Tomentosin selectively targets microglial pyroptosis to overcome fluoxetine-resistant depression: a network-based therapeutic discovery

Abstract

Treatment-resistant depression (TRD), a clinically challenging issue of major depressive disorder (MDD), affects up to one-third of patients and is associated with elevated suicide risk and limited treatment options. Cumulative evidence highlights pathological microglial state and inflammasome-derived pyroptosis as key contributors to TRD pathophysiology. This study aimed to validate tomentosin as a network-based identified antidepressants and anti-microglial candidate and to define its mechanisms in suppressing microglial pyroptosis. Through a network-based multiscale interactome screening of a large terpenoid library exceeding 170,000 compounds, we identified tomentosin, a brain-penetrant sesquiterpene lactone with favorable drug-likeness and network relevance. In mice unresponsive to fluoxetine (called as FRD, fluoxetine-resistant depression), tomentosin (20 mg/kg) significantly alleviated depressive behaviors and normalized reactive microglial states in the anterior cingulate cortex (ACC). These pharmacological effects were observed in systemic and intracerebral inflammation-induced depressive mice models. We found that tomentosin mechanistically targeted the suppression of microglial NOD-like receptor protein-3 (NLRP3)/caspase-1/gasdermin D (GSDMD) signaling pathway. This inflammasome-specific suppressive effect was confirmed by the absence of pharmacological effects in caspase-1 knockout (Casp1 KO) mice. Underlying mechanisms were further validated through molecular interaction analyses, comparative studies with inhibitors, and overexpression vector transfections. Our findings suggest that tomentosin is a novel agent that selectively modulates inflammasome-associated microglia in FRD, primarily by suppressing pyroptosis in the ACC.

Introduction

Major depressive disorder (MDD) is a debilitating mental illness affecting 4.4% of the global population [1]. Its prevalence has steadily increased over the past few decades, making it a leading cause of disability worldwide [2] Although antidepressants are prescribed widely, about 50% of patients with MDD do not respond to first-line antidepressants, and one-third remain unresponsive even after adding a second antidepressant [3, 4]. Treatment-resistant depression (TRD) has long been recognized as a critical challenge, with patients exhibiting a 3-fold higher risk of suicide and self-harm compared to those with non-resistant forms of depression [5].

The pathophysiology of TRD is believed to be multifactorial, involving impaired transmission of monoamine neurotransmitters, dysregulation of hypothalamus–pituitary–adrenal (HPA), neuroinflammation, and neuronal dysconnectivity [6, 7]. Then, increasing evidence supports the microglial-targeted interventions against TRD [8, 9]. Individuals with TRD have been shown to exhibit greater translocator protein (TSPO, a marker of microglial activity) density than healthy and non-TRD patients, particularly in the anterior cingulate cortex (ACC) and prefrontal cortex (PFC) [10–12]. One hundred sixty-three young individuals who attempted suicide showed smaller frontal regions compared to 323 who had not [13]. Activated microglia cause neuroinflammation and can induce programmed neuronal cell death, known as pyroptosis, around microglia in which gasdermin D (GSDMD) pores have formed [14]. Whole-blood mRNA profiling of TRD patients has revealed an upregulation of inflammasome-associated genes [15]. Unfortunately, the anti-inflammatory candidates for TRD have yielded largely disappointing results in RCTs, particularly with lack of microglia-specificity [16, 17].

On the other hand, network pharmacology-based approaches, particularly multiscale interactome analysis, have emerged as powerful tools to identify high-specific candidate compounds that act on disease-relevant pathways [18]. The plant-derived volatile terpenoids have recently drawn attention for their potential central nervous system (CNS) activity [19, 20]. Leveraging this approach, we explored associations between natural terpenoids and microglia-dominant TRD networks and identified tomentosin–a sesquiterpene lactone derived from Inulae Flos (Inula japonica Thunb.)–as a promising therapeutic candidate. Inulae Flos has been traditionally used for its antidepressant properties [21], and tomentosin has previously demonstrated anti-inflammatory, anti-neuroexcitotoxicity, and neuroprotective effects [22–24]. Moreover, recent evidence confirms that Tomentosin can cross the blood-brain barrier (BBB) [25], further supporting is potential as a central nervous system (CNS)-active compound.

Building upon our network-guided findings, we experimentally validated the therapeutic potential of tomentosin in a fluoxetine-resistant depression (FRD) model by focusing on microglia-specific mechanisms. To elucidate its mode of action, we employed an integrated approach combined in silico network pharmacology with serial in vivo models and in vitro functional assays using BV2 and HMC3 microglial cells. These findings establish a mechanistic foundation for the further development of tomentosin as a novel treatment candidate for TRD, at least fluoxetine-unresponsive depression.

Material and methods

Compound-target-disease network construction

To systematically identify potential terpenoid candidates targeting MDD-associated neuroinflammation, we first retrieved proteins associated with both “MDD” and “neuroinflammation” using PubMed-based text mining via the Cytoscape-StringApp. Terpenoid compounds were sourced from TeroKit (http://terokit.qmclab.com/), a specialized database for terpenome research [26]. Experimentally validated compound–target interactions were compiled from DrugBank, search tool for interactions of chemicals (STITCH), and Therapeutic Target Database (TTD). After mapping these compounds to valid PubChem IDs, the dataset was curated to include only those terpenoids for which sufficient target (≥2) information was available. BBB permeability was predicted using SwissADME (http://www.swissadme.ch/), and quantitative estimation of drug-likeness (QED) was computed via the PubChemPy module in Python. The QED threshold was set to 0.35 based on the comparative analysis [27], where this cutoff exhibited performance superior to Veber’s rule (Fig. 1A).

Fig. 1. Network pharmacology-based identification of tomentosin as a novel candidate for treating TRD.

Compound–target–disease network construction was performed with assembling MDD–associated neuroinflammatory proteins (left), and 173,622 terpenoids were mapped to validated targets and filtered by quantitative estimate of drug-likeness (QED) and blood–brain barrier (BBB) permeability, yielding 615 candidates (right) A. Disease (red) and terpenoid (blue) targets were embedded in a multiscale interactome analysis; diffusion profiles were computed and Pearson correlations (0-1) used to quantify network-level similarity B. The top 50 candidates were prioritized by correlation score and displayed with circos plot, and then tomentosin was selected by exclusion criteria and further in vitro assay-based experimental validation C.

