Genome-directed discovery of an octalin-containing polyether ionophore improves diabetic skin wound healing

Abstract

Chronic hard-to-heal skin wounds in diabetes mellitus pose a significant clinical challenge, underscoring the urgent need for additional therapeutics. Herein, we report the identification of a polyether biosynthetic gene cluster (pdm BGC) from the marine-derived Streptomyces marincola SCSIO 03032 through genome mining. By overexpressing the positive regulator gene pdmRI, we increase the product of the pdm BGC 48-fold to enable the isolation, structural determination, and biosynthesis investigation of an octalin-containing polyether ionophore polydecalinmycin (PDM). Functionally, PDM demonstrates remarkable efficacy in promoting acute skin wound healing in BALB/c mice and significantly accelerating chronic wound closure in a diabetic mouse (db/db, male) model. Mechanistically, PDM induces an elevation of mitochondrial Na⁺ levels to trigger a transient burst of reactive oxygen species, a critical signal for initiating acute wound closure. Upon prolonged exposure, PDM activates a robust Nrf2-mediated antioxidant response, restoring mitochondrial homeostasis and conferring cytoprotection against high-glucose-induced oxidative stress. These insights highlight the therapeutic potential of PDM as a natural polyether candidate for developing ionophore-based strategies for diabetic wound healing.

Chronic skin wounds in diabetes mellitus pose a clinical challenge that requires additional therapeutics. Here the authors report a polyether biosynthetic gene cluster, engineer a bioproduction platform for the ionophore polydecalinmycin, and demonstrate its wound healing abilities.

Introduction

As the external epithelial barrier, the skin endures and repairs injuries throughout life. This vital role is hindered by diverse factors that affect skin healing and integrity. Chronic hard-to-heal skin wounds represent a major complication of diabetes, affecting approximately 2.5% of the global population^1,2. While wounding triggers an orchestrated healing response ascribed to restorative growth, the regeneration competence progress of the injured tissue is often delayed in diabetes, leading to hard-to-heal or non-healing wounds³. For diabetic chronic wounds, contemporary therapeutic strategies primarily focus on systemic glycemic control, microvascular improvement, and infection management⁴. Consensus is lacking on optimal strategies for diabetic wound treatment using the limited wound healing products that are moderately effective in the clinic⁵, highlighting the unmet medical needs for innovative therapies and medicines.

Natural products have historically been vital sources for new drug development, with over 64% of U.S. Food and Drug Administration (FDA)-approved small-molecule drugs derived from them⁶. Among these, polyether ionophores are a group of lipophilic polyketides featuring an oxygen-rich polycyclic skeleton composed of tetrahydrofuran and tetrahydropyran rings⁷ (Supplementary Fig. 1). These structural properties allow polyethers to transport monovalent and divalent cations freely across lipid membranes, thereby disturbing intracellular cation homeostasis⁸. Consequently, natural polyether ionophores exhibit diverse pharmacological activities, including antiparasitic, antibacterial, antifungal, and antitumor^7,9. Notably, several polyether ionophores, such as lasalocid, monensin, and salinomycin, have been approved by the FDA for use in veterinary practice for decades¹⁰. Beyond their established anticoccidial indications, these commercial polyether ionophores have been widely used as growth promoters in livestock production, particularly as feed additives for ruminants and poultry¹¹.

High rediscovery rates, complex structural identification and synthesis, and low native production yields have limited the discovery and pharmaceutical development of novel natural polyether ionophores despite their desirable drug-like properties⁷. Recent advancements in bioinformatics, genome mining, and bioengineering for unlocking cryptic biosynthetic gene clusters (BGCs) offer crucial opportunities to address these challenges¹². By employing a genome mining approach, we characterize a polyether (pdm) BGC from the deep-sea-derived Streptomyces marincola SCSIO 03032¹³, leading to the discovery of an octalin-containing polyether metabolite, polydecalinmycin (PDM), through the regulator overexpression of the pdm BGC. Our bioactivity studies demonstrate that PDM potentiates productive repair programs to improve skin wound healing. Mechanistic insights reveal that PDM induces moderate mitochondrial Na⁺ leaks. This mode of action initially promotes controlled superoxide production in keratinocytes, which subsequently activates an Nrf2-mediated antioxidant response, thereby restoring mitochondrial homeostasis and protecting repair cells against high-glucose-induced oxidative stress. Our findings have the potential to pave the way for the development of PDM-based innovative therapy for skin wound healing.

Results

Genome mining reveals an uncharacterized polyether BGC

Biosynthetic studies have demonstrated that polyether ionophores are biosynthesized from a type I polyketide synthase (PKS) system¹⁴. Biochemical and crystallographic experiments have elucidated that their typical polyether rings (tetrahydrofuran and tetrahydropyran) are formed by epoxidase and epoxide hydrolase^15,16. To discover new polyether ionophores, we used type I PKS proteins (PF00698 and PF08659), epoxidase Lsd18, and epoxide hydrolase Lsd19 as queries, and searched for neighboring genes within ±15 loci from genomes in the NCBI database. A total of 250 type I PKS BGCs were identified across 239 actinomycetal genomes, and a gene cluster similarity network was constructed based on the Jaccard Index (Fig. 1a). Notably, 164 polyether BGCs, including 97 nigericin (nig), 20 salinomycin (sln), 8 tetronomycin (tmn), 7 monensin (mon), 6 lasalocid (lsd), 6 laidlomycin (lad), 3 nanchangmycin (nan), 3 ionostatin (ion), 2 tetromadurin (tem), 1 maduramicin (mad), 1 tetronasin (tsn), and 10 unknown BGCs, were found to be gathered in the largest cluster (Fig. 1a, and Supplementary Fig. 2). AntiSMASH¹⁷ analysis of the 10 unknown BGCs identified two uncharacterized polyether BGCs in S. marincola SCSIO 64649¹⁸ and S. marincola SCSIO 03032¹³ (Fig. 1a), which are characterized by large and consecutive PKS modules, indicating their potential to biosynthesize structurally more complex polyethers (Supplementary Fig. 3). As the two polyether BGCs exhibit a high degree of sequence homology, we selected the pdm BGC (Fig. 1b) from the deep-sea sediment (−3412 m) derived S. marincola SCSIO 03032¹³, a productive strain reported to produce piericidins, heronamides, and spiroindimicins¹³, for further investigation.

Fig. 1. Genome mining of the pdm BGC.

a Gene cluster similarity network of 250 type I PKS BGCs based on the Jaccard Index. The 164 polyether BGCs are labeled with different colors. The pdm BGCs from S. marincola SCSIO 03032 and S. marincola SCSIO 64649 are highlighted in enlarged nodes. b Genetic organization of the pdm BGC.

The pdm BGC encodes five type I PKSs (PdmAI−PdmAV), an epoxidase (PdmC), two epoxide hydrolases (PdmBI and PdmBII), an oxidoreductase (PdmO), two regulators (PdmRI and PdmRII), and two transporters (PdmTI and PdmTII) (Fig. 1b). PdmAI−PdmAV contain 18 type I PKS modules, which are divergent from those in previously reported polyether BGCs (Supplementary Fig. 2). PdmC, PdmBI, and PdmBII show high similarities to Lsd18 (69%), Lsd19A (57%), and Lsd19B (66%) in the biosynthesis pathway of lasalocid¹⁹, respectively (Supplementary Table 1). These three enzymes have been well characterized as being responsible for the formation of two polyether rings in lasalocid¹⁹. The oxidoreductase (PdmO) displays 71% similarity to InoJ, a proposed dehydrogenase in the biosynthesis of ionomycin²⁰ (Supplementary Table 1). Therefore, the pdm BGC was putatively involved in the biosynthesis of an unusual polyether.

Heterologous expression of the pdm BGC leads to a polyether product

Initially, we examined the fermentation broth of S. marincola SCSIO 03032 under the regular culture condition for piericidins, heronamides, and spiroindimicins¹³. However, no polyether-like compounds were observed in HPLC analysis (Supplementary Fig. 4). To explore the product of the pdm BGC, we employed a heterologous expression strategy. A pMSBBAC2-based²¹ BAC library of S. marincola SCSIO 03032 was constructed, and screening of the library afforded the plasmid pCSG8135 covering the entire pdm BGC (Supplementary Fig. 5). Introduction of pCSG8135 into the heterologous host Streptomyces albus Del14²² led to a product (1), which was absent in the control strain S. albus Del14/pSET152 (Fig. 2a). However, the characterization of 1 from S. albus Del14/pCSG8135 failed due to its extremely low yield.

