Computationally guided cis-targeting design of the TrME enhances logic-gated macrophage activation
Productive macrophage-mediated phagocytosis of tumor cells is strictly predicated on the ratiometric activating and inhibitory signaling within individual macrophages (Fig. 1a). CRT specifically binds to asialoglycoproteins containing triantennary and multivalent type II (Galβ1→4GlcNAc) chain epitopes (Tri/m-II) on target cells, facilitating recognition by macrophage LRP1 (refs. 36,39). This interaction initiates downstream signaling cascades that promote phagocytosis40 (Extended Data Fig. 1a). CD47 inhibits phagocytosis through ligation of SIRPα expressed on phagocytes, leading to activation of tyrosine phosphatases and inhibition of myosin accumulation at the submembrane assembly site of the phagocytic synapse. To further investigate the interplay between CRT–LRP1 and CD47–SIRPα signaling, we incubated macrophages with fluorescent beads conjugated to varying ratios of recombinant murine CRT and CD47. We observed a positive correlation between phagocytic efficiency and the median fluorescence intensity ratio (MFICRT/MFICD47) on the fluorescent beads, indicating that the prophagocytic CRT–LRP1 and antiphagocytic CD47–SIRPα signals functioned in a ratio-dependent manner (Extended Data Fig. 1b). To enhance macrophage cytotoxicity by achieving ratiometric activating and inhibitory signaling within individual macrophages, we sought to establish an AND logic gate on the surface of single effector cells that simultaneously activates LRP1 and blocks SIRPα. We predicted the spatial structure of CRT using AlphaFold and identified a rigid α-helical region at its C terminus that tended to spatially separate adjacent domains and thereby promote interactions between different cells rather than within the same cell (Fig. 1b). As we aimed to enable the cis-targeting of a single macrophage by fusing the anti-SIRPα scFv to CRT, we required a linker with appropriate flexibility to permit coordinated cis-engagement. To avoid interference from the C-terminal rigid helix, the anti-SIRPα scFv was, therefore, fused to the N terminus of CRT. We simplified the format architecture by tandemly linking monovalent CRT and anti-SIRPα scFv (anti-SIRPα–CRT), thereby enabling us to modulate their mechanism of action toward stable cis-targeting through the linkers (Fig. 1c). This design allowed for logical gating of LRP1 activation and SIRPα blockade, while avoiding the undesired side effects of trans-targeting, such as cell crosslinking and lack of synergy (Extended Data Fig. 1c). The mechanism of action of anti-SIRPα–CRT is critically determined by the length and flexibility of linkers, which dictate the structural stability and spatial orientation of its constituent domains. Classical flexible (GGGGS)n and rigid helical (EAAAK)n linkers are insufficient for fine-tuning flexibility. Therefore, to screen optimal linkers for stable cis-targeting, we designed a chimeric linker library of varying lengths based on the GGGGS and EAAAK motifs, encompassing a full spectrum of flexibility, from highly flexible (GGGGS)n to rigid (EAAAK)n (Fig. 1d).
Fig. 1: Overall design of the TrME.
a, Schematic diagram of the efficient macrophage activation program. Macrophages determine the initiation of effector functions by integrating signals sensed and transduced through activating and inhibitory receptors. A high ratio of activating to inhibitory signals is critical for triggering phagocytosis in macrophages. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). b, The structure of CRT, predicted by AlphaFold. pLDDT, predicted local distance difference test. c, Design of anti-SIRPα–CRT for cis-targeting. Anti-SIRPα–CRT cotargets SIRPα and LRP1 on the surface of macrophages to establish an ‘activate and block’ logic gate, inducing a proportional input of activating and inhibitory signals and thereby robustly activating macrophages. The anti-SIRPα–CRT mechanism of action can be precisely directed by adjusting the linker properties. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). d, Design of the chimeric linker library. Each chimeric linker consisted of n(2–7) motifs, with each motif randomly selected from either a flexible motif F (GGGGS) or a rigid motif R (EAAAK). This design generated a linker library spanning the full spectrum of flexibility, from flexible (GGGGS)n to rigid (EAAAK)n, allowing optimization of linker length and rigidity to enable stable cis-targeting by anti-SIRPα–CRT. e, Schematic diagram of the three-tiered computational workflow for identifying cis-targeting anti-SIRPα–CRT constructs. In the first tier, we used AlphaFold to predict the spatial conformation of the anti-SIRPα–CRT–SIRPα complex and measured the minimum distance between SIRPα and CRT. A 3.5-Å cutoff was applied as the criterion. In the second tier, we calculated the angle formed among three key points in the most stable conformation of anti-SIRPα–CRT constructs: the SIRPα-binding site of anti-SIRPα scFv, the centroid of the chimeric linker and the LRP1-binding site of CRT. Constructs exhibiting an angle > 90° were excluded. In the third tier, 200-ns MD simulations were performed for anti-SIRPα–CRT constructs to evaluate the temporal stability of the cis-orientation. f, The minimum distance between SIRPα and CRT in the anti-SIRPα–CRT–SIRPα complex. Detailed candidates are presented in Supplementary Table 1. g, The angle formed by the three key points of the most stable anti-SIRPα–CRT construct conformation: the SIRPα-binding site of anti-SIRPα scFv, the centroid of the chimeric linker and the LRP1-binding site of CRT. Detailed candidates are presented in Supplementary Table 1. h, The 200-ns MD simulations of anti-SIRPα–CRT constructs. i, Representative flow cytometry and quantitative analyses of the MFILRP1-act in CFSE-labeled SIRPα-preblocked macrophages, which were cocultured with CTR-labeled macrophages, under treatment with anti-SIRPα–CRT (0.05 μg ml−1) for 0.5 h (n = 6 cell samples per group, biological replicates). Box plots show the median (center line), the 25th and 75th percentiles (box bounds) and minimum and maximum values (whiskers). Individual data points are shown. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). j, Confocal images of macrophages incubated with CD47-conjugated fluorescent beads and anti-SIRPα–CRT for 2 h (n = 5 cell samples per group, biological replicates). Green, beads; magenta, macrophage membrane; blue (DAPI), nuclei. k, Design of the TrME. TrME redirects anti-SIRPα–CRT-activated macrophages to tumors through anti-TAA specificity. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). l, Flow cytometry analysis of the phagocytosis of GL261 cells by BMDMs after 4 h of incubation in the presence of TrMEB7H3 (0.1 μg ml−1).
Source data
We established a three-tiered computational screening workflow to systematically identify anti-SIRPα–CRT constructs capable of achieving cis-targeting (Fig. 1e). In the first tier, we used AlphaFold to predict the spatial configuration of the anti-SIRPα–CRT–SIRPα complex and measured the minimum atomic distance between SIRPα and CRT. A 3.5-Å threshold was selected as the cutoff, representing the upper limit of the van der Waals interaction distance—a well-established standard in protein–protein interaction modeling. Constructs with interdomain distances greater than 3.5 Å were considered to exhibit minimal steric interference between SIRPα and CRT, ensuring the potential for independent engagement of anti-SIRPα and CRT on the same macrophage surface (Fig. 1f and Extended Data Fig. 1d). In the second tier, we quantified the spatial orientation of each anti-SIRPα–CRT construct by calculating the angle formed between three key points in its lowest-energy conformation: the SIRPα-binding site of anti-SIRPα scFv, the centroid of the chimeric linker and the LRP1-binding site of CRT. Constructs with angles <90° were predicted to orient both functional domains toward the same side of the molecule, favoring cis-activation within a single macrophage, whereas those with larger angles were predicted to adopt trans-like geometries and were excluded from further analysis (Fig. 1g and Extended Data Fig. 1e). In the third tier, we conducted 200-ns molecular dynamics (MD) simulations for all the anti-SIRPα–CRT constructs with predicted angles < 90° to assess the temporal stability of their cis-orientation. Among all the candidates, the FRF-linked anti-SIRPα–CRT construct maintained a cis-oriented configuration (<90° interdomain angle) throughout 100% of the simulation time, demonstrating the highest conformational stability (Fig. 1h). On the basis of these results, the FRF linker was selected as the optimal design for promoting stable cis-targeting between the anti-SIRPα and CRT modules.
