Prodrug-inspired adenosine triphosphate–activatable celastrol-Fe(III) chelate for cancer therapy

Abstract

Celastrol (CEL), an active compound isolated from the root of Tripterygium wilfordii, exhibits broad anticancer activities. However, its poor stability, narrow therapeutic window and numerous adverse effects limit its applications in vivo. In this study, an adenosine triphosphate (ATP) activatable CEL-Fe(III) chelate was designed, synthesized, and then encapsulated with a reactive oxygen species (ROS)–responsive polymer to obtain CEL-Fe nanoparticles (CEL-Fe NPs). In normal tissues, CEL-Fe NPs maintain structural stability and exhibit reduced systemic toxicity, while at the tumor site, an ATP-ROS–rich tumor microenvironment, drug release is triggered by ROS, and antitumor potency is restored by competitive binding of ATP. This intelligent CEL delivery system improves the biosafety and bioavailability of CEL for cancer therapy. Such a CEL-metal chelate strategy not only mitigates the challenges associated with CEL but also opens avenues for the generation of CEL derivatives, thereby expanding the therapeutic potential of CEL in clinical settings.

An intelligent drug delivery system improves the biosafety and bioavailability of celastrol for cancer therapy.

INTRODUCTION

Celastrol (CEL), a quinone methide triterpene isolated from Tripterygium wilfordii Hook F (a vine commonly known as thunder god vine used in traditional Chinese medicine), is one of the valuable natural bioactive compounds that are most likely to be developed as a modern drug (1). Over the past few years, many preclinical studies have demonstrated that CEL inhibits tumor growth (2), promotes apoptosis (3, 4), suppresses metastasis (5, 6) and angiogenesis (7), and exhibits potent anticancer effects in multiple in vitro or in vivo models (8, 9). However, the clinical applications of CEL are still limited because of its poor water stability, low bioavailability, and off-target effects, which result in high systemic toxicity (10, 11), including hepatotoxicity (12, 13), cardiotoxicity (14), and hematopoietic system toxicity (15, 16). Although the exact underlying mechanisms are still under investigation, emerging evidence suggests that the ternary complex of (heat shock protein 90) Hsp90, co-chaperone, and client proteins plays a crucial role in various biological processes in both healthy and cancer cells, which may be involved in the pharmacological action and toxicological effect of CEL (17, 18). CEL can bind to the Cdc37 and Hsp90 interaction site, leading to the disruption of the Hsp90-Cdc37 binding and the proper protein folding of client proteins, such as epidermal growth factor receptor (EGFR), protein kinase B (Akt), cyclin-dependent kinase 4 (Cdk4), and tyrosine-protein kinase Met (c-Met), thereby suppressing growth and survival signaling in cancer cells (19, 20).

To achieve better clinical application of CEL with increased tumor selectivity and enhanced biosafety, several strategies have been developed, mainly including the development of its derivatives and nanoscale drug delivery systems (NDDSs). At present, modifications of CEL have mainly been focused on the C-20 carboxylic acid functionality, alteration at the A ring, and C-6-modification at the B ring (21, 22). For example, introducing additional electrons, hydrogen bonding groups, or urea groups at the C-20 carboxyl via esterification or amidation can potentially improve the pharmacological and antitumor properties (23, 24). However, complex organic reactions, laborious separation, and purification processes lead to low yields, and some derivatives may even have poor stability (25). In addition, the structure-effect relationship remains controversial (26). Nanotechnology also holds promise for improving the pharmacokinetics, biodistribution, and the side effect profile of CEL. Nanodrug delivery systems such as liposomes (27), polymeric micelles (6), and nanoparticles (NPs) (28) can be designed to respond to the tumor microenvironment (TME) for controlling the release of the payload. Furthermore, converting the drug into a TME-activatable prodrug can further reduce physiological toxicity caused by premature release (29, 30). TME stimuli that can trigger release or drug reactivation include different metabolite environments (31, 32) and altered levels of nutrients (33) and biomolecule [e.g., hydrogen peroxide (34), glutathione (35), and adenosine triphosphate (ATP) (36)]. As a ubiquitous biogenic biomolecule in living organisms, ATP plays an essential role in cellular energy metabolism processes. In particular, the concentration of ATP in the extracellular environment of tumors is 10³ to 10⁴ times higher than normal tissue levels (0.1 to 0.5 mM versus 10 to 100 nM). Meanwhile, the intracellular concentrations of ATP in tumors are further increased to 1 to 10 mM. Numerous studies have been conducted taking advantage of these concentration differences to develop innovative and efficient drug delivery platforms (37), enabling controlled release of small-molecule drugs (38) and gene therapy (39).

The hydroxyl and carbonyl groups on ring A of CEL are critical for forming hydrogen bonds with Hsp90/Cdc37 (23, 40), which not only result in antitumor activity but also can lead to physiological toxicity. The phenolic-like structure of CEL enables binding to metal ions like Fe(III), Zn(II), and Cu(II) (41, 42). Inspired by prodrug strategies (29, 30), we developed a CEL delivery system based on the coordination of CEL and Fe(III). The Fe(III) was selected for coordination stability and safety. As shown in Fig. 1A, Fe(III) coordinates with the C-2 carbonyl and C-3 hydroxyl groups of CEL to form an inactive CEL-Fe chelate prodrug (denoted as CEL-Fe hereafter). This coordination attenuates the binding ability of CEL with proteins (Hsp90/Cdc37), for reducing toxicity to normal tissues. In tumor tissues, high ATP concentrations can competitively displace CEL-Fe coordination, to reactivate CEL’s pharmacological activities. To further enhance pharmacokinetics and the tumor stimuli-response effect, CEL-Fe was encapsulated in polymeric micelles of polyethylene glycol–grafted polymer containing thioketal groups (denoted as TK-PEG) and polyethylenimine (PEI)–modified F127 polymer (denoted as F127-PEI). Such an encapsulation enables tumor accumulation through the enhanced permeability and retention (EPR) effect (Fig. 1B) (43). In response to high reactive oxygen species (ROS) levels, drug release behavior is triggered by the degradation of thioketal groups, leading to rapid drug activation in response to high concentrations of ATP. In summary, our system combines dual strategies of coordination detoxification and responsive release polymer encapsulation, thus effectively contributing to reducing the physiological toxicity of CEL and improving its bioavailability.

Fig. 1. Schematic illustration of the intelligent CEL delivery system.

(A) Schematic illustration of the formation and dissociation of the CEL-Fe(III) chelate (CEL-Fe). According to the corresponding structure-activity relationships, CEL-Fe chelate was proposed as a toxicity reduction strategy based on coordination bonds. The reduction in hydrogen bonds leads to a decrease in the binding affinity of CEL-Fe for its target proteins, resulting in reduced toxicity. ATP can be used as a competitive binding agent to Fe(III). When coordination bonds between CEL and Fe(III) are broken, the antitumor efficacy of CEL is activated. (B) Schematic illustration of the formation of CEL-Fe NPs and their responsiveness to ATP and ROS at the tumor site. This NP formulation enables tumor accumulation through the EPR effect and leads to drug release and drug activation triggered by ROS and ATP, respectively.

RESULTS

The formation of the CEL-Fe(III) chelate and ATP-triggered dissociation

We first investigated the coordination ability of CEL. The process in which a ligand donates a pair of electrons to a metal ion is referred to as LMCT (ligand-to-metal charge transfer), and it is usually characterized by an ultraviolet-visible (UV-vis) spectrum (44, 45). The addition of Fe(III) (1:1 molar ratio) to CEL changed the solution color from orange to dark green, and broad LMCT bands were observed in the UV-vis spectrum (Fig. 2, A and B, and movie S1), indicating the formation of coordination bonds between CEL and Fe(III). The absorbance of the CEL/Fe(III) mixture at 600 nm increased with higher Fe(III) ratios, plateauing at 1:1 (fig. S1), suggesting that CEL-Fe chelate was formed at a complex ratio of 1:1 (29). Furthermore, we explored the coordination versatility of CEL with various metals (CEL-M). All selected metal ions except Ca(II), Mg(II), Sr(II), and Pt(V) exhibited LMCT bands (fig. S2A) and color changes (fig. S2B), indicating that these metal ions generated the CEL-M coordination. However, Fe(III) was chosen because of its endogenous nature (46), relatively good biocompatibility (47), and high coordination stability (48, 49).

