Study on the Potential Molecular Mechanisms of Sodium Dehydroacetate (Na‐DHA) Interfering With Bone Metabolism and Inducing Osteoporosis Based on Network Toxicology, Molecular Docking, and In Vitro Experimental Validation

ABSTRACT

Sodium Dehydroacetate (Na‐DHA), a widely used food additive, has raised concerns about the chronic health risks associated with long‐term exposure. However, the potential impact of Na‐DHA on bone metabolism, its contribution to osteoporosis risk, and the specific molecular mechanisms remain unclear. This study aims to systematically elucidate the molecular mechanisms through which Na‐DHA induces osteoporosis by integrating network toxicology, molecular docking, and in vitro experiments. Potential targets of Na‐DHA were identified through multi‐database screening. Osteoporosis‐related genes were extracted from the GEO database (GSE156508) and subjected to differential and enrichment analyses. Common targets between Na‐DHA and osteoporosis were identified using a Venn diagram. A protein–protein interaction (PPI) network was constructed using STRING, and core targets were selected through random forest analysis. Molecular docking of core targets with Na‐DHA was performed using AutoDock Vina. Human bone marrow mesenchymal stem cells (hBMSCs) were used as a model to assess cell viability using the CCK‐8 assay, observe osteogenic/adipogenic differentiation phenotypes through Alizarin Red S and Oil Red O staining, and validate the expression of core targets and osteogenic genes by qRT‐PCR and Western blot. Multi‐database screening identified 325 potential Na‐DHA targets and 500 osteoporosis‐related differential genes. Of these, 34 common key targets were identified, which were mainly enriched in pathways related to lipid metabolism, autophagy, and steroid biosynthesis. Random forest analysis identified LCMT1, ARHGEF11, and VCAM1 as core targets, and molecular docking revealed potential binding interactions between Na‐DHA and all three targets. In vitro experiments demonstrated that 10 μM Na‐DHA significantly inhibited hBMSCs viability, reduced calcium deposition in Alizarin Red S staining, increased lipid droplet accumulation in Oil Red O staining, and downregulated the expression of key osteogenic genes (BGLAP, SP7, RUNX2, ALPL). Moreover, Na‐DHA significantly upregulated the mRNA and protein expression of LCMT1 and downregulated the mRNA and protein expression of ARHGEF11 and VCAM1. Na‐DHA may increase osteoporosis risk by upregulating LCMT1, downregulating ARHGEF11 and VCAM1, and disrupting lipid metabolism and the balance between osteogenesis and adipogenesis, ultimately disrupting bone metabolic homeostasis. This study provides scientific evidence for assessing the skeletal health risks of Na‐DHA and exploring potential intervention targets for bone toxicity.

The main workflow and core design concept of the current study.

1. Introduction

In recent years, the rapid development of the food industry has significantly increased the use and frequency of food additives, with long‐term low‐dose mixed exposures becoming a core public and academic concern due to the associated chronic health risks. Previous studies have confirmed that certain food additives can disrupt metabolic balance (Warner 2024), impair immune function (Rinninella et al. 2019), and interfere with organ development through subtle mechanisms, increasing the risk of chronic diseases such as obesity, cardiovascular diseases, and cancer. This has prompted a shift in toxicological assessments of food additives from “acute toxicity” to “long‐term organ‐specific damage” (Raposa et al. 2016; Zhang et al. ²⁰²⁵; Mafe and Büsselberg ²⁰²⁵).

Among various food additives, Sodium Dehydroacetate (Na‐DHA) is widely used in baked goods, dairy products, beverages, and sauces due to its broad antimicrobial activity (Jin et al. 2025). However, its high exposure rate and potential toxicity have raised concerns (Izawa et al. 2018). Monitoring data show that the detection rate of Na‐DHA in baked goods exceeds 30% in some regions (Mei‐Ling et al. ²⁰²²). Toxicological studies have confirmed that Na‐DHA has subchronic oral toxicity, acute toxicity, and can damage the cardiovascular system, liver, and kidney functions (Fang et al. 2022; Spencer et al. ¹⁹⁵⁰; Huang et al. ²⁰²¹; Du et al. ²⁰²³). However, current toxicological research on Na‐DHA mainly focuses on general toxicity and food safety limit evaluations, with no systematic investigation into its potential interference with bone metabolism and increased osteoporosis risk, a significant gap given the high population exposure levels (Zhang, Du, et al. ²⁰²⁴).

Osteoporosis, a prevalent metabolic bone disease, is characterized by an imbalance between bone resorption (dominated by osteoclasts) and bone formation (driven by osteoblasts), leading to bone mass loss, increased bone fragility, and heightened fracture risk (Lane et al. 2000; Fischer and Haffner‐Luntzer ²⁰²²). Traditionally, aging, estrogen deficiency, and poor lifestyle choices are recognized as the primary risk factors, but increasing evidence suggests that long‐term exposure to environmental chemicals, such as certain food additives and heavy metals, has become a significant external contributor (Snega Priya et al. 2023; Yu et al. ²⁰²³). These substances can directly inhibit osteoblast activity, enhance osteoclast function, or indirectly induce immune‐inflammatory responses that disrupt bone metabolic homeostasis, thereby exacerbating osteoporosis risk. This understanding provides a theoretical basis for exploring the association between everyday environmental exposures like Na‐DHA and bone metabolic diseases. Overall, Na‐DHA has the molecular characteristics of a potential bone metabolism‐disrupting chemical (BMDC), but there is a lack of systematic research and direct experimental evidence regarding whether it substantially disrupts bone metabolic balance and increases the risk of osteoporosis, as well as the specific molecular mechanisms involved.

