In vivo nanoparticle-mediated radiopharmaceutical-excited fluorescence molecular imaging

Cerenkov luminescence imaging utilizes visible photons emitted from radiopharmaceuticals to achieve in vivo optical molecular-derived signals. Since Cerenkov radiation is weak, non-optimum for tissue penetration and continuous regardless of biological interactions, it is challenging to detect this signal with a diagnostic dose. Therefore, it is challenging to achieve useful activated optical imaging for the acquisition of direct molecular information. Here we introduce a novel imaging strategy, which converts γ and Cerenkov radiation from radioisotopes into fluorescence through europium oxide nanoparticles. After a series of imaging studies, we demonstrate that this approach provides strong optical signals with high signal-to-background ratios, an ideal tissue penetration spectrum and activatable imaging ability. In comparison with present imaging techniques, it detects tumour lesions with low radioactive tracer uptake or small tumour lesions more effectively. We believe it will facilitate the development of nuclear and optical molecular imaging for new, highly sensitive imaging applications.

Despite these achievements, CLI faces many challenges. Cerenkov luminescence is weak with an intensity that is several orders of magnitude less than that of standard fluorescence imaging 17,18 . This is because most of the radiated energy is in the form of g photons and/or other emitted particles generated during radioactive decay. The CL spectrum is weighted towards ultraviolet and blue 15,19 , which leads to high absorption in biological tissues and limits the imaging depth. Like g radiation in nuclear imaging, CL is not 'triggered' light but is continuously generated during radioactive decay regardless of biological interactions. To achieve 'activatable' imaging of nuclear agents, whose optical signal is activated on relevant biological interactions, the fluorescent conversion of CL has been proposed [19][20][21][22][23] . However, the overall optical signal must be killed for energy transformation, which limits imaging sensitivity.
To overcome these challenges, especially, to achieve activatable imaging and enhance the optical signal, we present a novel in vivo imaging strategy, radiopharmaceutical-excited fluorescence imaging (REFI). It utilizes europium oxide (EO) nanoparticles to convert g-radiation (major) and CL (minor) into fluorescence. This is similar to a process called radioluminescence, which uses ionizing radiation (X-or g-radiation) to irradiate nanophosphors and generate light. Phosphorous materials have been used in radiation detectors for many years. A recent publication reported measurements of Eu(II) and Eu(III) in the new CsSrI3:Eu scintillator using X-ray absorption spectroscopy 24 . However, adopting this concept for bioimaging is quite recent and underexplored [25][26][27][28] . There is no comparison between this approach and other imaging techniques for highly sensitive in vivo tumour detection yet. Our technique combines the merits of CLI and FMI to boost the light intensity and signal-tobackground ratio and achieve internal activatable imaging via the excitation from two different electromagnetic radiations. To investigate the excitation mechanisms, the signal enhancement, the emission spectrum, the tissue penetration ability and the difference in outcomes between REFI, CLI, FMI and PET, a series of in vitro phantom and in vivo xenograft studies are conducted. The benefits of REFI for highly sensitive tumour imaging are demonstrated in two scenarios. We believe that by combining the strengths of CLI and FMI, signal-enhanced optical imaging through internal radionuclide excitation with dual electromagnetic radiations will benefit preclinical research and clinical diagnosis in the future.

Results
EO morphology and spectrophotometry. Scanning electron microscopy (SEM) showed EO nanoparticles with peanut-like morphology (Fig. 1a). The average diameter of EO nanoparticles was 85±22 nm. The z potential was 28.6 mV. The excitation profile is shown in Fig. 1b. There were multiple characteristic absorption peaks at 301, 323, 365, 384, 396, 467 and 536 nm. The emitted fluorescence spectra of EO excited by 400-and 535-nm lasers are shown in Fig. 1c,d. The peak emission was 620 nm in both cases. The absolute quantum yield of EO was 39%.
Our previous study illustrated that an excitation range from ultraviolet to blue can excite EO to generate red emission (620 nm). This indicated that CL (350-450 nm weighted) was also a b d c 2  capable of exciting EO. However, it was not known whether it was the only radiation from radionuclides causing the excitation. Figure 2 demonstrates that neither blocking Cerenkov light (Row II) nor blocking b þ radiation (Row III) significantly influenced the induction of fluorescence from EO. Conversely, the fluorescent signal decreased dramatically when the g radiation from 18 F-FDG or 99m Tc-MDP was blocked (Row IV). Therefore, g radiation was the major cause of EO excitation. Both high (511 keV) and low energy (140 keV) g-rays were able to achieve EO excitation.
