In 1989, Gross et al. first engineered T cells expressing CARs, which recognize antigens of interest without MHC restrictions (38). Currently, CARs can recognize a wide range of antigens, including tumor-associated antigens (TAAs) and tumor-specific antigens (TSAs) (39–41). The antigen-binding domain of CAR is typically derived from the single-chain variable fragment (scFv) of a monoclonal antibody. Its affinity optimization is needed to balance T-cell activation and toxicity (42, 43). CAR-T evolution from 1st to 5th generation reflects a shift from basic to complex signaling platforms, which improves T-cell activation, proliferation and persistence (44). Current CARs integrate multiple co-stimulatory domains and signaling elements, including components that enable cytokine secretion. Thus far, there have been seven FDA-approved CAR-T products (45–47).

CAR-T therapy has demonstrated remarkable clinical benefits, especially in hematologic malignancies (3). However, its application in solid tumors remains limited owing to challenges such as antigen escape, insufficient trafficking and infiltration, and the immunosuppressive TME. Strategies to enhance efficacy in solid tumors include optimizing target antigen selection, designing CARs that recognize TSAs, improving CAR architecture, modulating the TME, and advancing cell manufacturing and delivery techniques.

CRS and ICANS are common and life-threatening toxicities of CAR-T cell therapy (48). CRS occurs when the activation of CAR-T cells, which release cytokines such as IFN-γ, tumor necrosis factor-α (TNF-α), IL-6, IL-1, IL-10, and granulocyte-macrophage colony-stimulating factor (GM-CSF), is triggered (49). These cytokines activate bystander monocytes and Mφs, amplifying the inflammatory response. TNF-α activates the nuclear factor-kB signaling pathway, inducing the expression of anti-apoptotic and pro-inflammatory genes (50, 51). IL-6 can activate other immune cells, further activating T cells and establishing a “positive feedback loop” (52). Additionally, IL-6 and other cytokines can induce endothelial activation, leading to systemic inflammatory reactions (53). Once the blood-brain barrier (BBB) is disrupted, peripheral cytokines can enter the central nervous system (CNS). Additionally, cluster of differentiation (CD)19 CAR-T cells may recognize CD19⁺ brain mural cells, thereby directly activating CNS immune cells and inducing cytokine release (54, 55).

TCR-T cells express engineered TCRs that initiate intrinsic activation upon recognizing specific peptide-MHC (pMHC) complexes. This activation is primarily mediated through the CD3 complex (56, 57). Engineered TCRs with high affinity for pMHC complexes can activate T cells more efficiently and elicit stronger signaling responses (32, 58, 59). Unlike CARs, TCRs recognize both intracellular and surface antigens; they can also respond to lower antigen densities (4).

TCR-T therapy has shown promising efficacy in acute myeloid leukemia (AML) (60). Morelli et al. conducted a first-in-human trial testing a TCR-T product, which recognizes shared KRAS and TP53 mutations in solid tumors. The trial, which involved patients with solid tumors, showed a favorable overall response (OR) lasting up to 6 months (61) ( Table 1 ). In August 2024, the FDA approved Tecelra (afamitresgene autoleucel), the first TCR-T product, for the treatment of synovial sarcoma (72).

Despite these findings, the complete response rate (CRR) of TCR-T therapy remains low; many patients experience only transient responses followed by relapse. The development of TCR-T therapy faces multiple challenges. Apart from the T-cell exhaustion and immunosuppressive TME hindering CAR-T therapy, TCR-T therapy is limited by the challenge of producing T cells that precisely express TCRs to identify heterogeneous tumor cells (4).

