the risk of relapse and failed responses attributed to tumor antigen escape, it may be necessary to conduct more precise screening of the patient’s genetic spectrum before initiating a new CAR-T cell therapy (52).

CAR-T cell exhaustion, a significant factor contributing to CAR-T cell resistance, often leads to antigen-positive relapse and results from various factors. Firstly, the differentiation status of T cells is crucial for maintaining the functional persistence of CAR-T cells. Previous studies have shown that less differentiated T cells have greater expansion potential and prolonged persistence compared to fully differentiated effector T cells (53). Most current CAR-T cell products are autologous, and due to factors, such as the presence of tumors, prior cytotoxic treatments, and prolonged ex vivo cultivation, these cells often exhibit an exhausted phenotype characterized by excessive differentiation (54, 55). Additionally, immunosuppressive components of the tumor immune microenvironment, including regulatory T cells, myeloid-derived suppressor cells, tumor-associated macrophages (TAMs), and immunosuppressive ligands, also contribute to CAR-T cell exhaustion (25, 56, 57). Furthermore, tonic CAR signaling transduction, mainly associated with the costimulatory domain, plays a key role in CAR-T cell exhaustion. Insufficient signaling compromises cell persistence, while excessive signaling leads to exhaustion (58). For example, compared to 4-1BB, CD28 CAR-T cells demonstrate faster and stronger cellular effector functions, but this rapid and intense signal transduction can induce CAR-T cell exhaustion, thereby limiting persistence (59–61). Achieving potent and sustained activity against specific tumor targets requires careful selection of optimal CAR designs. Recent studies have underscored the importance of positively charged plaques (PCPs) on CAR in mediating CAR aggregation at the antigen-binding domain surface, thereby facilitating sustained CAR signaling (62, 63). Regulating PCPs offers a means to optimize CAR-T cell function. For CARs with high sustained signaling, such as GD2.CAR and CSPG4.CAR, reducing PCPs during in vitro expansion or enhancing ionic strength in culture can diminish spontaneous CAR activation and mitigate CAR-T cell exhaustion. Conversely, for CARs with weak tonic signaling like CD19.CAR, augmenting PCPs on the CAR surface can enhance in vivo persistence and anti-tumor efficacy (62). Recently, the same team developed CAR-Toner, an artificial intelligence (AI)-based PCP score calculator and optimizer (64). Taking the camel single-domain nanobody (VHH) targeting the acute myeloid leukemia (AML) tumor-associated antigen CLL1 as an example, the authors optimized the CAR design using CAR-Toner to systematically reduce their PCP scores. The results showed that an intermediate tonic signaling strength optimally benefits CAR-T cell function. As an AI-based tool, CAR-Toner is capable of not only conducting PCP calculations but also offering optimization recommendations for PCP scores. This groundbreaking tool is anticipated to catalyze progress in the field of CAR-T design and significantly contribute to the advancement of AI-driven CAR-T design.

Various strategies have emerged to combat CAR-T cell exhaustion, such as blocking exhaustion-promoting signaling pathways, inhibiting downstream effectors, alleviating immunosuppression within the tumor microenvironment (TME), and converting inhibitory signals into stimulatory ones (52, 61). When coupled with AI-based tools, these approaches facilitate the development of more potent CAR designs, thus augmenting the efficacy of CAR-T cell therapy (65).

Recent evidence indicates a concerning association between CAR-T cell therapy and the development of secondary T-cell malignancies, prompting regulatory measures by the FDA to enhance safety oversight (66, 67). By the close of 2023, 22 reported cases have documented the emergence of T-cell cancers subsequent to CAR-T treatment. Notably, in three instances, the CAR transgene was identified within the malignant clone, strongly implicating the therapy in the genesis of T-cell cancer (28). Presently, CAR-T products approved by regulatory agencies employ T cells engineered via viral transduction to convey the genetic construct. However, the use of current retroviral vectors still harbors potential risks of oncogenesis through genomic integration or related mechanisms. For example, lentiviral vector constructs, while integrating into the genome in a semi-random manner, display an affinity for genomic regions characterized by active gene expression, thereby heightening the risk of insertional oncogenesis (68). To mitigate these risks in the future, strategies such as precision targeting of CAR construct insertion to specific genomic loci or the utilization of transient CAR mRNA delivery to the cytoplasm offer promising avenues for enhancing safety in CAR-T therapy (68).