Fig. 2. Heterologous expression of the pdm BGC and engineering for the high-yield production of PDM.

a HPLC analysis of metabolite profiles of S. albus Del14 strains harboring (i) pSET152 (plasmid vector), (ii) pCSG8135 (pdm BGC), (iii) pCSG8136 (Δorf(−1)), (iv) pCSG8137 (ΔpdmRI), (v) pCSG8138 (ΔpdmRII), (vi) pCSG8139 (ΔpdmO), (vii) pCSG8140 (Δorf1), (viii) pCSG8135 and pCSG8141, and (ix) pCSG8135 and pCSG8142. b, c HPLC analysis of metabolite profiles and the yield of PDM from S. marincola SCSIO 03032 (WT), S. marincola SCSIO 03032/pCSG8143 (WT::pCSG8143), and S. marincola SCSIO 03032/pCSG8144 (WT::pCSG8144). Data are represented as mean ± SD (n = 3 independent experiments). ^****P < 0.0001. d Chemical structure of PDM. Source data are provided as a Source data file.

We then compared the pdm BGC with other polyether BGCs (Supplementary Table 1) and proposed that the upstream and downstream boundaries of pdm BGC are pdmRI and pdmO, respectively. To prove this hypothesis, gene inactivation experiments of orf(−1), pdmRI, pdmO, and orf1 were performed in S. albus Del14/pCSG8135, respectively (Supplementary Fig. 6). As expected, the production of 1 was unaffected in the Δorf(−1) and Δorf1 mutants, while the production of 1 was completely abolished in the ΔpdmRI and ΔpdmO mutants (Fig. 2a). Thus, the pdm BGC presumably contains thirteen orfs from pdmRI to pdmO (Fig. 1b).

Enhanced production of the polyether facilitates its structural elucidation

Genetic manipulations of regulators are effective strategies to increase the production of metabolites²³. Two regulators (PdmRI and PdmRII) are present in the pdm BGC (Fig. 1b). Bioinformatic analysis shows that PdmRI belongs to the LuxR family of transcriptional regulators, which are reported as pathway-specific activators²⁴. PdmRII is a member of the MarR family of transcriptional regulators, which have been used for improving the production of salinomycin²⁵. Indeed, the inactivation of pdmRI and pdmRII completely abolished the production of 1 (Fig. 2a). We then attempted to overexpress pdmRI and pdmRII to enhance the yield of 1. To achieve this, the plasmids pCSG8141 (carrying pdmRI) and pCSG8142 (carrying pdmRII) were constructed and individually introduced into S. albus Del14/pCSG8135, but no increase in the production of 1 was detected in either the pdmRI or pdmRII overexpression strains (Fig. 2a). Alternatively, we tried to individually overexpress the regulator genes pdmRI and pdmRII in the wild-type (WT) strain S. marincola SCSIO 03032. Plasmids pCSG8143 (carrying pdmRI) and pCSG8144 (carrying pdmRII) were constructed and individually introduced into the WT strain. To our delight, the yield of 1 was increased in both of the pdmRI and pdmRII overexpression strains (Fig. 2b, c). Significantly, a 48-fold improvement was observed in S. marincola SCSIO 03032/pCSG8143 (overexpressing pdmRI, 48.0 ± 3.0 mg/L), compared to that in WT (1.0 ± 0.05 mg/L).

Subsequently, 10 L-scale cultures of S. marincola SCSIO 03032/pCSG8143 were performed, allowing the isolation of 1 as pale-yellow oil. The molecular formula of 1 was determined to be C₄₃H₇₀O₈ by HRESIMS (Supplementary Fig. 8a), indicating 9 degrees of unsaturation. The planar structure of 1 was established by comprehensive analysis of the NMR data (Supplementary Table 2, and Supplementary Fig. 8b−g). The geometries of the double bonds in 1 were determined to be 8Z, 17Z, and 20E from the ³J_HH coupling constants and NOESY experiments (Supplementary Table 2, and Supplementary Fig. 8h). Similarly, the relative configurations of the octalin (10R*, 11S*, 16R*, and 19R*), tetrahydrofuran (25S* and 28S*), and tetrahydropyran rings (29R* and 33S*) in 1 were elucidated based on the intensive NOESY analysis (Supplementary Fig. 8h). In addition, the relative configuration between CH₃-38/CH₃-39 was established as anti, while those between CH₃-39/CH₃-40, CH₃-41/23-OH, and 23-OH/25-OH were concluded to be syn by J-based configuration analysis (JBCA)²⁶ (Supplementary Fig. 9). The detailed structural elucidation of 1 is placed in Supplementary Note 1. Cumulatively, 1 (designated polydecalinmycin, PDM) was assigned as an octalin-containing polyether with established relative configurations (Fig. 2d).

Biosynthetic logic and chemical derivatization determine the absolute configuration of PDM

A biosynthetic pathway of PDM (1) is proposed (Fig. 3) based on the comparative bioinformatic analysis of pdm and other polyether BGCs^14,27. The formation of the acyl carrier protein (ACP)-tethered polyketide chain by PKS proteins PdmAI−PdmAV matches well with the collinearity rule of PKS²⁸. Epoxidation and epoxide hydrolysis by epoxidase (PdmC) and epoxide hydrolases (PdmBI and PdmBII) are proposed to occur on the ACP-tethered polyketide chain to generate the putative intermediate IM2, analogous to those proposed in the biosynthesis of lasalocid²⁹, nanchangmycin³⁰, and ionostatin²⁷. The cyclized intermediate IM2 is then proposed to be hydrolyzed from the ACP by the TE domain to produce the putative intermediate IM3. The oxidoreductase PdmO, showing 60% identity to InoJ, an oxidoreductase involved in the ionomycin biosynthesis²⁰, may catalyze both the dehydrogenation of C-8/C-9 and the [4 + 2] cycloaddition between C-11/C-16 and C-10/C-19 to complete the biosynthesis of PDM (1).

Fig. 3. Proposed biosynthetic pathway for PDM.

KS: ketosynthase, AT: acyltransferase, ACP: acyl carrier protein, DH: dehydratase, KR: ketoreductase, ER: enoylreductase, TE: thioesterase, IM: proposed intermediate compound, structurally uncharacterized. Solid arrows denote experimentally supported conversions; dashed arrows denote hypothetical or unverified steps.

Bioinformatic analysis has emerged as a confident tool for assigning the absolute configurations of ortho-PKS chains^31,32. The absolute configurations of ionomycin predicted through bioinformatic analysis of the ketoreductase (KR) and enoylreductase (ER) domains matched those confirmed by X-ray crystallographic analysis of its calcium complex^20,33 (Supplementary Fig. 10). Similarly, based on the bioinformatic investigation of the KR and ER domains from the five pdm PKSs (Supplementary Fig. 11), the absolute configurations of the ACP-tethered polyketide chain were proposed to be 2S, 4R, 6S, 9S, 22S, 23S, and 25S (Fig. 3), which were consistent with the corresponding configuration assigned by NMR. The two epoxides at C-28/C-29 and C-32/C-33 formed by PdmC in IM1 were predicted to have the configurations of R instead of S. This assignment was supported by the NOESY correlation between H-25 and H-29 (Supplementary Fig. 12) and was in accordance with the stereocontrol mechanism of epoxidase that trisubstituted alkenes yield (R,R) epoxides¹⁶. Furthermore, the sequential epoxide hydrolysis by PdmBI and PdmBII generated IM2 with the configurations of 25S, 28S, 29R, 32R, and 33S, which was similar to the confirmed process of lasalocid^15,34. Subsequently, the absolute configuration of PDM (1) was assigned as 2S, 4R, 6S, 10R, 11S, 16R, 19R, 22S, 23S, 25S, 28S, 29R, 32R, and 33S by comparing the experimental and calculated ECD spectra (Supplementary Fig. 13, and Supplementary Data 2 and 3). Finally, the 23S configuration was confirmed by the modified Mosher’s method on chemically derivatized methyl-PDM (Supplementary Tables 3 and 4, and Supplementary Figs. 16−18).

In vivo genetic and in vitro biochemical experiments implicate the function of PdmO in PDM biosynthesis

To probe the function of tailoring genes in the pdm BGC, we performed individual gene inactivation of four pdm genes (pdmO, pdmC, pdmBI, and pdmBII, Fig. 1b) in S. marincola SCSIO 03032 (Fig. 4a). Multiple independent attempts (>5) failed to inactivate the oxidoreductase-encoding gene pdmO, suggesting potential toxicity of accumulated intermediates in the putative ΔpdmO mutant. The ΔpdmC mutant was successfully generated but did not accumulate any detectable products, consistent with the hypothesis that the epoxidase PdmC might act on the ACP-tethered polyketide chain (Fig. 3). The epoxide hydrolase-encoding genes pdmBI and pdmBII are co-transcribed within the same operon (Fig. 1b). Notably, both single-gene inactivation mutants (ΔpdmBI and ΔpdmBII) yielded the same product 2. The structure of 2, designated prepolydecalinmycin A, was determined through comparative NMR analysis with 1 (Supplementary Figs. 8 and 19, and Supplementary Table 5). The detailed structural elucidation of 2 is placed in Supplementary Note 1. Unlike the octalin, tetrahydrofuran, and tetrahydropyran ring systems present in 1 (Fig. 2d), the structure of 2 was established to contain four double bonds and two epoxide groups (Fig. 4b). Despite the formation of tetrahydrofuran and tetrahydropyran rings, the lack of an unsaturation at C-8/C-9 in 2, compared to 1, indicating that the biosynthetic route from 2 to 1 requires a dehydrogenation step, along with the octalin cyclization.