To confirm that the FRF-linked anti-SIRPα–CRT construct had the highest cis-targeting efficiency, we developed a cis-versus-trans assay. The macrophages were divided into two groups and labeled with two distinct dyes: carboxyfluorescein succinimidyl ester (CFSE) and Cell Tracker red (CTR). The CFSE-labeled macrophages were further subdivided into two groups, one of which was preincubated with a saturating concentration of the parental anti-SIRPα scFv before coculture with CTR-labeled macrophages. Each CFSE-labeled macrophage subgroup was then coincubated with CTR-labeled macrophages under different treatment conditions and the levels of activated LRP1 (LRP1-act) induced by the CRT–LRP1 interaction were subsequently analyzed across the four macrophage populations. For ideal cis-targeting anti-SIRPα–CRT, SIRPα preblockade reduced CRT–LRP1 binding, thereby lowering the level of LRP1-act. Before conducting the cis-versus-trans assay, we verified that all the anti-SIRPα–CRT constructs bound to macrophages preincubated with anti-SIRPα scFv at comparable efficiencies (Extended Data Fig. 2a). Moreover, fluorescence analysis using anti-SIRPα–FITC confirmed only minimal and statistically insignificant binding to macrophages that were pretreated with anti-SIRPα scFv (Extended Data Fig. 2b). These findings indicated that the anti-SIRPα–CRT constructs might engage in minimal competition with the anti-SIRPα scFv used for blocking but such minor interactions did not compromise the integrity of cis-signal versus trans-signal discrimination. To further validate the accuracy of our computational screening workflow, in addition to the FRF construct, we selected several control linkers that were excluded at different tiers of the screening workflow: the FFFF linker (excluded at the first tier because of a minimum interdomain distance < 3.5 Å), the RRR linker (excluded at the second tier for forming an interdomain angle > 90°) and the FF, RF and FFR linkers (excluded at the third tier for various stabilities of cis-angles during MD simulations). These constructs were included as controls in the cis-versus-trans assay. Upon exposure to the six anti-SIRPα–CRT variants, the FRF-linked construct induced the most significant decrease in LRP1-act levels in CFSE-labeled macrophages preblocked with anti-SIRPα scFv, reaching levels nearly equivalent to those observed with CRT stimulation alone. In contrast, the MFILRP1-act in CFSE-labeled macrophages without SIRPα blockade remained consistent with that in CTR-labeled macrophages (Fig. 1i). These results demonstrated that the FRF-linked anti-SIRPα–CRT construct preferentially delivered CRT-mediated agonistic signaling in cis on the same macrophage upon SIRPα binding, rather than trans-targeting neighboring cells. Furthermore, treatment with the FFFF-linked construct resulted in lower levels of LRP1-act in all four macrophage populations than in those induced by CRT alone, indicating potential interference between the anti-SIRPα and CRT modules. The level of LRP1-act in the presence of the RRR-linked construct showed almost no reduction in CFSE-labeled macrophages preblocked with anti-SIRPα scFv, suggesting stable trans-targeting. The FF-linked, RF-linked and FFR-linked constructs reduced LRP1-act levels to varying degrees in CFSE-labeled macrophages preblocked with anti-SIRPα scFv, reflecting different levels of cis-targeting efficiency. Collectively, these findings further validated the accuracy and predictive value of our computational modeling workflow.
Next, we prepared SIRPα-knockout and LRP1-knockout macrophages, mixed them and treated them with different anti-SIRPα–CRT constructs. We then measured LRP1-act levels in the SIRPα-knockout macrophage subset, which could only occur through trans interactions. The results showed that FRF-linked anti-SIRPα–CRT exhibited the lowest level of trans-induced LRP1 activation, which was comparable to the baseline observed with soluble CRT treatment alone. These findings further indicated that the FRF linker configuration favored cis-engagement within individual macrophages, thereby minimizing trans-signaling (Extended Data Fig. 2c,d). To evaluate the ability of anti-SIRPα–CRT to modulate phagocytosis-associated signaling, we incubated anti-SIRPα–CRT with macrophages and added beads conjugated with recombinant murine CD47. Flow cytometry was used to assess LRP1-act induced by CRT–LRP1 interaction and activated SIRPα (SIRPα-act) triggered by CD47–SIRPα signaling. We observed that FRF-linked anti-SIRPα–CRT induced a significant increase in MFILRP1-act /MFISIRPα-act, indicating logically gated control of ratiometric activation and inhibitory signaling in macrophages (Extended Data Fig. 2e–g). Moreover, FRF-linked anti-SIRPα–CRT significantly enhanced the uptake of CD47-conjugated fluorescent beads (Fig. 1j and Extended Data Fig. 2h,i). We further generated a reverse fusion construct by linking the anti-SIRPα scFv to the C terminus of CRT through the FRF linker (CRT–anti-SIRPα). This reverse fusion markedly reduced the ratiometric signaling associated with macrophage activation and induced macrophage aggregation, indicating that the rigid C-terminal α-helix of CRT disrupted the optimal flexibility provided by the FRF linker, thereby impairing cis-targeting efficiency (Extended Data Fig. 2j,k). By contrast, among the CRTtrun–anti-SIRPα constructs generated by fusing anti-SIRPα scFv to the C-terminal–helix–truncated CRT using different linkers, the FRF-linked construct consistently mediated the highest macrophage uptake of CD47-conjugated fluorescent beads (Extended Data Fig. 2l,m). Collectively, these data demonstrated that the FRF linker conferred optimal flexibility for CRT and anti-SIRPα scFv to preferential cis-target the same macrophage, thereby establishing a ratiometric balance between prophagocytic and antiphagocytic signaling and effectively initiating the phagocytic response.