To address this critical gap in understanding the mechanisms of Na‐DHA‐induced skeletal toxicity, this study employs an integrated research framework combining network toxicology, molecular docking, and in vitro experimental validation. As an interdisciplinary research strategy, network toxicology allows for the systematic identification of chemical‐gene‐pathway interactions by integrating toxicogenomics and disease‐related databases, overcoming the limitations of traditional single‐target studies (Zhang 2016; Wang et al. ²⁰²⁵; Cheng et al. ²⁰²⁶). In this study, we first used this approach to systematically screen osteoporosis‐related targets affected by Na‐DHA, then applied machine learning algorithms to identify core candidate genes. Molecular docking was subsequently performed to predict the binding modes of these core genes with DHA, identifying the key binding sites and binding affinities. This multi‐step computational biology strategy not only selects key targets for subsequent in vitro validation but also provides a theoretical foundation for understanding how Na‐DHA and similar environmental chemicals interfere with bone homeostasis.

Based on this, we used human bone marrow mesenchymal stem cells (hBMSCs) as a research model to verify the interference effects of Na‐DHA on bone metabolism through osteogenic/adipogenic differentiation experiments, qRT‐PCR, and Western blotting. The study aims to elucidate the molecular mechanism by which Na‐DHA induces osteoporosis. The results of this study will provide new perspectives for evaluating the skeletal health risks of Na‐DHA and lay the groundwork for identifying potential targets to mitigate its bone toxicity.

2. Materials and Methods

2.1. Prediction of Potential Targets for Na‐DHA

In this study, a multi‐database screening strategy was employed to identify potential targets for Na‐DHA, ensuring both comprehensive coverage and reliability. Specifically, in the CHEMBL database, “Dehydroacetic acid sodium salt” was used as the search keyword, with filters set to “pChEMBL ≥ 5.0” (indicating significant affinity between the compound and the target), “species = Homo sapiens ” and “target type = Protein”. This yielded Na‐DHA's active related targets (Zdrazil et al. 2024; Davies et al. ²⁰¹⁵). In the CTD database, “Dehydroacetic acid” was searched, and only “Directly associated” human gene targets were retained, excluding indirect associations (Wiegers et al. 2025). The PharmMapper database was used to upload Na‐DHA's 2D structure, with settings for “species =  Homo sapiens ” and “Top 300 potential targets” (Wang et al. 2017). In the SEA database, the SMILE file of Na‐DHA was uploaded, and human targets with a p‐value < 0.05 were filtered (Keiser et al. 2007). The SWISS Target Prediction database was used to upload Na‐DHA's structure, selecting human targets with a probability ≥ 50% (Gfeller et al. 2013; Daina and Zoete ²⁰²⁴; Daina et al. ²⁰¹⁹). In the TargetNet database, Na‐DHA's structure was uploaded, with “Include models with AUG ≥ 0.7” (indicating a prediction accuracy ≥ 70%) and “Fingerprint type = ECFP4” (using ECFP4 molecular fingerprint matching) (Yao et al. 2016). All targets obtained from these six databases were compiled in Excel, and duplicate entries were removed using the “Remove Duplicates” function to generate the final set of potential targets for Na‐DHA.

2.2. Osteoporosis Disease Targets and Enrichment Analysis

The GEO database dataset GSE156508 was downloaded, and differential expression analysis was conducted using the “limma” package in R, with the criteria “|log2FC| ≥ 1” and “adj. P < 0.05” to obtain differentially expressed genes (DEGs) related to osteoporosis. Metascape was then used for Gene Ontology (GO) functional enrichment analysis (including biological processes, cellular components, and molecular functions) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis.

2.3. Identification of Common Targets and Interaction Network Construction

Using Venny 2.1.0, the intersection between Na‐DHA targets and osteoporosis‐related targets was identified. The common target genes were then imported into the STRING database, with parameters set to “species = Homo sapiens ” and “Combined score ≥ 0.7” to construct a protein–protein interaction (PPI) network. GO and KEGG enrichment analysis of the common targets was performed using the SangBox3.0 database (http://sangerbox.com/login.html), with corrected p‐values < 0.05.

2.4. Machine Learning for Key Target Selection

Based on the expression data of the common target genes, a random forest algorithm on the HiPlot platform was used to rank the importance of 34 potential targets, identifying core genes involved in the process. Two measures were used to quantify the importance of each feature (gene). The first measure was Mean Decrease Accuracy, which quantifies the drop in classification accuracy after randomly permuting the expression values of each gene. A greater drop indicates a higher contribution of the gene to distinguishing the treatment and control groups. The second measure was Mean Decrease Gini, which measures the reduction in Gini impurity (a measure of node impurity in decision trees) due to each gene. Genes that reduce Gini impurity the most are considered key to splitting (distinguishing) the sample groups in the decision tree.

Cross‐validation was used to assess the generalization ability of the model and optimize feature selection. A “Feature Count–Cross‐validation Error” curve was plotted to observe the trend of model error as the number of features (genes) increased, identifying the subset of features that minimized error and preventing overfitting.

Finally, feature importance was visualized using scatter plots with “Mean Decrease Accuracy” and “Mean Decrease Gini” on the x‐axis and gene names on the y‐axis. The differential expression of key genes was displayed in violin plots.

2.5. Molecular Docking Validation

The 3D structure of Na‐DHA (CID: 23721629) was downloaded from the PubChem database and preprocessed using PyMOL 2.5.2 (adding hydrogens and minimizing energy). Core target protein structures were obtained from the PDB database, and water molecules and ligands were removed using PyMOL. Gasteiger charges were added using AutoDockTools. Docking parameters were set for semi‐flexible docking, with the grid box covering the protein's active site. AutoDock Vina was used to calculate binding energy (kcal/mol), and hydrogen bonds were visualized using PyMOL.

2.6. Cell Culture and Differentiation

hBMSCs were purchased from Yaji Biotechnology Co. Ltd. and sourced from healthy donors. Flow cytometry confirmed that the hBMSCs expressed CD105, CD44, and CD29 markers (> 70%) and lacked CD45 and CD34 markers (< 5%). The hBMSCs were cultured in complete growth medium consisting of L‐DMEM, 10% FBS, and 1% penicillin–streptomycin and expanded to the 6th passage for experimentation.

