In vivo metallophilic self-assembly of a light-activated anticancer drug

Self-assembling molecular drugs combine the easy preparation typical of small-molecule chemotherapy and the tumour-targeting properties of drug–nanoparticle conjugates. However, they require a supramolecular interaction that survives the complex environment of a living animal. Here we report that the metallophilic interaction between cyclometalated palladium complexes generates supramolecular nanostructures in living mice that have a long circulation time (over 12 h) and efficient tumour accumulation rate (up to 10.2% of the injected dose per gram) in a skin melanoma tumour model. Green light activation leads to efficient tumour destruction due to the type I photodynamic effect generated by the self-assembled palladium complexes, as demonstrated in vitro by an up to 96-fold cytotoxicity increase upon irradiation. This work demonstrates that metallophilic interactions are well suited to generating stable supramolecular nanotherapeutics in vivo with exceptional tumour-targeting properties.

Self-assembling molecular drugs combine the easy preparation typical of small-molecule chemotherapy and the tumour-targeting properties of drug-nanoparticle conjugates. However, they require a supramolecular interaction that survives the complex environment of a living animal.
Here we report that the metallophilic interaction between cyclometalated palladium complexes generates supramolecular nanostructures in living mice that have a long circulation time (over 12 h) and efficient tumour accumulation rate (up to 10.2% of the injected dose per gram) in a skin melanoma tumour model. Green light activation leads to efficient tumour destruction due to the type I photodynamic effect generated by the self-assembled palladium complexes, as demonstrated in vitro by an up to 96-fold cytotoxicity increase upon irradiation. This work demonstrates that metallophilic interactions are well suited to generating stable s up ramo le cular n an otherapeutics in vivo with exceptional tumour-targeting properties.
Curing cancer is one of the toughest challenges of modern medicine 1 , and chemotherapy occupies the front line of cancer treatment [2][3][4] . Many chemotherapy agents are well-defined small molecules that typically lead to nonspecific delivery, rapid blood clearance and low accumulation in tumours, altogether generating severe side effects for patients with cancer 5,6 . To overcome these limitations, they have been conjugated to tumour-targeting nanocarriers, either covalently or supramolecularly, which in principle enhances drug delivery to the tumour [7][8][9][10][11][12][13][14][15][16][17][18] . However, many nanocarriers show comparatively low drug-loading capacity (typically <20%) 19,20 , while the resulting tumour accumulation remains disappointingly low: recent studies showed that a median 0.7% of the administered nanoformulated drug dose ends up in solid tumours 20,21 . In addition, achieving the reproducible preparation of drug-nanoparticle conjugates is often challenging, which restricts the clinical applications of nanodrugs.
Drug self-delivery systems (DSDSs) may solve this issue. They consist of small-molecule drugs that self-assemble into nanostructures without the assistance of dedicated nanocarriers 22 . These systems combine the easy preparation of small molecules and the tumour-targeting properties of nanoconjugates, achieving high drug-loading efficiency 2 . They also require supramolecular forces that hold in the complex biological environment of a living animal. DSDSs proposed to date rely on a combination of hydrophobic forces, hydrogen bonding and/ or metal-ligand coordination. The so-called metallophilic interaction is another kind of supramolecular interaction occurring specifically between d 8 or d 10 metal centres. It is well known in optoelectronics and material science 23 for its ability to modify the photophysical and photochemical properties of metal compounds. It has also been proposed for biological applications, but only in vitro demonstrations have been made 24 . In this Article, we demonstrate with a light-activated DSDS Article https://doi.org/10.1038/s41557-023-01199-w formation of a precipitate (Fig. 1g). In tetrahydrofuran/H 2 O solutions, similar self-assembly was observed, albeit with slower polymerization rates and different morphologies (Supplementary Figs. 7 and 8). The two series of self-assembled nanoparticles showed different powder X-ray diffraction patterns that were also different from the pattern calculated from the crystal structure or obtained from a bulk solid sample of PdL ( Supplementary Fig. 9). This result indicated that the four types of materials are different polymorphs. Overall, PdL appeared to self-assemble readily in the presence of water, although the resulting self-assembly depended entirely on the detailed composition of the solvent it was dissolved in.
The self-assembly of PdL was then studied in a cell-growing medium called Opti-MEM complete that contained 2.5 vol% foetal calf serum. At 25 µM, aggregation immediately occurred, as shown by a hydrodynamic diameter of approximately 164 nm determined by dynamic light scattering (DLS; Fig. 2a). After 30 min, the maximum hydrodynamic diameter had only slightly shifted to 190 nm, but the number of particles had increased significantly (Fig. 2b). The absorption of the solution (Fig. 2c) showed a gradual baseline increase during the first 2 h, which is characteristic of light scattering by nanoparticles, to remain constant over 24 h. The main nanostructures observed by TEM in the medium were nanodots (Fig. 2d), but these nanodots self-assembled as regular nanofibres that gradually lengthened. Cryogenic electron microscopy (Cryo-EM) imaging confirmed the formation of nanofibres in such medium, characterized by a well-ordered structure at a repeating distance of ~1.68 nm in the Fourier transform image (Fig. 2e). As for the monomer in DMSO, in medium the photoluminescence lifetime of aggregates of PdL was dramatically quenched by O 2 from 83.5 to 0.058 µs ( Supplementary  Fig. 5g,h), suggesting that the aggregates reacted even faster with O 2 than molecular PdL (in DMSO). Also, the different shapes of the nanorods in Fig. 2d compared with those in DMSO/H 2 O (Fig. 1f) or tetrahydrofuran/H 2 O mixtures ( Supplementary Fig. 7d) suggested that nanorod formation may involve proteins in the medium. Hence, we determined the protein content of the nanoaggregates formed in cell medium via protein gel and the commercial Pierce BCA Protein Assay Kit. The results ( Supplementary Fig. 10) indicated that the amount of protein involved in the stabilization of the nanoparticles in medium was very low, and that the different shapes observed under such conditions must be due to other factors, such as compound concentration, solvent polarity, pH, ionic force, the presence of salts, viscosity, and so on. Overall, in cell-growing medium, DLS, electron microscopy, ultraviolet-visible spectroscopy and protein determination demonstrated the time-dependent self-assembly of PdL into nanorods and nanoparticles, which, considering DFT and crystal structure analysis, must at least in part be a result of the metallophilic Pd⋯Pd interaction.
As the next step, the influence of self-assembly on the photochemical properties of PdL was considered. Photodynamic effects may occur either via a type I mechanism (electron transfer) or a type II (energy transfer) mechanism 29 . Direct detection of the near-infrared emission peak of 1 O 2 at 1,270 nm under blue light irradiation (450 nm) was only possible in CD 3 OD and hence for the PdL monomer. The corresponding 1 O 2 generation quantum yield was very low (φ Δ = 0.09; Fig. 2f and Supplementary Table 6). In Opti-MEM medium, and hence for the self-assembled form of PdL (25 µM), indirect 1 O 2 detection using the chemoselective chemical probe 9,10-anthracenediyl-bis(methy lene)dimalonic acid showed no decrease in the absorbance band at 378 nm upon green light irradiation, characteristic of the 1 O 2 adduct (Fig. 2g) 30