To achieve tumor-specific targeting, we focused on leveraging TAAs. In glioblastoma (GBM), a tumor type characterized by extensive macrophage infiltration (Supplementary Fig. 1), B7H3 is highly expressed in both surgical specimens and cell lines and its expression correlates significantly with survival outcomes (Supplementary Figs. 2–4). Our data revealed that B7H3 expression was markedly increased in tumor cells, whereas the expression of B7H3 was only minimal or negligible in endothelial cells, fibroblasts, macrophages, T cells and neutrophils (Supplementary Fig. 5). These results strongly supported the specificity and safety of targeting B7H3. Therefore, we tended to engineer TrMEB7H3 by fusing anti-B7H3 scFv to anti-SIRPα–CRT. Ensuring that the anti-B7H3 scFv does not interfere with the cis-targeting ability of the anti-SIRPα scFv and CRT was essential. To accomplish this requirement, a linker was introduced to maintain sufficient spatial separation while allowing flexible binding with tumor cells. The C-terminal α-helix of CRT naturally provided a rigid spacer that stabilized the spatial orientation of adjacent modules, effectively separating the macrophage activation module from the anti-B7H3 scFv. To introduce appropriate flexibility for tumor engagement, we further connected the anti-B7H3 scFv after this helical region using a short chimeric linker (Fig. 1k). Next, macrophages were coincubated with CFSE-labeled GL261 cells and treated with TrMEB7H3. Flow cytometry analysis revealed that TrMEB7H3 containing the RF chimeric linker mediated the most significant macrophage uptake of GL261 cells, indicating that the RF linker provided optimal flexibility for anti-B7H3 scFv binding to tumor cells (Fig. 1l). Next, we constructed several TrMEB7H3 variants: (1) CRT lacking the C-terminal α-helix, either directly fused to or connected through an RF linker to the anti-B7H3 scFv; (2) CRT with a half-truncated C-terminal α-helix, directly or RF-linked to the anti-B7H3 scFv; and (3) CRT with a full-length C-terminal α-helix, directly or RF-linked to the anti-B7H3 scFv. Comparative analysis revealed that retaining the complete C-terminal α-helix was crucial; constructs with partially or fully truncated helices showed reduced MFILRP1-act/MFISIRPα-act ratios, whereas the untruncated variants maintained activation levels comparable to those of the anti-SIRPα–CRT module. These results indicated that the intact α-helix effectively prevented spatial interference between the tumor-targeting and macrophage activation modules, ensuring that incorporation of the anti-TAA scFv did not compromise the ability of the TrME to establish a logic gate for ratiometric activation and inhibitory signaling (Extended Data Fig. 3a). Thus, the final form of TrMEB7H3 was constructed by fusing anti-SIRPα scFv and anti-B7H3 scFv to the N terminus and C terminus of CRT, respectively, using the FRF and RF chimeric linkers. To assess the superiority of cis-targeting, we introduced recombinant mouse CD47-conjugated beads into the macrophage and CFSE-labeled GL261 cell coculture system. We generated a trans-biased TrMEB7H3 variant by incorporating the RRR linker, which was identified as a trans-favoring configuration, as a control. In this assay, the antitumor efficacy of TrMEB7H3 was found to remain statistically unchanged upon the addition of CD47-conjugated beads. In contrast, the macrophage phagocytic efficiency induced by RRR-linked TrMEB7H3 decreased to 59% of the original level following the addition of CD47-conjugated beads. Furthermore, TrMEB7H3 induced minimal phagocytic activity in phagocytosis assays using red blood cells as targets, regardless of the presence or absence of CD47-conjugated beads (Extended Data Fig. 3b,c). These findings indicated that TrMEB7H3 effectively overcame the suppressive effects of microenvironmental ‘don’t eat me’ signals on macrophages and specifically promoted macrophage-mediated clearance of target cells expressing the corresponding TAA.
TrMEB7H3 resists immunosuppressive signaling and reprograms macrophages for antitumor immunity
For the large-scale production of TrMEB7H3, the culture supernatant from stably transfected Expi293F cells was purified using a Ni-NTA resin. The eluted fractions were analyzed by SDS–PAGE, revealing a band (~99 kDa) corresponding to TrMEB7H3, as confirmed by immunoblotting with anti-CRT and anti-6×His antibodies (Fig. 2a). The purified TrMEB7H3-protein preparation was analyzed by SDS–PAGE followed by Coomassie blue staining, which revealed a single major band corresponding to the expected molecular weight of TrMEB7H3. No detectable nonspecific or contaminating bands were observed, confirming the high purity of the TrMEB7H3-protein preparation (Supplementary Fig. 6). To evaluate tumor-specific targeting, TrMEB7H3 was incubated with GL261 cells and B7H3-knockout GL261 cells (GL261B7H3⁻ cells) (Supplementary Fig. 2). Cell imaging demonstrated that TrMEB7H3 exhibited selective binding affinity for B7H3high GL261 cells (Fig. 2b), confirming its specificity for tumor cells expressing high levels of the corresponding TAA. Next, we further generated macrophages with single knockouts of LRP1 or SIRPα and double knockouts of LRP1 and SIRPα. Flow cytometry analysis demonstrated that TrMEB7H3 bound to macrophages and tumor cells in a target-specific manner, with a marked reduction in binding observed in the corresponding knockout cell lines (Fig. 2c). To assess whether TrMEB7H3 facilitates the recognition and conjugation of macrophages with tumor cells, GL261 cells were coincubated with primary mouse bone-marrow-derived macrophages (BMDMs) at a 1:2 ratio. Confocal laser scanning microscopy revealed minimal interaction between tumor cells and BMDMs in the control group. However, treatment with 0.05 μg ml−1 TrMEB7H3 resulted in approximately 90% of BMDMs forming conjugates with tumor cells (Fig. 2d). These results demonstrated that TrMEB7H3 specifically bound to tumor cells in a TAA-dependent manner, prompting macrophages to recognize and engage these cells.
Fig. 2: TrME induces the targeted phagocytosis of cancer cells and subsequent macrophage activation.
a, Western blotting of purified fractions of TrMEB7H3 using anti-CRT and anti-6×His antibodies (n = 3 samples per group, biological replicates). b, Confocal images of B7H3high GL261 cells and GL261B7H3⁻ cells after incubation with TrMEB7H3 (0.1 μg ml−1) (n = 5 cell samples per group, biological replicates). Green, 6×His; magenta, membrane; blue (DAPI), nuclei. c, Flow cytometric analysis of the binding efficiency of TrMEB7H3 (0.1 μg ml−1) to GL261 cells and RAW264.7 cells, as determined using Alexa Fluor 488–His tag antibody. KO, knockout; DKO, double knockout. d, Confocal images and quantitative analyses of macrophage–tumor cell interaction under the treatment of TrMEB7H3 (0.05 μg ml−1). P values were determined using a two-sided unpaired t-test. Error bars denote the mean ± s.d. (n = 3 cell samples per group, biological replicates). e,f, Flow cytometry (e) and quantitative analysis (f) (n = 3 cell samples per group, biological replicates) of phagocytosis of B7H3-labeled beads or B7H3/CD47 double-labeled beads by BMDMs (after 2 h of incubation) in the presence of TrMEB7H3 (0.1 μg ml−1). P values were determined using a two-way ANOVA. Error bars denote the mean ± s.d. g, Representative dose–response curve showing TrMEB7H3-induced phagocytosis of B7H3/CD47 double-conjugated beads by BMDMs after 2 h of incubation. Error bars denote the mean ± s.d. (n = 3 cell samples per group, biological replicates). h, Quantification of the phagocytosis of GL261, GL261B7H3⁻, MB49 and 4T1 cells by BMDMs (after 4 h of incubation) in the presence of TrMEB7H3 (0.1 μg ml−1). P values were determined using a two-way ANOVA. Error bars denote the mean ± s.d. (n = 5 cell samples per group, biological replicates). i, Quantification of phagocytosis of GL261 cells by BMDMs (after 4 h of incubation) in the presence of TrMEB7H3 (0.1 μg ml−1). Blockade validation was performed using anti-B7H3 and anti-CRT antibodies (10 μg ml−1), as well as recombinant SIRPα protein (10 μg ml−1). P values were determined using a one-way ANOVA. Error bars denote the mean ± s.d. (n = 5 cell samples per group, biological replicates). j, Confocal images of the phagocytosis of U87 human GBM cells by human THP-1 cells in the presence of human TrMEB7H3 (n = 5 cell samples per group, biological replicates) Magenta, U87 cells; green, THP-1 cells. k, Flow cytometry analysis of U87 cell phagocytosis by THP-1 cells and GL261 cell phagocytosis by BMDMs (after 4 h of incubation) in the presence of TrMEB7H3 (0.05 μg ml−1). Boxes in the plots indicate the percentage of THP-1 cells and BMDMs that phagocytized tumor cells. l, Left, flow cytometry analysis of M1 macrophage markers (CD86) and M2 macrophage markers (CD206) in BMDMs following coincubation with TrMEB7H3 and GL261 cells. Right, quantification of the M1/M2 cell ratio. P values were determined using a one-way ANOVA. Error bars denote the mean ± s.d. (n = 3 cell samples per group, biological replicates). m, Volcano plot of differentially expressed genes in sorted BMDMs following coincubation with TrMEB7H3 and GL261 cells. Differential gene expression was analyzed using DESeq2 with two-sided statistical testing. P values were adjusted for multiple comparisons using the Benjamini–Hochberg false discovery rate method. Genes with Padj < 0.05 and |log2(fold change)| ≥ 1 were considered significantly differentially expressed. The deep-purple circles indicate significantly upregulated marker genes and the lilac circles indicate significantly downregulated marker genes (n = 3 cell samples per group, biological replicates). FC, fold change. n, Hierarchical clustering of gene expression in sorted BMDMs following coincubation with TrMEB7H3 and GL261 cells (n = 3 cell samples per group, biological replicates). NS, not significant (P > 0.05); *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Exact P values are indicated in the graph.