For osteogenesis/adipogenesis experiments, BMSCs were seeded in 12‐well plates at a density of 5 × 10⁴ cells/well and cultured until 80%–90% confluence. To assess the impact of Na‐DHA on BMSC lineage commitment, cells were exposed to 10 μM Na‐DHA during the entire differentiation period. The medium was replaced with Na‐DHA‐containing medium every 3 days to maintain the treatment condition. Solvent‐treated control groups were included in all experiments.

In osteogenic differentiation, BMSCs were cultured in osteogenic differentiation medium containing 10% FBS, 10 mM β‐glycerophosphate, 0.1 μM dexamethasone, and 50 μg/mL ascorbic acid (Sigma, USA) for 14 days, with medium changes every 3 days.

In adipogenic differentiation, BMSCs were cultured in adipogenic differentiation medium containing 10% FBS, 0.5 mM IBMX, 0.5 μM dexamethasone, and 0.2 mM insulin for 14 days.

2.7. Cell Viability Assay (CCK‐8 Method)

hBMSCs were seeded at 5 × 10³ cells/well in 96‐well plates. After attachment, cells were treated with 25, 50, 100, and 200 μM Na‐DHA for 72 h. CCK‐8 reagent (10 μL) was added to each well and incubated for 2 h at 37°C. Absorbance was measured at 450 nm using a microplate reader.

2.8. Alizarin Red Staining

After 14 days of osteogenic differentiation, cells were fixed with 4% paraformaldehyde for 30 min, washed with distilled water, and stained with 2% Alizarin Red S solution (pH 8.3) at room temperature for 30 min. Excess dye was removed by PBS wash, and images were taken under an inverted microscope.

2.9. Oil Red O Staining (Adipogenesis Differentiation)

After 14 days of adipogenic differentiation, cells were fixed with 4% paraformaldehyde for 30 min, incubated with 60% isopropanol for 5 min, and stained with Oil Red O solution (0.5% in isopropanol) at room temperature for 30 min. After washing with 60% isopropanol, images were captured under an inverted microscope.

2.10. Quantitative Real‐Time Polymerase Chain Reaction (qRT‐PCR)

RNA was extracted with TriQuick reagent after 7 days of induction. RNA concentration and purity were determined using Nanodrop 2000 (A260/A280 = 1.8–2.0). 1 μg of RNA was reverse transcribed into cDNA using the SureScript kit (reaction conditions: 42°C for 30 min, 85°C for 5 min). qPCR was performed with 2 × SYBR Premix Ex Taq, using a 20 μL reaction mixture: 10 μL SYBR Mix, 0.4 μL forward primer, 0.4 μL reverse primer, 2 μL cDNA, and 7.2 μL ddH2O. Reaction conditions: 95°C for 30 s, 95°C for 5 s, 60°C for 30 s, 40 cycles. The specificity of the primers was verified by melt curve analysis. The primer sequences for RUNX2, ALPL, BGLAP, SP7, LCMT1, ARHGEF11, VCAM1 and ACTB are listed in Table 1. Gene expression was quantified using the 2^−ΔΔCt method.

TABLE 1.

Primer sequences of genes for qRT‐PCR detection in human bone marrow mesenchymal stem cells (hBMSCs).

Gene name	Primer type	Primer sequence (5′ → 3′)
RUNX2	Forward	5′‐CGCCTCACAAACAACCACAG‐3′
	Reverse	5′‐TCACTGTGCTGAAGAGGCTG‐3′
ALPL	Forward	5′‐CCGTGACAGATGCCAACTTC‐3′
	Reverse	5′‐GGCGGCAGACTTTGGTTTC‐3′
BGLAP	Forward	5′‐AGGGCAGCGAGGTAGTGAAGAG‐3′
	Reverse	5′‐GCCGATGTGGTCAGCCAACTC‐3′
SP7	Forward	5′‐AGGTAGAGTGAGCTGGGGAG‐3′
	Reverse	5′‐TGCACAGAGTAGGCACTCAA‐3′
LCMT1	Forward	5′‐GCCACTAGGCAGAGGGAATC‐3′
	Reverse	5′‐AAACCCAGGTGACTGCAACA‐3′
ARHGEF11	Forward	5′‐CTCCTGCCTGCGAGATTAGG‐3′
	Reverse	5′‐TCCACACCAGACTCCTGGAA‐3′
VCAM1	Forward	5′‐GGACCACATCTACGCTGACA‐3′
	Reverse	5′‐TTGACTGTGATCGGCTTCCC‐3′
ACTB	Forward	5′‐AGCATCCCCCAAAGTTCACAA‐3′
	Reverse	5′‐TGGGGTGGCTTTTAGGATGG‐3′

2.11. Western Blot Analysis

Protein extraction was performed using RIPA buffer (containing 1% PMSF) after 14 days of induction. Protein concentration was measured using the BCA method, and samples were adjusted to the same concentration, mixed with 5× sample buffer, and denatured at 95°C for 5 min. SDS‐PAGE was performed, followed by transfer to PVDF membranes. After blocking the membrane with 5% blotting grade at room temperature for 2 h, the following primary antibodies were added and incubated overnight at 4°C: rabbit anti‐human LCMT1 polyclonal antibody (catalog number: ab169337, Abcam), rabbit anti‐human ARHGEF11 polyclonal antibody (catalog number: ab110059, Abcam), rabbit anti‐human VCAM1 polyclonal antibody (catalog number: ab134047, Abcam), and mouse anti‐human β‐actin monoclonal antibody (catalog number: YA823, MCE). All primary antibodies were diluted at 1:1000. After washing with TBST, HRP‐conjugated goat anti‐rabbit IgG secondary antibody (catalog number: A0208, Beyotime) or goat anti‐mouse IgG secondary antibody (catalog number: A0350, Beyotime) was added and incubated at room temperature for 1.5 h with a dilution of 1:2000.

