The eligibility criteria were predefined within the study protocol, aligning with each aspect of the PICOS framework (Population/Animals, Intervention, Comparator(s)/Control, Outcome Measure(s), and Study design):

To collect fitting studies, we conducted comprehensive research in the PubMed database and in Google Scholar and BASE (Bielefeld Academic Search Engine) search engines. The search strategy was developed in accordance with the step-by-step guide of Leenaars et al. (12). The full search strategy is available in the PROSPERO protocol and in the Supplementary Materials .

All identified records were imported into EndNote (RRID: SCR_014001) for the management of search records. Following the removal of duplicates, two reviewers (AM and EM) independently screened the titles and abstracts of the studies. Subsequently, two researchers (KZ and EM) independently reviewed the full-text articles for inclusion. In instances of disagreement, consensus on inclusion or exclusion was achieved through discussion, with involvement from a third researcher (AM) if necessary.

One or more of the baseline characteristics of the murine population (strain, age, sex, total number, and provider of the mice) were adequately reported in the majority of studies. Notably, 42% (14 out of 33) of the included studies provided comprehensive details across all five categories. The ages of the mice varied between studies, with the highest range being 6 to 20 weeks and a median age of 7 weeks. In 45% (15 out of 33) of the studies, exclusively female mice were used. In contrast, only 6% (2 out of 33) of studies employed male mice. 15% (5 out of 33) of experiments employed both male and female mice, whereas the remaining 33% (11 out of 33) studies did not disclose this information. ( Figure 3A ). The total number of mice was either provided directly or deduced from the experimental data. Otherwise, if the number could not be calculated, it was classified as ‘not reported,’ which accounted for 18% (6 of 33) of the studies. The overall number of mice enrolled in all the experiments was 1595, with a median of 50 per study.

Most of the included studies (28 out of 33) employed the NSG mouse model (NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ), while one study used the NOG mouse model (NOD.Cg-Prkdc^scidIl2rg^tm1Sug/ShiJic) and one study used the NXG mouse model (NOD-Prkdc^scidIl2rg^Tm1/Rj). These models are characterized by the absence of functional T and B lymphocytes and exhibit multifaceted defects in NK cell activity, macrophage function, complement activity, and dendritic cell function. Despite distinct Il2rg targeted mutations and different NOD backgrounds, these models are considered equivalent regarding overall biological/physiological characteristics and experimental applications (19). Additionally, two studies utilized athymic nude mice carrying the “nude” mutation (Foxn1^nu ), which results in defects in T cell development. One study used nude mice on a BALB/C background (BALB/c Nude), while the other on a NMRI background (NMRI-Foxn1^nu/Foxn1^nu ).

The study by Pennell et al. employed a synergic mouse model on a C57BL/6 background where mice B-cells exclusively express human CD19 (huCD19^Tg/0). These mice were then implanted with a mouse lymphoma cell line engineered to co-express human CD19 (TBL12.huCD19) and received syngeneic murine T cells expressing a human CD19-directed CAR. This experimental design aimed to replicate clinical cytokine release syndrome (CRS) and neurotoxicity.

Tumors can be established using either human tumor cell lines (cell line-derived tumor xenografts, CDX) or tumor samples from patients (patient-derived xenografts, PDX). For models of hematopoietic-origin tumors cells are typically injected systemically via intravenous (IV) route. Solid tumors can be established by subcutaneous (SC) or intraperitoneal (IP) inoculation. To create a more realistic tumor environment, orthotopic inoculation (OI) is used, while systemic inoculation via IV route is employed to mimic tumor invasion and metastasis.

Patient-derived xenografts (PDXs) generated from clinical tumor samples are valuable models that closely replicate the genetic and cellular profiles of the original tumors. Among the included studies, Loff et al. and Raj et al. utilized PDXs to study the anti-tumor activity of the UniCAR and switchable CAR (sCAR) platforms, respectively. Loff et al. developed a CD123-expressing B-ALL xenograft model to evaluate the anti-tumor activity of UniCAR-T cells against extramedullary leukemic bulks by subcutaneously transplanting tumor cells into the flanks of NSG mice (20). On the other hand, Raj et al. employed an orthotopic model of advanced pancreatic ductal adenocarcinoma (PDAC) by inoculating stage IV PDAC cells into the pancreas of immunocompromised NSG mice (21). These cells express HER2 at relatively modest levels, creating stringent conditions for CAR-T therapy. This model establishes highly aggressive primary tumors with rapid-onset metastasis to the liver and lungs, mimicking late-stage PDAC in patients.

