An Oxidative Stress Biomarker for Noise-Associated Stroke: Evidence from Human and Mice

Abstract

Background:

Occupational noise exposure is a recognized risk factor for stroke, yet biomarkers specific to noise-associated stroke (NAS) remain unclear. Oxidative stress (OS) plays a key mechanistic role in noise-induced cerebrovascular damage. We aim to identify OS biomarkers of NAS and conduct risk assessment.

Methods:

Candidate OS biomarkers were screened using Gene Expression Omnibus transcriptomic datasets integrated with eight machine-learning approaches, then verified in UK Biobank longitudinal data and in stroke model mice after 4-week noise exposure. In addition, the relationship between oxidative balance score (OBS) and stroke was evaluated among noise-exposed adults in the National Health and Nutrition Examination Survey (NHANES) 2005–2018 using logistic regression, and quartiles were applied to determine optimal cut-points.

Results:

Eight OS-related genes were identified, of which seven were validated in stroke mice. C-C Motif Chemokine Ligand 3 (CCL3) emerged as the key biomarker. In UK Biobank, occupational noise increased CCL3 expression (effect = 0.052, P = 0.038), and higher CCL3 levels predicted greater stroke risk (hazard ratio = 1.15, P < 0.001). Functionally, CCL3 drove neuroinflammation via chemokine signaling. In noise-exposed stroke mice, circulating CCL3 increased (P = 0.014) and infarct volume enlarged by 12.9% (P = 0.011). Among 3306 noise-exposed NHANES participants, higher OBS was protective (odd ratio [OR] = 0.97, P = 0.01), whereas OBS <10 markedly elevated stroke risk (OR = 2.40, P = 0.033) compared with the OBS >24 group.

Conclusion:

CCL3 is a validated OS biomarker linking occupational noise exposure to stroke. Maintaining OBS >18 through dietary and lifestyle strategies may mitigate NAS risk and offer a targeted neuroprotective approach for noise-exposed populations.

KEY MESSAGES

• (1)CCL3 is identified as a validated oxidative stress biomarker linking occupational noise exposure to ischemic stroke, indicating a mechanistic bridge between environmental stress and cerebrovascular injury.
• (2)Elevated CCL3 expression predicts higher stroke risk in both population-based cohorts and noise-exposed mice, suggesting its potential as a diagnostic and prognostic indicator for noise-associated stroke.
• (3)Maintaining an oxidative balance score above 18 through dietary and lifestyle interventions effectively mitigates stroke risk among noise-exposed populations, offering a feasible preventive strategy.

INTRODUCTION

Noise exposure is now among the most prevalent occupational hazards worldwide, with an estimated 22 million workers in the United States alone exposed to hazardous noise levels annually.[1] While its auditory effects, such as noise-induced hearing loss, are well established, increasing evidence highlights its broader systemic health implications. Epidemiological studies have demonstrated that chronic noise exposure significantly increases the risk of cardiovascular disorders, including hypertension, atherosclerosis, and stroke, through non-auditory pathways involving dysregulation of the autonomic nervous system and endocrine disturbances.[2,3] Pooled Scandinavian cohort studies have investigated the association between occupational noise exposure and stroke, and reported a non-significant trend toward an increased risk of ischemic stroke[4]; however, the underlying pathophysiological mechanisms of noise-associated stroke (NAS) remain inadequately understood.

NAS is defined here as cerebrovascular events in individuals with chronic occupational noise exposure (≥85 dB for prolonged durations), where noise serves as an environmental risk factor for increasing stroke susceptibility via oxidative stress and related pathways. While not a distinct diagnostic category in clinical practice, NAS diagnosis relies on standard stroke criteria,[5] combined with evidence of noise exposure history, as assessed in epidemiological cohorts.

Emerging clinical and experimental evidence implicates oxidative stress (OS) as a critical mediator linking noise exposure to cerebrovascular injury. A comprehensive review has summarized evidence indicating that noise exposure may induce endothelial dysfunction and oxidative stress, with increases in biomarkers such as 8-hydroxy-2’-deoxyguanosine and malondialdehyde.[6] A mechanistic study based on an animal model further revealed that noise exposure causes vascular inflammation and OS with subsequent endothelial dysfunction in the mouse brain.[7] Notably, OS is also a well-established hallmark of noise-induced hearing loss, where antioxidant therapies attenuate cochlear damage.[8] These findings suggest that common redox imbalance pathways may underlie both auditory and cerebrovascular dysfunctions in individuals exposed to occupational noise. Current approaches to stroke risk stratification rely primarily on conventional biomarkers such as C-reactive protein and lipid profiles. Although these markers are well established for evaluating general cardiovascular risk, they may have limited sensitivity in capturing early, mechanism-driven biological alterations induced by environmental and occupational exposures, including chronic noise exposure.[9,10] An increasing number of studies have demonstrated that noise exposure can trigger excessive production of ROS, leading to endothelial dysfunction, vascular inflammation, and remodeling—pathways that are not directly reflected by traditional inflammatory or lipid-based biomarkers.[11,12,13] By contrast, OS-related biomarkers are more closely aligned with these established pathophysiological mechanisms that link noise to cerebrovascular injury. Prior studies have shown that noise-induced oxidative imbalance precedes overt inflammatory responses and structural vascular damage, suggesting that oxidative stress markers may capture earlier and more proximal biological responses to noise exposure.[14,15,16] Nevertheless, oxidative stress markers are not inherently specific to noise-related health effects, and their clinical utility depends on contextualizing them within defined exposure settings. To date, no clinical biomarker has been specifically validated for predicting noise-associated stroke risk in occupationally noise-exposed populations, highlighting an important gap that warrants further investigation.