Multiscale interactome analysis

A multiscale interactome was adopted [28], comprising protein–protein, protein–biological function, and biological function–biological function associations. Relevant interaction data were gathered from the Biological General Repository for Interaction Datasets (BioGRID) and the Database of Interacting Proteins. Diffusion profiles were then computed for terpenoid targets and for the MDD–neuroinflammation protein set by employing a matrix-based power iteration (biased random-walker) algorithm, which incorporated restart probabilities and scalar weights for node transitions. Pearson’s correlation was utilized to measure the similarity between each compound’s diffusion profile and the disease diffusion profile. Finally, the top k proteins or biological functions (with k = 20) most influenced in these diffusion profiles were extracted to construct a subnetwork representing key mechanisms. This threshold was determined based on a multiscale interactome study demonstrating that high-ranking nodes in diffusion profiles correspond to the most strongly propagated drug effects and account for the majority of the overall visitation frequency.

Animals

A total of eighty male, specific pathogen-free C57BL/6J mice (8 weeks old, weighing 22–24 g) were purchased from Dae Han Bio Link (Co., Ltd., Eumseong, Korea). Additionally, 12 wild-type and 12 caspase-1 knockout (B6N.129S2-Casp1^tm1Fkv/J) mice were obtained from The Jackson Laboratory (Bar Harbor, ME, USA). The mice were housed under controlled conditions at 23 ± 2 °C (temperature) and 55 ± 10% (humidity) using a thermohygrostat (ALFFIZ, BuSung Co., Ltd, Seoul, Korea), with a 12-h light-dark cycle (09:00 to 21:00) and ad libitum access to food (Cargill Agri Furina, Gyeonggido, Korea).

All animal care and experimental protocols were approved by the Institutional Animal Care and Use Committee of Daejeon University (DJUARB2024-019) and were conducted in accordance with the Guide for the Care and Use of Laboratory Animals, as published by the National Institutes of Health (NIH).

Four mouse models and tomentosin administration

After a one-week acclimation period, the mice were allocated to three for the following experimental questions: (1) whether tomentosin overcomes FRD phenotype in an unpredictable chronic mild stress (UCMS) model (n = 24), (2) whether tomentosin exerts anti-depressive effects on lipopolysaccharide (LPS) injection-derived systemic inflammation model (n = 32), (3) whether the effects of tomentosin are dependent on the inflammasome using a Casp1 KO model challenged with LPS (n = 24). and (4) whether tomentosin inhibits microglial inflammasome-specific depression in a recombinant interleukin (rIL)-1β intracerebroventricular injection model (n = 24), respectively.

According to manufacturer’s guideline, the tomentosin was dissolved in 1% of DMSO. To prevent bias arising from different conditions, solvent and administrating route were evenly applied to all the designated groups.

UCMS model

Mice were randomly assigned to three groups as follows: normal (n = 6), UCMS (n = 6), and UCMS with fluoxetine (10 mg/kg, prepared in drinking water; PHR1394, Sigma-Aldrich, MO, USA; n = 12, one mouse death by drowning). After UCMS for 4 weeks, the fluoxetine-treated mice were divided into responder and non-responder called as TRD group following two behavioral tests such as tail suspension test (TST) and forced swimming test (FST). The TRD group was intraperitoneally administrated with tomentosin (20 mg/kg; BD013978, BLDpharm, Kaiserslautern, Germany) for one week, and then all of mice were re-examined behavioral tests evaluating depressive symptoms (Fig. 2A). The UCMS schedules were followed as shown in Table S1.

Fig. 2. Effects of tomentosin on microgliopathy-dominant FRD in a UCMS mouse model.

The schematic of the experimental schedule was designed A. After 4-week UCMS induction, the activity durations in TST B and FST C were evaluated to divide responder to fluoxetine and non-responder (FRD). After a 1-week UCMS with fluoxetine or tomentosin, the activity durations in TST D and FST E were re-evaluated. Iba-1 positive microglial activity and NeuN positive neuronal activity in the ACC were assessed by double-staining analysis F. The number of NeuN-positive cells in the ACC was quantified using ImageJ G. Microglial soma size H and process length I in the ACC were evaluated via Sholl analysis. Data are expressed as the mean ± SD (n = 3, 5 or 6/group). ^#p < 0.05 and ^##p < 0.01 compared to the normal mice; ^*p < 0.05 and ^**p < 0^.01 compared to the UCMS-subjected mice.

Systemic LPS injection model

Mice were randomly assigned to four groups (n = 8/group): saline, LPS (1 mg/kg, dissolved in normal saline; E. Coli O111:B4, L2630, Sigma-Aldrich, MO, USA), LPS with tomentosin, and LPS with BAY11-7082 (5 mg/kg, dissolved in normal saline; HY-13453, MCE, NJ, USA), respectively. After an injection of LPS, mice were intraperitoneally administrated with tomentosin or BAY11-7082 once daily for three days along with three sequential behavior tests such as nest building test (NBT), open field test (OFT) and FST (Fig. 3A).

Fig. 3. Effects of tomentosin on inflammation-mediated depression in systemic LPS-injected and Casp1 KO mouse models.

The schematic of the experimental schedule in the LPS-injected mouse model was designed A. After intraperitoneal LPS injection, the nest-building score in NBT B, time spent in the center zone in OFT C, and active duration in FST D were evaluated in sequence. In the ACC region, microglial inflammasome was assessed by Iba-1/GSDMD and neuronal pyroptosis was examined by NeuN/GSDMD double-positive signals E. The experimental design to validate the anti-inflammasome effects of tomentosin was established using a Casp1 KO model F. The nest-building score in the NBT G, time spent in the center zone in the OFT (H), and active duration in the FST I were evaluated. In the ACC, microglial inflammasome and neuronal pyroptosis were examined using Iba-1/GSDMD and NeuN/GSDMD double-staining J. Data are expressed as the mean ± SD (n = 3, 6 or 8/group). ^#p < 0.05 and ^##p < 0.01 compared to the saline-injected mice; ^*p < 0.05 and ^**p < 0^.01 compared to the LPS-injected mice.

Casp1 KO mouse model under systemic LPS injection

To verify the regulatory effects of tomentosin on the NLRP3/caspase-1/GSDMD inflammasome pathway, wild-type and Casp1 KO mice were each divided into two treatment groups (n = 6 per group): LPS (1 mg/kg) with saline or LPS with tomentosin (20 mg/kg), resulting in a total of four experimental groups. Under the same conditions described above, the mice were subjected to three sequential behavioral tests: NBT, TST, and FST (Fig. 3F).