In 1988, Rosenberg and colleagues reported the first clinical case in which TILs induced regression in a patient with metastatic melanoma (73). TILs possess multiple antigen-recognition abilities and strong tumor-homing properties, excelling in addressing tumor heterogeneity and infiltration. The first product, LN144, has received FDA approval for the treatment of melanoma. TILs also demonstrate efficacy in other ICI-resistant solid tumors, including breast cancer and non-small cell lung cancer (mNSCLC) (30, 74). Shared resistance mechanisms in ICI and TIL therapies include abnormal neoantigen presentation and reduced tumor mutational burden (TMB) (66, 75, 76). It is reasonable to speculate that TIL therapy would show limited efficacy in patients with ICI resistance. Recent studies have shown that ex vivo expanded TILs from ICI-resistant patients can still recognize and eliminate fresh tumor cells in advanced melanoma and mNSCLC. This may result from TILs’ ability to bypass ICI-specific resistance mechanisms, the absence of in vivo T-cell suppressors during manufacturing, and the numerical advantage of TIL therapy. However, ex vivo expanded TILs derived from ICI-resistant patients with advanced melanoma and mNSCLC are shown to retain the ability to recognize and eliminate fresh tumor digests. This may be attributed to the capacity of TILs to circumvent ICI-specific resistance mechanisms, the absence of in vivo suppressive factors during manufacturing, and the numerical advantage of TILs (25, 29, 30, 66). For example, responses to TIL therapy were observed in mNSCLC with programmed death-ligand 1 (PD-L1)–negative, TMB^low, or STK11-mutated, which typically indicate resistance to ICIs (30). Despite these benefits, patients previously exposed to programmed cell death protein 1(PD-1) inhibitors have a lower response rate to TIL therapy, likely owing to shared resistance mechanisms. Thus, if applied as a first-line treatment, therapeutic potential of TIL therapy may be optimized (66). Ongoing clinical trials (NCT05727904, NCT03645928) are currently evaluating the efficacy and safety of regimens that include lifileucel combined with pembrolizumab in ICI-naïve patients.

Despite its potential, TIL therapy faces challenges. It requires efficient identification and isolation of tumor antigen-specific lymphocytes (77). Additionally, the production process of TIL therapies must be tightly controlled and standardized to ensure the quality and efficacy of the cellular products. However, there are still deficiencies in the expansion process and the selection of specific T-cell subsets for TIL, as well as a lack of comparability in production processes between studies. The Study by Albrecht et al. has shown that the use of the Zellwerk ZRP bioreactor enables the automated control of key parameters in the culture process (e.g., temperature, pH, and pO₂), ensuring the stability and consistency of culture conditions (78).

NK cells play a crucial role in innate tumor surveillance, providing innate anti-tumor activity without prior sensitization. This makes them ideal for engineering into CAR-NK cells. With favorable safety profiles, rapid action, and potential for “off-the-shelf” allogeneic products (via reduced manufacturing time or costs), they possess enhanced clinical feasibility. Clinical trials have shown promising results ( Table 1 ). On the other hand, their safety and sourcing flexibility position them as an alternative to CAR-T (79). NK cells can induce tumor apoptosis even at low numbers, thereby reducing the risk of excessive cytokine release. Moreover, their MHC-unrestricted activation avoids graft-versus-host disease, even when transferred into allogeneic hosts (79).

Like other ACTs, NK therapy is also hindered by TME suppressive factors, notably transforming growth factor-beta (TGF-β). TGF-β suppresses NK cell function by activating SMAD signaling, inhibiting the secretion of perforin, granzyme B, and IFN-γ, and downregulating activating receptors (e.g., NKG2D, NKp30) through the SMAD signaling pathway, which is activated by the TGFβRI/II receptor complex (80, 81). Blocking this pathway via TGFβRII intracellular domain deletion preserves NK function, supporting trials in TGF-β-rich tumors, such as glioblastoma (82, 83). Although NK therapy also faces common ACT challenges, CAR-NK shows promise with innate advantages over CAR-T (84–86).

CIK cells comprise CD3⁺CD56⁺ NK-T cells, CD3⁺CD56^- T cells, and CD3^-CD56⁺ NK cells. These cells are generated by culturing peripheral blood-derived lymphocytes with anti-CD3 antibodies, IL-2, and IFN-γ (87, 88). CIK cells can recognize tumor cells through both MHC-restricted and MHC-unrestricted pathways. This can help CIK cells overcome tumor antigen escape (89).