While CAR-T cell therapy holds tremendous promise in the therapeutic realm of hematologic malignancies, the associated potential life-threatening toxicities remain a significant concern. Cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS) emerge as the two most common adverse events during CAR-T cell therapy ( Table 1 ), attributed to the overactivation of CAR-T cells and the massive release of cytokines (69–71). CRS typically manifests a few days after the initial infusion, with symptoms ranging from mild flu-like manifestations such as fever, fatigue, chills, and muscle pain to severe life-threatening complications including shock, hypotension, coagulation abnormalities, and multi-organ dysfunction (72, 73). On the other hand, ICANS typically occurs 1-3 weeks post-CAR-T cell infusion, characterized by symptoms such as aphasia, delirium, focal neurological deficits, tremors, seizures, or life-threatening cerebral edema (74, 75). The severity of CRS and ICANS is categorized from grade I to grade IV, depending on factors such as CAR structure, CAR-T cell dosage, tumor burden, treatment targets, and patients’ individual characteristics. Most patients receiving CAR-T cell therapy are likely to experience varying degrees of CRS, with nearly half developing ICANS ( Table 1 ). Among these patients, 10%-40% may experience severe toxicity reactions (≥ grade 3), necessitating admission to the intensive care unit (ICU) for life support and the use of additional medications such as the interleukin (IL)-6 receptor antagonist tocilizumab (76–78). Blocking cytokine networks or optimizing the structural design of CARs to reduce toxicity are potential strategies to mitigate CRS and ICANS triggered by CAR-T cells (79, 80).

Given that most CAR-T cell target antigens are not exclusively tumor-specific and are also expressed in normal cells, CAR-T cells may inadvertently cause damage to normal tissue and organs while targeting tumors (81). Long-term follow-up data indicate that on-target off-tumor toxicity, leading to B-cell aplasia and hypogammaglobulinemia, are the most common long-term adverse reactions following treatment with anti-CD19 CAR-T cells (14). Although these side effects can be managed through sequential intravenous immunoglobulin, the long-term repeated infusions may escalate treatment costs. Additionally, cytopenia, infections, and tumor lysis syndrome are also common side effects of CAR-T cell therapy for hematologic malignancies. While proper management and intervention can control these toxicities, the resulting disease and economic burden are significant factors limiting the widespread adoption of CAR-T cell therapy (77, 82). Additionally, multiple-targeting CAR-T cells are designed to mitigate the impact of on-target off-tumor effects. Intensive research has centered on developing diverse protein-based logic-circuit strategies to enhance the specificity of CAR T cell activation and cytotoxicity towards tumor cells (83–85). Additionally, locoregional administration of CAR T cells, aimed at concentrating antitumor activity within the tumor microenvironment, may offer a potential solution to mitigate off-tumor toxicity (86–88).

A major challenge of CAR-T cell therapy is its high cost and lengthy manufacturing process. Most FDA-approved and clinical trial CAR-T cell products are autologous, requiring personalized customization involving complex processes like T cell isolation, activation, CAR gene transduction, expansion, and reinfusion (89). This labor-intensive process occurs in specialized facilities, adding to expenses and time. Viral vector preparation, crucial for CAR transfection, also contributes to costs and production delays (90). Production of viral vectors requires at least 2 weeks, meeting cGMP standards and extensive safety testing (91–93). The cost of a single dose of approved CAR-T cell products ranges from $373,000 to $475,000 ( Table 1 ), with a production cycle of 2 to 4 weeks ( Figure 2 ), posing economic burdens and logistical challenges for patients (94). With advancements in related technologies, it is imperative to establish increasingly standardized and simplified manufacturing processes to mitigate the costs associated with CAR-T therapy (22, 95).

