Metal-organic framework-based fuel-driven chemical reaction network for ferroptosis therapy

Abstract

Ferroptosis is a prospective approach for cancer treatment. However, the efficacy of ferroptosis therapy is limited by three parallel ferroptosis defense pathways: the glutathione (GSH)-glutathione peroxidase 4 (GPX4) pathway, the ferroptosis suppressor protein 1 (FSP1)-ubiquinol (CoQH2) pathway, and the dihydroorotate dehydrogenase (DHODH)-CoQH2 pathway. Inspired by the principles of preprogrammed chemical reaction networks (CRNs), herein, a novel drug delivery system (SRF@Au@M NPs) was designed based on MIL-100(Fe) for predictable behaviors in tumor cells to break the three ferroptosis defense systems. SRF@Au@M NPs were fabricated through the size optimization of MIL-100(Fe), sorafenib (SRF) loading, in-situ growth of Au nanoparticles (Au NPs) and surface modification with dihydrolipoic acid derivatives. SRF@Au@M NPs disintegrate in the presence of high concentrations of GSH, releasing sorafenib (SRF) into tumor cells, which reduces GSH synthesis and inhibits GPX4 activity. The Au nanoparticles decompose glucose to produce H₂O₂, providing fuel for the Fenton reaction and disrupting carbohydrate metabolism to inhibit NAD(P)H generation. Particularly, a novel redox-CRN was formed between dihydrolipoic acid derivatives and iron ions, continuously promoting reactive oxygen species generation while concurrently consume NADH. The imbalance of NAD(P)H metabolic homeostasis impedes the recycling of CoQ to CoQH2, resulting in the simultaneous inhibition of the FSP1-CoQH2 and DHODH-CoQH2 pathways. Consequently, the SRF@Au@M NPs triggered a potent ferroptosis storm in 4T1 tumor cells and achieved an 92.5% tumor growth inhibition in tumor-bearing mice, significantly higher than that of other treatment groups. Our sophisticated strategy based on CRNs provides a new promising paradigm for ferroptosis activation and cancer treatment.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12951-026-04718-2.

Introduction

Ferroptosis is a unique type of programmed cell death triggered by iron overload and the accumulation of lethal lipid peroxidation (LPO) [1–3]. Over the past decade, extensive research has revealed that ferroptosis activation may serve as a potential therapeutic strategy for multiple diseases [4]. In particular, cancer cells resistant to traditional therapies or prone to metastasis are more susceptible to ferroptosis [5, 6], making it a fascinating avenue for cancer treatment. Although significant progress has been achieved to enhance the precision and efficacy of ferroptosis-based therapies, achieving robust and comprehensive ferroptosis induction remains a challenge. This difficulty is primarily attributed to the complex and inherent ferroptosis defense systems and the unique metabolism of cancer cells. Detailed mechanistic studies uncovered that there are three parallel ferroptosis defense systems that are essential for preventing LPO accumulation and subsequent ferroptosis: the glutathione (GSH)-glutathione peroxidase 4 (GPX4) system, the NAD(P)H-ferroptosis suppressor protein 1 (FSP1)-ubiquinol (CoQH2) system, and the dihydroorotate dehydrogenase (DHODH)-CoQH2 system [7–9]. In detail, the GSH-GPX4 pathway prevents lipid reactive oxygen species (ROS) from accumulating within lipid bilayers, while FSP1 prevents uncontrolled LPO by reducing ubiquinone (CoQ) to CoQH2 [10, 11]. Parallelly, DHODH neutralizes lipid peroxidation by reducing CoQ to CoQH2 in the inner mitochondrial membrane, operating independently of non-mitochondrial compartments to counteract ferroptosis [12–14]. Therefore, it is ‌an intriguing approach to inducing ferroptosis by simultaneously disrupting multiple defense systems while elevating iron levels for cancer treatment.

Activation of and resistance to ferroptosis involve multiple redox reactions, in which the products of one reaction serve as reactants for another. This interdependence forms intricate chemical reaction networks (CRNs) that exhibit rich and complex kinetic behaviors, such as oscillations or multiple stability [15, 16]. Inspired by the transformation of complex substances within cells, various artificial CRNs have been developed to impart lifelike properties to materials, including oscillations [17, 18], out-of-equilibrium shifts [19, 20], and dissipative self-assembly [21, 22]. These CRNs can be precisely preprogrammed, enabling the induction of a range of predictable behaviors [23–25]. For instance, Andreas Walther et al. constructed a fuel-driven enzymatic chemical reaction network (ERN) using lipoic acid (LA) and dihydrolipoic acid (DHLA) redox pairs to program an autonomous thiol/disulfide redox system [26]. They used lipoamide dehydrogenase and β-nicotinamide adenine dinucleotide hydrate (NADH) as auxiliary reducing agents to regenerate LA into DHLA, thus effectively avoiding the direct annihilation of the reducing agent by the oxidizing fuel H₂O₂ (i.e., short-circuit reaction (SCR)) (Fig. 1a, b). Moreover, the ERN allows for programming hydrogel lifetimes after being linked to thiol-terminated star polymers. It is noteworthy that redox imbalance and NAD(P)H levels are closely associated with the occurrence of ferroptosis in tumor cells. On the one hand, when oxidative stress occurs in tumor cells, ROS oxidize polyunsaturated fatty acids to produce LPO, but this process is partially reversed by the GSH-GPX4 system [12, 27]. On the other hand, carbohydrate metabolism can produce NAD(P)H to regenerate CoQH2, which is essential for maintaining the function of FSP1 and DHODH [7, 9, 28]. Inspired by the CRN and the work of Andreas Walther et al., we envisage fuel-driven CRN in living organisms to facilitate the formation of new reaction pathways that can continuously undergo Fenton reactions to produce ROS and simultaneously affect NAD(P)H metabolism, thereby disrupting the three ferroptosis defense systems of tumor cells and causing a powerful “ferroptosis storm.”

Fig. 1.

Fuel-driven CRN based on MIL-100(Fe) for ferroptosis therapy. (a) Short-circuit reaction (SCR) between oxidizing and reducing agents in a redox-CRN, and (b) previous redox-CRN to install dissipative disulfides without SCR. (c) Our CRN aims to amplify oxidative stress intracellularly. (d) Schematic diagram of the synthetic processes of SRF@Au@M NPs. (e) Schematic illustration of SRF@Au@M NPs for devastating three ferroptosis defense systems via the formation of a new CRN intracellularly. (f) The components that formed the SRF@Au@M NPs. Created with BioRender.com

Metal-organic frameworks (MOFs) have been widely employed as nano-sized drug delivery systems (DDSs) for efficient loading of various therapeutic agents to treat cancer due to their high porosity, biocompatibility and excellent stability [29–32]. In addition, considering that MOFs facilitate easy surface engineering, as a proof of concept, this property may enable MOFs to integrate the components that constitute CRNs, thus exhibiting multifunctionality in ferroptosis therapy. In previous studies, most MOF-based ferroptosis DDSs focused on enhancing the Fenton or Fenton-like reaction and GSH depletion. Ferroptosis activation through GPX4 inhibition has been achieved in various non-catalytic GSH depletion pathways, such as MOF disintegration and reduction of high-valent metal ions (e.g., Fe(III), Cu(II), Mn(IV)) [33–37]. It is often overlooked that the efficacy of exogenous iron in activating ferroptosis is not directly proportional to the amount of ions administered, largely because the cellular uptake efficiency of the DDSs also significantly influences the therapeutic outcome [38, 39]. Another conflicting problem is that the depletion of the endogenous reducing agent GSH will downregulate the turnover frequency of Fe³⁺ to Fe²⁺, thereby inhibiting the occurrence of the Fenton reaction [40, 41]. More importantly, these MOF-based ferroptosis nanomedicines primarily target the disruption of the GSH-GPX4 pathway, but largely overlook the FSP1-CoQH2 and DHODH-CoQH2 defense systems. Therefore, by optimizing the MOF particle size and then integrating new CRNs based on the optimal particle size to achieve continuous Fenton reaction and interference with the ferroptosis defense system, this strategy can maximize therapeutic outcomes for ferroptosis treatment.