CIK therapy exerts anti-tumor effects through multiple mechanisms. Classically, NK-like cells exert toxicity when their inhibitory receptors (e.g., KIR family members) fail to bind MHC-I molecules, known as the “missing self” mechanism. It happens when tumors downregulate MHC-I molecule expression to escape immune surveillance (90–92). NKG2D receptors on NK-like cells bind with NKG2D ligands (e.g., MICA/B) on tumors to trigger cytotoxicity. However, this process is impaired when soluble MICA produced by tumors neutralizes NKG2D receptors (93, 94). In such cases, NK-like cells activate compensatory pathways through co-expressed DNAM-1 receptors, which recognize CD155 molecules on tumors, forming a “dual-receptor recognition network” and helping to maintain anti-tumor function (95, 96).

The proportions of different CIK cell subsets may lead to inconsistent treatment outcomes. A higher proportion of CD3+CD56+ or CD4- CIK cells may correlate with a better response or diminished cytotoxicity (8, 9). Compared with solid tumors, CIK cells show limited efficacy due to the TME, and they tend to perform better in killing hematological malignancies (97). Enhanced manufacturing techniques and combination treatment modalities may be needed to improve the outcomes of CIK therapy.

Composed of the innate immune system, Mφs possess the capability of phagocytosis. Mφs polarize into pro-inflammatory M1 (promoting tumor killing) and immunosuppressive M2 (facilitating tumor progression) phenotypes. To address the immune escape mechanisms in the TME, such as the CD47/SIRPα “don’t eat me” axis (inhibiting phagocytosis) and TGF-β-driven M2 polarization, CAR-M was designed. In addition, unlike the first-generation CAR-M, which was “copied” from CAR-T, there is now a second-generation CAR tailored specifically for the Mφs to enhance its effectiveness (98). For example, Zhang et al. engineered induced pluripotent stem cell (iMACs)-derived Mφs with CD3ζ-TIR-CAR, which incorporates the intracellular Toll/IL-1 receptor (TIR) domain of Toll-like receptor 4. This CAR design features tandem CD3ζ-TIR dual signaling, enabling iMACs to exhibit both target cell phagocytic ability and antigen-dependent M1 polarization while resisting conversion to the M2 phenotype (99).

Clinical trials have shown that Mφ-based therapies are safe, with minimal side effects (e.g., fever, abdominal discomfort) (100) ( Table 1 ). The human epidermal growth factor receptor 2 (HER2) CAR-M product CT-0508 has shown efficacy in refractory HER2⁺ solid tumors receiving conventional treatment (71). Similar to NK cells, which have shorter lifespans and lower cytotoxicity compared to T cells, Mφs also hold potential for developing “off-the-shelf” products. However, Mφs are more difficult to use for ACT than T cells because of their weak proliferative capacity both in vivo and ex vivo (12).

Lymphocyte exhaustion contributes to poor ACT outcomes, and ICIs play a key role in overcoming this barrier. After initial T-cell activation via recognition of tumor antigens, ICPs are upregulated on lymphocytes and tumor cells. ICIs block these inhibitory signaling pathways and restore T-cell activity (136) ( Figure 2A ). In addition, ICIs modulate cytokine expression within the TME. IL-2 and IFN-γ are pivotal cytokines driving CD8⁺ T cell proliferation and activation. PD-1 engagement with PD-L1 markedly inhibits the secretion of these cytokines, a process that ICIs can potentially reverse (14, 15). Tumors often display elevated IL-10 production from Mφs and Treg cells. Notably, PD-1 deficiency in Mφs enhances IL-10 secretion, which in turn suppresses T helper type 1 (Th1) cells. ICIs may disrupt this axis by inhibiting PD-1-mediated IL-10 upregulation, thereby modulating the tumor immunosuppressive milieu (16).

