Co-delivery of chemokine CXCL9 and costimulatory ligand TNFSF9 by mesenchymal stem cells reprograms the immune microenvironment for triple-negative breast cancer.

Introduction
Triple-negative breast cancer (TNBC) has a high risk of relapse and metastasis, which is characterized by high heterogeneity, accounting for approximately 20% of breast cancer1,2. Conventional treatments, including surgery, chemotherapy, and radiotherapy, have limited efficacy, with a median overall survival of only 18 months3–5. Although immunotherapy has emerged as a promising strategy for TNBC, its effectiveness is constrained by insufficient tumor immune infiltration6,7. Hence, novel therapies to increase immune infiltration are urgently needed.
Immune cell migration is regulated by chemokines through chemotactic gradients. Among chemokines, the C-X-C motif chemokine ligand (CXCL9) has been shown a positive association with survival rates in patients receiving immune checkpoint blockade (ICB) therapy, which mainly recruits effector T cells and natural killer (NK) cells through the CXCL9/C-X-C motif chemokine receptor 3 (CXCR3) axis8–11. Apart from chemotactic immune cells, activated immune cells in the tumor microenvironment (TME) are equally necessary. Tumor necrosis factor superfamily member 9 (TNFSF9), a ligand that binds 4-1BB (a co-stimulatory molecule), stimulates T-cell and NK-cell activation12,13. Thus, mimics of TNFSF9 are recognized as promising candidates for immunotherapy. Yet, systemic administration of these chemokines or agonist antibodies inevitably causes severe adverse effects, such as myocarditis and hepatitis13. Targeted delivery of immunotherapeutic agents is therefore an ideal way to minimize systemic toxicity.
Mesenchymal stem cells (MSCs) have good biological characteristics including variable sources, ease of isolation and culture, and immunological naivety. And so, they have been widely used as vehicles in cell-based immunotherapy for cancer treatment in recent years14–16. Consistent with our prior findings in colorectal cancer mouse models, adipose tissue-derived MSCs exhibited good tumor-homing ability and successfully delivered CXCL9 and TNFSF4 to induce local anti-tumor immune responses17. In another preclinical study, we showed that human umbilical cord mesenchymal stem cells (hUC-MSCs) expressing TNFSF9 exhibited inhibitory effects on lung and colorectal cancers18. Although many studies have showed the utility of MSCs as anti-tumor drug carriers in TNBC treatment, MSC-based immunotherapy for TNBC has not been extensively investigated19.
In this study, we engineered hUC-MSCs to successfully express TNFSF9 and CXCL9 (MSC-T9C9). In TNBC mouse models, the engineered MSCs effectively inhibited tumor growth via enhancing the infiltration and activation of CD8+ T cells and NK cells in the TME, enabling a controlled immune activation without significant adverse effects. Furthermore, MSC-T9C9 synergized with ICB to improve the antitumor efficacy. Our conclusions provide a solid foundation for future clinical applications of MSC-T9C9 as an effective and safe immunotherapy for TNBC patients.

Results

The tumor-homing ability of hUC-MSCs
The hUC-MSCs were isolated from human umbilical cord and identified by flow cytometry analysis based on the typical markers (including positive and negative markers) of hUC-MSCs. The identification results have been shown in our previous publication18. To reveal the biodistribution of hUC-MSCs in the 4T1 tumor bearing mouse model at different post-injection time points, hUC-MSCs were transfected with lentiviruses and subsequently were able to express the GFP protein stably. 1 × 106 GFP-labeled hUC-MSCs were injected into tumor-bearing mice via tail vein. The tumors and other organs (including lung, liver, spleen, and kidney) were collected on days 3, 7, and 10 after injection, respectively. According to the results of immunofluorescence staining (Fig. 1A–C), the intensity of green fluorescence was highest in tumor tissues three days after injection. However, no significant green fluorescence was detected in other organ tissues. With the prolongation of time, the fluorescence intensity in tumor tissues gradually decreased. Consistently, flow cytometric analysis quantitatively confirmed the time-dependent homing of GFP-labeled hUC-MSCs to the tumor site, with the percentage of GFP-positive cells among live cells peaking on day 3 and declining by days 7 and 10 (Fig. 1D, E). These results collectively suggest that hUC-MSCs have tumor-homing ability and reside in tumors for a relatively long time, indicating the potential of hUC-MSCs as an efficient vehicle.