Most of the included studies (31 out of 33) established CDX models using popular cell lines that express the target antigen(s). For solid tumor establishment, the majority introduced the tumor antigen via the subcutaneous (SC) route (14 out of 33) or the intraperitoneal (IP) route (4 out of 33). Additionally, 8 studies utilized the intravenous (IV) route to establish hematopoietic-origin tumor models, with the exceptions of Cho et al. and Ruffo et al., which used human epidermal growth factor receptor 2 (HER2)-expressing cell lines to study the activity of CAR-T cells against metastatic tumor cells (22, 23).

The included studies targeted a range of hematologic and solid tumor antigens. Hematologic tumor antigens included CD33 (3 out of 33 studies), CD123 (3 out of 33 studies), CD20 (5 out of 33 studies), and CD19 (3 out of 33 studies). For solid tumors, the targeted antigens were HER2 (9 out of 33 studies), prostate stem cell antigen (PSCA) (2 out of 33 studies), receptor tyrosine kinase AXL (1 out of 33 studies), mesothelin (3 out of 33 studies), B7-H3 (CD276) (1 out of 33 studies), epidermal growth factor (EGFR) (4 out of 33 studies), cadherin-6 (CDH6) (1 out of 33 studies), folate receptor (FR) (2 out of 33 studies), carbonic anhydrase IX (CAIX) (1 out of 33 studies), prostate specific membrane antigen (PSMA) (2 out of 33 studies), glypican-3 (GPC3) (1 out of 33 studies), ROR1 (1 out of 33 studies), C-type lectin domain family 12 member A (CELC12A) (1 out of 33 studies), disialoganglioside (GD2) (1 out of 33 studies), and epithelial cell adhesion molecule (EpCAM) (1 out of 33 studies).

Discussion regarding the immunogenicity of CAR-T platform components was addressed in a relatively modest proportion of studies, with only 36% (12 out of 33) acknowledging or examining this aspect. Rodgers et al. specifically investigated the immunogenicity risk associated with engrafting a nonhuman sequence into an antibody fragment (Fab) to create a switch. They conducted an in silico immunogenicity analysis of the adapter linked to the N terminus of the light or heavy chains of a model, therapeutically approved antibody (trastuzumab). This analysis predicted that the PNE graft had a low likelihood of inducing an antibody response in the context of a typical antibody (24).

Regarding euthanization criteria, 54% (18 out of 33) of the studies presented explicit criteria, while merely 12% (4 out of 33) included comprehensive assessments of overall body condition and coat condition. Furthermore, a subset of studies, constituting 15% (5 out of 33), employed a secondary mouse strain for various specific purposes, including pharmacokinetics, T-cell depletion, pharmacokinetics-half life, pharmacokinetics-biodistribution, and breeding of the huCD19^Tg/0 strain. Detailed characteristics of the animal models are described in Table 2 .

The modular CAR platforms varied across studies depending on the adapter molecules used to facilitate the interaction between the soluble and signaling modules. Among the included studies, 18% (6 out of 33) utilized the anti-FITC CAR platform, wherein the universal CAR recognizes antibodies or ligands tagged with fluorescein isothiocyanate (FITC) molecules. Since FITC was shown to be safe for in-human use in conjugation with monoclinal antibodies for tumor-specific fluorescence imaging (25), it has the potential for use in cellular immunotherapy. However, the short half-life of the soluble module necessitates multiple injections at short intervals.