In hematologic malignancies, the combination of PD-1 blockades with CAR-T cell therapy enhances the response rate and CAR-T cell persistence. In a retrospective analysis, patients with relapsed/refractory (r/r) diffuse large B-cell lymphoma (DLBL) and TP53 mutations received CD19 or CD20 CAR T therapy individually or in combination with PD-1 inhibitors (Sintilimab/Tislelizumab). The results showed that the overall survival (OS) of the CAR-T + PD-1 inhibitors group was not reached, while the OS of the CAR-T group alone was 10.9 months (137). Similar benefits have been observed in cases of leukemia managed by CIK therapy combined with various ICIs (138). In solid tumors, these combination strategies have also yielded encouraging results (139). A phase 2 trial in ICIs-naive advanced melanoma resulted ORR of 63.6% and 5 CRs with TILs + pembrolizumab. After a 17.2-month follow-up, the median response duration remained unreached (29). A Phase 3 study (NCT05727904), still recruiting participants, will assess the efficacy and safety of the TIL product TILVANCE-301 plus pembrolizumab versus pembrolizumab alone for untreated unresectable or metastatic melanoma.

Miller et al. found that among the exhausted CD8⁺ TILs, a small population of progenitor or stem-like T cells (TCF1⁺TIM-3^-) can differentiate into highly toxic, terminally exhausted TILs. Furthermore, PD-1 blockades may increase their proportion in TILs by promoting proliferation (140). Huang et al. identified a subset of tumor-specific CD8⁺ cells in the tumor-draining lymph nodes, which exhibited typical memory characteristics and anti-tumor effects after adoptive transfer. These cells were identified as responders to PD-1/PD-L1 blockades (141). The outcomes of clinical trials evaluating TIL therapy plus pembrolizumab in patients with various solid tumors have been particularly promising ( Table 2 ).

Notably, ICI efficacy may be enhanced by ACT. Evidence shows that NKG2D.ζ-D NK cells can eliminate MDSCs, which reduces the effector function of ICIs (144). A novel strategy that modifies cells to secrete ICIs locally represents an alternative approach to exogenous ICIs. Chen et al. developed CAR-T cells capable of secreting a PD-1 scFv, which primarily accumulates at the tumor site alongside the CAR-T cells. This minimizes systemic toxicity associated with widespread ICI distribution (143).

Research concerning CAR-Mφ therapy in combination with ICIs is also ongoing (71). However, further investigations are needed to determine the feasibility of combining ACT with ICIs.

Antigen escape remains a significant obstacle to the long-term efficacy and durability of ACT. Under the selective pressure of antigen-specific adoptive cells, tumor cells may downregulate targeted antigens (1). Targeting multiple antigens can help mitigate antigen escape, extend the persistence of ACT, and reduce the likelihood of relapse ( Figure 2B ). One approach involves engineering “tandem CARs,” which incorporate two tumor-specific antigen–targeting scFvs within a single CAR construct. In r/r non-Hodgkin lymphoma (NHL), tandem CD19/CD20 CAR-T cell therapy has produced ORR ranging from 70% to 90%, with a CRR of approximately 70% (145–147). One tri-specific CAR-T product targeting HER2, IL-13 receptor α2 (IL-13Rα2) and EphA2 has shown encouraging results in the treatment of glioblastoma (159). Another strategy involves sequential administration of CAR-T cells targeting different antigens. For instance, reinfusion of CD22 CAR-T cells displayed substantial clinical benefit in patients with r/r B-cell acute lymphoblastic leukemia (B-ALL) who had previously relapsed due to CD19 antigen loss after CD19 CAR-T therapy (148). Additionally, approaches that combine two types of ACT may eliminate both tumor cells and immunosuppressive cell populations. In a preclinical study, CAR-T cells were co-administered with NKG2D.ζ NK cells. The engineered NK cells enhanced the anti-tumor activity of CAR-T cells within solid tumors by depleting MDSCs and secreting pro-inflammatory cytokines (144).