In this work, MIL-100(Fe) was prepared with controlled particle sizes ranging from 30 nm to 265 nm using a continuous dropwise method. Among these representative sizes, the 60 nm nanoparticles exhibited an optimal balance of stability, cellular uptake efficiency, drug loading capacity and toxicity, and were therefore selected for the subsequent construction of CRN and treatment of ferroptosis. Subsequently, a composite MIL-100(Fe)-based DDSs (denoted as SRF@Au@M NPs) was fabricated through sorafenib (SRF) loading, in situ growth of Au nanoparticles (Au NPs) and surface modification with dihydrolipoic acid derivatives (Fig. 1 d, f). Once internalized by cancer cells, the surface Au NPs mimic glucose oxidase (GOx) activity, consuming intracellular glucose and thereby disrupting key metabolic pathways, including glycolysis and the pentose phosphate pathway (PPP). This ultimately reduces NAD(P)H production in 4T1 cells. Consequently, this inhibits the recycling of CoQ to CoQH2, which acts as a critical antioxidant within the FSP1/DHODH-mediated anti-ferroptosis defense mechanism that traps lipid peroxyl radicals. Concurrently, the H₂O₂ generated from glucose decomposition serves as a fuel for the Fenton reaction, generating highly toxic hydroxyl radicals (•OH) in the presence of iron ions. Additionally, SRF@Au@M NPs disintegrate under high GSH concentrations, releasing SRF to inhibit system Xc⁻, thereby suppressing cystine uptake, GSH synthesis, and GPX4 activity, ultimately causing ferroptosis. Notably, the reduction of intracellular GSH content can effectively avoid the subsequent side reactions in CRNs, namely the aforementioned CRNs process between lipoic acid derivatives and GSH (Fig. 1b), thus facilitating the reduction of Fe³⁺ to Fe²⁺ within the CRN (Fig. 1c). Furthermore, the modification of dihydrolipoic acid derivatives on the SRF@Au@M surface serves three critical functions: (1) enhancing nanoparticle dispersibility and biocompatibility; (2) constructing CRNs via the DHLA/LA redox cycle, thereby sustaining the Fenton reaction and elevating ROS levels in tumor cells; (3) depleting NADH via the DHLA/LA redox cycle to lower CoQH2 levels, further downregulating the expression of the ferroptosis defense proteins DHODH and FSP1. Both in vitro and in vivo experiments revealed that the SRF@Au@M NPs efficiently suppressed tumor growth by simultaneously disrupting all three major ferroptosis defense systems (Fig. 1e). To the best of our knowledge, this represents the first report integrating CRNs into DDSs for ferroptosis therapy. The strategic implementation of CRNs in DDSs establishes a novel paradigm for tumor treatment.

Results and discussion

Size-controlled synthesis of MIL-100(Fe) nanoparticles

In accordance with the classical crystallization mechanism, the critical factor for synthesizing uniform MOFs is the separation of nucleation and growth [42, 43]. Typically, the process of nucleation and growth can be elucidated through the LaMer mechanism (Fig. 2a), which can be divided into three phases [44–46]. (I) A rapid increase in the concentration of free monomers in the solution leads to supersaturation, but no particles are present; (II) the monomer undergoes “burst nucleation”, whereby the monomer concentration is higher than the critical level of supersaturation and the concentration of free monomers in the solution is significantly reduced; (III) nanoparticles continue to grow and the concentration of monomer passes back below the critical level of supersaturation. Inspired by the LaMer mechanism, we employed a continuous-dropwise approach to achieve size-controlled synthesis of MIL-100(Fe) by leveraging its nucleation characteristics (Fig. 2b). In a typical synthesis, solutions containing the metal ions (FeCl₃·6H₂O) and organic linker (Benzene-1,3,5-tricarboxylic acid, BTC) were slowly and continuously added to a FeCl₃ solution at specific concentrations. Once the addition of the FeCl₃ and BTC was completed (this point is marked as 0 min), the feed of FeCl₃·6H₂O and BTC were halted but without stopping the reaction. A specific volume of the reaction solution was taken at 0 min, 10 min, 20 min, 30 min, and 40 min to monitor the size of the MOF. As displayed in Figure S4a-b, the nanoparticle size of MIL-100(Fe) remained stable at around 33 nm, indicating that particle growth ceased upon termination of the monomer supply. Consistent with this, the mass yield of MIL-100(Fe) also remained constant over time (Table S2). Specifically, the yield per 10 mL of reaction volume stabilized at around 15%, suggesting that the nanoparticles were uniformly dispersed in the system and did not undergo further growth or aggregation. Taken together, these results imply that the final MIL-100(Fe) particles in Stage III did not undergo Ostwald ripening. This further suggests that the reaction could be deliberately terminated at any point during Stage III, or alternatively, that the metal and linker could be replenished at any time to resume particle growth.

Fig. 2.

(a) The LaMer mechanism for controlled nanoparticle growth. (b) Continuous dropwise reactor for the synthesis of different sizes of MIL-100(Fe). (c) SEM images of MIL-100(Fe) with different sizes synthesized using the LaMer approach. The average particle size was measured by ImageJ software. At least 100 nanoparticles were measured for each sample. (d) Simulated and experimental PXRD patterns of MIL-100(Fe) of different sizes obtained from reactions with the 6 mM initial solutions and (e) 1.5 mM initial solutions. (f) Growth of MIL-100(Fe) versus the amount of FeCl₃ added in the parallel reaction batches. (g) Plot showing the linear dependence of D³ on the amount of FeCl₃ added. Diameters (D) are taken from the mean diameters in Fig. 2f. The “amount of FeCl₃” in this work is the amount of the reactants needed to be added into the initial solution to achieve the same reaction effect assuming that part of the reaction solution has not been taken out from the reaction bottle during the reaction process

In the nanoparticle growth stage, the total Gibbs free energy is primarily dependent on the bulk free energy (Figure S5) [43]. This indicates that the temperature (T), supersaturation of the solution (S), and its molar volume (V) exert a significant influence on the nanoparticle size. Based on above mechanism, different sizes of MIL-100(Fe) were obtained through a continuous-dropwise approach (for details see the support information and Table S1) with the growth temperature, supersaturation of the solution, and molar volume maintained constant throughout the process. As shown in Fig. 2c, eight different-sized products presented a uniform diameter distribution, and the size of MIL-100(Fe) increased with the addition of reactants. Powder X-ray diffraction (PXRD) results confirmed that the different sizes of MIL-100(Fe) exhibit a similar peak compared with the simulated patterns (Fig. 2d-e), indicating that MIL-100(Fe) has been successfully prepared. In addition, the initial solution concentration influenced the product size (Fig. 2f). 6 mM initial solution led to the formation of smaller nanoparticles (31 ± 4 nm) compared with 1.5 mM initial solution (96 ± 10 nm). Since nucleation occurs initially in the reaction, minor changes to the initial solution can significantly affect the nucleation process. A higher initial concentration of FeCl₃·6H₂O during the nucleation stage promotes the formation of a larger number of nuclei. Consequently, with the same total amount of reactants added, the average size of the resulting MIL-100(Fe) nanoparticles is reduced.

A linear correlation was observed between the cubic power of the average particle diameter (D³) of MIL-100(Fe) and the amount of FeCl₃ added during the growth stage (D³∝amount of added reactants, Fig. 2g). During this stage, the size of the particles is governed by the number of monomers. To maintain this linear relationship, two essential conditions must be satisfied: (I) The number of growing nanoparticles must remain constant, with no new nuclei forming. This prevents size polydispersity and is essential for preserving the linearity; (II) the supersaturation level (free monomer concentration) during growth must be kept constant, ensuring that monomer consumption is directly proportional to the amount of reactants added. The observed linear dependence of D³ on the added FeCl₃ confirms that nucleation and growth stages were successfully separated by controlling the degree of supersaturation. More importantly, this linear relationship allows us to calculate the exact amount of reactants required to synthesize MIL-100(Fe) of specific sizes over a wide range. Therefore, the growth of MIL-100(Fe) can be precisely predicted.

Cell uptake studies and the optimal particle size selection for MIL-100(Fe)

In ferroptosis therapy, the efficacy of exogenous iron-induced ferroptosis often does not show a linear correlation with the administered iron dose due to the suboptimal cellular uptake efficiency of nanoparticles. Therefore, identifying an optimal carrier size is crucial for maximizing therapeutic efficacy. While several studies have investigated the impact of MOF size on cellular uptake, no consensus has been reached. For instance, Park et al. successfully tuned the size of PCN-224 nanoparticles (30–190 nm) and demonstrated that 90 nm PCN-224 exhibited superior cellular uptake and photodynamic therapy (PDT) efficacy compared to other sizes [47]. Additionally, Liu et al. developed a completely aqueous method to precisely control the size of ZIF-8 at room temperature, revealing a size-dependent cellular uptake pattern in the order of 60 > 30 > 90 > 130 nm [48]. These findings highlight that cellular uptake selectivity can vary across different MOF types. Hence, it is necessary to optimize the MOF particle size to enhance therapeutic efficacy.