Another 15% (5 out of 33) of the studies adopted the UniCAR platform, developed by Michael Bachmann’s laboratory at Helmholtz-Zentrum Dresden-Rossendorf (HZDR) in Germany. This platform employs a 10-amino acid sequence derived from the nuclear protein La/SS-B (5B9 tag) as the adapter molecule. The primary goal of this design is to provide rapid and reversible control of CAR T cell activity, enable multiple targeting, and reduce the risk of developing antigen-free tumors during treatment. UniCAR contains humanized anti-La 5B9 scFv as the ectodomain of CAR, allowing CAR T cells to target multiple antigens either sequentially or concurrently. While this model offers significant flexibility, potential immunogenicity is a concern. The La protein is a known autoantigen in Sjögren’s syndrome and systemic lupus erythematosus, and autoantibodies against La epitopes are common in patients and may also exist in healthy populations (26).

Additionally, 15% (5 out of 33) of the studies employed the switchableCAR platform (sCAR), which features a 14-amino acid peptide neo-epitope (PNE) sequence from the GCN4 yeast transcription factor as the adapter. This platform is advantageous because the PNE sequence does not exist in the human proteome, has a high-affinity antibody, and has been found to be non-immunogenic. The anti-PNE CAR utilizes the PNE-antibody (52SR4) scFv as the ectodomain. However, despite the non-existence of the PNE sequence in the human proteome, antigenic mimicry may still result in off-target activation of CAR T-cells.

9% (3 out of 33) of the studies employed the conventional anti-CD19 CAR, equipped with a CD19-containing bridging protein (BP) capable of redirecting CAR19 T cells towards a range of target antigens. The BP comprises the extracellular domain of human CD19 coupled with disulfide-stabilized single-chain Fv (scFv) or nanobodies (Nb) targeting the respective antigen.

Furthermore, the following CAR platforms were each represented by two studies:

Concerning the structural characteristics of the signaling module, the vast majority of the included studies (29 out of 33) employed a second-generation CAR. These second-generation CARs featured a single costimulatory domain, which was either CD28 or 4-1BB. Only one study out of the 33 included used a first-generation CAR, which lacks a costimulatory domain. Moreover, three studies utilized a third-generation CAR with the two costimulatory domains CD28 and 4-1BB (3 out of 33 studies). The hinge domain varied widely, with CD8a-derived being the most commonly used in 48% of studies (16 out of 33), followed by CD28-derived in 36% (12 out of 33). Other variants included CD3-derived (1 out of 33), IgG4-derived (2 out of 33), and a mutant variant of the IgG4 hinge domain (2 out of 33). The transmembrane domain was primarily derived from CD28 in 55% of the studies (18 out of 33), while CD8a was the source in 42% of the studies (14 out of 33). In one study, the transmembrane domain was derived from CD3. In all included studies, the CD3ζ activation domain was utilized. The detailed structure of CAR is described in Supplementary Table S1 .

In the majority of cases (88%, 29/33), the intervention followed a specific sequence, with the tumor being injected first and allowed to establish itself, followed by the administration of CAR T-cells, either with or without the soluble module, and repeated doses of the soluble module and/or CAR T-cells as part of the treatment regimen. However, in two studies, the exact time of CAR administration in days after tumor establishment was not reported (43, 44). In the other two studies, CAR T-cells were administered prior to tumor establishment to create a low-burden tumor model and allow for the CAR T-cell engraftment of the peripheral blood (26, 36). Two other studies co-administered tumor cells with CAR-T cells (45, 46). Additionally, one study included a co-intervention in which all mice received intraperitoneal injections of 1,000–2,000 IU of IL-2 twice a week for 4 weeks (47).

In vivo antitumor activity against xenografts is a critical factor in determining the clinical development potential of specific CAR designs. To obtain reliable readouts that mimic clinical settings, xenografts are often treated with suboptimal doses of CAR T cells. In the studies reviewed, the CAR-T doses ranged between 1 to 40 million cells, with CAR-T cells typically administered once. The soluble module was administered multiple times in all studies except for those by Saleh et al., Landgraf, and Ochi, where it was given once alongside CAR-T cells, and Ambrose et al. and Rennert et al., where it was secreted by the CAR-T cells. Supplementary Table 1 contains more comprehensive details regarding the intervention employed in each study.

To evaluate CAR-T in vivo, there are critical parameters to be assessed and are related to the wide range of CAR-T-functions after encountering the target tumors in an organism. The primary measure of CAR-T cell functionality is their killing activity, which is assessed through changes in tumor burden or volume in xenograft models compared to negative or positive control groups. Although this measure cannot precisely predict the performance of CAR-T therapy in clinical context, it provides an estimation of the cell product’s ability to induce remission.

