Extracellular vesicles-based pre-targeting strategy enables multi-modal imaging of orthotopic colon cancer and image-guided surgery

Backgroud Colon cancer contributes to high mortality rates as the result of incomplete resection in tumor surgery. Multimodal imaging can provide preoperative evaluation and intraoperative image-guiding. As biocompatible nanocarriers, extracellular vesicles hold great promise for multimodal imaging. In this study, we aim to synthesized an extracellular vesicles-based nanoprobe to visualize colon cancer with positron-emission tomography/computed tomography (PET/CT) and near-infrared fluorescence (NIRF) imaging, and investigated its utility in image-guided surgery of colon cancer in animal models. Results Extracellular vesicles were successfully isolated from adipose-derived stem cells (ADSCs), and their membrane vesicles were observed under TEM. DLS detected that the hydrodynamic diameters of the extracellular vesicles were approximately 140 nm and the zeta potential was − 7.93 ± 0.24 mV. Confocal microscopy showed that extracellular vesicles had a strong binding ability to tumor cells. A click chemistry-based pre-targeting strategy was used to achieve PET imaging in vivo. PET images and the biodistribution results showed that the best pre-targeting time was 20 h, and the best imaging time was 2 h after the injection of 68 Ga-L-NETA-DBCO. The NIRF images showed that the tumor had clear images at all time points after administration of nanoparticles and the Tumor/Muscle ratio peaked at 20 h after injection. Our data also showed that both PET/CT and NIRF imaging clearly visualized the orthotopic colon cancer models, providing preoperative evaluation. Under real-time NIRF imaging, the tumor location and tumor boundary could be clearly observed. Conclusions In brief, this novel nanoprobe may be useful for multi-modal imaging of colon cancer and NIRF image-guided surgery. More importantly, this study provides a new possibility for clinical application of extracellular vesicles as nanocarriers. Graphic Abstract Supplementary Information The online version contains supplementary material available at 10.1186/s12951-021-00888-3.


Introduction
At present, surgery still plays a crucial role in colon cancer treatment, with more than 50% of cancer patients undergoing surgery. Complete resection is a major challenge in tumor surgery [1]. In response, multi-modal imaging has been performed to visualize cancer before and during surgical procedures to provide pre-or intraoperative disease-specific images in real time [2]. Multimodal imaging is a combination of two or more imaging modalities that offer biologically complementary information and may provide a better imaging solution than individual technologies [3]. Positron emission tomography/computed tomography (PET/CT) imaging can enable visualization of tumors and their regression or progression, offering high sensitivity, excellent penetration, and high spatial resolution [4]. Near-infrared fluorescence (NIRF, 650-1000 nm) imaging can provide anatomical information and real-time delineation of tumor because of the advantages of high temporal resolution, spatial resolution, super sensitivity and lower background [5][6][7]. Therefore, multi-modal PET/CT and NIRF imaging of colon cancer can be applied to obtain preoperative and intraoperative imaging of tumor.
Nanoparticle-based tracers have unique characteristics to serve as carriers for both radionuclide and NIRF dye labels during image-guided surgery. Recently, extracellular vesicles have attracted increasing attention, as they play important roles in physiology, pathology, and oncology. Extracellular vesicles are nanosized phospholipid bilayer vesicles secreted by various cells [8]. As biological nanoparticles, extracellular vesicles have been used to deliver drugs to specific cell types or tissues in vivo, especially tumor tissues [9][10][11]. For example, drugs like paclitaxel, imatinib, doxorubicin, curcumin, anthocyanidin, acridine orange, as well as nucleic acids, have been successfully loaded into extracellular vesicles and delivered to target cells [12]. Previous studies have confirmed that extracellular vesicles are ideal nanoparticles because of the following advantages: they are secreted by nearly all cell types and can be found in multiple types of extracellular fluids, such as the blood, urine, amniotic fluid, saliva, cerebrospinal fluid, and breast milk [13]. Second, the intrinsic small sizes of some extracellular vesicles facilitate their extravasation through tumor vessels and their subsequent diffusion into tumor tissues [14]. Third, the membrane structure of extracellular vesicles is similar to that of cells. They can be easily modified with functional groups and directly fuse with the membrane of target cells, thus improve the feasibility of modifying the target cell membrane [15]. Some human clinical trials performed using extracellular vesicles from dendritic cells for cancer therapy reported positive results regarding their feasibility and safety [16]. As reported, extracellular vesicles from various sources have different properties and can be applied to distinct functions [17]. Adipose-derived stem cells emerged as a stable source of extracellular vesicles due to the advantages of facile availability. Extracellular vesicles extracted from adipose-derived stem cells (ADSCs-EV) can be employed to engineer a multi-modal imaging probe because of their preferential tumor targeting ability [17].
To date, great efforts have been dedicated to investigating their applications as natural drug-delivery systems, but the tracing of extracellular vesicles has not been perfected, which stands to accelerate the clinical applications of extracellular vesicles-based drug delivery systems and the study of extracellular vesicles-based theranostic probes. Previous studies have traced extracellular vesicles-based drug delivery systems by optical imaging [18], but it was neither quantitative nor accurate because of the limited tissue penetration. Magnetic resonance (MR), PET, and single photon-emission computed tomography (SPECT) have been applied for tracking extracellular vesicles and analyzing their biodistribution [19][20][21][22]. Therefore, the possibility of extracellular vesicles used as nanocarriers for multi-modal imaging of colon cancer has been demonstrated in theory. However, most previous studies exhibited high tracer uptake in the liver, spleen, and kidneys, and low uptakes in tumor sites, making it difficult to compare the biodistribution of extracellular vesicles from different sources in vivo from images.
In this study, extracellular vesicles were isolated from ADSCs and modified with 1,2-distearoyl-sn-glycero-3-phosphoethanolamine-N-[azido(polyethylenegly col)-2000] (DSPE-PEG 2000 -N 3 ) and DSPE-PEG 2000 -Cy7, then the modified ADSCs-EV (N 3 -EV-Cy7) was obtained. A pre-targeting strategy are applied to label N 3 -EV-Cy7 for imaging with the goal of increasing the target/nontarget ratio and improving the image quality. Additionally, the strategy can shorten imaging time and reduce potential radiation damage. Then, in vivo PET/CT and NIRF imaging of tumor-bearing nude mice was performed to verify the potential application of extracellular vesicles. Subsequently, tumor resections were performed to under real-time NIRF imaging (Scheme 1). We believe our results confirm the feasibility of extracellular vesicles-based nanoprobes for multimodality imaging and image-guided surgery of colon cancer in animal models.