The three primary factors influencing nanoparticle cellular uptake are morphology, particle size, and surface physicochemical characteristics [49–51]. SEM analysis confirmed that the synthesized MIL-100(Fe) nanoparticles exhibit a uniform spherical morphology. Therefore, we further characterized their size and zeta potential. Four samples with distinct particle sizes (31, 58, 94, and 136 nm as determined by SEM) were selected and designated as xMIL-100(Fe), where “x” represents the specific particle size. Their hydrodynamic size, polydispersity index (PDI) and zeta potential were measured in both phosphate-buffered saline (PBS) and 10% fetal bovine serum (FBS) growth media. As shown in Figure S6 and Table 1, the hydrodynamic size in both media were significantly larger than the dry SEM sizes, and particles of different sizes exhibited distinct aggregation behaviors. For ₃₀MIL-100(Fe), the hydrodynamic size in PBS was approximately 3.5-fold greater than the SEM-measured size. In contrast, the hydrodynamic sizes of ₆₀MIL-100(Fe), ₉₀MIL-100(Fe) and ₁₃₅MIL-100(Fe) in PBS were approximately 2-fold larger than those measured by SEM. The greater aggregation of smallest nanoparticles may be attributed to the higher surface free energy. All four samples were well dispersed in PBS, with polydispersity index (PDI) values below 0.3. Aggregation was more pronounced in 10% FBS growth media compared to PBS. The hydrodynamic sizes of ₃₀MIL-100(Fe) and ₆₀MIL-100(Fe) increased approximately 6- and 4-fold, respectively, relative to their dry sizes, while those of ₉₀MIL-100(Fe) and ₁₃₅MIL-100(Fe) increased about 3-fold. These findings indicate that the smaller nanoparticles tend to exhibit a higher degree of aggregation. The PDI values were also generally higher in 10% FBS than in PBS. This phenomenon is likely due to the formation of a “protein corona” on the MOF surface, which can destabilize the particles and promote aggregation, leading to larger hydrodynamic sizes [52]. The zeta potential of the differently sized MIL-100(Fe) particles was approximately −18 mV in PBS and increased to around −11 mV in 10% FBS, following protein adsorption. These findings suggest that MIL-100(Fe) particles of different sizes exhibit nearly identical physicochemical properties. Consequently, we hypothesize that the observed differences in subsequent cellular uptake efficiency are primarily attributable to particle size.

Table 1.

Summary of physicochemical properties of MIL-100(Fe) with different particle sizes

Label	SEM (nm)	Size (nm)		PDI		z-potential (mV)
		PBS	10% FBS	PBS	10% FBS	PBS	10% FBS
30ML-100 (Fe)	32±6	114±10	203±6	0.1018	0.3774	−17.7±0.4	−11.9±0.8
60ML-100 (Fe)	58±7	120±10	234±5	0.1260	0.3504	−19±0.4	−11.1±0.3
90ML-100 (Fe)	94±11	199±30	297±11	0.2364	0.2806	−19.3±0.3	−10.8±0.7
135ML-100 (Fe)	136±14	282±25	386±4	0.2547	0.2357	−18.2±1.4	−11.5±0.5

To investigate the size-dependent cellular effects, four groups of NPs with distinct sizes (ranging from 30 to 135 nm as determined by SEM results) were applied to 4T1 cells. Quantitative analysis of nanoparticle uptake was performed via inductively coupled plasma optical emission spectrometry (ICP-OES) following cell digestion. As displayed in Fig. 3a, cellular uptake exhibited a time- and size-dependent pattern. The intracellular iron content gradually increased from 1 h to 24 h and displayed a highly size-dependent manner in the order of 30 > 60 > 90 > 135 nm. In addition, the cellular uptake reached a plateau around 12 h for 60, 90, and 135 nm fractions, whereas the intracellular level of ₃₀MIL-100(Fe) continued to increase significantly beyond this time point. The higher apparent cell uptake for smaller MOFs may be attributed to their larger specific surface area compared to larger ones, leading to stronger nanoparticle–cell interactions and enhanced internalization.

Fig. 3.

(a) Cellular uptake kinetics of MIL-100(Fe) nanoparticles of different sizes. (b) Confocal images and (c) corresponding flow cytometric analysis of 4T1 cells treated with DOX@M nanoparticles of different sizes at an equivalent DOX concentration (5 µg/mL) for 3 h. (d) SRF loading capacity of MIL-100(Fe) nanoparticles with different sizes. (e) Flow cytometric analysis of apoptosis in 4T1 cells under different treatment conditions. (f) Radar chart evaluating the multi-parameter performance of MIL-100(Fe) nanoparticles with different sizes

Subsequently, the doxorubicin (DOX), a chemotherapeutic agent with red fluorescence at 620 nm, was loaded into all the MOFs (labeled _XDOX@M, where x represents the particle size) to further investigate cellular uptake efficiency by confocal laser scanning microscopy (CLSM) and flow cytometry. Prior to drug loading, we measured the specific surface area and pore size distribution (PSD) of the carriers. As shown in Figure S7a, the N₂ adsorption-desorption isotherms indicate favorable porosity. Both the Brunauer–Emmett–Teller (BET) surface area and Langmuir surface area exceeded 1000 m²/g for all particle sizes, suggesting promising drug-loading potential. The PSD reveals that the primary pore diameters of MIL-100(Fe) across various particle sizes are below 2 nm (Figures S7b), confirming a predominantly microporous structure. Figure 3b shows the CLSM images of 4T1 cells incubated with each DOX@M for 3 h. It can be observed that ₃₀DOX@M presented the highest intracellular fluorescence, followed by ₆₀DOX@M, ₉₀DOX@M, and ₁₃₅DOX@M (Figure S8). Similarly, we found that the DOX fluorescence intensity in cytoplasm decreased with the increased size of DOX@M, as detected by flow cytometry (Fig. 3c). These results are consistent with the ICP-OES measurements, indicating that the loading of drugs does not affect the internalization of MIL-100(Fe).

To inform subsequent ferroptosis therapy, SRF was loaded into MIL-100(Fe) (labeled _XSRF@M, where x represents the particle size) and its cytotoxicity was investigated. SRF@M was prepared by mixing MIL-100(Fe) with SRF at a mass ratio of 1:2. To our surprise, the drug loading efficiency (DLE) of SRF exhibited unexpected differences. As displayed in Fig. 3d, the DLE was determined to be 2.8%, 6.9%, 6.8% and 7.5% for ₃₀SRF@M, ₆₀SRF@M, ₉₀SRF@M, and ₁₃₅SRF@M, respectively. Apparently, ₃₀SRF@M displayed the lowest drug loading efficiency. A probable explanation might be that the 30 nm nanoparticles are in a state where the crystal core has just formed, and the crystal has undergone only initial growth. As a result, their internal porous structure is not yet fully developed, which hinders the drug’s ability to enter the internal cavity. In this case, SRF may only be attached to the external surface or to potential superficial defects of the MIL-100(Fe). This makes the drug prone to premature release during blood circulation, potentially causing off-target toxicity. Actually, the PXRD patterns of MIL-100(Fe) showed much sharper and stronger peaks as the particle size raises (Fig. 2d and e), indicating that the crystallinity of MIL-100(Fe) gradually increases. During the subsequent crystal growth process, the DLE ceased to increase and stabilized at approximately 7%.

The anticancer capability of differently sized SRF@M particles was then evaluated in 4T1 cells at an SRF concentration of 6 µg/mL. As anticipated, the efficacy decreased monotonically decreased as the particle size increased. Cell mortality rates were 46.1%, 27.9%, 17.2%, and 12.8% for 30, 60, 90, and 135 nm particles, respectively (Fig.3e). It is noteworthy that free SRF resulted in 13.1% cell death, which was close to ₁₃₅SRF@M but lower than those of other SRF@M composites. This may be ascribed to the combination of poor solubility of SRF and higher uptake efficiency of MIL-100(Fe). These findings underscore that optimizing MOF particle size is critical for enhancing anticancer efficacy.