Out of the 33 studies analyzed, 19 reported measurements of tumor burden in treated animals, while 15 reported tumor volume measurements. However, only 10 studies for tumor burden and 11 for tumor volume were eligible for statistical synthesis. Exclusion criteria were applied due to missing reports of the central tendency measure (mean or median) and its error, absence of standard deviation or error, or normalization of outcome measures to unreported baseline values.

All of the included ten studies examined the impact of tumor burden on the efficacy of anti-tumor CAR-T therapy compared to various negative control cohorts, which included 1) untreated mice; 2) mice treated with only the switchable (soluble) module without CAR-T cells; 3) mice treated with only CAR-T cells without the switchable (soluble) module; and 4) mice treated with non-transduced T cells. Six of these studies also included a comparison of CAR-T efficacy with conventional CAR-T cells targeting the same antigen as a positive control. The negative control cohorts were combined, and two separate forest plots were utilized to represent the relevant efficacy of the included CAR-T platforms in eliminating the target tumor. The first plot compared the experimental group with the combined negative control group, while the second plot compared the experimental group to the positive control group. A total of 234 mice were included in these studies. One study by Peng et al. contributed data from two different cell lines for xenograft model establishment; accordingly, we added numerical identifiers to the study ID, as follows: 1) JeKo-1 mantle cell lymphoma cell line; and 2) HT-29 colorectal adenocarcinoma cell line (39).

Individual studies demonstrated varying degrees of tumor burden reduction when employing modular CAR systems, each influenced by specific CAR architectures, cancer types, and experimental conditions ( Figure 4A ). Substantial between-study heterogeneity was observed, with an I² value of 84.8% (95% CI: 74.4; 90.9%). Graphic Display of Heterogeneity (GOSH) plots identified three influential outliers: 35, 47; and 54. After excluding these outliers, there was still notable heterogeneity the I² value decreased to 76.6% (95% CI: 53.4; 88.3%), likely due to the use of different modular CAR-T platforms across studies. However, the overall trend suggests a potential benefit of the experimental modular CAR-T platforms in reducing tumor burden.

Comparison with the positive control group indicated varied results among individual studies regarding tumor burden. ( Figure 4B ). For instance, some studies showed a slight tendency towards greater tumor burden in the experimental group compared to the control, but this difference was not significant. Heterogeneity in this analysis was moderate, with an I² value of 56.8% (95% CI: 0.0%; 81.4%). No influential outliers were identified in the GOSH plots despite the diverse nature of the CAR-T platforms used.

Eleven studies examined tumor volume as the primary determinant of anti-tumor CAR-T activity, comparing it to negative control cohorts. Two of these studies also included comparisons with conventional CAR-T cells targeting the same antigen as a positive control. Negative control cohorts were consolidated, and forest plots were utilized to represent the relevant efficacy of the included CAR-T platforms in eliminating the target tumor comparing either to the combined negative control group or the positive control. A total of 361 mice were involved in these studies. Four studies were included multiple times: Bejestani et al. contributed twice due to the utilization of two distinct tumor burden xenograft models, varying in CAR-T administration timing (36). The identifiers 1) denote a low-burden model with CAR-T cells administered four weeks before tumor establishment, while 2) represent a high-burden model with CAR-T cells administered four weeks after tumor establishment. Karches et al. were included three times as they employed three different cell lines for xenograft model establishment: 1) Suit-2-MSLN, 2) MIA-PaCa-MSLN, and 3) MIA MSTO-MSLN (49). Lu et al. used two different cell lines for tumor establishment, so they were included twice, as follows: 1) MDA-MB-231 and 2) Hos-FRa (50). Lastly, Lee et al. (2018) (44) utilized the MDA-MB-23 cell line, which naturally expresses FRa, and the same cell line engineered to express PSMA, or carbonic anhydrase (CA IX) (44). The identifiers (1) (2) (3), were assigned to these models, respectively.