Synthesis and characterization of a light-sensitive DSDS
As in the padeliporfin photosensitizer recently approved for clinical photodynamic therapy (PDT) 25 , the PdL small molecule (Fig. 1a) contains a palladium(ii) metal centre. In contrast with padeliporfin, however, PdL is a bis-cyclometalated palladium compound characterized by the presence of two Pd-C covalent bonds (see Fig. 1b and full characterization in Supplementary Figs. 3 and 4 and Supplementary  Table 1). Its X-ray structure (Fig. 1b) shows head-to-tail dimers with a short interplanar distance of 3.4 Å and a short Pd⋯Pd distance of 3.518 Å, characteristic of metallophilic interactions. A density functional theory (DFT) model of the supramolecular dimer converged at a Pd⋯Pd distance of 3.26 Å (Fig. 1c), matching well with that observed in the crystal. The metallophilic interaction derives from the hybridization of both palladium d 4z 2 orbitals and π orbitals of the aromatic ligand in the highest occupied molecular orbital (HOMO) of the dimer. A dimerization energy of −40.0 kcal mol −1 was found for the dimer [PdL] 2 , while that of the ligand dimer [H 2 L] 2 (where H 2 L is (bis(3-(pyridin-2-yl) phenyl)amine)) was −33.7 kcal mol −1 . As the ligand H 2 L can only dimerize via π-π interactions, these calculations suggest that the overlap of the palladium d 4z 2 orbitals contributed ~16% to the supramolecular interaction between two PdL molecules in the dimer, while π-π stacking contributed ~84%. Time-dependent DFT (TDDFT) calculations at the same degree of theory further confirmed the decrease in the gap between the HOMO and the lowest unoccupied molecular orbital (LUMO) induced by supramolecular dimerization, with a bathochromic shift of the lowest-energy absorption band from 383 nm for the monomer to 502 nm for the dimer (Fig. 1d).
When dissolved in dimethyl sulfoxide (DMSO) in air, PdL showed a low emission quantum yield (φ p = 0.0016) with a short lifetime (τ = 0.295 µs) ( Fig. 1e and Supplementary Table 6). Under degassed conditions, the emission quantum yield and lifetime increased by 1-2 orders of magnitude (0.07 and 29.6 µs, respectively), demonstrating the triplet character of the excited state of PdL. The corresponding photoluminescence quenching efficiency was ~98%, which suggested that 3 PdL* efficiently interacts with molecular oxygen. Transient spectra under aerated and degassed conditions ( Supplementary Fig. 5) demonstrated that the initially formed singlet excited state 1 PdL* performs fast intersystem crossing to 3 PdL* (rate = 34.5 ns −1 ), which then reacts with O 2 via electron transfer to generate a PdL + cation radical. This result strongly suggested that PdL may engage in PDT type I reactivity to generate superoxide radicals (O 2  Table 7) 26,27 . In contrast, in a DMSO/H 2 O 1/9 mixture (100 µM), a rapid (<1 min) increase in the baseline and the generation of a new absorbance peak at 504 nm were observed (Fig. 1f), which are typical for metal-metal-to-ligand charge transfer excited states induced by Pd⋯Pd interactions 28 and altogether suggest efficient and fast self-assembly. This hypothesis was confirmed by transmission electron microscopy (TEM) images showing nanorods and nanocubes (Fig. 1f, insert). Usually, the formation of Pd⋯Pd supramolecular bonds is accompanied by a long-wavelength emission peak 28 , and indeed an increase in the H 2 O content in DMSO/H 2 O mixtures (f w = V water /V total is the ratio of the water volume V water and the total volume of the solution V total ) led to a gradual replacement of the monomeric emission peak at 564 nm (as observed in pure DMSO; f w = 0) by new emission maxima at 593 nm (f w = 0.5) and finally 610-670 nm (f w = 0.9) concomitant with the Article https://doi.org/10.1038/s41557-023-01199-w the presence of dihydroethidium (DHE)-a chemoselective chemical probe for superoxide-the oxidation product 2-hydroxyethidium was produced efficiently, as shown by its emission at 590-620 nm ( Fig. 2h and Supplementary Fig. 9) 33 . These results clearly demonstrated that PdL is capable of photochemically generating superoxide both in the monomeric and aggregated states, which suggested that as a light-activated DSDS it may behave as a PDT type I photosensitizer, generating its photodynamic effect by electron transfer.