Fig. 2. Chemical verification and cytotoxic changes for CEL-Fe chelate formation and dissociation.

(A) Illustration of the CEL/CEL-Fe conversion and ATP-triggered drug activation. (B) UV-vis spectra of the CEL solution (0.4 mM) with increasing molar ratios of FeCl₃·6H₂O. (C) UV-vis spectra analysis of the CEL-Fe solution [dimethyl sulfoxide (DMSO)] in the presence of 1 mM ATP (300 rpm). X-ray photoelectron spectroscopy (XPS) spectrum of O 1s (D), Raman spectra (E), and Fourier transform infrared (FTIR) spectroscopy spectra (F) of CEL, CEL-Fe, and CEL-Fe + ATP. (G to L) Cytotoxicity comparison of free drugs in normal versus tumor cell lines after incubation with various concentrations of CEL, CEL-Fe, and ATP-treated CEL-Fe for 24 hours. (G) A549 (human lung adenocarcinoma cells) and (J) BEAS-2B (human lung epithelial cells) from the lung, (H) HepG2 (human hepatocellular carcinoma cells) and (K) L02 (human normal liver cells) from the liver, and (I) 786-0 (human renal cell carcinoma) and (L) 293T (human embryonic kidney cells) from the kidney. Data are expressed as means ± SD (n = 5). **P < 0.01 and ***P < 0.001, ordinary two-way analysis of variance (ANOVA). a.u., arbitrary units.

Subsequently, the disruption of CEL-Fe coordination by ATP was investigated. As reported in many studies, ATP has a high binding affinity for metal ions such as Fe(III) (50, 51) and Zn(II) through metal ion-triphosphate coordination (37, 52). Here, we attempted to use this property of ATP to dissociate the CEL-Fe coordination (Fig. 2A), thereby achieving drug activation at the tumor site. The addition of 1 mM ATP to the CEL-Fe solution led to rapid color recovery and the disappearance of the LMCT band in the UV-vis spectrum (Fig. 2, A and C, and movie S1), indicating the dissociation of the coordination bonds. Similar observations after treatment with the strong chelator EDTA (fig. S3) supported that ATP dissociates CEL-Fe coordination through competitive binding. Considering the intracellular and extracellular ATP concentrations in healthy versus tumor cells, we examined the CEL-Fe response across an ATP concentration range from 10 nM to 1 mM (fig. S4, A to F) (36, 53). The observations indicated that extracellular ATP levels in healthy cells (10 to 100 nM) are insufficient to trigger effective CEL activation.

Next, the formation of the CEL-Fe chelate and ATP-triggered dissociation were further characterized by x-ray photoelectron spectroscopy (XPS), Raman spectroscopy, and Fourier transform infrared (FTIR) spectroscopy. XPS confirmed Fe(III) presence via peaks at 712.21 and 725.32 eV, which were attributed to 2p2/3 and 2p1/3 of Fe(III), respectively (fig. S5A) (54, 55). In addition, XPS analysis revealed an increased O 1s binding energy (BE) in CEL-Fe, indicating Fe-O coordination (56). However, this coordination was disrupted by ATP, as evidenced by the decreased O 1s BE (Fig. 2D and table S1). The Raman spectra showed the disappearance of CEL hydroxyl stretches at ~3008 cm⁻¹ and the emergence of Fe-O stretches at 547 to 566 cm⁻¹ in CEL-Fe (Fig. 2E and fig. S5B) (57, 58). Furthermore, FTIR analysis revealed a new Fe-O stretch at 601 cm⁻¹ in CEL-Fe, which was observed to disappear upon treatment with ATP (Fig. 2F) (59). In summary, we confirmed that a coordination bond could be formed between CEL and Fe(III), which could be competitively dissociated by ATP. Therefore, this ATP-dissociable coordination provides a viable option for the construction of a CEL derivative with adjustable cytotoxicity.

According to our design, coordination can lead to attenuating drug-protein interactions (DPIs) and cytotoxicity of CEL, while disruption by ATP should restore activity. To verify this, cell viability curves for CEL-, CEL-Fe–, and CEL-Fe + ATP–treated groups were obtained using the cell counting kit-8 (CCK-8) assay. Meanwhile, three tumor/normal cell line pairs from the lung (A549/BEAS-2B), liver (HepG2/L02), and kidney (786-0/293T) were selected to explore differential cytotoxicity (Fig. 2, G to L, and table S2). Across the different cell lines, 24-hour treatment with CEL-Fe displayed higher viability than CEL, while ATP treatment of CEL-Fe regained cytotoxicity close to free CEL levels. The half-maximal inhibitory concentration (IC₅₀) of the CEL-Fe group was over twofold higher than the CEL group in most of the chosen cell lines (table S2). In addition, CEL-Fe–treated normal cells showed even higher viability, which may be attributed to the lower intracellular ATP concentrations and Hsp90/Cdc37 expression levels in these nontumor cells (60, 61). These results preliminarily confirm the feasibility of the ATP-activatable CEL-Fe(III) chelate as a safe antitumor agent.

The mechanism of cytotoxicity diversity

In this part, we explored the specific mechanisms underlying the observed cytotoxicity changes to provide support for our strategy. At the gene expression level, RNA sequencing (RNA-seq) analysis was performed in A549 cells to explore the mechanism. Principal components analysis (PCA) of gene expression profiles showed a clear separation between the CEL-Fe and CEL groups. In contrast, the CEL-Fe + ATP group clustered with the CEL group, suggesting that ATP reactivated CEL-Fe cytotoxicity through similar mechanisms as CEL (Fig. 3A). Volcano plots were generated by calculating fold change and P values of gene expression, identifying 260 and 145 differentially expressed genes for CEL versus phosphate-buffered saline (PBS) and CEL versus CEL-Fe, respectively (Fig. 3, B and C). Meanwhile, no intergroup differences in gene expression were observed between the CEL- and CEL-Fe + ATP–treated groups, which revealed that ATP could activate the pharmacodynamic activity of CEL (Fig. 3D). The other group comparisons shown in fig. S6 (A to C) further confirmed the similarity between CEL and CEL-Fe + ATP and their differences between the CEL-Fe and PBS groups.

Fig. 3. RNA-seq analysis.

(A) PCA of gene expression. The first two principal components, explaining 99.09% and 0.57% of the total variance, respectively, were used to represent the data and visualize differences between groups. (B to D) Volcano plots of significant differential gene expression between cells treated with (B) CEL and PBS, (C) CEL and CEL-Fe, and (D) CEL-Fe + ATP and CEL. The fold change (FC) was calculated as the expression level ratio between treatment X and treatment Y (X versus Y). Genes with |log₂(FC)| > 0.5 and P < 0.05 were defined as significantly differentially expressed. (E and F) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. The size and color of the dots represent the gene ratio and P value, respectively. Rich factor refers to the ratio of the differentially expressed gene number to the total gene number in a certain pathway. A higher rich factor indicates more significant enrichment. (G to J) Results of the gene set enrichment analysis (GSEA) of differential expression genes. The enrichment score (ES) measures the extent to which a set of genes is disproportionately represented at the bottom or top of a ranked list of genes. Gene ranking metric is based on fold change. The normalized enrichment score (NES) is the ES normalized by gene set size. The horizontal bar graph shows the enriched gene sets/pathways, where red and blue indicate up-regulated and down-regulated pathways, respectively. Darker red and blue colors indicate higher NES.

Next, Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis and gene set enrichment analysis (GSEA) were performed to analyze pathways leading to the cytotoxicity differences. KEGG analysis of CEL versus PBS (Fig. 3E) revealed enrichment in pathways like tumor necrosis factor (TNF) signaling, protein processing in the endoplasmic reticulum (ER), p53 signaling, and cell cycle. These pathways substantially overlapped for CEL-Fe versus CEL (Fig. 3F), and protein processing in the ER highly enriched as well for CEL-Fe + ATP versus PBS and CEL-Fe + ATP versus CEL-Fe (fig. S6, D and E). GSEA showed ER protein processing and ribosome pathway enrichment, and compared to the PBS group, treatment with either CEL or CEL-Fe + ATP induced up-regulation of the proteasome pathway, consistent with reported mechanisms of CEL- and proteasome-mediated degradation (Fig. 3, G to J, and fig. S7) (62–64). Together, the enriched pathways in KEGG analysis are relevant to Hsp90/Cdc37 function (20, 65–67), and GSEA results imply that the mechanism of CEL relates to protein synthesis, processing, and degradation. These analyses confirm the rationality and efficacy of our choice of targeting Hsp90/Cdc37.