Elevated expression of TNFSF9 and CXCL9 is associated with increased immune infiltration and an improved prognosis in breast cancer patients
To explore the prognostic values of TNFSF9 and CXCL9 expression in breast cancer patients, the dataset was included from The Cancer Genome Atlas Program (TCGA, https://portal.gdc.cancer.gov). The results of Kaplan-Meier curve analysis revealed that high level of CXCL9 expression was associated with favorable overall survival (OS, p < 0.05) of BRCA patients, but no such association was found for TNFSF9 (Fig. S1A, C). Furthermore, the relationship between these genes and immune cell signatures was estimated by using the ESTIMATE and Xcell algorithms. Both TNFSF9 and CXCL9 expression levels were strongly correlated with elevated ImmuneScore and increased infiltration of CD4+ T cells, CD8+ T cells and NK cells (Fig. S1B, D). Subsequently, TNFSF9 and CXCL9 were combined into a gene set (T9C9) and the gene set score for each sample was calculated using single sample gene set enrichment analysis (ssGSEA). A higher T9C9 score was significantly related with improved OS (p = 0.037) and enhanced immune cell infiltration (Fig. 2A, B), suggesting the co-elevation of TNFSF9 and CXCL9 is more favorable for patient outcomes.
To validate these findings across breast cancer molecular subtypes, we stratified the TCGA-BRCA cohort into basal and luminal subtypes. Notably, the association between the T9C9 signature and both survival benefit and immune infiltration remained significant within these subtypes, corroborating the results observed in the overall cohort (Fig. S2A–D). Additionally, a heatmap intuitively demonstrates the strong correlation between T9C9 score and activated immune infiltration across various breast cancer datasets (Fig. S2E). The findings demonstrate that both TNFSF9 and CXCL9 are associated with anti-tumor immune signatures.

Construction and evaluation of MSC-T9C9 cells
To generate engineered hUC-MSCs (MSC-T9C9), the cells were transduced with lentivectors encoding both TNFSF9 and CXCL9. Secreted CXCL9 was measured by ELISA, which revealed that MSC-T9C9 produced nearly 4000 pg/ml of CXCL9 after 48 hours in culture, compared to negligible levels in the MSC-Vec control (Fig. 2C). Meanwhile, the expression of TNFSF9 on the membrane surface of MSCs was examined by flow cytometry. The results showed that TNFSF9 expression on MSC-T9C9 was approximately 2-fold higher than on MSC-Vec (Fig. 2D). Compared to MSC-Vec, MSC-T9C9 showed no change in proliferative capacity in vitro, eliminating any potential for result bias due to differences in cell number in subsequent experiments (Fig. S3A).
To evaluate the immunostimulatory function of MSC-T9C9, the cells were co-cultured with mouse splenic lymphocytes. Flow cytometric analysis revealed the expression of activation marker CD69 was obviously upregulated on CD4+ T, CD8+ T and NK cells when co-cultured with MSC-T9C9 as compared to MSC-Vec (Fig. 2E). Particularly, while a proportion of T cells remained unactivated in the MSC-Vec group, the vast majority were activated upon co-culture with MSC-T9C9. Additionally, a robust increase in Granzyme B (Gzmb) secretion by CD8+ T and NK cells was observed after co-cultured with MSC-T9C9 (Fig. 2F, G).
Then, subcutaneous tumorigenic ability was also evaluated. No tumor formation was observed in either the MSC-Vec or MSC-T9C9 groups during the 6-month period following the injection of 1 × 106 cells (Fig. S3B). As for tumor-homing ability, MSC-T9C9, similar to MSC-Vec (Fig. 1), displayed robust tumor tropism and accumulation, and their intratumoral retention decreased progressively over time (Fig. S3C–F). These results confirmed that MSC-T9C9 was successfully engineered and possessed a potent immunostimulatory function without altering its proliferative capacity, tumorigenic potential, or tumor-homing capability.

