Trojan bacteria cross blood-brain barrier for glioblastoma photothermal immunotherapy

Bacteria can bypass the blood-brain barrier (BBB) transcellularly, paracellularly and/or in infected phagocytes, suggesting the possibility of employment of bacteria for combating central nervous system (CNS) diseases. However, the bacteria-based drug delivery vehicle crossing the BBB is still vacant up to present. Herein, we develop an innovative bacteria-based drug delivery system (dubbed Trojan bacteria) for glioblastoma (GBM) photothermal immunotherapy. Typically, Trojan bacteria are made of therapeutics internalized into bacteria (e.g., attenuated Salmonella typhimurium, Escherichia coli). The therapeutics are composed of glucose polymer (GP) (e.g., poly[4-O-(α-D-glucopyranosyl)-D-glucopyranose])-conjugated and indocyanine green (ICG)-loaded silicon nanoparticles (GP-ICG-SiNPs). The GP-ICG-SiNPs can be selectively and robustly internalized into the bacterial intracellular volume through the bacteria-specic ATP-binding cassette (ABC) transporter. In an orthotopic GBM mouse model, we demonstrate that the intravenously injected Trojan bacteria could take therapeutics together not only to bypass the BBB, but also to target and penetrate GBM tissues. Under 808 nm-laser irradiation, the photothermal effects (PTT) produced by ICG allow the destruction of Trojan bacterial cells and the adjacent tumour cells. Furthermore, the bacterial debris as well as the tumour-associated antigens would promote antitumor immune responses that prolong the survival of GBM-bearing mice. Moreover, we demonstrate the residual Trojan bacteria could be effectively eliminated from the body due to the distinct photothermal effects. We anticipate the proposed Trojan bacteria system would catalyze innovative therapies for various CNS diseases. and common software of ImageJ (NIH Image; http;//rsbweb.nih.gov/ij/). Error bars represent the standard deviation obtained from three independent measurements. All the statistical analyses were performed using the Origin and GraphPad Prism 7 software. The statistical signicance of differences was determined by a one-way ANOVA analysis. p < 0.05 (*), p < 0.01 (**) and p < 0.001 (***) were used to indicate statistical difference.


Introduction
GP and the photothermal agent of ICG. As schematically illustrated in the synthetic route of GP-ICG-SiNPs ( Supplementary Fig. 1), the GP molecules are rstly conjugated to the SiNPs surface based on the Schiff base reaction between the aldehyde groups of GP and the amino SiNPs 41 . After that, the ICG molecules are loaded on GP-SiNPs through electrostatic adsorption. The constructed nanoagents can selectively enter intracellular volume of bacteria (e.g., VNP, EC) to form Trojan bacteria system through the bacteriaspeci c ABC transporter pathway [42][43][44][45][46] . Fig. 1b schematically illustrates the Trojan bacteria serving as a carrier to deliver therapeutics to cross the BBB to achieve light-triggered combination therapy of GBM. In a typical experiment, after intravenous injection of Trojan bacteria, they could take the internalized therapeutics together to cross the BBB, and eventually target and penetrate GBM tissues. Under 808-nm irradiation, the ICG molecules loaded on SiNPs could convert light energy into the su cient heat to lysis host bacterial cells and destruct tumour cells. As a result, a pool of tumour-associated antigens and bacterial residues could together promote antitumor immune responses [47][48][49][50][51] . Typically, immature dendritic cells (iDCs) recognize tumour-associated antigens and bacterial residues through pattern recognition receptors and become mature DCs (mDCs). Meanwhile, mDCs migrate to lymphoid organs to express a large number of cytokines and stable major histocompatibility complex to activate T cells (TCLs). As a consequence, the TCLs express cytokines like TNF-α and IFN-γ, leading to a remarkable immune therapeutic outcome.
Characterization of Trojan bacteria system. A series of experiments were carried out to demonstrate the successful preparation of GP-ICG-SiNPs including transmission electron microscopy (TEM), dynamic light scattering (DLS), UV-vis absorbance (UV), photoluminescence (PL) and ow cytometry. As shown in the TEM image in Supplementary Fig. 2, GP-ICG-SiNPs appear as spherical particles with a narrow size distribution of ~4.1 nm, which is slightly larger than that of bare SiNPs (e.g., ~2.7 nm). The DLS spectra in Supplementary Fig. 3 reveal the hydrodynamic diameter of GP-ICG-SiNPs is ~6.5 nm, also larger than the hydrodynamic diameter of bare SiNPs (~3.6 nm). As shown in Supplementary Fig. 4a, three distinct peaks located at 320 nm (assigned to SiNPs), 731 nm, and 790 nm (assigned to ICG) exist in the absorption spectrum of GP-ICG-SiNPs, con rming the successful loading of ICG molecules. As further revealed in Supplementary Fig. 4b, upon the treatment of phenol-sulfuric acid, the absorption spectrum of GP-ICG-SiNPs displays a new peak at 490 nm since the linked GP is hydrolyzed into furfural derivative, followed by the formation of furfural resin with phenol 52 . The new appearing peak con rms the successful conjugation of GP molecules. The amounts of linked GP and loaded ICG onto SiNPs can be quanti ed based on the corresponding calibration absorption curves ( Supplementary Fig. 5), respectively. Supplementary Figs. 6a and 6b give the PL spectra of GP-ICG-SiNPs under the excitation of 405 nm or 780 nm, respectively. Typically, the two characteristic emission peaks located at 485 nm and 810 nm are respectively corresponding to SiNPs and ICG. The photothermal curves in Supplementary Fig. 7 suggest that the temperature of GP-ICG-SiNPs solutions can be enhanced by 32°C during 300-sec 808-nm laser exposure when the loading concentration of ICG is or more than 150 µg/mL. These results together demonstrate the successful modi cation of ICG molecules with SiNPs.