For further validation at the molecular level, molecular docking and molecular dynamics (MD) simulations were performed to examine the differences in binding modes of CEL and CEL-Fe to the target proteins Hsp90/Cdc37 and how this may further influence the protein-protein interactions (PPIs) within the Hsp90-Cdc37 complex (Fig. 4A). In the Hsp90-Cdc37 complex [Protein Data Bank (PDB): 2K5B], key hydrogen bonds were identified between Gln¹³³ (Hsp90)–Arg¹⁶⁶/Arg¹⁶⁷ (Cdc37) and Arg⁴⁶/Glu⁴⁷ (Hsp90)–Asp170 (Cdc37) (fig. S8) (23, 68). CEL disrupts these bonds by occupying the hydrophobic pocket (40, 65). Docking revealed that CEL forms three hydrogen bond pairs with Hsp90/Cdc37 via the C-3 hydroxyl and C-20 carboxyl groups (Fig. 4, B and D, and table S3). Critically, bonds were identified between the C-3 hydroxyl and Arg⁴⁶ (Hsp90), C-20 carboxyl and Arg¹⁶⁷ (Cdc37), and C-20 carboxyl and Asp¹⁷⁰ (Cdc37), occupying key sites in Hsp90-Cdc37. In contrast, CEL-Fe only exhibited hydrogen bonds at the C-20 position, leading to reduced interference for the complex and cytotoxicity (Fig. 4, C and E, and table S3).

Fig. 4. Results of molecular docking and MD simulations.

(A) Illustration of the relationship between the DPI (CEL/CEL-Fe and Hsp90/Cdc37) and the PPI (Hsp90 and Cdc37). Because of attenuated DPIs, CEL-Fe exhibits less interference on PPIs, thereby exerting a detoxifying effect. (B to E) Molecular docking of CEL/CEL-Fe to Hsp90/Cdc37 (Hsp90N-Cdc37M; PDB ID: 2K5B). The protein is shown in a cartoon representation. The key side chains involved in hydrogen bond formation are shown in stick type, where C atoms are shown in pink, O atoms are shown in red, H atoms are shown in white, N atoms are shown in blue, and hydrogen bonds are shown in light green. CEL is shown in orange, and CEL-Fe is shown in dark green. (F and G) The number of hydrogen bonds formed between (F) CEL or (G) CEL-Fe and Hsp90-Cdc37 backbone during 50 ns of MD. (H and I) The distances of hydrogen bonds over time. The distances below 3.5 Å are considered indicative of proper hydrogen bond formation. (J and K) Local bonding patterns of CEL (J)/CEL-Fe (K) binding to Hsp90-Cdc37.

MD simulations further provided dynamic evidence. After ligand (CEL/CEL-Fe) and protein (Hsp90-Cdc37) binding stabilized, root mean square deviations (RMSDs) remained low (fig. S9, A to D). The trajectory overlay results demonstrated that coordination with Fe influenced the bonding stability of CEL’s C-20 carboxyl end (fig. S9, E and F). Hydrogen bond analysis from 10 to 50 ns showed that CEL formed approximately two bonds, on average, with Hsp90-Cdc37 (Fig. 4F), while CEL-Fe formed less than 0.5 bonds (Fig. 4G). Statistical analysis revealed significantly higher occupancy for four CEL-protein hydrogen bonds with residues Trp¹⁶⁸ and Asp¹⁶⁹ of Cdc37 (table S4 and Fig. 4J), all at the binding interface. Analysis of hydrogen bond distances over time showed distances between CEL’s O2, and residues Asp¹⁶⁹-OD1/OD2 remained under 3.5 Å, indicating stable bond formation during the simulation (Fig. 4, H and I, and fig. S9, G and H). In contrast, CEL-Fe exhibited a lack of sustained bonding. Fe(III) occupied hydrogen bond donor positions (CEL’s O1/O2) and acted as a new acceptor (Fig. 4K), disrupting ligand-protein interactions. Together with molecular docking, the MD simulations dynamically demonstrated that CEL can bind to Cdc37 and disrupt the formation of the Hsp90-Cdc37 complex.

In addition to the utilization of computational analysis, coimmunoprecipitation also revealed differences in Hsp90-Cdc37 binding after drug treatment (fig. S10A). Using Cdc37 as bait, CEL decreased Hsp90 binding, while CEL-Fe exhibited weaker effects due to reduced interference. However, CEL-Fe + ATP treatment consistently resulted in a decrease in the captured amount of Hsp90, similar to the CEL treatment group. Western blots (WB) evaluating client protein Cdk4 showed consistent results, with both CEL-Fe + ATP and CEL leading to decreased Cdk4 expression (fig. S10B). Because Cdk4 promotes G₁-S progression (69, 70), cell cycle arrest analysis further complemented the mechanisms (fig. S10, C and D). CEL led to increased G₀-G₁ proportions in A549 cells, inducing cycle arrest and inhibiting proliferation, and a similar cycle arrest was observed in the CEL-Fe + ATP group.

In summary, multiple lines of evidence at the cellular, gene expression, and molecular levels validated that: (i) CEL cytotoxicity involves protein synthesis/processing/degradation and many signaling pathways. (ii) CEL-Fe coordination can effectively reduce hydrogen bonding with target proteins Hsp90-Cdc37, which can lead to attenuating cytotoxic effects. (3) ATP competitively displaces CEL-Fe, releasing CEL to elicit cytotoxic changes via similar pathways.

Synthesis and characterization of the ATP-ROS dual-response CEL-Fe nanodelivery system

Encapsulation in polymeric micelles is a common strategy to improve the poor aqueous stability and bioavailability of hydrophobic drugs like CEL (71). The hydrophobic blocks can bind CEL through hydrophobic interactions, enabling micelle self-assembly (72, 73). First, a series of experiments were conducted to assess whether the self-assembly process of micelles is still feasible when the hydrophobic core is replaced by CEL-Fe. As shown in fig. S11 (A to C), the UV absorption spectra still exhibit LMCT bands after encapsulating, confirming no significant impact of encapsulation on coordination. Furthermore, dynamic light scattering (DLS) measurements indicate a predominantly monodisperse size distribution (fig. S11, E to G), demonstrating the formation of uniformly sized NPs. To fully exploit the unique properties of Fe, a ROS-responsive polymer, TK-PEG, was used to achieve TME-triggered and Fenton reaction–enhanced release (74, 75). TK-PEG contains ROS-labile thioketal moieties and combines with F127-PEI to achieve a balance between TME-responsive drug release and physiological stability.

TK-PEG and F127-PEI were synthesized and then characterized by proton nuclear magnetic resonance (¹H NMR; fig. S12). CEL-Fe NPs were prepared by mixing TK-PEG and F127-PEI (8:1 ratio) with CEL-Fe in dimethyl sulfoxide (DMSO), followed by nanoprecipitation in ddH₂O. CEL NPs were similarly prepared as a control. DLS analysis and transmission electron microscopy (TEM) imaging revealed sphere-like morphologies for both formulations, with diameters of ~110 nm for CEL NPs (fig. S13, A and B) and ~86 nm for CEL-Fe NPs (Fig. 5, A and B). Both NPs exhibited negative zeta potential, which can be attributed to the carboxyl groups of the polymers and the PEG chains (Fig. 5C). Inductively coupled plasma mass spectrometryI determined Fe(III) levels after aqua regia digestion, and a 423-nm absorbance standard curve quantified the CEL content in CEL-Fe NPs (fig. S13C), as coordination did not significantly affect the absorbance at 423 nm (Fig. 2C). Quantification revealed a ~1:1.1 CEL:Fe molar ratio (Fig. 5D), consistent with our preceding results (fig. S1). UV-vis spectroscopy showed an LMCT band (Fig. 5E), and XPS revealed Fe 2p peaks (fig. S13, D and E) in CEL-Fe NPs, confirming maintained coordination after encapsulation (44, 45).