Characteristics of ADSCs-EV
ADSCs-EV appeared membrane vesicles under TEM (Fig. 1a). Western blotting indicated that the ADSCs-EV positively expressed surface markers such as CD9 and CD63. While non-expression markers of ADSCs-EV, such as GM130 and β-tubulin (Fig. 1b), were negatively expressed. The mean hydrodynamic diameters of ADSCs-EV and Cy7-EV-N 3. were approximately 140 nm (Fig. 1c). The zeta potentials of the ADSCs-EV and Cy7-EV-N3 were − 7.93 ± 0.24 mV and − 9.68 ± 1.3 mV, respectively (Fig. 1d). There was no significant difference in hydrodynamic diameters and zeta potential between ADSCs-EV and Cy7-EV-N 3. The stability of ADSCs-EV at 4 °C was great over 8 days (Fig. 1e, f ). HCT116 cancer cells or adipose stem cell were co-incubated with ADSCs-EV at various concentrations (up to 100 μg/mL) and different time periods (up to 72 h). The results showed that the survival rate of cells in each group was greater than 90% (Additional file 2: Table S1-S2). ADSCs-EV had no obvious toxicity to HCT116 colon cancer cell and adipose stem cell.

Scheme 1
The schematics of multimodal PET and NIRF imaging and real-time NIRF intra-operation based on extracellular vesicles from ADSCs The zeta potential of ADSCs-EV and Cy7-EV-N 3 . e The changes of hydrodynamic diameters of ADSCs-EV at 4℃ over 8 days. f The changes of zeta potential of ADSCs-EV at 4℃ over 8 days

In vitro tumor cell binding
As exhibited in fluorescence images, the uptake of Cy5 and N 3 labeled ADSCs-EV (Cy5-EV-N 3 ) in HCT116 cell was increased over time (Fig. 2a). The quantification of the fluorescent intensity was consistent with images ( Fig. 2b). Confocal microscopy of HCT116 cells incubated with Cy5-EV-N 3 revealed strong fluorescence signals in the cell membrane and cytoplasm (Fig. 2c). L-NETA-DBCO was used as a bifunctional chelator in this study. The labeling procedure for 68 Ga-L-NETA-DBCO resulted in high radiochemical purities (> 95%; Additional file 2: Figure S1A). As shown in Additional file 2: Figure  S1B, the proportion of intact tracer exceeded 90% after incubation in serum at 37 °C for 2 h, indicative of excellent serum stability of our tracer. HCT116 cancer cells or adipose stem cell were co-incubated with 68 Ga-L-NETA-DBCO (37KBq/well) at different time periods (up to 72 h). The results showed that the survival rate of cells in each group was greater than 90% (Additional file 2: Table S3). The cell uptakes were displayed in Fig. 2d. The