Biological properties in vitro and in vivo
Considering the good absorption of PdL at 520 nm (ε = 915 M −1 cm −1 in DMSO) and its PDT type I properties, its cytotoxicity was evaluated first in vitro using two-dimensional (2D) monolayers of lung carcinoma (A549), epidermoid carcinoma (A431) and skin melanoma (A375) . 3b and Supplementary Fig. 11).
In the light-irradiated group, no toxicity was observed after 2 h, but after 4 h and 24 h, the numbers of apoptotic and necrotic cells had increased, suggesting that PdL induced cancer cell death 4 h after irradiation via both cell death mechanisms. The cytotoxicity of PdL in 3D multicellular tumour spheroid models (A549 and A375), which better mimic the physical penetration of light and drugs in three dimensions 34  To explain the phototoxicity observed in vitro, a few mechanistic studies were undertaken. First, the intracellular ROS (as demonstrated with the non-selective dichlorodihydrofluorescein diacetate (DCFDA) probe), and especially superoxide levels (as demonstrated with the DHE probe), were found to be significantly increased after light irradiation in the presence of PdL ( Supplementary Fig. 16), while the glutathione levels decreased (Supplementary Fig. 17). These results confirmed the conclusion of photophysical studies-that electron transfer (PDT type I) is probably the main source of the phototoxicity of PdL aggregates in cells. A cellular fractionation experiment was then realized to see where the palladium was mainly located (Supplementary Fig. 18); clearly, PdL accumulated in the membrane fraction (which includes cell, mitochondria and lysosome membranes) and the nuclei ( Supplementary  Fig. 18). This result was consistent with the idea that an endocytic uptake mechanism may be working for these aggregates; it also suggested that the DNA photocleavage properties of PdL should be tested. However, DNA agarose gel experiments at different concentrations (20-100 µM), irradiation times (5-60 min) and incubation times (1-24 h) after activation demonstrated that PdL neither interacted with DNA in the dark nor cleaved DNA upon irradiation ( Supplementary Fig. 19). Interestingly, after incubating PdL nanoparticles with pUC19 plasmid DNA (2 mg ml −1 ) at 37 °C in the dark or after green light irradiation, Article https://doi.org/10.1038/s41557-023-01199-w the PdL nanoparticles were found to be stable ( Supplementary  Fig. 20). Overall, these results suggested that the excellent phototoxicity of PdL may result more from increased ROS levels in membrane-rich organelles, such as the lysozome or mitochondria, than from DNA damage in the nucleus. Considering these promising in vitro results, further in vivo testing in mouse tumour models was undertaken.
Human skin melanoma is known to be prone to resisting PDT type II treatment by the combination of a hypoxic tumour microenvironment 35 and melanin-induced quenching of ROS 36 . Hence, PdL was evaluated in vivo using human skin melanoma (A375) tumour xenografts in nude mice. Following intravenous tail injection (100 µl; 420 µM in Dulbecco's modified Eagle's medium (DMEM) and 10% foetal bovine serum (FBS); 0.9 mg kg −1 ), the mice showed a constant body weight over 20 d (Fig. 3c) and the important organs remained healthy, as determined by haematoxylin and eosin staining ( Supplementary Fig. 21), suggesting low systemic toxicity at this compound dose. In the dark group, PdL showed moderate tumour growth inhibition, but green light irradiation (520 nm; 100 mW cm −2 ; 10 min; 60 J cm −2 ) performed 12 h after injection of the self-assembled PdL strongly inhibited tumour growth (Fig. 3d). Haematoxylin and eosin staining of the irradiated tumours on day 5 revealed that the tumour tissues were dramatically damaged in the PdL + light group, while the other groups did not show any remarkable effect. TUNEL staining also demonstrated a decrease in cancer cells in the irradiated tumour and cell killing via apoptosis (Extended Data Fig. 1b). Overall, these experiments demonstrated not only that PdL showed excellent antitumour efficacy in an A375 melanoma mouse model but also that it showed very low cytotoxicity to healthy organs, highlighting the high potential of PdL DSDSs for anticancer PDT application.