Fig. 5. Characterization of the nanoscale ATP/ROS–dual responsive CEL-Fe NPs.

(A) TEM image of CEL-Fe NPs. (B) Hydrodynamic size distribution of CEL-Fe NPs before and after H₂O₂ treatment. (C) Zeta potential of CEL NPs and CEL-Fe NPs. (D) Content of CEL and Fe(III) in CEL-Fe NPs. The molar ratio of CEL to Fe(III) was ~1:1.1. (E) UV-vis spectra of CEL-Fe NPs in the presence of ATP (10 mM). The coordination interactions are gradually broken down by ATP. The inserted graph represents the color changes in the samples (from left to right: CEL NPs, CEL-Fe NPs, and CEL-Fe NPs + ATP.) (F) Stability of the coordination bonds in CEL-Fe NPs at different ATP concentrations (0, 0.1, 0.5, 1, 5, and 10 mM) and pH value (4 or 6) or in 1× PBS buffer. (G) Drug release performances of CEL-Fe NPs in different buffers. ***P < 0.001, ordinary two-way ANOVA. (H) Electron spin resonance spectra of CEL NPs and CEL-Fe NPs [with added H₂O₂ (5 mM)]. Spin trapping agent: 5,5-dimethyl-1-pyrroline-N-oxide (DMPO; 2 μl). (I) Illustration of the characteristics of the TME-responsive behavior of CEL-Fe NPs. For [(C), (D), (F), and (G)], data are expressed as means ± SD (n = 3).

The stability of CEL-Fe NPs was then investigated. DLS showed stable NP size over 5 days (fig. S14, A and B). The polymeric shell enhanced coordination bond stability. As shown in Fig. 5E, the LMCT band remained visible after 2 hours in 10 mM ATP, and when treated with 1 mM ATP, a much slower response speed was observed (fig. S14C), which is likely due to reduced accessibility of ATP caused by electrostatic repulsion. Furthermore, CEL-Fe revealed strong stability in PBS (0.01 M) and acidic solutions (pH 4 and 6; Fig. 5F and fig. S14D), indicating that stable interaction was formed between CEL and Fe(III).

To further demonstrate the advantage of such a dual-responsive CEL delivery system, the ROS-responsive capacity of CEL-Fe NPs was evaluated. The DLS analysis results showed that after incubating CEL NPs and CEL-Fe NPs with H₂O₂ (10 mM) in Hepes buffer, the sharp peaks were transformed into multiple broad peaks and the polydisperse index (PDI) increased significantly (Fig. 5B and fig. S13B).

To simulate the TME and verify ROS-responsive release with the presence of Fe(III), drug release experiments were performed under mild acidic and H₂O₂-containing conditions. Drug release from CEL-Fe NPs was significantly enhanced under optimal Fenton reaction conditions (H₂O₂ 5 mM; pH 6; Fig. 5G). In contrast, CEL NPs showed no significant difference (fig. S14E). Electron spin resonance detected ·OH signals with the spin trap 5,5-dimethyl-1-pyrroline-N-oxide (DMPO) for H₂O₂-treated CEL-Fe NPs at pH 6 (Fig. 5H). This demonstrates that ·OH production accelerated release.

In summary, the nanoscale size facilitates passive tumor targeting via the EPR effect (76, 77), and the introduction of PEG chains can reduce the accessibility of ATP based on electrostatic repulsion, thereby delaying CEL-Fe toxicity within normal cells and further improving biocompatibility. In comparison, at the tumor site, the Fenton reaction accelerates ROS production, leading to the acceleration of drug release, making an important contribution to the rapid activation of CEL and ensuring antitumor efficacy (Fig. 5I).

In vitro evaluation of cell uptake behavior and ATP-responsive cytotoxicity of CEL-Fe NPs

First, effective cell uptake is a prerequisite for drug efficacy. Accordingly, after treating A549 cells with fluorescence Cyanine5.5 (Cy5.5)-labeled CEL-Fe NPs (CEL-Fe/Cy5.5 NPs) for 2, 4, and 8 hours, the cell uptake of NPs in A549 cells was measured by flow cytometry (FCM) and confocal laser scanning microscopy (Fig. 6, A to C). The results showed that the CEL-Fe/Cy5.5 NPs accumulated in A549 cells in a time-dependent manner and reached the peak concentration at approximately 4 hours. Regarding Cy5.5-labeled CEL NPs (CEL/Cy5.5 NPs), the cellular uptake performance was basically the same because of the similar surface charges and size (fig. S15). In addition, after Prussian blue staining, small blue particles could be observed in A549 cells after co-incubation with CEL-Fe NPs (Fig. 6D), indicating the increased intracellular Fe(III) concentration. The time-dependent increase in the intracellular drug concentration also provided a basis for the drug treatment time that we chose during subsequent studies.

Fig. 6. In vitro evaluation of cell uptake behavior and ATP-responsive cytotoxicity.

(A) FCM analysis of the intracellular fluorescence levels in A549 cells incubated with CEL-Fe/Cy5.5 NPs. (B) The corresponding mean fluorescence intensity (MFI) value of (A). (n = 3, ordinary one-way ANOVA). (C) Confocal fluorescence images of A549 cells after incubation with CEL-Fe/Cy5.5 NPs. The blue and red indicate cell nucleus and the CEL-Fe/Cy5.5 NPs, respectively. (D) Optical microscopy images of A549 cells after incubation with CEL-Fe NPs and staining with Perls staining kit. Cell viability of (E) A549 and (F) BEAS-2B after different treatment (n = 4, ordinary two-way ANOVA). (G) Fluorescence microscopy images of live/dead cell staining by Calcein-AM/propidium iodide [PI; c(CEL) = 6 μM, 4 hours]. (H) Western blot analysis of Hsp90 and Cdk4 expression in A549 cells after different treatments [c(CEL) = 6 μM, 6 hours]. (I) The corresponding mean relative grayscale values of Cdk4 (n = 3, ordinary one-way ANOVA). (J) FCM analysis of the intracellular caspase 3 activity by fluorochrome-labeled inhibitors of caspases [c(CEL) = 6 μM, 6 hours] and (K) relevant results of proportion of activated caspase 3 (n = 3, ordinary one-way ANOVA). (L) FCM analysis on A549 cell apoptosis levels. A549 cells after different treatments [c(CEL) = 6 μM, 6 hours] were stained with annexin V–fluorescein isothiocyanate (FITC) and PI. (M) Relevant results of proportion of cells in different states (n = 3, ordinary one-way ANOVA). In the above graphs, CEL, Fe, and ATP represent CEL NPs, CEL-Fe NPs, and ATP-treated CEL-Fe NPs, respectively. For [(E), (F), (I), (K), and (L)], data are expressed as means ± SD (*P < 0.05, **P < 0.01, and ***P < 0.001).

We also evaluated the cytotoxicity of the NPs using the CCK-8 assay. The results shown in Fig. 6E revealed that after 24 hours of treatments, CEL-Fe NPs showed a lower cytotoxicity against lung cancer cell A549 than CEL NPs. Additional support for the above conclusion was provided by the results of the live/dead cell staining assay, which was conducted to directly visualize cell viability, and in the ATP-treated CEL-Fe NPs group, cytotoxicity was restored because of reactivation of the CEL activity (Fig. 6G and fig. S16). When CEL-Fe NPs were treated with a stronger coordination competitor (EDTA) or a weaker one (citric acid), their cytotoxicity changed in a different way and implied that the formation and dissociation of coordination bonds led to changes in cytotoxicity (fig. S17, A and B). Furthermore, when CEL-Fe NPs were cocultured with normal human lung epithelium cell (BEAS-2B) cells to verify biosafety, the cell viability of the BEAS-2B was significantly higher than that of CEL-NPs treated for 24 or 48 hours (Fig. 6F and fig. S18B). However, compared with the results in Fig. 6E and fig. S18A, cell viability of A549 revealed no difference after being treated for 48 hours. The difference in cytotoxicity between normal and tumor cell lines may be due to differences in endogenous ATP concentration (37, 38).

