LIFUS-driven engineered bacteria reprogram immunosuppressive niches via mechano-NOTCH signaling

TL;DR

Scientists engineered bacteria to use ultrasound to create mechanical forces in tumors, which helps T cells fight cancer more effectively.

Contribution

A novel mechano-immunotherapy using engineered bacteria and ultrasound to disrupt immunosuppressive tumor niches via NOTCH signaling.

Findings

LIFUS-activated gas vesicles reduce cancer-associated fibroblasts and disrupt CAF-CD8+ T cell communication.

Mechanotransduction via NOTCH signaling enhances CD8+ T cell infiltration and cytotoxicity in tumors.

LIFUS-GV mechano-priming improves adoptive T cell therapy efficacy in solid tumors.

Abstract

Solid tumors impose coupled stromal and immunologic barriers that limit T cell infiltration and function. Here, we engineer Salmonella VNP20009 to express gas vesicles (GVs), creating an intratumoral cavitation source that converts low-intensity focused ultrasound (LIFUS) into localized mechanical forces. LIFUS-activated GVs remodel the tumor microenvironment by reducing cancer-associated fibroblast (CAF) abundance, decompressing the matrix, and selectively disrupting CAF-CD8+ T cell communication via a mechanosensitive Notch1-Jagged1 axis. Single-cell RNA sequencing reveals a redistribution of CD8+ T cell states, characterized by enrichment of cytotoxic effector populations and attenuation of NOTCH signaling in memory-associated cells. These biomechanical changes enhance intratumoral CD8+ T cell infiltration and restore effector cytokine production. Leveraging this mechanism, we develop a mechano-priming approach for adoptive T cell therapies. Pre-conditioning with LIFUS-driven GVs improves CD8+ T cell cytotoxicity, strengthens tumor cell adhesion, reduces exhaustion signatures, and achieves durable tumor control and extended survival in orthotopic and metastatic models.

•Engineered VNP/ARG-GVs enable ultrasound imaging and tumor mechanotransduction•LIFUS-activated GVs generate forces that disrupt CAF-CD8+ T cell NOTCH signaling•GV-mediated mechanotransduction reverses immunosuppression and boosts cytotoxicity•LIFUS-GV mechano-priming improves OT-1 and CAR-T efficacy in solid tumors

Engineered VNP/ARG-GVs enable ultrasound imaging and tumor mechanotransduction

LIFUS-activated GVs generate forces that disrupt CAF-CD8+ T cell NOTCH signaling

GV-mediated mechanotransduction reverses immunosuppression and boosts cytotoxicity

LIFUS-GV mechano-priming improves OT-1 and CAR-T efficacy in solid tumors

Lin et al. engineer bacteria expressing gas vesicles to convert low-intensity ultrasound into localized mechanical forces within tumors. This mechano-immunotherapy remodels the stromal-immune microenvironment, disrupts CAF-CD8+ T cell NOTCH signaling, and enhances T cell-based cancer immunotherapy efficacy.

Figures

Establishment and characteristics of engineered bacteria (*VNP*/ARG)(A) Schematic diagram of the transfer of ARG-encoding plasmid pBAD-bARGSer-axe-txe into *VNP*20009 to establish engineered bacteria *VNP*/ARG.(B and C) Electroporation (B) and arabinose-induced ARG expression. Robust *VNP*/ARG-GV expression was observed only after 0.2% arabinose induction, whereas *VNP*/ARG showed minimal signal (*n* = 3) (C).(D) Schematic diagram depicting LIFUS-triggered cavitation of intracellular GVs.(E) TEM visualization showing abundant spindle-shaped GVs occupying ∼50% of the cytoplasmic space in induce

Mechanical immunotherapy effect of LIFUS combined with *VNP*/ARG-GV for tumors(A) Schematic illustrating the five-cycle LIFUS + *VNP*/ARG-GVs treatment regimen over 21 days.(B and C) Repeated LIFUS activation of intratumoral *VNP*/ARG-GVs resulted in marked suppression of tumor growth compared with all control and monotherapy groups (∗∗∗∗*p* < 0.0001; *n* = 6) (B).(D) Kaplan-Meier survival analysis demonstrating significantly prolonged survival in the combination group (median 47.5 days), with several mice surviving beyond 60 days. All control and monotherapy cohorts succumbed by days 30–40 (c

Single-cell sequencing analysis identifies a treatment-enriched CD8^+^ effector subset and reveals mechanosensitive features of T cell activation(A) UMAP showing eight major TME cell populations across all groups: B cells, endothelial cells, epithelial cells, fibroblasts, myeloid cells, neutrophils, smooth muscle cells, and T/NK lymphocytes.(B) Quantification of T cell abundance, showing marked expansion of T lymphocyte populations in the combination group.(C and D) Subclassification identified CD8^+^ T cells as the predominant cytotoxic lymphocyte subset.(E) Unsupervised clustering identified

Mechano-regulation of CAF-CD8^+^ T cell NOTCH signaling interactions(A and B) Ultrasound elastography (A) shows a significant reduction in tumor stiffness with the combination group (12.95 ± 2.18 kPa) compared with controls (57.19 ± 15.02 kPa, ∗∗∗*p* < 0.005) and *VNP*/ARG-GV monotherapy (46.82 ± 8.13 kPa, ∗∗*p* < 0.01), LIFUS monotherapy group (36.93 ± 6.812 kPa), and the control group (57.19 ± 15.02 kPa, ∗*p* < 0.05) (*n* = 4) (B).(C) AFM further confirms reduced matrix elastic modulus (*n* = 12).(D) Collagen I distribution secreted by fibroblasts in Figure S4B supported the hypothesized cha

Dual mechanotransductive regulation of CAFs-CD8^+^ T cell crosstalk via NOTCH signaling(A) The schematic diagram of co-culture systems of CAFs and CD8^+^ T cells with GVs responsive to LIFUS constructs *in vitro*. Before 60 s of LIFUS exposure, cell co-incubation with purified GVs allowed sufficient GV-cell contact.(B and C) Confocal imaging revealed a significant reduction in Notch1 receptor and Jagged1 ligand expression in CAFs exposed to LIFUS-driven GVs (GVs + LIFUS group) compared to unstimulated controls (GVs − LIFUS group). Scale bars, 100 μm.(D and E) In CD8^+^ T cells, Jagged1 express

Mechanobiology-enhanced adoptive T cell therapy improves control of primary and metastatic tumors(A) Schematic representation of the B16-OVA orthotopic melanoma model and the OT-1-based treatment strategy. The GVs (1 × 10^6^ particles/mL) were injected intratumorally every cycle with LIFUS or without LIFUS before OT-1 cells (1 × 10^6^ cells) were adoptively transferred later.(B) Tumor growth curves over 15 days. OT-1^+LIFUS^ treatment resulted in the most pronounced tumor inhibition, exhibiting significantly smaller tumor volumes than both OT-1^−LIFUS^ and control groups (266.7 ± 127.2 mm^3^ v

Full text

Introduction

The tumor microenvironment (TME) imposes tightly interlinked physical and immunological barriers that fundamentally restrict therapeutic responses in solid tumors.1 Activated cancer-associated fibroblasts (CAFs) generate a dense, stiff extracellular matrix that increases interstitial pressure, impedes T cell infiltration, and limits drug penetration.2 Concurrently, hypoxia, acidosis, and nutrient deprivation drive CD8^+^ T cell dysfunction and exhaustion, enhancing immune evasion through stromal-derived inhibitory ligands such as Jagged2, which activate NOTCH signaling.3^,^4 As a result, even transformative immunotherapies, including immune checkpoint blockade, often fail in desmoplastic tumors with intense stromal remodeling.5^,^6

The distinct metabolic and vascular properties of tumors have renewed interest in bacteria as programmable therapeutic agents.7^,^8 The attenuated Salmonella strain VNP20009 preferentially accumulates in hypoxic tumor cores and activates both innate and adaptive immunity,9^,^10^,^11^,^12 but phase 1 trials revealed limited monotherapy efficacy despite acceptable safety.13^,^14^,^15 Consequently, strategies to enhance the immunomodulatory potential of engineered microbes are urgently needed. Meanwhile, advances in synthetic biology and ultrasound (US) have created opportunities for noninvasive, remote, and spatially precise control of microbial activity.16^,^17^,^18^,^19

US is fundamentally a mechanical wave capable of exerting non-thermal biophysical forces on tissues. Low-intensity focused US (LIFUS) generates shear stress, microstreaming, and cavitation while avoiding heat-induced toxicity.20^,^21^,^22^,^23 Mechanical cues are increasingly recognized as powerful regulators of immune function, modulating mechanosensitive receptors such as Piezo1, integrins, T cell receptors, and NOTCH.24 Yet, existing US-bacteria platforms primarily rely on thermal gene activation or microbubble-assisted delivery, highlighting that US-generated mechanical forces, rather than heat or payload release, can be directly harnessed to reprogram tumor immunobiology.

Gas vesicles (GVs) provide an enabling platform opportunity to address this gap. These genetically encoded, air-filled protein structures act as nanoscale cavitation nuclei that collapse under US and can be produced in situ within hypoxic tumors by engineered bacteria.25 Although GVs have been used as acoustic reporter genes (ARGs) for imaging,26^,^27^,^28 their potential as intratumoral mechanotransducers capable of modulating immune-stromal interactions remains unexplored.

Here, we engineer VNP20009 to express ARG-encoded GVs (VNP/ARG-GVs) and develop a LIFUS-responsive bacterial platform that integrates real-time acoustic imaging with precise intratumoral delivery of mechanical forces. Through single-cell RNA sequencing, US elastography, and functional validation, we identify a previously unrecognized “mechano-NOTCH axis” whereby LIFUS-induced GV cavitation disrupts CAF-CD8^+^ T cell communication by suppressing Notch1-Jagged1 signaling. This mechanical intervention softens the extracellular matrix, enhances CD8^+^ T cell infiltration, relieves exhaustion, and amplifies cytotoxic function. These insights further motivated a translational extension: mechano-priming of adoptive T cell therapies. Ex vivo LIFUS-GV stimulation enhances OT-1 and chimeric antigen receptor T (CAR-T) cell antitumor activity in both primary tumors and metastatic lung disease.