Fig. 4. Functional investigation of the enzymes involved in PDM biosynthesis.

a HPLC analysis of metabolite profiles of ΔpdmBI, ΔpdmBII, and ΔpdmC mutants. b HPLC analysis of in vitro enzyme assays of PdmO, IonCII, or InoJ with 2. Enzyme assays were performed in Tris-HCl buffer (50 mM, pH 8.0) containing 2 (50 μM), enzyme (PdmO, IonCII, or InoJ, 10 μM), and NAD⁺ (2 mM) at 30 °C for 2 h. c HPLC analysis of the metabolite profiles of the ΔpdmO mutant and its complementation strains.

It has been proposed that IonCII and InoJ, two homologs of PdmO, catalyze the dehydrogenation of a C–C single bond to form the characteristic enolized 1,3-diketone moiety in ionostatin²⁷ and ionomycin²⁰, respectively. The absence of this moiety in 2 (Fig. 4b) suggested that 2 could serve as a candidate substrate for these enzymes. To test this hypothesis, PdmO, IonCII, and InoJ were expressed and purified to homogeneity (Supplementary Fig. 20), and individually incubated with 2 in the presence of NAD⁺. Product 3 with an [M+H]⁺ ion of m/z 715.5137 was formed in the reactions containing PdmO or InoJ, but not with IonCII (Fig. 4b). Although the inherent instability of 3 prevented its full NMR characterization (Supplementary Fig. 21), the presence of a Δ^8,9 in 3 was inferred from its 2 Da lower mass than 2 and further supported by the comparative HRMS/MS analysis of 2 and 3 (Supplementary Fig. 22). Therefore, our biochemical assays provided experimental evidence to support the dehydrogenase activity of PdmO and InoJ.

Inactivation of pdmO in the heterologous strain S. albus Del14/pCSG8135 was successful, which abolished the production of 1 but accumulated no observable intermediates (Fig. 2a). We then introduced pdmO and its homologous genes ionCII and inoJ into the ΔpdmO mutant. However, only complementation with pdmO, but not with ionCII or inoJ, could restore the production of 1 in the ΔpdmO mutant (Fig. 4c). These results indicated that PdmO should be involved in the biosynthesis of 1 (Fig. 3), possibly playing an essential role in the octalin ring formation that could not be complemented by the characterized oxidoreductase InoJ. Structural comparison reveals that the putative biosynthetic precursors for ionostatin and ionomycin lack the three double bonds necessary for [4+2] cycloaddition in the precursor of PDM (Supplementary Fig. 23). Furthermore, the AlphaFold 3³⁵-predicted NAD⁺ bound structures of PdmO, InoJ, and IonCII reveal highly similar NAD-binding domains but notable differences in their substrate-binding regions (Supplementary Fig. 24). Specifically, InoJ and IonCII possess helices and loops that narrow the substrate entrance, forming smaller binding pockets than PdmO, which likely restricts the substrate access to the catalytic site.

PDM accelerates acute and chronic skin wound healing

We previously found that polyether ionophores enhance the proliferation of epidermal cells³⁶. Consistently, PDM significantly promoted the growth of diverse cell types, encompassing human epidermal keratinocytes (HaCaT), primary mouse dermal fibroblasts (PMDF), mouse embryonic fibroblasts (MEF), and HEK293 (Fig. 5a, and Supplementary Fig. 25). PDM displayed a superior capacity to promote the proliferation of HaCaT cells compared to monensin A, a representative natural polyether ionophore (Supplementary Fig. 26). Furthermore, we noted the proficiency of PDM in expediting the migration of monolayer HaCaT cells (Supplementary Fig. 25). PDM also exhibited the potential to augment human vascular endothelial cell (HUVEC) proliferation and tube formation in vitro (Supplementary Fig. 25). Its pro-angiogenic potential was further confirmed by a classic chick embryo chorioallantoic membrane model, where PDM treatment resulted in a vascular morphology characterized by stout main blood vessels and abundant dendritic branches (Supplementary Fig. 25). These findings underscore the multifaceted roles of PDM in key processes essential for wound healing, including proliferation, migration, and angiogenesis³⁷. Given these comprehensive effects, we embarked on evaluating the impact of PDM on murine wound healing. Topical application of PDM at the wound site substantially accelerated wound closure (Fig. 5b), demonstrating its enhanced efficacy over monensin A in promoting cutaneous repair (Supplementary Fig. 26). Histological analysis revealed that PDM treatment resulted in markedly thicker granulation tissues beneath the wound (Fig. 5c), accompanied by enhanced vascularization (Fig. 5d). These results collectively suggest that PDM holds strong potential as a therapeutic agent for promoting acute skin wound healing.

Fig. 5. PDM promotes skin wound healing.

a Effect of PDM on HaCaT cell proliferation (n = 4 independent experiments, ^**P = 0.0092, ^***P = 0.0003, ^***P < 0.0001, ^***P < 0.0001, ^***P < 0.0001). b Dynamic wound areas in BALB/c mice receiving the PDM treatment (n = 7 mice per group receiving vehicle or PDM 250 nM treatment; n = 6 mice treated with PDM 125 nM). Punch diameter: 6 mm. PDM 125 nM vs 0 nM: ^*P = 0.0218, ^*P = 0.0183, ^*P = 0.0138; PDM 250 nM vs 0 nM: ^*P = 0.0398, ^**P = 0.0013, ^***P = 0.0001, ^***P = 0.0001. c H&E staining of the excised skin wounds. Scale bar: 250 μm. n = 5 independent experiments, relative wound length: ^*P = 0.0438, ^***P = 0.0007; relative granulation thickness: ^**P = 0.0017, ^**P = 0.0029. d Immunofluorescence staining of the wound tissues for endothelial markers CD31 and α-SMA. Scale bar: 20 μm (20×). n = 5 tissues per group, CD31⁺: ^**P = 0.0013, ^***P < 0.0001; α-SMA⁺: ^***P = 0.0004, ^***P < 0.0001. e Cell viability of HaCaT cells under PDM treatment in high-glucose (200 mM) culture medium (n = 3, ^***P < 0.0001, ^***P < 0.0001, ^***P < 0.0001, ^***P < 0.0001). f Representative images and quantification showing the dynamic wound healing in db/db mice receiving the PDM or EGF (250 nM) regimen. (n = 7 mice treated with vehicle, PDM, or EGF). Punch diameter: 6 mm. PDM 125 nM vs vehicle: ^*P = 0.0123, ^**P = 0.0015; PDM 250 nM vs vehicle: ^**P = 0.0032, ^***P = 0.0004; EGF 250 nM vs vehicle: ^**P = 0.0054. g H&E staining of the excised skin wounds. Scale bar: 250 μm. n = 7, relative wound length: ^***P = 0.0007, ^***P = 0.0001, ^*P = 0.0340; relative granulation thickness: ^*P = 0.0206, ^*P = 0.0237, ^*P = 0.0240. h Immunofluorescence staining of the wound tissues for endothelial markers CD31 and α-SMA. Scale bar: 20 μm (20×). n = 7, CD31⁺: ^***P < 0.0001, ^***P < 0.0001, ^**P = 0.0095; α-SMA⁺: ^***P < 0.0001, ^***P < 0.0001, ^**P = 0.0019. i, j Transcriptomic profiling of skin wound healing margin of db/db mice (n = 3). Data are mean ± SD (a, e) or mean ± SEM (b–d, f–h). P values were calculated using two-sided Student’s t-test (a, c, d, e, g, h) or Two-way ANOVA with Tukey’s multiple-comparison test (b, f). The mouse schematic (b, f) was created in BioRender. Deng, L. (2026) https://BioRender.com/5plq0py. Source data are provided as a Source data file.