It should also be noted that the gaps in tumor vasculature typically range from 100 nm to 2 μm. For effective drug delivery, nanomedicines must not only extravasate from the blood vessels but also be retained within the tumor tissue. While smaller particles (e.g., ~ 30 nm) can easily extravasate, they can also readily diffuse back into the bloodstream, preventing effective enrichment at the tumor site [39, 53]. To comprehensively evaluate and compare the key properties—including stability, cellular uptake, drug loading efficiency, and cytotoxicity—of the nanomaterials with different particle sizes, we constructed a radar chart (Fig. 3f). The analysis revealed that the 60 nm nanomaterial exhibited the most balanced and favorable overall profile. Therefore, the 60 nm MIL-100(Fe) was selected for all subsequent ferroptosis therapy experiments.

Fabrication and characterization of SRF@Au@M NPs

MOF-based ferroptosis-induced nanoparticles were synthesized in three main steps, SRF loading, in situ growth of Au NPs, and subsequent surface modification (Fig. 1d). To enhance the drug loading efficiency, the influence of the mass ratio of MIL-100(Fe) to SRF was investigated. As shown in Table S3, an increased mass of SRF may facilitate enhanced drug loading. The highest loading efficiency of 11.9% was achieved via the SRF standard curve (Figure S9) when the mass ratio of MIL-100(Fe) to SRF was 2:1. Following loading into MIL-100(Fe), the sharp peak of SRF at 266 nm disappeared in the SRF@M formulation (Fig. 4a). Then, Au NPs were deposited on the SRF@M surface in situ by direct reduction of HAuCl₄ in ice water with NaBH₄. As exhibited in Figure S10, dark spots corresponding to the Au NPs (~2 nm in diameter) could be observed on the external surface of SRF@M, accompanied by a color change from brown to burgundy (Fig. 4a). Besides, compared to SRF@M, SRF@Au@M exhibits a new broad and shallow absorption peak in the 500–600 nm range, which is attributed to the Au NPs’ surface plasmon resonance (Fig. 4a) [54]. To further enhance the colloidal stability of the resultant nanoparticles and potentiate ferroptosis therapy, we synthesized two lipoic acid derivatives: sodium dihydrolipoic acid (DHLA-Na) and PEGylated dihydrolipoic acid (DHLA-PEG) for subsequent surface modifications (Figure S1-S3). Their structures were characterized by ¹H-NMR, ¹³C-NMR and electrospray ionization-high resolution mass spectrometry (Figure S11-S17). The aggregation of SRF@Au@M was effectively inhibited following surface modification with lipoic acid derivatives (Figure S18 and Table S4). The hydrodynamic size of the unmodified SRF@Au@M was 239 nm with a PDI of 0.3255, whereas the corresponding values for SRF@Au@M NPs were 151 nm and 0.2407, respectively (Fig. 4b). It is noteworthy that the PEGylation modification of SRF@Au@M NPs effectively reduces the adsorption of proteins in serum, which is conducive to the prolonged circulation of nanoparticles in vivo and thus maximizes their therapeutic efficacy (Fig. 4f).

Fig. 4.

(a) UV-vis spectra of SRF, Au NPs, SRF@M, SRF@Au@M and SRF@Au@M NPs, the inset shows the corresponding solution images. (b) DLS results of SRF@Au@M and SRF@Au@M NPs. (c) Zeta potentials of MIL-100(Fe) during various modifications. (d) TEM image of SRF@Au@M NPs. (e) Elemental mapping images of the SRF@Au@M NPs. (f) SDS-PAGE gel of adsorbed protein corona on SRF@Au@M and SRF@Au@M NPs incubated with bovine plasma. (g) XPS spectra of SRF@Au@M NPs. (h) High-resolution Au4f spectra of SRF@Au@M NPs. (i) Hydrophilic diameters and PDI of SRF@Au@M NPs dispersed in water and PBS for 7 days

Transmission electron microscopy (TEM) images reveal that SRF@Au@M NPs have a uniform spherical morphology with an average diameter of 60 nm (Fig. 4d), aligning with the optimized particle size of MIL-100(Fe). The nanoparticles surface is decorated with a layer of ultra-small gold nanoparticles. Energy-dispersive X-ray spectroscopy (EDS) elemental mapping confirms that the characteristic elements—C, Fe, and O (from MIL-100(Fe)), F (from SRF), Au (from the gold nanoparticles), and S (from the lipoic acid derivatives)—throughout the SRF@Au@M NPs (Fig. 4e), demonstrating successful drug loading onto the MIL-100(Fe) framework. Additionally, the mass fractions of Fe, Au and S measured by ICP-OES were 8.4%, 11% and 1.6%, respectively (Table S5). The UV-vis spectrum displayed that the SRF@Au@M NPs contain characteristic peaks of SRF@M from 200 to 400 nm and Au NPs from 500 to 600 nm (Fig. 4a). Meanwhile, the zeta potential changes reflected the fabrication process of the nanoparticles (Fig. 4c). The X-ray photoelectron spectroscopy (XPS) spectra of SRF@Au@M NPs were also obtained (Fig. 4g), and the C1s, Fe 2p, N 1 s, O 1 s, Au 4f, and S2p peaks were observed in the spectra. The high-resolution Au 4f spectra in Fig. 4h revealed two peaks at 83.8 and 87.5 eV, corresponding to the Au 4f7/2 and Au 4f5/2 signals of metallic gold, indicating that the Au NPs on the nanoparticles existed in the metallic Au⁰ state [55]. Finally, we found that SRF@Au@M NPs could be well dispersed and stored in water or PBS for 7 days with no significant alterations in size or PDI (Fig. 4i), indicating that the nanoparticles possessed good stability. All the above results demonstrate that SRF@Au@M NPs have been successfully prepared. The hydrodynamic size, zeta potential, and cellular uptake behavior of SRF@Au@M NPs with different particle sizes were also evaluated. As shown in Figure S19, the hydrodynamic size of SRF@Au@M NPs increased with the original MIL-100(Fe) size, and all formulations exhibited a PDI below 0.3, indicating good dispersion. The zeta potentials of all four sizes were negative, ranging from −17 to −22 mV. More importantly, after 3 h of incubation, the intracellular iron content in 4T1 cells still showed a clear size-dependent pattern, consistent with that observed before surface modification and Au deposition. These results confirm that particle size remains a critical factor influencing cellular uptake even after the introduction of Au NPs and dihydrolipoic acid derivatives.

The integration of Fe species and Au NPs into the MOFs system is anticipated to effectively deplete GSH and glucose levels within tumor cells, which plays ‌crucial‌ roles in the subsequent ferroptosis therapy. First, we evaluated the GSH-scavenging activity of the SRF@Au@M NPs using 5,5′-dithiobis-2-(nitrobenzoic acid) (DTNB), a compound that reacts with GSH to form yellow products with a characteristic absorption peak at 412 nm (Fig. 5a). As illustrated in Figure S20 and Fig. 5b, the absorbance at 412 nm decreased in a dose-dependent manner after 1 h and 4 h of incubation with SRF@Au@M NPs. At the highest nanoparticle concentration, 36% of GSH was depleted after 4 h. Meanwhile, after 12 h of incubation, a notable increase in hydrodynamic diameter (484.9 nm) and PDI (0.7755) was observed (Fig. 5c). TEM images also displayed an irregular, diffuse morphology instead of the uniform spherical morphology. These results suggested that the strong GSH-consuming ability of SRF@Au@M NPs. The release profiles of SRF from SRF@Au@M NPs under different conditions were investigated. As shown in Figure S21, the cumulative release of SRF was less than 15% at PBS within 24 h, indicating good stability under physiological conditions. In contrast, in the presence of 2 mM GSH, the cumulative release increased to approximately 26% within 24 h. When the GSH concentration was raised to 10 mM, the cumulative release was further enhanced, reaching about 65%. This demonstrates a clear dependence of SRF release on GSH concentration. Fe³⁺ can deplete GSH while being reduced to Fe²⁺, which exhibits Fenton reactions activity to generate •OH by catalyzing H₂O₂.To investigate the ability of the SRF@Au@M NPs to generate ROS under GSH trigger, 3,3’,5,5’-tetramethylbenzidine (TMB) served as the chromogenic probe to detect the generation of •OH, as its solution color turns blue with enhanced absorbance at 652 nm after oxidation (Fig. 5d). As shown in Fig. 5e, in the presence of both GSH and H₂O₂, the SRF@Au@M NPs were capable of oxidizing the transparent TMB to blue oxTMB, and the absorbance of oxTMB at 652 nm gradually increased over time (Figure S22), indicating that the SRF@Au@M NPs exhibit Fenton reaction activity triggered by GSH. To evaluate the catalytic performance, steady-state kinetic studies were conducted with different TMB and H₂O₂ concentrations. As shown in Fig. 5f-i, a typical Michaelis-Menten curve was obtained. The K_m and V_m values were calculated as 1.56 mM and 8.57 × 10⁻⁸ mol/s for H₂O₂, respectively, through the double reciprocal plot method. The K_m value was lower than those of reported ferrous-based nanoenzymes (Table S6), indicating a high affinity of SRF@Au@M NPs for H₂O₂. Besides, the electron spin resonance (ESR) spectra were recorded to distinguish the formation of •OH rather than other ROS. Figure 5j shows that the SRF@Au@M NPs+H₂O₂ group exhibited a typical four-line spectrum with characteristic peaks at relative intensities of 1:2:2:1, confirming that •OH was produced via the Fenton reaction.