Collectively, our work establishes mechano-immunotherapy as a conceptual and technical framework that leverages physical forces, not biochemical agents, to reprogram the TME and potentiate T cell-based therapies. By demonstrating that genetically encoded cavitation centers can deliver precisely localized mechanical cues, this study addresses an unmet need in the field and provides a programmable strategy for overcoming stromal-immune suppression (see graphical abstract).

Results

Construction of engineered bacteria (VNP/ARG) for acoustic response and optimization of LIFUS parameters

We engineered attenuated Salmonella typhimurium VNP20009 to express an ARG operon, enabling in situ production of GVs that function as intratumoral cavitation nuclei. The plasmid pBAD-bARGSer-axe-txe was successfully introduced into VNP20009 (Figures 1A and 1B). Agarose gel electrophoresis confirmed correct plasmid assembly (Figure 1B). Upon L-arabinose induction, VNP/ARG-GVs exhibited robust ARG expression, whereas VNP/ARG and wild-type VNP20009 showed minimal signal (Figure 1C). To better validate ARG-GVs expression at the protein level, Coomassie blue staining (Figure S1A) was performed to reveal that only VNP/ARG-GVs displayed robust protein-level expression of GV structural components (55–70 kDa). The slight signal observed in VNP20009 and VNP/ARG in Figure 1C is, therefore, attributable to nonspecific background rather than true GV expression. Transmission electron microscopy (TEM) also revealed abundant spindle-shaped vesicles occupying more than half of the cytoplasmic volume (Figure 1E). US-exposed samples showed disrupted vesicle morphology and membrane fragmentation (Figure S1B), consistent with cavitation-induced collapse (Figure 1D). GV diameters ranged from 0.5 to 1.8 μm (average 0.70 ± 0.18 μm) (Figure S1C), notably smaller than clinically used microbubbles, supporting improved penetration into dense tumor stroma.Figure 1. Establishment and characteristics of engineered bacteria (VNP/ARG)(A) Schematic diagram of the transfer of ARG-encoding plasmid pBAD-bARGSer-axe-txe into VNP20009 to establish engineered bacteria VNP/ARG.(B and C) Electroporation (B) and arabinose-induced ARG expression. Robust VNP/ARG-GV expression was observed only after 0.2% arabinose induction, whereas VNP/ARG showed minimal signal (n = 3) (C).(D) Schematic diagram depicting LIFUS-triggered cavitation of intracellular GVs.(E) TEM visualization showing abundant spindle-shaped GVs occupying ∼50% of the cytoplasmic space in induced VNP/ARG-GVs.(F and G) Optimization of in vitro acoustic imaging at 0.6 and 0.8 MPa. White arrows in (F) indicated VNP/ARG-GVs in agarose phantoms responsive to LIFUS. At 5 min, both pressures produced slightly different acoustic intensities (22.67 ± 0.92 dB vs. 24.75 ± 0.92 dB). At 15 min, 0.8 MPa caused significantly greater attenuation (2.00 ± 0.67 dB vs. 14.25 ± 2.92dB; ∗∗∗∗p < 0.0001), consistent with more complete GV cavitation (G) (n = 4).(H and I) In vivo acoustic imaging of VNP/ARG-GV in tumor-bearing mice. LIFUS at 0.8 MPa generated ∼2.5-fold stronger nonlinear signals (34.00 ± 10.00 dB) compared with 0.6 MPa (13.75 ± 9.19 dB; ∗∗∗∗p < 0.0001) (n = 4) (H). Representative images show a progressive signal following GVs’ collapse (I).(J) Colonization assays (Figures S1D and 1J) displayed reduced VNP/ARG-GV colonization in 0.8 MPa-treated samples (0.24 ± 0.09 vs. 0.6 MPa, 1.12 ± 0.11 OD600 units) (n = 4).(K and L) Live/dead staining quantifying ultrasound-induced bacterial lysis. Minimal death was observed at 0.6 MPa (7.15% ± 2.20%), whereas 0.8 MPa induced substantial bacterial death (68.88% ± 4.62%; ∗∗∗∗p < 0.0001) (n = 4) (K). Scale bars, 100 μm.Scale bars are indicated in the images. Data are representative of three independent experiments. Data are mean ± SD (C, G, H, J, and K). ∗p < 0.05*, ∗∗∗∗p < 0.0001.* Statistical analysis was performed using one-way ANOVA (H and K) or two-way ANOVA (G and J). See also Figure S1.

To optimize LIFUS-responsive imaging of GVs and avoid the influence of thermal effects, two acoustic pressure parameters, 0.6 and 0.8 MPa, were selected.26 In vitro, agarose phantoms containing *VNP/*ARG-GVs were exposed to LIFUS (Figure 1F). At 5 min post-sonication, both intensities produced similarly low acoustic signal intensities (22.67 ± 0.92 dB vs. 24.75 ± 0.92 dB, p < 0.05). However, prolonged exposure (15 min) resulted in significantly greater acoustic signal attenuation in the 0.8 MPa group (2.00 ± 0.67 dB vs. 14.25 ± 2.92 dB in 0.6 MPa; p < 0.0001) (Figure 1G), confirming more complete GV cavitation at higher pressure. Similarly, B mode imaging demonstrated robust GV cavitation in tumor-bearing mice. Quantitative analysis revealed that 0.8 MPa-induced signals (34.00 ± 10.00 dB) were ∼2.5-fold higher than 0.6-MPa signals (13.75 ± 9.19 dB; p < 0.005) (Figure 1H). Overall, acoustic imaging confirmed that acoustic cavitation was more complete at 0.8 MPa, resulting in improved contrast (Figure 1I).

Therefore, viability evaluations were conducted to assess the impact of these acoustic parameters on VNP/ARG colonization. Colonization assays demonstrated that LIFUS at either pressure did not affect VNP/ARG colonization in the absence of GVs (Figure S1D). In contrast, VNP/ARG-GVs displayed reduced colonization after LIFUS, particularly at 0.8 MPa (0.24 ± 0.09 vs. 1.12 ± 0.11 OD600 units at 0.6 MPa) (Figure 1J). This result suggests that the collapse of the substantial intracellular GV load within VNP/ARG at 0.8 MPa generates internal mechanical stress that suppresses bacterial colonization, rather than reflecting LIFUS-induced toxicity. Live/dead staining of VNP/ARG-GVs further confirmed minimal death at 0.6 MPa (7.15% ± 2.54%) but marked death at 0.8 MPa (68.88% ± 4.62%, p < 0.0001) (Figures 1K and 1L). This demonstrates that under 0.8-MPa LIFUS-GV conditions, a subset of VNP/ARG bacteria undergo direct lysis, preventing further proliferation in vivo and reducing potential toxicity. Meanwhile, surviving bacteria remained capable of supporting continued tumor colonization and repeated activation cycles, as detailed below. Cytotoxicity assays and histological evaluation (Figures S1E–S1G) further verified that VNP/ARG maintains a favorable biosafety profile.

Collectively, these results identify 0.8 MPa as the optimal LIFUS parameter, balancing robust GV cavitation for imaging and mechanostimulation with sufficient bacterial viability for longitudinal therapy.

Mechanical immunotherapy effect of LIFUS + VNP/ARG-GVs in breast tumors

We next evaluated whether LIFUS-mediated mechanostimulation of VNP/ARG-GVs enhances antitumor immunity using a 4T1 orthotopic breast cancer model (Figure 2A). To clarify the bacterial dosing and induction strategy, mice received a single intravenous (i.v.) injection of approximately 1 × 10^6^ cells/mL VNP/ARG bacteria, and acoustic cavitation imaging comparing i.v. injection vs. intratumoral (IT) injection of the same VNP/ARG dose at 24, 48, and 72 h was performed. The results showed that although only ∼21% of injected bacteria colonize initially, in situ bacterial expansion rapidly amplifies GV density, ultimately achieving cavitation-relevant acoustic intensities by 72 h (for details see Figures S2A and S2B). Spatially, i.v. delivery produced homogeneous whole-tumor distribution, whereas IT injection remained restricted to the needle tract, and the acoustic intensity was significantly lower than i.v. delivery by 72 h (36.34 ± 1.53 dB vs. 27.67 ± 4.04 dB, p < 0.005). These findings support the use of systemic administration for achieving sustained, uniform cavitation nuclei throughout the tumor.Figure 2. Mechanical immunotherapy effect of LIFUS combined with VNP/ARG-GV for tumors(A) Schematic illustrating the five-cycle LIFUS + VNP/ARG-GVs treatment regimen over 21 days.(B and C) Repeated LIFUS activation of intratumoral VNP/ARG-GVs resulted in marked suppression of tumor growth compared with all control and monotherapy groups (∗∗∗∗p < 0.0001; n = 6) (B).(D) Kaplan-Meier survival analysis demonstrating significantly prolonged survival in the combination group (median 47.5 days), with several mice surviving beyond 60 days. All control and monotherapy cohorts succumbed by days 30–40 (control and LIFUS alone: median 20 days, VNP/ARG: 27.5 days, VNP/ARG-GVs: 25 days, n = 10).(E and G) Flow cytometry analysis (E) showing differential immune remodeling across groups. The details of the flow cytometry gating strategy are shown in Figure S3. LIFUS alone did not alter CD4^+^/CD8^+^ infiltration. VNP/ARG monotherapy induced about 1.5-fold increase in CD8^+^ T cells. Combination treatment produced a pronounced expansion of cytotoxic CD8^+^ T cells (63.96% ± 4.90% vs. 14.31% ± 2.0% in control; ∗∗∗∗p < 0.0001; n = 4) (G).(F and H) The combination group showed the strongest suppression of FOXP3^+^ Tregs (7.76% ± 1.60%), significantly lower than controls (28.38% ± 2.59%; ∗∗∗p < 0.005; n = 4) (H).(I and J) ELISA quantification of effector cytokines. IFN-γ levels were highest in the combination group (37.4 ± 6.88 ng/mL), followed by LIFUS alone (12.1 ± 2.3 ng/mL) and control (5.6 ± 1.4 ng/mL) (n = 4) (I). TNF-α expression followed a similar trend (1.37 ± 0.41 ng/mL vs. 0.42 ± 0.15 ng/mL and 0.18 ± 0.06 ng/mL) (n = 4) (J).(K and L) Representative Ki67 and hematoxylin and eosin staining showing reduced proliferation and increased apoptosis across groups, with the strongest effects observed in the LIFUS + VNP/ARG-GVs cohort. Scale bars, 100 μm.Data are representative of three (B, D, I, and J) or two (G and H) independent experiments. Data are mean ± SD or chi-square (D). Statistical analysis was performed using one-way ANOVA (B and H–K), two-way ANOVA (G), or log rank test (D). See also Figure S2.