Intracranial rIL-1β injection model

Mice were randomly assigned to four groups (n = 6/group): artificial cerebrospinal fluid (aCSF), rIL-1β (10 ng/head, dissolved in aCSF; 1857-LC, R&D Systems, MN, USA), rIL-1β with tomentosin, and rIL-1β with BAY11-7082. The rIL-1β was injected into the left lateral ventricle using a robotic stereotactic drill and injection system (Neurostar, Tuebingen, Germany) at a flow rate of 1 μL/min for 4 min. The injection coordinates relative to bregma were as follows: AP: −0.5 mm, ML: −1.0 mm, DV: 2.5 mm. To prevent backflow, the syringe was left in place for 5 min after the injection. After intracranial rIL-1β injection, mice were intraperitoneally administrated with tomentosin or BAY11-7082 once daily for seven days. After a 3-day recovery period following intracranial injection, five consecutive behavioral tests were performed in order of increasing stress intensity (NBT, MBT; marble burying test, OFT, TST, and FST), with a 24-h recovery interval between tests to minimize cumulative interference (Fig. 4A).

Fig. 4. Effects of tomentosin on neuroinflammation-specific depression in the intracranial rIL-1β-injected mouse model.

The schematic of the experimental schedule was designed A. After the intracranial rIL-1β injection, the nest-building score in NBT B, number of buried marbles in MBT C, time spent in the center zone in OFT D, active duration in TST E and FST F were evaluated in sequence. In the ACC region, microglial inflammasome was assessed by Iba-1/GSDMD and neuronal pyroptosis was examined by NeuN/GSDMD double-positive signals G. The protein levels of neuronal activity markers (NeuN, c-Fos, BDNF, and PSD95) in the ACC were examined by western blotting analysis H. Data are expressed as the mean ± SD (n = 3 or 6/group). ^#p < 0.05 and ^##p < 0.01 compared to the aCSF-injected mice; ^*p < 0.05 and ^**p < 0^.01 compared to the rIL-1^β-injected mice.

Five behavioral tests and data analyses

All behavioral tests, including TST, FST, NBT, OFT, and MBT were followed as displayed in Figs. 2A–4A, and tracking data were recorded and calculated using a video camera connected to corresponding software (Smart 3.0, Panlab SL, Barcelona, Spain). All the performances were evaluated by researchers blinded to the experimental conditions. Detailed methods of all behavioral tests were described in supplementary materials.

Immunofluorescence analysis in ACC

Microglia activity (Iba-1; ionized calcium-binding adapter molecule 1, IL-1β; interleukin 1 beta, and GSDMD), astrocytic activity (GFAP; glial fibrillary acidic protein and S100β; S100 calcium-binding protein β) and neuronal integrity (NeuN; neuronal nuclei) in ACC were assessed using immunofluorescent staining analysis. To prepare brain sections for analyzing histologically, three independent mice per group were transcranially perfused with 0.05% heparin (10 units/mL in PBS) on the final day of tomentosin injection, followed by 4% paraformaldehyde (pH 6.9). The removed brains were gradually cryoprotected in 10, 20, and 30% sucrose for 24 h each and were subsequently embedded in an optimal cutting temperature (OCT) compound (Leica Microsystems, Bensheim, Germany) in liquid nitrogen. They were cut into frozen coronal sections (30 μm) using a cryostat (CM3050_S, Leica), and sections were stored in free-floating buffer.

In brief, the brain sections were incubated with blocking buffer (5% normal chicken serum in PBS and 0.3% Triton X-100 for 1 h at 4 °C), and then were adapted with anti-rabbit Iba-1 (1: 400, 019-19741, Wako Biologicals), anti-rabbit IL-1β (1:100, ab9722, Abcam), anti-rabbit GFAP (1:200, Z0334, Dako), anti-rabbit GSDMD (1:100, #46451, Cell Signaling), anti-mouse NeuN (1:200, MAB377, Merck-Millipore), and anti-rabbit S100β (1:100, ab52642, Abcam) primary antibodies overnight at 4 C. After washing with ice-cold PBS, the sections were incubated with a goat anti-rabbit (1:400; Alexa Fluor 488, ab150077), goat anti-rabbit (1:400; Alexa Fluor 594, ab150080) and goat anti-mouse (1:400; Alexa Fluor 594, ab150116) secondary antibodies for 2 h at 4 °C. The sections were subsequently exposed to DAPI to stain the cell nuclei. Immunofluorescence imaging was performed on a Axio-phot microscope (Carl Zeiss, Jena, Germany) and Olympus IX71 microscope equipped with TH4-200 illumination and DP74 digital camera (Olympus, Tokyo, Japan). Fluorescent signals and morphological characteristics of stained cells (cell body size/cell, dendritic process/cell, or cell number/mm²) were analyzed using ImageJ/FIJI 1.54 (Sholl analysis) and Image-Pro Plus 6.0 software (NIH, Bethesda, MD, USA and Media Cybernetics, Inc. Rockville, USA).

Blood-brain barrier permeability assay

To evaluate BBB penetrability of tomentosin, we used parallel artificial membrane permeability assay (PMBBB-096, BioAssay Systems, Eching, Germany), according to the manufacturer’s recommendations. Briefly, 500 μM of tomentosin and permeability controls (high and low) were dispensed into a donor plate well impregnated with BBB lipid solution. This donor plate was placed on the acceptor plate and then was incubated for 18 h at 37 °C. To minimize own absorbance of tomentosin and permeability controls, their equilibrium standards at 200 μM were prepared separately. The absorbance of acceptor solutions and equilibrium standards were measured using a NanoDrop device (Thermo Fisher Scientific, Madison, WI, USA). Permeability rate (cm/s) was calculated by equation provided and cut-off value of BBB permeability is 4.0 × 10⁻⁶cm/s.

BV2 and HMC3 microglial cell experiments and tomentosin cytotoxicity

To further investigate inhibitory effects of tomentosin on microglia specifically, the mouse BV2 (ATCC, CRL-2468, VA, USA) and human HMC3 (ATCC, CRL-3304, VA, USA) microglial cell line was utilized (Fig. 5A). The cells were maintained at 37 °C under 5% CO₂ and cultured in media (DMEM, LM001-05; WelGENE Inc., Daegu, Korea) supplemented with 10% FBS (S001-01, WelGENE Inc.) and 1% antibiotic-antimycotic solution (LS203-01, WelGENE Inc.).

Fig. 5. Effects of tomentosin on activated BV2 and HMC3 microglia and its byproducts.