Uptake, biodistribution and tumour targeting
The low systemic dark toxicity and high antitumour PDT efficacy of PdL stimulated us to check drug uptake in vitro and in vivo using inductively coupled plasma mass spectrometry (ICP-MS). The cellular uptake of PdL (2 µM) was found to be time dependent, with the Pd content in A375 cells increasing from 29 ngPd per million cells at 2 h to 172 ngPd per million cells at 24 h (Fig. 4a). It was also temperature dependent, with a decrease to 23 ngPd per million cells at 4 °C 2 h after treatment (5 µM) compared with 41 ngPd per million cells at 37 °C (Fig. 4b). Further coincubation experiments (Fig. 4b) showed that active internalization occurred via clathrin-mediated endocytosis (pitstop) and micropinocytosis (wortmannin). Altogether, these results highlighted that both energy-independent and energy-dependent cellular uptake took place in vitro, suggesting that PdL may pass through the cell membrane as both isolated molecules and nanoaggregates.
An essential question at this stage was to understand whether the nanoparticles formed by PdL in cell-growing medium in vitro would also form in a living mouse. Thus, the presence and morphology of nanostructures in the bodies of mice injected with PdL were investigated in more detail. First, blood samples taken from the eye socket of mice 5 min after intravenous tail injection of PdL in DMEM solution (100 µl; 0.9 mg kg −1 ) showed roughly spherical, high-contrast nanoparticles characterized by an average size of 181 ± 75 nm ( Supplementary Fig. 22). Similar to those found in the injected DMEM solution, these nanoparticles were rich in palladium according to energy-dispersive X-ray (EDX) element mapping analysis (Extended Data Fig. 2a), which strongly suggested that they contained PdL. As proof of this concept, we also injected the maximum permitted volume (10 µl) of a pure, non-aggregated DMSO solution of PdL (4.2 mM) into a mouse to check whether molecules of PdL would self-assemble as nanoparticles upon introduction in the blood. Blood collection after 5 min, followed by TEM, EDX and scanning electron microscopy analysis (Extended Data Fig. 2b and Supplementary  Fig. 23), demonstrated that palladium-rich round nanoparticles with a diameter of ~100 nm had indeed formed, similar to the nanoparticles observed upon intravenous injection of PdL in DMEM. Altogether, these results suggested that molecules of PdL spontaneously aggregated into nanoparticles upon injection in the blood, where they remained self-assembled during circulation. The biological half-time of these PdL nanoparticles in the blood was found to be ~7.9 h ( Supplementary  Fig. 26), which is rather long and notably much longer than that found for many small-molecule anticancer drugs (typically below 1 h). At 12 h after tail vein injection of PdL, the A375 tumour was sectioned and imaged by electron microscopy. These images (Extended Data Figs. 2c; 1 and 0.5 µm scales; indicated by red arrows) showed dark nanosized spots in the lysozomes of the cancer cells, with an average diameter of 260 ± 75 nm (slightly larger than the diameter of nanoparticles in the blood). These dark spots were not observed in the untreated control group (Supplementary Fig. 24); thus, we interpreted them as palladium-containing nanoparticles. To verify this hypothesis, we performed EDX analysis of these nanoparticles (Supplementary Fig. 25), which confirmed that the cancer cells in the tumour had indeed taken up PdL nanoparticles. Overall, the presence of nanoparticles both in the blood and in the tumour tissue of mice treated with PdL is proof that the Pd⋯Pd interaction causing the self-assembly of the molecule in medium is strong enough to keep the nanostructures in circulating blood, which leads to delivery of the prodrug to the tumour.
To quantify tumour delivery, the biodistribution of Pd was determined by ICP-MS in A375 mouse xenografts several hours (2, 6, 12, 20 and 24 h) after intravenous tail injection of PdL in DMEM. As shown in Fig. 4c, the complex showed low accumulation (below 0.27 µg g −1 tissue) in the heart, kidney and lung, while the liver showed significantly higher accumulation (above 1.0 µg per per gram of tissue), as expected considering its role in the detoxification and metabolism of exogenous substances. Noticeably, the accumulation level of PdL in the liver gradually decreased over time. Meanwhile, the tumour tissue showed an increasing Pd accumulation from 0.17 µg per gram of tissue after 2 h to a peak of 0.87 µg per gram of tissue at 12 h, which corresponded to an impressive 10.2% of the injected dose per gram (% ID g −1 ) (Fig. 4c,d), and finally decreased to 0.17 µg per gram of tissue at 24 h. These results highlight that the long circulation time in the blood (~8 h) leads to an extraordinary tumour accumulation rate of the PdL nanoparticles, which peaks at 12 h. These data suggest that the enhanced permeability and retention effect, conjugated with the high drug loading of the nanoparticle, explains the high tumour accumulation of the compound, as the PdL nanoparticles do not contain any active tumour-targeting molecules. In conclusion, PdL appears to be a particularly well-performing DSDS characterized by easy synthesis and formulation in biocompatible buffer, high drug-loading efficiency of the self-assembled nanoparticles, low systemic toxicity to the tumour-bearing mouse for these nanoparticles and excellent tumour accumulation and antitumour efficacy upon light irradiation using a drug-to-light interval of 12 h.