We next assessed whether LIFUS-induced GV rupture supports sustained tumor monitoring and therapeutic efficacy. Although 0.8-MPa LIFUS induced partial GV-dependent bacterial lysis (∼69%), a substantial viable bacterial population remained in tumors. Acoustic cavitation imaging showed that acoustic intensity sharply decreased after insonation but returned by 72 h (39.33 ± 3.06 dB) (for details, see Figures S2C and 2D), reflecting a reproducible “drop-and-recover” dynamic driven by the balance between GVs’ collapse and bacterial regrowth. Based on this kinetic pattern, we implemented a 3-day interval between insonations, enabling five consecutive treatment cycles without loss of intratumoral acoustic visibility.

Therapeutically, LIFUS+VNP/ARG-GVs elicited marked tumor growth inhibition (Figures 2B and 2C). In contrast, VNP/ARG alone, whether induced or uninduced, did not significantly suppress tumor progression (p > 0.05), underscoring the essential contribution of LIFUS-generated mechanical forces. To further determine whether GVs alone possess intrinsic therapeutic effects, we also administered purified GVs i.v. with or without LIFUS. As expected, neither GVs alone nor GVs combined with LIFUS reduced tumor burden (p > 0.05) (Figures S2E and S2F), consistent with the poor tumor accumulation and rapid systemic clearance of injected GVs, which restricts the intratumoral GV concentration during the treatment cycle (administered once over the 21-day treatment period to match the in vivo protocol) and limits the magnitude of LIFUS-induced mechanical forces. After completing five cycles, LIFUS treatment was discontinued, as tumors remained controlled without rebound growth. The combination therapy extended median survival to 47.5 days, significantly longer than control (20 days), LIFUS (20 days), VNP/ARG (27.5 days), or VNP/ARG-GV (25 days) groups (p < 0.0001; Figure 2D), with several mice surviving beyond 60 days.

Flow cytometry profiling revealed pronounced immune remodeling across treatment groups (Figures 2E–2H). In control, LIFUS, and VNP/ARG or VNP/ARG-GVs monotherapy tumors, the infiltrating lymphocyte compartment was dominated by CD4^+^ T cells. However, there was no significant difference (p > 0.05) in CD4^+^ or CD8^+^ T cell frequencies between the control and LIFUS groups, and both exhibited minimal T cell infiltration, consistent with the profoundly immunosuppressive nature of the TME. In contrast, the VNP/ARG and VNP/ARG-GV groups showed increased T cell infiltration, with both CD4^+^ and CD8^+^ T cell populations significantly elevated relative to control (p < 0.01), likely reflecting bacteria-driven immunogenic activation. Importantly, in the combination treatment group, CD8^+^ T cell levels increased dramatically (63.96% ± 4.90%) compared with both control (14.31% ± 2.0%, p < 0.0001) and VNP/ARG-GVs groups (24.65% ± 3.68%, p < 0.0001), demonstrating the strong immunomodulatory effect of LIFUS combined with VNP/ARG-GVs. This expansion was accompanied by a marked contraction of FOXP3^+^ regulatory T cells (7.76% ± 1.60%), shifting the TME toward a strongly effector-dominant immune landscape. Correspondingly, interferon-gamma (IFN-γ) and tumor necrosis factor-alpha (TNF-α) levels were significantly elevated (Figures 2I and 2J). Histological analysis revealed reduced Ki67 staining and widespread apoptosis specifically in the combination group (Figures 2K, 2L, and S2G).

Collectively, these findings show that VNP/ARG-GVs, in combination with LIFUS sustained by dynamic bacterial regrowth after each US pulse, generate a programmable cycle of mechanotransduction that enhances immune infiltration, amplifies cytokine activity, and achieves durable tumor suppression.

LIFUS-activated VNP/ARG-GVs remodel fibroblast-CD8+T cell crosstalk through NOTCH-mediated mechanotransduction

Given the strong antitumor activity observed in Figure 2, we performed single-cell RNA sequencing to investigate how LIFUS-activated VNP/ARG-GVs remodel immune-stromal interactions within the TME. Uniform manifold approximation and projection (UMAP) clustering identified eight major cell lineages (Figure 3A), and compositional analysis revealed a pronounced expansion of T and natural killer (NK) populations in the combination group relative to monotherapies (Figure 3B). Within lymphocytes, CD8^+^ T cells were the dominant cytotoxic subset (Figures 3C and 3D), prompting further subcluster analyses.Figure 3. Single-cell sequencing analysis identifies a treatment-enriched CD8^+^ effector subset and reveals mechanosensitive features of T cell activation(A) UMAP showing eight major TME cell populations across all groups: B cells, endothelial cells, epithelial cells, fibroblasts, myeloid cells, neutrophils, smooth muscle cells, and T/NK lymphocytes.(B) Quantification of T cell abundance, showing marked expansion of T lymphocyte populations in the combination group.(C and D) Subclassification identified CD8^+^ T cells as the predominant cytotoxic lymphocyte subset.(E) Unsupervised clustering identified 13 distinct subpopulations (Cd8t_00 to Cd8t_12). The proportions of Cd8t_00, Cd8t_01, and Cd8t_02 compared to the control group show an obvious increase, with the highest enrichment of Cd8t_02 (red line) in the combination group.(F) The UMAP projection showed the subset composition and expression of marker genes across clusters to annotate each cluster with specific biological identities.(G) Functional scoring demonstrates that Cd8t_02 shows high cytotoxicity with low inhibitory signatures, whereas Cd8t_01 displays increased cytotoxic scoring but is concurrently enriched for exhaustion markers.(H) GO enrichment shows Cd8t_02 transcriptional programs associated with adhesion, cytoskeletal organization, and T cell activation, features linked to biomechanical responsiveness. See also Figure S4.

Based on the single-cell RNA sequencing analysis of CD8^+^ T cells, we performed unsupervised clustering, which identified 13 distinct subpopulations (denoted as Cd8t_00 to Cd8t_12). In the combined treatment group, we observed a significant increase in the proportions of Cd8t_00, Cd8t_01, and Cd8t_02 compared to the control group (Figure 3E). To characterize these subpopulations, we examined their spatial distribution in the UMAP projection (Figure 3F) and analyzed the expression of marker genes across clusters (Figure S4A). This allowed us to annotate each cluster with specific biological identities: Cd8t_00 (Tm) corresponds to memory T cells, expressing markers such as CD27 and CD28; Cd8t_01 (Tex) represents exhausted CD8^+^ T cells, with high expression of exhaustion-related genes including Pdcd1^+^ (PD-1) and Havcr2^+^ (TIM-3); and Cd8t_02 is identified as effector memory T cells, exhibiting strong effector functions (Gzmb^+^ and Gzmk^+^) and memory potential. We further evaluated the cytotoxic and inhibitory activity of each cluster through functional scoring. Although Cd8t_01 also displayed a relatively high cytotoxicity score, its prominent exhaustion signature limits its contribution to antitumor immunity. In contrast, Cd8t_02 demonstrated not only significant expansion in the combination treatment group but also the most robust cytotoxic gene program, including the highest expression of Gzmb and Gzmk, indicating potent effector capability.

Therefore, we focused on Cd8t_02 as the core cytotoxic population for downstream analysis. Gene Ontology enrichment analysis revealed that this cluster is significantly enriched for terms related to cell-cell adhesion, actin cytoskeleton remodeling, and T cell activation, features characteristic of mechanosensitive effector T cells operating within a stiff stromal matrix (Figure 3H). Intercellular adhesion within the TME is modulated by the stiffness and elasticity of the dense stromal matrix, with relevant adhesion-associated proteins (e.g., collagen) primarily secreted by CAFs, the principal stromal constituents in tumor matrix composition.29^,^30^,^31