As expected, individual studies demonstrated varying degrees of tumor volume reduction when employing modular CAR systems. However, the overall tendency shows a reduction in the tumor volume after treatment compared to the negative control. The between-study heterogeneity variance was estimated at I² = 72.7% (95% CI: 55.8%; 83.2%) ( Figure 5A ). Graphic Display of Heterogeneity (GOSH) plots identified three influential outliers: 41; (1, 49); and (1, 44). The I² value dropped to 52.6% (95% CI: 13.0%; 74.2%) after outliers’ exclusion. This moderate heterogeneity probably stems from the use of different modular CAR-T platforms. In the two studies using positive control for comparison, no significant difference in tumor volume was observed ( Figure 5B ).

It is essential to comprehend the dynamic changes in the quantity of CAR T cells and their phenotype, which indicates the developmental and functional stages of CAR T cells. The phenotype of CAR T cells is defined by the expression of certain surface markers that occur during their differentiation, activation, and memory formation. Typically, CAR T cells with a memory-like phenotype, characterized by the expression of surface markers CD62L, CCR7, CD45RA, and CD45RO, exhibit superior efficacy. Analyzing these markers on CAR T cells, alongside other functional tests, would provide a thorough understanding of the underlying biology of how CAR T cells maintain or lose their antitumor function.

Six studies examined the CD3⁺ CAR⁺ T-cell count in peripheral blood following T-cell injection, with He et al. (30), Lu et al. (50), and Ma et al. (29) reporting events per μL of blood, while Liu et al. (51) and Meyer et al. (37) reported it as a percentage of total lymphocytes. The duration of observation varied across studies, ranging from 10 to 54 days post-CAR-T cell administration. Only Liu et al. and Lu et al. assessed the significance of the observed differences between the experimental and control groups. The former found no significant difference between the experimental groups and the conventional CAR-T control group. The latter found a significant difference (p<0.05) between the experimental group and the CAR-T-only group, with a tenfold higher prevalence of CAR-T cells in the blood of the experimental groups in both the THP1-FRβ and MDA-MB-231 models.

Ruffo et al. (23) reported the percentage of various CAR-T cell phenotypes in peripheral blood after 7 days of CAR-T administration, including T-central memory (TCM) CD62L⁺CD45RA^-, T-effector memory (TEM) CD62L^-CD45RA^-, terminal effector memory T cells (TEMRA) CD62L^-CD45RA⁺, and T-stem cell memory (TSCM) CD62L⁺CD45RA⁺, without performing statistical tests for comparison with conventional CAR-T control because the results were reported for one sample per each group. Stock et al. (48) included the following phenotypes: naïve-like T (TN) cells as CD45RA⁺ CCR7⁺, central memory-like T (TCM) cells as CD45RA^- CCR7⁺, effector memory-like T (TEM) cells as CD45RA^- CCR7^-, and effector-like T (Teff) cells as CD45RA⁺ CCR7^-. Compared to the conventional CAR-T control, significant differences were observed for the TN (p<0.05), TCM (p<0.001), and TEM (p<0.0001) populations, while no significant difference was observed for Teff. However, after 10 days of treatment, CAR-T cells in the experimental group predominantly exhibited the effector memory-like T (TEM) phenotype.

Rodgers et al. investigated the impact of different doses of the soluble module (0.05, 0.5, or 2.5 mg/kg) on the proportions of various memory phenotypes (TCM, TEM, TEMRA, TSCM) compared to conventional CAR T cells. They examined the expression of CD45RA and CD62-L on CD4+ and CD8+ T cells in peripheral blood after 21 days (5 days following the final dose). The group receiving 2.5 mg/kg exhibited significant expansion of the CD45RA+CD62L− terminal effector memory expressing CD45RA (TEMRA) compartment in both CD4+ and CD8+ T cells, compared to the lower-dose groups. Conversely, the lower-dose groups (0.05 and 0.5 mg/kg) had significantly larger populations of CD45RA−CD62L+ central memory cells, which is hypothesized to result from lower levels of stimulation during the initial dosing period. This finding highlights a key advantage of this platform: the dosing of the soluble module can influence the CAR-T memory phenotype in vivo (24).

A detailed representation of this outcome is represented in Supplementary Table 2 .