Fig. 5.

(a) Schematic of DTNB reduction by GSH. (b) GSH depletion by SRF@Au@M NPs characterized by the absorbance of DTNB at 412 nm. (c) DLS of SRF@Au@M NPs incubated with 10 mM GSH for 12 h. The inset shows the corresponding TEM images. (d) Schematic of TMB oxidation by •OH. (e) UV − vis absorption spectra of TMB in the presence of different formulations. The inset shows the corresponding photograph of solutions. (f-i) Michaelis−Menten kinetic analysis and Lineweaver-Burk plotting of POD-mimicking activity with H₂O₂ or TMB as substrates, respectively. Data presented as mean ± S.D. (n = 3). (j) ESR spectra for the detection of •OH radical in different groups. (k) The pH values of the SRF@Au@M NPs reacted with glucose at different time points. Data presented as mean ± S.D. (n = 3). (l) UV − Vis absorption spectra of TMB in the presence of different formulations

In our therapeutic platform, cascade reactions enable the use of glucose as a source of chemical fuel to generate •OH. Au NPs can act as GOx-mimicking enzymes, oxidizing glucose to produce gluconic acid and H₂O₂ [56]. To confirm the GOx-like ability of the Au NPs within SRF@Au@M NPs, SRF@Au@M NPs were incubated in a 5 mg/mL aqueous glucose solution, and the pH was monitored over time. As shown in Fig. 5k, the pH of the solution containing both glucose and SRF@Au@M NPs decreased from 6.78 to 4.94 after 24 h, indicating the effective catalytic oxidation of glucose to gluconic acid. Furthermore, we found that the Glu + SRF@Au@M NPs incubation with TMB resulted in a remarkable enhancement in the solution absorbance at 652 nm compared to the Glu + TMB or SRF@Au@M NPs + Glu groups (Fig. 5l). Additionally, the absorbance of TMB at 652 nm displayed a time-dependent increase trend (Figure S23), indicating the generation of H₂O₂. Collectively, these findings confirm that the Au NPs in our nanoparticle can efficiently mimic glucose oxidase to catalyze the production of H₂O₂ from glucose.

Fuel-driven chemical reaction network in vitro

Classical Fenton reactions refer to the decomposition of H₂O₂ initiated by metal ions in a low oxidation state, which results in the production of hydroxyl radicals. The high-valent ions formed in this process are subsequently reduced back to their low-valent state via H₂O₂. Although the reduction reaction is significantly slow, the catalyst regeneration is considered the most critical process because it reduces radical production and can deplete the availability of low-valent ions for the reaction [57]. In this work, we envisioned an effective strategy that introduces a dihydrolipoic acid derivative as a cocatalyst into the iron-based MOF to enhance the Fenton efficacy by accelerating the Fe³⁺/Fe²⁺ cycles. To verify this hypothesis, the Fe²⁺ contents in 4T1 cells incubated with two different formulations (SRF@Au@M and SRF@Au@M NPs) were investigated through Ferro Orange staining, a fluorescent probe selective for Fe²⁺. As shown in Fig. 6a, the intracellular Fe²⁺ fluorescence intensity was significantly higher in both treatment groups than in the control group, owing to the introduction of exogenous iron. After 4 h, the fluorescence intensity of Fe²⁺ in the nanoparticle group containing dihydrolipoic acid derivatives (SRF@Au@M NPs) was higher than that of the group without it (SRF@Au@M). To our surprise, after 12 h, the fluorescence intensity decreased in both groups, but it remained significantly higher in the SRF@Au@M NPs group than in the SRF@Au@M group. Quantitative analysis demonstrated that the fluorescence intensity of the SRF@Au@M NPs group was 1.9-fold and 1.6-fold higher than that of SRF@Au@M group at 4 h and 12 h, respectively (Fig. 6b). Based on the high concentration of GSH in tumor cells and the SRF loaded in our nanoparticles, we proposed the mechanistic hypothesis of “slow versus fast” to “weak versus strong” stepwise regulation of intracellular Fe²⁺ content (Fig. 6c). Specifically, upon administration, the nanoparticles rapidly disintegrate in the high-GSH tumor microenvironment, which reduces Fe³⁺ to Fe²⁺. At this initial stage, cells treated with SRF@Au@M NPs form DHLA/LA redox couples via lipoamide dehydrogenase, leading to faster Fe³⁺ reduction and, consequently, stronger red fluorescence at 4 h compared to the control. This constitutes the “slow versus fast” regulatory phase. Following nanocarrier disintegration, the released SRF inhibits cellular cystine uptake by targeting the membrane transporter SLC3A2, thereby reducing the intracellular synthesis of GSH. Consequently, the decreased GSH content impairs the conversion of Fe³⁺ to Fe²⁺, resulting in a decrease in fluorescence intensity in both groups after 12 h compared to 4 h. However, owing to the presence of DHLA/LA redox couples, the SRF@Au@M NPs group exhibited a stronger reduction ability, maintaining higher intracellular Fe²⁺ content. Accordingly, this later stage represents the “weak versus strong” regulatory process. To further verify this hypothesis, intracellular GSH contents were tested at different times. As illustrated in Fig. 6d, compared to the control group, GSH content in cells treated with either SRF@Au@M or SRF@Au@M NPs decreased significantly after 4 h and further declined to approximately 60% after 12 h. In contrast, the GSH level in 4T1 cells treated with blank MIL-100(Fe) decreased to about 85% at 4 h but recovered to nearly 96% by 12 h (Fig. 6e). These results reveal that intracellular GSH depletion is co-regulated by iron-based MOFs and the chemotherapeutic agent SRF. It is worth noting that intracellular GSH and GSSG (oxidized GSH) can form competing redox circuits with the DHLA/LA couple redox (Fig. 1b), which would hinder the reduction of Fe³⁺ to Fe²⁺. Therefore, depleting intracellular GSH content via SRF treatment can effectively minimize this competing side reaction, favoring Fe³⁺ reduction by DHLA. To confirm the conversion of DHLA to LA, we analyzed the cell lysate via HPLC-MS after incubation with SRF@Au@M NPs and 4T1 cells. As shown in Figure S24 and Fig. 6f, a distinct LA peak was observed at a retention time of 7.33 min, with a measured molecular weight of 205.0358, which matches the theoretical value for LA. This indicates that DHLA is smoothly converted to LA, concurrently reducing Fe³⁺.

Fig. 6.