The schematic of the experimental schedule was designed A. The cytotoxicity of tomentosin and bay were measured by the WST-8 cell viability assay in the BV2 B. IC₅₀ value of the tomentosin against LPS-induced NO productions was calculated C. Levels of NO, TNF-α, IL-6, and IL-1β under various stimulations (LPS, poly I:C, and IFN-γ) were measured D and E. Under LPS exposure, phagocytic activity was assessed using a FITC-fluorescent phagocytosis assay, and migratory activity was evaluated after 24 h, followed by semi-quantification in BV2 cells F. The cytotoxicity of tomentosin in HMC3 cells was further assessed by the WST-8 assay G. In HMC3 human microglial cells stimulated with two inducers (LPS or IFN- γ), the mRNA expression levels of inflammatory genes (IL1B, NLRP3, GSDMD, and RELA) were evaluated by RT-PCR H. Finally, the BBB permeability rate of tomentosin was assessed I. Data are expressed as the mean ± SD (n = 3 or 6/group). ^#p < 0.05 and ^##p < 0.01 compared to the vehicle-treated cells; ^*p < 0.05 and ^**p < 0^.01 compared to the stimulator-exposed cells.

To confirm whether tomentosin has microglial cytotoxicity, the BV2 and HMC3 cells (2 × 10⁴ cells/well) were seeded into a 96-well microplate for 12 h, and then tomentosin was treated for 26 h or 6 h, respectively. The cytotoxicity was evaluated using a WST-8 assay kit (EZ-Cytox, DoGenBio, Seoul, Korea). The absorbance was measured at 450 nm using a UV spectrophotometer (Molecular Devices, Sunnyvale, CA, USA).

Microglial inflammatory- and inflammasome-associated assay

To obtain half-maximal inhibitory concentration (IC₅₀) value of tomentosin against microglial nitric oxide (NO) production, the BV2 cells (2 × 10⁴ cells/well) were treated with tomentosin (5 to 100 μM) for 2 h before exposure to LPS (1 µg/mL, L2630, Sigma-Aldrich, MO, USA), poly I:C (200 μg/mL, P1530, Merck-Millipore, NJ, USA) or IFN-γ (10 pg/mL, 51-27606E, BD Biosciences, CA, USA). After incubation for 24 h, the supernatants were mixed with an equal volume of Griess reagent (1% sulfanilamide/0.1% N-(1-naphthyl)-ethylenediamine dihydrochloride/2.5% H₃PO₄) to determine levels of NO. After incubation for 15 min at 37 °C, the absorbance was measured at 540 nm using a UV spectrophotometer (Molecular Devices).

The BV2 and HMC3 cells were seeded into 6-well plates (2 × 10⁴ cells/well) and then incubated for 12 h. The cells were treated with tomentosin (20 μM) for 2 h before exposure to LPS (1 µg/mL) for 24 h. In BV2 experiment, cytokine levels of tumor necrosis factor-α (TNF-α, DY140), IL-6 (DY406), and IL-1β (DY400) in the supernatant were determined using an ELISA duo-set kit (R&D Systems, MN, USA). The absorbance was read at 450 nm using a UV spectrophotometer (Molecular Devices). In the HMC3 experiments, the mRNA levels of IL1B, NLRP3, GSDMD, RELA, and GAPDH were analyzed using real-time quantitative PCR (RT-qPCR). Total RNA was extracted using a RNeasy Mini Kit (Qiagen, CA, USA), and cDNA was subsequently synthesized using a High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific, MA, USA). Quantitative PCR was performed using SYBR Green PCR Master Mix (Thermo Fisher Scientific, MA, USA), and PCR amplification was conducted according to a standard protocol on a Rotor-Gene Q real-time PCR system (Qiagen, CA, USA). The detailed primer sequences are provided in Table S3.

Microglial phagocytosis and migration assay

To confirm inhibitory activity of tomentosin against pathological microglial phenotypes, we assessed phagocytic activity using a phagocytosis assay kit (500290, Cayman Chemical, MI, USA). Under the same conditions as for the cytokine assay, the BV2 cells were exposed to latex beads (1:200) conjugated with rabbit IgG-fluorescein 5-isothiocyanate (FITC) for 30 min. After fixation with 4% PFA, the phagocytosed fluorescent beads were counted using a microscope.

For assessing migration ability, the cells were wounded using a sterile 20 μL pipette tip. After incubation for 24 h, the migratory activity was determined by measuring the relative changes in the width of the wounds using a microscope. The degree of cell migration is expressed as the percentage of that of vehicle-treated cells.

Western blot analysis of microglial cell lysates and ACC homogenates

Protein expression of BV2 microglial cell lysates for the inflammatory (TLR4; toll-like receptor 4, iNOS; inducible nitric oxide synthase, NF-κB; nuclear factor kappa, ERK; extracellular signal-regulated kinases, and STAT3; signal transducer and activator of transcription 3) and inflammasome (NLRP3; NLR family pyrin domain containing 3, ASC; apoptosis-associated speck-like protein, caspase-1, GSDMD, and IL-1β) signaling pathway were assessed. Microglial activity (Iba-1), neuronal activity (NeuN, c-Fos, BDNF; brain-derived neurotrophic factor, and PSD95; postsynaptic density protein 95), and programmed cell death (pro- and cleaved caspase 1 or 3) in mouse ACC tissue was evaluated by western blotting analysis.

In brief, the harvested cells were lysed using a protein extraction solution (Pro-Prep, iNtRON Biotechnology, Sungnam, Korea) and removed ACC tissues were homogenized in radioimmunoprecipitation assay (RIPA) buffer (R0278, Sigma, MO, USA), supplemented with protease and phosphatase inhibitor cocktails (#1861284, Thermo Scientific, MA, USA). After equalizing the protein concentrations using a bicinchoninic acid protein assay kit (BCA1 and B9643, Sigma-Aldrich, MO, USA), the lysates and homogenates were separated by 10% polyacrylamide gel electrophoresis and then transferred to polyvinylidene fluoride (PVDF) membranes. To minimize non-specific binding, the membranes were blocked in 5% bovine serum albumin (BSA) for 1 h. The membranes were incubated overnight at 4 °C with primary antibodies, including antibodies against the anti-mouse NeuN (1:1000, MAB377, Merck-Millipore), anti-mouse c-Fos (1:1000, ab208942, Abcam), anti-rabbit BDNF (1:1000, ab108319, Abcam), anti-mouse PSD95 (1:1000, ab13552, Abcam), anti-rabbit NLRP3 (1:1000, ab214185, Abcam), anti-mouse apoptosis-associated speck-like protein (ASC, 1:1000, #67824, Cell Signaling), anti-rabbit caspase-1 (1:1000, NBP1-45433, Novus), anti-rabbit GSDMD (1:1000, #46451, Cell Signaling), anti-rabbit IL-1β (1:1000, ab9722, Abcam), anti-rabbit caspase-3 (1:1000, #9662, Cell Signaling), anti-rabbit cleaved caspase-3 (1:1000, #9664, Cell Signaling), anti-rabbit TLR4 (1:1000, #14358, Cell Signaling), anti-rabbit iNOS (1:1000, PA1-036, Thermo Fisher), anti-rabbit NF-κB (1:1000, #8242, Cell Signaling), anti-rabbit phospho-NF-κB (1:1000, ab86299, Abcam), anti-rabbit ERK (1:1000, #9102, Cell Signaling), anti-rabbit phospho-ERK (1:1000, #9101, Cell Signaling), anti-mouse STAT3 (1:1000, MA1-13042, Thermo Fisher). anti-mouse phospho-STAT3 (1:1000, MA5-15193, Thermo Fisher) and anti-mouse α-tubulin (1:2000, ab7291, Abcam). The membranes were incubated with an HRP-conjugated anti-rabbit or anti-mouse antibody (GeneTex, Inc., Irvine, CA) for 1 h. The bands were visualized with an advanced enhanced chemiluminescence (ECL) kit. The intensity was analyzed by ImageJ version 1.46 (NIH, Bethesda, MD, USA).