Given that adhesion- and cytoskeleton-associated programs often reflect mechanical engagement with the extracellular matrix, we next examined whether LIFUS-induced stromal remodeling contributes to CD8^+^ T cell activation. US elastography showed that LIFUS + VNP/ARG-GVs markedly decreased tumor stiffness (12.95 ± 2.18 kPa), compared with controls (57.19 ± 15.02 kPa, p < 0.005) and VNP/ARG-GV alone (46.82 ± 8.13 kPa, p < 0.01), a trend further corroborated by atomic force microscopy (AFM) measurements (Figures 4A–4C). Immunofluorescence demonstrated dense fibroblast accumulation in control tumors, unchanged fibroblast content with VNP/ARG-GVs alone, and a pronounced reduction following the combined treatment (Figures 4D and S4B). This pattern indicates that mechanical forces, rather than bacterial colonization alone, drive stromal decompression and CAF attenuation. Correspondingly, contrast-enhanced US demonstrated significantly improved perfusion in the combination group (Figures S4C and S4D), consistent with reduced matrix-induced vascular compression and increased tissue permeability.32^,^33Figure 4. Mechano-regulation of CAF-CD8^+^ T cell NOTCH signaling interactions(A and B) Ultrasound elastography (A) shows a significant reduction in tumor stiffness with the combination group (12.95 ± 2.18 kPa) compared with controls (57.19 ± 15.02 kPa, ∗∗∗p < 0.005) and VNP/ARG-GV monotherapy (46.82 ± 8.13 kPa, ∗∗p < 0.01), LIFUS monotherapy group (36.93 ± 6.812 kPa), and the control group (57.19 ± 15.02 kPa, ∗p < 0.05) (n = 4) (B).(C) AFM further confirms reduced matrix elastic modulus (n = 12).(D) Collagen I distribution secreted by fibroblasts in Figure S4B supported the hypothesized changes in fibroblasts within the TME. Quantitative assessment (D) showed a significant reduction in the combined therapy group (42.24 ± 8.46 mm^2^) compared with the control group (248.62 ± 9.19 mm^2^, ∗∗∗∗p < 0.0001; n = 3).(E) The differential analysis visualizes changes in interaction numbers and strength, where red indicates enhanced and blue indicates reduced interactions in the combined group vs. control. The highlighted box reveals Cd8t_04 in the differential number of interactions and Cd8t_00 in differential interaction strength with reduced interactions with CAF.(F) Relative information flow of major stromal-immune signaling pathways. Relative contribution of fibroblast-derived signaling pathways to Cd8t_04 cells in control and combination treatment groups, normalized to a scale of 0–1. The dashed line at 0.5 indicates equal contribution between groups. NOTCH signaling dominates CAF–Cd8t_04 cell communication in control tumors and is selectively reduced following combination treatment, whereas CXCL and IFN-II pathways show a relative increase.(G–I) Notch1 and Jagged1 expression at the transcript (G and H) (n = 6) and protein levels (I) are significantly reduced after combination therapy.Data are representative of three independent experiments. Data are mean ± SD. Statistical analysis was performed using one-way ANOVA (B–D) or two-tailed unpaired Student’s t test (G and H). See also Figure S4.

Building on the functional validation showing a reduction in fibroblast abundance and matrix stiffness within the combined treatment group, we next performed a cell-cell communication analysis to dissect the interactions between CD8^+^ T cell subsets and fibroblasts. The differential analysis (Figure 4E) visualizes changes in interaction numbers and strength of ligand-receptor interactions, where red indicates enhanced and blue indicates reduced interactions in the combined group vs. control, highlighting two CD8^+^ subsets, Cd8t_04 and Cd8t_00, that each showed marked reductions in the number and strength of interactions with fibroblasts. Annotation based on Figure S4A identifies that Cd8t_00 corresponds to conventional memory T cells and Cd8t_04 to central memory T cells with high effector potential; we hypothesized that decreased stromal contact might differentially affect their functional states. Gene Ontology enrichment of Cd8t_04 (Figure S4E) highlighted strong involvement in RNA metabolism and ribosome biogenesis, suggesting that this population possesses high translational activity, supporting a functionally primed state for rapid differentiation into effector cells once released from stromal suppression, such as NOTCH signaling derived from CAF.34^,^35

Pathway-level analysis further supported these findings. Among the major stromal-immune communication pathways, NOTCH signaling emerged as the dominant fibroblast-to-Cd8t_04 regulatory axis in control tumors, as reflected by its relative information flow exceeding the 0.5 threshold (Figure 4F). Upon combination treatment, NOTCH signaling was selectively and markedly suppressed, indicating a pronounced shift away from CAF-driven NOTCH-mediated regulation. In contrast, TGFβ and WNT signaling showed milder, non-significant change, while CXCL and IFN-II pathways exhibited slight increases, aligning consistent with a broader shift toward an immunostimulatory microenvironment. The quantification of absolute information flow (Figure S4F) reinforced that NOTCH had the most substantial reduction in overall signaling strength among all evaluated pathways, reinforcing its role as the primary mechanosensitive pathway. Concordantly, Notch1 and Jagged1 levels were significantly decreased at both the mRNA and protein levels in the combination group (Figures 4G–4I), establishing NOTCH suppression as a key mechanotransductive outcome.

Together, these data suggest that the combined therapy disrupts NOTCH-mediated CAF-CD8^+^ T subset crosstalk, where the NOTCH pathway plays a role in suppressing CD8^+^ T cell activity within the TME through CAFs, partially relieving Cd8t_04 cells from stromal suppression and permitting their expansion and differentiation into cytotoxic effectors, thereby enhancing antitumor immunity.36^,^37^,^38^,^39

LIFUS-driven GV forces exert a dual regulatory effect on CAFs and CD8+ T cells through NOTCH signaling

Our in vivo analyses revealed that LIFUS+VNP/ARG-GVs treatment reduces CAF abundance, lowers tumor stiffness, and weakens CAF-CD8^+^ T cell NOTCH communication. To dissect the mechanotransductive mechanisms underlying these observations, we established an in vitro CAF-CD8^+^ T cell co-culture system supplemented with purified GVs to avoid multiple confounding variables and exposed to LIFUS (Figure 5A). Confocal imaging showed that CAFs in the GVs-LIFUS (no LIFUS) group expressed high levels of Notch1 and Jagged1. In contrast, LIFUS-triggered GV (GVs + LIFUS group) cavitation markedly reduced both proteins (Figure 5B), consistent with western blotting results (Figure 5C). These findings indicate that mechanical forces directly repress NOTCH output in CAFs, aligning with the CAF depletion and stromal softening observed in vivo. Interestingly, CD8^+^ T cells responded differently to mechanostimulation: Jagged1 expression was reduced following GVs + LIFUS treatment, whereas Notch1 receptor levels remained unchanged (Figures 5D and 5E). Thus, this dual regulatory weakening of CAF-derived Jagged1 disrupts CAF-mediated suppression, while preserving CD8^+^ T cell capacity to receive pro-activation NOTCH cues.40Figure 5. Dual mechanotransductive regulation of CAFs-CD8^+^ T cell crosstalk via NOTCH signaling(A) The schematic diagram of co-culture systems of CAFs and CD8^+^ T cells with GVs responsive to LIFUS constructs in vitro. Before 60 s of LIFUS exposure, cell co-incubation with purified GVs allowed sufficient GV-cell contact.(B and C) Confocal imaging revealed a significant reduction in Notch1 receptor and Jagged1 ligand expression in CAFs exposed to LIFUS-driven GVs (GVs + LIFUS group) compared to unstimulated controls (GVs − LIFUS group). Scale bars, 100 μm.(D and E) In CD8^+^ T cells, Jagged1 expression was decreased following LIFUS-driven GV stimulation (GVs + LIFUS group), whereas Notch1 receptor levels remained unchanged, indicating asymmetric NOTCH pathway modulation between CAF and CD8^+^ T cell. Scale bars, 100 μm.(F) Experimental workflow for assessing CAFs’ functional changes following LIFUS-driven GV treatment.(G and H) Transwell migration assays revealed significantly reduced CAF motility in the LIFUS treatment group (56 ± 4.0 cells/field) compared with the control group (150.67 ± 15.04 cells/field, ∗∗∗∗p < 0.0001) and CAFs exposed to treatment without LIFUS (101.67 ± 3.79 cells/field, ∗∗p < 0.01; n = 3). Scale bars, 100 μm.(I and J) ELISA quantification of CAF-secreted factors showed decreased TGF-β (0.39 ± 0.07 vs. 0.68 ± 0.38 ng/mL in control, ∗∗∗p < 0.005) (n = 3) (I) and collagen I (0.27 ± 0.11 vs. 0.76 ± 0.06 ng/mL in control, ∗∗p < 0.01) (n = 3) (J) after LIFUS-driven GV stimulation, consistent with reduced CAF activation.(K) Schematic of co-culture assay evaluating CD8^+^ T cell (precondition with or without LIFUS-driven GVs) and 4T1 tumor cell interactions.(L and M) CD8^+^ T cells stimulated by LIFUS-driven GVs exhibited significantly enhanced adhesion to 4T1 tumor cells (51.89% ± 10.47% adherent cells/field) compared with control (3.25% ± 1.19%, ∗∗∗p < 0.005) and GV-only groups (15.72% ± 7.53%, ∗∗p < 0.01) (n = 3). Scale bars, 100 μm.(N) Functional validation showed the cytotoxicity of CD8^+^ T cells to 4T1-Luc cells (Figure S5C); the luminescence intensity, which represented the apoptosis of 4T1-Luc tumor cells, showed significant luminescence change at 12 h between CD8^+^ T^−LIFUS^ ((48 ± 6.3) × 10^5^ ps^−1^cm^−2^sr^−1^) and CD8^+^ T^+LIFUS^((17 ± 5.2) × 10^5^ ps^−1^cm^−2^sr^−1^, ∗∗p < 0.01) group, and a stronger decrease at 36 h compared to control group (∗∗∗∗p < 0.0001) (n = 3). A representative image is shown in Figure S5C.Data are representative of three independent experiments. Data are mean ± SD. Statistical analysis was performed using one-way ANOVA (H–J and M) or two-way ANOVA (N). Also see Figures S5A–S5C.

To determine the functional consequences of NOTCH suppression in CAFs, we assessed migration and profibrotic signaling (Figure 5F). Mechanical stimulation significantly reduced CAF migration across Transwell membranes (56 ± 4.0 vs. 150.67 ± 15.04 cells/field in control; p < 0.0001) (Figures 5G and 5H). Similarly, secretion of TGF-β and collagen I, key drivers of CAF activation and fibrosis, was markedly decreased after GVs + LIFUS treatment (Figures 5I and 5J). These results demonstrate that mechanical-force-induced NOTCH suppression attenuates CAF motility and profibrotic signaling, consistent with the decreased stromal stiffness observed in vivo.