We hypothesize that PDM may also represent a promising candidate for treating diabetic wounds, which are characterized by impaired healing capacity³⁸. Under high-glucose stress that mimics diabetic stress, PDM effectively restored the proliferation of HaCaT cells (Fig. 5e). In a diabetic mouse model with refractory wounds (Fig. 5f), PDM treatment significantly accelerated wound closure compared with vehicle controls and demonstrated superior efficacy relative to epidermal growth factor (EGF) (Fig. 5f). Histological analysis of excisional wounds revealed that PDM-treated mice exhibited a reduced wound gap and increased granulation tissue thickness compared to controls (Fig. 5g). PDM also robustly enhanced angiogenesis at the wound site (Fig. 5h). Transcriptomic profiling of the healing margins from PDM-treated db/db mice revealed that PDM-induced transcriptional changes were enriched in gene programs associated with epithelial proliferation, cytoskeletal remodeling, cell-cell adhesion, receptor signaling, and developmental processes (Fig. 5i, j, and Supplementary Fig. 27). Although Kyoto encyclopedia of genes and genomes (KEGG) pathway analysis indicated moderate enrichment of inflammatory pathways, such as NF-κB and IL-17 signaling, PDM administration did not induce hypertrophic scar formation in either db/db or BALB/c mice (Supplementary Fig. 28). Moreover, fasting blood glucose levels remained stable in PDM-treated db/db and BALB/c mice (Supplementary Fig. 29), indicating that the observed wound-healing benefits were not confounded by glycemic alteration. These findings highlight the potential application of PDM in accelerating the healing of diabetic skin wounds.

PDM enhances ROS generation for initiating acute wound healing

The structural resemblance of PDM to the polyether ionophores monensin A and lasalocid leads to the hypothesis that PDM may transport Na⁺ across the membrane (Supplementary Fig. 1)^36,39. To evaluate the binding affinity between PDM and Na⁺, isothermal titration calorimetry (ITC) experiments were performed in Tris-HCl buffer (pH 7.0), revealing an exothermic binding between PDM and Na⁺ with a dissociation constant (Kd) of 2.25 ± 0.88 μM (Fig. 6a). In PDM-treated HaCaT and HEK293 cells, we observed a conspicuous enhancement of fluorescent signal emitted by a specific Na⁺ probe (Fig. 6b, c). Previous studies have identified that the intracellular Na⁺ elevation increases the production of mitochondrial reactive oxygen species (ROS)^40,41. Of note, Na⁺ levels within the respiring mitochondrial matrix are typically lower than those in the cytosol and extracellular compartment⁴². We hypothesize that PDM increases the Na⁺ content within the mitochondrial matrix by freely transporting Na⁺ across the mitochondrial membrane. As expected, the isolated mitochondria treated with PDM showed elevated Na⁺ levels (Fig. 6d, e). Moreover, PDM retained its ability to elevate intracellular Na⁺ at 4 °C compared with 37 °C, indicating that PDM-induced increase of intracellular Na⁺ is not primarily dependent on active mitochondrial depolarization or endogenous ion channel transport (Supplementary Fig. 30). Within the mitochondrial matrix, it has been reported that Na⁺ binds with phospholipids in the inner mitochondrial membrane (IMM), reducing membrane fluidity and thereby prolonging semiquinone lifetime at the complex III Qo site and promoting ROS generation^41,43. We employed fluorescence recovery after photobleaching (FRAP) to investigate the effect of PDM on IMM fluidity. PDM markedly slowed down the compensation of the bleached areas in both HaCaT and HEK293 cells (Fig. 6f), suggesting a reduction in IMM fluidity mediated by PDM. The PDM-treated HaCaT cells showed a significant increase in ROS, and this PDM-induced superoxide generation was markedly inhibited by the complex III Qo site inhibitor, myxothiazol (Fig. 6g). Moreover, the ROS induced by PDM could be scavenged by N-acetyl cysteine (NAC) (Fig. 6h). Consistently, the proliferation activity of PDM was nullified by myxothiazol and NAC (Fig. 6i, j). Compared to PDM monotherapy, combining PDM with NAC reduced the efficacy of PDM in murine wounds (Fig. 6k), suggesting that PDM-induced ROS act as a key effector during the initiation of acute skin wound healing.

Fig. 6. PDM elevates mitochondrial Na⁺ to generate ROS and initiate acute wound healing.

a Thermodynamics and fitted curve of the titrating PDM (20 μM) to NaCl (200 μM) in Tris-HCl buffer (pH = 7.0) using an isothermal titration calorimetry. b Representative confocal images showing the intracellular Na⁺ content probed by ING-2 in the PDM-treated HaCaT (left) and HEK293 (right) (n = 5 sections per group). Scale bar: 10 μm. The experiment was repeated independently three times with similar results. c Microplate reader-based quantification of intracellular Na⁺ levels in HaCaT (n = 3 independent experiments, ^*P = 0.0123, ^*P = 0.0221, ^***P = 0.0001, ^***P < 0.0001) and HEK293 (n = 3 independent experiments, ^*P = 0.0167, ^**P = 0.0019, ^**P = 0.0017, ^***P = 0.0002) cells incubated with PDM. d Expression of mitochondrial HSP60 and cytoplasmic β-actin. The experiment was repeated independently three times with similar results. e Mitochondrial Na⁺ contents probed by ING-2 (n = 3 independent experiments, ^**P = 0.0080, ^***P = 0.0004). f, FRAP analysis in the HaCaT (n = 6 independent experiments, ^***P < 0.0001) and HEK293 (n = 6 independent experiments, ^***P < 0.0001) cells exposed to PDM (500 nM) for 1 h. The blocking effect of myxothiazol (g, n = 3 independent experiments, myxothiazol 0 nM, PDM vs 0: ^**P = 0.0042, ^**P = 0.0013, ^***P = 0.0002; myxothiazol 10 nM vs myxothiazol 0 nM: ^###P = 0.0004, ^###P = 0.0001) and NAC (h, HaCaT, n = 3 independent experiments, NAC 0 mM, PDM vs 0: ^**P = 0.0076, ^***P = 0.0006, ^***P = 0.0001, ^***P = 0.0001; NAC 20 mM vs NAC 0 mM: ^###P = 0.0007, ^###P = 0.0002, ^##P = 0.0021; HEK293, n = 3, NAC 0 mM, PDM vs 0: ^**P = 0.0094, ^**P = 0.0058, ^**P = 0.0027, ^***P < 0.0001; NAC 10 mM vs NAC 0 mM: ^##P = 0.0031, ^#P = 0.0173, ^##P = 0.0030, ^###P = 0.0001) on the PDM-induced ROS was evaluated in HaCaT or HEK293 cells. Cell viability of HaCaT cells treated with PDM and myxothiazol (i, n = 3 independent experiments, ^*P = 0.0249, ^**P = 0.0021, ^***P = 0.0002, ^***P = 0.0004) or NAC (j, n = 3 independent experiments, ^***P < 0.0001, ^***P < 0.0001, ^**P = 0.0010, ^***P < 0.0001, ^***P = 0.0001) combination. k Effect of NAC (20 mM) on PDM-mediated wound healing in BALB/c mice. The relative wound areas are represented as mean ± SEM (n = 5 mice per group). Punch diameter: 6 mm. ^**P = 0.0020, ^***P = 0.0003, ^*P = 0.0118. Data are expressed as mean ± SD (c, e, g–j) from at least triplicate independent experiments or mean ± SEM (k). Two-tailed unpaired Student’s t-test (c, e–j) and Two-way ANOVA with Tukey’s multiple-comparison test (k) were employed to analyze the statistical significance. The mouse schematic (k) was created in BioRender. Deng, L. (2026) https://BioRender.com/5plq0py. Source data are provided as a Source data file.

Beyond Na⁺, we examined the effects of PDM on other macronutrient metal ions (Ca²⁺, Mg²⁺, and K⁺) in standard culture media. PDM treatment increased intracellular Ca²⁺ (Supplementary Fig. 31) and K⁺ (Supplementary Fig. 31) levels, while intracellular Mg²⁺ levels remained unchanged (Supplementary Fig. 31). In contrast, ionomycin, a canonical Ca²⁺ ionophore that was able to increase intracellular Ca²⁺ levels⁴⁴, did not stimulate cell proliferation (Supplementary Fig. 31). Notably, PDM retained its ability to promote cell proliferation even in Ca²⁺-free medium (Supplementary Fig. 31), indicating that PDM-mediated Ca²⁺ influx is not sufficient to drive this proliferative effect in vitro. In contrast to the K⁺ ionophore nigericin that induces K⁺ efflux due to high intracellular K⁺ context (Supplementary Fig. 31), the increase of intracellular K⁺ levels observed following PDM treatment is likely driven by Na⁺ influx-dependent Na⁺/K⁺ exchange via Na⁺/K⁺-ATPase, resulting in enhanced K⁺ uptake⁴⁵. To assess whether elevated intracellular K⁺ contributes to cell proliferation, KCl was added to the culture medium to increase intracellular K⁺ levels (Supplementary Fig. 31). This intervention failed to promote HaCaT cell proliferation (Supplementary Fig. 31), indicating that K⁺ accumulation is not a driver of cell proliferation. In summary, similar to other ionophores, PDM lacks strict Na⁺ selectivity and may allow the transport of other cations, such as Ca²⁺.