The protein bands were visualized using ECL, and band intensity was quantified using ImageJ software, with β‐actin as the loading control.

2.12. Statistical Analysis

All experimental data are presented as “mean ± standard deviation (x±s)”. Statistical analysis was performed using PRISM 9.5.1. One‐way analysis of variance (ANOVA) was used for comparisons between multiple groups, and Tukey's post hoc test was applied for pairwise comparisons. Independent t‐tests were used for comparisons between two groups. p < 0.05 was considered statistically significant. GraphPad Prism 9.5.1 was used to create statistical graphs.

3. Results

3.1. Multi‐Database Combined Screening of Na‐DHA Potential Targets and Protein–Protein Interaction Network Analysis

To systematically identify the potential targets of Na‐DHA and clarify the interactions among these targets, a multi‐database screening strategy was employed. Databases including CHEMBL, CTD, PharmMapper (PM), SEA, SWISSTargetPrediction, and Targetnet were used. The CHEMBL database identified 73 targets, Targetnet 77, SWISS 75, PharmMapper 90, CTD 8, and SEA 2 potential targets (Figure 1A). Subsequently, the STRING database was utilized to construct a PPI network for these potential targets. The results revealed dense interaction lines among the targets, indicating that Na‐DHA's potential targets are broadly interconnected at the molecular level (Figure 1B).

FIGURE 1.

Multi‐database screening and protein–protein interaction (PPI) network analysis of potential targets for sodium dehydroacetate (Na‐DHA). (A) Distribution of potential Na‐DHA targets identified through combined multi‐database screening (CHEMBL, CTD, PharmMapper (PM), SEA, SWISSTargetPrediction, Targetnet databases). (B) PPI network constructed based on the STRING database for Na‐DHA's potential targets. Nodes represent target proteins, and edges represent the interactions between proteins.

3.2. Differential Gene Analysis and Enrichment of Osteoporosis

To identify differential genes associated with osteoporosis and analyze their functional and pathway characteristics, we extracted gene expression data related to osteoporosis from the GEO database and performed differential analysis. Enrichment analysis was conducted using Metascape. The volcano plot results revealed a large number of DEGs (Figure 2A). The heatmap in Figure 2B visually presents the expression pattern differences of these genes between the two sample groups. The enrichment network shows that these differential genes are involved in core biological processes, including “ribosomal complex biogenesis”, “DNA metabolic processes”, as well as “lipid metabolism”, “skeletal system development”, and “autophagy” (Figure 2C). Notably, disruption in lipid metabolism can disturb the differentiation balance between osteoblasts and adipocytes, which is a significant factor contributing to the “insufficient osteogenesis and fat infiltration” phenotype seen in osteoporosis.

FIGURE 2.

Differential gene analysis and enrichment results for osteoporosis. (A) Volcano plot of differential genes, where red triangles represent significantly upregulated genes, blue inverted triangles represent significantly downregulated genes, and black circles represent genes with no significant differential expression (the x‐axis represents log₂ (FoldChange), the y‐axis represents −log₁₀ (p‐value), and the dashed line represents the significance threshold). (B) Heatmap of differential gene expression, where red indicates high expression and purple indicates low expression. Columns correspond to samples, and rows correspond to differential genes. (C) Metascape enrichment network for differential genes. Nodes of different colors represent different functional enrichment modules, and the lines between nodes reflect the relationships between the modules. The legend on the right labels the biological processes or pathways involved in each module, including lipid metabolism, skeletal system development, and autophagy.

3.3. Common Targets of Na‐DHA and Osteoporosis and Functional Enrichment

To identify common targets between Na‐DHA and osteoporosis and analyze their functional and pathway characteristics, a Venn diagram was used to find the intersection between Na‐DHA targets and the differential genes associated with osteoporosis. GO and KEGG enrichment analysis of the intersecting genes was conducted. The results, shown in Figure 3A, reveal that Na‐DHA and osteoporosis share 34 common targets. GO enrichment analysis for these common targets (Figure 3B) highlighted significant enrichment in biological processes such as “response to oxygen‐containing compounds” and “drug metabolic processes”, while cellular component enrichment (Figure 3C) focused on “cytoplasm” and “peroxisomal membranes”. Molecular function enrichment (Figure 3D) revealed “catalytic activity” and “cation binding”, and KEGG pathway analysis (Figure 3E) pointed to pathways like “autophagy” and “steroid biosynthesis”, suggesting that Na‐DHA may affect bone metabolism by interfering with these processes.

FIGURE 3.

Common targets between Na‐DHA and osteoporosis and functional enrichment analysis. (A) Venn diagram and localized protein–protein interaction (PPI) network for common targets. The purple circle represents osteoporosis‐related differential genes, the blue circle represents Na‐DHA potential targets, and the intersection represents the common targets. The inset shows the interaction relationships between the common targets, where nodes represent target proteins, and the lines between nodes represent PPIs. (B) Gene Ontology (GO) biological process enrichment bubble chart. The x‐axis represents the gene ratio (GeneRatio), the y‐axis represents the enriched biological process terms, bubble size corresponds to gene count (Count), and bubble color represents −log₁₀ (p‐value). (C) GO cellular component enrichment bubble chart, showing the intracellular distribution features of the common targets. The parameters are the same as in (B). (D) GO molecular function enrichment bubble chart, presenting the molecular function characteristics of the common targets. The parameters are the same as in (B). (E) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment bubble chart. The x‐axis represents −log₁₀ (p‐value), the y‐axis represents the enriched pathway terms, bubble size corresponds to gene count (Count), and bubble color represents the pathway category.