To test the generality of the proposed Trojan bacteria strategy, two representative bacteria of VNP and EC were selected for the following experiments. The VNP or EC were incubated with the synthesized GP-ICG-SiNPs at 37 o C for 2 h, and then washed with phosphate buffered saline (PBS) buffer for several times.
As revealed in the scanning electron microscope (SEM) images in Fig. 2a, the surface and the morphology of Trojan bacteria is similar to that of pristine bacteria. As further con rmed by elemental mapping in high-angle annular dark eld-scanning transmission election microscope (HAADF-STEM) images (Fig. 2b), the silicon element exists only in Trojan EC or Trojan VNP rather than in pure EC or VNP. SEM and TEM images together prove that the prepared therapeutics indeed enter bacteria rather than nonspeci cally absorb on the bacterial surface. Moreover, Trojan bacteria have the same growth curve compared with the untreated bacteria ( Supplementary Fig. 8a), and their survival rate could maintain as high as 90% or more ( Supplementary Fig. 8b), indicating the growth and activity of host bacteria would not be greatly in uenced by the internalized therapeutics.
As shown in the confocal laser scanning microscope (CLSM) images in Fig. 2c, the green uorescence from SiNPs ( rst column, λ ex = 405 nm, λ em = 500-550 nm) and the red uorescence from ICG (second row, λ ex = 633 nm, λ em = 700-800 nm) could be simultaneously observed in the Trojan bacteria. On the contrary, no uorescent signals are detected when EC or VNP are treated with ICG-SiNPs under the same conditions (Fig. 2d). Quantitatively, the uptake e ciency of nanoagents by EC or VNP cells was further determined by ow cytometry. As revealed in Fig. 2e, the uptake e ciency of GP-ICG-SiNPs by EC or VNP after 2 h of incubation is gradually rising when increasing the concentration of GP-ICG-SiNPs. Typically, when the GP-ICG-SiNPs is 10 mg/mL, the uptake e ciency is 56.8% for EC and 50.5% for VNP. If further enhancing the concentration to 15 mg/mL, the uptake e ciency does not improve signi cantly, e.g., 62.1% for EC and 55.7% for VNP. As such, 10 mg/mL GP-ICG-SiNPs is employed in the following experiments.
We further performed inhibition assay and competition assay to con rm whether GP-ICG-SiNPs entering bacteria was through the ABC transporter pathway. For the inhibition assay, no uorescent signals are measured when EC or VNP are co-incubated with sodium azide (NaN 3 ) and GP-ICG-SiNPs, in which NaN 3 serves as the inhibitor of ABC transporter ( Supplementary Fig. 9) 53 . For the competition assay, EC or VNP were rst incubated with 0, 5 and 20 mg/mL of GP and then incubated with GP-ICG-SiNPs. As indicated in Supplementary Fig. 10, the uorescence of bacteria becomes gradually weaken with the increase of GP concentrations. To testify the selectivity of GP-ICG-SiNPs for bacteria over mammalian cells, the mixture of human blood and bacteria were incubated with GP-ICG-SiNPs for 2 h. As shown in Fig. 2d and Supplementary Fig. 11, uorescence signals are only observed in EC or VNP rather than in red blood cells (RBCs). These results together demonstrate that GP-ICG-SiNPs can be internalized into bacteria to form the Trojan system via the bacteria-speci c ABC transporter pathway.
Trojan bacteria against tumour in vitro. We next studied the photothermal ability of Trojan bacteria. Under 808-nm laser irradiation, the constructed Trojan bacteria (Trojan EC or Trojan VNP) could heat up to 55℃ within 400 s, slightly lower than the temperature achieved by the equivalent free GP-ICG-SiNPs (Fig. 3a).
Accordingly, the bacterial cell viability gradually decreases as the temperature increases from 48 to 55 o C ( Fig. 3b). In particular, the bacterial cell viability is only 37% at 52 o C. Meanwhile, the overall morphology of the Trojan bacteria begins to rupture when the temperature rises to 48 o C, as shown in the SEM image in Fig. 3c. These results manifest that the constructed Trojan bacteria feature good photothermal activity.
Next, we investigated whether the photothermal effects produced by the Trojan bacteria could destroy glioblastoma G422 cells. In live-dead cell staining, the red uorescence is only observed in most of Trojan bacteria-treated G422 cells (Fig. 3d). In MTT assay, under the irradiation of 808 nm for 5 min, the cell viability of G422 cells dramatically decreases to less than ~20% when they are incubated with Trojan bacteria or equivalent free GP-ICG-SiNPs for 6 h, much lower than the 100% of other control groups (p <0.001) (Fig. 3e).