Source data
To evaluate the resilience of TrMEB7H3 to phagocytic inhibition mediated by the CD47–SIRPα axis, fluorescent beads conjugated with recombinant B7H3 and recombinant CD47 (B7H3:CD47 = 1:2) were used as phagocytic targets. TrMEB7H3-mediated phagocytosis of B7H3-labeled and CD47-labeled fluorescent beads was substantially greater than that induced by any dual-component combination, CRT–anti-B7H3 + anti-SIRPα antibody or RRR-linked TrMEB7H3 (Fig. 2e,f). These results clearly demonstrated that cis-configured TrMEB7H3 activated macrophages more effectively than its trans counterpart and each domain (anti-SIRPα, CRT and anti-B7H3) had a distinct and indispensable role in coordinating efficient macrophage phagocytosis, underscoring the functional necessity of integrating all three modules within a single molecule. Removal of CD47 from the fluorescent beads eliminated the observed differences between groups, indicating that TrMEB7H3 effectively disrupts the CD47–SIRPα axis. In the phagocytosis dose–response assay, a Y-shaped anti-SIRPα/B7H3 bsAb with an active Fc domain was used as a control, representing an early-generation cis-designed bsAb that inadvertently exerted its activity through a trans mechanism23,41,42. TrMEB7H3 induced macrophage-mediated engulfment of B7H3-labeled and CD47-labeled fluorescent beads in a dose-dependent manner. At concentrations of 0.1 nM, TrMEB7H3-mediated enhancement of macrophage phagocytosis reached its maximum, whereas the rigid anti-SIRPα/B7H3 bsAb (with an active Fc domain) achieved its peak effect only at a tenfold higher concentration, with a maximum response that was only 64% of that induced by TrMEB7H3 (Fig. 2g). We then cultured macrophages in ultralow-attachment plates in serum-free medium and treated them with TrMEB7H3. Cell imaging demonstrated that, compared to the anti-SIRPα/B7H3 group, the TrMEB7H3-treated group exhibited a significantly lower rate of macrophage crosslinking, indicating that TrMEB7H3 effectively minimized intercellular aggregation (Extended Data Fig. 3d). In addition, we generated a CD70-targeting TrME variant (TrMECD70). Consistent with the above findings, TrMECD70 markedly enhanced the phagocytosis of CD70-expressing GL261 cells compared with the anti-SIRPα/FcγR/CD70 antibody and resulted in significantly decreased macrophage crosslinking (Extended Data Fig. 3d,e). These findings highlighted the superiority of the cis-targeting TrME design in enhancing macrophage activation over early-generation cis-designed bispecific antibodies, particularly at lower doses, while minimizing intercellular aggregation. Next, we evaluated the specificity of TrMEB7H3-mediated phagocytosis in tumor cell lines with varying B7H3 expression profiles (Supplementary Fig. 2). No significant increase in macrophage-mediated phagocytosis was observed against GL261B7H3⁻ cells or B7H3low 4T1 murine breast cancer cells under TrMEB7H3 treatment and anti-SIRPα–CRT alone failed to enhance macrophage phagocytosis of tumor cells. These findings confirmed that the TrME-mediated modulation of macrophage phagocytosis depended on the tumor-targeting module recognizing TAAs expressed on tumor cells (Fig. 2h). Consistently, blocking the interaction between macrophages and cancer cells with recombinant SIRPα protein, anti-B7H3 antibodies or anti-CRT antibodies abolished the prophagocytic effect of TrMEB7H3 (Fig. 2i). Furthermore, incorporating anti-human B7H3 and SIRPα scFvs, along with human CRT, into TrMEB7H3 induced antigen-specific phagocytosis of human GBM U87 cells by THP-1 cells (Fig. 2j,k).
After phagocytosing tumor cells, macrophages present tumor-derived antigens and elicit T cell responses43. To determine whether TrMEB7H3 bridges the innate and adaptive immune responses, GL261 cells expressing cytoplasmic ovalbumin (GL261-cOVA) were cocultured with BMDMs. Consistent with enhanced macrophage phagocytosis, BMDMs cocultured with B7H3high GL261-cOVA cells exhibited increased presentation of OVA peptide on major histocompatibility complex I (MHC I; the H2kb–SIINFEKL complex) following TrMEB7H3 treatment (Extended Data Fig. 3f). ELISpot analysis demonstrated that, when T cells from OT-Ⅰ mice were introduced into the coculture system of BMDMs and GL261-cOVA cells, TrMEB7H3 significantly enhanced T cell activation, as indicated by increased interferon-γ (IFNγ) production (Extended Data Fig. 3g,h).