The fluorescent emission of EO increased linearly with increasing activity of radioactive tracers (Fig. 3a,b). The excitation efficiency of 99m Tc-MDP was better than that of 18 F-FDG. Fluorescent emission strongly correlated with tracer activity with R 2 values of 0.97 ( 18 F-FDG) and 0.99 ( 99m Tc-MDP). For both 18 F-FDG and 99m Tc-MDP excitation, the emitted fluorescent intensity of EO increased exponentially with an increase in its mass (Fig. 3c,d). Fluorescent emission strongly correlated with the amount of EO with an R 2 value of 0.99. With increasing excitation distance, the EO emission intensity decreased exponentially (Fig. 3e,f). Fluorescent emission strongly and inversely correlated with distance with R 2 values of 0.99 for 18 F-FDG excitation and 0.97 for 99m Tc-MDP excitation. The fluorescent signal was almost undetectable when the excitation distance between 18 F-FDG and EO was over 25 mm.
The fractions of gand Cerenkov excitation in REFI are dependent on radioactivity, gphoton energy, EO mass, and excitation distance. For a single study (100 mCi 18 F-FDG þ 10 mg EO with 10 mm distance), we employed a lead partition and two sets of mirrors (Fig. 3g,h) to obtain dual-radiation excitation and CL radiation excitation separately. The induced mean optical signal of EO was (100.5±5) Â 10 4 p s À 1 and (4.6 ± 0.1) Â 10 4 p s À 1 , respectively (Fig. 3i). Therefore, the fractions of gand CL excitation were 95.4 and 4.6% in this case, which proved g-radiation was the major source of excitation.
Different excitation efficiencies. Figure 3b indicates that different radioisotopes with the same activity caused different excitation efficiencies, and 99m Tc-MDP with lower g photon energy (140 keV) was better than 18 F-FDG (511 keV), when the radiotracer was separated with EO for excitation. However, when we co-mixed different radiotracers, 140 mCi 18 F-FDG, 99m Tc-MDP and 131 I-NaI ( 131 I-sodium iodide, major g energy 364 keV), with 10 mg EO, respectively, and investigate their emission spectra, we found the opposite (Fig. 4a). 18 F-FDG þ EO showed the highest signal, and 99m Tc-MDP þ EO was the lowest. It was a surprise that 18 F and 99m Tc showed different relative excitation efficiency in conditions of separation and mixture. We hypothesized that the photoelectric interactions and Compton scattering of europium and g-photons were related to the distance between radionuclide and europium. When mixing the 18 F with EO, both b þ particles and g-photons transferred energy to extranuclear electrons of EO and resulted in visible photons. However, 99m Tc only emits g photons, thus its excitation efficiency was lower than 18 F in case of mixing with EO. When radiotracer was separated from EO for certain distance, the probability of b þ particles reaching EO was almost zero. It was g photons that dominated the excitation for both isotopes. However, the probability of Compton scattering for 511 keV g photon might be higher than that for 140 keV g photon, which did not induce optical signal. The probability of photoelectric effect for 140 keV g-photon might be higher than that for 511 keV g-photon, which induced optical signal. That was the reason why separating and mixing radiotracers with EO showed different excitation efficiency.
To further explore the relationship between emission intensity and excitation photon energy below 140 keV, we built a hybrid X-ray and optical system ( Supplementary Fig. 1a). X-ray with the energy range of 40 to 130 keV was used to excite the EO ( Supplementary Fig. 1b), and the emission intensity increased linearly with an R 2 value of 0.95 ( Supplementary Fig. 1c).
Optical signal enhancement and spectral red shift. With the presence of 10 mg EO, the optical signal enhancement showed a marked effect (Fig. 4a)  magnitude more than the normal Cerenkov luminescence, even with the only 2-min exposure (Note that it took 5-min exposure time for the rest of our studies of REFI and CLI.). This together with our previous mechanism studies (Fig. 2) proved that EO was utilized to achieve fluorescent conversion of g-radiation emitted from radiopharmaceuticals to boost the overall optical signal. Besides that, all three combinations ( 18 F-FDG, 99m Tc-MDP or 131 I-NaI þ EO) showed similar spectra between 570 and 740 nm. The peak of the emission profile was at 620 nm followed by a smaller peak at 700 nm. It is notable that for wavelength o570 nm, the overall optical signal was weaker than Cerenkov signal emitted from 18 F or 131 I along. This indicated that the blue Cerenkov light also contributed the excitation of EO, therefore part of its energy was shifted from blue to red. The radiopharmaceutical induced fluorescence showed greater signal intensity and shifted the optical spectrum from blue towards red. Both improvements suggested better ability to penetrate tissues. Since the emission profiles were similar for all three radiotracers, and the most clinically used 18 F-FDG demonstrated the best excitation efficiency in the case of co-mixing EO, we focused on the 18 F-FDG with EO for the rest of our studies. Figure 4b,c shows the difference in tissue penetration ability between CLI and REFI. The Cerenkov luminescent signal of the control well (50 mCi 18 F-FDG) was nearly ablated (from 6.4 to 0.4 Â 10 5 p s À 1 ) after covering with a piece of porcine gastric mucosa tissue (1-mm thick). However, the REFI signal was greater in Wells 1 to 4, which contained a range of amounts of EO (0.15-0.60 mg) mixed with 18 F-FDG (50 mCi each). The signal intensity of Well 4 (50 mCi 18 F-FDG þ 0.6 mg EO) was approximately fourfold greater (1.8 Â 10 5 p s À 1 ) than that of the control (Fig. 4d).