PDM activates an Nrf2-mediated antioxidant response to restore mitochondrial homeostasis and protect cells from high-glucose-induced oxidative stress

To delve deeper into the underlying mechanisms of PDM in wound healing, we performed bulk RNA sequencing to obtain a comprehensive overview of how PDM affects the transcriptome in HaCaT cells (Fig. 7a, and Supplementary Fig. 32). Bioinformatic analysis using the Reactome pathway database revealed that the top-regulated genes were highly enriched in mitochondrial DNA (mtDNA)-encoded mitochondrial genes (Supplementary Fig. 32), with most of these genes showing upregulated (Fig. 7b, c). We further validated the active expression of a substantial portion (9 out of 15) of mitochondrial genes in PDM-treated HaCaT cells (Fig. 7d). These included genes encoding components of mitochondrial complex I, complex III, complex IV, complex V, and mitochondrial rRNA (Fig. 7d). Moreover, PDM profoundly elevated the mtDNA copy number in HaCaT cells (Fig. 7e), suggesting a potential contribution to the active expression of the respiratory genes⁴⁶.

Fig. 7. PDM improves mitochondrial functions and activates an Nrf2-mediated antioxidant adaptive response.

The differentially expressed genes (a) between the DMSO and PDM-treated HaCaT cells, and the up-regulated pathways (b) enriched in pathway enrichment analysis. c Heatmap showing the up-regulated mitochondrial gene cluster. d RT-PCR-based expression analysis of mitochondrial genes (n = 3 independent experiments, ^*P = 0.0126, ^**P = 0.0081, ^*P = 0.0248, ^**P = 0.0036, ^**P = 0.0044, ^**P = 0.0055, ^*P = 0.0163, ^**P = 0.0041, ^*P = 0.0165, ^**P = 0.0031, ^***P = 0.00098). e Alteration of mitochondrial DNA copy number induced by PDM in HaCaT cells (n = 3 independent experiments, SLCO2B1-ND1, ^*P = 0.0145, ^***P = 0.0001; SERPINA1-ND5, ^*P = 0.0161, ^**P = 0.0018, ^***P = 0.0001). f Seahorse-based real-time oxygen consumption assay on HaCaT cells under PDM or FCCP (1000 nM) treatment. g Glucose consumption rate in HaCaT cells in the presence or absence of PDM at the indicated time points (n = 3 independent experiments, PDM 250 nM vs 0 nM, ^**P = 0.0094, ^**P = 0.0028; PDM 500 nM vs 0 nM, ^**P = 0.0012, ^**P = 0.0015). h Analysis of intracellular ATP content in HaCaT cells incubated with PDM (n = 3 independent experiments, ^***P = 0.0001, ^***P < 0.0001, ^***P < 0.0001). i PDM-mediated temporal dynamics of Na⁺ and ROS in HaCaT cells under normoglycemic or high-glucose conditions. j Alteration of Nrf2 expression induced by PDM in HaCaT cells. The experiment was repeated independently three times with similar results. k Immunofluorescent staining of Nrf2 in HaCaT cells with PDM treatment for 48 h. Scale bar: 10 μm. The experiment was repeated independently three times with similar results. l RT-PCR-based expression analysis of Nrf2-regulated genes (GPX2, PRDX1, SRX1, and TXN1) in the PDM-treated HaCaT cells (GPX2, PDM 500 nM vs 0 nM: ^*P = 0.0129, ^**P = 0.0026, high glucose vs normal glucose: ^**P = 0.0095; PRDX1, PDM 500 nM vs 0 nM: ^***P < 0.0001, ^***P < 0.0001, high glucose vs normal glucose: ^*P = 0.0286; SRX1, PDM 500 nM vs 0 nM: ^**P = 0.0045, ^***P = 0.0008, high glucose vs normal glucose: ^**P = 0.0097; TXN1, PDM 500 nM vs 0 nM: ^***P < 0.0001, ^***P < 0.0001, high glucose vs normal glucose: ^**P = 0.0094). m Cell viability of HaCaT cells under normoglycemic or high-glucose conditions exposed to different concentrations of PDM for 72 h. Data are displayed as mean ± SD (n = 3 independent experiments). Two-tailed Student’s t-test (d, e, h, l) or Two-way ANOVA with Tukey’s multiple-comparison test (g). Source data are provided as a Source data file.

In pursuit of deciphering the identity of PDM and its potential role in orchestrating diverse mitochondrial gene expressions for supporting high-intensity mitochondrial respiratory, we focused on mitochondrial functions widely acknowledged as essential for driving productive wound healing progress⁴⁷. When mitochondrial complex V (ATPase) was inhibited by oligomycin to prevent proton inflow, we observed that the real-time oxygen consumption rate (OCR) in HaCaT cells significantly expedited following PDM treatment (Fig. 7f). Notably, the pace of oxygen consumption was finally braked when the mitochondrial complexes I and III were blocked using rotenone and antimycin A (Fig. 7f). Consistently, within a 24-h cell proliferation cycle, PDM significantly promoted glucose consumption in HaCaT cells (Fig. 7g). Moreover, PDM substantially boosted ATP levels in HaCaT cells within 24 h post-treatment (Fig. 7h). Together, these results demonstrate that PDM enhances mitochondrial function in keratinocytes.

ROS are beneficial for initiating acute wound healing but become chronically detrimental in diabetic wounds, where excessive oxidative stress contributes to hard-to-heal wounds⁴⁸. Consistently, we found that scavenging ROS within HaCaT cells using NAC promoted the expression of multiple mitochondrial respiratory chain genes (Supplementary Fig. 33). To clarify how PDM enhances wound healing, especially under the high-glucose condition, we prolonged the detection period to systematically characterize the temporal dynamics of intracellular Na⁺ and ROS in normoglycemic and high-glucose conditions. Intracellular Na⁺ levels exhibited a dynamic trajectory, featuring an initial rapid increase followed by a decline and stabilization above the baseline (Fig. 7i). Notably, intracellular ROS exhibited a rapid surge within the first 30 min, followed by a gradual decrease to levels below the baseline that were sustained over time (Fig. 7i). The subsequent reduction in ROS is consistent with Nrf2-mediated redox adaptation. PDM increased both Nrf2 expression (Fig. 7j) and its nuclear localization (Fig. 7k), accompanied by pronounced upregulation of downstream Nrf2-regulated antioxidant genes, including GPX2, PRDX1, SRX1, and TXN1 (Fig. 7l). Short interfering RNA-mediated silencing of Nrf2 significantly attenuated the pro-proliferative effect of PDM (Supplementary Fig. 33). Compared with normoglycemic conditions, high-glucose stress amplified PDM-triggered Nrf2 activation and its downstream antioxidant response (Fig. 7j–l). Compared with the normoglycemic condition, PDM displayed a substantially superior capacity to promote cell proliferation under the high-glucose environment (Fig. 7m). Collectively, these findings indicate that PDM promotes diabetic wound healing by activating an Nrf2-mediated antioxidant response.

Discussion

The comprehensive evidence presented in this study has unveiled several important findings. Firstly, we employ the genome mining approach to allow the efficient identification of a polyether gene cluster from a deep-sea-derived Streptomyces, and the overexpression of a positive regulator leads to the successful isolation of a novel polyether PDM with an approximate 48-fold production increase compared to the WT. Moreover, our studies demonstrate that PDM possesses significant pharmaceutical potential, effectively facilitating both acute and chronic wound healing. Finally, we elucidate the mechanistic basis of PDM bioactivity, showing that PDM enables Na⁺ influx to promote ROS production, which subsequently activates an Nrf2-mediated antioxidant adaptive response. This response restores mitochondrial homeostasis and provides cytoprotective effects against high-glucose-induced oxidative stress.

It is well established that secondary metabolism in Streptomyces is controlled by a complex regulatory network, including pathway-specific regulators, global regulators, and small signaling molecules⁴⁹. Pathway-specific regulators frequently rely on native regulatory networks to function efficiently⁵⁰. In this work, overexpressing the positive regulator genes pdmRI and pdmRII boosted the PDM production in the native producer S. marincola SCSIO 03032 but not in the heterologous host S. albus Del14 (Fig. 2a, b). This discrepancy therefore leads us to speculate that some unknown regulatory elements or interactions present in S. marincola SCSIO 03032—but absent in S. albus Del14—are necessary for PdmRI and PdmRII to upregulate the pdm BGC. In contrast to the laboratory conditions, the native deep-sea sediment environment (−3412 m in the Indian Ocean) from which S. marincola SCSIO 03032¹³ was isolated is characterized by high salinity, high pressure, low temperature, darkness, and nutrient limitation. The harsh living conditions and complex microbial communities may serve as natural triggers for the expression of pdm BGC in the wild-type strain. The ecological significance of this hypothesis, however, remains difficult to assess due to the limited feasibility of conducting in situ studies.