In cellular imaging of phagocytosis, most of the macrophages treated with TrMEB7H3 exhibited a ‘fried egg’ morphology (round-shaped cells with large nuclei centered in the cytoplasm), resembling the morphology of M1 macrophages induced by combined LPS and IFNγ. By contrast, untreated macrophages displayed a slender morphology, similar to that of M2 macrophages induced by interleukins 4 and 13 (IL-4 and IL-13)44 (Fig. 2j, Extended Data Fig. 3f and Supplementary Fig. 7). To investigate the phenotypic changes in macrophages, we analyzed macrophages cocultured with GL261 cells after TrMEB7H3 treatment. Flow cytometry revealed a significant increase in the M1 macrophage marker CD86 and a decrease in the M2 macrophage marker CD206, with an average M1/M2 ratio of 4.43 (Fig. 2l). RNA sequencing data showed that TrMEB7H3-treated macrophages clustered distinctly from untreated macrophages, with upregulation of numerous M1 macrophage-related genes and downregulation of M2 macrophage-related genes (Fig. 2m,n and Extended Data Fig. 4a). We further examined the transcriptional changes in BMDMs treated with TrMEB7H3 alone, using untreated BMDMs and BMDMs cocultured with GL261 cells in the presence of TrMEB7H3 as controls. The results showed that TrMEB7H3 alone induced moderate activation of BMDMs, whereas coculture with GL261 cells in the presence of TrMEB7H3 elicited a complete M1-polarized transcriptional profile, indicating that robust macrophage activation by TrME required the presence of targeted tumor cells (Extended Data Fig. 4b). Moreover, TrMEB7H3 treatment upregulated MHC II (H2-Ab1) and costimulatory molecules (4-1BBL and OX40L)45, downregulated coinhibitory molecules (PDL2 and CLEC1)46,47 and activated signaling pathways contributing to antitumor efficacy (Extended Data Fig. 4c–h). In sum, these findings demonstrated that TrMEB7H3 effectively reprogrammed macrophages toward an antitumor phenotype.
Single-dose TrMEB7H3 mRNA–LNP shows improved intracranial pharmacokinetics and prolongs survival in GBM-bearing mice
To evaluate the in vivo efficacy of TrME, our initial strategy involved direct intratumoral delivery of TrME protein into the brains of GBM-bearing mice, thereby bypassing the blood–brain barrier. A single dose of TrMEB7H3 (2.5 mg kg−1) was administered through intratumoral injection 7 days after tumor implantation (Extended Data Fig. 5a). Luciferase-based imaging revealed that, compared to the PBS-treated group, a single administration of TrMEB7H3-protein markedly suppressed tumor growth throughout the observation period (Extended Data Fig. 5b,c). Survival analysis revealed that a single dose of TrMEB7H3 significantly prolonged survival compared to the control group, indicating effective control of tumor progression (Extended Data Fig. 5d). The LNP–mRNA system has been shown to enable rapid and efficient production of therapeutic proteins48,49, with the advantage of providing prolonged in vivo exposure, thereby enhancing overall therapeutic efficacy50. To investigate the efficacy of TrMEB7H3 produced in vivo, we used LNPs formulated with commercially available Dlin-MC3-DMA to encapsulate mRNA encoding TrMEB7H3, generating the TrMEB7H3-MC3 formulation. Additionally, a highly efficient signal peptide was engineered at the N terminus of TrMEB7H3 to promote the effective secretion of the protein from the cytoplasm to the extracellular space (Fig. 3a).
Fig. 3: BCT-A5 LNPs demonstrate superior in vivo delivery of TrME mRNA compared to commercial LNPs.
a, Schematic design of the mRNA encoding secreted TrMEB7H3 and the process of TrMEB7H3 generation and secretion from cells. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). b, Schematic diagram of the LNP structure, which is composed of an ionizable lipid, cholesterol, a PEG lipid and a helper lipid. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). c, Illustration showing the ring skeletons of cis-ionizable lipids and tail chains of different lengths. d, Compositions of LNPs based on cis-ionizable lipids. e, GL261 cells were treated with mLuc-loaded LNPs based on cis-ionizable lipids. The luminescence intensity of luciferase after 24 h of incubation with mLuc-LNPs is shown in the heat map. Compositions A, E, F and D are detailed in Fig. 3d. Each set of 16 numbers corresponds to the lipid molar ratios listed in Supplementary Fig. 9c. f, The best-performing BCT-A5 LNP contained BC-12, cholesterol, C14PEG2K and DOPE at a ratio of 15:30:0.5:40. g, Confocal images of GL261 cells transfected with GFP mRNA by BCT-A5 (n = 5 cell samples per group, biological replicates). Green, GFP; blue (DAPI), nuclei. h, Confocal images of GL261 cells transfected with aVHH mRNA by BCT-A5 (n = 5 cell samples per group, biological replicates). Magenta, aVHH; blue (DAPI), nuclei. i, Confocal images of GL261 cells cocultured with BMDMs transfected with CRT–anti-B7H3 mRNA using BCT-A5 (n = 5 cell samples per group, biological replicates). Magenta, CRT–anti-B7H3; green, BMDMs; blue (DAPI), nuclei. j, Immunofluorescence of TrMEB7H3 in the brains of GBM-bearing mice 12 h after the intratumoral injection of TrMEB7H3-BCT-A5 (n = 3 tumor samples per group, biological replicates). Green, TrMEB7H3; blue (DAPI), nuclei. k, TrMEB7H3 concentrations in the serum of mice after the repeated administration of TrMEB7H3-BCT-A5 and TrMEB7H3-MC3 were determined by ELISA at 12 h and 108 h after each dose. The vertical dotted lines indicate injections (n = 3 serum samples per time point, biological replicates).
Source data
To assess single-dose pharmacokinetics, TrMEB7H3-MC3 (approximately 1.5 mg kg−1) or TrMEB7H3-protein (approximately 2.5 mg kg−1) was administered through intratumoral injection into GBM-bearing mice (Supplementary Fig. 8a). Compared to TrMEB7H3-protein, TrMEB7H3-MC3 prolonged the retention of TrMEB7H3 in brain homogenates (TrMEB7H3-MC3: Cmax = 53.78 ± 2.09 μg ml−1, terminal t1/2 = 34.04 h, ~0.005 μg ml−1 after 6 days; TrMEB7H3-protein: Cmax = 47.13 ± 4.04 μg ml−1, terminal t1/2 = 0.52 h, ~0.005 μg ml−1 after 3 days) (Supplementary Fig. 8b). Moreover, TrMEB7H3-MC3 treatment significantly improved tumor control in GBM-bearing mice, increasing the median survival from 38.5 to 46.5 days. These findings indicated that the LNP–mRNA system enhanced the efficacy of TrME by maintaining elevated levels of TrME over an extended duration (Extended Data Fig. 5e–h).
Rational design of ionizable lipids and LNP compositions enhances in vivo TrME mRNA delivery
The synthesis of numerous classes of ionizable lipids has expanded the chemical space for mRNA delivery51,52,53,54,55,56. However, the impact of structural configurations, particularly cis–trans stereochemistry, on the delivery capabilities of ionizable lipids remains underexplored57,58. To enhance the in vivo delivery of TrME mRNA, we synthesized cis-B-9 (BC-9) and trans-B-9 (BT-9) to investigate the influence of cis–trans stereochemistry on mRNA delivery efficiency (Supplementary Fig. 9a). LNPs are typically composed of four components: (1) an ionizable lipid; (2) a PEG lipid; (3) cholesterol; and (4) a helper lipid48 (Fig. 3b). To ensure that our findings were not specific to a single LNP formulation, we used four distinct compositions (A, B, C, and D) (Supplementary Fig. 9b) and 16 molar ratios (Supplementary Fig. 9c) to create 128 distinct LNPs, each loaded with firefly luciferase mRNA (mLuc). Transfection efficiency was assessed in GL261 cells using the Luciferase-Glo assay (Extended Data Fig. 6a). Our results demonstrated that BC-9 LNPs exhibited significantly higher transfection efficiency compared to BT-9 LNPs when using compositions A or B, although neither showed substantial efficiency with other compositions (Extended Data Fig. 6b). Furthermore, LNPs incorporating DOPE as the helper lipid consistently outperformed those with DPPC (Extended Data Fig. 6c). While both PEG lipids performed similarly, the five most efficient LNPs used C14PEG2K rather than DMG-PEG2k (Extended Data Fig. 6d). These findings suggest that cis-ionizable lipids might offer superior delivery efficiencies compared to their trans counterparts and highlight the critical role of LNP composition in determining delivery performance.