Combining the merits of CLI and FMI. CLI does not need external excitation as FMI does, which excludes background reflection and autofluorescence. Thus, an improved signal-tobackground ratio is one of the strengths of CLI. REFI relies on radiation from a radiopharmaceutical to excite EO internally. Combining REFI with CLI confers the absence of external excitation and gives REFI the merit of a high signal-to-background. In both in vitro and in vivo phantom studies, the signalto-background ratio of REFI was significantly larger than that of   The activatable imaging strategy of external excitation in FMI is advantageous in terms of molecular specificity. Conversely, radioactive probes are constantly emitting CL during decay regardless of their interactions with biological molecules. This excludes CLI from the benefits of biologically specific, activatable imaging. However, REFI incorporates g-radiation and CL to excite EO internally for fluorescent emission. This strategy conferred the merit of FMI for activatable imaging to REFI. For in vitro phantom imaging ( Fig. 5a-h), the activated REFI signal was distinguished from conventional CLI by either optical signal enhancement (Fig. 5c,d, Po0.001) or 620-nm filtering (Fig. 5e,f, Po0.001). For in vivo phantom imaging ( Fig. 5i-n), PET showed no significant differences (P40.05) between the two implanted capillary tubes (Fig. 5i), which indicated that the two tubes were filled with the same dose of 18 F-FDG (50 mCi). However, the signal intensity was significantly different between CLI and REFI with no filtering (Fig. 4j, Po0.05) or filtering (Fig. 4k, Po0.001).
In vivo validation of the superiorities of REFI. EO nanoparticles were directly injected into the tumours of Bcap-37 (human breast cancer cell) xenograft mice to simulate tumour-targeted nanoparticle delivery. Ten hours later, 18 F-FDG was tail-vein injected. Figure 6a shows a comparison of REFI, CLI, and FMI in vivo. All three imaging techniques successfully detected optical signals from tumours. However, REFI exhibited both high optical intensity (REFI versus CLI: 8.43 ± 1.35 versus 3.69 ± 0.84, unit: 10 5 p s À 1 , Po0.01) and high signal-to-background ratio (REFI versus FMI: 1.74 ± 0.17 versus 0.94 ± 0.14, Po0.01). Therefore, REFI offered the best tumour to normal tissue contrast in all cases. The superiorities of REFI were consistent with the results of the previous phantom studies. We also performed a comparison between REFI and CLI, in which EO was tumour injected after the tail-vein injection of 18 F-FDG. The signal enhancement effect was significant too (Po0.01, Supplementary Fig. 2).
REFI and CLI were compared using U87MG (human gliomablastoma cell) and Bcap-37 xenograft mice models to verify the passive accumulation of EO nanoparticles in tumours via enhanced permeability and retention (EPR) effect after intravenous (i.v.) injection.
For in vivo longitudinal imaging of Bcap-37 xenografts, the consistent phenomenon was observed. For both tumour and bladder, the optical signal of REFI was significantly stronger than that of CLI at all time points (Fig. 6c). This indicated that EO particles may partially excrete through kidney and urinary a b c d   We also performed the comparison between FMI using ICG (indocyanine green) and REFI using 11 C-CHO (Choline) with EO through HepG2 (human hepatic cancer cell) orthotopic liver tumour mice models. The results further proved that REFI offered much better tumour to normal tissue contrast than FMI did (P o 0.01, Supplementary Fig. 3).
Multimodality comparison for tumour detection. To investigate the advantages of REFI for sensitive imaging of tumour lesions, 4T1-luc2 (luciferase-expressing mouse adenocarcinoma cell) xenografts were used for multimodality imaging comparison. PET, CLI, FMI and REFI were applied for two scenarios, and REFI offered the best imaging performance in both cases. The same dose of 18 F-FDG was i.v. injected into the xenografts for PET, CLI and REFI. All quantifications of optical intensity and signal-to-background ratio are listed in Figs 7 and 8, and Supplementary Table 2.