In vivo animal PET imaging
Having confirmed the labeling of 68 Ga-L-NETA-DBCO, we performed PET imaging based on different imaging time periods (1, 2, and 3 h) and different pre-targeting time periods (10, 20, 30 h). Comparing different imaging time points, maximum tumor uptake was obtained at 2 h after the injection of 68 Ga-L-NETA-DBCO (Fig. 3a). The T/M ratios displayed in Fig. 3b were consistent with the PET imaging. Then PET imaging with different pretargeting time points (10, 20, 30 h) was performed at the best imaging time (2 h). Comparing different pretargeting time points, PET imaging results indicated that maximum tumor uptake was displayed at 20 h after the injection of Cy7-EV-N 3 (Fig. 3c). The tumor-to-muscle ratios (T/M ratios) were maximal at 20 h after the injection (11.41 ± 1.78; Fig. 2d). The chelator group (only injected with 68 Ga-L-NETA-DBCO) did not show any obvious signal at the tumor site (Additional file 2: Figure  S2).

In vivo biodistribution
Biodistribution was conducted to further quantitative analysis of the distribution of this nanoprobe in vivo. The data (Table 1) was consistent with the imaging. Tissue uptakes at different pre-targeting time are shown in Fig. 4a. Tumor uptake (1.37 ± 0.05%ID/g; Fig. 4b) and T/M ratio (6.68 ± 0.68; Fig. 4c) peaked at 20 h after injection. Liver and spleen showed minimum tracer uptake with values lower than 1%ID/g at all time points. The tumor/liver ratio (1.329 ± 0.26; Fig. 4d) and tumor/spleen ratio (1.71 ± 0.16; Fig. 4e) were peaked at 20 h after injection.

In vivo NIRF imaging
In vivo and ex vivo NIRF images ( Fig. 5a and c) indicated that high tumor uptakes were displayed at all time points. The T/M ratio peaked at 20 h after injection (4.63 ± 0.90; Fig. 5b).

Multi-modal animal PET/CT and NIRF imaging of orthotopic colon cancer
For further studying the nanoprobe, orthotopic colon cancer was used as models for multi-modal imaging. Figure 6 shows that the tumor is clearly visible not only in the primary lesion (right abdomen, Fig. 6a, b), but also in a metastatic site (liver, Fig. 6c). Histology with HE staining confirmed the colon cancer and liver metastasis (Fig. 6d, e). Orthotopic colon cancer showed higher expression level of CD31, while subcutaneous colon tumor tissues showed lower expression of CD31 (Fig. 6f ).

Real-time NIRF imaging for intraoperative guidance
Having done Multi-modal animal PET/CT and NIRF imaging of orthotopic colon cancer, we further performed the tumor surgery under real-time NIRF imaging. Under the guidance of real-time NIRF imaging, the location of the tumor could be identified preoperatively, the boundary of the tumor can be confirmed intraoperatively, and the absence of residual tumor can be observed postoperatively (Fig. 6g). The image-guided video is shown in Additional file 1: Video S1.

In vivo toxicity studies
BALB/c mice (n = 5) received an i.v. injection of N 3 -EV-Cy7 + 68 Ga-L-NETA-DBCO or PBS to evaluate the in vivo potential toxicity. No significant hepatic or renal toxicity was observed from the indicating normal Table 1 The biodistribution at different pre-targeting time points (10, 20 and 30 h). Data are expressed as mean ± standard deviation (%ID/g, n = 4)  values of liver and kidney function markers, including ALT, AST, ALP, BUN and CRE ( Fig. 7a-d). We did not observe significantly evidence of major organ damage from the H&E staining sections (Fig. 7e).

