Carrier-free nanoparticles in microneedles disrupt endosomal acidification suppress ferroptosis in radiation-induced oral mucositis Authors: Yanxuan Xiao, Pengcheng Sun, Wenjie Xu, Xinyan Hao, Tiantian Tang, Xinying Liu, Hai Huang, Zhaojian Gong, Junyong Wu Published: 2026 ABSTRACT AbstractRadiation-induced oral mucositis (RIOM) is the most common complication in patients receiving radiotherapy for head and neck, to which ferroptosis is an important contributor. Transferrin receptor (TFRC) -mediated endocytosis is a critical source of iron. Accordingly, we proposed that inhibiting ferroptosis by disrupting TFRC mediated-iron uptake. In this study, we developed endosome-targeting self-assembling nanoparticles (CB NPs) integrated into a detachable microneedle patch (CBT@MNs) for precise and synergistic RIOM therapy. This system integrated the clinically conventional drugs chloroquine and baicalin and loaded into MN patch for mucosal drug delivery. In the tongue mucosa, the MN tips released CB NPs. Upon entering endosomes through the endocytosis pathway, CB NPs disrupted endosomal acidification through the proton sponge effect, which may block TFRC recycling, achieving long-term inhibition of ferroptosis and inflammation efficacy. In rat models of RIOM, CBT@MNs dramatically attenuated mucosal ulceration and accelerated healing via anti-inflammation, angiogenesis, and collagen deposition. Transcriptomic analysis further revealed that CBT@MNs suppressed ferroptosis and inflammatory signaling. This radioprotective strategy for inhibiting ferroptosis and regulating redox homeostasis provided a novel approach for designing tissue engineering materials.Keywords: Endosome, Transferrin receptor recycling, Carrier-free nanoparticles, Microneedle patch, Radiation-induced oral mucositis Introduction In nearly all patients receiving radiotherapy for head and neck malignancies, the highly radiosensitive oral mucosa develops radiation-induced oral mucositis (RIOM), with approximately 65% of cases progressing to severe oral mucositis [1]. Severe RIOM manifests as confluent ulceration, intense pain, and bleeding, frequently forcing radiotherapy interruptions [2]. Each day of treatment delay may reduce locoregional control and overall survival by roughly 1% [3]. The economic burden is equally stark, with additional direct medical costs of 5000–30,000 per patient [4]. Despite this, current treatments remain largely symptomatic—anti-inflammatory agents, analgesics, and wound-healing agents—and offer minimal benefit in severe cases. Identifying radioprotective strategies that prevent severe RIOM is therefore imperative to sustain treatment adherence, improve prognosis, and enable uninterrupted radiotherapy. Recent studies identified ferroptosis as the principal mechanism driving RIOM progression [[5], [6], [7]]. Clinically, nasopharyngeal carcinoma patients exhibited markedly elevated serum iron and transferrin (TF) saturation, together with reduced unsaturated iron-binding capacity after radiotherapy [8]. Single-cell sequencing by Yu et al. confirmed that ferroptosis was the dominant mode of ionizing radiation (IR)–induced epithelial cell death [5]. IR generated abundant reactive oxygen species (ROS) that fueled Fenton reactions with ferrous iron (Fe2+), causing lipid peroxidation, mitochondrial damage, and ferroptosis. Central to this process was transferrin receptor (TFRC) recycling: the TFRC-TF-Fe3+ complex was internalized via clathrin-mediated endocytosis, and the endosomal metalloreductase STEAP3 released and reduced Fe3+ to Fe2+ before mitochondrial transport. The iron-free TF-TFRC complex then recycled to the membrane for further iron uptake (Fig. 1B). Ferroptotic cells released large amounts of Damage-Associated Molecular Patterns (DAMPs), lipid peroxidation products, and chemokines, that amplified inflammatory cascades, exacerbating mucosal injury and driving progression to severe mucositis. We therefore proposed a dual-action strategy combining anti-ferroptotic and anti-inflammatory therapy. By preventing endosomal acidification, we aimed to block Fe2+ reduction and TFRC recycling, thereby starving epithelial cells of iron and protecting them from ferroptosis. Coupled with anti-inflammatory intervention, this approach would disrupt the ferroptosis–inflammation amplification loop and reduce severe mucositis incidence. Schematic Diagram of the Therapeutic Strategy. (A) The synthesis process of CBT@MNs. UV, ultraviolet. PDMS, MNs mold. (B) Schematic of therapeutic mechanisms of CBT@MNs for RIOM. IR caused oral mucositis by producing ferroptosis and excess inflammatory cytokine. CBT@MNs inhibited ferroptosis and inflammation by disrupting iron uptake, thereby protecting oral mucosa. In recent years, self-assembled small-molecule nanomedicines have emerged as a promising and biocompatible platform for co-delivery of multiple drugs in combination therapy [9]. Unlike conventional delivery systems, these nanomedicines represent an advanced, integrated co-delivery strategy with several key advantages, including favorable biocompatibility, high drug loading, and reduced risk of drug leakage [10]. Chloroquine (CQ) is a promising radioprotective agent. Within the acidic endosomal compartment, CQ becomes protonated and positively charged, leading to substantial intraluminal accumulation and disruption of endosomal acidification [11]. Baicalin (BA), the principal flavonoid constituent extracted from the root of Scutellaria baicalensis Georgi, possesses broad-spectrum anti-inflammatory and antioxidant activities and has demonstrated efficacy in promoting wound healing, including irradiated wounds [12,13]. Given their complementary therapeutic effects, co-delivery of CQ and BA offers a promising radioprotective strategy, aiming to enhance therapeutic efficacy by concurrently suppressing ferroptosis and inflammation in mucosal tissue and thereby reducing the incidence of severe mucositis. Herein, we successfully developed a carrier-free nanomedicine driven by hydrogen bonding and π–π stacking interactions for co-delivery of CQ and BA, designed to modulate the microenvironment of RIOM and achieve efficacious and safe treatment (Fig. 1A). CB NPs were internalized into endosomes, where CQ disrupted endosomal acidification through a proton sponge effect and reduced iron uptake, thereby preventing radiation-induced ferroptosis in epithelial cells. Concurrently, this system attenuated the radiation-triggered inflammatory cascade. Given that RIOM was a superficial, localized mucosal injury, we designed a detachable microneedle (MN) system. The backing layer of the MN patch was fabricated from polyvinylpyrrolidone (PVP), selected for its solubility, hydrophilicity, and biocompatibility. Upon contact with the oral mucosa, the backing layer dissolved rapidly, leaving the needle tips embedded in the mucosa to achieve drug release. The needle tips were composed of methacrylated hyaluronic acid (HAMA), tannic acid (TA), and CB NPs. UV-triggered photo-crosslinking of HAMA provided a degree of resistance to the humid oral environment, but this network alone lacked sufficient mechanical integrity for prolonged residence. Moreover, CB NPs simply entrapped within HAMA would rapidly diffuse out upon saliva contact. The multiple galloyl groups of TA formed robust hydrogen bonds and π–π stacking interactions with the phenolic hydroxyls of BA, locking CB NPs within the network. Simultaneously, the abundant phenolic hydroxyls of TA were engaged in dense hydrogen-bond crosslinking with HAMA backbones, markedly enhancing the mechanical properties of the patch. This study aimed to evaluate the therapeutic efficacy of the CBT@MNs patch against RIOM. We characterized the physicochemical and mechanical properties of CB NPs and CBT@MNs and investigated their capacity to rescue post-irradiation cell viability and their underlying mechanisms. The overarching hypothesis was that CBT@MNs conferred therapeutic benefit by coordinately suppressing ferroptosis and inflammation (Fig. 1B). Results Fabrication and characterization of CB NPs and CBT@MNs CB NPs were prepared via a solvent–antisolvent precipitation method, yielding spherical solid nanoparticles (Fig. 2A–B). The mean particle diameter was 134 ± 6.39 nm, with a zeta potential of −21 ± 2.25 mV and a polydispersity index (PDI) of 0.12 ± 0.01 (Table S1 and Fig. 2C). The resulting suspension appeared as a clear yellow solution (Fig. S1A–B). The loading capacity and encapsulation efficiency for CQ were 35% ± 8.07% and 84.18% ± 2.34%, and those for BA were 38.74% ± 4.93% and 89.88% ± 4.75%, respectively (Table S2). Assessment of CB NPs stability revealed that both particle size and PDI remained largely unchanged over a period of 14 days (Fig. 2D). Cytotoxicity was evaluated in HOK and HUVEC cells, as shown in Fig. S1C–D. The safe concentration of CB NPs substantially exceeded the intended therapeutic dose. To identify the optimal self-assembly ratio, formulations with CQ:BA ratios of 1:1, 2:1, and 4:1 were compared with respect to morphology, size, and PDI (Table S1 and Fig. S1E–F). Increasing the CQ:BA ratio led to progressive increases in diameter and PDI, accompanied by irregular particle morphology. These findings suggested that 1:1 might represent the optimal ratio for self-assembly, which would be further validated in subsequent in vitro experiments. Preparation