We also investigated the mechanism of the changes in cytotoxicity of CEL-Fe. The results of other experiments in this study also indicate that CEL-Fe may lead to attenuating DPI compared to CEL and thus may be less able to interfere with the formation of the Hsp90-Cdc37 complex. As shown in Fig. 6 (H and I), the expression level of the Cdk4 (a client protein of the Hsp90-Cdc37 complex) follows the same trend as the cytotoxicity. However, the expression level of Hsp90 is unaffected, which is consistent with the results of previous studies (23, 78, 79). In addition, we also performed FCM analysis to measure the activity of caspase 3 (Fig. 6, J and K) and the total rate of dead (apoptotic) cells (Fig. 6, L and M), which found that the difference in cytotoxicity is correlated with the degree of apoptosis triggered by the different treatments. In the group treated with CEL-Fe NPs, the activation rate of caspase 3 and the total percentage of Q1-UR and Q1-LR in the annexin V–fluorescein isothiocyanate (FITC)/propidium iodide (PI) double staining FCM analysis results were revealed to be lower than those of the groups treated with CEL NPs or CEL-Fe NPs + ATP (Fig. 6, J to M). Furthermore, in the group treated with CEL-Fe NPs + ATP, the degree of apoptosis returned to a level similar to that of the groups treated with CEL NPs. All these results demonstrated that the CEL-Fe chelate–based strategy was able to reduce toxicity and can be reactivated by ATP, the underlying mechanism correlated with the activity of the Hsp90-Cdc37 complex and the degree of apoptosis.

Study of biosafety and biodistribution

In this study, we propose an intelligent CEL delivery system that is based on the coordination of CEL and Fe(III). The improvement of biosafety is a key aspect that needs systematic evaluation. Biosafety is crucial for intravenous administration in in vivo therapeutic experiments to assess the feasibility of NDDSs.

The hemolysis assay (Fig. 7A and fig. S19) showed that hemolysis occurred in the groups treated with CEL and CEL NPs. The encapsulation of the polymer had no significant preventive effect on the occurrence of hemolysis. In contrast, no significant hemolysis was observed in the CEL-Fe and CEL-Fe NPs groups in the concentration range that we used in this assay. The hemolysis rate was lower than 5% even when the concentration of CEL-Fe NPs reached 500 μM, indicating that CEL-Fe NPs had less interaction with the membrane of red blood cells. The results also revealed that, because of the interference of salt ions, CEL-Fe was even less water soluble than CEL and was precipitated after centrifugation; thus, CEL-Fe may not be feasible to be administered by intravenous injection.

Fig. 7. Study of biosafety and biodistribution.

(A) Hemolysis assay of CEL, CEL NPs, CEL-Fe, and CEL-Fe NPs. (B) H&E histological staining results of tissue slides of mouse heart, liver, spleen, lung, and kidney. The arrows in the graphs indicate some of the obvious toxic features. The unit of the scale bar in the figure is micrometers (μm). (C) Pathology score of the H&E stains of each organ in different groups. The pathology scores were assigned on the basis of the slice information, considering cell degeneration, necrosis, and inflammatory cell infiltration. Higher scores indicate more severe damage in the corresponding organ. (D to G) Fluorescence images of in vivo stage (0 to 48 hours) and ex vivo organs at 48 hours and the corresponding MFI values. Mice were intravenously injected with Cy7-labeled NPs [c(Cy7) = 40 μM, 100 μl; n = 3; Ex = 740 nm and Em = 770 nm; the radiant efficacy was calculated by emission light (photons s⁻¹ cm⁻² sr⁻¹)/excitation light (μW/cm²)]. [(D) and (F)] BALB/c nude mice bearing the A549 cell-derived xenograft (CDX). [(E) and (G)] BALB/c nude mice bearing the patient-derived xenograft (PDX) of lung cancer. For [(F) and (G)], data are expressed as means ± SD (n = 3).

We also confirmed the biosafety of CEL-Fe NPs in vivo. The dosage of CEL is usually in the range of 1 to 3 mg/kg (10) when administered by intravenous injection. In this study, we used a stress test (total dosage was increased to 9 mg/kg) to determine whether intravenous administration of NPs into healthy BALB/c mice resulted in significant toxicity or adverse effects. Ten days after the administration, we collected the major organs and stained them with hematoxylin and eosin (H&E). The H&E-stained tissue sections, shown in Fig. 7B, underwent scoring based on slice information regarding cell degeneration, necrosis, and inflammatory cell infiltration, with higher scores indicating more severe damage. Organ scoring results are presented in Fig. 7C, and detailed organ scoring criteria can be found in table S5. Significant tissue toxicity was observed in the CEL and CEL NP treatment groups. Specifically, liver slices exhibit significant hepatocellular edema and punctate necrosis, accompanied by infiltration of lymphomonocytes and neutrophils in the portal area. Spleen slices exhibited a decreased number of splenic nodules, leading to reduced immune functions. Lung slices displayed a significant widening of the alveolar septum and infiltration of the interstitial space with inflammatory cells. Kidney slices from the CEL-treated group showed infiltration of focal inflammatory cells around the interstitial vessels. However, observations from the group receiving controlled dosing of CEL-Fe NPs revealed that, except for the spleen showing a decrease in splenic white pulp lymphoid nodules, the results in other organs were essentially consistent with those of the PBS treatment group. In addition, given our focus on lung cancer research, the impact of CEL-Fe NPs on normal lung cells has drawn our attention. We conducted a thorough investigation to discern the distinct effects of CEL and CEL-Fe NPs on lung function using the FinePointe whole-body plethysmography system. Respiratory rate, tidal volume, and enhanced pause (Penh), an index of airway resistance reflecting airway stenosis, were recorded in each group to evaluate the pulmonary condition of mice. Results shown in fig. S20 revealed that mice treated with CEL potentially experienced some degree of inflammation or injury in the lungs, leading to abnormal respiratory function. Conversely, in the CEL-Fe NPs group, there were no significant differences observed in these three indices compared to the PBS group. In summary, our results confirmed that the chelate strategy and the protective effect of the polymeric shell can lead to attenuating the toxic effects of CEL and help improve biosafety.

In addition, the biodistribution of NPs assembled by hybrid polymers was assessed to confirm enhanced tumor accumulation. To this end, A549 cell-derived xenograft (CDX) and the subcutaneous transplantation tumor model of the patient-derived xenograft (PDX) of lung cancer were established in BALB/c nude female mice. First, Cy7-labeled NPs were prepared and administered to tumor-bearing BALB/c mice through the tail vein, and the biodistribution of NPs at different time points was evaluated using an in vivo imaging system. The fluorescence intensity at the tumor site gradually increased over time in these two models (Fig. 7, D and E, and fig. S21), suggesting that Cy7-labeled NPs accumulated at the tumor site delivered through blood circulation. The mice were euthanized after 48 hours, and the main organs were dissected and fluorescence-imaged (Fig. 7, D and E). The fluorescence quantification results of dissected organs aligned with in vivo fluorescence imaging, confirming that the tumor sites show the strongest fluorescence intensity in both CDX and PDX models (Fig. 7, F and G). Therefore, it is inferred that, following our strategy, the accumulation of CEL-loaded NPs at the tumor site lays an important foundation for the antitumor efficacy of this strategy.

In vivo evaluation of the antitumor efficacy

After validating the biosafety and biodistribution of the CEL-Fe NPs on A549 CDX and PDX mouse models, we further studied its antitumor efficacy separately on these two models.