Discussion
In this study, we prepared a multi-modal nanoprobe with extracellular vesicles as the nanocarrier. This nanoprobe exhibited suitable size, good stability, and was non-cytotoxic and bound well to tumors. Multi-modal imaging with PET and NIRF were successfully achieved with this nanoprobe, and the surgery was conducted successfully under real-time NIRF imaging. To the best of our knowledge, pre-targeting strategy have not been applied for labeling extracellular vesicles-based nanoprobe. In other words, this is the first report of extracellular vesicles as the nanocarriers for multi-modal imaging and surgery of tumors with the application of pre-targeting strategy. Multimodality imaging, a combination of two or more imaging modalities, can provide more complementary information for imaging than individual techniques. To date, many researchers have constructed multimodality nanoprobes with the application of synthetic nanoparticles, such as liposomes, metal nanoparticles, and magnetic nanoparticles, because of their advantages, such as large carrying capacity and facile surface modification [23,24]. However, the widespread use of artificial drug carriers has been prevented by their potential toxicity, immunogenicity, and inability to penetrate and target specific organs. These disadvantages may be largely avoided when these drug carriers are in the form of biological structures.
The use of extracellular vesicles as nanocarriers has attracted wide attention because of their excellent biocompatibility [25]. Extracellular vesicles could facilitate their extravasation through tumor vessels and their subsequent diffusion into tumor tissues. They have the advantages of membrane-permeability, good stability in vivo, easy surface modification, and large loading capacity. Most importantly, unlike extracellular vesicles from cancer cells, normal cell-derived extracellular vesicles almost have no toxicity and are well tolerated by human [26]. Therefore, extracellular vesicles are favorable candidate nanocarriers. In this study, our data also confirmed the advantages of the extracellular vesicles-based nanoprobe. The extracellular vesicles-based nanoprobe was stable as revealed by its unchanged size and zeta potential over 8 d. The nanoprobe exhibited no obvious cytotoxicity, as shown by CCK8 assay and in vivo toxicity studies. Additionally, fluorescence imaging and confocal imaging showed that extracellular vesicles had great tumor cell binding ability and excellent cell internalization. Therefore, extracellular vesicles-based nanoprobes have great promise for multi-modal imaging.
As we all know, the half-life of 68 Ga is short (67.7 min), which does not match the ADSCs-EV with high . f Immunohistochemistry assay of CD31 in orthotopic and subcutaneous colon cancer. g Representative NIRF images of tumor-bearing mice pre-, intra-, and postoperatively. The red arrows point to the tumor molecular weight. According to the report [27], pretargeting strategy can combine short half-life nuclide and long half-life molecular. Therefore, the pre-targeting strategy based on click reaction (N 3 -DBCO) was applied in this study [28]. The results showed a maximum tumor uptake when the pre-targeting time was 20 h and the imaging time was 2 h. Low uptakes in liver and spleen were displayed at all time points compared with the study by Shi et al. [22]. Although PET/CT imaging offered an abundance of information needed for diagnosis and preoperative planning, it cannot be applied for intraoperative detection of tumor tissues.
Optical image-guided surgery provides the accurate analysis of biological lesions through optical molecular probes, allowing more precise surgery. Compared with traditional optical imaging, NIRF imaging has been applied for intraoperative tumor resection studies because of some advantages, such as high temporal resolution, super sensitivity and lower tissue auto-fluorescence. After comparing the NIRF images acquired at different time periods, the appropriate imaging time for image-guided surgery has been suggested to be 10-30 h. Our surgery was performed under real-time NIRF imaging to further confirm the functions of this nanoprobe. During the operation, we could observe the tumor location, tumor boundaries, and residual after tumor resection clearly.
The orthotopic colon cancer model was designed to demonstrate the advantages of the multimodal nanoprobe further. The biological characteristics of orthotopic colon cancer are more similar to human tumors than subcutaneous colon cancer, such as the position, the morphology, the surrounding tissues, the blood supply, and the environment of colorectal tumors [29]. Excitingly, the orthotopic transplantation tumors showed higher tracer uptake than the subcutaneous tumors in our study according to the PET images. According to CD31 expression, the orthotopic tumor tissues exhibited higher density of blood vessels than subcutaneous tumor tissues.
Some issues remain to be addressed in this study. First, the production and isolation of extracellular vesicles remain challenging. For example, the differential centrifugation method we applied in this study has the advantages of high yield and low cost, but this method cannot be applied in the clinic because of the low purity and complex procedures. The extracellular vesicles from ADSCs are also limited by low yield and cannot be employed in clinical applications. Recently, some researchers tried to isolate extracellular vesicles from milk and fruits to increase the yield [30]. Second, the composition, function, specific targeting, and metabolism in vivo of extracellular vesicles have not yet been fully elucidated. Additionally, various DSPE-PEG 2000 functionalized head-groups can be modified on extracellular vesicles to increase the tumor-targeting ability of the nanoprobe using hydrophobic interaction approach. Briefly, further in vitro and in vivo animal studies based on extracellular vesicles as a nanocarrier are required before clinical translation.