3.4. Random Forest Screening of Core Targets and Expression Validation

To identify the core targets through which Na‐DHA interferes with bone metabolism, random forest analysis was conducted on the 34 common targets identified from the network toxicology results. The analysis ranked genes based on “model prediction contribution” and “node classification efficacy”. Violin plots were used to validate the differential expression of these core targets between disease and control samples. The random forest analysis revealed that, in the Mean Decrease Accuracy evaluation (reflecting each gene's contribution to model prediction accuracy), genes such as LCMT1, ARHGEF11, TDP2, GRWD1, CAT, and VCAM1 had higher scores than others (Figure 4A). In the Mean Decrease Gini evaluation (measuring the reduction in Gini impurity), LCMT1, CAT, ARHGEF11, VCAM1, SEPSECS, and ANTXP2 ranked the highest, with LCMT1 ranking first in both dimensions, suggesting its pivotal role in regulating Na‐DHA‐induced osteoporosis (Figure 4B). Cross‐validation (Figure 4C) demonstrated that as the number of features (genes) increased, the model's error decreased, indicating that the core gene subset could effectively improve the model's generalizability. Furthermore, the Venn diagram (Figure 4D) showed that LCMT1, ARHGEF11, VCAM1, and CAT were common targets in both evaluation methods. Violin plot analysis (Figure 4E) indicated that LCMT1 expression in the osteoporosis group (red) was significantly higher than in the control group (blue), while ARHGEF11 and VCAM1 showed significantly lower expression in the osteoporosis group, with statistical significance (adj. p < 0.05).

FIGURE 4.

Random forest selection of potential core targets for sodium dehydroacetate (Na‐DHA) interference in bone metabolism and expression validation. (A) Gene importance ranking based on “Mean Decrease Accuracy”. The x‐axis represents the mean decrease accuracy value, and the y‐axis represents the genes. Larger values indicate a greater contribution of the gene to distinguishing between the “Na‐DHA treatment group” and the “control group” in the model. (B) Gene importance ranking based on “Mean Decrease Gini”. The x‐axis represents the mean decrease Gini value, and the y‐axis represents the genes. Larger values indicate the critical role of the gene in decision tree node splitting (differentiating sample groups). (C) Cross‐validation error curve of the random forest model. The x‐axis represents the number of features (genes) included, and the y‐axis represents the cross‐validation error. The curve decreases as the number of features increases, suggesting that the core target subset improves the model's generalization ability. (D) Venn diagram showing the intersection of core targets identified by the “Mean Decrease Accuracy” and “Mean Decrease Gini” methods. The intersecting genes (LCMT1, ARHGEF11, CAT, VCAM1) are the key targets identified by both methods. (E) Violin plot showing the expression distribution differences of core targets between the osteoporosis‐related group (red, Osteoblast group) and the control group (blue, Osteoarthritis group). The y‐axis represents gene expression levels, and the distribution and dispersion reflect the statistical significance of expression differences between groups (adj p < 0.05).

3.5. Molecular Docking of Core Targets With Na‐DHA

Based on the results above, LCMT1, ARHGEF11, and VCAM1 were identified as potential core targets for Na‐DHA‐induced osteoporosis. Molecular docking was conducted to investigate the binding potential and interaction modes between these targets and Na‐DHA. For LCMT1 (Figure 5A), the binding energy of Na‐DHA and LCMT1 was −5.0 kcal/mol. The docking model revealed that Na‐DHA formed hydrogen bonds with the Lys37 and Arg73 residues of LCMT1, with hydrogen bond distances of approximately 2.2 Å and 2.4 Å, respectively. For ARHGEF11 (Figure 5B), the binding energy was −4.8 kcal/mol, and Na‐DHA formed a hydrogen bond with the Ser123 residue, with a hydrogen bond distance of about 2.0 Å. For VCAM1 (Figure 5C), the binding energy was −4.4 kcal/mol, and Na‐DHA formed a hydrogen bond with the Ile68 residue, with a hydrogen bond distance of approximately 2.3 Å.

FIGURE 5.

Molecular docking results of core targets with Na‐DHA. (A) Molecular docking between LCMT1 and Na‐DHA. The left side shows the overall structure of LCMT1 protein (surface transparency displayed), and the right side shows an enlarged view of the binding region. Na‐DHA (cyan) forms hydrogen bonds with the Lys37 and Arg73 residues of LCMT1 (yellow dashed lines indicate hydrogen bond distances). (B) Molecular docking between ARHGEF11 and Na‐DHA. The left side shows the overall structure of ARHGEF11 protein, and the right side shows an enlarged view of the binding region. Na‐DHA (green) forms a hydrogen bond with the Ser123 residue of ARHGEF11. (C) Molecular docking between VCAM1 and Na‐DHA. The left side shows the overall structure of VCAM1 protein, and the right side shows an enlarged view of the binding region. Na‐DHA (yellow) forms a hydrogen bond with the Ile68 residue of VCAM1.

3.6. Effects of Na‐DHA on hBMSCs Viability and Differentiation Phenotypes

To assess the cytotoxic effects of Na‐DHA, hBMSCs were treated with various concentrations of Na‐DHA. The results showed that both 10 μM and 100 μM Na‐DHA significantly reduced cell viability (Figure 6A). To minimize cytotoxic interference while maintaining observable biological effects, 10 μM Na‐DHA was used for subsequent experiments. Further analysis of hBMSC differentiation potential revealed reduced calcium deposition in Alizarin Red S staining (Figure 6B) in the Na‐DHA exposure group. In contrast, adipogenic differentiation was enhanced, as shown by increased intracellular lipid droplet accumulation in Oil Red O staining (Figure 6C). qRT‐PCR results (Figure 6D) showed that osteogenic markers (BGLAP, SP7, RUNX2, ALPL) were significantly downregulated following Na‐DHA treatment (p < 0.05).

FIGURE 6.

Effects of Na‐DHA on human bone marrow mesenchymal stem cells (hBMSCs) viability and differentiation phenotypes. (A) Cell viability of hBMSCs after treatment with different concentrations of Na‐DHA. (B) Alizarin red S staining results for osteogenic differentiation (×200). Red indicates calcium deposition nodules, and the Na‐DHA treatment group shows fewer calcium nodules compared to the control group. (C) Oil red O staining results for adipogenic differentiation (×200). Red indicates intracellular lipid droplets, and the Na‐DHA treatment group shows more lipid droplets compared to the control group. (D) qRT‐PCR results for key osteogenic genes. “*”, “**”, “***”, and “#” represent p < 0.05, p < 0.01, p < 0.001, and p < 0.0001, respectively, compared to the control group.