Allogeneic CAR-T cells sourced from healthy donors offer advantages such as scalable production, lower manufacturing costs, and immediate availability (96). As illustrated in Figure 3 , allogeneic universal CAR-T cells represent off-the-shelf cell production. They are undergoing testing in numerous clinical trials, with promising outcomes documented in Table 2 . For instance, a meta-analysis showed CR rates of 70% in R/R ALL and 52% in non-Hodgkin lymphoma (NHL) with CD19 CAR-T cells derived from healthy donors (97). However, allogeneic CAR-T cells face challenges like T-cell receptor (TCR)-mediated graft-versus-host disease (GVHD) and human leukocyte antigen (HLA) related host-versus-graft (HvG) response due to their allogeneic nature (96). Targeted knockout of TCR, HLA, and related molecules using gene editing techniques is an effective strategy. Various gene editing tools have been explored, including zinc finger nucleases (ZFNs) (98), transcription activator-like effector nucleases (TALENs) (99), and CRISPR-Cas9 (100). CRISPR-Cas9 stands out for its versatility, allowing for simultaneous editing of multiple genes. It has been extensively used in allogeneic CAR-T cells, targeting TCR and HLA class I molecules, and guiding CAR insertion to specific loci, enhancing CAR-T cell potency and stability (101). However, CRISPR-Cas9 editing has limitations such as off-target editing and large-scale genomic rearrangements. Base editing technology, a precise gene editing tool based on CRISPR, offers new possibilities for overcoming these challenges without inducing double-stranded DNA breaks (DSBs) (102, 103). Adenine base editors (ABEs) and cytosine base editors (CBEs) enable the conversion of specific DNA bases without DSBs (104). Recent studies have shown the feasibility of highly specific knockout of T cell genes using base editing, paving the way for the development of allogeneic CAR-T cells with enhanced safety and efficacy (105). While allogeneic CAR-T cell therapy holds promise, ongoing clinical trials are essential to further evaluate its safety and efficacy. Longer-term follow-up data from these trials will provide valuable insights into the durability of responses and the potential for adverse events. Continued research efforts are necessary to optimize allogeneic CAR-T cell therapies and address any challenges that may arise, ultimately advancing their clinical utility in treating various malignancies (106).

While significant progress has been made in the exploration of allogeneic and in vivo CAR-T cells, they remain in continuous proof-of-concept stages without clinical approval. Therefore, enhancing product quality, refining manufacturing processes, and reducing self-production time are crucial to address the challenges faced by autologous CAR-T cells. Shortening the ex vivo culture time has been shown to yield CAR-T cells with improved effector function and reduced production costs (54). Several promising methods for rapid CAR-T cell production have been explored, such as the FasTCAR platform by Gracell Biotechnologies, which produces CAR-T cells within a day and has demonstrated efficacy in preclinical and clinical evaluations for R/R B-ALL (35). Additionally, researchers at the University of Pennsylvania have successfully prepared functional CAR-T cells within 24 hours by directly transducing non-activated quiescent T cells (by lentivirus vector), extending the survival of tumor-bearing mice (107). Agarwalla et al. have developed an implantable Multifunctional Alginate Scaffold for T Cell Engineering and Release (MASTER), which integrates T cell activation, reprogramming, and in vivo expansion, reducing manufacturing time to 1 day. CAR-T cells produced using the MASTER scaffold have shown promising anti-tumor activity in mouse xenograft models of lymphoma (36). These rapidly manufactured CAR-T cells exhibit superior anti-tumor activity and greater persistence compared to conventional CAR-T cells, potentially offering a more cost-effective approach.