Conclusion
In this study, a novel extracellular vesicles-based nanoprobe was successfully engineered for multimodal PET/ CT/NIRF imaging and image-guided surgery of colon cancer. This nanoprobe showed promise for preoperative evaluation and intraoperative surgical guidance. Our data verified that a pre-targeting strategy can be applied to gain high quality images. This research also proved that extracellular vesicles are also potential high-quality nanocarriers for multimodality imaging and have broad application prospects.

Cell culture
This study was approved by the Ethics Committee at the Tongji Medical College of Huazhong University of Science and Technology (No. 2018-S288). ADSCs were isolated from the subcutaneous fat from patients without cancers. The adipose tissue was cut into small pieces and the connected fascial tissue was removed. Then the adipose tissue was digested for 1 h with 0.2% collagenase type I (Sigma, St Louis MO, USA) and centrifuged for 4 min at 1000 rpm. The cell pellets obtained were passed through a 70-μm filter (Corning, Rochester NY, USA), and then cultivated in Dulbecco's modified Eagle's medium (DMEM; Gibco, Gaithersburg MD, USA) with 10% fetal bovine serum (FBS, Serapro, Naila, Germany). The human colon cancer cell line HCT116 and mouse colon cancer CT26 were preserved in our laboratory (Hubei Province Key Laboratory of Molecular Imaging) and propagated in an RPMI-1640 medium (Gibco) supplemented with 10% FBS.

ADSCs-EV isolation
Extracellular vesicles from serum-free ADSCs culture supernatant were obtained by differential centrifugation. Dead cells and cell fragments were removed by centrifugation at 3000 g for 30 min. Then the supernatants were centrifuged at 13,000 g for 70 min. The supernatants were concentrated by an Amicon ® Ultra-15 Centrifugal Filter Device (100 kDa molecular weight, Millipore, USA). Finally, the supernatants were centrifuged at 120,000 g for 70 min to obtain ADSCs-EV, which were subsequently suspended in PBS, passed through a 0.22 μm filter, and quantified by surface proteins with a BCA Protein Assay Kit (Beyotime, Shanghai, China) and stored at − 80 °C.

ADSCs-EV characterization
Extracellular vesicles were examined by transmission electron microscopy (TEM, Hitachi, Japan) and dynamic light scattering (DLS, Malvern Instruments Ltd., Worcestershire, UK). Then western blot was carried out to identify the ADSCs-EV by several surface markers. Briefly, equal amounts of total protein samples from ADSCs-EV (30 µg) were loaded in each well of SDS-PAGE gels. The gels were subsequently transferred onto PVDF membranes and incubated overnight at 4 °C with following primary antibodys: CD9 (Cat #A1703; ABclonal), GM130 (Cat #11308-1-AP; Proteintech), CD63 (Cat #ab134045; Abcam) and β-tublin (Cat #10094-1-AP; Proteintech). On the following day, the membranes were incubated with HRP-conjugated antibody (Aspen, China) for 1 h. After incubation, the membranes were again washed 3 times with PBS and exposed to X-ray films using ECL detection reagents (#WP20005, Thermo Fisher). Changes in hydrodynamic diameters were monitored for 8 d by DLS to test the stability of the ADSCs-EV in vitro. A Cell Counting Kit-8 (CCK8) kit (SAB, College Park, MD, USA) was used to identify the cytotoxicity of different concentrations of ADSCs-EV and 68 Ga-L-NETA-DBCO in HCT116 human colon cancer cells.

Tumor-bearing nude mouse models
The protocol of mouse experiments was reviewed and approved by the Animal Care Committee of Tongji Medical College, Huazhong University of Science and Technology. HCT116 cells (5 × 10 6 ) suspended in 100 μL PBS were subcutaneously injected into the right upper limb of BALB/C nude mice (male, 4 weeks old, Beijing HFK Bioscience co., Ltd, China). After the tumor size reached approximately 0.8 cm, the mice were prepared for study. Orthotopic colon cancer models were also prepared with the following protocol. The 5-week BALB/C nude mice were laparotomized to expose the cecum, then HCT116 cells (5 × 10 6 ) suspended in 50 μL PBS were injected into the serosal layer of the colon. Four weeks after cell injections, the mice were prepared for study.