We next assessed the ability of CD8^+^ T cells (GVs ± LIFUS treatment) to inhibit 4T1 tumor cells. To ensure homogeneous US exposure while minimizing acoustic attenuation, CD8^+^ T cells and 4T1 tumor cells were directly co-cultured within a Transwell insert whose diameter matched that of the US transducer. As shown in Figure 5K, CD8^+^ T cells preconditioned with GVs + LIFUS exhibited significantly enhanced adhesion to 4T1 tumor cells (51.89% ± 10.46% vs. 3.25% ± 1.19% in controls; Figures 5L and 5M), consistent with emerging evidence linking mechanical cues to T cell force-dependent killing.41^,^42^,^43 Mechanoprimed CD8^+^ T cells also produced markedly higher levels of granzyme A (GZMA) and IFN-γ (Figures S5A and S5B), confirming enhanced effector function. To further evaluate tumor cell killing, apoptosis of 4T1-Luc tumor cells was examined at 12 and 36 h post-treatment. Figures S4C and 5N indicate a significant luminescence change at 12 h between the CD8^+^ T^−LIFUS^ and CD8^+^ T^+LIFUS^ groups ([48 ± 6.3] × 10^5^ ps^−1^cm^−2^sr^−1^ vs. [17 ± 5.2] × 10^5^ ps^−1^cm^−2^sr^−1^; p < 0.01) and a stronger decrease at 36 h compared with the control group (p < 0.0001). These functional validations confirm a dual regulatory mechanism in CAFs and CD8^+^ T cells, where mechanotransduction-mediated NOTCH modulation, activation in CD8^+^ T cells, and inhibition in CAFs, synergistically enhance antitumor immunity, creating a critical perspective for the utilization of CD8^+^ T cells in preclinical tumor treatment.

Preclinical translation of mechanobiology-enhanced CD8+ T cell therapy

Having established that LIFUS-induced GV cavitation enhances CD8^+^ T cell cytotoxicity in vitro, we next evaluated the translational relevance of this mechanobiological strategy in adoptive T cell therapy. We employed two complementary tumor models: an orthotopic B16-ovalbumin (OVA) melanoma model treated with OT-1 cells and a 4T1-luciferase metastatic lung model treated with CAR-T cells.

As illustrated in Figure 6A, B16-OVA tumors were preconditioned with GVs + LIFUS in the OT-1^+LIFUS^ group before OT-1 infusion, while the comparison group received OT-1 cells without LIFUS priming (OT-1^−LIFUS^). After 15 days, the OT-1^+LIFUS^ group showed the most pronounced tumor regression (Figure 6B) and significantly reduced tumor weight (0.18 ± 0.04 g vs. 0.71 ± 0.04 g in controls; p < 0.0001) (Figure 6C). Importantly, mouse body weight remained stable during the treatment period (16.55 ± 0.15 g on day 15 vs. 16.57 ± 0.11 g at baseline; p > 0.05) (Figure 6D). Flow cytometry of tumor-infiltrating OT-1 cells revealed a marked restoration of effector function following mechanobiological priming. The fraction of exhausted PD-1^+^Tim-3^+^ OT-1 cells dropped to 2.55% ± 1.27% in the OT-1^+LIFUS^ group compared with 43.0% ± 4.60% in control (p < 0.0001) (Figure 6E). Correspondingly, TNF-α^+^ and IFN-γ^+^ effector OT-1 subsets were significantly increased (Figure 6F), and the proportion of Ki-67^+^ proliferating OT-1 cells rose to 63.20% ± 8.43%, compared with 10.57% ± 2.25% in control (p < 0.0001) (Figure 6G). Together, these findings indicate that LIFUS-driven GVs’ mechanical priming strongly reinvigorates tumor-infiltrating OT-1 cells, enhancing both proliferative fitness and cytotoxic function.Figure 6. Mechanobiology-enhanced adoptive T cell therapy improves control of primary and metastatic tumors(A) Schematic representation of the B16-OVA orthotopic melanoma model and the OT-1-based treatment strategy. The GVs (1 × 10^6^ particles/mL) were injected intratumorally every cycle with LIFUS or without LIFUS before OT-1 cells (1 × 10^6^ cells) were adoptively transferred later.(B) Tumor growth curves over 15 days. OT-1^+LIFUS^ treatment resulted in the most pronounced tumor inhibition, exhibiting significantly smaller tumor volumes than both OT-1^−LIFUS^ and control groups (266.7 ± 127.2 mm^3^ vs. control 1,800 ± 1.003 mm^3^, ∗∗∗∗p < 0.0001, vs. OT-1^−LIFUS^ 796.7 ± 613.2 mm^3^; ∗∗∗p < 0.005; n = 3).(C and D) Tumor (C) and body weight (D) measurements. The OT-1^+LIFUS^ group exhibited a marked reduction in tumor weight (0.18 ± 0.04 g vs. 0.71 ± 0.04 g in controls; ∗∗∗∗p < 0.0001; n = 5). In contrast, body weight in OT-1^+LIFUS^ group remained stable throughout the 15-day treatment period (16.55 ± 0.15 g on day 15 vs. 16.57 ± 0.11 g at baseline; p > 0.05; n = 3).(E) Flow cytometric quantification of exhausted OT-1 cells within tumors. The OT-1^+LIFUS^ group exhibited a 94.1% reduction in PD-1^+^Tim-3^+^-co-expressing OT-1 cells compared with controls (2.55% ± 1.27% vs. 43.0% ± 4.60%; ∗∗∗∗p < 0.0001; n = 5).(F) Intracellular cytokine staining of TNF-α and IFN-γ. OT-1^+LIFUS^ significantly enhanced effector cytokine production relative to both comparison groups (∗∗∗∗p < 0.0001; n = 5).(G) Proliferative activity of intratumoral OT-1 cells, assessed by Ki-67 expression. OT-1^+LIFUS^ showed the highest proportion of Ki-67^+^ OT-1 cells (63.20% ± 8.43%), indicating improved expansion and functional fitness (∗∗∗∗p < 0.0001; n = 5).(H) Schematic of the 4T1-luciferase lung metastasis model and CAR-T-based treatment strategy. CAR-T cells were incubated with GVs (1 × 10^6^ particles/mL, 1:1 ratio) for 10 min and exposed to LIFUS 60 s before i.v. injection.(I) Bioluminescence analysis of metastatic tumor burden. Photon flux in the CAR-T^+LIFUS^ group was reduced the most relative to controls (mean (4.23 ± 2.09) × 10^7^ ps^−1^cm^−2^sr^−1^ at the endpoint), consistent with substantially attenuated metastatic progression (n = 5).(J) Representative longitudinal bioluminescence imaging. Detectable thoracic luciferase signal at day 0 confirmed successful establishment of lung metastases. Metastatic burden increased rapidly in controls beginning at day 5, whereas CAR-T^+LIFUS^ exhibited minimal signal on days 10–15 and near-complete regression by day 25 ((1.6 ± 9.3) × 10^4^ ps^−1^cm^−2^sr^−1^ vs. (5.3 ± 2.3) × 10^7^ ps^−1^cm^−2^sr^−1^ in the control group, ∗∗∗∗p < 0.0001). Tails were covered to prevent signal interference from freshly administered D-luciferin (n = 5) (see Figure S5D).(K) Kaplan-Meier survival analysis. CAR-T^+LIFUS^ markedly prolonged survival (median 63.5 days) compared with CAR-T monotherapy (47.5 days; ∗∗p < 0.01) and controls (35.5 days; ∗∗∗p < 0.005) (n = 6).Data are representative of two (B–G, I, and K) independent experiments. Data are mean ± SD (B–G), the mean value (I), or the Chi square (K). Statistical analysis was performed using one-way ANOVA (C and E–G), two-way ANOVA (B and D), or log rank test (K). Also see Figures S5D–S5F.

To evaluate therapeutic benefit in metastatic disease, which poses greater physical and immunologic barriers, a 4T1-luc lung metastasis model was established (Figure 6H). Given that the aerated lung parenchyma poses a fundamental barrier to US by scattering and attenuating acoustic waves, consistent in vivo insonation of metastatic deposits is not feasible. Therefore, we employed a GV-based mechano-priming approach to activate CAR-T cells ex vivo prior to administration. Longitudinal bioluminescence imaging of lung tissue revealed rapid metastatic progression in the control group (mean [4.23 ± 2.09] × 10^7^ ps^−1^cm^−2^sr^−1^ at the endpoint), whereas CAR-T^+LIFUS^ mice exhibited a 92.7% reduction in metastatic burden, with markedly suppressed photon flux over time (Figure 6I), consistent with the minimal metastatic nodules observed in lung tissue. Representative images and quantitative photon flux measurements of significant lung metastasis via small animal imaging (Figures 6J and S5D) showed thoracic luciferase signals in each group at day 0, followed by rapid expansion in the control group by days 5–15, with all mice in this group succumbing by day 25. CAR-T^–LIFUS^ group mice maintained moderate tumor loads early but succumbed around day 20. In the CAR-T^+LIFUS^ group, mild lung metastasis was observed on days 10 and 15. However, by day 25, the fluorescence intensity had almost vanished ([1.6 ± 9.3] × 10^4^ ps^−1^cm^−2^sr^−1^), representing a substantial reduction compared to the control group ([5.3 ± 2.3] × 10^7^ ps^−1^cm^−2^sr^−1^, p < 0.0001) and a significant decrease relative to the CAR-T^−LIFUS^ group ([12.3 ± 9.1] × 10^6^ ps^−1^cm^−2^sr^−1^, p < 0.05). Notably, no deaths were observed in the group through day 45. These imaging data align with ex vivo lung pathology (Figure 6I).

Survival analysis confirmed these therapeutic gains (Figure 6K). CAR-T^+LIFUS^ prolonged median survival to 63.5 days, significantly longer than CAR-T monotherapy (47.5 days) and control animals (35.5 days; p < 0.005).