Conclusion
One main advantage of DSDSs is their high drug-loading capacity, as the nanoparticles are mostly composed of drug molecules. However, with traditional DSDSs, such high drug-loading capacity may come with high toxicity to blood-filtering organs such as the liver or kidneys 37 , as each nanoparticle brings into a cell many toxic molecules. Light-activated DSDSs bring a solution to this problem, as the self-assembling drug only becomes toxic after light irradiation 38 . As the liver remains in the dark, high hepatic uptake is not a problem for PdL, provided liver clearance occurs, which is suggested in Fig. 4c. Light activation hence provides a dramatic advantage compared with traditional DSDSs.
Light-activated DSDSs can also be analysed from the point of view of PDT. Of course, PDT treatment offers a patient-friendly alternative to chemotherapy, with the potential to inhibit tumour proliferation while minimizing side effects by selective light irradiation of tumour tissue 39 . However, photosensitizer molecules taken up in healthy tissues also lead to undesired photosensitivity (for example, of the skin) for the patient-a typical side effect of PDT with Photofrin 40 . It is hence Article https://doi.org/10.1038/s41557-023-01199-w essential that the photosensitizer be delivered with high efficacy to the tumour tissue. Many reports have demonstrated that nanoconjugates enable an increase in the tumour accumulation of molecular drugs, including PDT photosensitizers 41,42 . However, the low average drug-loading capacity (typically 20 wt%) and tumour accumulation efficacy (a median of 0.7% ID) achieved by classical drug delivery nanosystems 20 suggested the urgency to develop new drug delivery nanoconjugates, also in the context of PDT.
The drug delivery properties of PdL include a long circulation time (>12 h) with good levels of accumulation in the tumour tissue (up to 10.2% ID g −1 ). Photochemically speaking, unlike porphyrins, the PdL complex maintains excellent photodynamic properties in the aggregated state, even in a hypoxic tumour microenvironment. With these results in hand, we conclude that the metallophilic interaction can potentially be used to build high-performance supramolecular nanocarriers with improved tumour accumulation and that the Pd⋯Pd interaction observed with complexes such as PdL can generate photodynamic systems that conserve their phototoxic properties under hypoxia.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41557-023-01199-w.