MSC-T9C9 suppressed TNBC growth by immune activation
The therapeutic effects of MSC-T9C9 were evaluated in the TNBC mouse models. Mice in each group were intravenously administered every 3 days for a total of 3 times and finally analyzed at 18th day after first injection (Fig. 3A). Based on our previous correlative findings linking the T9C9 signature to favorable prognosis and immune infiltration in BRCA patients (Fig. 2A, B), we first utilized immunodeficient NOG mice to determine the immune-dependency of MSC-T9C9. In this model, tumor growth was comparable between the MSC-T9C9 and MSC-Vec groups (Fig. 3B), indicating that the anti-tumor effects of MSC-T9C9 require a functional immune system. We next employed two immunocompetent TNBC models (4T1 and EMT6 orthotopic syngeneic models) to investigate the mechanism of MSC-T9C9 in vivo. During tumor volume monitoring, MSC-T9C9 substantially inhibited tumor growth in both models, whereas MSC-Vec had no apparent effect compared to the PBS control (Fig. 3C, D). Consistently, final tumor images and weights confirmed significantly smaller tumors in the MSC-T9C9 group (Fig. 3C, D). Furthermore, H&E and TUNEL staining revealed extensive areas of apoptosis/necrosis in MSC-T9C9-treated tumors from both of 4T1 and EMT6 models (Fig. 3E, F).
TNFSF9, as a ligand of 4-1BB expressed on the surface of activated T and NK cells, could enhance their proliferation and cytokine secretion10,20–22. CXCL9, as one of the T helper 1 (Th1)-type chemokines, could recruit T and NK cells8,9,23. To further evaluate the intratumoral immune infiltration, tumors from each group were harvested at the experimental endpoint and analyzed by immunofluorescence staining. Consistent with the tumor suppression observed, the results revealed apparently higher infiltration of CD8+ T cells (cyan immunofluorescence) and NK cells (orange immunofluorescence) in the tumor tissues after MSC-T9C9 injection (Fig. 3G–J). Collectively, the results from both immunodeficient and immunocompetent mouse models underscore the pivotal role of immune activation in mediating the anti-tumor effects of MSC-T9C9.

MSC-T9C9 reshapes the tumor immune microenvironment
For a comprehensive quantitative assessment of the tumor immune microenvironment, we performed flow cytometry (the gating strategy is shown in Fig. S4). In the 4T1 mouse model, the percentage of tumor-infiltrated CD45+ cells was apparently increased after MSC-T9C9 injections (Figs. 4A and S5A). Furthermore, MSC-T9C9 treatment significantly expanded tumor-infiltrating lymphocyte subsets, with the percentages of CD4⁺ T cells, CD8⁺ T cells, and NK cells among CD45⁺ cells increasing by approximately 6-fold, 9.5-fold, and 4.1-fold, respectively, relative to the PBS control (Fig. 4B–D, G–I). In contrast, the MSC-Vec group exhibited no significant change in either CD8+ T or NK cell ratios compared to the PBS group. To further quantify this expansion, we calculated absolute cell counts, which confirmed a significant increase in CD4⁺ T, CD8⁺ T, and NK cells per standard number of live cells (Fig. S5B–D). These effects were specific to the tumor microenvironment, as no significant differences were observed in splenic lymphocytes among the groups (Fig. S5E–G). The immunophenotype of the tumor microenvironment was recapitulated in the EMT6 model (Fig. S6A–D, G–L).
The immunostimulatory potential of MSC-T9C9 was further investigated. The frequency of CD8+ T cells producing Gzmb was nearly 4.5-fold higher in the MSC-T9C9 group than in the PBS group (1.38% ± 0.87% versus 6.17% ± 1.14% g; p < 0.0001) (Fig. 4E, J). At the same time, MSC-T9C9 treatment also induced an obvious increase the proportion of Gzmb+ NK cells (Fig. 4F, K). Quantitative analysis confirmed a dramatic rise in the absolute numbers of Gzmb+CD8+ T and Gzmb+NK cells per tumor (Fig. 4L, M). Consistent results were obtained in the EMT6 mouse model (Fig. S6E, F, M–P).
Apart from increasing the expression of Gzmb, once activated, CD8+ T cells and NK cells can secrete many cytokines, including tumor necrosis factor (TNF)-α, interleukin (IL)-2, and interferon-γ (IFN-γ)24–26. The concentrations of TNF-α, IFN-γ and IL-2 in the tumors of MSC-T9C9 treated mice were apparently higher than those of the other two groups of mice by ELISA assay (Fig. 5N–P). No statistical difference was found between the PBS and MSC-Vec groups. In line with these findings, flow cytometric analysis revealed that MSC-T9C9 treatment enhanced the capacity of CD8⁺ T cells to produce TNF-α and IFN-γ (Fig. S5I, J).
Additionally, the effect of MSC-T9C9 on key immunosuppressive cell populations was evaluated, including regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSCs), and tumor-associated macrophages (TAMs). 4-1BB/4-1BBL axis has a complex role in Treg expansion and activity27–29. Following MSC-T9C9 treatment, the proportion of Foxp3+ cells among CD4+ T cells was significantly reduced, although their absolute number per tumor increased (Fig. S5K, L). suggesting that the marked expansion of effector T cells diluted the Treg population. For MDSCs, flow cytometric analysis revealed no significant differences among the groups (Fig. S5M). In contrast, MSC-T9C9 treatment strikingly expanded the macrophage population and skewed their polarization toward a pro-inflammatory M1 phenotype, as evidenced by an increased M1/M2 ratio (Fig. S5N, O). These findings align with our prior bioinformatics analysis (Fig. S2E), validating the ability of MSC-T9C9 to reverse immunosuppressive landscape of TNBC.
Collectively, these data demonstrate that MSC-T9C9 effectively remodels the TNBC tumor immune microenvironment into an anti-tumor state by recruiting and activating T and NK cells.