3.7. Effects of Na‐DHA on Potential Core Targets

qRT‐PCR results (Figure 7A) further demonstrated that Na‐DHA treatment significantly upregulated LCMT1 expression (*p < 0.05), while osteogenic regulatory genes such as ARHGEF11 showed a significant downregulation (**p < 0.01). Additionally, VCAM1, a key gene involved in cell adhesion and migration, exhibited a marked decrease in mRNA expression (***p < 0.001). Western blot experiments (Figure 7B) confirmed these findings at the protein level, with LCMT1 protein expression significantly higher in the Na‐DHA treatment group (**p < 0.01), while ARHGEF11 and VCAM1 protein levels were significantly reduced (ARHGEF11: **p < 0.01; VCAM1: *p < 0.001).

FIGURE 7.

Effects of sodium dehydroacetate (Na‐DHA) on the expression of potential core targets. (A) qRT‐PCR analysis of relative mRNA expression levels of core targets. (B) Western blot analysis of protein expression levels of core targets (left: Protein bands, right: Quantification bar chart). β‐Actin was used as the internal control. *p < 0.05, **p < 0.01, ***p < 0.001 vs. control group.

4. Discussion

This study employed an integrated strategy combining network toxicology, molecular docking, and in vitro experiments to systematically analyze the molecular mechanisms by which Na‐DHA interferes with bone metabolism. Network toxicology screening identified common targets between Na‐DHA and osteoporosis, with enrichment analysis pointing to metabolic‐related pathways. The random forest algorithm identified LCMT1, ARHGEF11, and VCAM1 as potential core targets, which were further validated through molecular docking. In vitro experiments confirmed that Na‐DHA inhibited osteogenic differentiation and promoted adipogenic differentiation in hBMSCs and downregulated the expression of key osteogenic genes.

The enrichment analysis in this study revealed significant enrichment in lipid metabolism‐related pathways such as “steroid biosynthesis” and “propionate metabolism”, which correlate with the enhanced adipogenic differentiation phenotype in hBMSCs exposed to Na‐DHA. Bone metabolism and lipid metabolism are tightly interregulated: osteoblasts and adipocytes both originate from hBMSCs, and their differentiation follows a “one reduces as the other increases” balance (transdifferentiation phenomenon) (Xiao et al. 2024). Additionally, lipid metabolism provides essential energy and signaling molecules for osteoblast function (Alekos et al. 2020). Steroids (cholesterol derivatives) are involved in the synthesis of osteogenic hormones, and propionate metabolism affects fatty acid oxidation, which fuels osteogenic differentiation. By interfering with these metabolic pathways, Na‐DHA may disrupt the balance between osteogenesis and adipogenesis. This disruption is consistent with the phenotypes observed in Oil Red O staining (increased lipid droplets) and Alizarin Red staining (decreased calcium nodules), suggesting that “metabolic reprogramming” is a key mechanism through which Na‐DHA interferes with bone metabolism, offering new perspectives on the metabolic regulation of bone homeostasis by food additives. Notably, the observed phenotype of Na‐DHA promoting adipogenic differentiation while inhibiting osteogenic differentiation in hBMSCs is not an isolated case. Previous studies have shown that parabens (methyl‐ and butyl‐paraben), commonly used as food and personal care preservatives, can also significantly influence the differentiation fate of multipotent stem cells (C3H10T1/2). Parabens promote adipogenesis through activation of the PPARγ pathway while simultaneously inhibiting osteogenic and chondrogenic differentiation (Hu et al. 2017). This suggests that synthetic chemical preservatives may share a common biological effect of “disrupting stem cell differentiation balance”, rather than this effect being specific to a particular class of preservatives. It also helps explain why various types of chemical preservatives may increase the risk of osteoporosis. Notably, parabens have been detected in human placentas and breast milk (Darbre et al. 2004; González‐Palacios et al. ²⁰²⁵; Castillero‐Rosales et al. ²⁰²⁴; Ji et al. ²⁰²⁴), and Na‐DHA is frequently found in food products. This indicates that prolonged exposure to such preservatives could have a lasting impact on bone health at different stages of life (e.g., early development and adulthood). Future research should incorporate population exposure data to conduct more comprehensive risk assessments.

The core targets identified by random forest analysis—LCMT1, ARHGEF11, and VCAM1—are likely involved in Na‐DHA‐induced bone metabolic disorder, playing crucial roles in bone metabolism. Specifically, LCMT1, as a leucine carboxyl methyltransferase, catalyzes the C‐terminal methylation of the catalytic subunit of protein phosphatase 2A (PP2Ac). This post‐translational modification is a critical regulatory node in maintaining PP2A activity and plays a central “molecular switch” role in metabolic disorders induced by various environmental chemicals. For instance, in studies of benzo (a)pyrene (BaP)‐induced liver fat deposition, low‐dose BaP exposure significantly upregulated LCMT1 expression, and LCMT1‐mediated methylation of PP2Ac stabilized the interaction between PP2A's catalytic and regulatory subunits. This stabilized PP2A's dephosphorylation activity on autophagy‐related substrates (e.g., Atg5, mTOR), resulting in disrupted autophagy and exacerbated hepatic lipid accumulation (Li et al. 2023). In bone metabolism, LCMT1‐PP2Ac methylation has been shown to influence BM‐MSC differentiation. When BM‐MSCs differentiate into adipocytes, PP2Ac methylation levels decrease significantly, and inhibition of PP2A demethylase increased PP2Ac methylation, enhancing adipogenesis (Ikeda et al. 2020). In this study, Na‐DHA treatment significantly upregulated LCMT1 expression in hBMSCs, supporting the conclusion that Na‐DHA may enhance LCMT1‐mediated methylation of PP2Ac, increasing PP2Ac methylation levels and promoting adipogenesis over osteogenesis. This aligns with the phenotypic changes observed in Oil Red O staining (increased lipid droplets) and Alizarin Red staining (decreased calcium nodules) and corroborates the enrichment of lipid metabolism pathways such as “steroid biosynthesis” and “propionate metabolism”, highlighting the disruption of the osteogenesis‐adipogenesis balance by Na‐DHA. It is noteworthy that LCMT1 plays a common regulatory role in chemical‐induced metabolic disorders. In a study by Li et al. (Li et al. 2023) they found that low‐dose benzo [a]pyrene (BaP) induced liver lipid deposition by inhibiting autophagy through the LCMT1/PP2Ac pathway, ultimately promoting lipid accumulation in hepatocytes. This mechanism mirrors the way Na‐DHA interferes with lipid metabolism and promotes adipogenic differentiation of hBMSCs through LCMT1 in our study. These findings suggest that LCMT1 is a common target for various exogenous chemicals that disrupt metabolic homeostasis. The pathways regulated by LCMT1, including PP2Ac methylation, autophagy, and lipid metabolism, may constitute a central molecular axis in the metabolic disorders induced by chemicals in different tissues (e.g., bone and liver).