Ethics declarations
This study was conducted following the Guide for the Care and Use of Laboratory Animals 43 . All animal experiments were performed under guidelines approved by the ethics committee of the Dalian University of Technology. The maximal tumour size permitted was 2,000 mm 3 (and the maximum burden was 10% of the body weight), but this limit was never reached in the experiments described below; the maximum tumour size in the experiments described below was ~1,000 mm 3 .

General materials
All reagents were purchased from commercial vendors. The reactants and solvents were used without further purification. All 1 H NMR and 13 C attached proton test NMR results were obtained on Bruker DPX 300 spectrometers. Chemical shifts are indicated in ppm relative to the residual solvent peak. Electrospray ionization mass spectra (ESI-MS) were recorded using an MSQ Plus spectrometer in positive ionization mode. The TEM experiments were carried using a JEOL 1010 (10 kV) transmission electron microscope with a Formvar/carbon-coated copper grid from Polysciences. Ultraviolet-visible spectra were recorded on a Cary 50 spectrometer from Varian. The emission spectra were measured using an FLS900 spectrometer from Edinburgh Instruments. The singlet oxygen emission spectra were measured on a special custom-built setup that was described previously 31 . The spectra data were analysed using Origin 8. The human cancer cell lines A549 (lung carcinoma), A431 (skin carcinoma) and A375 (malignant melanoma) were distributed by the European Collection of Authenticated Cell Cultures and purchased from Sigma-Aldrich. DMEM (with and without phenol red and without glutamine), glutamine-S (200 mm), tris(hydroxymethyl)aminomethane (Tris base), trichloroacetic acid, glacial acetic acid and sulforhodamine B were purchased from Sigma-Aldrich. Opti-MEM Reduced Serum Medium without phenol red was obtained from Gibco. The measurements of complexes on photocytotoxicity were performed according to the literature 31 . The annexin V/propidium iodide double staining assay was purchased from Bio-Connect. The FractionPREP Cell Fractionation Kit was obtained from BioVision Incorporated. The cytotoxicity data were analysed with GraphPad 8. The confocal microscopy and electron microscopy images were analysed using ImageJ 1.52v.

Cryo-EM measurement
We applied 6 µl of the sample ([PdL] = 25 µM) to a freshly glow-discharged carbon 200 mesh Cu grid (Lacey Carbon Film; Electron Microscopy Sciences, Aurion). Grids were blotted after a 10-s wait for 3 s at 99% humidity in a Vitrobot plunge freezer (FEI Vitrobot Mark III; Thermo Fisher Scientific). Cryo-EM images were collected on a Titan Krios operating at 300 kV at a nominal magnification of 33,000× or 81,000×, yielding a pixel size at the specimen of 3.5 or 1.4 Å, respectively (The Netherlands Centre for Electron Nanoscopy, Leiden University).