Together, these preclinical results demonstrate that LIFUS-mediated mechanobiological stimulation, delivered either as tumor preconditioning or as direct T cell priming, substantially enhances the efficacy of adoptive T cell therapies across both primary and metastatic tumor settings. This establishes LIFUS-driven mechanotransduction as a potent, generalizable strategy to augment T-cell-based immunotherapy for solid and metastatic tumors.

Discussion

In this study, we demonstrate that GVs encoded by tumor-colonizing engineered Salmonella VNP20009 can serve not merely as acoustic reporters, but as intratumoral mechanotransducers that convert LIFUS energy into biologically meaningful mechanical cues. This LIFUS-GV interface reprograms both the physical and immunoregulatory landscapes of desmoplastic tumors by simultaneously decompressing the stromal matrix and disrupting NOTCH-mediated CAF-CD8^+^ T cell communication, effects that cannot be achieved by bacteria- or LIFUS-based platforms alone. Through integrated analyses including nonlinear US imaging, single-cell RNA sequencing, elastography, and in vitro mechanistic studies, we outline a mechanistic framework in which mechanical forces operate as a programmable immunomodulatory input within the TME.

Unlike previous US-responsive bacterial systems that rely primarily on drug release, thermal gene switches, or passive acoustic reporting,44^,^45^,^46 our platform leverages GVs as intracellular structures capable of collapsing under LIFUS to generate localized mechanical stress. The NOTCH pathway operates through receptor-ligand interactions, triggering release of the Notch intracellular domain to regulate target genes and maintain cellular homeostasis.47^,^48 This mechanical event directly attenuates CAF-derived Notch1-Jagged1 signaling, a stromal axis known to restrain CD8^+^ T cell activation and maintain T cell quiescence within stiff matrices.49^,^50 Single-cell profiling revealed that this disruption particularly affects CD8t_04 central-memory-like cells, highly dependent on fibroblast-derived cues, while promoting their rapid differentiation.51^,^52 In parallel, LIFUS-driven GV cavitation reduces matrix stiffness and improves perfusion, thereby alleviating the biophysical constraints that otherwise prevent immune infiltration.53^,^54 Together, these molecular and biophysical changes establish a previously unrecognized “mechano-NOTCH axis” governing T cell fate in solid tumors.

To isolate the mechanical component of this phenomenon, we performed in vitro CAF-T cell assays using purified GV instead of the engineered bacteria. This design excluded confounders such as pathogen-associated molecular patterns (PAMPs) or metabolic by-products and showed that mechanical forces alone are sufficient to repress CAF-NOTCH signaling, diminish profibrotic activity, and potentiate CD8^+^ T cell adhesion and cytotoxicity. These results provided the conceptual basis for extending the approach into a translational “mechano-priming” strategy,55^,^56 where either tumors or therapeutic T cells are exposed to LIFUS-driven GV cavitation before adoptive transfer. This mechano-priming strategy that potentiates adoptive T cell therapies (e.g., CAR-T) is particularly important from a translational perspective because current CAR-T therapies face substantial challenges, including complex and costly manufacturing, as well as loss of T cell potency caused by excessive responsiveness to the suppressive TME and dense stromal matrix.

This work also highlights a practical advantage of using engineered bacteria as endogenous GV bioreactors. Previous studies have indicated that an acoustic intensity threshold exceeding ∼0.72 MPa enables simultaneous US imaging.27 Our experimental validation confirmed that a parameter of 0.8 MPa ensured optimal LIFUS-mediated cavitation imaging of GVs while effectively controlling the survival period of the VNP/ARG in vivo. Following partial LIFUS-induced GV collapse and bacterial lysis, the surviving intratumoral population repopulates over 72 h, restoring acoustic visibility and enabling repeated, self-sustaining cycles of mechanical activation. This dynamic equilibrium, directly validated through in vivo imaging, addresses a key translational gap in US-controlled therapeutics by providing a stable and renewable intratumoral mechanical actuation source within deep tumor tissue.

Beyond mechanistic insights, this work carries important scientific and engineering implications. Mechanical force emerges as a precisely tunable immunoregulatory input, analogous to biochemical stimulation. Engineered bacteria function as renewable mechanical actuators embedded within tumors, enabling spatially controlled immunomodulation. Moreover, the mechano-priming approach offers a modular augmentation for adoptive cell therapies, potentially improving their performance in solid tumors where T cell exhaustion and stromal exclusion remain major bottlenecks. Collectively, this study uncovers a central role of mechanical forces in modulating the CAF-0CD8^+^ T cell NOTCH axis, remodels tumor biomechanics, and demonstrates potent synergy with adoptive T cell therapies.

Limitations of the study

Despite these advances, some limitations merit consideration. While NOTCH emerged as the most significantly altered signaling axis, other mechanosensitive pathways likely contribute to the observed remodeling and warrant deeper mapping using spatially resolved transcriptomics. Although LIFUS-induced partial bacterial lysis enhances biosafety by preventing uncontrolled proliferation, clinical translation will require additional containment strategies and large-animal verification. Moreover, anatomical constraints imposed by lung and air-tissue interfaces limit insonation in certain settings, motivating the development of adaptive phased-array US systems. Finally, integrating mechano-priming T cells or TME may further improve therapeutic durability.

Resource availability

Lead contact

Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Huixiong Xu (xu.huixiong@zs-hospital.sh.cn).

Materials availability

Engineered plasmids, purified GVs, and bacterial strains generated in this study will be made available with a completed materials transfer agreement by the lead contact upon request.

Data and code availability

• •All data generated or analyzed during this study are included in this manuscript (including supplementary information). The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive (GSA: CRA033977). All sequencing data are publicly available as of the date of publication.
• •No original code was generated.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was supported by the Nation National Science Foundation of China grants to H.X. (82430064), L.L. (82001830), H.Y. (82302206), X.G. (82402263), and C.Z. (82572222); Scientific Research and Development Fund of Zhongshan Hospital of Fudan University grant to H.X. (2022ZSQD07); Shanghai Sailing Program to H.Y. (23YF1441600); 10.13039/100007219Shanghai Natural Science Foundation to X.L. (24ZR1410700); 10.13039/501100002858China Postdoctoral Science Foundation to H.Y. (2023TQ0073); and Postdoctoral Fellowship Program of CPSF to L.L. (GZC20230499).

Author contributions

H.X. and H.Y. designed and supported the project; L.L. contributed to operations of experiment and paper writing; X.L. and W.G. provided guidance on gas vesicle engineering and immunofluorescence staining; X.L. and C.Z. provided access to LIFUS equipment; J.X. and X.G. analyzed the data.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesMouse anti-CD45ProteintechCat# 98035-1-RR; RRID: AB_3672181Mouse anti-CD3ProteintechCat# 17617-1-AP; RRID: AB_1939430Mouse anti-CD4ProteintechCat# 67786-1-Ig; RRID: AB_2918550Mouse anti-CD8ProteintechCat# 66868-1-Ig; RRID: AB_2882205Anti-FOXP3 antibodyProteintechCat# 22228-1-AP; RRID: AB_11182376Anti-Ki67 antibodyProteintechCat# 28074-1-AP; RRID: AB_2918145Anti-Collagen I antibodyProteintechCat# 66786-1-Ig; RRID: AB_2882131Anti-Notch1 antibodyProteintechCat# 10062-2-AP; RRID: AB_2153338Anti-Jagged1 antibodyProteintechCat# 66890-1-Ig; RRID: AB_2882220Anti-PD-1 antibodyProteintechCat# 98080-1-RR; RRID: AB_3672227Anti-Tim-3 antibodyProteintechCat# 82836-1-RR; RRID: AB_3670571Anti-mouse IFN-γ antibodyProteintechCat# 30293-1-AP; RRID: AB_3669709Anti-mouse TNF- α antibodyProteintechCat# 17590-1-AP; RRID: AB_2271853Anti-mouse Granzyme B antibodyProteintechCat# 13588-1-AP; RRID: AB_2114429Bacterial and virus strainsSalmonella enterica serovar Typhimurium VNP20009 (attenuated)Shanghai YuanChuang BiotechnologyN/ALentiviral/retroviral vector pRVKM-FMC63-CD28-RFPShanghai Jiman BiotechnologyN/ABiological samples4T1 tumor tissueBALB/cIsolated from mice raised in SPF environmentBlood sample of miceBALB/cIsolated from mice raised in SPF environmentMain organs of miceBALB/cIsolated from mice raised in SPF environmentChemicals, peptides, and recombinant proteinsL-arabinoseAladdinCat# A106196LB brothBeyotime Biotechnology Co.(Shanghai, China)Cat# ST156ChloramphenicolBeyotimeCat# ST1150KanamycinBeyotimeCat# ST102RPMI-1640 mediumGibcoC11875500BTRIPA lysis bufferBeyotimeCat# C90204LysozymeBeyotimeCat# ST206D-luciferin (potassium salt)AladdinCat# L120798CCK-8 Cell Counting KitBeyotimeCat# C0042SYTO 9/Propidium Iodide Live/Dead stainBeyotimeCat# C1399STumor Dissociation KitBeyotimeCat# C3702ACK Lysing BufferGibcoCat# A1049201DNase IRocheCat# 10104159001Dynabeads Human T-Activator CD3/CD28GibcoCat# 11161DStreptavidin Magnetic BeadsSelleckCat# B90011Recombinant mouse IL-2ProteintechCat# CK24-10UGParaformaldehyde (4%)BeyotimeCat# P0099-3LTriton X-100BeyotimeCat# D7046SBSA (bovine serum albumin)BeyotimeCat# AD1461DAB substrate kitBeyotimeCat# C0085SCritical commercial assaysChromium Single Cell 3′ Library & Gel Bead Kit10x GenomicsCat# PN-120237Chromium Single Cell A Chip Kit10x GenomicsCat# 1000265Cell Ranger software package10x GenomicsN/ADeposited datascRNA-seq data from 4T1 tumorsThis paperGSA: CRA033977Experimental models: Cell lines4T1 murine breast carcinoma cellsShanghai Tenth People’s Clinical and Translational Research Center)N/A4T1-Luc (luciferase-expressing 4T1) cellsShanghai Tenth People’s Clinical and Translational Research Center)N/AB16-OVA melanoma cellsShanghai Tenth People’s Clinical and Translational Research Center)N/APrimary mouse CAFs (tumor-derived fibroblasts)BALB/cIsolated from mice raised in SPF environmentPrimary mouse CD8^+^ T cells (tumor and splenic)BALB/cIsolated from mice raised in SPF environmentOT-1 CD8^+^ T cellsC57BL/6N/ACAR-T cells (FMC63-CD28-RFP)Shanghai Jiman BiotechnologyN/AExperimental models: Organisms/strainsMouse: BALB/c, female, 5–6 weeksShanghai Jihui Experimental Animal Breeding CenterN/AMouse: C57BL/6, female, 6–7 weeksShanghai Jihui Experimental Animal Breeding CenterN/AMouse: CD45.1^+^ wild-type recipientsShanghai Jihui Experimental Animal Breeding CenterN/AOligonucleotidesFor genetic construct validation, see Table S1This paperN/ARecombinant DNAPlasmid: pBAD-bARGSer-axe-txeShapiro et al.18Addgene Cat#192473Software and algorithmsGraphPad PrismGraphPad SoftwareVersion 10.2 https://www.graphpad.com/; RRID: SCR_002798SeuratSeuratVersion 4.0 https://satijalab.org/seurat/articles/get_started.html; RRID:SCR_016341Cell Ranger10x GenomicsVersion 6.0 https://support.10xgenomics.com/single-cell-gene-expression/software/pipelines/latest/; RRID:SCR_017344CellChat (R package)10x GenomicsVersion 6.0 https://github.com/sqjin/CellChat; RRID:SCR_021946ImageJImage, JavaVersion 1.8 https://imagej.net/ij/; RRID:SCR_003070Flow cytometry analysis software (e.g., FlowJo)BD BioscienceVersion 10 https://www.flowjo.com/; RRID:SCR_008520Ultrasound imaging and therapy systemMindray (China)Resona R9Shear Wave Elastography module (SWTQ)Mindray (China)Resona R9Small animal in vivo imaging systemPerkinElmerAniView100