All 18 studies included in the analysis reported survival rates in treated animals compared to negative control groups, with 5 studies also using conventional CAR-T as a positive control for comparison. Therefore, two separate forest plots were constructed to show the median survival relative to the combined negative group or positive group. A total of 518 mice were involved in the statistical synthesis. Seven studies were included multiple times: Bejestani et al. contributed twice due to using two distinct tumor burden xenograft models with varying CAR-T administration timing (36). The identifier (1) denotes a low-burden model with CAR-T cells administered four weeks before tumor establishment, while (2) represents a high-burden model with CAR-T cells administered four weeks after tumor establishment. Benmebarek et al. used (1) MV4-11-LUC-GFP and (2) THP-1-LUC-GFP cell lines to establish two xenograft models (52). Kuo et al. targeted two different tumor antigens (1): EGFR/HER3, and (2) CDH6 (28). Loff et al. used two different xenograft models (1): cell-line-derived and (2) patient-derived (20). Peng et al. contributed data from two different cell lines for xenograft model establishment (1): JeKo-1 mantle cell lymphoma cell line; and (2) HT-29 colorectal adenocarcinoma cell line with R12 N Fab soluble model; and (3) with 324 N Fab soluble model (39). Rennet et al. used two cell lines for xenograft model establishment (1): U937 and (2) PL21 (42). Sun et al. also used two cell lines (1): MKN45 and (2) MGC803 (34).

The median survival varied among studies with a clear tendency towards a higher survival probability compared to the combined negative group at the study endpoint ( Figure 6A ). However, substantial between-study heterogeneity was observed, with an I² value of 93.3% (95% CI: 91.2%; 94.8%). Graphic Display of Heterogeneity (GOSH) plots identified three influential outliers (1, 26, 28, 41). Excluding these outliers, the I² value remained high at 93.8% (95% CI: 91.8%; 95.3%), indicating considerable heterogeneity presumably related to the use of different modular CAR-T platforms across studies, as well as the different target antigens and experimental setups. Comparing to the positive control group, there is no significant difference in survival rate, with a minor tendency toward a lower survival rate ( Figure 6B ). Heterogeneity in this analysis was moderate, with an I² value of 63.8% (95% CI: 18.2%; 84.0%). No influential outliers were identified in the GOSH plots.

Body weight was assessed either in grams or as a percentage change in body weight at the study endpoint. Eight studies provided data on this measure, and a standardized mean difference was computed to accommodate the varied measurement scales [citations]. The findings indicated either a decrease in body weight or no change compared to the combined negative control group. This trend aligns with the reduction in tumor burden and/or tumor volume observed in the experimental group ( Supplementary Figure 1A ).

Understanding the cytokine profile of CAR T cells is crucial for determining efficacy and toxicity. Certain cytokines, such as IL-2, IL-15, IFN-γ, and TNF-α, can promote CAR T cell expansion and killing, activating other immune populations. Others, like IL-6, IL-10, and TGF-β, can inhibit antitumor immune function. Human cytokine levels in peripheral blood were measured in picograms per milliliter (pg/mL), with data reported in 7 out of 33 studies. Among these, 3 studies compared cytokine levels to a conventional CAR-T against the target antigen as a positive control. All the assessed cytokines were pro-inflammatory, including IFN-γ, TNF, IL-2, IL-18, and IL-6. IFN-γ was reported in all seven studies, TNF in 4 out of 7, IL-2 in 3 out of 7, and IL-18 and IL-6 were each reported in one study. Standardized mean differences were calculated, considering either a positive control ( Supplementary Figure 1B ) or a combined negative control ( Supplementary Figure 1C ). Notably, cytokine expression was generally higher in the experimental group compared to the negative control group, while expression levels were lower or showed no significant change compared to the positive control group. In the study by Liu et al. cytokine production was noticeably lower than conventional CAR-T cells (51).