MSC-T9C9 drives an effector phenotype in CD8+ T cells
For a more detailed analysis of CD8+ T cell phenotypes, we analyzed the cells based on the surface expression of CD44 and CD62L. Although the frequency of effector (CD44+CD62L-) CD8+ T cells was lower in the MSC-T9C9 group, their absolute numbers were significantly increased (Figs. 5A, B and S5H). To investigate the underlying mechanism, we analyzed the expression of 4-1BB and PD-1 on the effector CD8+ T cell population. The mean fluorescence intensity (MFI) of 4-1BB was markedly elevated after MSC-T9C9 treatment (Fig. 5C), indicating robust antigen-driven activation and a demand for co-stimulation22,27. A concurrent increase in PD-1 expression was also detected (Fig. 5D), demonstrating the critical role of 4-1BB signaling in promoting the expansion and differentiation of exhausted T (Tex) cells30. Additionally, the proportions of both naïve (CD44-CD62L+) and central memory (CD44+CD62L+) CD8+ T cells were significantly increased (Fig. 5E, F), suggesting that CXCL9 secreted by MSC-T9C9 recruits diverse T cell subsets from the circulation. Overall, MSC-T9C9 administration potently augments anti-tumor immunity by driving T cell differentiation towards an effector state. The complementary actions of CXCL9 and TNFSF9 create a synergistic loop, leading to a broadly amplified and sustained immune response.

Biosafety of MSC-based immunotherapy
The safety evaluation of in vivo application is essential for MSC-based therapies. During the treatment period, all mice receiving MSC-T9C9 maintained normal activity and did not exhibit any signs of distress (e.g., piloerection, hunched posture, or lethargy). Most importantly, no significant body weight loss was observed in the treatment groups compared to the control groups. At the end of the experiment, blood samples and major organs were collected from each group of mice for further analysis. Given a major concern of cytokine storms in immunotherapy, serum levels of key cytokines (TNF-α, IL-2 and IFN-γ) were assessed in TNBC mouse models. Crucially, no significant differences were found between the MSC-T9C9-treated and control groups, indicating that the potent immune activation triggered by MSC-T9C9 is confined to the tumor microenvironment and does not provoke a harmful systemic inflammatory response (Figs. 5G and S7C). Based on the results of blood routine tests, no abnormalities were found in white blood cell (WBC), hemoglobin (HGB) and platelet (PLT) values, suggesting that MSC-T9C9 therapy did not have significant hematologic toxicity (Fig. S7A, D). Histological examination (H&E staining) of major organs showed no noticeable damage or metastatic lesions (Fig. S7B). Additionally, blood biochemical analysis showed that alanine aminotransferase (ALT), aspartate aminotransferase (AST) and creatinine (CREA) showed no obvious difference between the MSC-T9C9 group and the control groups across both the 4T1 and EMT6 mouse models, with all parameters within normal ranges, supporting the absence of hepatotoxicity or nephrotoxicity (Figs. 5H and S7D). In summary, MSC-T9C9 treatment demonstrates a favorable safety profile, with no evident adverse effects, which supports its therapeutic promise.