In vivo tumour inhibition experiments
Female 3-week-old BALB/c mice were originally purchased from Vital River Laboratory Animal Center (Beijing, China). The mice were kept under specific pathogen-free conditions with free access to standard food and water for 2 weeks at 20-21 °C, under 40-60% relative humidity and with a 12 h dark/12 h light cycle. The mice were used for experiments when their weight reached ~20 g. This study was conducted following the Guide for the Care and Use of Laboratory Animals 43  The tumour volume (V) was calculated using formula V = L/2 × W 2 after measuring the tumour length (L) and width (W) 3 . The mice were then randomly divided into four groups (vehicle control, 520 nm light, PdL and PdL + 520 nm light, with four mice in each group). The injectable PdL solution was prepared by diluting the PdL stock DMSO solution (4.2 µM) to 420 µM using DMEM medium containing 10% vol/vol FBS and 1% vol/vol penicillin/streptomycin. The mice were treated through tail intravenous injection with DMEM (vehicle control and 520 nm light groups) or PdL (2.1 µmol kg −1 ; 420 µM; 100 µl DMEM medium (10% FBS); 0.9 mg kg −1 ) (PdL dark and PdL + 520 nm light groups). After 12 h injection, 520 nm irradiation (100 mW cm −2 ; 5 min) was then carried out twice, with an interval of 5 min, for the light groups. Thus,

Nature Chemistry
Article https://doi.org/10.1038/s41557-023-01199-w the total light dose for each treatment was 100 mW cm −2 for 10 min at 60 J cm −2 . These treatment and irradiation steps were replicated at days 0, 7 and 14, respectively. On day 5, one mouse in each group was killed and the tumour was taken up and fixed with paraformaldehyde (10% vol/vol), then sectioned into slices and analysed via haematoxylin and eosin or TUNEL protocols to evaluate the tumour cell damage and apoptosis conditions. The tumour volume and body weight of the remaining mice (n = 3) were measured and recorded and the average tumour volume and body weight were calculated over 20 d. Finally, these mice were also killed and the healthy organs were removed, fixed with paraformaldehyde (10% vol/vol) and then sectioned into slices and analysed via the haematoxylin and eosin protocol to determine the side effects of treatment.

Mouse tumour electron microscopy imaging experiments
One tumour-bearing mouse was treated with PdL (2.1 µmol kg −1 ; 420 µM; 100 µl DMEM medium (10% FBS); 0.9 mg kg −1 ) through intravenous tail injection. After 12 h, the mouse was sacrificed, the tumour tissue was collected and then fixed with a biological TEM fixation solution (Wuhan Servicebio). Next, the tumour tissue was split into small pieces with a volume of ~1 mm 3 and fixed again using 1% osmic acid phosphate buffer solution for 2 h, followed by dehydration with ethanol (vol/vol = 30, 50, 70, 80, 95 or 100%; 20 min per group) and acetone twice (15 min). The prepared samples were then treated with acetone/epon-812 embedding medium at a ratio of 1:1 for 2 h and 1:2 for 12 h, and then pure epon-812 solution for another 5 h at 37 °C. Next, the tissue-containing embedding medium was filled in the embedding mould for 24 h at 37 °C and another 48 h at 60 °C. The obtained tissue-containing resin was then sectioned into slices with a thickness of ~60-80 nm via ultramicrotome (Leica EM UC7) and moved to the copper grid (150 mesh). The obtained grids were stained with 2% uranyl acetate ethanol solution for 8 min and 2.6% lead citrate solution for another 8 min. Next, the grids were dried at room temperature and observed using a JEOL JEM-2100 transmission electron microscope ( Japan).

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
All of the relevant data gathered during this study have been deposited in the Zenodo repository (https://zenodo.org/record/7032728) or can be obtained from the corresponding authors upon request. Crystallographic data for the structure reported in this Article have been deposited at the Cambridge Crystallographic Data Centre under deposition number CCDC 2142714 (PdL). Source data are provided with this paper.
Corresponding author(s): Wen Sun, Sylvestre Bonnet Last updated by author(s): Dec 29, 2022 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy The data relevant with this study are shown in the manuscript and its supplementary Information, or are available in the general repositories Zenodo. Source data are provided with this paper.