Six days after subcutaneous injection of 4T1-luc2 tumour cells, the tumour lesions were clearly visible on the upper and lower back of each mouse (Fig. 7a) with average diameter of  (Fig. 7b white arrows). This was because the brown adipose tissue (high 18 F-FDG uptake) close to the upper tumour influenced its uptake of the radiotracer. CLI confirmed this phenomenon (Fig. 7c, right mouse). Optical signal were detected from brown adipose tissue and lower tumour lesion, but upper tumour was still not visualized. However, with the presence of EO and 18 F-FDG in the same tumour lesion, the internal radiopharmaceutical excitation was activated. Significant optical signal enhancement (REFI versus CLI: upper tumour: Po0.01, lower tumour: Po0.01) was detected in both tumours for REFI (Fig. 7c, left mouse). After 620-nm filtering, most of the Cerenkov luminescence was blocked (Fig. 7d, right mouse), including the optical signal from the brown adipose tissue (Fig. 7d, left and right mice). However, REFI offered significant optical signal for upper (REFI versus CLI: Po0.001) and lower (REFI versus CLI: Po0.05) tumour lesions (Fig. 7d, left mouse).
In previous phantom imaging and in vivo imaging studies, the FMI showed relatively worse performance partially due to the low quantum yield of EO. To make the comparison fair and thorough, the quantum dot 620 (QD620) was used in this study. It is 'untargeted', same as EO, but with absolute quantum yield of more than 80% (more than twice of EO). Besides that the dose of QD620 (0.1 ml, 10 mg ml À 1 ) was 10 times higher than the dose of EO (0.1 ml, 1 mg ml À 1 ) for in vivo imaging. However, REFI still demonstrated significant better signal-to-background ratio with the same excitation and emission setting up (REFI versus FMI: upper tumour: 2.74 ± 0.59 versus 0.55 ± 0.02, Po0.05, lower tumour: 3.07±0.63 versus 0.43±0.05, Po0.05, Fig. 7e).
Sixty-five hours after subcutaneous injection of 4T1-luc2 tumour cells, the tumour size was only 2.1±0.3 mm (Fig. 8a,  red arrow). BLI visualized the tumour lesion (Fig. 8b), but PET offered a negative scan (Fig. 8c). This insufficient sensitivity of PET is probably due to the partial volume effects that caused significant underestimation of radiotracer concentration in small lesions [29][30][31] . CLI confirmed this again (Fig. 8d,e, black arrows). It was really difficult to detect the week Cerenkov signal from such a small lesion. However, REFI successfully achieved a positive imaging in both no filtering and 620-nm filtering modes (Fig. 8d,e red arrows, REFI versus CLI: both Po0.01).
To further validate above findings that REFI was more sensitive than PET for small tumour detection, another in vitro phantom and in vivo breast cancer mice model studies were conducted ( Supplementary Figs 4 and 5). The results also proved that REFI was indeed more sensitive in small tumour detection.
Biodistribution of EO and toxicity evaluation. Cytotoxicity assay of EO was performed. Human umbilical cord mesenchymal stem cell morphology after 24 h of incubation with 50 and 400 mg ml À 1 EO was captured using fluorescence microscopy (Fig. 9a). Neither significant change of cell morphology nor cell aggregation was observed in the experimental sample compared with that of the control sample (Fig. 9a) without nanoparticles.
The biodistribution of EO was measured 40 min after the tail-vein injection of EO (0.1 ml, 1 mg ml À 1 ). Vital organs, such as heart, kidney, liver, lung and spleen, were ground using tissue-grinding pestles. Each sample (0.1 ml), including the sample of blood and urine, was mixed with 18 F-FDG (0.1 ml, 100 mCi) respectively, so that the REFI signal was induced. By comparing the optical intensity with the control samples obtained from mice without the injection of EO (pure CL signal), the signal difference reflected the biodistribution of EO per 0.1 ml (EO concentration). The results showed the spleen and heart had the highest EO concentration. Blood, liver, kidney and urine were very similar, and lung had the lowest EO concentration (Fig. 9b). This proved that EO was partially accumulated in liver, and partially excreted in urine through kidney. Haematoxylin and eosin (H&E) microscopy ( Fig. 9c) showed no obvious structural change in kidneys, lungs, spleen, liver, heart or tumour of the experiment group with EO injection. Compared with QD620, the EO nanoparticles were much less toxic, as all mice injected with QD620 died within the next 4 h, but all mice injected with EO lived more than 4 weeks except killed mice.