Allogeneic universal CAR-T cells, while offering solutions to immune rejection through gene editing, also raise safety concerns. A previous clinical trial of allogeneic CAR-T therapy was halted by the FDA due to the emergence of chromosomal abnormalities (108). Although investigations suggested the abnormality wasn’t related to gene editing, safety concerns persist. A potential solution lies in direct CAR-T cell generation in patients via a universally applicable medicinal product containing the CAR gene. The illustration of in vivo generation of CAR-T cells is depicted in Figure 4 (109). This approach’s combination of simplicity, speed, and cost-effectiveness makes it an attractive option for CAR-T cell therapy. By leveraging the body’s natural processes, in vivo CAR-T cell generation eliminates the requirement for ex vivo cell manipulation and lengthy manufacturing processes. This streamlined method not only reduces production time but also lowers associated costs, potentially improving the accessibility of CAR-T cell therapy to a broader patient population. Various vector platforms are under exploration, including lentiviral vectors (LVs) and nanoparticles (NPs) (110, 111). Lentiviral vectors (LVs) stand as the predominant choice for ex vivo CAR-T cell transduction, boasting stable gene integration, high transduction efficiency, and a broad host range. Pioneering work by Buchholz et al. introduced the pseudotyping of vectors with modified envelope proteins of Nipah virus (Niv), enabling specific targeting of CD8 by fusing the envelope protein to a CD8-specific single-chain variable fragment (scFv) (112). Their study demonstrated that CD8-LV could directly generate human CD19-CAR-T cells in NGS mice in vivo, showcasing potent anti-tumor activity. Furthermore, Buchholz et al. developed Niv-LV targeting CD3 and CD4, capable of directly generating functional CAR-T cells in mice (113). Despite LVs’ high transduction efficiency, the potential risk of insertional mutagenesis remains a concern (114). In recent years, researchers have explored virus-like particles (VLPs) as a novel vector. VLPs retain viral proteins without containing a packaged genome, combining viral vector targeting specificity with the transient delivery advantages of non-viral vectors. Hamilton et al. demonstrated the generation of gene-edited CAR-T cells in vivo by packaging Cas9 RNPs into retroviral VLPs, offering a new direction for future in vivo CAR-T cell generation, albeit with potentially lower in vivo transduction efficiency compared to LVs (115). Nanoparticles (NPs) have garnered significant attention as gene delivery vehicles due to their low immunogenicity, cost-effectiveness, and customizable production (116). Unlike the complex machinery of viruses, gene delivery using NPs relies on the physicochemical properties of the particles and payloads, offering advantages such as payload flexibility and ease of modification (117). The two primary types of nanocarriers used in CAR-T cell development are lipid- and polymer-based NPs. Numerous preclinical studies have successfully employed these carriers to deliver CAR gene-containing DNA or mRNA into T cells in vivo, leading to the on-site generation of CAR-T cells and effective cytotoxic functions in small animal models. For instance, Matthias T. Stephan’s team designed polymer nanoparticles encapsulating CAR DNA and mRNA (118, 119), resulting in anti-tumor efficacy comparable to CAR-T cells prepared using traditional lentiviral vectors. In addition to antibody-targeted nanocarriers, Daniel J. Siegwart’s team recently introduced a Selective Organ Targeting (SORT) LNP capable of delivering mRNA to the spleen and locally generating CAR-T cells in a controlled manner (120). The research into in vivo generation of CAR-T cells has the potential to transform CAR-T cell therapy into a widely adopted pharmaceutical treatment (109). This advancement could significantly enhance clinical compliance and substantially reduce costs (121). Amid the FDA’s recent warning regarding secondary cancer risks following CAR-T cell infusion treatments, the application of in vivo editing to transiently modify CAR-T cells using mRNA offers a promising, potentially safer, and cost-effective solution to address the existing challenges associated with CAR-T cell therapy (122). By harnessing AI-based tools to refine CAR design, the advancement of in vivo CAR-T cell development stands poised to accelerate, offering a more streamlined and effective approach toward overcoming the current challenges in CAR-T cell therapy (123, 124).