Experimental model and study participant details

Bacterial strain

The attenuated Salmonella typhimurium VNP20009 strain was provided by Shanghai Yuan Chuang Biotechnology. The engineered VNP/ARG strain was provided by Addgene, the complete nucleotide sequence refer to Shapiro, M et al.18

Cell lines

The 4T1 (mouse breast cancer), luciferase-expressing 4T1 (4T1-Luc) cell lines were acquired from Shanghai Tenth People’s Clinical and Translational Research Center. Ovalbumin-expressing B16 (B16-OVA) was established as this paper described. All cell lines were screened for mycoplasma contamination, and no contamination was detected. All cell lines were cultured in Roswell Park Memorial Institute 1640 medium (RPMI-1640) or Dulbecco’s modified Eagle medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin at 37°C in a 5% carbon dioxide incubator.

Mouse models and care

BALB/c mice: Female BALB/c mice (5–6 weeks old) were used for 4T1 tumor models, safety assessments, and the 4T1 lung metastasis model. Mice were obtained from Shanghai Ji Hui Experimental Animal Breeding Center.

C57BL/6 mice: Female C57BL/6 mice (6–7 weeks old) were used for the B16-OVA tumor model.

All mice were housed under specific pathogen-free conditions. All animal experiments were approved by the Shanghai Tenth People’s Hospital Animal Ethics Committee (SHDSYY-2025-Y3216-01) and conducted in accordance with institutional guidelines.

Method details

Construction of VNP/ARG engineered bacteria

The ARG sequence-containing plasmid (Addgene) was digested with appropriate restriction enzymes. The purified ARG fragment was ligated into a VNP20009-compatible vector. Recombinant plasmids were screened, verified by Sanger sequencing, and electroporated into the attenuated Salmonella typhimurium VNP20009 strain. Successful ARG integration was confirmed via polymerase chain reaction. Primer sequences used for qPCR are listed in Table S1.

Coomassie blue staining for GV protein expression

VNP/ARG bacteria were cultured in Luria-Bertani medium to an OD_600_ of approximately 0.13. Cultures were divided into control (VNP20009 wild-type), non-induced (VNP/ARG, -Ara), and induced (VNP/ARG-GVs, +0.2% L-arabinose for 4 h) groups. Bacterial pellets were collected by centrifugation (600× g, 30 min), washed with phosphate-buffered saline (PBS), and lysed in RIPA buffer containing protease inhibitors. Lysates were sonicated on ice (4 pulses of 5 s each) and clarified by centrifugation (12,000× g, 10 min, 4°C). Total protein was quantified bicinchoninic acid assay. Twenty micrograms of total protein per sample were separated on a 12% sodium dodecyl sulfate–polyacrylamide electrophoresis gel. The gel was stained with 0.1% Coomassie Brilliant Blue R-250 in 40% methanol and 10% acetic acid for 1 h and destained until clear bands were visible. GV-associated protein bands in the ∼55–75 kDa region were compared across groups.

Purification of GVs

Induced VNP/ARG cultures (0.2% arabinose, 4 h) were pelleted (4°C, 600 × g, 30 min). The supernatant was digested with lysozyme for 1 h and sequentially filtered through 0.45 μm and 0.22 μm polyvinylidene fluoride membranes. GVs were purified via flotation centrifugation (12,000× g, 3 h, 4°C, swinging-bucket rotor). The top 0.5–1 mL buoyant layer containing GVs was collected and washed three times with ice-cold PBS using repeated flotation centrifugation. The final GV suspension was resuspended in PBS. GV morphology was confirmed by TEM, and concentration was standardized by measuring OD_500_, with OD_500_ = 0.3 used for all in vitro assays. GVs were stored at 4°C and used within one week.

In vitro US imaging

A 1% (w/v) agarose phantom was prepared in PBS. Induced VNP/ARG-GVs (1 × 10^8^ cells/mL) were mixed with liquid agarose (∼43°C) and poured into a custom-made dish after defoaming. US imaging was performed using a Resona R9 system with a 1 MHz transducer at acoustic pressures of 0.6 MPa and 0.8 MPa (8 kHz pulse repetition frequency, 16 cycles). Acoustic cavitation images were recorded at 2 min and 5 min post-sonication.

Bacterial viability under acoustic intensities

VNP/ARG bacteria (1 × 10^6^ cells/mL) embedded in molded agarose gels were exposed to 0.6 MPa or 0.8 MPa acoustic pressures for 5 min. Live/dead staining was performed using SYTO 9 and propidium iodide according to the manufacturer’s protocol. Live (green) and dead (red) bacteria were visualized and quantified using fluorescence microscopy.

Bacterial colonization assay

VNP/ARG (1 × 10^6^ cells/mL) were plated on Luria-Bertani agar. When colonies reached 70–80 CFU per plate (∼2 days), they were divided into non-induced and induced (0.2% L-arabinose) groups. Plates were exposed to different acoustic intensities for 5 min, and colony growth was observed after 24 h.

Cytotoxicity (cell counting kit-8) assay

The 4T1 cells (5 × 10^3^ cells/well) were seeded in 96-well plates and pre-cultured for 24 h. Cells were treated as follows: tumor cells only, VNP/ARG (1 × 10^6^ cells/mL), LIFUS + tumor cells, 4T1 + VNP/ARG-GVs, 4T1 + VNP/ARG-GVs + LIFUS, and 4T1 + VNP/ARG. After 24 h, the medium was replaced, and the cell counting kit-8 assay (Beyotime, C0042) was performed according to the manufacturer’s instructions. Absorbance was measured at 460 nm. Toxicity was calculated as: Toxicity (%) = [(A - A_0_)/(B - A_0_)] × 100%, where A is the experimental well, A_0_ is the blank, and B is the control.

In vivo safety assessment

BALB/c mice were intravenously injected with 1 × 10^6^ cells/mL of VNP/ARG or PBS (control). Body weight was monitored weekly. Blood samples were collected on days 1 and 3 post-injection for complete hematological analysis (including inflammatory markers and liver/kidney function tests). At the endpoint, heart, liver, and kidney tissues were collected for hematoxylin and eosin staining to assess any histopathological changes.

In vivo US imaging

4T1 tumors were established in the right flank of BALB/c mice. When tumor volume reached ∼100 mm^3^, 1 × 10^6^ cells/mL of induced VNP/ARG-GVs were administered intratumorally. US imaging was performed at a mechanical index of 0.02.

4T1 subcutaneous tumor therapy

BALB/c mice bearing 4T1 tumors were randomly assigned to five groups (n ≥ 5): Control, LIFUS only, VNP/ARG (uninduced), VNP/ARG-GVs (induced), and VNP/ARG-GVs + LIFUS. Mice were intravenously injected with 1 × 10^6^ cells/mL VNP/ARG. Three days later, 0.2% L-arabinose (50 μL) was administered intratumorally to induce GV expression. One day post-induction, US imaging was performed, followed by acoustic cavitation therapy (0.8 MPa, 8 kHz pulse repetition frequency, 16 cycles, mechanical index = 0.4) for 10 min. This cycle (arabinose induction followed by LIFUS therapy) was repeated on days 11, 15, and 19 (with LIFUS applied on days 12, 16, 20). Tumor volume (V = 0.52 × L × W^2^) and body weight were measured every three days. Survival was monitored for 60 days.

Flow cytometry for tumor-infiltrating T cells

Tumors and spleens were harvested and processed into single-cell suspensions. Cells (100 μL per sample) were stained with the following antibodies: CD45 (Proteintech, 240356D1), CD3 (Proteintech, 17617-1-AP), CD4 (Proteintech, 67786-1-lg), CD8 (Proteintech, 66868-1-lg), and FOXP3 (Gibco, 11161D). Cells were incubated for 15 min at room temperature in the dark, washed, and analyzed by flow cytometry.