Molecular docking analysis associated with inflammasome-related proteins

Molecular docking simulations were performed to identify potential targets and validate specific binding modes through a multi-stage approach. The 3D structures of target proteins and ligand compounds, including tomentosin (PubChem CID: 155173), were retrieved from the RCSB PDB (Table S2), AlphaFold Protein Structure Database, or PubChem database, respectively. Prior to analysis, all ligand structures were energy-minimized and converted to PDBQT format using OpenBabel. Blind docking was initially performed on multiple candidate proteins using both the CB-Dock2 server and local AutoDock Vina software.

Subsequently, site-specific docking was conducted to elucidate the specific inhibitory mechanism against GSDMD using AutoDock Vina with the AlphaFold-predicted GSDMD structure (Chain A). For this specific analysis, a grid box was explicitly defined around the Cys191 residue and the known inhibitory pocket with an exhaustive parameter of 64. The top-scoring binding poses were identified based on binding affinity, and molecular interactions were visualized using PyMOL 3.0.4 and LigPlus software.

Overexpressed GSDMD vector construction and cell transfection

To verify the working mechanism of tomentosin, the clonal gene (pCMV-Puro-Gsdmd, insert length: 1464 bp) for GSDMD overexpression (NM_026960) was constructed using a Twist vector (Twist Bioscience, CA, USA).

The BV2 microglial cells were seeded at 2 × 10⁴ cells/well into the 6 well plates and then incubated until grown to 70 ~ 80% confluency. The cells were transfected with 1 μg of plasmid DNA using Lipofectamine 3000 (L3000015, Thermo Fisher Scientific, MA, USA) diluted in Opti-MEM reduced serum medium (31985-062, Gibco, MA, USA) for 24 h. Puromycin (2 μg/mL) was treated to select transfected cell for 24 h. Underwent verification whether transfected using Western blotting analysis, the GSDMD overexpressed BV2 microglial cells were seeded at 2 × 10⁴ cells/well into the 6 well plates, and after 12 h, the LPS (1 μg/mL) and/or tomentosin (20 μM) were treated for 24 h to evaluate the inhibitory activity against GSDMD protein expression.

Comparative with LDC7559 inhibitor and chemical stability evaluation

To compare the inhibitory activity of tomentosin with LDC7559 (GSDMD inhibitor, HY-111674, MCE NJ, USA) against activated microglia, cells were treated with tomentosin (20 μM) or LDC7559 (5, 10 and 20 μM) for 2 h before exposure to LPS. In addition, to evaluate the chemical stability, the tomentosin was on heating at 40 and 45 °C or acidification with HCl at pH 3.0, and then cells were treated with heated or acidified tomentosin (20 μM) for 2 h before exposure to LPS. After exposing 24 h to LPS, cell supernatants were mixed with Griess reagent to measure NO levels, and absorbance was measured at 540 nm using a UV spectrophotometer. Under the same conditions without LPS stimulation, the cytotoxicity of above was evaluated using a WST-8 assay kit. The absorbance was measured at 450 nm using a UV spectrophotometer (Molecular Devices).

Statistical analysis

All results are expressed as the mean ± standard deviation (SD). Statistical significance was analyzed by unpaired Student’s t-test or one-way analysis of variance (ANOVA) followed by post hoc analysis by Dunnet t-test using Prism 7 software (GraphPad). Differences at p < 0.05 were considered significant.

Results

Network pharmacology-based identification of tomentosin

To discover novel terpenoids with potential therapeutic effects against neuroinflammation-driven MDD, we performed a network pharmacology analysis. Out of 173,622 terpenoids surveyed, their interactions with target molecules were analyzed using predictive tools. In addition to evaluating their CNS-pharmacological applicability, we computed their drug-likeness (based on QED scores incorporating molecular weight, solubility, hydrogen bond of acceptors/donors, polar surface area, rotatable bonds, and aromatic ring) and BBB permeability. A total of 615 candidates were identified that met all three of the following criteria: (1) molecular target diversity (>2), (2) drug-likeness (>0.3), (3) BBB penetrability (“pass”) (Fig. 1A).

Next, we performed a multiscale interactome analysis, calculating correlation scores (0–1) by diffusion profiles for each terpenoid target and comparing them with MDD-associated neuroinflammatory proteins (Fig. 1B). Among these, the top 50 candidates were displayed in circos plot, with the largest proportion of sesquiterpenoids (32%) (Fig. 1C). Next, the candidates were prioritized by three exclusion criteria: (1) low correlation score, (2) known antidepressant effects, and (3) unavailable commercially. Then, the top 10 candidates were evaluated by in vitro screening that considered IC₅₀ against NO production in LPS-stimulated microglia, and non-cytotoxicity. Finally, we identified that tomentosin has a favorable QED (0.438), strong network correlation (score = 0.03), and low IC₅₀ (16.1 μM) value, making it a compelling candidate for further investigation.

Tomentosin alleviates FRD pathophysiology

Mice exposed to UCMS for 4 weeks exhibited predominant MDD-like behaviors compared to normal group, as evidenced by significant reductions in activity duration of TST (p < 0.01) and FST (p < 0.05). Among these UCMS-subjected 11 mice, six mice showed positive responses to treatment of antidepressants (fluoxetine, 10 mg/kg) from their behavioral tests, meanwhile other 5 mice didn’t respond to fluoxetine treatment (referred to FRD group) (Figs. 2B, C and S2A, B). However, when FRD mice were administered with tomentosin (20 mg/kg) for a week, their depression-like behaviors were significantly alleviated, as level of fluoxetine-responding mice (p < 0.05 or p < 0.01; Fig. 2D, E).