Discussion
This study establishes REFI as a novel imaging strategy. Using EO nanoparticles and clinical radiopharmaceuticals, the g-(major) and Cerenkov (minor) radiation can be converted to fluorescence to achieve internally excited optical imaging. Our series of in vitro studies identified the excitation source and illustrated the relationship between the emission intensity and various parameters, such as radioactivity, EO mass and excitation distance. The excitation efficiency of different radiotracers and X-ray photon energies was investigated. The optical signal enhancement, spectra red shift and signal tissue penetration were also demonstrated.
REFI utilized both gand Cerenkov radiation from radioactive tracers to achieve internal activatable imaging. This unique feature eliminates the adverse effects of autofluorescence and reflection from external excitation as occurs in conventional fluorescence molecular imaging and enhances emission signals through fluorescent conversion of g-radiation to overcome the challenge of detecting a weak Cerenkov signal from radioisotopes. Comparison of REFI, CLI and FMI through phantom in vitro and Bcap-37, U87MG and 4T1-luc2 xenograft in vivo studies clearly demonstrated that REFI combined the merits of CLI and FMI with superb signal-to-background ratio and internal activatable imaging ability. Especially in the 4T1-luc2 xenograft study of small tumour detection, REFI showed significant better signal-tobackground ratio than FMI did, no matter it was QD620 (higher quantum yield) or RJ2-DG750 (better targeting specificity) that was applied in FMI. Furthermore, in comparison of CLI, the optical signal of REFI was boosted remarkably and was spectrally shifted towards the deeper tissue-penetrating red range. All these features benefited the optical imaging performance in our in vitro and in vivo studies.
REFI employed both radiotracer and EO nanoparticles, and the internal signal activation highly relies on the distance of them. Therefore, if tumour lesion shows similar uptake of one tracer but ARTICLE higher uptake of the other, in comparison with surrounding normal tissues, the tumour to normal tissue contrast will stand out. Our dual tumour 4T1-luc2 xenograft study demonstrated this phenomenon through multimodality comparison of REFI, PET and CLI. The uptake of 18 F-FDG in upper back tumour was influenced by brown adipose tissue nearby, but REFI was able to visualize the tumour unaffectedly, whereas PET and CLI failed. The sensitivity of PET is superior to many other imaging modalities, but it is still limited by the low resolution partially due to the partial volume effects [29][30][31] . For small tumour lesion with the size smaller or close to its spatial resolution, the underestimation of radiotracer uptake becomes significant. The high superficial resolution of optical imaging, the signal enhancement effect and the better signal-to-background ratio empowered REFI to detect small subcutaneous tumour lesions more effectively. Our 4T1-luc2 xenograft study indicated that REFI was able to detect tumour lesions with the size o2 mm, which were o3 days after tumour transplanting in nude mice. This suggested a great potential of applying REFI for highly sensitive early tumour detection and tumour metastasis imaging, even with passive delivery of the nanoparticle. Different from bioluminescence imaging, REFI does not require incorporation of immunogenic proteins 32,33 , such as luciferase and GFP (green fluorescent protein). It thus can be applied to a wider range of animal tumour models and holds better clinical translation potential.
Other studies of Cerenkov-induced fluorescence imaging (SCIFI) have already demonstrated the mechanism of utilizing Cerenkov light solely for excitation and applications of SCIFI in tumour marker detection using targetted fluorescent probes 19,22 . The fundamental difference between REFI and SCIFI (or other similar approaches) is the utilization of g-radiation for optical excitation and the resulted signal enhancement, as the overall optical signal of SCIFI was even weaker than Cerenkov light because of the inevitable energy loss during excitation. There were also studies of using radioluminescent nanoparticles and radiotracers, such as 18 F-FDG, for biomedical imaging applications 23,[25][26][27] , but these works remained in phantom studies and did not reveal the dual-radiation excited imaging mechanism. To the best of our knowledge, this is the first in vivo small animal tumour model study using REFI or other similar approaches.
The EO nanoparticle is not the only choice for applying the strategy of REFI. The lanthanide-doped nanoparticle is likely to provide similar fluorescence conversion of both gand Cerenkov radiation from radioisotopes 26,27 . These can be excited by collimated X-rays (approximately the same energy magnitude as g-radiation from 99m Tc-MDP) and have already been applied in X-ray/optical imaging modalities, such as X-ray luminescence computed tomography 27,34,35 . Therefore, our imaging technique converts external high-energy (X-ray) and low-energy (optical light) electromagnetic excitation to internal gand CL excitation via radiopharmaceuticals for in vivo optical molecular imaging.