Combination treatment with MSC-T9C9 and anti-PD-1 antibody enhances the antitumor efficacy
As the primary immunotherapy strategy, the efficacy of ICB is limited by many factors. To further explore the clinical significance of T9C9 in immunotherapy, survival analysis was performed in the Kim cohort and the IMvigor210 cohort31,32. High T9C9 expression strongly correlated with improved survival in both datasets (Figs. 6A and S8A) and was associated with a higher gene set score in patients responsive to ICB (Figs. 6B and S8B). Based on the original clinical annotations, the Kim cohort was stratified into responders and non-responders to ICB therapy, allowing for further comparative analysis of immune infiltration between the two groups. The heatmap visually demonstrated responders exhibited markedly increased infiltration of effector T cells, whereas non-responders were characterized by immunosuppressive populations (Fig. S8C). The enrichment network revealed a more interconnected anti-tumor immune ecosystem in ICB-sensitive patients, aligns with the T cell infiltration pattern (Fig. S9). The results are consistent with previous reports33–35.
Since MSC-T9C9 treatment increased PD-1 expression on effector T cells (Fig. 5D), we hypothesized that it might have potential to synergize with ICB. We tested this by administering anti-PD-1 antibody one day after MSC-T9C9 treatment (Fig. 6C). In the 4T1 mouse model, this combination therapy achieved superior tumor inhibition and a significant survival benefit compared to all monotherapies (Fig. 6D, E). On day 18, tumor tissues from each group were collected and analyzed by flow cytometry. Compared to the control groups, both MSC-T9C9 and combination therapy significantly increased CD8⁺ T cell infiltration and Gzmb expression, with the highest proportion of Gzmb⁺CD8⁺ T cells observed in the combination group (Fig. 6F, G). A moderate increase Gzmb⁺CD8⁺ T cells were also observed in the aPD-1-alone group compared to the control.
These findings indicate that while ICB can activate CD8⁺ T cells, its monotherapy efficacy is limited by inadequate immune infiltration. MSC-T9C9 effectively primes the tumor immune microenvironment, and its combination with PD-1 antibodies unleashes a potent synergistic anti-tumor response.