ARHGEF11, a key regulator of the Rho GTPase superfamily, catalyzes the activation of Rho family GTPases (such as RhoA, Cdc42, Rac1), which regulate critical cellular processes like cytoskeletal rearrangement, polarity, migration, adhesion, and differentiation (Bjornson and Cahill 2025; Bjornson et al. ²⁰²⁵; Du et al. ²⁰²⁰). Although direct studies on ARHGEF11 in bone metabolism are limited, its molecular function is closely associated with osteoblast physiology. Osteoblasts migrate from the bone marrow to sites of bone formation, forming polarized shapes to secrete bone matrix (e.g., type I collagen) and maintaining intercellular signaling during bone matrix mineralization (Thiel et al. 2018). These processes depend on Rho GTPase‐mediated cytoskeletal dynamics. Existing studies have confirmed that Rgnef, a member of the RhoGEF family, directly regulates osteoblast differentiation and mineralization by activating the RhoA/Rac1 signaling pathway (Lee et al. 2026). Thus, RhoGEF family proteins (including ARHGEF11 and Rgnef), as activators of Rho GTPases, likely play important roles in the regulation of osteoblast function. In this study, Na‐DHA appears to exacerbate bone metabolic imbalance by inhibiting ARHGEF11, providing theoretical evidence for its role in osteoporosis. In this study, Na‐DHA appears to exacerbate bone metabolic imbalance by inhibiting ARHGEF11, providing theoretical evidence for its role in osteoporosis.

VCAM1, a key adhesion molecule, plays a critical role in maintaining bone metabolic balance. In OVX‐induced osteoporosis models, VCAM1 acts as a “molecular anchor” that supports the migration of BM‐MSCs (the same lineage as hBMSCs in this study) to bone formation sites by mediating their adhesion to bone marrow stromal and endothelial cells, thereby ensuring the accurate homing of osteoprogenitor cells for bone remodeling and repair (Teng et al. 2018). In this study, molecular docking experiments showed that Na‐DHA likely binds to VCAM1, which may alter VCAM1's spatial conformation (e.g., disrupting its adhesion domain) or hinder its interaction with downstream ligands (e.g., integrin α4β1), directly inhibiting VCAM1's biological activity (Scholten et al. 2025). In addition, in vitro cell experiments confirmed a significant downregulation of VCAM1 expression in Na‐DHA‐treated hBMSCs. This “direct binding inhibition and expression downregulation” dual effect suggests that Na‐DHA may weaken VCAM1 function in both “activity” and “abundance” dimensions. Together, LCMT1, ARHGEF11, and VCAM1 form a multi‐node mechanism network through which Na‐DHA interferes with bone metabolism, influencing stem cell differentiation, migration, and cellular metabolic support. This network leads to the imbalance of bone homeostasis, providing a clearer molecular understanding of Na‐DHA's interference in bone metabolism.

In the in vitro experiments, Na‐DHA demonstrated a concentration‐dependent inhibition of hBMSC viability, as well as a shift from osteogenesis to adipogenesis, clearly indicating its characteristics as a “Bone Metabolism Disrupting Chemical (BMDC)”. The disruption of the osteogenesis‐adipogenesis balance is a key pathological basis for osteoporosis. When hBMSCs differentiate more into adipocytes, the supply and function of osteoblasts are weakened, leading to bone loss. Given the high detection rate of Na‐DHA in food, this study suggests that long‐term exposure to Na‐DHA may increase the risk of osteoporosis by disturbing the differentiation balance of hBMSCs. This finding provides experimental evidence for assessing the skeletal health risks of food additives and lays the foundation for future in vivo studies and epidemiological investigations. Consistent with the results observed in our in vitro experiments, where Na‐DHA inhibited osteogenic differentiation of hBMSCs, a study by Zhang et al. further confirmed the skeletal toxicity of Na‐DHA in vivo in broiler chickens. Their research found that supplementing the feed with Na‐DHA reduced bone index, disrupted the microstructure of trabecular bone, and disturbed the balance between osteoblast and osteoclast activity (Zhang, Du, et al. ²⁰²⁴). This is highly consistent with the phenotypic changes observed in our study, including decreased calcium deposition and enhanced adipogenic differentiation in hBMSCs. This cross‐species consistency strongly supports the conclusion that Na‐DHA can exert broad‐spectrum adverse effects on bone health by interfering with key aspects of bone metabolism. It is important to note that the skeletal toxicity of Na‐DHA is not an isolated effect. Previous studies have also confirmed its ability to induce liver and kidney damage, reproductive toxicity, and coagulation dysfunction (Zhou et al. 2023; Zhang, Zhang, et al. ²⁰²⁴; Yang et al. ²⁰²⁶; Li et al. ²⁰²⁵). This indicates that Na‐DHA is a multi‐organ toxicant, with bone metabolism disruption likely representing one of its systemic effects. The connection between these systemic toxicities underscores the need for future safety assessments of Na‐DHA to consider the long‐term health risks across multiple organs, including the bones, liver, and kidneys.