(a) CLSM images showing intracellular Fe²⁺ (detected by Ferro Orange probe) in 4T1 cells treated with SRF@Au@M (I) and SRF@Au@M NPs (II). (b) Quantification of Fe²⁺ levels analyzed using ImageJ. Data presented as mean ± S.D. (n = 3). (c) Mechanism of intracellular Fe²⁺ regulation. (d) Intracellular relevant GSH content after treatment with SRF@Au@M or SRF@Au@M NPs at different time points. (e) Intracellular relevant GSH content after treatment with MIL-100(Fe) at different time points. (f) ESI-HRMS spectrum of 4T1 cell lysates after co-incubation with SRF@Au@M NPs. (g) CLSM images of intracellular H₂O₂ level after different treatments. (h) Relative fluorescence intensity of H₂O₂ in 4T1 cells treated with different formulations. (i) Flow cytometric analysis of cellular ROS content (DCFH-DA probe) in 4T1 cells after various treatments. All data were presented as mean ± S.D. (n = 3). *p < 0.05, **p < 0.01, ***p < 0.001

As described above, Au NPs exhibit GOx-like activity, oxidizing glucose to produce H₂O₂, which serves as the chemical fuel for the Fe²⁺-mediated Fenton reaction. Based on this property, we first examined the H₂O₂ levels in 4T1 cells using the Amplex Red (ADHP) probe following treatment with different formulations. As exhibited in Fig. 6g-h, compared to the control, SRF, MIL-100(Fe), and SRF@M groups, cells treated with SRF@Au@M showed significantly higher red fluorescence, confirming that H₂O₂ levels can be elevated through Au NPs-catalyzed glucose oxidation. In addition, we found that SRF@Au@M NPs treatment groups exhibited higher red fluorescence than other groups but lower than the SRF@Au@M treatment groups. This phenomenon can be attributed to the accelerated Fe³⁺/Fe²⁺ cycles facilitated by DHLA/LA redox couple, which accelerates the decomposition of the generated H₂O₂. Furthermore, the intracellular ROS generation capacity of the two nanoparticle formulations was evaluated using the fluorescent indicator 2,7-dichlorodihydrofluorescein diacetate (DCFH-DA). As presented in Fig. 6i, flow cytometry results indicated that SRF@Au@M and SRF@Au@M NPs increased ROS levels in 4T1 cells from 10.2% to 27.1% and 50.4%, respectively, after 4 h of treatment. However, after 12 h, the ROS level in the SRF@Au@M group returned to control levels, whereas it remained approximately 1.6-fold higher than the control in the SRF@Au@M NPs group. It is worth emphasizing that ROS levels typically decrease rapidly once they reach the highest level, due to the intrinsic antioxidant and antiferroptotic properties of cancer cells [58, 59]. The sustained high ROS levels observed in our system highlight the ability of CRNs to maintain ROS production over time.

Transcriptomic analysis of 4T1 cells after treatment with SRF@Au@M NPs

The construction of novel CRNs has the potential to influence anabolism, catabolism and energy metabolism of tumor cells. In particular, the reprogramming of carbohydrate and amino acid metabolism in cancer cells may facilitate the induction of ferroptosis [28]. To gain more comprehensive understanding the effects of the CRNs based on MIL-100(Fe), as constructed, on tumor cell metabolism, a whole transcriptome analysis of 4T1 cells was performed following incubation with SRF@Au@M NPs. As shown in Figure S25, Spearman correlation coefficients between parallel samples were greater than 0.92, indicating the high quality and reliability of the data, which could be applied to further analysis. The overall distribution of differentially genes was characterized by a volcano plot, as displayed in Figs. 7a and 2820 significantly differentially expressed genes (|log2FoldChange|>=1, p < 0.01) were detected in 4T1 cells after treatment with SRF@Au@M NPs, with 1260 genes upregulated and 1560 genes downregulated. Meanwhile, the Venn diagram demonstrated that 565 novel genes were unique to 4T1 cells after treatment with SRF@Au@M NPs (Fig. 7b). Furthermore, Gene Ontology (GO) enrichment analysis revealed that these differentially genes associated with “cellular response to starvation”, “regulation of ferroptosis”, “cell redox hemostasis”, “glucose metabolic process”, “response to oxidative stress”, and “carbohydrate metabolic process” (Fig. 7d). This indicates that our constructed CRNs have a profound impact on cellular metabolic processes. Firstly, enhanced oxidative stress disrupts intracellular redox homeostasis, promoting lipid peroxidation and inducing ferroptosis. Secondly, glucose deprivation may impair the PPP, reducing NADPH levels and thereby weakening the CoQH2-mediated defense system against ferroptosis. Next, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was conducted to identify significantly altered signaling pathways. As shown in Fig. 7e, SRF@Au@M NPs evidently influenced the PPP and ferroptosis-related pathways. Clustered heatmap analysis demonstrated that a total of 16 genes associated with ferroptosis were significantly altered in 4T1 cells after treatment with SRF@Au@M NPs (Fig. 7c). These results revealed that SRF@Au@M NPs trigger ferroptosis in 4T1 cancer cells through multiple mechanisms.

Fig. 7.

(a) The volcano plot of differentially expressed genes in 4T1 cells treated with SRF@Au@M NPs. (b) Venn diagram of various cell populations. (c) Clustering heatmap of differentially expressed genes from ferroptosis. (d) Gene Ontology (GO) functional enrichment analysis of tumor cells after treatment with SRF@Au@M NPs. (e) KEGG enrichment analysis of differentially expressed genes in 4T1 cells treated with SRF@Au@M NPs

In vitro mechanism of ferroptosis and cytotoxicity

Given the importance of ROS generation in ferroptosis therapy, we further investigated the ROS content in 4T1 cells using the DCFH-DA probe after treatment with different formulations. As shown in Fig. 8a, formulations containing Au NPs (SRF@Au@M and SRF@Au@M NPs) exhibited notably stronger green fluorescence compared to the other groups, indicating higher ROS production. Notably, the SRF@Au@M NPs groups showed the strongest ROS upregulation, with a mean fluorescence intensity 9.6-fold higher than that of the control and significantly higher than the SRF@Au@M group, indicating its superior ability to ROS-inducing ability (Figure S26). In addition, intracellular •OH, which is generated via the Fenton reaction and subsequently oxidizes polyunsaturated fatty acids (PUFAs) to induce LPO, was monitored using coumarin-3-carboxylic acid (3-CCA). Consistent with the intracellular ROS analysis, the SRF@Au@M NPs-treated group displayed the highest •OH levels among all groups (Fig. 8a, b). Encouraged by the efficient ROS generation and GSH depletion abilities of SRF@Au@M NPs, the crucial protein GPX4 in GSH-GPX4 system for defense against ferroptosis was examined. As depicted in Fig. 8h and Figure S29a, compared with the control, SRF, MIL-100(Fe), SRF@M and SRF@Au@M groups, the SRF@Au@M NPs group exhibited the lowest GPX4 expression. This indicates that the GSH consumption and ROS generation induced by SRF@Au@M NPs act synergistically to suppress GPX4 expression in 4T1 cells.

Fig. 8.

(a) DCFH-DA and 3-CCA stained of 4T1 cells after treatment with different formulations. (b) Flow cytometric analysis of intracellular •OH content (3-CCA probe). (c-f) Detection of NADPH content, NADH content, CoQH2 content, and CoQH2 relative to CoQ content in 4T1 cells treated with control, SRF, MIL-100(Fe), SRF@M, SRF@Au@M, and SRF@Au@M NPs. (g) Cellular ATP level in 4 T1 cells incubated with SRF@Au@M NPs. (h) Western blot analysis of GPX4, FSP1 and DHODH expression. (i) Detection of intracellular LPO using the C11-BODIPY^581/591 probe. (j) Intracellular MDA levels after different treatments. (k) Cytotoxicity assays of 4T1 cells with different treatments. (l) Cell viabilities determined upon coincubation with SRF@Au@M NPs in the presence of different inhibitors, including z-VAD (5 µM), NER (50 µM), CQ (1 µM), DFO (100 µM), and Lip-1 (25 µM). All data were presented as mean ± S.D. (n = 3). *p < 0.05, **p < 0.01, ***p < 0.001