Encouraged by the promising results of the BC-9, we synthesized three additional cis-ionizable lipids (BC-12, BC-14 and BC-16) by varying the length of lipid tails (Fig. 3c). To identify the optimal formulation, we selected C18PEG2K and DOTAP as alternative PEG and helper lipids, respectively, and prepared three additional compositions (Fig. 3d). A total of 256 LNPs with 16 different molar ratios were synthesized and evaluated for transfection efficiency. Among these, the BC-12 LNP (BCT-A5) showed the highest luminescence intensity (Fig. 3e). Notably, LNPs with composition A consistently outperformed those with other compositions, regardless of the specific cis-ionizable lipid used, further underscoring the importance of LNP composition (Extended Data Fig. 6e). Additional analyses revealed that PEG and helper lipids might interact synergistically to enhance LNP performance. Specifically, when DOPE was used as the helper lipid, C14PEG2k LNPs exhibited higher luminescence than C18PEG2k LNPs. Conversely, when C18PEG2k was the PEG lipid, DOTAP LNPs outperformed DOPE LNPs in transfection efficiency (Extended Data Fig. 6f,g). These results suggest that the lipid components within LNPs influenced delivery efficiency both independently and through complex interactions. To exclude the influence of mRNA length, NanoLuc mRNA (655 nt) was used as a base construct and two additional variants were generated by inserting noncoding sequences after the stop codon, yielding total lengths of 1,150 nt and 2,955 nt, with the latter being comparable to that of TrME mRNA (2,941 nt). Using BC-9, BC-12, BC-14 and BC-16 formulations from composition A, we evaluated the delivery efficiency of all three mRNAs (Supplementary Fig. 10). The results demonstrated that shorter mRNAs exhibited higher delivery efficiency, yet the top three LNP formulations consistently performed best across all mRNAs. These findings indicate that, while mRNA length influenced encapsulation efficiency, the BCT-A5 formulation had a high total loading capacity, making it well suited for mRNAs of varying lengths and supporting its broad adaptability for mRNA therapeutics.
On the basis of these findings, we selected BCT-A5 as the delivery system for TrME mRNA, formulated with BC-12, cholesterol, C14PEG2K and DOPE at a molar ratio of 15:30:0.5:40 (Fig. 3f). To validate the BCT-A5 system, we synthesized the trans-ionizable lipid BT-12 and formed BTT-A5 LNPs with the same composition. As anticipated, BCT-A5 demonstrated superior delivery efficiency compared to BTT-A5, with luciferase reporter assays revealing that the luminescence intensity of BCT-A5 LNPs was approximately fivefold higher than that of MC3 LNPs (Supplementary Fig. 11). To test the universal applicability of BCT-A5, we delivered mRNAs encoding various functional proteins, including a cytoplasmic protein (GFP), a membrane protein (glycosylphosphatidylinositol-anchored camelid VHH antibody, aVHH)59 and a secreted protein (CRT–anti-B7H3). Confocal imaging confirmed the successful intracellular delivery and translation of these proteins (Fig. 3g–i and Supplementary Fig. 12). Furthermore, we performed flow cytometry analysis using BCT-A5 LNPs encapsulating GFP mRNA to evaluate the cell transfection efficiency in vitro. The results showed that BCT-A5 LNPs efficiently transfected GL261 tumor cells and B cells but exhibited low efficiency in T cells and macrophages and no transfection in endothelial cells (Supplementary Fig. 13).
Next, we evaluated the ability of BCT-A5 to mediate the in vivo production of TrME. TrMEB7H3-BCT-A5, formulated by encapsulating TrMEB7H3 mRNA within BCT-A5, effectively expressed TrMEB7H3 following intratumoral injection into GBM-bearing mice (Fig. 3j). In addition, in vivo delivery of mLuc using BCT-A5 maintained detectable signals for at least 5 days (Supplementary Fig. 14). To assess the pharmacokinetic superiority of BCT-A5, mice were intravenously administered 0.5 mg kg−1 TrMEB7H3-BCT-A5 or TrMEB7H3-MC3 every 5 days for a total of three doses. Both formulations maintained high serum concentrations of TrMEB7H3 but TrMEB7H3-BCT-A5 achieved a peak concentration approximately 1.69-fold higher than that of TrMEB7H3-MC3 (Fig. 3k). Safety evaluations showed that TrMEB7H3-BCT-A5 did not induce significant erythrocyte lysis at mRNA concentrations up to 20 μg ml−1 (Supplementary Fig. 15). No clinical signs or organ damage were observed in mice administered 0.5 mg kg−1 TrMEB7H3-BCT-A5 intravenously (Supplementary Fig. 16). Serum biochemical markers remained unchanged, further supporting the safety profile of TrMEB7H3-BCT-A5 (Supplementary Fig. 17). We further analyzed serum cytokine levels (IL-6, tumor necrosis factor, IFNγ and IL-1α) after 0.5 mg kg−1 BCT-A5 LNP administration to evaluate systemic immune activation. The results revealed no significant cytokine induction, except for a mild and transient increase in serum IL-6, which returned to baseline within 72 h (Supplementary Fig. 18). These data collectively demonstrate that TrMEB7H3-BCT-A5 conferred pharmacokinetic advantages over the MC3-based formulation, while eliciting no detectable systemic cytokine elevations or histopathological abnormalities.
To further evaluate any TrMEB7H3-associated toxicity, we conducted dose-escalation studies for both the TrMEB7H3-protein and the TrMEB7H3-BCT-A5 formulation, using body weight and organ coefficients (liver and spleen) as primary indicators of systemic toxicity. The results showed that the administration of up to 50 mg kg−1, TrMEB7H3-protein did not induce significant changes in body weight or hepatosplenomegaly. Similarly, TrMEB7H3-BCT-A5, administered up to 5 mg kg−1, caused no apparent weight loss or liver enlargement, although mild splenomegaly was observed at this dose—likely reflecting transient immune activation because of high-dose mRNA (Supplementary Fig. 19).
Moreover, we performed comprehensive complete blood count (CBC) analyses following treatment with TrMEB7H3-protein and TrMEB7H3-BCT-A5, using anti-SIRPα–CRT and anti-CD47 antibodies as controls. Neither the TrMEB7H3 formulation nor anti-SIRPα–CRT induced anemia, thrombocytopenia or leukocyte depletion and all major immune cell populations (neutrophils, lymphocytes and monocytes) remained stable (Extended Data Fig. 7a–d). We also conducted a hematological toxicity assessment following repeated doses of the TrMEB7H3-protein. CBC analyses were performed following both single and repeated intravenous administrations of TrMEB7H3 in mice, with anti-CD47 antibody treatment serving as a control. Repeated administrations of TrMEB7H3 did not cause anemia, thrombocytopenia or T cell depletion, as reflected by stable red blood cell, hemoglobin, platelet and white blood cell subset counts after both dosing regimens. In contrast, anti-CD47 antibody treatment led to mild reductions in red blood cell, hemoglobin and platelet counts (Extended Data Fig. 7e–h). These findings confirm that the hematological safety profile of TrMEB7H3 was favorable, which is consistent with its tumor-selective design.