For A549 CDX models, nude mice with A549 xenograft were divided into six groups (n = 5) and treated with PBS, CEL, CEL-Fe, CEL NPs, CEL-Fe NPs, and NPs, respectively. Ensuring uniform dosages, an equivalent amount of CEL or carrier substrates was administered in each group. When the tumor volume reached 100 mm³, different treatments were administered intravenously on days 1, 3, 7, and 10. The tumor volume and weight of the mice were monitored daily (Fig. 8A). On day 12, the mice were euthanized, and the tumors were weighed and photographed. As depicted in Fig. 8 (B and C), the PBS and NP groups showed rapid tumor growth, which increased more than 15-fold within 12 days. Meanwhile, the tumor in the CEL- or CEL-Fe–treated group also increased by more than 10-fold and 13-fold. However, the group treated with CEL NPs and CEL-Fe NPs had significantly reduced the rate of tumor growth. Notably, the group treated with CEL-Fe NPs exhibited the smallest tumor volume and weight among all the groups, despite the nonsignificant difference compared to CEL NPs. This implies a potential for improved antitumor efficacy with CEL-Fe NPs (Fig. 8, B to E). Moreover, the expression of Cdk4, necrosis, and apoptosis in each group was analyzed using immunohistochemistry (IHC), H&E staining, and terminal deoxynucleotidyl transferase deoxyuridine triphosphate nick end labeling (TUNEL) assay on tumor tissue sections. The results showed that the group treated with CEL-Fe NPs had the lowest Cdk4 expression, the highest levels of apoptosis, and elevated tumor necrosis. This indicates that CEL-Fe NPs can effectively restore cytotoxicity, inhibit tumor proliferation, and enhance cytotoxicity to tumors (Fig. 8F).

Fig. 8. In vivo antitumor effect of CEL-Fe NPs in the CDX model.

(A) Schematic illustration of the establishment of the A549 CDX model and treatment schedules. (B) Tumor volume of each group of mice treated with PBS, CEL, CEL-Fe, CEL NPs, CEL-Fe NPs and NPs (terms as groups 1 to 6, respectively). (C) Tumor volume curves 12 days after different treatments (data are expressed as means ± SD; n = 5; *P < 0.05, **P < 0.01, and ***P < 0.001, ordinary two-way ANOVA). Ex vivo tumor weight (D) and images (E) of obtained tumors on day 12 after different treatments (data are expressed as means ± SD; n = 5; **P < 0.01 and ***P < 0.001, ordinary one-way ANOVA). (F) Representative Cdk4 immunohistochemical, TUNEL, and H&E staining of tumors obtained on day 12 after treatment. i.v., intravenous injection.

To preserve the genetic and pathological features of the cancer cells from the patient and realistic model to test the efficacy of the drug (80–82), the antitumor efficacy of different formulations was also evaluated in BALB/c nude mice bearing PDX of human lung cancer. Upon reaching a tumor volume of 100 mm³, mice were randomly divided into four groups (n = 5). Intravenous administration of different treatments (PBS, CEL, CEL NPs, and CEL-Fe NPs) occurred on days 0 and 3, and tumor volumes and weights were monitored every other day, as detailed in Fig. 9A. After 2 weeks, the mice were euthanized, and the tumors were weighed and photographed. As illustrated in Fig. 9 (B and C), the PBS group exhibited rapid tumor growth, increasing more than 10-fold after 14 days. In addition, because of lower bioavailability and poor tumor accumulation, tumors in the CEL-treated group also grew nearly eightfold. However, treatment with CEL NPs significantly reduced the rate of tumor growth, while the CEL-Fe NPs group demonstrated the most effective antitumor efficacy (Fig. 9, B to E). At the conclusion of the experiments, the group treated with CEL-Fe NPs exhibited the lowest weight, showing a statistically significant difference compared to the group treated with CEL NPs (Fig. 9D). Moreover, tumor necrosis analysis by H&E staining and apoptosis analysis by the TUNEL assay performed on tumor slices showed that the group treated with CEL-Fe NPs had the highest levels of apoptosis and tumor necrosis (Fig. 9F). In conclusion, the reactivation of CEL by competitive bonding with ATP at a high concentration in the tumor site was demonstrated both in CDX and PDX mouse models. Meanwhile, the enhanced antitumor effect of CEL-Fe NPs may also be attributed to their TME-responsive release behavior.

Fig. 9. In vivo evaluation of CEL antitumor effect in the PDX model.

(A) Schematic illustration of the establishment of the animal model and treatment schedules. (B) Tumor volume of each group of mice treated with PBS, CEL, CEL NPs, and CEL-Fe NPs. (C) Tumor volume curves during 14 days after different treatments (data are expressed as means ± SD; n = 5; *P < 0.05, **P < 0.01, and ***P < 0.001, ordinary two-way ANOVA). The red line in the graph represents the initial tumor volume (100 mm³). Ex vivo tumor weight (D) and images (E) of obtained tumors on day 14 after different treatments (data are expressed as means ± SD; n = 5; **P < 0.01 and ***P < 0.001, ordinary one-way ANOVA). (F) Representative H&E staining and TUNEL assay of tumors obtained on day 14 after treatment.

DISCUSSION

In this study, we have successfully developed a dual-responsive nano prodrug system (CEL-Fe NPs) for efficient delivery of CEL for cancer treatment. In normal tissues, exploiting the coordination of CEL and Fe(III) and the protective effect provided by the polymeric shell (6, 83), CEL-Fe NPs maintain their structural stability while their systemic toxicity is reduced. While the CEL-Fe NPs accumulate at the tumor site, an ATP-ROS–rich microenvironment, the thioketal groups in the polymeric shell are broken by ROS, and additionally enhanced drug release is achieved through Fenton reaction (84). Furthermore, CEL-Fe is reactivated by abundant ATP (36, 53), leading to the release of CEL from the CEL-Fe chelate with reactivatable antitumor activity. This strategy shows well the therapeutic efficacy and safety in both A549 CDX and lung cancer PDX models. In comparison with other CEL delivery systems, our strategy not only proposed the concept of Fe-coordination prodrug for CEL but also take full advantage of the tumor microenvironment to achieve the best antitumor effect and minimize side effects on major organs.

A coordination bonding approach has been used in many fields, for example, polyphenol-Fe based systems have been used as an effective carrier-free strategy to develop many applications (41, 56, 85). In addition, taking advantage of drug-metal ion complexes (such as doxorubicin-Fe), combination therapy can also be achieved through the chemotherapeutic effects of drugs and specific functions of metal ions (e.g., Fenton effect) (86–88). However, a CEL-based chelate has never been reported, and its potential applications have been ignored. In this study, we developed an intelligent CEL delivery system based on chelation strategy to improve its biosafety and bioavailability. The mechanisms underlying the reduction of toxicity were assessed through molecular docking and MD simulation, revealing that the differences in interactions between CEL/CEL-Fe and targeted proteins (Hsp90/Cdc37) lead to different inhibition efficiencies. In addition, RNA-seq was used to verify the differences in pharmacological mechanisms in vitro. In conclusion, this Fe(III)-based coordination prodrug strategy blocks the binding activities of CEL to Hsp90/Cdc37 and can be reactivated by ATP competitive binding to Fe(III).

Although this study mainly focused on CEL and did not extensively explore the role of metal ions, our approach is noteworthy for its simplicity, scalability, and potential to be extended to other drugs containing coordination sites, enhancing selectivity (18). Moreover, considering the variety of metal ions available for coordination, applications built upon this chelate strategy can be further advanced to develop a more integrated and multifunctional system for the successful utilization of CEL. For instance, combining Gd³⁺ enables diagnostic imaging and radiation sensitization (89), while using Zn²⁺ to activate cGAS/STING signals up-regulates innate immunity (90). This approach may also pave the way for the development of strategies involving other cytotoxic drugs.

MATERIALS AND METHODS

Materials

CEL [(2R,4aS,6aS,12bR,14aS,14bR)-10-hydroxy-2,4a,6a,9,12b,14a-hexamethyl-11-oxo-1,2,3,4,4a,5,6,6a,11,12b,13,14,14a,14b-tetradecahydropicene-2-carboxylic acid] was purchased from Bide Pharm (Shanghai, China). Ferric chloride hexahydrate (FeCl₃·6H₂O), N,N′-carbonyl diimidazole (CDI), and Tween 80 were purchased from Aladdin (Shanghai, China). ATP was purchased from Macklin (Shanghai, China). PEG-300 and triethylamine (TEA) were purchased from Sigma-Aldrich Co. Ltd. (MO, United States). 1,2,4,5-Cyclohe-xane tetracarboxylic dianhydride (CHTA), mPEG5000-OH, and Pluronic F127 were purchased from Energy Chemistry (Shanghai, China). Ethylene imine polymer-600 (PEI-600) was purchased from RHAWN (Shanghai, China). Cy7 and Cy5.5 were purchased from Yuan ye (Shanghai, China). DMPO was purchased from TGI (Shanghai, China). RPMI 1640 media, Dulbecco’s modified Eagle’s medium (DMEM), fetal bovine serum (FBS), and trypsin were purchased from Gibco (BRL, MD, USA). Whole protein extraction kit, bicinchoninic acid protein assay kit, Prussian blue staining kit, 4′,6-diamidino-2-phenylindole (DAPI), and CCK-8 were purchased from Solarbio (Beijing, China). Annexin V–FITC/PI apoptosis detection kit, Calcein/PI cell viability/cytotoxicity assay kit, and 4% paraformaldehyde were purchased from Beyotime (Shanghai, China).