Regarding pathological alterations, UCMS induced a significant increase in Iba-1-positive signals (a marker of microglia) (p < 0.01; Figs. 2F and S3A) and microglial soma size (p < 0.01; Figs. 2G to I and S3E, F), alongside an elevated number of GFAP/S100β-positive reactive astrocytes (p < 0.05; Fig. S4A, B). These changes were associated with a significant decline in neuronal integrity within the ACC, evidenced by a reduction in NeuN-positive neurons (p < 0.05; Figs. 2F and S3A) and the expression of neuroactive proteins, including BDNF, c-Fos, and PSD95 (p < 0.01; Fig. S1D and F). Furthermore, a marked elevation in serum C-reactive protein (CRP) levels (p < 0.01; Fig. S1C) was observed in the UCMS group. While both fluoxetine and tomentosin successfully reversed UCMS-induced neuronal dysfunctions (p < 0.05 or p < 0.01 for all alteration), only tomentosin exhibited a systemic anti-inflammatory effect by attenuating serum CRP levels (p < 0.01). Notably, tomentosin did not affect the reactive astrocytes.

Tomentosin inhibits microgliopathy-related depressive behaviors

Given that microgliopathy is an emerging pathophysiological hypothesis of TRD, we adapted inflammation-mediated MDD mouse models. Mice exposed to systemic LPS (1 mg/kg) injections exhibited depressive- and anxiety-like behaviors compared to saline-injected controls, as evidenced by significant reductions in nest-building scores (NBT; p < 0.05), time spent in the central area (OFT; p < 0.05), and active duration (FST; p < 0.05). These behaviors were considerably improved by administration of tomentosin compared to LPS-injected mice (p < 0.05 or p < 0.01; Figs. 3B to D and S2C, D). Notably, while the antidepressant and anti-neuroinflammatory effects of tomentosin were robust in wild-type mice (p < 0.05 or p < 0.01), these effects were completely abolished in LPS-challenged Casp1 KO mice (Figs. 3G to I and S2E, F).

To further evaluate the inhibitory effects of tomentosin under brain-specific neuroinflammatory conditions, mice were subjected to intracranial injection of rIL-1β (10 ng/head). Tomentosin treatment significantly improved depressive behaviors in four tests (NBT, MBT, TST, and FST) compared to aCSF-injected mice, except OFT (p < 0.05 or p < 0.01; Figs. 4B to F and S2G to I). Overall, the effect size of tomentosin was generally superior to that of BAY11-7082 (5 mg/kg).

Tomentosin suppresses microglial-derived neuronal pyroptosis in the ACC

We further examined the selective inhibitory activity of tomentosin on microglial-mediated inflammasome and neuronal pyroptosis in the ACC. Systemic LPS challenge significantly enhanced microglial inflammasome signaling, as evidenced by increased Iba-1/IL-1β co-localization (p < 0.01; Fig. S4C and D), Iba-1 protein expression (p < 0.05; Fig. S5A and B), and Iba-1/GSDMD double-positive signals (p < 0.01; Figs. 3E and S3B) compared to saline-injected mice. Sequentially, neuronal pyroptosis (NeuN-positive neurons around GSDMD, p < 0.01; Figs. 3E and S3B) and neuronal dysfunction (NeuN, BDNF, c-Fos, and PSD95 protein expressions, p < 0.05 or p < 0.01; Fig. S5A, B) were observed. While tomentosin significantly attenuated these pathological alterations (p <p < 0.05 or p < 0.01 for all signals), its protective effects were completely abolished in Casp1 KO mice (Figs. 3J and S3C, 4E).

Correspondingly, mice subjected to intracranial rIL-1β injection exhibited microglial GSDMD pore formation, as evidenced by increased Iba-1/IL-1β (p < 0.01; Fig. S4F and G) and Iba-1/GSDMD (p < 0.01; Figs. 4G and S3D) positive signaling in the ACC. In parallel, neuronal pyroptosis, characterized by NeuN/GSDMD double-positive signals (p < 0.01; Figs. 4G and S3D), and reduced expression of neuroactive proteins (p < 0.01; Figs. 4H and S5E) were observed in the ACC. Tomentosin blocked microglial-derived pyroptosis compared to aCSF-injected mice (p < 0.05 or p < 0.01 for all signals). Specifically, along with FRD model (Fig. S1D to F), tomentosin inhibited the cleavage of caspase-1 in the ACC of both MDD models (Fig. S5A to D). however, this was not observed for caspase-3.

Tomentosin mitigates activated phenotype of microglia and their byproducts

To evaluate whether tomentosin effectively suppress stimulus-evoked microglial inflammatory responses, the mouse BV2 microglial cells were applied. As expected, the IC₅₀ value of tomentosin for the LPS-induced NO levels was 16.09 μM in this cell line, denoting that effective and non-cytotoxic dose of tomentosin is around 20 μM (Fig. 5B and C). As a marker of activated microglia, NO production was notably increased by non-cytotoxic dose of three inducers (Fig. S6A), such as LPS (approximately 4.1-fold), poly I:C (approximately 2-fold), and IFN-γ (approximately 2.9-fold), whereas these were significantly inhibited by tomentosin treatment (p < 0.01 for each; Fig. 5D). These microglial inhibitory effects were concordantly supported by suppressing production of inflammatory cytokines (TNF-α, IL-6, and IL-1β; p < 0.05 or p < 0.01; Fig. 5E). Furthermore, tomentosin strongly reduced both FITC-bead phagocytosis and migration into scratched areas (p < 0.01 for both; Figs. 5F and S6B, C).

For cross-validation, the activity of tomentosin was further assessed in HMC3 human microglial cells. Non-cytotoxic doses of tomentosin (Fig. 5G) significantly suppressed the upregulation of inflammasome-related genes (IL1B, NLRP3, and GSDMD) induced by LPS or IFN-γ (p < 0.05 or p < 0.01; Figs. 5H and S7D), although no significant effect was observed for RELA (p65).

We confirmed that tomentosin had a high BBB permeability rate (approximately 4 times cut-off levels) which was compared to equal dose of permeability controls (Fig. 5I).