In the constructed Trojan bacteria system, tumour-associated antigens and bacterial residues produced by PTT might trigger an effective immune response. To testify this hypothesis, we designed a transwell system to study this effect in vitro. As shown in the Fig. 3f, G422 cells with different treatments are placed in the upper chamber of the transwell system, and DCs are seeded in the lower chamber. After that, the maturation of DCs are evaluated by ow cytometry. The up-regulation of co-stimulatory factors CD80 and CD86 as the typical markers on the surface of DCs indicate the level of DCs maturation 54,55 . As revealed in Figs. 3g-3h, compared with the control groups, the level of DCs maturation can be signi cantly improved when G422 cells were treated with Trojan bacteria system under 808-nm irradiation (e.g., 60.6% DC maturation in Trojan EC group, 64.8% DC maturation in Trojan VNP group). These results indicate that the tumour-associated antigens and bacterial residues produced by PTT could trigger effective DC maturation.
In vivo behaviors of bacteria in mice. Before cancer treatment, we have systematically studied the behavior of bacteria in mice after tail vein injection. We rst determined the safe dose of bacteria injected into the mice. The body weights were measured for healthy mice injected with EC or VNP at ~1×10 6 , 1×10 7 and ~1×10 8 CFU (Fig. 4a). Specially, at the high dose of ~1×10 8 CFU EC or VNP, the mouse body weights rapidly dropped, and one or two of ve mice died respectively in the EC or VNP-treated groups, implying the severe toxicity of bacteria at such a high dose. Under lower doses, i.e., ~1×10 6 and ~1×10 7 CFU per mouse, the mouse body weights did not change signi cantly, and no mice died in the corresponding group. Therefore, EC or VNP at a moderate dose (~1×10 7 CFU per mouse) with a tolerable side effect were employed in the following experiments.
To reveal the in situ and real-time location of bacteria in vivo, we transformed the pRSETB-mCherry plasmids into EC (mCherry@EC) or VNP (mCherry@VNP) to express red uorescence protein of mCherry ( Supplementary Fig. 12). Afterwards, the female health Balb/c mice were injected with these engineered VNP or EC through the tail vein at the dose of ~1×10 7 CFU per mouse. The mice were then sacri ced at 12, 24, 72, 120 and 360 hours after intravenous injection to obtain their main organs (e.g., brain, heart, liver, spleen, lung and kidney), followed by the detection of red uorescence of mCherry by an in vivo optical imaging system (IVIS Lumina III). As shown in Figs. 4b-4d, the uorescence signal mainly exists in the liver, and is gradually weakened over time, which is basically undetectable at 15 days. We then homogenized the extracted organs, serially diluted (10-fold), and plated them on solid LB agar plates. In consistent with the results of ex vivo images, bacteria mainly accumulated in the liver and were quickly cleared from all extracted organs. The total elimination was basically achieved at 15 days (Figs. 4e-4g).
To further ensure the safety of bacterial injection, routine blood tests including complete blood and serum biochemical analysis were performed on the tested dose, i.e., ~1x10 7 CFU per mouse (Figs. 4h-4q). Compared with untreated healthy mice, all serum biochemical parameters and most blood routine data were within the normal range at the rst day of bacterial injection, except for a decrease in white blood cell (WBC) and platelet count. At the fth day of bacterial injection, the WBC count and platelet count of the bacteria-treated mice returned to normal ranges, indicating that the acute in ammation caused by EC and VNP infection was mild and tolerated by the mice and did not develop chronic toxicity within 5 days.
Trojan bacteria crossing BBB, targeting and penetrating glioblastoma. Next, we performed a series of experiments to demonstrate the constructed Trojan bacteria could cross the BBB. We rst built an in vitro human brain microvascular endothelial cell (HBMEC) model to investigate whether the Trojan bacteria could cross the BBB (Fig. 5a) 56, 57 . The construction of HBMEC model was evaluated by the measurement of transepithelial electrical resistance (TEER) ( Supplementary Fig. 13a). Experimentally, Trojan bacteria were inoculated with HBMEC cells in the upper chamber (apical chamber) of transwell at a dose of ~8×10 4 CFU/well, followed by collecting 10 µL of culture medium from the lower chamber at 1h, 2h, 3h and 4h, respectively. The penetration rate of Trojan bacteria crossing the BBB was determined by counting the number of colonies on the LB solid medium the next day ( Supplementary Fig. 13b). As revealed in Fig. 5b, the penetration rate of EC and VNP increases gradually with time, climbing to 49.7% and 60% at 4h, respectively. These results demonstrate the Trojan bacteria can cross the BBB in the in vitro model. Afterwards, the female Balb/c mice with in situ GBM were injected with these mCherry@VNP or mCherry@EC through the tail vein at the dose of ~1×10 7 CFU per mouse. As revealed in ex vivo images in Figs. 5c-5e, the red uorescence of bacteria could be found in GBM after 12 h of bacteria injection, gradually increasing following the time. By contrast, the uorescence from bacteria in other organs like liver decreases exponentially following the time. Such difference might be resulted from the selective proliferation of bacteria in the hypoxic, immunosuppressive, and biochemically unique glioblastomas microenvironment 33,58−60 . And through the results of the plates, the amount of EC or VNP in brain counted in each plate was much higher than that in other organs. Speci cally, the CFU of bacteria in brain reached its peak up to ~0.51×10 6 CFU/g for EC and ~0.74×10 6 CFU/g for VNP at the 120 h post-injection with bacteria (Figs. 5f-5h).