We believe that REFI and its principle hold great promise for in vivo activatable imaging that can detect molecular targets or events and provide quantitative information of pathological processes on molecular levels with proper modification of the EO or other lanthanide-doped nanoparticle to overcome the issue of target specificity. REFI, with its intrinsic superiorities, will benefit the imaging of tumour-to-tumour molecular heterogeneity, which can further facilitate the development of precision medicine and personalized patient care. Therefore, this imaging technique will expand the applicability of activatable nuclear and optical molecular imaging.
In conclusion, through utilizing EO nanoparticles mediators, we have achieved internal conversion of continuous radiopharmaceutical radiation to activatable fluorescence for molecular imaging. REFI can merge the advantages of nuclear and optical molecular imaging techniques in vivo. This creates a strong motivation for further modification of the EO nanoparticle to obtain biomarker specificity and application of this highly sensitive imaging technique to early and small tumour detection in the near future.

Methods
Nanoparticles and radionuclide tracers. The EO nanoparticle (Eu2O3, 99.9% metal basis, molecular weight ¼ 351.91) was purchased from the Aladdin Chemistry Co. Ltd. 18 F-FDG, 99m Tc-MDP and 131 I-NaI was provided by the department of nuclear medicine, Chinese PLA General Hospital, Beijing, China. The QD620 was purchased from China Beijing Beida Jubang Science & Technology Co. Ltd, and the RJ2-DG750 probe was purchased from PerkinElmer. The absolute quantum yield of EO and QD620 was measured using FLS980 fluorescence spectrometer (Edinburgh Instruments). Both kinds of nanoparticle were dissolved in PBS for measurements.
SEM and spectrophotometry. The size and morphology of EO nanoparticles were determined by SEM (Hitachi S-4700). As-prepared EO powder samples were dispersed and dropped on a copper grid for scanning. The particle size was quantified using Image J. The EO nanoparticles were characterized for fluorescent properties using an EnSpire Multimode Plate Readers (PerkinElmer). The samples were read using a transparent 96-well plate. The excitation profile was obtained using a 620-nm emission filtre. The fluorescence profiles were obtained with excitation at 400 and 535 nm.
Mechanistic study of radiopharmaceutical excitation. The imaging system used for these studies was the IVIS Spectrum system (Caliper Life Sciences). The imaging parameters were binning: 4, exposure: 5 min and aperture: f1, unless otherwise indicated.
In the identification of the excitation source experiment, two 1.5-ml tubes filled with 0.2 g EO and 600 mCi of 18 F-FDG were placed 15 mm apart inside the imaging chamber and a sequence of images were acquired. The first image was taken in the normal view. The second was taken with a black cardboard box covering the 18 F-FDG tube. The third and fourth images were taken with an aluminium foil and a lead torus placed between the two tubes, respectively. Finally, a normal view of the two tubes was acquired again. Images of tubes with 990 mCi of 99m Tc-MDP and 0.20 g of EO were acquired following the same protocol.
In the emission intensity versus radioactivity experiment, EO (10 mg) was excited using 0.1 ml of 18 F-FDG and 0.1 ml of 99m Tc-MDP with 11 different activities (2,5,10,19,30,63,140,240, 521, 1,010 and 1,990 mCi). The images were analysed quantitatively to obtain the relationship between the emission intensity and radioactivity. The excitation distance was 10 mm.
In the emission intensity versus EO mass experiment, 18 F-FDG and 99m Tc-MDP of 100 mCi of were used to excite EO in eight different amounts (0.0012, 0.0052, 0.0099, 0.0511, 0.1082, 0.1501, 0.2042 and 0.2505 g). The excitation distance was 10 mm.
In the emission intensity versus excitation distance experiment, the excitation distance was set to 1,3,5,7,10,15,20,25,43 and 60 mm, respectively. The radioactivity of 18 F-FDG and 99m Tc-MDP was 250 mCi in each acquisition, and 10 mg of EO was used.
For the assessment of g and Cerenkov excitation fraction, a pair of mirrors was set next to the two EP tubes containing 18 F-FDG (100 mCi) and EO (10 mg), respectively. The tips of two EP tubes were 12 mm apart. After taking optical images, a lead partition (12-mm thick) was put in between the EP tubes to block g-radiations from 18 F. However, the CL was partially reflected to the EO because of the two mirrors reflection. Then optical images were taken again for pure CL excitation.