Beyond biomarker discovery, translating insights into preventive strategies in identified high-risk cohorts is a paramount clinical objective. The oxidative balance score (OBS) is a comprehensive indicator integrating 15 antioxidant and 5 pro-oxidant factors, encompassing 16 dietary and 4 lifestyle variables, that collectively assess an individual’s OS status.[17] Typically, higher OBS values are associated with reduced OS rates and are strongly correlated with lower mortality risk from chronic diseases, including diabetes mellitus,[17] chronic kidney disease,[18] and cardiovascular disease.[19] Consequently, sustained elevation of OBS may, therefore, attenuate NAS susceptibility, guiding precision dietary and lifestyle interventions in noise-exposed populations.

Despite accumulating evidence linking occupational noise exposure to OS and increased stroke risk, existing research approaches face crucial methodological limitations. Population-based epidemiological studies are well-suited for identifying exposure–disease associations and evaluating risk patterns at the population level, yet they often rely on indirect exposure assessments and observational designs, which limit mechanistic interpretation and causal inference.[20,21,22] Conversely, experimental animal studies allow for controlled exposure conditions and mechanistic exploration but little related research has been limited.[23] Therefore, neither approach alone is sufficient to identify robust, biologically meaningful biomarkers or to establish exposure-relevant preventive thresholds for noise-associated stroke. An integrative strategy that combines large-scale human population analyses with experimental animal validation offers a complementary framework to bridge this gap. Such an approach enables the identification of candidate biomarkers with population-level relevance while simultaneously providing experimental evidence to support biological plausibility and translational significance.

In this study, we integrated population databases and experimental models to identify OS biomarkers for predicting NAS and mechanistically validated the potential of exogenous modulation of OS burden to reduce NAS risk.

MATERIALS AND METHODS

Study Design

The National Health and Nutrition Examination Survey (NHANES) is a biennial survey that evaluates the health and nutritional status of the United States population. Detailed information regarding the survey methodology and consent procedures is available on the NHANES website.[24] The NHANES protocols have been authorized by the National Center for Health Statistics Research Ethics Review Board. This study utilized data from the NHANES 2005–2018 cycle (n = 70,190). Participants without a history of occupational noise exposure (n = 65,613) were excluded. Among the 4577 individuals exposed to noise, 1271 were excluded due to incomplete covariate information, leaving 3306 adults in the final sample.

Experimental Animals

Ten male C57BL/6 mice (8 weeks old, weighing 22–25 g) were obtained from the SPF Biotechnology Laboratory Animal Field (Beijing, China). Animals were housed at 24 °C with a 12 hours light/dark cycle with continuously supplied water and standard laboratory chows (Shulaibao, Wuhan, China). All animal experiments were approved by the ethics committee of the Shanghai Jiao Tong University (permit number A2024333-001).

Measurements of OBS, Stroke, and Covariates

OBS was determined by integrating 16 dietary factors and 4 lifestyle components, comprising 5 pro-oxidants and 15 antioxidants.[25] The definition of OBS is based on the latest literature.[26] Antioxidants were scored in three tertiles (0–2), while pro-oxidants were inversely scored [Table 1]. Higher OBS scores indicate greater substantial antioxidant effects. This study included participants with a total count of at least 16 out of the 20 OBS components. In cases where OBS components were missing, the corresponding component was assigned a score of 0, regardless of its antioxidant or pro-oxidant nature.

Table 1.