With an aim to study the intratumoural transport of constructed system, we rst constructed an ex vivo model of three dimensional cultured multicellular spheroids (MCSs), which were made of U87MG cells when the diameter of MCSs arrived at ~750 µm. PBS, GP-ICG-SiNPs (8 mg/kg ICG), mCherry@VNP (~1.0x10 7 CFU) and Trojan bacteria (GP-ICG-SiNPs (8 mg/ml ICG) internalized into ~1.0x10 7 CFU mCherry@VNP) were co-incubated with U87MG MCSs for 12h, respectively. Fluorescence signals at different depths of MCSs were collected by CLSM. The three-dimensional confocal images show that the MCSs treated by Trojan bacteria display distinct green (from SiNPs) and red (from mCherry@VNP) uorescence signals at both the edge and internal space, with a penetration depth of ~260 µm (Fig. 5i). In addition, the distribution of Trojan bacteria in MCSs is basically consistent with that of pure mCherry@VNP, but much deeper than that of free GP-ICG-SiNPs. These results indicate that the constructed Trojan bacteria could penetrate deep GBM tissues in vitro.
To verify the constructed Trojan bacteria could penetrate the deep GBM tissues in vivo, the female Balb/c mice with in situ GBM were intravenously injected with PBS, GP-ICG-SiNPs, EC, VNP, Trojan EC or Trojan VNP, respectively. Afterwards, the excision and section of deep GBM tissue was performed at 12 h post injection, followed by hypoxia inducible factor-α (HIF-α) analysis and the bacterial 16S RNA uorescence in situ hybridization (FISH) analysis. The expression level of HIF-α indicates the hypoxia situation in GBM tissues and the expression of 16S RNA indicates the location of injected bacteria in GBM tissues. As revealed in Fig. 5j, we can observe distinct green uorescence signal of HIF-α in all groups, suggesting the high hypoxia level of deep GBM tissues. In addition, the distinct red uorescence signals of 16S RNA are only found in EC, VNP, Trojan EC and Trojan VNP-treated groups, suggesting Trojan bacteria with hypoxiatargeting ability indeed penetrate GBM tissues in vivo.
The uorescence signals of ICG can be employed for monitoring the dynamic distribution of GP-ICG-SiNPs as well as Trojan bacteria in the body. Supplementary Fig. 14 shows the uorescence signals of ICG in the brain peaked at ~12-hour post-injection of Trojan bacteria. Therefore, at the 12-hour postinjection of Trojan bacteria, the brains of those mice were suffered by an 808 nm irradiation (1.2W/cm 2 , 5 min). As revealed in the photothermal images record by an IR camera (Fig. 6b), the rapid GBM temperature rising only occurs in Trojan EC or Trojan VNP-treated groups. In particular, the GBM-surface temperature can increase to 50.7 o C in Trojan EC group after 5-min irradiation, and 51.7 o C in Trojan VNP group after 5 min irradiation (Fig. 6c). By contrast, signi cant heating is not observed in other control groups under the same conditions. Next, at 5-day and 10-day post-injection, GP-ICG-SiNPs, Trojan EC, Trojan VNP at the same dose were intravenously injected into the mice again. Analogously, at 5-day and 12-hour post-injection or 10-day and 12-hour post-injection, the GBM sites of these mice were irradiated by 808 nm laser and the temperature of GBM sites was stabilized at 50 o C for 5 min by adjusting the 808 nm laser power. Afterwards, bioluminescence imaging was applied to visualize the anti-tumor effect every four days. As displayed in Figs. 6d-6e, the bioluminescence signals of luc-G422 cells in Trojan EC or Trojan VNP groups are much weaker than that of the other treatment groups. The quantitative analysis shows that Trojan bacteria system has obvious inhibitory effect on tumour growth, and the corresponding inhibition rates are 66.25% of Trojan EC and 70.11% of Trojan VNP, respectively, which are much better than 16.14% of EC, 11.88% of VNP and 8.01% of GP-ICG-SiNPs (Fig. 6f). Also, survival analysis shows that survival time of mice has been signi cantly prolonged in Trojan bacteria group compared with other control groups (Fig. 6g). To further evaluate the antitumor effects of Trojan bacteria system, H&E staining of GBM tissues were performed. As shown in Fig. 6h, the most apparent tumour cell destruction, tissue necrosis and nuclear pyknosis are found in Trojan bacteria-treated group. Taken together, these therapeutic data demonstrated the adaptable anticancer ability of Trojan bacteria in vivo.