Emission spectra assessment. 18 F-FDG, 99m Tc-MDP and 131 I-NaI (each 140 mCi) were mixed with EO nanoparticles (10 mg), respectively. Their emission spectra together with the Cerenkov luminescent spectra of 18 F and 131 I (both 140 mCi) were measured using the IVIS system (exposure: 2 min). PBS was used as the Control.
X-ray excitation of EO nanoparticles. During the X-ray excitation process, the voltage of the X-ray tube (Hamamatsu L9181-02 Microfocus X-ray Source, Japan) was increased gradually, while the tube current was kept constant. Therefore, X-ray photons with different emission energies (40, 60, 80, 100, 110, 120 and 130 keV) were obtained to excite EO nanoparticles (2 mg) inside an EP tube ( Supplementary  Fig. 1b), but the total photon number of the X-ray beam was kept constant for each excitation. To protect the EMCCD (DU888 þ , Andor, UK), the optical imaging system was perpendicular to the X-ray tube (Supplementary Fig. 1a). The white light image was acquired with a 0.1-s exposure inside room light, and the X-rayexcited fluorescent image was acquired with 5-s exposure inside a light sealed environment. The PBS was used as control.
Excitation efficiency of different radiopharmaceuticals. 18 F-FDG, 99m Tc-MDP and 131 I-NaI (each 100 mCi) were used to excite 10 mg EO, respectively (mixing excitation). The optical images were acquired using the IVIS Spectrum system (Caliper Life Sciences) with binning: 4, exposure: 2 min and aperture: f1. The excitation efficiencies of using different radiotracers were compared ( Supplementary Fig. 1d).
Biological tissue penetration assessment. The biological tissue penetration ability of the fluorescence emission was evaluated using a piece of 1-mm thick porcine gastric mucosa tissue dissected from a freshly butchered porcine stomach. 18 F-FDG of 50 mCi was injected into one well as the control. The other four wells were injected with the mixture of 18 F-FDG (50 mCi) and EO (0.15, 0.30, 0.45 or 0.60 mg). Images without and with tissue blocking were taken (binning: 8, exposure: 2 min).
Comparison of optical imaging techniques via phantoms. Two identical tissuemimicking phantoms were made from high-density polyethylene to simulate the optical properties of mouse muscle 36 . Each was a cube that was 40 Â 40 Â 40 mm 3 in dimension. A small hole (diameter 2.5 mm, length 20 mm) was drilled 2 mm beneath the surface (Fig. 4a). A 0.1-ml mixture of 18 F-FDG (100 mCi) and EO (1 mg) or 0.1 ml 18 F-FDG (100 mCi) alone was injected into each hole. CLI, REFI and FMI were performed for comparison. For CLI and REFI, both 620-nm filtered and unfiltered images were acquired (exposure 5 min). For FMI, the excitation and emission wavelengths were 465 and 620 nm (exposure 1 s). The experiment was performed in triplicate.
Two glass tubes were filled with 0.1 ml of 18 F-FDG (50 mCi) or a 0.1-ml mixture of 18 F-FDG (50 mCi) and EO (0.15 mg). They were then subcutaneously implanted into a euthanized nude mouse. PET (Genisys PET, SofieBiosciences), CLI, REFI and FMI were performed. The experiment was performed in triplicate.
Animal experiments. All animal experiments were conducted in compliance with the guidelines of the Institutional Animal Care and Use Committee of General Hospital of Chinese People's Armed Police Forces. All animal procedures were performed isoflurane gas anaesthesia (3% isoflurane-air mixture), and all efforts were made to minimize suffering. The Balb/c nude mice were obtained from the Laboratory Animal Center of the Chinese Academy of Medical Sciences. Seven-to eight-week-old mice were used for experiments. Female mice were applied for Bcap-37 and 4T1-luc2 studies, and male mice were applied for U87MG and HepG2 studies. The subcutaneous tumour and breast cancer mice models were established by subcutaneously injecting 5 Â 10 6 tumour cells in nude mice. The orthotopic liver tumour mice models were established by performing a laparotomy in mice under isoflurane gas anaesthesia and injecting 5 Â 10 6 HepG2 cells into the liver.
In vivo validation of REFI. In the experiments of the intratumoural administration route, the control and experimental groups (three Bcap-37 xenografts per group) were tail-vein injected with 800 mCi of 18 F-FDG. The tumour tissue of the experimental group received direct injection of 0.05 ml of EO solution (diluted in normal saline, 1 mg ml À 1 ) 10 h prior to tail-vein injection. Forty minutes after the injection of 18 F-FDG, CLI and REFI were performed for both groups (exposure 5 min) and the FMI was performed for the experiment group (exposure 2 s). 18 F-FDG (100 mCi) was tail-vein injected into three 4T1-luc2 xenografts. CLI was performed 40 min after the injection (binning 4, exposure 5 min and aperture f1). Then, the EO solution (25 ml, 1 mg ml À 1 ) was locally injected into the tumour lesions, and REFI was taken immediately (Supplementary Fig. 2).