The percentage of tumor-free mice is another parameter to assess the ability of the CAR-T platforms to eliminate tumors. This outcome measure was calculated as the proportion of mice without tumors relative to the initial number of animals at the start of the experiment. It was assessed in only two studies: one using anti-FITC CAR targeting EGFR with the soluble module FITC-cetuximab (Ctx), and another using SpyCatcher CAR-T targeting hGPC3 with anti-hGPC3 scFv-SpyTag (51, 53). Both studies showed that the mice treated with modular CAR-T remained tumor-free until the end of the experiment. Tamada et al. found that anti-FITC CAR with FITC-Ctx effectively prevented tumor formation until day 26 in the experimental group, while control groups had tumors by day 15 (53). In Liu et al.’s study, the highest concentration of anti-hGPC3 scFv-SpyTag kept mice tumor-free for 50 days, whereas lower concentrations and control groups did not (51).

Five studies explored modular CAR-T therapy efficacy against tumor metastasis in different experimental models. Lu et al. utilized a xenograft model of acute myeloid leukemia (AML) to investigate metastasis, where THP1-FRβ tumor cells were administered intravenously, leading to widespread dissemination of the tumor cells, forming both liver and non-liver metastatic lesions, most of which localized to the mouse ovary. Treatment with Spy-Catcher CAR cells plus EC17 effectively controlled liver tumor metastases compared to control groups (54). In contrast, Meyer et al. found that UniCAR-T-treated animals developed extramedullary AML disease. The treated animals exhibited numerous subcutaneous and organ metastases, emphasizing the aggressive nature of the disease. All studied metastases had a high CD33⁺ AML chimerism and CD123 positivity (37).

Ochi et al. observed prolonged survival in mice treated with cCD16ζ-T cells and rituximab, with significant suppression of tumor growth in the liver, spleen, lung/heart, and uterus/ovary/fallopian tube compared to control groups (38). Raj et al. used switchable CAR-T cells targeting HER2 in a patient-derived xenograft model of stage IV advanced pancreatic ductal adenocarcinoma (PDAC) to simulate the significant liver and lung metastases seen at this stage. Throughout the experiment, the mice treated with sCAR-T cells and HER2-specific switches remained tumor-free, as did mice treated with conventional HER2 CAR-T, demonstrating that the sCAR platform is promising against aggressive and diffuse tumors derived from patients with advanced PDAC (21).

Lastly, Sun et al. mimicked clinical scenarios by establishing a disseminated peritoneal tumor model. They found that combined administration of the soluble module F-AgNPs and TRUE CAR-T cells effectively controlled tumor growth and cleared metastatic regions (34).

The earliest platform was developed by Powell Jr. lab in the University of Pennsylvania and it is based on the high-affinity interaction between avidin and its natural binding partner, biotin. The chimeric receptor, called biotin-binding immune receptor (BBIR), consists of the second-generation CAR structure with avidin as the ectodomain so it specifically recognizes biotinylated antibodies. Interestingly, CAR T-cells were only activated when biotinylated antibodies bound to their target. Free biotin, on the other hand, can bind to the receptor and render CAR T-cells inactive (55). This strategy has been proven effective in the elimination of tumors expressing EpCAM and CD20 both in vitro and in vivo (55, 56). However, Avidin is considered xenogeneic, thus there are some concerns about its immunogenicity. The authors claimed that the presence of a preconditioned environment (lymphocyte depletion) can reduce the risk of developing inhibitory immunogenicity (55).

One of the innovative universal CAR platforms is called multiple component-logically gated CAR or CO-LOCKER CAR. It uses the Latching Orthogonal Cage–Key pRotein (LOCKR) switch which consists of a structural “Cage” protein that uses a “Latch” domain to sequester a functional peptide in an inactive conformation until binding of a separate “Key” protein induces a conformational change that permits binding to an “Effector” protein. This allows for the logical gating of CAR activity (AND, OR, NOT gates). The platform was only tested in vitro as a “proof-of-principle”. In vivo studies, however, have not been conducted yet (57).