and characterization of CB NPs and CBT@MNs. (A-B) TEM image of CB NPs, with scale bars of 200 and 100 nm. (C) The diameter distribution, PDI and zeta potential of CB NPs. (D) In vitro stability of CB NPs within 14 d in 5% glucose solution containing 10% fetal bovine serum. (E) UV−vis absorption spectra comparing CB NPs and free drugs. (F) UV−vis absorbance of CB NPs in SDS, Urea, and Sodium Chloride at 80 mM. (G) Changes in particle size of CB NPs after incubation with various concentrations of SDS, Urea, and Sodium Chloride. (H) Dynamic snapshot CB NPs self-assembly from 0 ns to 50 ns (I) Intermolecular interaction patterns, including hydrogen bonding and π-π stacking. (J) Schematic illustration of the CBT@MNs synthesis. (K-L) SEM images of CBT@MNs, scale bars 500 and 200 μm. (M) The release profile of CQ and BA in vitro from CBT@MNs. (N) Compression-displacement curve of C@MNs and CBT@MNs. Digital photograph of tongue of a normal rat (O) and the tongue after CBT@MNs insertion (P). (Q) H&E staining images revealed the epithelial structure of the rat tongue dorsum under normal conditions, as well as following C@MNs or CBT@MNs administration, scale bars 100 μm. Statistical significance was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Data are presented as mean ± SD. n = 3. The self-assembly mechanism of CB NPs was further analyzed by UV-vis spectroscopy. Compared with the simple superposition of absorbance values of the two free drugs (Fig. 2E), the absorption peak of CB NPs at 262 nm exhibited a red shift, indicative of intermolecular conjugate force during assembly. As shown in Fig. 2F, addition of sodium dodecyl sulfate (SDS) induced a slight blue shift in the UV–vis spectrum. Concurrently, SDS treatment led to a marked increase in NPs size, whereas exposure to Urea or Sodium Chloride had negligible effects (Fig. 2G). Collectively, these findings indicated that hydrophobic interaction constituted the predominant driving force governing the self-assembly process. Molecular dynamics simulations were next employed to further elucidate the self-assembly mechanism of CB NPs. Fig. 2H presented representative snapshots of the assembly process at selected time points. Over the course of the 20 ns simulation, the system progressively coalesced into stable nanoclusters. Fig. 2I provided a schematic depiction of the intermolecular interactions within the CB NP assembly, highlighting π–π stacking and hydrogen bonding as principal driving forces. The root-mean-square deviation (RMSD), solvent-accessible surface area (SASA), number of intermolecular hydrogen bonds, and radius of gyration (Rg) of the system all stabilized at approximately 20 ns (Fig. S2A–D), indicating that the system reached overall equilibrium and formed a compact and stable self-assembled structure by this time point. Fig. 2J illustrated the fabrication procedure for CBT@MNs. A mixDed solution containing CB NPs, TA, and HAMA was first introduced into a PDMS mold. Needle tips were formed by sequential centrifugation, UV irradiation, and oven drying. Subsequently, a PVP solution was applied to the mold, followed by centrifugation and air-drying at room temperature to yield the final CBT@MNs patch. A representative photograph of the fabricated patch was shown in Fig. S3A. Scanning electron microscopy revealed that each pyramidal microneedle measures 1500 μm in height, with a base width of 640 μm (Fig. 2K). The resulting square patch comprised a uniform 10 × 10 array of conical MNs (Fig. 2L). In vitro drug release studies demonstrated cumulative release rates approaching 80% for both CQ and BA after 24 h (Fig. 2M), establishing a robust foundation for subsequent therapeutic efficacy of the CBT@MNs. The mechanical properties of CBT@MNs were evaluated using electronic universal material testing machine under compressive loading. As shown in Fig. 2N, both C@MNs (no TA) and CBT@MNs underwent continuous deformation with increasing load and displacement, without evidence of fracture. Previous studies reported that the force was required for a single MN to penetrate the stratum corneum and insert into skin ranges from approximately 0.1 to 3 N [14]. In the present study, the compressive force exceeded 3 N at a displacement of 1.5 mm. Notably, the CBT@MNs group exhibited superior mechanical strength relative to the C@MNs groups. The dorsal tongue mucosa comprises epithelium, basement membrane, and lamina propria, with a mean thickness of 789.4 μm [15]. Mucosal penetration poses a greater challenge than skin insertion owing to tissue viscoelasticity and the underlying musculature, necessitating MNs with adequate mechanical strength. To assess mucosal penetration capability of CBT@MNs, Sprague–Dawley (SD) rat tongues were used. Fig. 2O depicted an untreated rat tongue, whereas Fig. 2P showed the tongue following insertion of CBT@MNs stained with trypan blue. The presence of a well-defined array of puncture sites confirmed successful insertion. Harvested tongues were subsequently processed for histopathological analysis. Hematoxylin and eosin (H&E) staining revealed an intact epithelial architecture in control, contrasted by disruption of the mucosal barrier in MNs-treated specimens (Fig. 2Q). Compared with the C@MNs group, which produced only partial mucosal disruption, CBT@MNs created discontinuities in the tongue epithelium with evident epithelial defects. Biocompatibility of PVP, HAMA, and TA was assessed in HOK cells. No appreciable cytotoxicity was observed at concentrations up to 100 μg/mL (Fig. S3B). The in vivo degradation performance of CBT@MNs was shown in Fig. S4. These findings demonstrated that CBT@MNs could effectively penetrate the mucosal barrier with adequate mechanical strength. The radiation protection of CB NPs in vitro Both epithelial and endothelial cells are recognized as radiosensitive cell types and are frequently employed as in vitro models for RIOM research [16,17]. Accordingly, HUVEC and HOK cells were selected for subsequent experimentation. To evaluate the radioprotective efficacy of CB NPs in vitro, we first determined the cytotoxicity profiles of CQ and BA. As shown in Fig. S3C–D, concentrations up to 10 μM were well tolerated by both HOK and HUVEC cells. A previous study reported that CQ suppressed ferroptosis at relatively low concentrations [18]. We therefore assessed the effect of low-dose CQ on Erastin-induced ferroptosis. The data presented in Fig. 3A confirmed that low concentrations CQ effectively attenuated ferroptosis. A concentration of 1 μM was identified as optimal and exhibited efficacy comparable to the ferroptosis inhibitor ferrostatin-1. In parallel, we examined the ferroptosis-inhibitory activity of CB NPs formulated at varying CQ:BA ratios. As shown in Fig. 3B, the 1:1 ratio demonstrated superior performance, further corroborating for the optimal self-assembly ratio. In vitro radioprotective effects of CB NPs. (A) Effects of various concentrations of CQ (0.125–2 μM) on the viability (CCK8 assay) of HOK cells following 18 h incubation with Erastin (10 μM); Fer-1 co-incubation served as controls. (B) Effects of different CB and CB NP ratios on HOK cell viability under the same Erastin treatment (18 h). (C) HOK cells incubation with CB NPs after 6 h and 24 h. Red, lysotracker; Green, FITC; Blue, Hoechst. (D-E) Lysotracker stained cell images taken 18 h after IR, with corresponding bar graphs showing fluorescence intensity. Scale bars, 10 μm. (F-G) Cell viability was assessed 72 h after IR. (H-K) Cell death was evaluated by Calcein-AM/PI staining 48h after IR.HOK cells and HUVEC were exposed to 12 Gy IR. The concentration of CQ, BA, CB and CB NPs were 1 μM. Scale bars, 200 μm. Statistical significance was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Data are presented as mean ± SD. n = 3, *P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Free small-molecule drugs enter cells mainly through passive diffusion and distribute broadly within the cell, with a portion of CQ possibly accumulating in acidic organelles [19]. In contrast, CB NPs were actively internalized via endocytosis, enabling direct and efficient delivery CQ to acidic organelles such as endosomes, thereby precisely modulating acidity [20]. Lysotracker probe selectively label acidic organelles in living cells [21]. To assess endosomal targeting, we stained HOK cells with lysotracker and CB NPs/FITC. Fig. 3C exhibited that CB NPs were effectively taken up by cells and delivered to endosomes. As shown in Fig. 3D–E, fluorescence intensity was markedly attenuated in the CB NPs group relative to the Erastin group. Moreover, fluorescence intensity in the CB NPs group was statistically different from that in the CB (CQ + BA) group, confirming that CB NPs reduced acidity more effectively than free CQ. The radioprotective efficacy of CB NPs was subsequently evaluated using CCK-8 assays and live/dead cell staining. Following irradiation, the viability of HUVEC and HOK cells declined markedly, reaching approximately 28.9% and 35%, respectively (Fig. 3F–G). All treatment groups partially restored cell viability. Notably, the CB group exhibited statistically difference than CQ group, validating the therapeutic benefit of the combination strategy. CB NPs group further outperformed CB group, elevating viability to 86% and 85% in the respective cell lines. Live/dead staining corroborated