To the best of our knowledge, PDM is the first polyether ionophore containing an octalin ring, which is proposed to be formed through [4+2] Diels-Alder cycloaddition⁵¹. The discovery of only one PDM stereoisomer leads us to propose that the SDR family oxidoreductase PdmO may function similarly to the bifunctional oxidoreductase/Diels-Alderases MalC/PhqE⁵² and CtdP⁵³, catalyzing both dehydrogenation and [4+2] cycloaddition to construct the strict stereoselectivity during the octalin ring formation (Fig. 3). While both PdmO and its homolog InoJ were confirmed to possess dehydrogenation activity (converting 2 to 3, Fig. 4b), direct biochemical evidence for verifying the [4+2] cycloaddition activity of PdmO was not obtained, as no products containing the octalin ring were observed (Fig. 4b). Nevertheless, the cycloaddition activity of PdmO was inferred indirectly from the successful restoration of PDM production in the ΔpdmO mutant by complementation with pdmO, but not with inoJ (Fig. 4c), despite both enzymes sharing the dehydrogenation function. These observations suggest that while 2 served as a surrogate substrate for assessing the dehydrogenation activity of both PdmO and InoJ, it might not be the native substrate required for PdmO to catalyze the [4+2] cycloaddition. Based on the biosynthetic logic, we propose that the native PdmO substrate is likely compound IM3 (Fig. 3), which contains the characteristic tetrahydrofuran and tetrahydropyran rings.

Cumulative evidence from studies on various organisms has underscored the significance of ROS in initiating wound repair, functioning as second messengers that drive cell adaptation^54,55. Our findings establish that PDM orchestrates a biphasic response to accelerate both acute and diabetic wound healing. One of the key mechanistic insights from this study is the role of transient ROS generation in initiating the healing process. In the acute phase, PDM elevates mitochondrial Na⁺ levels, leading to reduced IMM fluidity and increased ROS production at the complex III Qo site (Fig. 7a–e). This ROS surge is essential for PDM-mediated acute wound repair, as scavenging ROS with NAC abolished its therapeutic efficacy (Fig. 7f–j). These observations align with the established role of ROS as vital signaling molecules that promote early-stage cellular recruitment and signal activation during wound healing⁵⁶. However, excessive oxidative stress is a key feature of diabetic wounds and a major barrier to healing⁴⁸. We found that PDM ultimately resolves this “oxidative paradox” by activating the Nrf2-mediated antioxidant response, a critical transcription factor that regulates redox, protein, and metabolic homeostasis⁵⁷. Of note, although Nrf2 displays basal activity, it is dysregulated and functionally insufficient in diabetic skin wounds subjected to hyperglycemia-associated chronic oxidative stress⁵⁸. With extended PDM exposure, Nrf2-mediated antioxidant signaling becomes predominant, resulting in a dominant antioxidant effect and a decline in ROS levels below baseline (Fig. 7i). Under high-glucose stress, this adaptive program is further amplified, effectively shielding cells from oxidative damage and restoring cellular proliferative capacity (Fig. 7m). Compared with current clinical treatments, PDM offers several distinct advantages. Unlike exogenous growth factors such as EGF, which have limited stability and efficacy in the protease-rich environment of chronic wounds⁵⁹, PDM demonstrated superior recovery rates in the refractory diabetic model (Fig. 5f–h). Furthermore, PDM exerts pleiotropic effects during wound healing, facilitating key processes such as proliferation, migration, and angiogenesis (Fig. 5, and Supplementary Fig. 25). Although PDM-mediated wound-healing activity appears to be associated with Na⁺ influx, its comprehensive ion transport specificity remains to be fully characterized. Transport of other ions mediated by PDM, particularly Ca²⁺ and K⁺ (Supplementary Fig. 31), may also contribute to promoting wound healing through complementary or synergistic mechanisms. Notably, preliminary safety evaluation revealed that repeated administration of PDM elicited no detectable toxicity in BALB/c mice, as evidenced by normal body weight, unremarkable serum biochemical parameters, and the absence of obvious histopathological abnormalities in major organs (Supplementary Fig. 34). In addition, PDM did not induce hypertrophic scar formation in treated db/db and BALB/c mice (Supplementary Fig. 28). These findings support PDM as a potential candidate for expediting wound healing. More comprehensive investigations of repeated topical exposure, long-term local safety, systemic exposure, therapeutic window, and ionophore-related safety of PDM will be essential before clinical translation.

In summary, PDM, a novel octalin-containing polyether ionophore discovered through genome mining of a polyether BGC, induces mitochondrial Na⁺ influx to enhance ROS production, activates Nrf2-mediated antioxidant signaling, and improves mitochondrial function in wounded skin repair cells. These responses contribute to accelerated wound healing, particularly in hard-to-heal diabetic wounds. Given the limited availability of FDA-approved therapies for hard-to-heal diabetic wounds, our study provides a basis for further development of ionophore-based approaches for wound healing.

Methods

Ethical statement

All animal studies were carried out in accordance with relevant institutional and national guidelines and were approved by the Sun Yat-sen University Institutional Animal Care and Use Committee (IACUC-2025102301, 20191028-010, 20190723-001, 20190313-001). BALB/c mice (male and female, 5–8 weeks) were purchased from Vital River (Beijing, China), Leptin receptor-deficient db/db mice (male, 5–8 weeks) were derived from Gempharmatech (Jiangsu, China). Mice were housed in Individually Ventilated Cages (IVCs) under SPF conditions with a 12 h light/dark cycle, temperature of 22 ± 2°C, and humidity of 55 ± 10%. In vitro experiments were conducted using established cell lines, which were exempt from institutional ethical approval.

Bacterial strains, plasmids, and reagents

Bacterial strains, plasmids, and primers used and constructed in this study are listed in Supplementary Table 6 and Supplementary Data 1. S. marincola SCSIO 03032 was isolated from a sediment sample (E 87^o59.7’, N 9^o59.3’) at a depth of 3412 m from the Bay of Bengal in the Indian Ocean¹³. The BAC library of S. marincola SCSIO 03032 was constructed using the BAC vector pMSBBAC2²¹ by Eight Star Bio-tech company. Chemicals, enzymes, and other molecular biological reagents were purchased from standard commercial sources and used according to the manufacturer’s recommendations.

DNA manipulation and bioinformatic analysis

DNA manipulation in Escherichiacoli and actinomycetes was carried out using standard molecular biology procedures⁶⁰. Primers, genes, and siRNAs used in this study were synthesized at the Beijing Genomics Institution Co., Ltd. PCR amplifications were carried out on an Authorized Thermal Cycler (Eppendorf AG). DNA sequencing was performed at Beijing Tsingke Biotech Co., Ltd. The end-sequencing was performed at the Eight Star Bio-tech company (Wuhan). Protein functional predictions and homologous protein sequence alignments were obtained using a BLASTp algorithm. Multiple sequence alignments were performed using online CLUSTALW and ESPript 3.0 software. The protein structures were predicted using AlphaFold 3 and aligned with PyMOL software.

General experimental procedures

NMR spectra were recorded on a Bruker AV-700 MHz NMR spectrometer with TMS as the internal standard. Deuterated NMR solvents were purchased from Cambridge Isotopes. HRESIMS data were measured on a Bruker Maxis 4G UHR-TOFMS spectrometer and analyzed with DataAnalysis 4.4. Materials for column chromatography (CC) were silica gel (100−200 mesh; Jiangyou Silica Gel Development, Inc.) and Sephadex LH-20 (40−70 μm; Amersham Pharmacia Biotech AB). HPLC analysis was carried out on an Agilent 1200 workstation with UV detection at 280 nm under the following program: solvent A, H₂O supplemented with 0.1% trifluoroacetic acid; solvent B, CH₃CN; 5% B to 100% B (0−15 min), 100% B (16−25 min), 100% B to 5% B (25−26 min), 5% B (26−30 min); flow rate 1 mL min⁻¹.

Genome mining of polyether BGCs

All genomes were retrieved from the NCBI databases. To search for polyether BGCs, the HMMER suite package was applied to locate PF00698 and PF08659 containing CDSs (type I PKS protein), and homologs of Lsd18 (GenBank accession no. BAG85033) and Lsd19 (GenBank accession no. BAG85034) with more than 40% identity. Candidate BGCs containing these marker genes in close genomic proximity were retrieved. Pairwise Jaccard Indexes (Equation 1 in Supplementary Method 1) were calculated between each BGC based on the homologous gene content, and a cutoff of 0.6 was used to construct the similarity network. Each subgraph in the network was assigned an ID, and all singletons were removed. The pdm BGC was prioritized through integrated ClusterBlast (antiSMASH) and BLASTp (NCBI) analyzes. Detailed descriptions of the genome dataset, search parameters, network construction, and BGC prioritization criteria are provided in the Supplementary Methods 1–14.