From the perspective of nutrition and food science, this study is the first to reveal the disruptive effects of Na‐DHA on bone metabolism and its potential molecular mechanisms through network toxicology. It adds valuable data to the research on the chronic toxicity profile of this food additive, providing new experimental support for reassessing the safety of Na‐DHA consumption. Our findings confirm that Na‐DHA significantly disrupts the osteogenic‐adipogenic differentiation balance of hBMSCs, thereby disturbing bone metabolism homeostasis. This suggests that the long‐term consumption of foods containing Na‐DHA, particularly in baked goods where Na‐DHA is widely used, may serve as a potential risk factor for osteoporosis, especially for high‐risk groups such as the elderly and postmenopausal women. In terms of regulatory standards for Na‐DHA use in the food industry, we recommend that relevant regulatory authorities further refine exposure limits. For instance, they could use population dietary intake survey data to establish more targeted maximum allowable concentrations of Na‐DHA in food products. We also encourage both industry and researchers to actively develop natural preservatives as alternatives to Na‐DHA, such as plant‐derived antimicrobial peptides or lactic acid bacterial fermentation products. These alternatives could help preserve food while reducing the health risks associated with chemical preservatives. Additionally, this study offers new insights into nutritional intervention strategies. Future research could explore dietary approaches (e.g., increasing calcium, vitamin D, and omega‐3 fatty acid intake) to mitigate Na‐DHA‐induced bone damage, providing scientific support for balancing the development of the baking and other food industries with public bone health. In summary, this study offers a comprehensive analysis of the molecular mechanisms through which Na‐DHA interferes with bone metabolism, from network prediction to molecular validation and cellular phenotyping. It provides new scientific insights into the chronic health risks of food additives and environmental factors contributing to osteoporosis.

In summary, this study offers a comprehensive analysis of the molecular mechanisms through which Na‐DHA interferes with bone metabolism, from network prediction to molecular validation and cellular phenotyping. It provides new scientific insights into the chronic health risks of food additives and environmental factors contributing to osteoporosis.

However, it must be acknowledged that this study has certain limitations. First, we only validated the effects of Na‐DHA on bone metabolism using in vitro hBMSCs, lacking in vivo animal studies and clinical epidemiological evidence to support a causal relationship between Na‐DHA exposure and osteoporosis. Second, the specific mechanisms by which the core targets (LCMT1, ARHGEF11, VCAM1) regulate bone metabolism remain incompletely understood, and further investigation is needed into the downstream signaling pathways of these targets. Third, this study only examined a single concentration of Na‐DHA (10 μM) for functional validation; the dose‐dependent effects of Na‐DHA on bone metabolism need to be explored with a broader concentration gradient.

To address these limitations, future research could explore multiple avenues: in vivo animal models, such as osteoporotic mice or zebrafish, could be used to evaluate the long‐term effects of Na‐DHA exposure on bone mass, microstructure, and serum bone metabolism markers. Core target functions in vivo could be validated through knockout or overexpression strategies. Further studies should also investigate the downstream signaling pathways through which LCMT1, ARHGEF11, and VCAM1 regulate osteogenic‐adipogenic differentiation in hBMSCs and elucidate their interactions with lipid metabolism pathways. Epidemiological studies could analyze the correlation between Na‐DHA intake in the diet and the incidence of osteoporosis in human populations. Additionally, screening for small‐molecule inhibitors that block the binding of Na‐DHA to core targets could provide theoretical support for preventing and treating Na‐DHA‐induced bone damage.

5. Conclusion

In conclusion, this study employed an integrated approach combining network toxicology, molecular docking, and in vitro experiments to investigate the molecular mechanisms by which Na‐DHA, a widely used food additive, disrupts bone metabolism and induces osteoporosis.

Bioinformatics analysis identified 34 common targets between Na‐DHA and osteoporosis, primarily enriched in lipid metabolism, autophagy, and steroid biosynthesis pathways. LCMT1, ARHGEF11, and VCAM1 were identified as core regulatory targets through random forest analysis and validated by molecular docking, which confirmed stable binding interactions between Na‐DHA and these targets. In vitro experiments with hBMSCs revealed that 10 μM Na‐DHA inhibited cell viability, suppressed osteogenic differentiation, and promoted adipogenic differentiation. Additionally, Na‐DHA treatment upregulated LCMT1 expression while downregulating ARHGEF11 and VCAM1.

Overall, our findings suggest that Na‐DHA disrupts the balance between osteogenesis and adipogenesis in hBMSCs by regulating LCMT1, ARHGEF11, and VCAM1 expression, thereby impairing bone metabolic homeostasis and increasing osteoporosis risk. This study provides valuable experimental evidence for assessing the skeletal health risks of Na‐DHA exposure and highlights the need for further investigation into the long‐term toxicity of food additives, as well as the mechanisms underlying environmental chemical‐induced bone diseases.

Author Contributions

Weihong Qian: conceptualization, writing – original draft, methodology, investigation, funding acquisition. Qingqing Bao: investigation, data analysis. Xiaoqing Tang: investigation, statistical analysis. Wuchao Lu: investigation, data collection. Jiaxin Huang: investigation, literature review. Jiapeng Bao: writing – review editing. Zhihong Yao: writing – review editing, supervision.

Funding

This research was supported by Zhejiang Province Traditional Chinese Medicine Science and Technology Plan Project (2014ZL1110) and Zhejiang Province Medicine and Health Science and Technology Plan Project (2014XL179).

Ethics Statement

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The GEO public datasets used in this study and available in the data repository have the following accession number: GSE156508 microarray data.