TrMEB7H3-BCT-A5 drives TAMs to initiate M1 polarization and phagocytosis-dependent tumor eradication
Given the ability of BCT-A5 to prolong in vivo exposure to TrMEB7H3, we next evaluated the antitumor efficacy of TrMEB7H3-BCT-A5 in an orthotopic GBM mouse model using B7H3high GL261-Fluc cells. Starting on day 7 after tumor implantation, mice received intratumoral injections of saline, free mRNA, anti-SIRPα–anti-B7H3, CRT–anti-B7H3, anti-SIRPα–CRT, anti-SIRPα/B7H3 (with an active Fc domain), TrMEB7H3-protein or TrMEB7H3-BCT-A5 every 5 days for a total of three doses. Tumor progression was monitored using luciferase-based imaging (Fig. 4a). Compared to monotherapy with CRT–anti-B7H3 or anti-SIRPα/B7H3, the TrMEB7H3-based regimen significantly suppressed tumor growth during the observation period, with both TrMEB7H3-protein and TrMEB7H3-BCT-A5 demonstrating efficacy (Fig. 4b,c). On day 22 after inoculation, magnetic resonance imaging (MRI) and histological analysis of tumor tissue using hematoxylin and eosin (H&E) staining revealed that the tumor area was smallest in the TrMEB7H3-BCT-A5 treatment group (Extended Data Fig. 8a,b). Notably, TrMEB7H3-BCT-A5 provided significantly longer-term protection compared to TrMEB7H3-protein, with ~90% of mice in the TrMEB7H3-BCT-A5 group surviving at the 60-day mark, whereas all TrMEB7H3-protein-treated mice succumbed within 56 days (Fig. 4d). To accurately evaluate therapeutic durability, we performed continued dosing studies with both TrMEB7H3-protein and TrMEB7H3-BCT-A5, which were administered every 5 days over a 60-day observation period. There were no deaths in either treatment group during the observation period (Extended Data Fig. 8c). Moreover, we examined the levels of TrMEB7H3 in the brains of mice that no longer exhibited tumor control. Tumor tissues were collected at various time points following a single administration of TrMEB7H3-BCT-A5. TrMEB7H3 was nearly undetectable by day 10 after injection, indicating that the loss of efficacy was attributable to drug depletion rather than functional exhaustion (Extended Data Fig. 8d).
Fig. 4: TrMEB7H3-BCT-A5 reshapes the phenotype of tumor-associated macrophages and enhances the adaptive T cell response in GBMs.
a, Experimental schedule for in vivo studies with repeated doses of TrMEB7H3-BCT-A5. Created in BioRender. A, H. (2026) https://BioRender.com/k650onf. b,c, IVIS imaging (b) and quantification of the bioluminescence signal intensity (c) of GBM-bearing mice following the designated treatments on days 7, 12, 17 and 22. P values were determined using a two-way ANOVA. Error bars denote the mean ± s.d. (n = 5 mice per group, biological replicates). d, Survival of GBM-bearing mice that received the indicated treatments. Survival was estimated using the Kaplan‒Meier method and compared using the log-rank test (n = 10 mice per group, biological replicates). e, Quantification of tumor-infiltrating immune cell subsets. P values were determined using a two-way ANOVA. Error bars denote the mean ± s.d. (n = 6 mice per group, biological replicates). f, Quantification of CD86+ and CD206+ cells among tumor-infiltrating CD11b+ F4/80+ cells. P values were determined using a two-way ANOVA. Error bars denote the mean ± s.d. (n = 6 mice per group, biological replicates). g–j, Expression of CD44 (g), CD69 (h), granzyme B (i) and IFNγ (j) in tumor-infiltrating CD8+ T cells (n = 6 mice per group, biological replicates in g). P values were determined using a one-way ANOVA. Error bars denote the mean ± s.d. k, ELISpot (left) and quantification (right) of IFNγ+ cell in tumor-infiltrating CD8+ T cells. P values were determined using a one-way ANOVA. Error bars denote the mean ± s.d. (n = 3 mice per group, biological replicates). The number of IFNγ+ spots is shown in the bottom left corners of the ELISpot images. l, Quantification of monocyte MDSCs (CD11b+Ly6G−Ly6Chi) and polymorphonuclear MDSCs (CD11b+Ly6G+Ly6Clo) among tumor-infiltrating CD11b+ cells. P values were determined using a two-way ANOVA. Error bars denote the mean ± s.d. (n = 6 mice per group, biological replicates). The detailed flow cytometry gating strategy was provided in Supplementary Fig. 26. *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Exact P values are indicated in the graph.
Source data
Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) staining of tumor sections indicated elevated levels of tumor cell apoptosis in response to TrMEB7H3, suggesting that, in addition to enhancing macrophage-mediated phagocytosis, TrMEB7H3 might induce alternative tumor cell-killing mechanisms (Extended Data Fig. 8e). To elucidate the antitumor mechanism of TrMEB7H3, we analyzed the immune infiltration profiles of orthotopic GBM tumors following TrMEB7H3 treatment, including both the proportions and absolute numbers of various immune cell populations. Treatment with TrMEB7H3-BCT-A5 resulted in increased macrophage infiltration and promoted polarization toward the antitumor M1 phenotype (Fig. 4e,f and Extended Data Fig. 8g,h). In addition, compared to saline control, TrMEB7H3-BCT-A5 significantly increased CD8+ T cell infiltration and upregulated the expression of T cell activation markers CD69 (ref. 60) and CD44 (ref. 61) (Fig. 4e,g,h and Extended Data Fig. 8f,g,i). Consistent with these findings, TrMEB7H3-BCT-A5 enhanced effector CD8+ T cell activity, as evidenced by the increased production of granzyme B and IFNγ (Fig. 4i–k and Extended Data Fig. 8f,i), which are critical mediators of tumor cell apoptosis. Moreover, TrMEB7H3 therapy reduced the presence of immunosuppressive cells, including CD4+ regulatory T (Treg) cells and myeloid-derived suppressor cells (MDSCs), within the TME (Fig. 4e,l and Extended Data Fig. 8j). These results indicate that TrMEB7H3 initiated innate and adaptive immune responses through macrophage–T cell interactions43, promoting myeloid cell migration, T cell stimulation and IFNγ production.
Modular TrME induces TAA-specific phagocytosis for broad-spectrum solid tumors
To evaluate the broad-spectrum antitumor activity of TrMEB7H3, we used the B7H3high MB49 bladder cancer model, which exhibits distinct B7H3 and CD47 expression profiles compared to GL261 GBM model (Supplementary Figs. 2 and 20). We analyzed the intratumoral concentration–time profile of TrMEB7H3 following intravenous administration of TrMEB7H3-BCT-A5 in bladder-cancer-bearing mice. A single injection maintained high intratumoral exposure to TrMEB7H3 for up to 5 days (Supplementary Fig. 21a). Consistent with the observations in the GBM mouse model, the administration of TrMEB7H3-BCT-A5 markedly increased macrophage infiltration into bladder tumors in mice. Notably, these macrophages penetrated deep into the tumor core rather than merely accumulating in perivascular regions (Supplementary Fig. 22). Furthermore, we evaluated antidrug antibody (ADA) formation in mouse serum during a dosing regimen of once every 5 days for four total administrations and detected no measurable ADA response. These data confirmed that TrMEB7H3-BCT-A5 enabled prolonged intratumoral exposure without an ADA response (Supplementary Fig. 21b). To further investigate the in vivo generation of TrMEB7H3, we isolated endothelial cells, macrophages, T cells, B cells, neutrophils and tumor cells from bladder cancer-bearing mice treated with TrMEB7H3-BCT-A5. Western blot analysis of cell culture supernatants revealed that tumor cells and B cells were the primary producers of TrMEB7H3, whereas endothelial cells, macrophages, T cells and neutrophils produced negligible amounts (Supplementary Fig. 23). Importantly, functional assays demonstrated that TrMEB7H3 derived from different cell types exhibited comparable macrophage-mediated tumor cell phagocytosis, indicating that the cellular origin of TrMEB7H3 did not affect its potency in vivo (Supplementary Fig. 24).