Chemotherapy can directly kill tumors, while also depleted Treg cells and MDSCs that inhibit effector T (Teff) cells. It is worth noting that cyclophosphamide and carboplatin have been proven to be more cytotoxic to Treg cells than to Teff cells (160, 161).Furthermore, chemotherapeutic agents induce tumor cell death and promote the release of DAMPs, including adenosine triphosphate, high-mobility group box 1 (HMGB1), and type 1 IFN (IFN-I). These DAMPs contribute to dendritic cell (DC) activation and facilitate antigen presentation to lymphocytes (17) ( Figure 2C ). This form of tumor apoptosis, which involves innate immune responses, is termed immunogenic cell death (ICD). After ICD, DCs and T cells are more likely to accumulate within the TME and secrete TNF-γ (162). A study on ErbB CAR-T cell therapy demonstrated that pretreatment with carboplatin enhanced tumor regression, even at lower CAR-T cell doses (160). Additionally, CIK or DC/CIK therapy plus chemotherapy has been shown to improve OS and progression-free survival (PFS) across various malignancies (89).

Chemical agent-based lymphodepletion before cell infusion is a critical step in ACT, as it enhances the tumor reactivity and persistence of transferred lymphocytes. By depleting endogenous lymphocytes, lymphodepletion reduces competition for cytokines and resources, thus facilitating interactions between infused T cells and tumor antigens (1). In a study by Curran et al., preconditioning lymphodepletion via high-dose cyclophosphamide led to enhanced therapeutic responses and CAR-T cell expansion without greater toxicity when combined with CD19 CAR-T therapy (150).

RT enhances the efficacy of CAR-T cell therapy through several mechanisms ( Figure 2D ). RT promotes ICD by increasing DAMP production and modulating the TME. After radiation-induced DNA damage, activation of the cyclic guanosine monophosphate/adenosine monophosphate synthase–stimulator of interferon genes (cGAS-STING) pathway leads to upregulation of IFN-I expression. Local RT in combination with ACT has been shown to improve T-cell infiltration by elevating chemokine and cytokine levels (18, 152) (18). Moreover, RT can downregulate the expression of CD47 (the “do not eat me” signal) while simultaneously upregulating MHC-I molecules and death receptors on tumor cells, thus promoting APC activation and enhancing immune recognition.

Indications for RT in the context of CAR-T cell therapy include reducing recurrence or progression at high-risk sites, serving as a palliative intervention, and managing tumor deposits located near vital structures (163). RT can be administered either before or after CAR-T cell infusion. Pre-treatment RT may reduce tumor burden in high-risk cases. For example, two patients with rapidly progressing inferior vena cava tumor thrombus underwent Gamma Knife treatment, followed by infusion of glypican-3 (GPC3) CAR-T cells, which resulted in favorable clinical outcomes (151). Salvage RT is increasingly used in cases of post-CAR-T disease progression. Patients with DLBL (164) or multiple myeloma (MM) (165) who experience local recurrence after CAR-T cell therapy may benefit from subsequent RT.

Tumor vaccines, administered in the form of whole cells, oncolytic viruses or molecular agents such as peptides or RNA, can enhance ACT by priming cancer antigen presentation. This process increases the reactivity and persistence of adoptively transferred cells ( Figure 2E ).

Through presenting TAA to T cells, DC vaccines activate and promote the proliferation of those cells (166). The administration of DC vaccines cultured with EGFR pathway substrate 8 (Eps8)-derived peptide or Wilms’ tumor 1 protein (WT1) after CD19 CAR-T therapy has been shown to increase the proportion of central memory T (Tcm) cells, stimulate CAR-T cell expansion, and improve therapeutic efficacy in r/r B-ALL (19). Similar effects were also observed with the combination of DC vaccines and CIK therapy (89). In addition to DCs, tumor cells themselves can be modified to serve as vaccines. For instance, a whole-cell vaccine based on K562 cells engineered to express cytomegalovirus (CMV)-derived phosphoprotein 65 facilitates (pp65) CAR-T cell proliferation and activation by presenting CMV 65 and co-expressing immunostimulatory molecules (e.g., CD40L and OX40L) (153).