Discussion
With the increasing development of sophisticated cell culture methods and cell engineering technologies, cell-based therapies have been widely used in clinical applications, including but not limited to chimeric antigen receptor (CAR) T cells, NK cells, dendritic cell (DC)-cytokine-induced killer (CIK) cells, and MSCs. Among them, MSCs exhibit tremendous potential in the treatment of various diseases, such as orthopedic injuries, cardiovascular diseases, and neurodegenerative diseases36–38. Due to the tumor-homing ability, MSCs are often used as carriers in anti-tumor therapies. In the treatment of TNBC, it has been reported that MSCs could be loaded with drugs (e.g., gemcitabine, paclitaxel and doxorubicin) and could be engineered to produce agents (e.g., tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) and interleukin-12)19. In the present study, the tumor-homing ability of MSCs was also verified in orthotopic TNBC mice. Even 10 days after injection, a small number of MSCs were still observed at the tumor site. This result is similar to our previous publications17,18. Consequently, MSCs can be used as a desirable tool for TNBC therapy.
As a prominent milestone in immunotherapy, immune checkpoint inhibitors (ICIs) are bringing light to TNBC patients. According to available clinical trial data, however, monotherapy has demonstrated limited clinical efficacy. The phase III KEYNOTE-119 trial (NCT02555657) demonstrated the overall response rate (ORR) was less than 10% in patients receiving single-agent pembrolizumab39. The main factor leading to these results is poor immune infiltration, especially tumor-infiltrating lymphocytes (TILs) which exert crucial functions in the immune response40,41. After a median follow-up of 10.6 years, the density of TILs was reported to be inversely associated with the risk of recurrence and death42. Based on the TCGA-BRCA cohort, analysis results revealed that CXCL9 was significantly positively correlated with multiple immune cell infiltration and long survival time. Hence, for increasing immune infiltration, the chemokine CXCL9 was chosen to establish the chemotactic gradient at the tumor site and recruit CD8+ T cells and NK cells in this study. Apparently, MSC-T9C9 recruited a large number of immune cells (especially CD8+ T cells and NK cells) to the TNBC tumor site with efficacy.
Since activated immune cells play a central role in anti-tumor immunity, it is more important whether the recruited immune cells are activated. In our past work, after comparing the efficacy of TNF superfamily ligands (including TNFSF4, TNFSF9, and TNFSF18) delivered by MSCs, we found that TNFSF9 and TNFSF18 exhibited relatively stronger lymphocyte-stimulating activities18. And it has been reported that TNFSF9, serving as an intracellular signal domain for the construction of chimeric antigen receptor (CAR) T cells, could stimulate lymphocyte activation43. A critical consequence stemming from MSC-T9C9 therapy was the acquisition of CD8+ T-cell and NK-cell effector function, as evidenced by an increase in Gzmb production and cytokine (TNF-α, IFN-γ and IL-2) secretion. Our findings are independently supported by and consistent with the results of a recent study30. Pichler et al. found the higher expression of 4-1BB on Tex cells by analyzing the public datasets of breast cancer. And they identified 4-1BB signaling regulates Tex cell proliferation and differentiation. T cell exhaustion is often considered as a mechanism of immune escape, however, Tex cells are not inert and have residual functions. This population (4-1BB+) exhibits a heightened state of activation and possesses potent effector capabilities. Hence, on the basis that CXCL9 increases multiple immune cell infiltration, Tex cell expansion by TNFSF9 (as an anti-4-1BB agonist) is also beneficial for antitumor therapy.
Due to the low tumor mutation burden and lack of immune cell infiltration, TNBC patients have a better survival benefit from ICI therapy when combined with chemotherapy39,44,45. In this study, the findings show a strong association between the T9C9 gene set score and immune cell signatures. And the patients who respond to ICI therapy have a higher gene set score, suggesting that T9C9 may be a predictive biomarker for immunotherapy. Previous studies have shown that anti-4-1BB agonists and CXCL9/10-DC therapy could enhance the efficacy of ICI in tumor mouse models39,46. Correspondingly, MSC-T9C9 plus anti-PD-1 antibody combination therapy achieves better tumor response and more survival benefit compared to either monotherapy, and deeply activates tumor-infiltrating CD8+ T cells into a more effector state. Mechanistically, MSC-T9C9 remodels the tumor microenvironment by enhancing immune infiltration and promoting the differentiation of T cells toward an effector state. The resulting effector T cell population exhibited high PD-1 expression, thereby unleashing a potent anti-tumor response upon PD-1 pathway blockade. Thus, MSC-T9C9 may have the potential to overcome immunotherapy resistance in breast cancer patients.
In addition to efficacy, safety is another crucial consideration for clinical translation. For example, urelumab (an anti-4-1BB agonist antibody) was initially effective in patients with melanoma and lymphoma, but some patients still developed severe hepatitis47. Clearly, the MSC-targeted delivery platform effectively avoided the toxicities of systemic application of the chemokine CXCL9 or TNFSF9. No evident adverse effects were observed in any of the treated mice. And based on existing clinical trials, MSC-based therapy is relatively safe48,49. Furthermore, our study prompts consideration of the dynamic interplay between MSC-T9C9 infiltration levels and immune reprogramming. While our data robustly demonstrate that MSC-T9C9 successfully reshapes the TME and enhances anti-tumor immunity, the quantitative relationship between the level of MSC infiltration and the intensity of the immune response remains to be fully defined. Future studies employing advanced lineage-tracing models, in vivo imaging, or high-dimensional spatial omics will be crucial to definitively map MSC trafficking and interaction networks. Establishing such a quantitative link will not only solidify the prognostic value of MSCs but also be pivotal for refining dosing and maximizing clinical benefit.
Given the paucity of research in the field of cancer therapy, MSC-based therapies still face several challenges, including the need to optimize modification techniques, administration regimens, and standardization processes. While the current study demonstrates the potent efficacy of MSC-T9C9, future studies comparing the therapeutic potential of MSCs engineered with different cytokine combinations would provide valuable insights into optimizing treatment strategies. Such comparative analyses would help identify the most effective cytokine profiles while minimizing potential side effects, ultimately contributing to the development of standardized treatment protocols for TNBC. To address these concerns, our future work will focus on key translational aspects, including rigorous evaluation of MSC-T9C9 in humanized or PDX models and comprehensive GLP-compliant safety, biodistribution, and toxicology studies as essential prerequisites for clinical trials. Further systematic investigation remains warranted to establish standardized treatment protocols for TNBC using this promising therapeutic approach.
In summary, we have engineered hUC-MSCs to develop a novel immunotherapy for TNBC. The modified MSCs had effective expression of the chemokine CXCL9 and the costimulatory ligand TNFSF9, which broke through the lymphocyte exclusion barrier and reprogrammed the immune microenvironment in TNBC mouse models without side effects (Fig. 7). And the combination of MSC-T9C9 and ICB had a better curative effect. As a result, MSC-T9C9 therapy has high potential to be applied in immunotherapy for TNBC patients clinically.