Tomentosin normalizes molecular alterations of activated microglia

To figure out the underlying mechanisms of tomentosin, we investigated microglial neuroinflammation-related pathways. LPS exposure led to sequential activation of molecular cascades involved in NF-κB translocation and the NLRP3 inflammasome, while they were substantially alleviated by tomentosin treatment (p < 0.05 or p < 0.01; Fig. 6A). Tomentosin primarily targeted the NLRP3/caspase-1/GSDMD pathway rather than the TLR4/NF-κB pathway, as indicated by the protein expression patterns (Fig. S7A to C).

Fig. 6. Effects of tomentosin on activated microglial-derived molecular alterations.

Levels of protein-associated with NLRP3/caspase-1/GSDMD inflammasome pathway in LPS-stimulated BV2 microglia were examined by Western blotting analysis A. Binding affinity, binding sites, and amino acid residues of both mouse and human GSDMD to tomentosin were analyzed by Vina binding energy score B. The pTwist-CMV-Puro-Gsdmd vector map was displayed C. Under conditions of GSDMD overexpression and LPS stimulation, the inhibitory effect of tomentosin on GSDMD cleavages was assessed D. These inhibitory effects were compared to LDC7559 (a potent GSDMD inhibitor) E. Chemical and effective stability of tomentosin was assessed under various environmental conditions (e.g., heating up to 45 °C and acidic pH 3.0) F. Data are expressed as the mean ± SD (n = 5 or 6/group). ^##p < 0.01 compared to the vehicle-treated cells; ^**p < 0.01 compared to the LPS-treated cells.

From the Gene Ontology and KEGG pathway enrichment analyses of tomentosin’s network targets, we identified major pathways related to inflammation and cell survival/apoptosis (Fig. S1A and B).

Tomentosin exerts high binding interaction with GSDMD and is competitive to inhibitor

To verify tomentosin’s interaction with GSDMD, molecular docking analysis was performed using AutoDock Vina, which assesses binding affinities with multiple target proteins. Among the proteins tested, tomentosin exhibited the highest binding affinity for GSDMD (Mus musculus: −7.6 kcal/mol and Homo sapiens: −7.7 kcal/mol) surpassing other potential targets such as NLRP3 (−7.0 kcal/mol), IL-1β (−6.2 kcal/mol), caspase-1 (−5.6 kcal/mol), and STAT3 (−6.6 kcal/mol). The docking model revealed that tomentosin forms stable hydrophobic contacts within a specific binding pocket of GSDMD, a finding consistently supported by independent local docking simulations (Fig. 6B and Table S2). Moreover, comparative docking analysis further demonstrated that tomentosin occupies the same binding pocket as disulfiram, specifically targeting the reactive Cys191 residue (Fig. S8).

We additionally validated the microglial GSDMD-specific inhibition of tomentosin in intensified conditions of GSDMD expression. As expected, tomentosin significantly inhibited GSDMD cleavage in LPS-stimulated microglial cells underwent transfection of GSDMD overexpression vector (p < 0.01; Figs. 6C, D and S6D). Notably, when compared to a known GSDMD inhibitor (LDC7559), tomentosin showed an even stronger inhibition of microglial NO production (Figs. 6E and S6E), indicating competitive efficacy.

Regarding chemical stability, tomentosin maintained its microglial inhibitory effects even after heating to 45 °C or exposure to acidic conditions (pH 3.0), further supporting its potential as a robust therapeutic candidate (p < 0.01 for all; Figs. 6F and S6F).

Discussion

Current therapeutic strategies for managing MDD remain controversial, particularly given that up to half of patients do not respond to first-line antidepressants enhancing serotonergic transmission [29]. Consistent with this high non-response rate to selective serotonin reuptake inhibitors (SSRIs), our current UCMS-induced MDD model also exhibited a subset of mice unresponsive to fluoxetine, categorized as fluoxetine-resistant depression (FRD) (Fig. 2A to C). Despite the widespread prescription of SSRIs, the continuously increasing prevalence of MDD has led to growing skepticism toward the serotonin-targeted treatments [30, 31]. Furthermore, a prospective positron emission tomography (PET) study reported no significant difference in 5-HT transporter binding between patients with TRD and those with non-TRD, suggesting that serotonergic dysfunction may not be the primary mechanism underlying TRD [32].

Accumulating data have proposed that microglia-driven neuroinflammation are key contributors to the pathophysiology of TRD [33–37]. Although previous clinical trials using anti-inflammatory agents (e.g., minocycline, infliximab) demonstrated limited overall efficacy, modest benefits were predominantly observed in TRD patients with elevated serum c-reactive protein (CRP) levels (>3 or 5 mg/L) [16, 17]. This suggests that targeting brain-specific neuroinflammation, particularly microglial activation, may offer a more effective therapeutic strategy. In this study, high serum CRP levels and depressive behaviors were observed in FRD model, and it was dramatically normalized by tomentosin (Fig. S1C). Microglia are essential for maintaining neuronal homeostasis; however, they polarize into activated phenotypes in pathological conditions [38, 39]. This ‘microgliopathy’, commonly links to neurotoxicity leading to functional and/or structural impairment, especially in regions of ACC and PFC [40, 41]. As expected, tomentosin treatment (20 mg/kg) notably attenuated microglial morphological changes in soma size and neuronal shrinkage (evidenced by reduced cell numbers and neuroactive proteins) in the ACC of FRD mice (Figs. 2D to I and S1), but it had no effect on reactive astrocytes (Fig. S4A and B). These findings are particularly relevant as TRD patients exhibit more severe microglial activation and immunological dysregulation in the PFC-ACC region than non-TRD patients [10, 42, 43].

Both ACC and PFC have been shown to exhibit heightened sensitivity to peripheral inflammatory signals—as reflected by elevated serum CRP levels—owing to their relatively weak BBB and elevated expression of receptors for inflammatory cytokines, including IL-1 and TNF-α [44, 45]. We initially employed a systemic LPS-induced inflammation model (Fig. 3A), in which tomentosin reduced microglia-associated neuroinflammation in the ACC and improved depressive-like behaviors (Fig. 3B to E), independent of sickness behavior (Fig. S2A to I). A recent comprehensive review integrating neuroimaging features of TRD indicated the reduced fronto-cingulate connectivity [46]. Particularly, the prefrontal regions are known to be areas where microglia are easily activated by chronic stress and inflammatory conditions [47, 48]. Based on the caspase-1-specific inhibitory activity of tomentosin (Fig. S1D and F), we validated its pharmacological specificity for pyroptosis using a Casp1 KO mouse model (Figs. 3F to J and S3C) and intracranial injection of rIL-1β (Figs. 4B to H and S3D). The ACC plays a central role in emotion regulation and self-referential processing, and its functional-structural deficits are commonly observed in suicidal individuals [49–51]. Postmortem evidence indicates that high microglial priming and inflammasome activation in the ACC contribute to suicidality in depressed individuals [41, 52]. These may be relevant to explain the increased suicide risk in TRD [53, 54], although a cause-and-effect relationship between microglia and suicidality has not yet been established. Additionally, in vitro assays demonstrated that tomentosin remarkably suppressed highly activated microglia (BV-2 and HMC3 cells) as well as their inflammatory and inflammasome-related molecular alterations (Fig. 5B to H).