Previous studies have shown that certain cancer cell lines retain resistance to ferroptosis even after GPX4 inactivation, implying the presence of extra ferroptosis defense mechanisms. FSP1 is another anti-ferroptosis protein localized on the plasma membrane, acting independently of GPX4. FSP1 functions as an NAD(P)H-dependent oxidoreductase capable of reducing CoQ to CoQH2, which can scavenge lipid peroxyl radicals, thereby suppressing LPO and ferroptosis. DHODH is another ferroptosis defense system situated in mitochondrial compartments. When GPX4 is acutely inactivated, DHODH activity is significantly enhanced, leading to increased CoQH2 production, which suppresses LPO and protects mitochondria from ferroptosis. Therefore, simultaneous inactivation of both the GPX4 and the CoQH2 systems may trigger a robust ferroptosis storm. The PPP, a glucose metabolic pathway parallel to glycolysis, serves as a major source of intracellular NADPH through its oxidative decarboxylation phase [28, 60]. This pathway supplies reducing equivalents for various biosynthetic reactions, including the reduction of CoQ to CoQH2. As discussed above, the presence of Au NPs enables SRF@Au@M NPs to effectively catalyze glucose oxidation into gluconic acid. We found that modification with Au NPs effectively reduced the intracellular glucose level to approximately 80%, which was significantly lower than that in the other groups (Figure S27), leading to a decrease in PPP flux and thus in NADPH production. Herein, the NADPH contents were tested in 4T1 cells after treatments with different formulations. As shown in Fig. 8c, we found that MIL-100(Fe) and SRF@M displayed NADPH-depletion capacity compared to the control and SRF groups, which may be attributed to the electron transfer from NADPH to Fe³⁺. Upon co-incubation of MIL-100(Fe) with NADPH, the absorbance of NADPH at 340 nm gradually decreased over time, demonstrating that MIL-100(Fe) possesses intrinsic NADPH oxidase-mimicking activity. Notably, after in situ growth of Au NPs and modification with dihydrolipoic acid derivatives, SRF@Au@M NPs retained this activity (Figure S28a and c).The SRF@Au@M and SRF@Au@M NPs groups further reduced the intracellular NADPH to 33.7% and 32.2%, respectively, which was significantly lower than those in the MIL-100(Fe) and SRF@Au@M groups. This indicates that Au NPs could consume glucose and markedly affect the PPP, thereby inhibiting NADPH generation. Of note, NADPH serves as a cofactor for glutathione reductase, which is responsible for reducing GSSG back to GSH [61]. Therefore, modification with Au NPs leads to a further decrease in GPX4 expression (Fig. 8h).

Another cofactor involved in the reduction of CoQ to CoQH2 is NADH. In Our CRNs, lipoamide dehydrogenase utilizes NADH as an auxiliary substrate to regenerate dihydrolipoic acid (DHLA) from lipoic acid (LA), which serves as the reducing age. Consequently, the cycling of the DHLA/LA redox pair leads to continuous consumption of intracellular NADH. We next evaluated the NADH content in 4T1 cells. As shown in Fig. 8 d, SRF-treated cells exhibited a slight increase in NADH content compared with untreated controls, suggesting that cells can maintain NADH metabolism homeostasis under constant chemotherapeutic stimulation. Similarly, MIL-100(Fe) and SRF@M reduced the intracellular NADH to 69% and 88.7%, respectively, which were significantly lower than the control group. Previous studies have shown that iron-containing materials possess excellent NADH-like oxidase mimic activity, effectively oxidizing NADH to NAD⁺ [62]. In addition, Moran et al. recently revealed that Fe^2+/3+ can act as a Lewis acid, coordinating with pyruvate and pre-organizing its interaction with NADH, thereby explaining the observed stereoselectivity resulting from an intramolecular hydride transfer pathway [63].We also found that MIL-100(Fe) and SRF@Au@M NPs possess NADH oxidase-like activity(Figure S28b and d). These findings could account for the NADH-scavenging functions of MIL-100(Fe) and SRF@M. SRF@Au@M further reduced the intracellular NADH content to 54%. This may be attributed to the suppression of the glycolysis pathway via Au NPs and the consequent decline in NADH production. The NADH content was significantly lower than the other groups after treatment with SRF@Au@M NPs, indicating that the introduced dihydrolipoic acid contributes to additional intracellular NADH consumption. A decrease in NAD(P)H may impair ATP production by affecting the oxidative phosphorylation process or substrate-level phosphorylation processes. As shown in Fig. 8g, treatment with SRF@Au@M NPs reduced the relative ATP content to 53.4%, indicating that NAD(P)H depletion compromised ATP generation.

Next, we determined the change in cellular CoQH2 content. As shown in Fig. 8e, the trend was consistent with the change in NAD(P)H levels.The CoQH2 level in 4T1 cells treated with SRF@Au@M was significantly lower than in those treated with Control, SRF, MIL-100(Fe) or SRF@M. This reduction can be attributed to the synergistic effect of iron and Au NPs on NAD(P)H consumption. In addition, the SRF@Au@M NPs group showed a superior ability to reduce CoQH2 content compared to the SRF@Au@M group. This confirms that the presence of dihydrolipoic acid derivatives enhances NADH consumption. Correspondingly, the SRF@Au@M NPs group showed the lowest CoQH2/CoQ ratio (Fig. 8f). Subsequently, we investigated the effect of CoQH2 imbalance on ferroptosis by examining the expression of FSP1 and DHODH in 4T1 cells after different treatments. As showed in Fig. 8h and Figure S29b-c, western blot results revealed that both SRF@Au@M and SRF@Au@M NPs remarkably decreased the FSP1 and DHODH expression, with the lowest levels observed in the SRF@Au@M NPs group. This phenomenon showed that our constructed DDSs can disrupt both the FSP1-CoQH2 system and the DHODH-CoQH2 system by NAD(P)H homeostasis in tumor cells.

The results above indicate that our constructed CRNs based on MIL-100(Fe) could destroy the three ferroptosis defense systems in 4T1 cells, leading to substantial LPO production. Next, cellular LPO was examined using the C11-BODIPY^581/591 probe (Fig. 8i). Compared with the control, SRF, and MIL-100(Fe) groups, the SRF@M group showed moderate LPO generation. Meanwhile, SRF@Au@M treatment further enhanced LPO generation. Specifically, SRF@Au@M NPs-treated cells displayed higher green fluorescence and lower red signals than SRF@Au@M-treated cells, demonstrating that dihydrolipoic acid derivatives in our DDSs can promote LPO induction. Moreover, malondialdehyde (MDA), the terminal smaller molecular byproducts of LPO decomposition, was also tested to assess ferroptosis. As shown in Fig. 8j, the MDA content in cells treated with SRF@Au@M NPs was 32.8 µM/g protein, approximately three times higher than that in the control group and significantly higher than that in the SRF@Au@M group. These results confirm that the MOF-based CRNs comprehensively and synergistically induce ferroptosis.

Consequently, the designed SRF@Au@M NPs achieved a powerful “ferroptosis storm” by simultaneously disrupting three defense systems. We then assessed the therapeutic efficacy of SRF@Au@M NPs in vitro using a CCK-8 assay. As presented in Fig. 8k, the 4T1 cells treated with the SRF@Au@M NPs exhibited the lowest viability among all groups, demonstrating the synergistic effect of its components. Meanwhile, the calcein-AM and propidium iodide (PI) staining (Figure S30) showed that the SRF@Au@M NPs group had fewer living cells (green color) and more dead cells (red color), further confirming the strongest antitumor efficacy of SRF@Au@M NPs treatment. To verify that cell death occurred primarily via ferroptosis, inhibitor rescue experiments were conducted. As displayed in Fig. 8l, inhibitors of apoptosis (Z-VAD), autophagy (chloroquine, CQ), and necroptosis (necrostatin-1, Nec-1) all failed to protect the cancer cells. In contrast, the ferroptosis inhibitors deferoxamine (DFO) and liproxstatin-1 (Lip-1) significantly rescued cell viability, increasing survival rates from 64.9% to 80.3% and 83.0%, respectively. Together, these results clearly indicate that the excellent‌ antitumor efficacy of the DDSs-mediated ferroptosis therapy was achieved through the proposed mechanism.