The gating strategy involved sequentially selecting cell populations to ensure accurate analysis. First, cells were gated on FSC-A versus SSC-A to exclude debris, followed by FSC-H versus FSC-A to select singlets. Viability dye-negative cells were then identified as live cells, and among these, CD45^+^ leukocytes were selected, followed by CD3^+^ T cells, which were further gated into CD4^+^ or CD8^+^ subsets. Finally, intracellular FOXP3 expression was specifically assessed within the CD4^+^ T cell population.

Enzyme-linked immunosorbent assay

Serum levels of IFN-γ and TNF-α were determined using commercial ELISA kits (Proteintech, 30293-1-AP) according to the manufacturer’s instructions. Absorbance was measured, and cytokine concentrations were derived from a standard curve.

Immunohistochemistry

For Ki67 staining (Proteintech, 28074-1-AP) and Collagen I (Proteintech, 66786-1-Ig), tumor sections were deparaffinized, and antigen retrieval was performed using citrate buffer. Sections were blocked, incubated overnight at 4°C with primary antibodies, and then incubated with horseradish peroxidase-conjugated secondary antibodies. Staining was visualized with 3,3′-diaminobenzidine, and nuclei were counterstained with hematoxylin.

Single-cell RNA sequencing

Tumors were dissociated into single-cell suspensions using a Tumor Dissociation Kit (Miltenyi). Red blood cells were lysed with ammonium-chloride-potassium lysis buffer. Single-cell capture and library construction were performed using the 10x Genomics Chromium platform, targeting 5,000–10,000 cells per sample. Sequencing libraries were prepared for 3′-end transcriptomes. Raw sequencing data were processed with Cell Ranger, and downstream analysis, including normalization, principal component analysis, UMAP visualization, and clustering, was performed with the Seurat R package (v4.0). T cells were identified by CD3D/CD3E expression, and CD8^+^ T cell subsets were extracted for sub-clustering.

Cell-cell interaction analysis

The Cell Chat R package was used to infer cell-cell communication. Seurat objects with annotated cell types were imported into CellChat, and the software’s built-in ligand-receptor interaction database was used to compute and visualize intercellular signaling strength and network dynamics.

US elastography

Tumor stiffness was assessed using Shear Wave Elastography on a Resona R9 US system with an 8 MHz probe. Regions of interest were positioned at a depth of 1.5 cm. For each animal, four consecutive valid acquisitions were obtained and averaged. Quantitative analysis was performed using Shear Wave Transient Quantification to determine tissue elasticity.

Fibroblast and CD8+ T cell isolation

CAFs: Tumor tissue from an 8-week-old BALB/c mouse was finely minced and enzymatically digested with pancreatin to obtain a single-cell suspension. Following red blood cell lysis, fibroblasts were resuspended in RPMI-1640 medium and cultured under standard conditions.

CD8^+^ T Cells: The spleen from an 8-week-old BALB/c mouse was mechanically dissociated and filtered through a 70 μm cell strainer. After red blood cell lysis, cells were resuspended in sorting buffer. CD8^+^ T cells were isolated using a negative selection kit (Selleck, B90011) according to the manufacturer’s protocol. Purified T cells were activated with Dynabeads Human T-Activator CD3/CD28 (Gibco, 11161D) and supplemented with interleukin-2 (Proteintech, CK24-10UG).

Co-culture of CAFs and CD8+ T cells

CAFs were seeded and grown to 70–80% confluence. Isolated CD8^+^ T cells were added at a 1:5 (CAFs: T cells) ratio in RPMI-1640 medium. After 24 h, co-cultures were incubated with purified GVs at a 1:1 T cells ratio for 10 min, followed by LIFUS exposure (1.0 MHz, 0.8 MPa, 16% duty cycle, 60 s). For transwell experiments, GV-loaded CD8^+^ T cells were placed in the upper chamber (0.4 μm pore membrane) and exposed to LIFUS under the same conditions.

Immunofluorescence for Notch1 and Jagged1

After co-culture, CAFs and CD8^+^ T cells were separated. Cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and blocked with bovine serum albumin. Samples were incubated with primary antibodies against Notch1 (Proteintech, KHC1061) and Jagged1 (Proteintech, 66890-1-lg) overnight at 4°C, followed by fluorochrome-conjugated secondary antibodies. Nuclei were stained with 4′,6-diamidino-2-phenylindole, and samples were visualized using fluorescence microscopy.

CAF migration assay

CAFs (from T^+LIFUS^ and T^−LIFUS^ groups) were resuspended in serum-free medium at a density of 1–5 × 10^5^ cells/mL. Two hundred microliters (200 μL) of the cell suspension was added to the upper chamber of a Transwell insert, while the lower chamber was filled with 600 μL of medium containing 10% fetal bovine serum as a chemoattractant. After 12 h, non-migrated cells were carefully removed from the upper surface of the membrane, and migrated cells on the lower surface were fixed, stained with 0.1% crystal violet, and counted under a microscope.

CD8+ T cell – 4T1 co-culture and functional assays

To ensure that CD8^+^ T cells were uniformly exposed to the ultrasound field while minimizing acoustic attenuation, we first employed Transwell chamber system whose diameter matched that of the ultrasound transducer (24 mm) for assessing CD8^+^ T cell functional cytotoxicity. Levels of GZMA and IFN-γ in the supernatant were measured by ELISA.

Red fluorescent protein (RFP)-expressing CD8^+^ T cells (CD8–RFP) were co-cultured with 4T1 cells at a 5:1 ratio on glass-bottom dishes. Where indicated, co-cultures were pretreated with GV ± LIFUS. After 4 h, non-adherent cells were washed off, and the remaining cells were fixed. Adherent RFP^+^ T cells were counted, and the adhesion index was calculated as (RFP^+^CD8^+^ T cells attached to 4T1 cells/total number of RFP^+^CD8^+^ T cells) × 100%.

For cytotoxicity, 4T1-Luc cells were co-cultured with differently treated CD8^+^ T cells. After 12 h and 36 h, D-luciferin was added, and bioluminescence was imaged to assess tumor cell viability. Levels of granzyme A and IFN-γ in the supernatant were measured by ELISA. The quantitative analysis of the bioluminescence signals was record to represent apoptosis of 4T1-Luc cells.

B16-OVA tumor model and treatment

C57BL/6 mice were sensitized with OVA via intraperitoneal injection for three consecutive days. Three weeks later, mice were challenged with an OVA nasal drip, followed by subcutaneous inoculation of B16-OVA cells. Mice were divided into three groups: Control (WT), OT-1^−LIFUS^, and OT-1^+LIFUS^ groups. OVA-specific CD8^+^ T cells (OT-1 cells) were prepared. GVs (1 × 10^6^ particles/mL) were injected intratumorally every cycle, and OT-1 cells (1 × 10^6^ cells/mL) were adoptively transferred later. The treatment consisted of 5 cycles. Tumor growth was monitored.

Flow cytometry for B16-OVA model

Tumor-infiltrating immune cells were isolated as described. Cells were stained with a viability dye and the following antibodies: anti-PD-1 (Proteintech, 240389B8), anti-Tim-3 (Proteintech, 82836-1-RR), anti-TNF-α, anti-IFN-γ, and anti-Ki67. Cells were analyzed by flow cytometry. Gating strategy: single, live, CD45^+^, CD3^+^, CD8^+^ T cells were analyzed for PD-1/Tim-3, TNF-α/IFN-γ, and Ki-67 expression.

4T1 lung metastasis model and CAR-T cell therapy

BALB/c Mice were intravenously injected with 1 × 10^5^ 4T1-Luc cells. CAR-T cells were generated by transfecting activated CD8^+^ T cells with the pRVKM-FMC63-CD28-RFP plasmid. For mechano-priming, CAR-T cells were incubated with GVs (1 × 10^6^ particles/mL, 1:1 ratio) for 10 min and exposed to LIFUS (1.0 MHz, 0.8 MPa, 16% duty cycle, 60 s). The treatment consisted of 5 cycles. Mechanoprimed CAR-T cells (CAR-T^+LIFUS^, 1 × 10^6^ cells) were washed and infused intravenously. Lung metastasis was monitored every 5 days by bioluminescence imaging.

Live/dead staining of CAR-T cells

CAR-T cells were harvested and washed once with PBS. Following the designated experimental treatments (co-incubation with GVs 1:1 for 10min then exposure to LIFUS 60s), cells were centrifuged at 300×g for 5 min and resuspended in pre-warmed, dye-free complete culture medium. Cells viability was assessed using a two-color fluorescence assay based on membrane integrity. The cell suspension was incubated with 2 μM Calcein-AM (for live cells) and 4 μM propidium iodide (PI, for dead cells) for 20 min at 37°C in the dark. After incubation, cells were washed twice with PBS to remove excess dye and finally resuspended in a small volume of PBS for imaging. Fluorescence microscope were acquired to captured images from multiple random fields for each sample using a 20× objective.

In vivo bioluminescence imaging

Mice were injected intraperitoneally with D-luciferin (150 mg/kg) 10 min prior to imaging. Bioluminescence was acquired using an AniView100 system (exposure time: 1 min). A fixed region of interest over the thoracic area was used to quantify total photon flux, and the background signal was subtracted. Metastatic suppression was calculated as: [(mean bioluminescence of control - mean of treatment)/mean of control] × 100%.

Quantification and Statistical analysis

Statistical analyses were performed using GraphPad Prism 10.2. Data are presented as mean ± standard deviation, as specified in the figure legends. Statistical significance was determined using one-way or two-way analysis of variance, as appropriate. A p-value <0.05 was considered statistically significant; ns, no significant. Most experiments were independently repeated at least two or three times with consistent results. Specific sample sizes (n) for each experiment are provided in the corresponding figure legends.