these findings, demonstrating that both CB and CB NPs substantially alleviated cell death after IR. The CB NPs group exerted the greatest protective effect, outperforming the CQ group (Fig. 3H–K). This disparity was likely attributable to enhanced endosomal targeting by CB NPs. In summary, these results indicated that CB NPs rescued cell viability after IR and outperformed the CB group in radioprotective efficacy. CB NPs disturbed endosomal acidification The weakly basic moiety within the CQ structure is known to disrupt endosomal acidification [22]. To measure acidity accurately, we used Lysosensor probe and Magic Red fluorogenic substrate (Fig. 4A). Under neutral conditions, Lysosensor emits predominantly blue fluorescence, whereas in acidic environments—such as functional endosomes—it exhibits yellow emission. Lysosensor is pH-sensitive and can be used to image acidic organelles. However, intracellular acidic organelles also include late endosomes, secondary lysosomes, and others, and probe cannot specifically differentiate among these organelles. Therefore, in the relevant experimental results, the acidic organelles we observed represented a mixed population. As shown in Fig. 4B, both CQ and the CB groups elicited partial blue fluorescence, whereas CB NPs treatment resulted in a substantially greater proportion of blue signal, indicating effective elevation of endosomal pH. To evaluate the functional consequences of altered endosomal acidity, we assessed the activity of cathepsin B, a major hydrolase distributed in late endosomes with pronounced pH-dependent activity, using the Magic Red fluorogenic substrate. Cleavage of this substrate by active cathepsin B liberated red fluorescence, the intensity of which was proportional to enzymatic activity. Fig. 4C–D demonstrated that both CQ and CB groups markedly reduced cathepsin B activity, while CB NPs group produced an even more pronounced attenuation that differed significantly from CQ group. These findings established that CB NPs effectively impeded endosomal acidification and intraluminal hydrolase activity. CB NPs disturbed endosomal acidification. (A) Schematic of endosomal acidification detection. (B) Endosomal acidity was assessed by Lysosensor Yellow/Blue DND-160 staining 6h after IR in HOK cells. Scale bars, 5 μm. (C-D) Cathepsin B activity was assessed using the Magic Red assay kit staining 6h after IR in HOK cells, with corresponding bar graphs showing fluorescence intensity. Scale bars, 10 μm. (E) Immunoblotting analyses detected TFRC protein levels in RAW264.7 and HOK cells. (F) Schematic of CB NPs disturbed TFRC recycling to membrane. (G) Immunofluorescence analysis of TFRC expression on the membrane of RAW264.7 cells. Scale bars, 2.5 μm. Statistical significance was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Data are presented as mean ± SD. n = 3, *P < 0.05, ***P < 0.001, ns, not significant. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) To further investigate the impact of CB NPs on TFRC recycling, we assessed TFRC expression levels on the plasma membrane of HOK and RAW264.7 cells (Fig. 4F). Western blot (WB) analysis of cell membrane fractions demonstrated decreased TFRC abundance in all treatment groups, with the lowest levels detected in the CB NPs group (Fig. 4E). Immunofluorescence staining revealed TFRC localization at the cell membrane, with distinct membrane-associated fluorescence in control and IR groups (Fig. 4G). In contrast, CQ or CB groups displayed blurred membrane boundaries and partial redistribution of fluorescence to the cytoplasm. In the CB NPs group, membrane boundaries were largely absent, with the majority of fluorescence localized intracellularly, indicating a marked reduction TFRC levels in cell membrane. These data indicated that CB NPs diminished the amount of TFRC available at the plasma membrane. CB NPs inhibited IR-induced ferroptosis Ferroptosis is characterized by iron overload and lipid peroxidation [23]. We quantified cellular iron uptake using the FerroOrange probe. IR markedly elevated intracellular Fe2+ levels in both HOK, HUVEC and RAW264.7 cells (Fig. 5A and Fig. S6A). Treatment with CQ or the CB combination attenuated radiation-induced iron accumulation, whereas CB NPs produced the greatest reduction in fluorescence intensity (Fig. 5B–C and Fig. S6B). This result was consistent with effective suppression of iron uptake. We assessed lipid peroxidation levels using the C11-BODIPY probe and malondialdehyde (MDA) quantification in HOK and HUVEC cells after IR. As shown in Fig. 5D, green fluorescence indicative of oxidized lipids was markedly elevated in irradiated HUVEC cells, whereas CB and CB NPs substantially reduced the oxidized lipids levels. MDA, a stable end product of lipid peroxidation, directly reflects plasma membrane oxidative damage [24]. IR significantly increased MDA content in both HOK and HUVEC cells (Fig. 5E–F). This phenomenon was attenuated by CQ and CB groups, with the CB NPs group exhibiting the lowest MDA levels. CB NPs inhibited IR-induced ferroptosis. (A-C) Images of FerroOrange stained cell 48 h after IR. Bar graphs show relative fluorescence intensity. Scale bars, 25 μm. (D) C11-BODIPY staining of HUVEC cells 6 h after IR, showing reduced (red) and oxidized (green) BODIPY. Scale bars, 25 μm. (E-F) MDA levels were quantified 48 h after IR in HOK cells and HUVEC. (G) Schematic of CB NPs inhibited IR-induced ferroptosis. (H) TEM images of mitochondrial morphology in HOK cells. Scale bar, 100 nm. (I-J) Mitochondrial size and membrane density derived from TEM micrographs using ImageJ software. (K-L) Lysotracker and Mitotracker stained cell images taken 72 h after IR, with corresponding bar graphs showing fluorescence intensity. Scale bars, 10 μm. Statistical significance was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Data are presented as mean ± SD. n = 3, *P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Mitochondria served as critical hubs for intracellular iron storage and metabolism. Iron-catalyzed Fenton reactions within the mitochondrial matrix and membranes generate excessive reactive oxygen species (ROS), leading to mitochondrial shrinkage, cristae loss or disappearance, outer membrane rupture, and aberrantly increase membrane density—hallmarks of ferroptosis (Fig. 5G). Transmission electron microscopy (TEM) revealed normal mitochondrial size and membrane density in control cells, whereas IR group displayed mitochondrial contraction, cristae depletion, and elevated membrane density, indicative of mitochondrial dysfunction (Fig. 5H). CB NPs group markedly ameliorated these ultrastructural abnormalities, significantly restoring mitochondrial size and reducing membrane density (Fig. 5I–J). We further examined the effect of CB NPs on mitochondria in irradiated HUVEC cells using Mitotracker and Lysotracker probes. In unirradiated cells, the majority of mitochondria exhibited a tubular morphology (Fig. 5K). IR induced pronounced mitochondrial fragmentation and swelling, early indicators of cellular stress potentially linked to elevated mitochondrial ROS. Both CB and CB NPs groups substantially improved mitochondrial morphology. Moreover, CB NPs significantly reduced Lysotracker fluorescence intensity (Fig. 5L), confirming concomitant effects on endosomal acidification in irradiated cells. Meanwhile, we included the ferroptosis inhibitor DFO as a control group to compare its efficacy with that of CB NPs. The results in Fig. S5A and S5B showed that DFO effectively reduced the radiation-induced elevation of intracellular iron levels and lipid peroxidation in HOK and HUVEC cells. The results in Fig. S5C demonstrated that DFO alleviated radiation-induced mitochondrial swelling. However, Lysotracker fluorescence intensity showed no significant change, indicating that DFO did not affect the acidity of acidic organelles. As a whole, these findings demonstrated that CB NPs suppressed IR-induced ferroptosis. The anti-inflammatory activity of CB NPs in vitro We next examined anti-inflammatory properties of CB NPs. Excessive ROS generated by IR exacerbated mucosal inflammation, prompting us to first evaluate the ROS-scavenging capacity of CB NPs using the DCFH-DA fluorescent probe. As shown in Fig. 6A–C, IR markedly increased green fluorescence intensity, indicating elevated ROS accumulation. Both CB and CB NPs attenuated the fluorescence signal, with the CB NPs group exhibiting the lowest intensity in HOK and HUVEC cells. As shown in Fig. S6C–D, CB NPs scavenged excessive ROS in RAW264.7 cells and it was superior to the DFO group. In vitro anti-inflammatory effects of CB NPs. (A) Cellular ROS level was detected using the DCFH-DA probe in HUVEC and HOK cells 6 h after IR. Scale bars, 200 μm. (B-C) Bar graphs show relative DCFH-DA fluorescence intensity. (D) Immunoblotting analysis of STAT3 and pSTAT3 protein levels in RAW264.7 cells. (E-G) RAW264.7 cells were pretreated with CQ, BA, CB and CB NPs (1 μM) for 1 h, then stimulated with LPS (1 μg/mL) for 24 h. The levels of (E) TNF-α, (F) IL-6 and (G) IL-1β in culture supernatants were quantified by ELISA. (H, J) Immunofluorescence staining of STAT3 and pSTAT3 expression in RAW264.7 cells. Scale bars, 5 μm. (I, K-L) Bar graphs show relative fluorescence intensity of STAT3, pSTAT3 and STAT3/pSTAT3 ratio. Statistical significance