Synthesis and identification of 68 Ga-L-NETA-DBCO
68 GaCl 3 was obtained from the 68 Ge/ 68 Ga generator with HCl (0.05 M) as eluent. Sodium acetate (1.25 M, pH = 8.6) was added to 500 μL of 68 GaCl 3 (187 MBq) to adjust the pH to 3.7. L-NETA-DBCO (5 nmol) was used to chelate the radionuclide 68 Ga, and the reaction was maintained 10 min at 100 °C. After the mixture was cooled, a C18 column was used to purify 68 Ga-L-NETA-DBCO. 68 Ga-L-NETA-DBCO was conjugated with N 3 -modifed ADSCs-EV by in vivo click reaction, which enables PET imaging. The radiochemical purity and in vitro stability of the probe (2 h in fetal bovine serum) were measured by high-performance liquid chromatography (HPLC).

In vivo PET imaging
In order to identify the best time points, Cy7-EV-N 3 (200 μg) was intravenously injected into the HCT116 tumor-bearing nude mice at different pre-targeting time points (10, 20, and 30 h). 68 Ga-L-NETA-DBCO (3.7 MBq) were injected into the mice (n = 3 per group) via the tail vein. Mice were anesthetized with 2% isoflurane, and micro-PET static imaging was performed at 1, 2 and 3 h after the injection of 68 Ga -L-NETA-DBCO. Static PET images were collected for 10 min using a small-animal PET scanner (TransPET BioCaliburn 700, Raycan Technology Co., Ltd, Suzhou, China). PET images were reconstructed with the ordered-subset expectation maximization three-dimensional/maximum a posteriori probability algorithm, and then image analysis was performed using Amide (http:// amide. sourc eforge. net) and Carimas 2.10 (Turku PET Centre, Finland) software.

NIRF imaging
Cy7-EV-N 3 was injected into the mice bearing HCT116 tumor grafts (n = 3 per group) via the tail vein for NIRF imaging. Mice were anesthetized with 2% isoflurane, and NIRF imaging was performed at different times (1,5,10,20,30, and 50 h). The static NIRF images were acquired with 750-nm excitation and 790-nm emission filters using an IVIS spectrum imaging system (In-Vivo FX PRO, Bruker, Germany). NIRF images were analyzed by Bruker MI (Germany).

Multimodal PET/CT and NIRF imaging of orthotopic colon transplantation tumor
Twenty hours after Cy7-EV-N 3 injection, 68 Ga-L-NETA-DBCO were injected into the orthotopic colon transplantation tumor model mice via the tail vein for click chemistry in vivo. 2 h after 68 Ga-L-NETA-DBCO injection, the mice's bladders were emptied by compression and the mice were anesthetized using 2% isoflurane. Static PET/CT images were collected for 10 min using a small-animal PET scanner (TransPET Discoverist 180, Raycan Technology Co., Ltd, Suzhou, China). PET/ CT images were reconstructed with the ordered-subset expectation maximization three-dimensional/maximum a posteriori probability algorithm, and then the analysis of images was done using Amide (http:// amide. sourc eforge. net) and Carimas 2.10 software. After PET/CT imaging was completed, NIRF imaging was subsequently performed and analyzed as described above. The mice were then sacrificed and the tumor tissue was collected for HE staining of pathological sections.

Immunohistochemistry analysis
The CD31 immunohistochemistry analysis was performed to evaluate the vascularization of the tumors. The HCT116 tumor tissues were collected, fixed in 4% paraformaldehyde and embedded in paraffin. The tumor Sects. (5 μm) were dewaxed, rinsed with EDTA buffer (pH 9.0), and blocked with 3% hydrogen peroxide. The tumor sections were incubated with anti-CD31 antibody (Abcam, Cambridge MA, USA) at 4 °C overnight. Then the tumor slices were incubated with secondary antibody (HRP-labeled goat anti-rabbit IgG, Abbkine, Redlands CA, USA) at room temperature for 30 min. The sections were stained with 3, 3'-diaminobenzidine (DAB, Beyotime, Hangzhou, China) for 8 min, subsequently by counterstaining with hematoxylin (Beyotime) for 2 min and were observed under microscopy.

Real-time NIRF imaging for intraoperative guidance
Twenty hours after the injection of Cy7-EV-N 3 , resection of the subcutaneous tumors in mice was performed using a real-time IVIS spectrum imaging system (Premium Imaging FB800, Premium imaging, California, USA).

Statistical analysis
Data are shown as the mean ± standard deviation (SD). Comparisons between groups were evaluated with the unpaired Student's t-test. p < 0.05 was considered to be statistically significant. Statistical analysis was conducted using GraphPad Prism v8.0 software.