Methods

Cells and animals
Murine breast cancer 4T1 and EMT6 cells were purchased from SIBS (Shanghai, China) and cultured in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin (P/S) under a condition of 5% CO2 at 37 °C. Female Balb/c mice (7 weeks old, 20 ± 2 g) were obtained from SLAC (Shanghai, China). Female NOG mice (7 weeks old, 20 ± 2 g) were obtained from Vital River (Beijing, China). At the experimental endpoint, mice were euthanized by carbon dioxide (CO₂) inhalation, followed by cervical dislocation to ensure death. All animal procedures were approved by the Guide for the Care and Use of Laboratory Animals of Tongji University (ethical approval number: TJBB00723103).

Kaplan–Meier plotter analysis and immune cell infiltration analysis
Analyses of the association of TNFSF9/CXCL9 expression with overall survival (OS) or progression-free survival (PFS) were conducted using R software (version 4.4.1) and the Biomarker Exploration for Solid Tumors platform (https://rookieutopia.com/)50. The analysis included the following publicly available cohorts: TCGA-BRCA, Kim et al. (n = 27; for ICB response stratification, 8 responders vs. 19 non-responders, based on original clinical annotations)31, and IMvigor210 (n = 348; for ICB response stratification, 68 responders vs. 230 non-responders, based on original clinical annotations)32. Gene set scores for TNFSF9 and CXCL9 were calculated using the single-sample Gene Set Enrichment Analysis (ssGSEA) algorithm. Immune cell infiltration was estimated by the ESTIMATE and Xcell algorithms. Pearson correlation coefficients were utilized to assess the association between TNFSF9/CXCL9 expression and abundance of immune cells.

Isolation, culture, and identification of MSCs from human umbilical cord
The human umbilical cord mesenchymal stem cells (hUC-MSCs) were isolated from the human umbilical cord and cultured in TransStemTM Serum-Free, Xeno-Free Human Mesenchymal Stromal Cell Medium (MM101-01, TransGen Biotech). And then hUC-MSCs were verified by flow cytometry according to the expression of mesenchymal surface markers. The specific experimental procedures were performed as we have described previously18. Collection of the human umbilical cords was approved by the Renji Hospital Ethics Committee (No. KY2021-027).

In vivo biodistribution of hUC-MSCs
To establish the breast cancer mouse model, 5 × 105 4T1 cells were injected subcutaneously into the right mammary fat pad of 7-week-old female Balb/c mice. When the volume of tumors grew to 150–200 mm3, 1 × 106 GFP-labeled MSCs/MSC-T9C9 in 100 μL PBS were intravenously injected into tumor-bearing mice. Mice were sacrificed after 3, 7 and 10 days, respectively. And lung, liver, spleen, kidney and tumor tissue of each mouse were harvested and were fixed with 4.0% paraformaldehyde. GFP immunofluorescence staining was performed by using GFP antibody (300943, Zen-Bioscience) and was imaged by fluorescence microscopy (ZEISS Axio Vert A1). To quantitatively assess the homing of GFP-labeled MSCs to tumors, single-cell suspensions were prepared from harvested tumors at each time point. Following digestion and filtration through a 70-μm strainer, red blood cells were lysed using red blood cells lysis solution (ZYFB006-0500, ZUNYAN, NanJing, China), and cell viability was stained using the Zombie NIRTM Fixable Viability Kit (423106, BioLegend). The proportion of GFP-positive cells within the live cell population was then determined by flow cytometry (CytoFLEX LX, USA).

Lentivirus transfection of hUC-MSCs
The lentiviral transfer plasmid was based on the third-generation, self-inactivating pLenti backbone. It featured a bicistronic expression cassette under the control of the CMV promoter, which co-expresses TNFSF9 and CXCL9 linked by a P2A peptide. A counterpart vector expressing only the fluorescent reporter gene was used as a blank control. Lentivirus production and titration were performed by OBiO Technology (Shanghai). Lentiviral transduction of MSCs was conducted at a multiplicity of infection (MOI) of 60 with the presence of 8 μg/ml polybrene (Sigma Aldrich). Transduction efficiency of >90% was verified by GFP expression under a fluorescence microscope, which was co-expressed by the lentivector. Successful expression of TNFSF9 and CXCL9 in MSCs was validated by flow cytometry analysis and ELISA, respectively.

In vitro cell proliferation assay
The cell counting kit-8 (CCK-8) assay (KGA9305, KeyGEN BioTECH) was used to assess cell proliferation capacity of MSC-Vec and MSC-T9C9 according to the manufacturer’s protocols. Briefly, MSCs were seeded in 96-well plates, CCK-8 solution was added at each time point. Followed by incubation for 2 h, the absorbance of each well was measured at 450 nm.