was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Data are presented as mean ± SD. n = 3, *P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant. To further characterize anti-inflammatory efficacy of CB NPs, we constructed in vitro inflammatory models by stimulating RAW264.7 cells with LPS. STAT3, a key member of the STAT protein family, drives the expression of multiple pro-inflammatory cytokines through the JAK/STAT signaling pathway. LPS stimulation typically activates STAT3, a molecular indicator of successful inflammatory model establishment. The p-STAT3/STAT3 ratio accurately reflects the degree of STAT3 activation [25]. WB analysis confirmed successful model construction, as evidenced by a pronounced increase in phosphorylated STAT3 (p-STAT3) levels in the LPS group (Fig. 6D). BA, CB and CB NPs groups abrogated this effect, significantly reducing p-STAT3 expression. Immunofluorescence staining of STAT3 and p-STAT3 in RAW264.7 cells corroborated these findings. CB NPs attenuated the LPS-induced upregulation of both STAT3 and p-STAT3 (Fig. 6H and J). Quantitative analysis further confirmed that p-STAT3/STAT3 ratio was significantly lower in the CB NP group (Fig. 6I and 6K-L). Culture supernatants were subsequently collected for quantification of pro-inflammatory cytokines by ELISA. LPS stimulation substantially elevated the secretion of TNF-α, IL-1β, and IL-6 (Fig. 6E–G), an effect that was reversed by BA, CB and CB NPs groups. Meanwhile, we examined the effect of CB NPs on the polarization of RAW264.7 cells. Macrophages were identified based on F4/80 expression and then stained for the M2 marker CD206 and the M1 marker CD86. The results showed that CB NPs reduced the number of M1 RAW264.7 cells (Fig. S7A). However, the number of M2 RAW264.7 cells showed no obvious change across all groups. Overall, CB NPs possessed robust anti-inflammatory activity in vitro. The radiation protection of CBT@MNs in vivo We evaluated the in vivo radioprotective efficacy of CBT@MNs using RIOM rat models. Based on previous studies [26,27] in which a single topical dose of baicalin-loaded MNs was approximately 10 μg, we tested three single-dose levels: 10, 20, and 30 μg per patch. After 7 days, blood biochemistry and organ toxicity were examined, and no significant toxicity was observed at any of the three doses (Fig. S8A–C). Therefore, 20 μg per patch was selected as the single dose. The experimental timeline was depicted in Fig. 7A. MNs were administered on days 0, 2, 4, and 6. Intraoral photographs were captured every two days from day 4 to day 14 to record the progression of oral mucositis. Tongue tissues were harvested on day 14 for histopathological analysis. Rats were randomly allocated into six groups as illustrated in Fig. 7B: Control, IR, IR + CQ@MNs, IR + BA@MNs, IR + CB@MNs, and IR + CBT@MNs. The IR group exhibited severe manifestations on day 14, including extensive lingual ulceration and pseudomembrane formation. In contrast, treatment groups (IR + CQ@MNs, IR + BA@MNs, IR + CB@MNs, and IR + CBT@MNs) displayed varying degrees of symptomatic alleviation. Notably, the IR + CBT@MNs group showed the most pronounced recovery, with near-complete healing of ulcers and erosive lesions on the tongue. Histological evaluation via H&E staining revealed marked epithelial damage in the IR group, characterized by epithelial thinning, flattening of lingual papillae, complete epithelial loss in ulcerated regions, and dense inflammatory cell infiltration (Fig. 7C). Treatment groups exhibited attenuated epithelial injury. Localized epithelial damage and inflammation were evident in the IR + CQ@MNs and IR + BA@MNs groups, whereas the IR + CB@MNs and IR + CBT@MNs groups displayed comparatively intact epithelium with limited inflammatory infiltration. Restoration of the characteristic lingual papillary architecture was observed in the IR + CBT@MNs group. Quantitative analysis of epithelial thickness (Fig. 7G) confirmed recovery in all treatment groups relative to IR controls, among them the CBT@MNs group approaching values comparable to those of normal tissue. Oral mucositis scoring (Fig. 7F) was highest in the IR group and significantly lower across all treatment groups. Notably, the IR + CBT@MNs group showed the lowest score. In vivo radioprotective effects of CBT@MNs. (A-J) The dorsal tongue of rats was exposed to 25 Gy IR. (A) Schematic diagram of in vivo radioprotection by CBT@MNs. IR was implemented on day 0. followed by CBT@MNs administration on day 0, 2, 4 and 6. Tongue were collected on day 14. (B) Representative images of tongues at various time points for each group. (C, G) H&E staining sections showed structural changes and inflammation of the dorsal tongue epithelium. Scale bar, 125 μm. Epithelial thickness was derived from H&E images. (D, H-I) Masson staining showed the repair of muscle and collagen fibers in tongue tissue, and quantitative analysis demonstrated the cross-sectional area diameter of muscle fibers and collagen volume fraction. Scale bar, 80 μm. (E, J) CD31 IHC staining showed the status of vascular proliferation in tongue tissue, and quantitative analysis demonstrated numbers of positive blood vessels. Scale bar, 40 μm. (F) Oral mucositis scoring over time per group. Statistical significance was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Data are presented as mean ± SD. n = 3, *P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant. To evaluate the therapeutic effect of CBT@MNs on promoting tongue tissue repair following IR, Masson's and CD31 Immunohistochemical (IHC) staining were performed. As shown in Fig. 7D, irradiated tongue tissues exhibited marked atrophy of muscle fibers, with a reduction in both collagen fiber area and density, along with a looser collagen distribution. In contrast, treatment groups promoted muscle repair. Notably, the CBT@MN group showed increased collagen fiber deposition and the most compact collagen distribution among all treatment groups (Fig. 7H and I). CD31, a marker of vascular endothelial cells, is widely used to assess angiogenesis. As illustrated in Fig. 7E and J, IR caused significant vascular atrophy, whereas the CBT@MN group demonstrated the strongest pro-angiogenic capacity. Together, these results indicated that CBT@MNs facilitated the regeneration of muscle fibers, collagen fibers, and blood vessels in irradiated tongue tissues. To assess the in vivo biosafety of CBT@MNs, hematological and biochemical analyses were performed on day 14. As shown in Fig. S9A-D, IR group exhibited markedly elevated proportions of white blood cells, neutrophils, lymphocytes, and monocytes, indicative of an acute systemic inflammatory state. These parameters were significantly attenuated in all treatment groups, with the CBT@MNs group demonstrating statistically significant differences relative to the IR group, confirming that CBT@MNs treatment effectively reduced systemic inflammation. Hepatic and renal function markers remained within normal limits across all groups, with no significant differences detected between the control and CBT@MNs groups (Fig. S9E-G). Histopathological examination of major organs—including kidney, liver, heart, lung, and spleen—revealed no evidence of appreciable systemic toxicity in the CBT@MNs group compared with controls (Fig. S10A). Body weight changes were presented in Fig. S10B. The IR group experienced a precipitous decline in body weight beginning on day 6, likely attributable to impaired feeding capacity secondary to progressive RIOM. Treatment groups exhibited transient weight loss followed by gradual recovery, consistent with amelioration of oral symptoms. These findings supported the short-term safety profile of CBT@MNs. Nevertheless, long-term biosafety is paramount for clinical translation, and additional studies are warranted to comprehensively evaluate the overall safety of the CBT@MNs. Mechanistic validation of CBT@MNs radioprotection Transcriptomic profiling was performed on tongue tissues from rats treated with CB@MNs and CBT@MNs to elucidate the molecular mechanisms underlying therapeutic efficacy. Comparable data quality across all groups confirmed the reliability of global expression distributions and subsequent differential analyses (Fig. S11A). Analysis of differentially expressed gene (DEG) overlap revealed widespread transcriptional dysregulation in the IR group (A2) relative to the control (A1). MNs treatment (A3–A4) partially reversed these alterations (Fig. S11B). Unsupervised clustering, heatmap visualization, and enrichment analyses identified four distinct DEG modules (Fig. 8A), which were functionally associated with inflammatory signaling pathways (C2), ferroptosis (C3), endosomal acidification (C1), and endocytosis (C4). These modules were robustly activated in the A2 group, indicating that IR might promote ferroptosis and chronic inflammation in tongue tissue. Both CQ@MNs and CBT@MNs partially corrected these imbalances, with the CBT@MNs group exhibiting the most pronounced normalization toward baseline expression patterns. Volcano plot analysis comparing the A4 and A2 groups identified 2299 upregulated and 2717 downregulated DEGs (Fig. S11C), a substantially larger number than that observed in the A3 group. This finding indicated that CBT@MNs treatment elicited more extensive and pronounced transcriptomic alterations than CQ@MNs alone, warranting further functional