In the experiments of i.v. administration, EO solution (0.1 ml, 1 mg ml À 1 ) was tail-vein injected into three U87MG-xenografted mice as the experiment group. Twenty-four hours later, both control and experimental groups received injections of 0.1 ml of 18 F-FDG (500 mCi). After another 40 min, CLI and REFI were conducted with 620-nm filtering and no filtering (exposure 5 min).
EO solution (0.1 ml, 1 mg ml À 1 ) and 0.1 ml of 18 F-FDG (480 mCi) were mixed and tail-vein injected into three Bcap-37 xenografted mice. The control group only received the injection of 18  In vivo comparison between REFI and FMI using ICG. Three HepG2 liver tumour mice models were used to compare the FMI and REFI. Volume of 0.1 ml ICG solution (1 mg ml À 1 ) was tail-vein injected into the mice. Twenty-four hours later, EO (0.1 ml, 1 mg ml À 1 ) and 11 C-CHO (250 mCi) was tail-vein injected into the same mice. Forty minutes later, the abdominal cavity of the mice was surgically opened to expose the liver, and REFI was performed with binning 4, exposure 5 min, 620-nm filtering ( Supplementary Fig. 3b). Then, FMI was performed with excitation 795 nm and emission 835 nm (binning 4, exposure 2 s, Supplementary Fig. 3a).
In vitro phantom comparison between REFI and PET. A glass bottle was filled with 12 ml 18 F-FDG with the concentration of 3.75 mCi ml À 1 . Then, a glass capillary 1 mm in diameter was inserted inside. The tip of the capillary was filled with 0.5 ml EO (1 mg ml À 1 ) and 0.5 ml 18 F-FDG (4.17 mCi ml À 1 ). The capillary tip was used to simulate a very small tumour lesion with a volume of 1 mm 3 , and the glass bottle was used to simulate a mouse body. The small lesion had a slight higher uptake of 18 F-FDG and EO than its surroundings. Then, the phantom was imaged using REFI (binning 4, exposure 5 min, no filtering) and PET respectively ( Supplementary Fig. 4).
In vivo multimodality comparison. EO solution (0.1 ml, 1 mg ml À 1 ) and 0.1 ml of 18 F-FDG (280 mCi) were mixed and tail-vein injected into three 4T1-luc2 dual tumour mice models. The mice received subcutaneous injection of 4T1-luc2 tumour cells 6 days ago. Volume of 0.1 ml of 18 F-FDG (280 mCi) was injected into another three identical models. For each mouse, PET was scanned 35 min after injection with 10 min data acquisition, and then REFI and CLI were immediately performed with no filtering and 620 nm filtering (exposure 5 min). For FMI, three identical models received i.v. injection of QD620 (0.1 ml, 10 mg ml À 1 ) and were scanned 50 min later, using system with 465 nm excitation and 620 nm emission (binning, 4 and exposure, 2 s).
In the experiments of in vivo small tumour lesion imaging, for PET, CLI and REFI, the protocols were exactly the same as the protocols of the dual tumour mice models imaging study, except that all tumour mice models were imaged 65 h after subcutaneous injection of 4T1-luc2 cells into the left lower abdominal mammary fat pad. For FMI, same protocol was applied for using QD620. However, for targetted fluorescence imaging, 0.1 ml (100 nmol ml À 1 ) RJ2-DG750 was i.v. injected into the mice models. The mice were scanned 2 h later with 745-nm excitation and 800-nm emission (binning, 4 and exposure, 2 s).
Six 4T1-luc2 orthotopic breast cancer mice models were established to perform the comparison. Thirty-six hours after tumour transplantation, BLI was applied to verify the location of the tumour. Then PET, CLI and REFI were performed to detect the tumour. Sixty-five hours after tumour transplantation, the same procedure was applied again ( Supplementary Fig. 5). For PET, CLI and REFI, the tail-vein injection of 100 mCi 18 F-FDG was applied. For REFI (three of the six mice models), the tail-vein injection of 0.1 ml EO (1 mg ml À 1 ) was applied.
Cytotoxicity assay. In vitro cytotoxicity was measured by performing MTT assays on human umbilical cord mesenchymal stem cells. Different concentrations of EO (0, 50, 200 and 400 mg ml À 1 , diluted in PBS) were added to the wells. The samples were observed 24 h later using a conventional fluorescence microscope (Olympus IX71).