Evaluation of TrMEB7H3 efficacy began on day 5 after tumor implantation, with intravenous administration every 5 days for a total of three doses (Fig. 5a). Tumor progression was monitored using ultrasound imaging on days 5, 10, 15 and 20. TrMEB7H3-based treatment significantly enhanced tumor inhibition. Compared to the saline control, TrMEB7H3-protein slowed tumor progression, while TrMEB7H3-BCT-A5 demonstrated superior tumor growth suppression relative to both free BCT-A5 mRNA and TrMEB7H3-protein (Fig. 5b). Then, 5 days after the final dose, tumors were isolated for further analyses. TrMEB7H3-BCT-A5 exhibited the most pronounced efficacy, as evidenced by reductions in tumor size, weight and cross-sectional area (Fig. 5c–e). In addition, survival outcomes correlated with tumor suppression, with TrMEB7H3-BCT-A5 significantly prolonging survival; only two mice succumbed during the 60-day observation period (Fig. 5f). These data highlight the broad antitumor potential of TrME in tumors expressing high levels of the relevant TAA. Considering the enhanced in vivo exposure efficiency conferred by the mRNA–LNP delivery system, we further compared the therapeutic efficacy of TrMEB7H3-BCT-A5 with that of control proteins delivered in the form of mRNA-BCT-A5 LNPs. TrMEB7H3-BCT-A5 maintained the most significant tumor control, confirming the superior potency of TrMEB7H3 (Extended Data Fig. 9a,b). We also compared the efficacy of protein-based anti-SIRPα/B7H3 and CRT–anti-B7H3 with that of their mRNA–LNP counterparts. Compared to protein administration, the mRNA-BCT-A5 LNP delivery system enhanced tumor suppression and prolonged survival, supporting the broad potential of BCT-A5 LNPs for improving protein drug efficacy (Extended Data Fig. 9c,d). Furthermore, we evaluated the therapeutic efficacy of the TrMEB7H3-protein, CRT–anti-B7H3 combined with anti-CD47 and CRT–anti-B7H3 combined with anti-SIRPα. The TrMEB7H3-protein demonstrated significantly greater tumor inhibition and survival benefits, confirming that the integration of all three functional domains within a single construct resulted in stronger phagocytic activation compared to separate modulation of individual signaling axes (Extended Data Fig. 9e,f and Supplementary Fig. 25).
Fig. 5: TrME induces tumor regression in a TAA-dependent manner.
a, Schematic diagram of the treatment regimen of TrMEB7H3 in mice bearing MB49 bladder cancer. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). b, Tumor images obtained by ultrasonography. Representative pictures are from n = 6 mice per group, biological replicates. c, Photographic images of tumors obtained from mice treated with different therapies. The absence of tumors in certain treatment groups was because of the failure of tumor-bearing mice to survive until the sampling time point (n = 6 mice per group, biological replicates). The scale bar represents 1 cm. d, Statistical graph of the tumor weights. In the saline and free mRNA groups, n = 5 mice per group, biological replicates; in the other treatment groups, n = 6 mice per group, biological replicates. P values were determined using a one-way ANOVA. Error bars denote the mean ± s.d. e, H&E staining of mouse tumor sections after different treatments. Representative pictures are from n = 6 mice per group, biological replicates. f, Survival of bladder cancer model mice after different treatments. Survival was estimated using the Kaplan‒Meier method and compared using the log-rank test (n = 6 mice per group, biological replicates). g, Schematic diagram of the treatment regimen of TrMEHER2 in 4T1 and 4T1-HER2 breast cancer mice. Created in BioRender. A, H. https://BioRender.com/k650onf (2026). h, Photographic images of breast tumors obtained from mice after different treatments (n = 3 mice per group, biological replicates. i, Individual 4T1 or 4T1-HER2 tumor growth curves for mice after different treatments (n = 3 mice per group, biological replicates). j, Survival of 4T1 and 4T1-HER2 breast cancer model mice after different treatments. Survival was estimated using the Kaplan‒Meier method and compared using the log-rank test (n = 6 mice per group, biological replicates). *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001. Exact P values are indicated in the graph.
Source data
Given the modular design of the anti-TAA scFv structure in TrME, we hypothesized that TrME could be adapted to target a wider range of solid tumors by modifying its TAA specificity. To test this, we replaced the anti-B7H3 scFv in TrMEB7H3 with an anti-HER2 scFv, generating TrMEHER2. Using the murine triple-negative breast cancer cell line 4T1, we engineered 4T1 cells expressing HER2 (4T1-HER2), enabling TrMEHER2 to specifically bind these tumor cells. Orthotopic breast cancer models were established using both 4T1 and 4T1-HER2 cells to determine whether TrMEHER2 functions in a TAA-dependent manner, rather than exerting nonspecific activity. Following the same three-dose treatment regimen initiated on day 7 after tumor induction (Fig. 5g), TrMEHER2-BCT-A5 significantly inhibited tumor growth, reduced the tumor burden and prolonged survival in mice bearing HER2high 4T1-HER2 tumors (Fig. 5h–j). Conversely, no significant antitumor effect was observed in mice injected with HER2-negative 4T1 cells, with all mice reaching tumor burden endpoints within 50 days (Fig. 5j). The limited response in HER2-negative tumors may be attributed to alterations in macrophage phenotype or blockade of the ‘don’t eat me’ signal, as observed in prior in vitro experiments. These results demonstrated that TrME initiated antitumor immune responses in a TAA-dependent manner and could be adapted to target a broader spectrum of solid tumors by modifying the anti-TAA scFv.
To evaluate the potential abscopal effects of TrME therapy, we performed dual mammary fat pad implantation experiments. Mice were established with (1) bilateral HER2+ tumors; (2) bilateral HER2− tumors; or (3) HER2+ (right) and HER2− (left) tumors. Following TrMEB7H3-BCT-A5 treatment, we observed that TrMEB7H3 treatment significantly reduced tumor size in both flanks of mice bearing bilateral HER2+ tumors but was not effective in mice bearing bilateral HER2− tumors. Importantly, in mice harboring HER2+ (right) and HER2− (left) tumors, TrMEB7H3-BCT-A5 treatment not only suppressed HER2+ tumors but also markedly reduced the size of contralateral HER2− tumors, indicating effective epitope spreading and immune-mediated clearance of antigen-negative tumors (Extended Data Fig. 10a,b). These findings substantiated the ability of TrME to induce systemic antitumor immunity beyond direct antigen recognition.