enrichment analyses to delineate the biological significance of these differences. Transcriptomics revealed that CBT@MNs dampened ferroptosis and inflammation. (A) Unsupervised clustering, heatmap, and enrichment analyses of differential gene modules among the four groups. Left: boxplots showing expression trends of representative gene clusters. Middle: heatmap of standardized expression profiles per cluster. Right: enrichment analysis summarizing key biological pathways associated with each cluster. (B) Heatmap of relative expression of genes associated with ferroptosis, TFRC recycling, endosome acidification and inflammatory signaling pathway in tongue from rats subjected to different treatments. (C) GO biological process enrichment (IR + CBT@MNs versus IR) highlighting regulation of iron metabolism, endosome acidification and inflammatory response. (D) KEGG chord diagram linking DEGs to enriched pathways, with emphasis on down-regulation of ferroptosis and inflammatory signaling pathway. Groups: A1, Control; A2, IR; A3, IR + CB@MNs; A4, IR + CBT@MNs. We next focused specifically on the effects of CBT@MNs on endosomal acidification and TFRC recycling. Modules C1 and C4, which were associated with endosomal acidification and endocytosis, were markedly upregulated in the A2 group and subsequently attenuated following MNs treatment. The most pronounced downregulation was observed in the A4 group (Fig. 8A). At the gene level, expression of proteins implicated in endosomal acidification was significantly elevated in the A2 group and declined in both the A3 and A4 groups (Fig. 8B). Concurrently, genes encoding proteins involved in TFRC recycling were upregulated in the A2 group: Ap2a2 recognized the cytoplasmic tail of TFRC to initiate endocytosis, Ehd4 regulated TFRC transport from early endosomes to recycling endosomes, and Rab8a/Rab8b drove exocytosis of TFRC-containing vesicles to the plasma membrane. Expression of these proteins was substantially downregulated in the A4 group. Gene Ontology (GO) analysis corroborated these observations, revealing enrichment of downregulated genes in the A4 group for terms such as “early endosome to recycling endosome transport” “endosome to plasma membrane transport vesicle” “endosomal lumen acidification” and “endocytic recycling” (Fig. 8C). These data further supported that CBT@MNs might interfere TFRC recycling and endosomal acidification. KEGG pathway analysis additionally highlighted normalization of endocytosis, endosome, and phagosome-related signaling in IR + CBT@MNs group (Fig. 8D). CBT@MNs also exerted potent anti-inflammatory effects. Module C2, enriched for JAK/STAT and TNF signaling pathways, was strongly induced in the A2 group and sharply suppressed following treatment, with the A4 group exhibiting the most profound reduction, approaching or falling below baseline expression levels (Fig. 8A). Pro-inflammatory mediators were elevated in the A2 group and robustly downregulated in the A4 group (Fig. 8B). These changes were reflected in GO terms associated with inflammatory response, wound healing, and epithelial cell proliferation (Fig. 8C), and in KEGG analysis emphasizing suppression of multiple inflammatory signaling cascades (Fig. 8D). Furthermore, CBT@MNs demonstrated robust properties of anti-ferroptosis. Module C3, enriched for iron ion, transferrin transport pathways and ferroptosis, was significantly activated in the A2 group and broadly inhibited by MNs treatment. The most pronounced suppression was evident in the A4 group (Fig. 8A). Correspondingly, GPX4—a key ferroptosis suppressor—was downregulated in the A2 group and restored in the A4 group, whereas ACSL4 and PTGS2 were upregulated in the A2 group and normalized in the A4 group. Genes involved in iron trafficking were also markedly downregulated in the A3 and A4 groups relative to A2 (Fig. 8B): STEAP3 reduces Fe3+ to Fe2+ within endosomes, SLC11A2 transports Fe2+ into the cytoplasm, and SLC25A37 mediates mitochondrial iron import, thereby influencing iron homeostasis and lipid peroxidation. MNs treatment, particularly in the A4 group, significantly interfered with both iron reduction and transport processes. DEGs were notably enriched for biological processes and molecular functions related to iron ion transport, transferrin receptor binding, and intracellular iron homeostasis (Fig. 8C). KEGG enrichment analysis confirmed significant suppression of ferroptosis and unsaturated fatty acid biosynthesis pathways (Fig. 8D). These integrated transcriptomic data indicated that CBT@MNs remodeled the post-irradiation microenvironment through a dual mechanism combining inhibition of inflammation and ferroptosis. To validate the transcriptomic findings, we assessed ferroptosis and inflammation levels in rats tongue tissue. Immunohistochemical staining revealed brown-positive signals for interleukin-1 beta (IL-1β) and interleukin-6 (IL-6) localized predominantly in macrophages within deep tongue tissue and perivascular regions (Fig. 9A). Expression of both cytokines was markedly elevated in the IR group and progressively diminished across treatment groups. Notably, the CBT@MNs group exhibited the lowest levels of these pro-inflammatory mediators (Fig. 9D–E). Next, we evaluated macrophage polarization in vivo. iNOS served as an M1 macrophage marker, and Arg-1 as an M2 macrophage marker. The results revealed that M1 macrophages were highly expressed in IR group, whereas their proportion gradually decreased in the treatment groups, with the most pronounced reduction observed in the CBT@MNs group. M2 macrophages were detected in all groups at a certain level (Fig. S7B). CBT@MNs attenuated ferroptosis and inflammation in vivo. Representative IHC images of tongue tissue sections stained with antibodies of IL-1β and IL-6 (A) at day 14. Scale bar, 100 μm. Bar graphs showed IL-1β and IL-6 positive cells (D-E). (B) Representative images of tongue tissue sections stained for Prussian blue. The brown spots indicated the iron deposits in tongue tissue. Scale bars, 400 μm. (F) Bar graphs showed positive staining area. (C, G-H) Representative Immunofluorescence images of tongue tissue sections stained with antibodies of GPX4 and ACSL4 at day 14. Scale bar, 100 μm. Bar graphs showed relative fluorescence intensity of GPX4 and ACSL4. Statistical significance was determined using one-way ANOVA followed by Tukey's multiple comparisons test. Data are presented as mean ± SD. n = 3, *P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) Tissue iron content was subsequently evaluated via Prussian Blue staining. Iron in tissue reacted with the staining solution to form blue iron precipitates. Subsequent addition of DAB solution produced brown chromogen via redox reaction, enabling detection of trace iron deposits. As shown in Fig. 9B, the IR group displayed extensive iron deposition (brown puncta) localized to damaged epithelial regions, whereas iron content was substantially reduced in all treatment groups, with the CBT@MNs group exhibiting the lowest iron levels (Fig. 9F). GPX4 (glutathione peroxidase 4), a critical ferroptosis protein, protects cell membranes from oxidative damage by catalyzing the reduction of lipid peroxides [28]. ACSL4, conversely, is a key lipid peroxidation marker that accumulates prominently during ferroptosis [29]. Immunofluorescence staining (Fig. 9C and 9G–H) confirmed that GPX4 expression was significantly diminished in the IR group relative to controls and partially restored in the CBT@MNs group. In parallel, ACSL4 was robustly upregulated in the IR group and markedly suppressed in the CBT@MNs group, consistent with the transcriptomic data. Collectively, IR induced pronounced ferroptosis and inflammation in tongue tissue, and that CBT@MNs treatment effectively attenuated both processes, achieving favorable therapeutic outcomes in RIOM. Discussion Multiple lines of evidence indicated that IR induces ferroptosis, a form of programmed cell death characterized by iron accumulation and lipid peroxidation [[30], [31], [32]]. Ferroptosis may represent the predominant mode of radiation-induced oral epithelial cell death [5]. We previously found that MNs loaded with the ferroptosis inhibitor DFO could alleviate RIOM in rats [33]. However, DFO was often limited by systemic toxicity and poor bioavailability, necessitating the exploration of alternative approaches to modulate ferroptosis. Recent studies revealed that endocytosis of TF was an essential step for ferroptosis [34,35]. This process depended on TFRC recycling. TF-Fe3+ engaged TFRC at the plasma membrane, and the resulting complex was internalized via endocytosis into early endosomes. Within the acidic endosomal lumen, iron dissociated from TFRC, and the TF–TFRC complex recycled back to the cell membrane. Concurrently, Fe3+ was reduced to Fe2+ within endosomes and subsequently transported into the cytoplasm and mitochondria, where it fueled the Fenton reaction to drive ferroptosis. We therefore hypothesized that modulating endosomal pH would disrupt iron dissociation from TFRC, impair TFRC recycling, and attenuate cellular iron uptake. Additionally, interfering with endosomal iron reduction might further suppress downstream Fenton reaction. Through these combined effects, ferroptosis in irradiated epithelial cells could be mitigated, offering therapeutic benefit for RIOM. CQ, a widely