Oncolytic viruses selectively infect and lyse tumor cells, but they can also stimulate systemic immune responses after intratumorally injection. In a study by Evgin et al., mice received CAR-T cells preloaded with oncolytic viruses such as vesicular stomatitis virus or reovirus bound to specific receptors. Upon arrival at the tumor site, the viruses replicated within tumor cells and triggered localized inflammation. This approach increased T-cell specificity and potency, thereby enhancing proliferation and anti-tumor activity (167).

Compared with conventional protein- or peptide-based cancer vaccines, mRNA vaccines offer advantages in both personalization and large-scale production. They address the unique therapeutic needs of individual patients without reliance on specific human leukocyte antigen haplotypes (168). Promising outcomes were observed regarding the combination of oncofetal antigen claudin 6 (CLDN6) CAR-T cell therapy and an amplifying RNA vaccine in solid tumors. The modified nanoparticulate vaccine, CARVac, incorporates CAR target antigen–encoding mRNA encapsulated in lipid-based structures, known as RNA-lipoplexes. Upon administration, the vaccine facilitates systemic delivery of the RNA to APCs in lymphoid tissues (154).

Cytokines serve as signaling molecules that regulate immune cell homeostasis and orchestrate signal-dependent immune responses (169). In the context of conventional ACT, cytokines such as IL-2, IL-7, IL-12, and IL-15 have been used through ex vivo preconditioning or in vivo co-administration (170) ( Figure 2F ). In CAR-T cell therapy, two innovative strategies use genetic engineering to enhance therapeutic function. The first strategy involves engineering CAR-T cells to produce specific cytokines autonomously. These fourth-generation CAR-T cells, or “armored” CAR-T cells, secrete cytokines (IL-2, IL-7, IL-15, IL-18, IL-21, IL-36) or chemokines (e.g., CC motif chemokine ligand [CCL]19) to improve anti-tumor activity in the TME (170). The second strategy involves modifying the architecture of cytokine receptors to alter intracellular signaling pathways. For example, by fusing the extracellular domain of the IL-4R with the intracellular domain of the IL-7R, researchers developed a novel inverted cytokine receptor (ICR), termed IL-4/7 ICR. Co-expression of this ICR with a CAR in T cells leads to enhanced expansion and anti-tumor activity (157). Similar results were observed with IL-4/21 ICR, which combines the IL-4R and IL-21R (158).

CRS and ICANS are the most common severe toxicities after CAR-T therapy. Endothelial activation plays a central role in their pathogenesis (190). Teachey et al. developed predictive models using cytokine levels, tumor burden, and other variables to estimate the likelihood of grade 4–5 CRS (191). Hong et al. constructed a decision tree model incorporating soluble VCAM-1, the angiopoietin (Ang)-2:Ang-1 ratio, and soluble ICAM-1 to predict CRS occurrence and severity (190). IL-6 is regarded as a cytokine involved in endothelial permeability and CRS development. IL-6 receptor inhibitors, such as tocilizumab, may be administered before the IL-6 level peaks, guided by predictive modeling (192).

ICANS occurring after CAR-T infusion may be linked to blood–brain barrier disruption and increased permeability (193). Platelets release Ang-1, which stabilizes endothelial cells; thus, thrombocytopenia contributes to capillary leakage and is associated with severe CRS and ICANS (194). A high Ang-2:Ang-1 ratio and elevated von Willebrand factor (vWF) levels have been observed in patients with grade ≥4 ICANS (194). ICANS-related dysregulation can also be reflected by cerebrospinal fluid (CSF) cytokine levels, including IL-8, IL-10, and monocyte chemoattractant protein-1 (MCP-1) (35, 195). Biomarkers of CNS injury measurable before CAR-T infusion, such as neurofilament light chain, S100 calcium-binding protein B (S100B), and glial fibrillary acidic protein, are detectable in blood and directly reflect their CSF concentrations. These biomarkers may serve as predictors of high-grade ICANS (193, 196, 197).