As previously reported, the lysates of tumour and bacterial cells can function as tumour vaccines and initiate tumour-speci c immune responses 33,61 . Hence, the in vivo DCs maturation triggered by Trojan bacteria system was evaluated. On the third day of the last treatment, we harvested the carotid lymph nodes and homogenized them into single cell suspensions. After staining with uorophore labeled anti-CD11c, anti-CD86 and anti-CD80 antibodies, the cell suspensions were collected for ow cytometry analysis. As shown in Figs. 6i and 6j, compared with control groups, the highest level of DC maturation was observed in the groups of Trojan bacteria+laser (e.g., 39.4% DC maturation for Trojan EC, 42.7% DC maturation for Trojan VNP), which could be attributed to the release of tumor-associated antigens and bacterial residues upon photothermal effect. On the other aspect, splenocytes from each group were costained with anti-CD3, anti-CD4 and anti-CD8a antibodies, and the percentage of CD3 + CD8a + T cells was measured by ow cytometry. As shown in Supplementary Fig. 15 and Fig. 6k, no signi cant difference on CD8a + T cell response was observed between GP-ICG-SiNPs group (21.2%) and PBS control (19.8%). On the contrary, the percentage of CD8a + T cells in splenocytes after the combined photothermal ablation with Trojan EC treatment (30.6%) and Trojan VNP treatment (34.1%) were much higher than that of EC (27.3%) and VNP (26.3%) with laser irradiation. At the same time, we assessed the percentage of CD3 + CD8a + T cells in the tumours by ow cytometry. As revealed in Supplementary Fig. 16 and Fig. 6l, no signi cant difference on CD8a + T cell response was observed between GP-ICG-SiNPs group (8.31%) and PBS control (7.06%). On the contrary, the percentage of CD8a + T cells in splenocytes after the combined photothermal ablation with Trojan EC treatment (19.1%) and Trojan VNP treatment (22.4%) were much higher than that of EC (10.8%) and VNP (11.4%) with laser irradiation. In addition, we also evaluated the cytokine levels of TNF-α and IFN-γ in serum samples of mice after various treatments. In the cytokine assay of serum samples, levels of both cytokines (IFN-γ and TNF-α) were higher in mice treated with Trojan bacteria than those in mice treated with pure bacteria and nanoagents (Figs. 6m and 6n). The increasing TNF-α and IFN-γ levels combined with the elevation of the percentage of CD8a + T cells further con rm the boosted immunity in the mice injected with Trojan bacteria under laser irradiation.
The elimination of residual bacteria after photothermal immunotherapy. Next, we examined whether the photothermal immunotherapy induced by Trojan bacteria could facilitate the elimination of bacteria from the GBM-bearing mice. In detail, the GBM-bearing mice were intravenously injected with (G1) PBS, (G2) 1×10 7 CFU mCherry@EC, (G3) ~1×10 7 CFU mCherry@VNP, (G4) Trojan mCherry@EC (e.g., GP-ICG-SiNPs (8 mg/kg ICG) internalized into ~1×10 7 CFU mCherry@EC) or (G5) Trojan mCherry@VNP (e.g., GP-ICG-SiNPs (8 mg/kg ICG) internalized into 1×10 7 CFU mCherry@VNP), respectively. At the 12-hour postinjection, the brains of those mice were suffered by an 808 nm irradiation (1.2W/cm 2 , 5 min), followed by ex vivo imaging of the main organs at 5-day post-injection. As shown in Figs. 7a and 7b, relatively strong red uorescence signals of mCherry could be detected in GBMs and livers in G2-G5 groups. Of note, the uorescence intensity in the Trojan bacteria treatment groups (G4 and G5) is signi cantly lower than that in pure bacteria treatment groups (G2 and G3). These ndings are further con rmed by major organs and GBM tissues harvesting, homogenization and then culturing the homogenates on plates (Figs. 7c and  7d). It is found that sporadic colonies grew from homogenates of GBM and livers in the Trojan bacteria treatment groups (G4 and G5), much less than those in pure bacteria treatment groups (G2 and G3). Accumulating evidence demonstrated only Trojan bacteria-treated mice under NIR irradiation could help to eliminate bacteria from the body.

Discussion
Due to the existence of the BBB in GBM, the accumulation of peripherally administered drugs into brain is seriously hindered. It has been reported that bacteria can bypass the BBB transcellularly, paracellularly and/or in infected phagocytes [34][35][36] . Intriguingly, cancer therapeutics internalized into bacteria would potentially circumvent drug delivery issues in GBM therapy. However, this progress is still in its infancy due to the lack of a natural connection between therapeutics and the microbiome. Herein, the developed Trojan bacteria system naturally bridge the divide between therapeutics and bacteria. We have previously demonstrated that bacteria including Gram-negative as well as Gram-positive bacteria actively swallowed GP-conjugated nanoparticles through bacteria-speci c ABC transporter pathway for ultrasensitive diagnosis of bacterial infections 42 . In this context, we have successfully constructed Trojan bacteria as drug delivery vehicles for GBM therapy, in which the Trojan bacteria system was made of GP-ICG-SiNPs internalized into VNP or EC. Consistently with previous reports, the as-prepared GP-ICG-SiNPs could be internalized into the host bacteria, con rmed by TEM, SEM and confocal images. We also performed an inhibition assay as well as a competition assay to further verify GP-ICG-SiNP accessing into the bacterial intracellular volume was through the ABC transporter pathway.
Subsequently, we demonstrated the notion that the constructed Trojan bacteria could cross BBB, target and penetrate GBM. As such, in the constructed Trojan bacteria, the bacteria could take therapeutics together to cross BBB, target and penetrate GBM tissues. Consequently, under the irradiation of 808-nm laser, photothermal treatment of ICG can induce tumour cell destruction and bacterial rupture. The produced lysates of tumour and bacterial cells upon PTT effects can act as immune stimulants to enhance the anti-tumor immune response. Compared with the use of equivalent free GP-ICG-SiNPs or pure bacteria for GBM therapy, this constructed Trojan bacteria system not only greatly augmented targeted delivery of GP-ICG-SiNPs towards the GBM, but also synergistically promoted antitumor immune responses that prolonged the survival of GBM-bearing mice. In addition, distinguished from pure bacteria therapy which would preserve a lot of residual bacteria after treatment, that residual bacteria could be effectively eliminated from the body after the Trojan bacteria treatment. Accumulating evidence demonstrated that the therapeutic system of Trojan bacteria could achieve photothermal immunotherapeutic effects and safety pro les under modulation of NIR light irradiation, providing a plausible microbiota-based therapeutic strategy against CNS diseases.