used antimalarial agent, also functions as a classical endosomal alkalinizing compound. In this study, we sought to delineate its efficacy of inhibiting ferroptosis and evaluate its potential as a radioprotective agent with broadened clinical indications. Notably, we found that CQ and the bioactive flavonoid BA, derived from Scutellaria baicalensis Georgi, underwent self-assembly to form carrier-free CB NPs, substantially improving drug stability and solubility. In vitro experiments demonstrated that CB NPs reduced endosomal acidity, limited cellular iron uptake, and ultimately attenuated IR-induced ferroptosis. Moreover, CB NPs exhibited potent anti-inflammatory activity in vitro. Given the limited local drug concentration and risk of retinal toxicity associated with systemic CQ administration, we engineered a novel detachable microneedle patch (CBT@MNs) for localized delivery of CB NPs into the oral submucosa. In vivo studies confirmed that CBT@MNs effectively suppressed both ferroptosis and inflammation, achieving favorable therapeutic outcomes in rat models of RIOM. Transcriptomic profiling further revealed that CBT@MNs downregulated ferroptosis- and inflammation-associated signaling pathways in tongue tissue. These findings underscored the translational potential of CBT@MNs for radioprotection. The relationship between mitochondria and ferroptosis is complex and context dependent. On the one hand, mitochondria are a major source of cellular ROS and play a central role in iron metabolism. Their high labile iron pool and robust ROS generation make them an ideal site for initiating ferroptosis. On the other hand, functionally intact mitochondria contribute to the maintenance of redox homeostasis, and mitochondrial quality control mechanisms such as mitophagy can protect cells from ferroptosis [[36], [37], [38], [39]]. In the present study, we observed characteristic mitochondrial ultrastructural changes in the irradiated cell, including increased membrane density, reduced cristae, and outer membrane rupture. These changes were significantly attenuated by CB NPs treatment. It was important to note that we employed mitochondrial morphology solely as a ferroptosis marker. The improvement in mitochondrial ultrastructure was most likely a secondary consequence of suppressed ferroptosis, although a potential direct protective effect on mitochondria cannot be entirely excluded and needed further investigation. Despite these encouraging results, several limitations and challenges remained. First, although CB NPs demonstrated therapeutic promise against radiation injury, the current evidence for its disruption of TFRC recycling remained indirect. Beyond endosomal alkalization, CQ is known to exert multiple biological effects, including immunomodulation, autophagy inhibition, and regulation of cellular metabolism. We were therefore unable to definitively conclude that the primary mechanism underlying the therapeutic efficacy of CB NPs was the disruption of TFRC recycling. We could only speculate that TFRC recycling may represent an important target in the treatment of RIOM. In future studies, we planned to directly assess TFRC recycling efficiency using a TFRC-Alexa Fluor pulse-chase assay and examine the impact of TFRC knockdown and overexpression on the effects of CB NPs, in order to more precisely elucidate the principal mechanism. Second, the RIOM rat models employed here utilized a single high-dose radiation regimen, which differed from the fractionated low-dose schedules typical of clinical radiotherapy. Single high-dose irradiation caused severe, irreparable DNA double-strand breaks and might excessively amplify ferroptosis. In contrast, fractionated low-dose irradiation allowed normal tissues to undergo DNA repair and repopulation between fractions, leading to a more dynamic pattern of cell death. Consequently, the single high-dose RIOM model used in this study might increase the relative contribution of ferroptosis, thereby potentially amplifying the therapeutic effect of CBT@MNs. In our model, prophylactic treatment was initiated immediately after irradiation and covered the acute injury phase. In the fractionated radiotherapy, the optimal therapeutic window would need to be redefined. In future studies, we plan to refine the fractionated irradiation regimen by reducing the dose and increasing the number of fractions to lower mortality, establishing a stable oral mucositis model. Third, the biosafety evaluation in the present study was limited to a 14-day observation period, which covered the acute phase of RIOM but did not allow assessment of potential chronic or delayed toxicities. For instance, although the total dose of CQ administered per rat was far below the reported toxic threshold for systemic use, CQ had a long terminal half-life and known retinal and cardiac toxicities upon chronic systemic exposure. Furthermore, the impact of repeated MN insertion on the long-term integrity of the oral mucosal barrier, particularly in the context of radiation-induced vascular damage and fibrotic predisposition, remained to be elucidated. Future studies should extend the monitoring period to at least 3–6 months and include comprehensive assessments of organ function, detailed retinal examinations, and analysis of local tissue fibrosis or hyperplasia. CB NPs exhibited excellent serum stability, maintained uniform particle size without significant aggregation, and retained in vitro bioactivity after prolonged storage. A major concern for any oral drug delivery system was the potential compromise of retention by salivary flow and tongue movement. The detachable design of CBT@MNs physically separated the drug-loaded tips from the backing layer immediately after application to prolong retention time. Furthermore, our treatment protocol of reapplication every 48 h increased the drug concentration on the tongue. For future clinical administration, the use of mucoadhesive backing materials or optimized needle shape could further prolong residence time in oral cavity. Conclusion We developed a novel detachable MN patch CBT@MNs for the treatment of RIOM. The backing layer, composed of soluble PVP, dissolved rapidly upon contact with the moist oral environment, leaving the drug-loaded needle tips embedded within the oral mucosa. The needle tips integrated CB NPs, TA, and HAMA. CB NPs reduced endosomal acidity and limited cellular iron uptake, thereby attenuating radiation-induced ferroptosis in epithelial cells. At the same time, CB NPs exerted broad-spectrum anti-inflammatory and antioxidant activities, preventing amplification of inflammatory signaling. This strategy reversed radiation-induced oral mucosal injury. It therefore holds promise for maintaining patient adherence to radiotherapy and enhancing curative outcomes in the treatment of malignant tumors. Materials and methods Materials CQ(HY-17589A), dimethyl sulfoxide (DMSO) (HY-Y0320), UREA (HY-Y0271), SDS (HY-Y0316B), NaCl (HY-Y0344D), Tocofersolan (TPGS) (HY-B0717) and D-α-Tocopherol Succinate (α-Tos) (HY-131553) were purchased from MedChemExpress (USA). HAMA (MW = 150,000 Da) and Lithium Phenyl (2,4,6-trimethylbenzoyl) phosphinate (LAP) were purchased from the EFL Reagent Company (China). BA (HY-N0197), Acetone (D167514), Methanol (M433283) and PVP (MW = 1300000 Da, K88-96), TA (T432042) were purchased from Aladdin (China). Antibodies recognizing STAT3(Phospho Ser727) (AWA00668), STAT3 (AWA02296), β-actin (AWA80113), IL-1β (AWA46641) and IL-6 (AWA46788) were purchased from Abiowell (China). Antibodies recognizing TFRC (ab269513), GPX4(ab252833), ACSL4 (ab15582) and Goat Anti-Rabbit antibody (ab288151) were purchased from Abcam (UK). Polydimethylsiloxane (PDMS) was purchased from Taizhou Microchip Pharmaceutical Technology LTD (China). Synthesis and characterization of CB NPs To prepare CB NPs, 5 mg of CQ and 5 mg of BA were first dissolved in solutions (200 μL DMSO with 2 mg of α-Tos and 0.5 mg of TPGS). The aqueous phase comprised 4.8 mL of ultrapure water. Under stirring at room temperature, the organic solution was added dropwise into the aqueous phase. After 15 min of stirring, the mixture underwent dialysis purification for 2 h using a membrane with a molecular weight cut-off of 5000 Da. Spontaneous assembly was followed by the use of ultrafiltration centrifuge tubes to remove unencapsulated free drug and residual organic solvent. The morphology was observed using TEM (Titan G2 60−300, FEI, USA). The particle size, zeta potential and PDI were determined using dynamic light scattering (DLS, Zetasizer ZS90, Malvern, UK). Particle size was monitored over a 14-day period to assess stability. Loading efficiency (LE) was calculated according to the formula: LE (%) = (mass of drug incorporated/mass of nanoparticles) × 100%. To explore the factors governing nanoparticle assembly, the resulting mixtures were scanned by a microplate reader over the wavelength range of 235–450 nm to obtain UV spectra. CB NPs was mixed with various concentrations of UREA, SDS and NaCl (0, 20, 40, 60 and 80 mM). Simultaneously, the evolution of particle size in each mixture was recorded at multiple time points. Molecular dynamic simulation of CB NPs Molecular dynamics simulations were carried out using the GROMACS 2020.6 package, while structural visualization was performed with VMD (version 1.9.3). ORCA 6.0.1 was employed