In vivo biodistribution and therapeutic evaluation of SRF@Au@M NPs

To assess the in vivo distribution of the DDSs, the nanoparticles were labeled with indocyanine green (ICG) and intravenously injected into 4T1 tumor-bearing mice via the tail vein. One group of mice received free ICG, while another group received an equivalent dose of ICG-labeled SRF@Au@M NPs (ICG@SRF@Au@M NPs). The IVIS system was used to record the fluorescence images of the mice at different time points (0, 3, 12, 24, 36, and 48 h). As shown in Fig. 9a and c, the fluorescence signal and corresponding average fluorescence intensity at the tumor site for both free ICG and the ICG@SRF@Au@M NPs group reached their maximal values at 3 h and then gradually declined. The ICG@SRF@Au@M NPs were retained at the tumor sites longer than free ICG, maintaining a significantly higher average fluorescence intensity for up to 48 h. At 48 h, the major organs and tumor tissues were extracted from each group for fluorescence imaging. As exhibited in Fig. 9b, negligible fluorescence signals were detected in the major organs of the mice from either group, indicating that the nanomedicine had been largely cleared from the body by this time point. However, a remarkable fluorescent signal was observed in the tumor tissues of the ICG@SRF@Au@M NPs administration group. The fluorescence intensity at the tumor site was 2.2-fold higher than that in the free-ICG-treated group (Fig. 9 d), suggesting that SRF@Au@M NPs can effectively accumulate in tumor tissue and enhance the tumor-targeting ability of therapeutic agents. The biodistribution of SRF@Au@M NPs over time was further quantified in major organs and tumors using ICP-OES. As shown in Fig.9e, the SRF@Au@M NPs were primarily found in the liver and spleen, probably due to the hepatic metabolic pathway and the reticuloendothelial system uptake pathway. Notably, the tumor Fe content peaked at 5.39% ID g⁻¹ at 3 h and remained at a substantial level of 2.43% ID g⁻¹ even at 48 h post-injection, confirming effective tumor targeting through the enhanced permeability and retention (EPR) effect. Notably, ICG is cleared more rapidly than iron ions from the liver and kidneys, probably because of its quicker release from the MIL-100(Fe) and its higher water solubility

Fig. 9.

(a) In vivo biodistribution images and (c) corresponding average radiant efficiency at tumor sites in 4T1 tumor-bearing mice after intravenous injection of free ICG and ICG-labeled SRF@Au@M NPs. Data presented as mean ± S.D. (n = 3). (b) Ex vivo fluorescence imaging and (d) corresponding average radiant efficiency at 48 h post injection. Data presented as mean ± S.D. (n = 3). (e) Biodistribution of SRF@Au@M NPs in major organs and tumors at different time points. Data presented as mean ± S.D. (n = 3). (f) Therapeutic schedule for 4T1 tumor-bearing mice. (g) Individual and (i) total tumor growth curves after different treatments. (h) Photos and (j) weights of tumors after treatment for 14 days. (k) Body weights of mice during therapy. (l) Staining images of tumor sections H&E, TUNEL, GPX4, FSP1 and DHODH from different groups. All tumor therapy data are presented as mean ± S.D. (n = 5). *p < 0.05, **p < 0.01, ***p < 0.001

Inspired by the outstanding in vitro ferroptosis-inducing efficacy and remarkable tumor accumulation ability, we evaluated the in vivo antitumor efficacy of SRF@Au@M NPs in 4T1 tumor-bearing mice. When the tumors reached ∼100 mm³, the mice were randomly assigned to 6 groups (n = 5 per group): (1) control, (2) SRF, (3) MIL-100(Fe), (4) SRF@M, (5) SRF@Au@M and (6) SRF@Au@M NPs. The therapeutic schedule involved intravenous injection every 2 days for two weeks (Fig. 9f). Figure 9 g and i illustrate the curves for individual and average tumor volume changes across all study groups, which were used to assess the therapeutic efficacy. The results clearly show that blank MIL-100(Fe) had negligible anti-tumor efficacy, while free SRF displayed a moderate therapeutic effect. SRF@M exhibited superior anti-tumor efficacy compared to free SRF, likely due to the poor solubility of the free SRF. Tumor growth was further inhibited in mice treated with SRF@Au@M, which is consistent with the in vitro cytotoxicity findings. In comparison, the SRF@Au@M NPs group exhibited the lowest tumor volume at the end of the two-week treatment period, which was significantly lower than that in all other groups. The remarkable tumor inhibition was further quantified by measuring the weights of excised tumors and analyzing corresponding images. (Fig. 9j, h). Treatment with SRF@Au@M NPs achieved a tumor growth inhibition rate of 92.5%, which was significantly higher than that of all other treatment groups (Figure S31). These results collectively demonstrate that SRF@Au@M NPs possess the greatest therapeutic efficacy against tumor cell, mainly owing to the collaborative ferroptosis-therapeutic effect of their components. In addition, hematoxylin and eosin (H&E) staining and terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assays revealed that mice treated with SRF@Au@M NPs group caused severe cell shrinkage and an increase in dead cells (red fluorescence, Fig. 9l). Key indicators of ferroptosis in vivo were evaluated through the detection of the typical ferroptosis markers. As shown in Figure S32, tumor tissue from the SRF@Au@M NPs group exhibited the lowest GSH content and the highest ROS levels, indicating that SRF@Au@M NPs can effectively deplete GSH and maximally induce the Fenton reaction within solid tumors. Immunohistochemical analysis revealed that the lowest expression of GPX4, FSP1 and DHODH in tumor tissues treated with SRF@Au@M NPs (Fig. 9l), confirming the effective induction of potent ferroptosis in the tumor microenvironment

Furthermore, the systemic biosafety of the different formulations was systematically evaluated in vivo. First, the hemolysis assay demonstrated that the SRF@Au@M NPs exhibited a very low hemolysis ratio (< 5%) even at concentrations up to 800 µg/mL (Figure S33), indicating excellent blood circulation stability and biocompatibility. Throughout the two-week treatment period, no significant loss in average body weight was observed (Fig. 9k), confirming no apparent systemic toxicity. Moreover, H&E staining of major organs displayed no pathological abnormalities in any treatment group (Figure S34). The blood biochemical markers, including aspartate aminotransferase (AST), alanine aminotransferase (ALT), alkaline phosphatase (AKP), creatinine (CRE) and uric acid (UA), showed no obvious changes from the control group after SRF@Au@M NPs treatment (Figure S35), implying that our nanoparticles have no subacute toxicity for liver and kidney.

Conclusion

In summary, from four different particle sizes of MIL-100(Fe), we selected the optimal particle size of 60 nm to construct a multifunctional DDS platform. Within tumor cells, SRF@Au@M NPs form a complex and predictable fuel-driven chemical reaction network that effectively disrupts the three major defense pathways against ferroptosis. Upon cellular uptake, the combined action of GSH depletion by MIL-100(Fe) and SRF-mediated inhibition of system Xc⁻ leads to a sharp decrease in intracellular GSH concentration. This directly inactivates GPX4 and minimizes competing side reactions within the subsequent CRNs. A novel redox-CRN established between dihydrolipoic acid derivatives and iron ions continuously drives the Fenton reaction, generating abundant ROS. The synergistic effect of GSH depletion and ROS surge decisively disrupts the primary GSH-GPX4 antioxidant defense. Besides, the GOx-like activity of the Au NPs catalyzes intracellular glucose to produce H₂O₂, which serves as fuel for the CRNs. Concurrent glucose depletion disrupts the PPP and glycolysis, thereby inhibiting NAD(P)H production. Meanwhile, the conversion of LA to DHLA consumes additional NADH, impeding the regeneration of CoQ to CoQH2. The resulting decline in CoQH2 pool concurrently incapacitates both the FSP1-CoQH2 and DHODH-CoQH2 ferroptosis defense systems. Consequently, the MIL-100(Fe)-based CRNs triggered a potent ferroptosis storm in 4T1 tumor cells and displayed a substantial tumor suppression effect in tumor-bearing mice. Overall, this nanoplatform breaks the redox and NAD(P)H metabolic homeostasis in the tumor microenvironment, thereby achieving comprehensive ferroptosis therapy. Our sophisticated strategy based on “chemical reaction networks” provides a novel and inspired approach for cancer treatment.

Author contributions

Ding Guo: Writing–original draft, Writing–review & editing, Methodology, Validation, Investigation, Conceptualization. Qian Lin: Visualization, Software, Methodology, Investigation, Data curation. Quanyi Jin: Writing–review & editing, Validation, Methodology, Data curation. Lu Zhang: Validation, Methodology, Formal analysis. Nian Liu: Methodology, Investigation. Min Jiang: Data curation, Investigation, Validation. Shanshan Peng: Formal analysis, Investigation. Jiafeng Xu: Methodology, Investigation. Xueming Gao: Software, Methodology. Lili Zong: Validation, Supervision, Formal analysis, Funding acquisition. Xuan Zhu: Writing–review & editing, Validation, Supervision, Funding acquisition, Conceptualization.

Funding

This research was supported by the National Nature Science Foundation of China (No. 82273872) and Xiamen University (Start-up Grant).

Data availability

The datasets supporting the conclusions of this article are included within the article and its supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.