Heterologous expression of pdm BGC

The positive BAC clone (E. coli DH10B/pCSG8135) containing the entire pdm BGC was screened from the BAC library of S. marincola SCSIO 03032 and confirmed by end-sequencing. The plasmid pCSG8135 was transferred into the heterologous host S. albus Del14 via triparental conjugation with the help of E. coli ET12567/pUB307. Briefly, the donor strain E. coli DH10B/pCSG8135, the helper strain E. coli ET12567/pUB307, and the recipient strain S. albus Del14 were mixed with a ratio of 1:1:2 and spread on a modified ISP4 medium plate containing 20 mM MgCl₂. After incubation at 28 °C for 18−20 h, the plate was covered with 50 μg mL⁻¹ of apramycin and 100 μg mL⁻¹ of trimethoprim and incubated at 28 °C for 5−7 days. The exconjugants were identified by diagnostic PCR, and three positive clones were randomly selected for small-scale fermentation and subjected to metabolite analyzes by HPLC.

Construction of gene inactivation mutants

Target genes were inactivated by the lambda-RED-mediated gene replacements⁶¹. Details for individual gene disruptions are depicted in Supplementary Fig. 6. The Δorf(−1), ΔpdmRI, ΔpdmRII, ΔpdmO, and Δorf1 mutants were constructed by replacing the targeted gene with an oriT and aadA cassette in pCSG8135, and then transferring the corresponding BACs carrying the mutated alleles into heterologous host S. albus Del14 via triparental conjugation as described above. The ΔpdmBI, ΔpdmBII, and ΔpdmC mutants were constructed by replacing the targeted gene with an oriT and aac(3)IV cassette in pCSG8145, and then transferring the corresponding cosmids carrying the mutated alleles into wild-type S. marincola SCSIO 03032 via biparental conjugation to obtain double crossover mutants. The exconjugants were identified by diagnostic PCR and three positive clones were randomly selected for small-scale fermentation and subjected to metabolite analyzes by HPLC.

Construction of gene overexpression strains

The ermE*p+pdmRI and ermE*p+pdmRII fragments were PCR amplified and inserted into pPM927 or pSET152 linearized with XbaI/BamHI to afford plasmids pCSG8141/pCSG8142 (in pPM927) and pCSG8143/pCSG8144 (in pSET152) after sequence confirmation. The plasmids pCSG8141 and pCSG8142 were then transferred into the pdm BGC heterologous strain S. albus Del14/pCSG8135 via triparental conjugation as described above, respectively. The plasmids pCSG8143 and pCSG8144 were then transferred into the wild-type S. marincola SCSIO 03032 via biparental conjugation as described above, respectively. The exconjugants were identified by diagnostic PCR, and three positive clones were randomly selected for small-scale fermentation and subjected to metabolite analyzes by HPLC.

Construction of gene complementation strains

The pdmO, ionCII, and inoJ fragments were PCR amplified and inserted into pPWW50A linearized with NdeI/BamHI to afford plasmids pCSG8152, pCSG8153, and pCSG8154 after sequence confirmation, respectively. These plasmids were then individually transferred into the ΔpdmO mutants S. albus Del14/pCSG8139 via triparental conjugation with the help of E. coli ET12567/pUB307 as described above. The exconjugants were identified by diagnostic PCR, and three positive clones were randomly selected for small-scale fermentation and subjected to metabolite analyzes by HPLC.

Fermentation and metabolite analysis

Media used in this study are listed in Supplementary Table 7. S. albus Del14 and its recombination strains were cultured on SFM medium agar plate and their fermentations were carried out using 50 mL of N4 medium by shaking at 200 rpm and 28 °C for 5‒7 days. S. marincola SCSIO 03032 and its recombination strains were cultured on 38# medium agar plates, and their fermentations were performed in 50 mL of 1# medium by shaking at 28 °C, 200 rpm for 3−4 days; After growing to the logarithmic growth phase, 5 mL of seed cultures were transferred to 50 mL of ISP4 medium and incubated at 28 °C, 200 rpm for 5‒7 days. The fermentation broth was extracted with butanone and then concentrated under vacuum. The residues were redissolved in dimethyl sulfoxide and subjected to HPLC analysis as described above. The yield of 1 was calculated from the standard curve (Supplementary Fig. 7).

Large-scale fermentation, extraction, and isolation

The large-scale fermentation of S. marincola SCSIO 03032/pCSG8143 and S. marincola SCSIO 03032/ΔpdmBI was carried out using 50 mL of 1# seed medium and 200 mL of ISP4 fermentation medium supplemented with 10 mL of Amberlite XAD-16 resin. A total of 10 L cultures were prepared. Subsequently, the XAD-16 resin was filtered from the fermentation cultures and extracted with acetone to afford the crude extract. The crude extract of S. marincola SCSIO 03032/pCSG8143 (2.4 g) was separated by Sephadex LH-20 CC and eluted with CHCl₃/CH₃OH (1:1) to give five subfractions (Fr1 to Fr5). Fr1 was further subjected to 100−200 mesh silica gel CC, eluting with a gradient of CHCl₃/CH₃OH (100/0, 98/2, 95/5, 90/10, 80/20, 0/100) to yield 1 (500.0 mg). The crude extract of S. marincola SCSIO 03032/ΔpdmBI (1.5 g) was separated by Sephadex LH-20 CC and eluted with CHCl₃/CH₃OH (1:1) to give eight subfractions (Fr1 to Fr8). Fr2 was further purified by semipreparative HPLC (H₂O/CH₃CN, 35/65, flow rate 2.5 mL min⁻¹) to yield 2 (2.7 mg).

Biochemical assays

The pdmO, ionCII, and inoJ genes were overexpressed in E. coli BL21(DE3) and the corresponding proteins were purified by Ni²⁺-NTA affinity chromatography according to the manufacturer’s manual (Novagen, Germany). A typical enzyme reaction was conducted in a 100 μL reaction mixture consisting of 50 μM 2, 10 μM PdmO, IonCII, or InoJ, and 2 mM NAD⁺ in Tris-HCl buffer (50 mM, pH 8.0). The reactions were incubated at 30 °C for 2 h and then stopped by adding 100 μL of ice-cold CH₃OH. The enzyme assays were monitored by HPLC analysis as described above.

Isothermal titration calorimetry (ITC)

ITC experiments were performed using a PEAQ-ITC titration calorimeter (MicroCal). PDM (20 μM) was titrated into NaCl solution (200 μM) at 25 °C in Tris-HCl buffer (20 mM, pH 7.0). Heat effects arising from dilution were measured in control experiments and subtracted to obtain the net enthalpy of binding.

Cell lines and cell culture

Human epidermal keratinocytes HaCaT (SCSP-5091) were purchased from the National Collection of Authenticated Cell Cultures (Shanghai, China). Human embryonic kidney 293 cells (HEK293, CRL-1573), mouse embryonic fibroblasts (MEF, CRL‑2991), and primary umbilical vein endothelial cells (HUVEC, CRL‑1730) were obtained from the American Type Culture Collection. All the cell lines were confirmed by Short tandem repeat (STR) profiling. PMDF with positive vimentin phenotype validation were isolated from newborn (1–3 days) BALB/c mice (Cloud-Clone Corp, CSI088Mu01)⁶². We periodically tested the mycoplasma to ensure no contamination.

Cell viability assay

Cells (2000–5000 cells/well) were seeded on 96-well plates for 24 h and then treated with PDM at different concentrations for 72 h. Cell Counting Kit 8 (CCK-8, APExBIO)⁶³ was employed to quantify cell viability. Briefly, water-soluble tetrazolium salt (WST-8) in the kit was added to cells and incubated for 4 h at 37  °C. The absorbance at 450 nm was measured using in a microplate reader, and cell viability was calculated relative to vehicle-treated controls.

Animal experiments

To develop the mouse full-thickness wound model, two excisional wounds on the dorsal skin were created using a 5-mm sterile biopsy punch in anesthetized mice after hair removal and disinfection treatment⁶⁴. Wounds were treated with vehicle (PBS, 20 μL per wound), PDM (250 nM or 125 nM in PBS, 20 μL per wound), or recombinant murine EGF (250 nM in PBS, 20 μL per wound; Sigma-Aldrich), and then covered with Tegaderm film dressing (3 M). Treatments were applied every 2–4 days. Dynamic wounds were digitally photographed every 2–4 days, and all wound areas were normalized to peripheral punched silica (6 mm diameter) using ImageJ. Wound healing dynamics are expressed as percentages of the corresponding initial wound areas. The excised wound tissues were subjected to H&E staining and immunofluorescence analysis according to standard protocols. The stained sections were scanned using a high-throughput whole-slide imaging system (Qiangsheng, Shenzhen). Each biological replicate corresponded to one mouse (5–7 mice per group). Two wounds on each mouse were analyzed as technical replicates and averaged to generate one value per mouse before statistical analysis.

In the biosafety study, healthy BALB/c mice were repeatedly subjected (5 injections) to PDM (0, 0.1, 0.2, 1.0, 2.0 mg/kg in PBS, i.p., every 2 days). For the biosafety assay, blood and organ (heart, liver, spleen, lung, and kidney) samples were collected from the mice 48 h after drug discontinuation.