for structural optimization and single-point energy calculations, and the corresponding RESP charges were derived using Multiwfn. All molecular parameters were generated based on the GAFF2 force field. A cubic water box measuring 8 nm × 8 nm × 8 nm was constructed, into which a specified number of CQ and BA molecules were randomly introduced. The system then underwent 5000 steps of energy minimization, followed by 100 ps of equilibration under both NVT and NPT ensembles. Subsequently, a 100 ns production run was executed. The cutoff distance was set to 1.0 nm, with a time step of 2 fs. Data analysis included the RMSD, Rg, SASA, and the number of hydrogen bonds. Fabrication and characterization of CBT@MNs 4-(Bromomethyl) phenylboronic acid (1.5 g) was dissolved in DMF (60 mL) at 60 °C. Subsequently, N, N, N′, N′-tetramethyl-1,3-propanediamine (0.15 g) was added, and the mixture was stirred for 24 h. The resulting yellow solution was then poured into THF (150 mL) to induce precipitation. The precipitate was collected, washed three times with THF, and lyophilized to afford pure TS-PBA. To prepare CBT@MNs, solutions were sequentially introduced into a PDMS mold. First, a mixture containing CB NPs (1 mg/mL), HAMA (100 mg/mL), LAP (2.5 mg/mL) and TS-PBA (5 mg/mL) aqueous solution was added to the mold. After centrifugation for 10 min, the mold was exposed to ultraviolet light for 20 min to induce cross-linking, followed by drying in a blast oven at 30 °C for 5 h to solidify the needle tips. Subsequently, repeating the above procedures. Finally, a PVP solution (100 mg/mL) was poured into the mold and centrifuged for another 10 min to form the backing layer. The mold was then left to dry overnight at room temperature, allowing the CBT@MNs to detach. The CBT@MNs were characterized for their morphology using a digital camera (D7100, Nikon, Japan) and SEM (TM 3030, Hitachi, Japan). Mechanical properties were evaluated with an electronic universal material testing machine (Instron 5967, USA). In detail, CBT@MNs patch was positioned on a platform, and a vertical compressive force was applied by a mechanical sensor at a rate of 10 μm/s. Force readings were recorded continuously until the maximum compression point was reached. The in vivo degradation performance of CBT@MNs was recorded by Super depth 3D microscope system (DSX1000, Olympus Japan). Mucosal penetration assay To evaluate mucosal penetration, the CB@MNs and CBT@MNs were pressed onto the dorsal surface of rat tongue mucosa for 1 min. Following tongue mucosa collection, the penetration performance of CB@MNs and CBT@MNs was determined through tissue sectioning combined with H&E staining. The extent of epithelial continuity disruption of the MNs was examined. High-performance liquid chromatography (HPLC) The HPLC (UltiMate 3000, ThermoFisher, USA) was used to assess in vitro release profiles of CQ and BA in PBS and H202. CQ parameters were as follows: C18 phase column, potassium dihydrogen phosphate 1.2 g/L: acetonitrile (65%: 35%) as the mobile phase, flow rate of 1.2 mL/min, detection wavelength of 343 nm. BA parameters were as follows: methyl alcohol: phosphoric acid (46%: 54%), detection wavelength of 280 nm. Cell culture and IR Human Umbilical Vein Endothelial Cells (HUVEC) and RAW264.7 cells obtained from ATCC, were cultured in Dulbecco's Modified Eagle's Medium (DMEM) with 10% fetal bovine serum. DMEM was obtained from Gibco (USA), while fetal bovine serum (FBS) was purchased from ExCell Bio (Jiangsu, China). The HOK cell line (Human Normal Oral Epithelial Keratinocytes), a gift from Dr. JS Gutkind (National Institute of Dental and Craniofacial Research, MD, USA), was maintained in Keratinocyte Serum-Free Medium (Thermo Fisher Scientific, USA). Both cell types were irradiated using an X-RAD 225 irradiator (Precision X-ray Inc., North Branford, CT) at a dose of 12 Gy. Cell viability and cell death assays HOK cells (1 × 105 cells/mL) and HUVEC (5 × 104 cells/mL) were seeded in 96-well plates (100 μL/well) and incubated overnight. The cells were then pretreated for 24 h with CQ (1 μM), BA (1 μM) or CB NPs (1 μM) before receiving 12 Gy irradiation. After 72 h, the Cell Counting Kit-8 (CCK8, CK04, Dojindo) and Calcein-AM/PI staining (C2015S, Beyotime) were used to assess cell viability and cell death, respectively, following the manufacturers’ instructions. LysoTracker and LysoSensor staining LysoTracker Red DND-99 (100 nM, L7528, Invitrogen) was added to the culture medium of HOK cells and incubated at 37 °C for 30 min, followed by fluorescence imaging using confocal laser scanning microscopy (CLSM, Leica STELLARIS 5) with excitation at 577 nm. The fluorescence intensity was quantified using ImageJ. Separately, cells were incubated with LysoSensor Yellow/Blue DND-160 (1 μM, L7545, Invitrogen) under the same conditions (37 °C, 30 min), and fluorescence was captured at an excitation wavelength of 329⁄440 nm using CLSM. Cathepsin B activity Cathepsin B activity was assessed using the Magic Red assay kit (ab270772, Abcam). Following the manufacturer's protocol, a stock solution was prepared, and HOK cells were incubated with the reagent at a 1:1000 dilution for 30 min at 37 °C. Fluorescence images were captured with CLSM using an excitation wavelength of 577 nm. The fluorescence intensity was quantified using ImageJ. Immunofluorescence RAW264.7 cells on glass coverslips were fixed (4% paraformaldehyde, 15 min, room temperature), washed three times with PBS, and permeabilized with 0.05% Triton X-100 for 10 min. After three PBS washes (5 min each), the cells were blocked with 5% BSA for 30 min and then incubated overnight at 4 °C with primary antibodies (TFRC, STAT3, pSTAT3). The coverslips were washed three times with PBST (5 min per wash), followed by 1 h incubation with anti-rabbit Alexa Fluor 488 secondary antibody (room temperature, dark). After three additional PBS washes (5 min each), nuclei were stained with DAPI. CLSM was used for imaging, and ImageJ for fluorescence quantification. Measurement of cellular ROS and lipid peroxidation To detect ROS and lipid peroxidation, HUVEC, HOK and RAW264.7 cells were incubated with 5 μM DCFH-DA (Biosharp) and 10 μM BODIPY 581/591C11 (Thermo Fisher). MDA levels were quantified in both cell samples and rat tongue tissues (harvested on day 14 post-IR) using the MDA Content Assay kit (BC0025, Solarbio) on a microplate reader (532/600 nm), following the manufacturer's instructions. Determination of iron in cell Intracellular Fe2+ levels in HOK, HUVEC and RAW264.7 cells were assessed using the FerroOrange probe (F374, Dojindo). After 24 h, cells were incubated with 1 μM FerroOrange for 30 min at 37 °C, washed with PBS, and then imaged by CLSM with an excitation wavelength of 577 nm. Fluorescence intensities were quantified with ImageJ. Mitochondrial morphology detection Mito-Tracker Green probe (C1048, Beyotime) was employed to label mitochondria. After 24 h, HUVEC cells were stained with 100 nM of the probe (30 min, 37 °C), washed with PBS, and imaged under CLSM. Transmission electron microscopy (TEM) HUVEC cells were fixed with PBS containing 2.5% (w/v) glutaraldehyde overnight. Samples were post-fixed with 1% osmium, rinsed, stained with 2% uranium acetate, dehydrated, and embedded. Ultra-thin sections (0.07–0.09 μm) were cut with an ultramicrotome (EM UC7, Leica) equipped with a diamond knife and then stained with lead citrate (Reynolds solution). Images were acquired via an FEI Tecnai G2 Spirit BioTwin microscope (Thermo Fisher) with a digital camera (Morada, Olympus Soft Image Solutions). Western blot analysis A standard Western blotting protocol (described previously) was followed. Cells were lysed in a RIPA buffer (P0013B, Beyotime) and centrifuged at 10,000 g for 10 min. The BCA method (CW0014S, CWBIO) was used to quantify protein. 20 μg of each lysate was mixed with loading buffer (P0015, Beyotime), boiled for 10 min, separated on 10% SDS-PAGE gels, and transferred onto 0.45-μm PVDF membranes (IPVH00010, Millipore) on ice. Membranes were blocked with 5% non-fat milk and then probed at 4 °C overnight with primary antibodies: anti-TFRC (1:5000), anti β-actin (1:5000), anti-STAT3 (1:2000) and anti-pSTAT3 (1:2000). After washing, membranes were incubated with secondary antibodies (goat anti-rabbit, 1:5000) for 2 h at room temperature. Immunoreactivity was visualized with an ECL substrate (Millipore, WBKLS0500). ELISA Cell culture supernatants were collected and centrifuged at 3000 rpm for 15 min at 2–8 °C. Concentrations of TNF-α, IL-1β, and IL-6 were determined using a chemiluminescence immunoassay. Before measurement, the chemiluminescence analyzer was prepared by checking the adequacy of wash solution, reaction cuvettes, tips, pre-trigger solution, trigger solution, and waste containers. Wash solution was diluted 20-fold with purified water, and pre-trigger/trigger solutions were equilibrated to room temperature. Calibration was performed using low- and high-value calibrators. Subsequently, the prepared supernatants were placed on the sample rack to measure the chemiluminescent signals. The concentrations of TNF-α, IL-1β, and IL-6 in each sample were automatically calculated. Flow cytometry RAW264.7 cells were pretreated with CQ, BA, CB and CB NPs (1 μM) for 1 h, then stimulated with LPS (1 μg/mL) for 24 h. The cells were harvested and stained with fluorescently labeled monoclonal antibodies PE-F4/80, PE-CY7/CD86 and BV711/CD206. Finally, the M1 and M2 polarization of RAW264.7 cells was analyzed by flow cytometry. Construction and treatment of the RIOM rat models SD rats were obtained from Hunan Silaikejingda Experimental Animal Co., Ltd. (China) and maintained under specific pathogen-free conditions with free access to food and water. After one week of acclimatization, all animals were weighed and ranked by body weight. Block randomization was performed using a computer-generated random number sequence to assign animals to each group, ensuring comparable mean body weight across groups before the start of the experiment. 