Building on the above therapeutic potential of tomentosin, we further explored its molecular mechanisms, focusing on inflammasome-related microglial dysfunction. Recently, microglial NLRP3 inflammasome and its downstream pyroptotic signaling have been found as key mediators in depression [55–57]. A clinical study found that TRD patients and responsive patients could be clearly separated by differential expression of NLRP3-related genes in peripheral blood transcriptomic analyses [58]. Notably, antidepressant efficacy has been observed only in microglia-specific, not astrocyte-specific, Nlrp3 knockout mice [59]. Typically, microglial NLRP3 inflammasome activation induces caspase-1-dependent GSDMD cleavage, resulting in microglial pyroptosis and the release of pro-inflammatory cytokines that exacerbate neuronal damage [14, 60]. As expected, tomentosin treatment suppressed GSDMD cleavage and microglia-derived IL-1β in the ACC across all inflammasome-related animal models, as shown by immunofluorescence staining (Figs. 3E, J, 4G and S4C to G), and further supported by reduced levels of NLRP3/caspase-1/GSDMD signaling-related genes and proteins (Figs. 5H and 6A). Molecular docking analysis revealed tomentosin’s strong binding affinity to GSDMD primarily via hydrophobic interactions (Fig. 6B), which are likely to interfere with its pore-forming N-terminal domain [61, 62]. Notably, tomentosin occupies the same binding pocket as the specific GSDMD inhibitor disulfiram and exhibits similar binding affinity by interacting specifically with the reactive Cys191 residue within the GSDMD N-terminal domain (Fig. S8). These interactions were validated under GSDMD-overexpressing conditions, where tomentosin exhibited superior inhibitory efficacy compared to a known GSDMD inhibitor (Fig. 6C to E). Importantly, tomentosin also exhibited strong thermal and acidic stability, further enhancing its pharmacological potential (Fig. 6F). A comprehensive summary of the proposed mechanism is illustrated in Fig. S9.

To find a novel candidate compound overcoming TRD, we applied a computational network-based pharmacology system, thus we could identify tomentosin having the highest correlation score targeted microglial pathology (Fig. 1). We initially focused on terpenoids due to their higher BBB penetrability compared to flavonoids and alkaloids [19], along with their structurally rich hydrocarbon chains bearing abundant bioactive functional groups [63]. Many essential oils comprising terpenoids are known to improve depressive/anxious moods [64], and terpenoids are considered promising antidepressant candidates because of their strong inhibitory properties against microglial-derived neuroinflammation [65]. Of note, terpenoids preferentially inhibit pathological microglial phenotypes without affecting the physiological function of resting microglia [66]. D-limonene, a monoterpene, evidenced the anti-anxiety effects by a phase 2 clinical trial (NCT06378957) [67]. Despite these powerful properties of terpenoids, discovering an optimal drug candidate remains challenging due to its largest diversity to be 5 times more than that of polyphenols and alkaloids [68, 69].

Although numerous natural compounds have been proposed as antidepressant candidates, few have been specifically investigated for use in TRD [70]. In CNS drug development, along with effect size, brain bioavailability and safety (i.e., non-toxicity) are equally important. Esketamine was approved in 2019 for TRD [71]; however, its clinical utility is limited by severe side effects such as dissociation, which affects over 40% of patients [72]. While minocycline is known to inhibit the microglial NLRP3 inflammasome, it has been criticized for exerting non-specific glial-suppressive effects even under physiological conditions [73, 74]. Tomentosin is abundant in the leaves of Inula viscosa and Inula japonica, with a high isolated yield (0.64%) [22]. Previous studies have reported its neuroprotective effects on animal models of cerebral ischemia and neuro-excitotoxicity [23, 24]. We confirmed that tomentosin has high BBB penetrability and does not interfere with the physiological function of resting microglia (Fig. 5D and I). In this study, BAY 11-7082 was used as a positive control because it inhibits IκBα phosphorylation and suppresses NF-κB-dependent inflammasome priming, thereby serving as an upstream comparator. Tomentosin demonstrated superior overall efficacy compared with this positive control (Figs. 3 and 4).

Our results suggest that tomentosin exerts a selective effect on microglia-derived inflammasome, making it a promising candidate for TRD. Notably, our AI-driven approach combined with experimental validation demonstrated both time efficiency and predictive accuracy in addressing unmet clinical needs for TRD drug discovery. Although this is the first study to explore the antidepressant potential of tomentosin specifically in the context of FRD, several limitations should be acknowledged: (1) serotonergic modulation was not evaluated; (2) long-term safety and pharmacokinetic profiles remain to be determined; (3) potential interactions with conventional antidepressants have not been explored.

In conclusion, our study provides strong preclinical evidence that tomentosin is a promising candidate for FRD, primarily through selective inhibition of pathological microglia in the ACC via suppression of NLRP3/caspase-1/GSDMD signaling. These findings support further translational efforts, with tomentosin positioned as a first-in-class compound. Future studies should explore its serotonergic activity, long-term safety, and efficacy in diverse TRD models. Moreover, comprehensive validation will require time-resolved molecular analyses, dose–response characterization, larger animal sample sizes, anhedonia-focused behavioral assessments, and the inclusion of female subjects to improve translational applicability.

Author contributions

Jin-Seok Lee and Ji-Yun Kang contributed to the conceptualization, data curation, investigation, visualization, formal analysis, and writing – original draft. Ji-Yeon Gu and Tae-Wook Woo contributed to support the data curation and investigation. Won-Yung Lee contributed to the molecular docking analysis. Chang-Gue Son was responsible for supervision and contributed writing – review & editing.

Funding

This research was supported by National Research Foundation of Korea (NRF) grants funded by the Ministry of Science, ICT & Future Planning (NRF-2022R1A2C1013084 and NRF-2018R1A6A1A03025221).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interest.

Ethics approval

All animal care and experimental protocols were approved by the Institutional Animal Care and Use Committee of Daejeon University (Approval number: DJUARB2024-019). This study involved only animal subjects and did not include human participants; therefore, informed consent to participate is not applicable. This study does not contain any identifiable images or data from human participants; therefore, informed consent for publication is not applicable.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-04092-5.