In vitro coculture of hUC-MSCs and lymphocytes
Followed red blood cell lysis, lymphocytes were isolated from spleens of Balb/c mice and cultured in RPMI 1640 containing 10% FBS, 1% P/S, and 20 ng/ml recombinant mouse interleukin (IL)-2 (CK24, Novoprotein, Shanghai, China) overnight. The MSC-T9C9 or MSC-Vec (GFP-labeled MSC) cells were then co-cultured with the lymphocytes at a 1:1 ratio in 12-well plates. After 48 h of co-culture, all cells were harvested and analyzed by flow cytometry to evaluate the expression of CD69 and Granzyme B on lymphocyte subpopulations.

In vivo anti-tumor therapy
7 days post-inoculation with 4T1 or EMT6 cells, all tumor-bearing mice were pooled and randomized into groups to minimize bias. For the 4T1 model, two independent experiments were conducted, n = 6 in each experiment. The groups then received intravenous (i.v.) injection of PBS, 1 × 106 MSC-Vec (GFP-labeled MSCs) or 1 × 106 MSC-T9C9 cells once 3 days for three consecutive treatments. The body weight and tumor size were monitored every 3 days. The tumor volume was calculated using the formula: tumor volume = (length × width2)/2. After 18 days after first injection, the mice were sacrificed and the tumors were collected for further analysis. To evaluate the effectiveness of combination therapy, 4T1 tumor bearing mice were randomly divided into four groups. Anti-PD-1 antibody (10 mg/kg/dose, clone RMP1-14, Bio X Cell) was administered intraperitoneally (i.p.) the day after the treatment of MSC-T9C9 for a total of three injections. Survival was defined as the time from treatment initiation to the endpoint, which was reached when tumor volume exceeded 1500 mm3 or if mice exhibited signs of severe morbidity.

Flow cytometry analysis
At the endpoint, mice from each group were randomly selected for sacrifice to obtain a representative sample for subsequent analysis. Throughout the experiment and data analysis, investigators were blinded to the group assignments. Then, tumor and splenic tissues were collected for flow cytometry analysis. Samples were digested into single-cell suspensions and stained with viability dye to exclude dead cells. Followed by FcR blockade, the cells were incubated with fluorescently labeled surface antibodies for 30 min at 4 °C. For intracellular staining, resuspended cells were fixed and permeabilized using True-NuclearTM Transcription Factor Buffer Set (424401, BioLegend). All samples were analyzed by flow cytometry (CytoFLEX LX, USA).

H&E analysis and TUNEL
After 18th day’s treatments, all mice in each group were euthanized. Subsequently, the tumors and major organs (including heart, liver, spleen, lung and kidney) were collected and fixed in 4.0% paraformaldehyde solutions. For hematoxylin and eosin (H&E) analysis, the tissue sections were stained using Hematoxylin and Eosin Staining Kit (G1121, Solarbio). Moreover, the tumor sections were stained via TUNEL assay kit (KGA1400-100, KeyGEN BioTECH) based on the manufacturer’s protocols.

Immunofluorescence analysis of tumor immune microenvironment
To evaluate the tumor immune microenvironment, the tumor tissues were fixed and cut into 5 μm sections. Following exposure to anti-CD8 and anti-CD335 antibodies, tissue sections were incubated with fluoroprobe-labeled secondary antibodies and counterstained with DAPI (KGA1808-50, KeyGEN BioTECH). Finally, the images were obtained via fluorescence microscopy (ZEISS Axio Vert A1). Immunofluorescence images were analyzed using ImageJ (Version 1.53k).

Analysis of cytokines by ELISA
To quantify cytokine levels, serum of mice was collected and tumor tissues were homogenized. Then, tumor necrosis factor (TNF) -α, IL-2 and interferon-γ (IFN-γ) were analyzed by ELISA kits (MM-0132M1, MM-0701M1, MM-45169M1, Jiangsu Meimian Industrial Co., Ltd) according to manufacturer’s instruction.

Blood routine test and blood biochemical analysis
At the end of the experiment, whole blood samples were collected by removing the eyes of mice. Whole blood samples were performed blood routine tests. Followed centrifugation, blood biochemical analysis was done (including hepatic and renal functions).

Statistical analysis
Two-tailed Student t test or two-way ANOVA was carried out for statistical analyses and the data were presented as mean ± SD. Statistical analyses were conducted by GraphPad Prism version 8.0. Statistically significant differences were expressed as *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001.

Supplementary information