36 rats were randomly allocated to six groups (n = 6/group): Control, IR, IR + CQ@MNs, IR + BA@MNs, IR + CB@MNs (CQ + BA) and IR + CBT@MNs. The RIOM model was induced by a single 25 Gy irradiation to the dorsal surface of the tongue (X-RAD 225 irradiator) with lead shielding of the rest of the body. Sample size was determined based on previous studies [5,40]. The drug dose per MN patch was estimated based on the total volume of 100 pyramidal needles (20.8 μL) and a CB NPs concentration of 1 mg/mL, yielding a theoretical load of approximately 20.8 μg CB NPs per patch. The patches were applied topically to the rat dorsal tongue. MN treatment was administered once every two days, beginning on the day of 25 Gy irradiation (day 0), with the first patch applied within 1 h after IR, for a total of four doses on days 0, 2, 4, and 6. All treatments were performed by one investigator, while all subsequent assessments—including mucosal scoring, histological grading, and image quantification—were carried out by a second investigator blinded to group allocation. All animals were monitored daily for general health status. Body weight was recorded every two days at the same time of day. Oral mucosal appearance was evaluated under brief isoflurane anesthesia on days 4, 6, 8, 10, 12, and 14. Healthy male SD rats (6–8 weeks old, weighing 200–250 g) without oral mucosal lesions were included. Rats with pre-existing oral lesions or >10% weight loss during acclimation were excluded before IR. Rats that died prematurely from anesthesia complications or severe self-mutilation were excluded from final analysis. Humane endpoints were predefined as: (1) body weight loss exceeding 25% of baseline; (2) inability to eat or drink normally for >48 h; (3) severe systemic infection or uncontrolled pain. Animals meeting any of these criteria were immediately euthanized by CO2 asphyxiation. All animal experiments were conducted in accordance with protocols reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the Second Xiangya Hospital, Central South University, China (reference: 20250921). Oral mucositis scoring system Oral mucositis severity was assessed every two days using a scoring system adapted from the RTOG criteria (https://www.rtog.org) and modified based on observations in irradiated rats. The scoring scale was defined as follows: 0 = normal mucosa; 0.5 = faint pink discoloration; 1 = mild erythema with local vasodilation; 2 = marked erythema and vasodilation accompanied by punctate erosion; 3 = focal desquamation with ulcer formation covering less than 25% of the area; 4 = exudate involving less than half of the irradiated mucosa, with ulcers occupying 25–50% of the area; 5 = nearly complete ulceration of mucosa. In vivo safety study The in vivo biosafety of CBT@MNs was evaluated through histopathological and blood biochemical analyses. On day 14, the major organs (heart, liver, spleen, lungs, and kidneys) were collected from the rats across all six groups, fixed in 10% paraformaldehyde, embedded in paraffin, and sectioned. Tissue sections were stained with H&E to assess tissue morphology and potential toxicity. Whole blood and serum were collected from the rats; complete blood count parameters were obtained using a hematology analyzer, and levels of alkaline phosphatase (ALP), alanine aminotransferase (ALT), and creatinine (CR) were measured using an automatic biochemical analyzer. To determine an appropriate dosing regimen, single doses of 10, 20, and 30 μg of MNs were applied to the dorsal tongue of rats on days 0, 2, 4, and 6. On day 7, the major organs and serum were collected for the aforementioned assays to evaluate dose-related toxicity. Transcriptome sequencing and data analysis On day 14, tongue tissue collected from rat were sent to Jiangsu Sanshu Biotechnology Co., LTD for transcriptome sequencing. Total RNA was extracted from the tongue mucosa (n = 3 biological replicates per group). A NanoDrop 2000 spectrophotometer was used to evaluate the purity and concentration of total RNA, whereas the Agilent 4200 system served to determine RNA integrity and quantity. The R package Lima (Voom) was utilized for differential expression analysis. To control the false discovery rate, P-values were adjusted using the Benjamini-Hochberg method. Genes were considered differentially expressed if they satisfied both conditions: |log2 (Fold Change) | > 1 and an adjusted P-value (q-value) < 0.05. H&E, Masson, Prussian blue, IHC and immunofluorescence staining On day 14 after IR, tongue tissues were collected and fixed overnight in 4% paraformaldehyde. The samples were subsequently embedded in paraffin, cut into 5-μm sections. For each animal, three non-consecutive sections were stained with H&E, Masson, Prussian blue, and IHC. Stained slides were imaged with an Olympus BX63 microscope. All quantitative measurements were performed by an independent investigator blinded to group allocation. Representative images for each group were selected based on the median value of the corresponding quantitative parameter and no animals or tissue sections were excluded from the quantitative analyses. Three animals per group were used for histological analyses. For H&E-stained sections, epithelial thickness was measured at three evenly spaced points along the dorsal tongue epithelium per section using ImageJ (v1.54, NIH), and the mean of three sections was taken as the value for that animal. On Masson-stained sections, muscle fiber diameter was determined by measuring the minimal diameter of six transversely sectioned fibers per animal and averaging. Collagen volume fraction (CVF) was quantified as the percentage of blue-stained (collagen) area relative to the total tissue area within each field, using the color thresholding function of ImageJ. Prussian blue staining was performed with a Prussian Blue Iron Stain Kit (G1428, Solarbio, China) according to the manufacturer's instructions. Iron deposition was expressed as the percentage of blue-positive area per field via color deconvolution or thresholding in ImageJ. IHC staining followed a previously described protocol. IL-6- and IL-1β-positive cells were counted manually in three fields per section using the ImageJ Cell Counter plugin, and relative numbers per group were calculated. CD31-positive blood vessels were counted in three non-overlapping fields per section. For immunofluorescence, sections were stained overnight with GPX4 (ab125066, Abcam, 1:200), ACSL4 (ab155282, Abcam, 1:200) iNOS (ab178945, Abcam, 1:250) and CD206 (ab300621, Abcam, 1:500). GPX4 and ACSL4 relative fluorescence intensity were calculated by ImageJ. Quantification and statistical analysis Statistical analyses were performed using GraphPad Prism 10 (GraphPad Software, San Diego, CA, USA). All data are expressed as the mean ± standard deviation (SD). For comparisons among three or more groups, a one-way analysis of variance (ANOVA) was used. Homogeneity of variances was assessed using the Brown–Forsythe test. To reduce the likelihood of false-positive results, Tukey's correction for multiple comparisons was implemented. P < 0.05 was considered statistically significant. CRediT authorship contribution statement Yanxuan Xiao: Writing – original draft, Investigation. Pengcheng Sun: Writing – review & editing, Investigation. Wenjie Xu: Writing – review & editing, Investigation. Xinyan Hao: Writing – review & editing, Investigation. Tiantian Tang: Validation, Data curation. Xinying Liu: Validation, Data curation. Hai Huang: Validation, Data curation. Zhaojian Gong: Funding acquisition, Conceptualization. Junyong Wu: Funding acquisition, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements This research was supported by The Science and Technology Innovation Program of Hunan Province 2024RC3061 (J.W.), the Scientific Research Launch Project for new employees of the Second Xiangya Hospital of Central South University QH20230269 (J.W.), the Hunan Provincial Natural Science Foundation of China 2025JJ50546 (Z.G.) and the Medical Discipline Construction Project of Hunan Province (The department of oral and maxillofacial surgery of Second Xiangya Hospital, Central South University, 2025). Footnotes Supplementary data to this article can be found online at https://doi.org/10.1016/j.mtbio.2026.103642. Contributor Information Zhaojian Gong, Email: gongzhaojian4458@csu.edu.cn. Junyong Wu, Email: wujunyong@csu.edu.cn. Supplementary data The following is/are the supplementary data to this article: Multimedia component 1 Data availability Data will be made available on request. References Associated Data Supplementary Materials Multimedia component 1 Data Availability Statement Data will be made available on request.