Characterization of the formation of coordination bond

The CEL (20 mM and 9 mg/ml) and FeCl_{`</sub>6H<sub>2</sub>O (5 mM and 1.35 mg/ml) solutions were prepared in ethanol. Then, 20 μl of CEL solution was added with varying amounts of FeCl<sub>`}6H₂O solutions in different molar ratios: 0, 1:10, 1:6, 1:3 1:2, 1:1, 2:1, and 3:1 (Fe:CEL). Ethanol was added to make up the total volume of the solutions at 1 ml. Then, photographs of solutions were taken, and UV-vis spectra were recorded using a UV-vis spectrometer (Thermo Fisher Scientific Inc., Waltham, MA, USA).

A similar scheme was used for the coordination characterization of CEL with different metals. Different metals were added to the DMSO solution of CEL at the same concentration according to the molar ratio of 1:1. UV-vis spectra of the samples were tested, and photos were taken.

Characterization on the dissociation of coordination bond

Because ATP is insoluble in ethanol, ATP response coordination dissociation experiments were performed in DMSO. CEL (475 μl at 0.25 mM) was mixed with 475 μl of FeCl·6H₂O (0.5 mM); then, a rapid addition of 50 μl of ATP solutions with different concentration (200 nM, 2 μM, 20 μM, 200 μM, 2 mM, and 20 mM) was performed before the time-dependent UV-vis spectra were recorded using the NanoDrop One ultra-micro spectrometer (Thermo Fisher Scientific Inc.) (time gap, 10 s). During the testing process, we have included small magnetic stir bars in the cuvettes and maintained a constant stirring speed of 300 rpm to ensure the uniformity of the solution. A similar protocol was used for the response experiment on coordination bonds using EDTA as a competitive binding reagent. The final concentration of EDTA in the mixed solution was 1 mM, and the UV-vis change was immediately recorded (Thermo Fisher Scientific Inc.).

Characterization of the structural features

CEL-Fe–and ATP-treated CEL-Fe were synthesized in dichloromethane, obtained after purification, and dried at room temperature. The XPS, FTIR spectroscopy, and Raman spectroscopy spectra of the samples were recorded on an Axis Supra x-ray photoelectron spectrometer (Kratos Analytical Ltd., Manchester, UK), a Nicolet FTIR spectrometer (Thermo Fisher Scientific Inc.), and a confocal Raman microscope LabRAM HR Evolution (HORIBA France SAS, Longjumeau, Paris, France), respectively.

Molecule docking simulation

The crystal structures of the Hsp90-Cdc37 complex (PDB: 2K5B) were downloaded from the RCSB PDB (www.pdb.org/). The main interactions between Hsp90 and Cdc37 were analyzed by Ligplot (European Molecular Biology Laboratory, European Bioinformatics Institute, Cambridgeshire, UK). Then, Cdc37 and Hsp90 were separated in PyMOL (Schrödinger Inc., NY, USA), and hydrogen atoms were added. Subsequently, the binding sites of CEL/CEL-Fe were defined at the interface of the Hsp90-Cdc37 complex, and the molecular docking was performed in AutoDock software (repeated 50 times in each group; Center for Computational Structural Biology, CA, USA). The result with the lowest BE in each group was selected as a representative and analyzed in PyMOL.

Molecular dynamics simulation

The structure of the Hsp90-Cdc37 complex was retrieved from the PDB with accession number 2K5B. Excess crystallographic water molecules were removed from the PDB file. The protein complex and small-molecule ligands were subjected to molecular docking using Yasara (version 21.12.19,), selecting docking poses with lower binding energies and rational conformations for MD simulations. The electrostatic potential charges of the small-molecule ligands were fitted using the restrained electro static potential (RESP) method with Multiwfn_3.8, and the ligand topology files were generated using Sobtop. The entire complex was simulated using GROMACS 2022.3 with the AMBER99SB-ILDN protein and nucleic AMBER94 force field. The system was enclosed in a cubic box with a 1.0-nm buffer space for the addition of solvent molecules (TIP3P water molecules), and counterions were added to maintain neutrality, with eight Na⁺ ions for the CEL_system and seven Na⁺ ions for the CEL-Fe_system. A restraint potential of 1000 kJ mol⁻¹ nm⁻² was applied to fix the positions of the protein and ligands. An energy minimization of 10,000 steps was performed using the conjugate gradient method. The particle mesh Ewald method was used to compute electrostatic interactions with a cutoff of 1.0 nm, and the van der Waals interactions were also truncated at 1.0 nm. The Berendsen thermostat and barostat were used to regulate temperature and pressure, with bonds involving hydrogen atoms constrained with a time step of 1 fs, respectively. The system was maintained at 298.15 K for a run of 1000 ps. Subsequently, the protein and ligand were separated and indexed, considering the protein-ligand complex as a single group for temperature coupling and translational-rotational removal over 50 ns of conventional dynamics.

The protein backbone and ligand-to-protein backbone RMSD were analyzed using the “rms” module in GROMACS, examining the RMSD of both the protein and the ligand molecules. The “hbonds” module was used to monitor the variation in the number of hydrogen bonds between the protein and the ligand over time.

Synthesis of the TK-PEG polymer

The TK-PEG polymer was synthesized as previously described (91). ROS-liable [2,2′-(propane-2,2-diylbis (sulfanediyl)) bis (ethan-1-ol), 392.12 mg, and 2 mmol] and CHTA (1,2,4,5 cyclohexane tetracarboxylic dianhydride, 493.17 mg and 2.2 mmol) were suspended in 5 ml of anhydrous N,N′-dimethylformamide, after stirring for 24 hours, and then mPEG5000-OH (0.2 mmol and 1 g) was added. Then, the above mixture (5 ml) was added to 15 ml of ddH₂O and followed by dialysis in a dialysis bag [molecular Da] for 72 hours. Last, the polymer TK-PEG was obtained by freeze drying. The ¹H NMR spectra of polymer TK were measured using an NMR spectrometer (Bruker BioSpin GmbH, Rheinstetten, Germany).

Synthesis of the F127-PEI polymer

The F127-PEI polymer was synthesized as previously described (56). F127 (12.7 g and 1 mmol) and CDI (0.25 g and 1.5 mmol) were mixed in 40 ml of DMSO; then, the reaction flask was filled with N₂, and 0.2 ml of TEA was added. The mixture solution was stirred under N₂ in the dark for 4 hours to obtain F127-CDI. PEI-600 (0.9 g and 1.5 mmol) was dissolved in 10 ml of DMSO and then slowly dropped into the previously prepared F127-CDI solution with stirring for 12 hours. The final F127-PEI powder was obtained after dialyzing for 48 hours in a dialysis bag (MWCO: 3500 Da) and freeze drying. The ¹H NMR spectra of F127-PEI were measured using an NMR spectrometer (Bruker BioSpin GmbH).

Synthesis and characterization of NPs

The TK-PEG and F127-PEI polymers were mixed in DMSO at a mass ratio of 8:1 to form a mixture with a total content of 40 mg/ml. To synthesize CEL-Fe NPs, 200 μl of CEL solution (10 mM, in DMSO) and FeCl_·6H₂O solution (20 mM, in DMSO) were mixed and added to 250 μl of the mixed polymer solution. The above mixture was added dropwise to 10 ml of ddH₂O, stirred for 30 min, dialyzed against ddH₂O (MWCO: 7000 Da) overnight to remove DMSO, and centrifuged at 6000 rpm for 15 min. CEL NPs and NPs encapsulated by other polymers were synthesized in a similar way.