The AdCAR-T platform employs a two-component signal transduction system based on a split recognition/activation design, where labeled monoclonal antibodies transmit antigen recognition into T-cell activation via an anti-label CAR. The soluble module in this system is generated through biotinylation using specific linker chemistry, resulting in a molecule comprising an antigen-binding moiety, a linker moiety, and a label moiety (biotin). The AdCAR is based on the unique characteristics of the monoclonal antibody mBio3, which binds to biotin in the context of a specific linker, referred to as a Linker-Label Epitope (LLE). The authors tested the in vivo activity of this platform using a rapidly progressive xenograft model of Burkitt’s lymphoma (Raji cell line) and utilized the therapeutic mAb Rituximab as the antigen-binding domain to demonstrate the potential for direct clinical translation of the AdCAR-T technology. They did not observe signs of GvHD in the mouse models. Tumor burden was assessed by in vivo bioluminescence imaging (BLI). Remarkably, AdCAR-T in combination with LLE-rituximab completely eradicated disseminated lymphoma, proving as efficient as conventional CD20-CAR-T, although with slightly delayed kinetics. Mice remained in complete remission, as demonstrated by BLI and flow cytometry of bone marrow, even after LLE-rituximab administration was terminated. In contrast, neither AdCAR-T nor LLE-rituximab alone had a significant effect on tumor burden (58).

The ARC-SparX platform combines a genetically modified T cell expressing an antigen-receptor complex (ARC-T) with a soluble protein antigen-receptor X-linker (SparX) that links the ARC-T to tumor cells. ARC-T cells feature a 73 amino acid synthetic protein D-domain in their extracellular domain, specifically binding to the TAG domain within the SparX adapter, which includes the third domain of human alpha-fetoprotein (AFP domain III) due to its stability and human origin. The SparX adapter also utilizes D-domains engineered to bind various tumor antigens. The platform’s selectivity ensures ARC-T cell activation only in the presence of SparX and antigen-expressing tumor cells.

A key advantage of the ARC-SparX platform is the controllability of ARC-T cells through SparX protein administration. In vivo studies on NALM6-BCMA-bearing NSG mice revealed that BCMA SparX proteins bound to tumor cells were detectable up to 8 hours post-injection. Cytokine levels in serum, such as IFN-γ, peaked at 8 hours and dropped to baseline within 48 hours after BCMA SparX administration, demonstrating intermittent cytokine activation. This pattern differed from the traditional BCMA-CAR, where cytokine levels steadily increased.

Comparative studies using a xenograft model showed that ARC-T cells with daily doses of bivalent BCMA SparX cleared tumors as effectively and rapidly as BCMA-CAR. Both cell types expanded in vivo to a peak on day 7, with BCMA-CAR cells expanding more extensively. At peak expansion, both cell types displayed similar effector phenotypes and increased effector memory (TEM) and effector T cells (TEMRA), while reducing stem-cell memory phenotype T cells (59).

The main advantage of the conduit CAR platform is that it does not require the introduction of any novel engineered antigen on T cells, but rather utilizes an existing feature present in most clinical CAR T-cell therapies, the flexible ScFv linker. The soluble module is a bispecific antibody (BsAb) that targets the (GGGGS)n or (G4S)n linker found on most existing CARs and the tumor-associated antigen. Borrok et al. demonstrated that CAR T cells expressing either a germline antibody ScFv (with no known specificity) or a CD19-targeting CAR can be redirected to target prostate tumor cells via a bispecific soluble module. This technology offers clinical advantages due to its adaptability to target novel TAAs, potential toxicity control through dosage adjustment, and compatibility with existing clinical CARs. However, the study did not include in vivo experiments (60).

This platform relies on the interaction between folate and the folate receptor and combines the application of a bispecific antibody (BsAb) with T-cells genetically engineered to express a unique BsAb-binding immune receptor (BsAb-IR). The BsAb-IR consists of a portion of an extracellular folate receptor (FR; 231aa) fused to intracellular TCR and CD28 costimulatory signaling domains in tandem. It can be bound and activated by an anti-FR antibody arm of a unique BsAb that bridges the FR and tumor antigen (frBsAb). The study included a proof-of-concept in vitro experiment showing that tumor antigen-specific frBsAbs bind specifically to target antigens on human tumor cells. Upon co-engagement of the BsAb-IR on engineered T-cells, this binding delivers simultaneous TCR CD3 activation and CD28 costimulation signals in a target-dependent manner. This results in the selective augmentation of activation, proliferation, and antitumor activity of the BsAb-IR T-cell subset. However, no in vivo studies were conducted (61).