Methods
Synthesis of GP-ICG-SiNPs. SiNPs were synthesized by mixing 1,8-naphthalimide and 3aminopropyltrimethoxysilane, followed by 365 nm UV irradiation at room temperature for 40 min. The resulted solution was then centrifuged at 6000 rpm for 20 min to remove unreacted reagents, and further puri ed by dialysis (MWCO, 1000, Spectra/Pro). The as-synthesized SiNPs (200 µL, 20 mg/mL) were mixed with GP (100 µL, 10 mg/mL) at 70 o C for 6 h, followed by the addition of 0.02 mg of NaBH 4 . After reacting 12 h at room temperature, the stable GP-modi ed SiNPs were obtained. To remove unreacted GP, Nanosep centrifugal devices (MW cutoff, 3 kDa; Millipore) were used to lter the reaction solution at 7500 rpm for 15 min. After that, ICG were co-incubated with GP-SiNPs, and stirred at 4 o C overnight. Then, the unreacted ICG were centrifuged by Nanosep centrifugal devices (MW cutoff, 3 kDa; Millipore) at 7000 rpm for 15 min. Then the products of GP-ICG-SiNPs were harvested and stored at 4 o C in the dark for the following experiments. Transmission electronic microscopy (TEM, Philips CM 200) with 200 kV was used for the characterization of the morphology and size of the nanoagents. UV-vis absorption spectra of nanoagents were measured by A 750 UV-vis near-infrared spectrophotometer (Perkin-Elmer lambda). Photoluminescence (PL) spectra of nanoagents were recorded by a spectro-uorimeter (HORIBA JOBIN YVON FLUORMAX-4). Dynamic light scattering (DLS) of nanoagents was analyzed by a Delsa™ nano submicron particle size analyzer (Beckman Coulter, Inc).
Bacterial culture. Attenuated Salmonella strain VNP20009 (VNP) and Escherichia coli 25922 (EC) were were purchased from American Type Culture Collection (ATCC). These bacterial cells were harvested at the exponential growth phase when they were cultured in LB liquid medium (250 rpm, 37 o C). Afterwards, the bacterial suspensions were washed twice and re-suspended in PBS buffer for the following experiments. The bacteria count in solution was determined by the measurement of corresponding optical density (OD) at 600 nm. The numbers of bacterial colonies were counted by a colony counting instrument (Czone 8).
Construction of mCherry@VNP and mCherry@EC. To construct mCherry@VNP, monoclonal VNP was selected and inoculated in 2 mL of LB liquid medium without ampicillin. Then, 0.5 mL of bacterial solution was added into 50 mL LB liquid medium and cultured at 37 o C until its OD 600 reached 0.5. After that, the shake ask was ice-bathed for 30 min, and 25 mL of bacterial solution was added to a pre-cooled 50 mL round bottom centrifuge tube, and centrifuged at 4 o C and 1000 rpm for 15 min to obtain bacterial pellets. Then the bacterial precipitation was washed twice with pre-cooled sterile water. The bacterial pellet was resuspended with 10 mL of 10% sterile glycerin, and centrifuged at 4 o C and 1000 rpm for 15 min. Then 50 µL 10% sterile glycerin was added to make the bacterial precipitation fully suspended and transferred to a precooled 0.5 mL centrifuge tube to form electrically transformed competent cells. We took 1 ng of mCherry plasmid and added it to VNP competent cells, mixed it gently, ice bath for 1 min, transferred it into a cold electroporation cup (Bio-Rad), and placed it on the electrode for electric shock transformation (Electric shock parameter: voltage = 2.5 kV, capacitance = 25 µF, resistance = 200Ω). After the electric shock, we added immediately 1 mL of LB liquid medium to the electroporation cup, mixed, and transferred all to a sterile 1. Cellular experiments in vitro. The G422 mouse glioblastoma cell line was obtained from Shanghai Zhong Qiao Xin Zhou Biotechnology and cultured under appropriate conditions. Dendritic cells (DCs) were isolated from the bone marrow of female Balb/c mice about 6-8 weeks old. G422 cells were inoculated in a 96-well plate, co-incubated with EC, VNP, GP-ICG-SiNPs Trojan EC or Trojan VNP for 6 h, irradiated with or without 808-nm laser for 5 min, and then cleaned with sterile PBS. MTT assay was used to determine the cellular viability. On the other aspect, the treated cells were stained with Calcein-AM (CAM) and propidium iodide (PI), and then analyzed by confocal microscopy (CAM: λ ex = 488 nm, λ em = 500-545 nm; PI: λ ex = 543 nm, λ em = 560-620 nm). In order to evaluate the maturation of DCs in vitro, we designed a transwell system. The DCs in the lower chamber were collected, stained with anti-CD11c-FITC, anti-CD86-PE and anti-CD80-APC, and sorted by ow cytometry.
Human brain microvascular endothelial cell model. A 12-well transwell plate with 2 µm of mean pore size membrane was used to construct the in vitro human brain microvascular endothelial cell (HBMEC) model.
The HBMEC cells (1.0×10 5 cells/well) were seeded in the transwell insert with 12 mm diameter. The transendothelial electrical resistance (TEER) values were detected by a Millicell-ERS volt-ohmmeter to monitor the cell monolayer integrity during the cell culture process. A TEER value between 150 and 300 Ω cm 2 was suitable for the further experiments.