Intracellular and mitochondrial Na⁺ detection

The adherent HaCaT cells or resuspended mitochondrial pellets were sustained in a complete medium or mitochondrial stock solution (Beyotime), respectively, followed by vehicle or PDM treatment for 0.5 h or a series of time points (5, 15, 30 min, 1, 2, 3, 6, 12, 24, 48, and 72 h). Simultaneously, a Na⁺-sensitive probe ION NaTRIUM Green-2 AM (ING-2, 5 μM) was added to detect the intracellular and mitochondrial Na⁺ content. We initially visualized the PDM-induced fluorescent alteration of the Na⁺ probe in HaCaT using an Ar laser (488 nm) scanning confocal microscope (Carl Zeiss), and the living cells were maintained at 37 °C with 5% CO_2, supported by its attached device. To quantify the intracellular and mitochondrial Na⁺ levels, we employed a microplate reader to measure the ING-2 fluorescence in PDM-treated HaCaT cells and mitochondria seeded in 96-well plates suitable for fluorescence-based assays. The probe was excited at 488 nm, and the microplate reader recorded its corresponding emission wavelength at 550 ± 5 nm. Experiments were performed in three independent biological replicates (n = 3), each with 3 technical replicates that were averaged for statistical analysis.

Fluorescence recovery after photobleaching (FRAP) analysis

Living HaCaT and HEK293 cells grown in confocal-based glass-bottom dishes were incubated in a complete medium containing PDM (500 nM) for 1 h. A cardiolipin probe, acridine orange 10-nonyl bromide (NAO; Ex: 489 nm, Em: 525 nm), was added to the cells for labeling the IMM at a final concentration of 50  nM⁶⁵. In the FRAP experiment, the focal fluorescent areas of the dye were bleached through high-energy irradiation with 15 scans at 20% laser power using confocal microscopy (Carl Zeiss). The dynamic fluorescence in the bleached areas was monitored over 40–60 scans (HACAT: 57 scans; HEK293: 43 scans; 1 s/scan). Data are presented as background-subtracted fluorescence intensities for each scan. FRAP analyzes were performed in six independent biological replicates (n = 6).

ROS measurement

HaCaT and HEK293 cells seeded in black opaque 96-well plates were treated with PDM for the indicated periods. Intracellular ROS was indicated by the cell-permeant H₂DCFDA (2’,7’-Dichlorodihydrofluorescein diacetate, ThermoFisher) at a final concentration of 37.5 μM for 30 min. Upon cleavage of the acetate groups in the probe by intracellular Esterases and oxidation, its cleaved product, 2’,7’-dichlorofluorescein, was tested at a 520 nm wavelength after exciting at 495 nm using a microplate reader. Experiments were performed in three independent biological replicates (n = 3). For each biological replicate, 3 technical replicates were averaged before statistical analysis.

RNA sequencing and bioinformatics analysis

The total RNA of excised wound tissues derived from db/db mice treated with PDM (250 nM, 20 μL, 5 repetitions, once every 2 days) or HaCaT cells treated with PDM (500 nM) for 3 h was extracted using a column-based total RNA purification kit following its recommended instructions. Bulk RNA sequencing was profiled by HaploX (Shenzhen, China) (mouse wound tissue samples) or Genenergy (Shanghai, China) (HaCaT cell samples) using the Illumina platform. Raw reads were aligned to the reference genome, and gene-level raw read counts were generated for differential expression analysis. Fragments Per Kilobase of transcript per Million mapped reads (FPKM) values were calculated by StringTie to represent gene expression abundance and were transformed to log₂(FPKM + 1) for expression visualization. Differential expression testing was performed using DESeq2 based on raw read counts, with P-values adjusted via the Benjamini-Hochberg procedure. Differentially expressed genes were defined by |log₂ fold change| ≥ 1 and adjusted P < 0.05. Volcano plots were generated using DESeq2-derived differential statistics, and heatmaps were constructed using log₂-transformed FPKM values of differentially expressed genes. Pathway enrichment analysis was subsequently performed against the Reactome Pathway Database. In the validation study, we performed RT-PCR to validate the expression of mitochondrial genes in HaCaT cells under the PDM treatment for 24 h. Single-stranded RNA was reverse-transcribed to cDNA using a reverse transcription kit (TOYOBO, Japan) with gDNA removed. The PCR reactions were conducted in a master mixed system containing SYBR Green I (Roche, Switzerland), cDNA template (20 ng/well), and custom-designed primers (0.4 μM).

Mitochondrial DNA copy number detection

HaCaT cells were incubated with PDM or monensin A for the indicated time points. Genomic DNA was isolated using a commercial kit (QIAGEN, Netherlands)⁶⁶. Briefly, cells were lysed, and genomic DNA was purified by binding to the column, followed by washing and elution in the recommended buffer before qPCR analysis. Relative mtDNA copy number was quantified using nuclear DNA content as a standard by RT-PCR. The PCR reaction mix contained the DNA Master SYBR Green I (Roche), Genomic DNA (20 ng/well), and primers (0.4 μM) specifically targeting mtDNA or nuclear DNA. Primers were derived from a Human mtDNA Monitoring Primer Set (TaKaRa), which is composed of primers for mtDNA (ND1 and ND5) and nuclear DNA (SLCO2B1 and SERPINA1). The relative mtDNA copy number was determined by calculating the difference in the Ct values for the ND1-SLCO2B1 and ND5-SERPINA1 pair.

Seahorse-based oxygen consumption rate measurement

HaCaT cells seeded in 24-well XF microplates were used to perform the Seahorse XF assay. Cells were incubated at 37 °C in CO₂-free conditions for 1 h to allow them to settle before the assay. In the detection process, the dissolved oxygen in the cell supernatant was measured using a Seahorse XFe24 extracellular flux analyzer (Agilent, USA)⁶⁵. Briefly, during the standardized assay, oligomycin (1 μM, Selleck), FCCP (2 μM, Selleck), and rotenone (0.5 μM, Sigma Aldrich) plus antimycin A (0.5 μM, Selleck) were sequentially added to the cell supernatant after the basal respiration was measured. To test the oxygen consumption rate expedited by PDM, we replaced FCCP with PDM at the excitation step.

Statistics and reproducibility

No statistical method was used to predetermine sample size. No data were excluded from the analyzes. The experiments were not randomized. An unpaired Student’s t-test and one-way/two-way ANOVA with Tukey’s multiple-comparison test were used as indicated for comparisons between the groups. Except for animal studies and RNA-seq, experiments were performed using at least three independent biological replicates with similar results, as indicated in the figure legend. Multiple technical replicates were acquired within each biological replicate and averaged before statistical analysis. Unless otherwise stated, data were represented as mean ± SD (standard deviation). All P values are two-sided, and a P < 0.05 is considered statistically significant, and “ns” indicates no significance (P > 0.05). All analyzes were performed using GraphPad Prism software (v.9.5.0). The Investigators were not blinded to allocation during experiments and outcome assessment.

Reporting summary

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

Author contributions

C.F. performed the heterologous expression, gene inactivation and complementation, regulator overexpression, and biochemical experiments. Z.L. performed the compound isolation and structural determination experiments. L.B., Y. Li, and J.N. conducted compound activity evaluation and mechanistic studies. W.X., C.F., and L.M. performed bioinformatics analysis and genome mining of the pdm gene cluster. L.S. conducted the TDDFT-ECD calculations. W.Z. and Z.L. analyzed the NMR data and guided the structural determination. Y.Z. guided the increased production experiments of PDM. C.F., Z.L., and L.B. analyzed the data and wrote the manuscript. C.Z., L.W., Y. Luo, and Z.L. directed the research and revised the manuscript. C.F., Z.L., and L.B. contributed equally to this work.

Peer review

Peer review information

Nature Communications thanks Luoyi Wang, Nguan Soon Tan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This research was supported in part by the National Key Research and Development Program of China 2023YFA0914200 (C. Z.), 2024YFC2816000 (Y. Z.); the National Natural Science Foundation of China 82173067 (Y. Luo), 82372715 (Y. Luo); the China Postdoctoral Science Foundation GZC20250585 and 2025M770863 (C. F.); the Guangzhou Basic and Applied Basic Research Foundation No. 2024A04J3912 (Y. Li); the Youth Innovation Promotion Association of the Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences O6515502 (Y. Li).

Data availability

The data generated in this study are available within the article, Supplementary Information, and Source Data file. Additional figures, tables, characterization data for compounds, and supporting data are available in the Supplementary Information. The pdm BGC from S. marincola SCSIO 03032 has been deposited in GenBank under accession number PX380304. Raw RNAseq data have been deposited in the Gene Expression Omnibus (GEO) under accession numbers GSE315226. These are available in the online version of the paper. Annotated DNA plasmid sequences and numerical source data are provided in this paper. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77444-w.