To obtain ATP-treated CEL-Fe NPs, ATP (100 mM) was added to CEL-Fe NP solutions and incubated overnight, then washed with water to remove residual Fe(III) and ATP, centrifuged at 4000 rpm in an ultrafiltration tube (Millipore, 50 kDa) to concentrate for 15 min, and repeated after adding ddH₂O at least five times.

Subsequently, the samples were examined by XPS to demonstrate complete removal of Fe(III).The morphology of NPs was obtained by a JEOL JEM-1400 transmission electron microscope (Japan Electronics Co. Ltd., Tokyo, Japan). The hydrodynamic diameter, zeta potential, and stability of the samples were tested with a Malvern Nano-ZS90 particle sizer (Malvern Panalytical Ltd., Malvern, UK).

Cytotoxicity assay

The CCK-8 assay was used to evaluate the cytotoxicity of different CEL-based solutions. For the CCK-8 assay depicted in Fig. 2 (G to L), aiming to explore the differences in cytotoxicity between tumor and nontumor cells from the same tissue, we selected three pairs of cell lines: A549 (human lung adenocarcinoma cells) and BEAS-2B (human lung epithelial cells) from lung, HepG2 (human hepatocellular carcinoma cells) and L02 (human normal liver cells) from liver, and 786-0 (human renal cell carcinoma) and 293T (human embryonic kidney cells) from kidney. Cells were seeded in 96-well plates (Thermo Fisher Scientific Inc.) at a density of ~8000 cells per well and incubated in basic cell culture media (RPMI 1640 for A549, L02, and 786-0 and DMEM for BEAS, HepG2, and 293T) supplemented with 10% FBS at 37°C overnight. Afterward, the medium was replaced by 100 μl of phenol red–free media containing PBS, CEL, CEL-Fe, and CEL-Fe + ATP at a concentration of CEL ranging from 2, 4, 6, 8, and 15 μM, further incubated for 24 hours.

For the assay in Fig. 6 (E and G) and fig. S18, we explored the trend of cytotoxicity changes of CEL NPs and CEL-Fe NPs in healthy epithelial cells (BEAS-2B) and tumor cells (A549) in a time-dependent way. In general agreement with the experimental protocol described above, cells were grown in 96-well plates to a coverage of about 80%, and CEL NPs/CEL-Fe NPs were diluted with phenol red–free media to 4/8 μM, and then, nanodrugs were used for incubation for different times (10, 24, 36, and 48 hours).

Following incubation, cellular viability was assessed using a CCK-8 assay (Solarbio, Beijing, China). Briefly, CCK-8 reagent (1×) was added to each well and further incubated with cells for another 2 hours. The absorbance at 450 nm was measured on a Tecan I–control microplate reader (Tecan Trading AG, Männedorf, Switzerland). The cytotoxicity of CEL NPs and CEL-Fe NPs against normal human kidney cells 293 was determined following the same protocol with the concentration of CEL ranging from 2, 4, 6, 8, and 15 μM. Cell viability (%) was calculated with the following formula

$$\text{Cell viability} (\%)=(A_{\text{sample}}-A_{\text{blank}})/ (A_{\text{NC}}-A_{\text{blank}})\times 100\%$$

Here, A_sample represents the average absorbance at 450 nm of the drug-treated groups. A_blank represents the average absorbance at 450 nm of wells with added PBS (without cells and drugs), and A_NC represents the average absorbance at 450 nm of the negative control group (cells without drug treatment).

Animals

The mice were maintained in an environment with a temperature of 22 ± 1°C, a relative humidity of 50 ± 1%, a light/dark cycle of 12 hours/12 hours, and provided with sterile food and water. All animal experiments were carried out under the guidelines evaluated and approved by the ethics committee of Peking University, P.R. China (LA2021316).

In vivo biosafety test

BALB/c mice at approximately 5 weeks of age (female and body weight of approximately 18 g) were purchased from Huafukang Biotechnology Co. Ltd. (Beijing, China). Concentrated samples of CEL NPs and CEL-Fe NPs were obtained by ultrafiltration. The sample denoted as CEL (free drug) was prepared as follows: PEG-300 (800 μl) was added to 200 μl of CEL solution (9 mg/ml in DMSO); then, 100 μl of Tween 80 was added, and last, saline was added to a total volume of 2 ml. After measuring the absorbance at 423 nm of each sample, they were then diluted with saline to the same concentration. The dosage for each mouse was 3 mg/kg (150 μl each time, intravenously), and the total dosage of 9 mg/kg was achieved by administering the dosage three times (24 hours between each dose). At the end of the experiment (day 10), the mice were euthanized, and the major organs (heart, liver, spleen, lung, and kidney) were collected, cryo-sectioned, and stained by H&E, and then, the slices were scanned to evaluate the toxicity to the organs.

On the basis of the obtained tissue section results, we scored them according to relevant literature (92), primarily focusing on cell degeneration, necrosis, and inflammatory cell infiltration, with scores ranging from 0, 1, 2, to 3, indicating none, mild, moderate, and severe, respectively. The individual scores for each parameter were summed to calculate a total score.

Lung-related physiological function tests were conducted on female BALB/c mice, approximately 5 weeks old with a body weight of around 18 g, obtained from Huafukang Biotechnology Co. Ltd. (Beijing, China). The dosage administered for these tests was consistent with the biosafety test. At the conclusion of the experiment, the mice were placed in the Whole Body Plethysmography (WBP; DSI Buxco, Harvard Bioscience Inc., Boston, MA, USA) apparatus, ensuring quiet, temperature-controlled, and dimly lit conditions to promote their comfort and acclimatization to the surroundings. Following their settling, the WBP equipment was activated, and parameters were recorded for each group of mice in a resting state. These parameters, including respiratory rate, tidal volume, and enhanced pause (Penh), were monitored using Buxco FinePointe software, with each parameter continuously measured for a minimum of 5 min to ensure the reliability of the data.

In vivo therapy

The CDX models were established by inoculating A549 cells [1 × 10⁶cells/100 μl in 1:1 (v/v) PBS and Matrigel; BD Bioscience] into the upper right legs of BALB/c nude mice (Huafukang Biotechnology Co. Ltd.). The PDX models were established as previously described (80) in 5-week-old BALB/c nude mice (Huafukang Biotechnology Co. Ltd.). Briefly, lung cancer patient-derived tumors were freshly collected, cut into small fragments (3 mm by 3 mm by 3 mm), and transplanted subcutaneously into BALB/c nude mice. Lung cancer tissues were obtained from the Fifth Medical Center of Chinese People’s Liberation Army General Hospital, with approval from the hospital’s ethics committee and in compliance with all relevant ethical regulations, following informed consent from the donor.

After the establishment of the A549 CDX and lung cancer PDX model and when the tumor volume reached approximately 100 mm³, BALB/c nude mice bearing tumor were randomly divided into several groups, with five animals per group.

For A549 CDX models, the six groups are as follows: PBS, CEL, CEL-Fe, CEL NPs, CEL-Fe NPs, and NPs. When the tumor volume reached 100 mm³, different treatments were administered intravenously on days 1, 3, 7, and 10 (3 mg of CEL kg⁻¹). The tumor volume and weight of the mice were monitored every day. After 12 days, the mice were euthanized, and the tumors were weighed and photographed.

For PDX models, the four groups are as follows: PBS, CEL, CEL NPs, and CEL-Fe NPs. The mice were treated twice with different samples (3 mg of CEL kg⁻¹) by intravenous injection on days 0 and 3. The tumor volume and weight of the mice were monitored every other day. After 2 weeks, the mice were euthanized, and the tumors were weighed and photographed.

Tumor volume (V) was calculated as follows

$$V=L\times W^2/2$$

Here, L (in mm) represents the length of the tumor. W (in mm) represents the width of the tumor, and V (in mm³) represents the volume of the tumor.

Statistical analysis

The results were presented as means ± SD. The level of significance in all statistical analyses was set at P < 0.05. Statistical comparisons were performed using one-way or two-way analysis of variance (ANOVA), and if more than two groups were involved, the Tukey test was used to correct for multiple comparisons using GraphPad Prism 8.0 software (Dotmatics, San Diego, CA, USA).

Supplementary Materials

This PDF file includes:

Instruments and methods

Figs. S1 to S21

Tables S1 to S5

Legend for movie S1

Other Supplementary Material for this manuscript includes the following:

Movie S1