Multicellular spheroids model. The 3D tumour spheroids of U87MG cells were obtained using a liquid overlay method. Each well of 96-well plates was pre-coated with 100 µL of the FBS free medium containing sterile agarose (2%, w:v). Subsequently, U87MG cells (5000 cells/well) were seeded into each well and cultured in the medium containing FBS (10%, v:v). The tumour spheroids were allowed to grow up to attain the diameter about 750 µm for 8 days at 37°C. The Z-stack scanning was performed on the U87MG multicellular spheroids from top to bottom with 40 µm per section by CLSM.
Animal experiments. All animal experimental procedures were performed according to the Guideline for Animal Experimentation with the approval of the animal care committee of Soochow University. In addition, the studies in human blood were conducted under a protocol that was reviewed and approved by the ethics committee of Soochow University. To construct the GBM-bearing mice model, the G422 cells (8×10 5 /per mouse) in a 5 mL volume were stereotactic injected using a 10 mL Hamilton syringe into the left striatum over 1 min into the following coordinates: 0.5 mm anterior, 2 mm left lateral from bregma, and 3.1 mm deep. To study the in vivo distribution of bacteria, mCherry@EC or mCherry@VNP (~1.0×10 7 CFU per mouse) was intravenously injected into the healthy mice or GBM-bearing mice after inoculation with Luc-G422 cells (10 5 /per mouse) for 7 days. Afterwards, the mice were sacri ced at speci c time points (12,24,36, 72 and 120 h). Accordingly, the organs including heart, liver, spleen, lung and kidney as well as brain were extracted, followed by imaging via an in vivo optical imaging system (IVIS Lumina III). Meanwhile, the excised organs were homogenized in sterile PBS, in which the suspension solution was collected from the tissue dispersions by centrifugation at 1000 rpm to remove tissue fragments. Finally, the collected suspension solution was diluted by PBS buffer and cultured on LB solid medium at 37°C for 12 h, followed by counting bacterial colonies with a colony counting instrument (Czone 8). On the other aspect, the GBM-bearing mice were respectively injected with (G1) PBS, (G2) ~1×10 7 CFU EC, (G3) 1×10 7 CFU VNP, (G4) GP-ICG-SiNPs (8 mg/kg ICG), (G5) Trojan EC (e.g., GP-ICG-SiNPs (8 mg/kg ICG) internalized into ~1×10 7 CFU EC) and (G6) Trojan VNP (e.g., GP-ICG-SiNPs (8 mg/kg ICG) internalized into 1×10 7 CFU VNP) after inoculation for 7 days (n=5, female, 6-7 weeks old). After intravenous injection for 12 h, the GBM sites were irradiated by 808-nm laser (1.2 W/cm 2 , 5 min). The laser spot was adjusted according to the burr hole on the skull. The temperature of GBM sites was stabilized at 50 o C for 5 min by adjusting the 808 nm laser power. Next, at 5-day and 10-day post-injection with bacteria, GP-ICG-SiNPs (containing 8 mg/kg of ICG), Trojan EC, Trojan VNP were intravenously injected into the mice again. Analogously, at 5-day and 12-hour post-injection or 10-day and 12-hour post-injection, the GBM sites of these mice were irradiated by 808 nm laser and the temperature of GBM sites was stabilized at 50 o C for 5 min by adjusting the 808 nm laser power. Afterwards, bioluminescence imaging was applied to visualize the anti-tumor effect every four days. On the other aspect, we harvested the carotid lymph nodes after 3 days post-last injection and homogenized them into single cell suspensions. Lymphocytes were stained with anti-CD11c-FITC, anti-CD80-APC and anti-CD86-PE for DC maturation analysis, while splenocytes labeled with anti-CD3-FITC anti-CD4-APC and anti-CD8a-PerCP was used to examine CD8a + T cell response. Then, the mice were sacri ced at 16-day. GBM were harvested, and of which single cell suspensions were labeled with CD3-FITC and CD8a-PerCP was used to examine CD8a + T cell response by ow cytometry. Meanwhile, serum samples were isolated from mice after various treatments and diluted for analysis. What's more, the tumors were harvested, xed in 4% paraformaldehyde for 24 h, and embedded in para n. Tissues were sectioned, stained with H&E and then observed with an optical microscope. To examine whether residual bacteria were eliminated from the body after treatment, the excised organs after cancer treatment were imaged by the in vivo optical imaging system (IVIS Lumina III). Meanwhile, the harvested organs were homogenized in sterile PBS and the residual bacterial dispersions were collected by centrifugation. Then the collected bacterial solutions were diluted and cultured on an agarose medium for 12 h. The numbers of bacterial colonies were obtained by a colony counting instrument (Czone 8).
Statistical analysis. The confocal images were processed by the commercial image analysis software (Leica Application Suite Advanced Fluorescence Lite, LAS AF Lite) and common software of ImageJ (NIH Image; http;//rsbweb.nih.gov/ij/). Error bars represent the standard deviation obtained from three independent measurements. All the statistical analyses were performed using the Origin and GraphPad Prism 7 software. The statistical signi cance of differences was determined by a one-way ANOVA analysis. p < 0.05 (*), p < 0.01 (**) and p < 0.001 (***) were used to indicate statistical difference.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Declarations
Data availability. The main data supporting the results in this study are available within the paper and its Supplementary Information. The raw and analyzed datasets generated during the study are too large to be publicly shared, but are available for research purposes from the corresponding author on reasonable request.

Figure 2
Construction and characterization of Trojan bacteria system. a, SEM images of EC, Trojan EC, VNP and Trojan VNP. Scale bars: 200 nm. b, Elemental mapping in HAADF-STEM images of EC, Trojan EC, VNP and Trojan VNP. Scale bars: 500 nm. c, CLSM images of EC and VNP incubated with GP-ICG-SiNPs. Scale bars: 25 μm. d, CLSM images of EC and VNP incubated with ICG-SiNPs. Scale bars: 25 μm. e-g, Flow cytometry analysis of the uptake rates of EC and VNP incubated with different concentrations of GP-ICG-SiNPs for 2 h (e) and corresponding quantitative analysis of uptake rates of different concentrations of GP-ICG-SiNPs by EC (f) and VNP (g). h, CLSM images of the mixture of human blood and EC after incubation with GP-ICG-SiNPs. Arrows indicate red blood cells (RBCs). Scale bars: 25 μm. The EC or VNP were incubated with the synthesized nanoagents ([SiNPs] = 12 mg/mL, [ICG] = 600 μg/mL) at 37 oC for 2 h. After incubation, the treated bacteria were rinsed with PBS buffer for several times. The bacterial cell concentration is ~1.0 ×107 CFU. All error bars represent the standard deviation determined from three independent assays. All data are presented as means ± SD. Statistical signi cance is calculated via oneway analysis of variance (ANOVA) with a Tukey post-hoc test (***P< 0.001).  In vivo behaviors of bacteria in mice. a, Average body weights of healthy mice injected with EC or VNP with different concentrations. The red arrow indicates the death of mice. Data are presented as the mean ± SEM. b-d, Ex vivo uorescence images of major organs (heart, liver, spleen, lung, kidney and brain) of healthy mice after the intravenous injection with mCherry@EC and mCherry@VNP at the dose of ~1×107 CFU for 12, 24, 72, 120 and 360 h (b) and corresponding uorescence intensity in mCherry@EC group (c) Figure 5 Trojan bacteria crossing BBB, targeting and penetrating GBM. a, Schematic diagram of in vitro BBB model for evaluating whether the Trojan bacteria could cross the BBB. b, The corresponding penetration rates of Trojan EC or Trojan VNP at 1, 2, 3 and 4h (mean ± SD, n = 3). c-e, Ex vivo uorescence images of major organs (heart, liver, spleen, lung, kidney and brain) of GBM-bearing mice after intravenous injection with mCherry@EC and mCherry@VNP at the dose of ~1×107 CFU for 12, 24, 72 and 120 h (c) and corresponding uorescence intensity in mCherry@EC group (d) and mCherry@VNP group (e) (mean ± SD, n = 3). f-h, Homogenates of major organs of GBM-bearing mice after intravenous injection with mCherry@EC (left) and mCherry@VNP (right) for 12, 24, 72 and 120 h cultured on the solid LB agar (n=3) (f) and corresponding quanti cation of bacterial colonization in mCherry@EC group (g) and mCherry@VNP group (h) (mean ± SD, n = 3). i, Penetration of Trojan bacteria in 3D tumor microspheres. j, In situ hybridization uorescence image of GBM tissues. The nucleus, hypoxic zone and bacteria were stained with DAPI (blue), anti-HIF-α antibody (green) and 16S RNA probe (red), respectively. Scale bars: 100 μm.  k, Quali cation of CD3+ CD8a+ T cells in splenocytes by ow cytometry (mean ± SD, n = 3, **P< 0.01). l, Quali cation of CD3+ CD8a+ T cells in tumour by ow cytometry (mean ± SD, n = 3, **P< 0.01). m, Quali cation of IFN-γ in serum of GBM tumour-bearing mice on the 16th day after treatment (mean ± SD, n = 3, ***P< 0.001). n, Quali cation of IFN-γ in serum of GBM tumour-bearing mice on the 16th day after treatment (mean ± SD, n = 3, ***P< 0.001). Statistical signi cance was calculated via one-way analysis of variance (ANOVA) with a Tukey post-hoc test. The elimination of residual bacteria after photothermal immunotherapy. a, The uorescence distribution in the main organs (heart, liver, spleen, lung, kidney and brain) of GBM-bearing mice after photothermal immunotherapy. The mice were intravenously injected with (G1) PBS, (G2) ~1×107 CFU mCherry@EC (m@EC), (G3) ~1×107 CFU mCherry@VNP (m@VNP), (G4) Trojan m@EC (e.g., GP-ICG-SiNPs (8 mg/kg ICG) internalized into ~1×107 CFU m@EC) or (G5) Trojan m@VNP (e.g., GP-ICG-SiNPs (8 mg/kg ICG) internalized into 1×107 CFU m@VNP), respectively. At the 12-hour post-injection, the brains of those mice were suffered by an 808 nm irradiation (1.2W/cm2, 5 min), followed by ex vivo imaging of the main organs at 5-day. b, The corresponding quantitative analysis of uorescence intensity of main organs in different groups after photothermal immunotherapy (mean ± SD, n = 3, ***P< 0.001). c, Homogenates of major organs of GBM-bearing mice in different groups after photothermal immunotherapy cultured on the solid LB agar. d, Corresponding quanti cation of bacterial colonization on LB solid plates in different treatment groups (mean ± SD, n = 3, ***P< 0.001). Statistical signi cance was calculated via one-way analysis of variance (ANOVA) with a Tukey post-hoc test.