Ingredients that make up the oxidative balance score

OBS components	Property	Female			Male
		0	1	2	0	1	2
Dietary OBS
Dietary fiber (g/d)	A	<11.45	11.45–17.40	≥17.40	<13.50	13.50–20.80	≥20.80
Carotene (RE/d)	A	<616.00	616.00–2184.00	≥2184.00	<601.50	601.50–1890.00	≥1890.00
Vitamin B6 (mg/d)	A	<1.53	1.53–2.98	≥2.98	<1.99	1.99-3.46	≥3.46
Vitamin B12 (mcg/d)	A	<3.45	3.45–8.71	≥8.71	<4.46	4.45-9.69	≥9.69
Riboflavin (mg/d)	A	<1.64	1.64–2.73	≥2.73	<2.02	2.012-3.19	≥3.19
Total folate (mcg/d)	A	<259.50	259.50–389.50	≥389.50	<330.50	330.50-495.00	≥495.00
Niacin (mg/d)	A	<19.25	19.25–30.64	≥30.64	<26.31	26.31-39.34	≥39.34
Vitamin C (mg/d)	A	<52.40	52.40–132.05	≥132.05	<49.45	49.45-132.50	≥132.50
Vitamin E (ATE) (mg/d)	A	<5.29	5.29–8.29	≥8.29	<6.37	6.37-9.92	≥9.92
Calcium (mg/d)	A	<750.50	750.50–1206.00	≥1206.00	<850.00	850.00-1308.00	≥1308.00
Magnesium (mg/d)	A	<225.00	225.00–322.00	≥322.00	<277.00	277.00-391.00	≥391.00
Zinc (mg/d)	A	<8.43	8.43–14.5	≥14.5	<11.28	11.28-17.76	≥17.76
Copper (mg/d)	A	<0.95	0.95–1.50	≥1.50	<1.12	1.1215-1.73	≥1.73
Selenium (mcg/d)	A	<82.15	82.15–118.75	≥118.75	<113.05	113.05-163.55	≥163.55
Iron (mg/d)	P	≥17.09	10.73–17.09	<10.73	≥19.54	13.14-19.54	<13.14
Total fat (gm/d)	P	≥77.54	54.15–77.54	<54.15	≥104.58	72.01-104.58	<72.01
Lifestyle OBS
Physical activity(MET-minutes/week)	A	<280.00	280.00–1920.00	≥1920.00	<960.00	960.00-4320.00	≥4320.00
cotinine(ng/mL)	P	≥0.08	0.01–0.08	<0.01	≥0.99	0.02-0.99	<0.02
Alcohol(g/d)	P	≥15	(0, 15)	non	≥30	(0, 30)	non
Body mass index(kg/m2)	P	≥30	[25, 30)	<25	≥30	[25, 30)	<25

Note: A stood for the antioxidant, P for the pro-oxidant, RE for the retinal equivalent, ATE for the alpha-tocopherol equivalent, and MET for the metabolic equivalent. OBS, oxidative balance score.

Stroke was defined as a self-reported previous diagnosis by a physician during a face-to-face interview. Participants who answered “yes” to the following question were classified as stroke cases: “Have you ever been told by a physician or a health professional that you had a stroke?” Although self-reported measures induce potential recall bias, and NHANES lacks relevant information on stroke type, given the relatively higher prevalence of ischemic stroke among stroke patients as well as its closer close relationship with OS,[27,28] most participants with stroke included in this study likely developed ischemic stroke.

Covariates were classified according to demographic and medical information. Demographic information, including age, sex, marital status, education, ethnicity, and tobacco consumption, and medical information, such as glucose level and blood pressure, were collected. The participants in our analytical sample were adults aged 20–80 years, and we categorized them into the following age groups for analysis: under 40 years, 40–60 years, and over 60 years. Ethnicity was stratified into five distinct categories: White people, Black people, Mexican Americans, and other races. Educational attainment was classified as below high school, high school, or above high school. Marital status was classified as never married, divorced/separated/widowed, or married/like married. Body mass index (BMI) was dichotomized at 25 kg/m². These covariates were defined in previous studies.[29,30]

Bioinformatics Analysis Using Machine Learning

Transcriptomic sequencing data were obtained from the Gene Expression Omnibus platform. GSE22255 included 20 ischemic stroke patients and 20 controls, whereas GSE137482 is a stroke dataset based on a mouse model as a validation set. OS-related genes were retrieved from the Gene database (https://www.genecards.org/) with a relevance score greater than 7.[31] Differential gene expression analysis was performed using the limma package [3.52.2] in R software (version 4.3.1; R Core Team, Vienna, Austria), with genes filtered based on |log (FC)| > 0.5 and adj. P-value <0.05. A Venn diagram was used to display the intersecting genes. Differentially expressed OS-related genes were screened for key molecules using eight machine learning algorithms (Random Forest [RF], Support Vector Machine [SVM], XGBoost [XGB], Generalized Linear Model [GLM], Gradient Boosting Machine [GBM], K-Nearest Neighbors [KNN], Neural Networks [NNET], and Decision Tree [DT]. To ensure robustness and reduce model-dependent bias, the final biomarker selection was not determined by a single cutoff value or a predefined weighting rule from any individual machine learning algorithm. Instead, a multi-model consistency–based selection strategy was employed. Specifically, oxidative stress–related candidate biomarkers were evaluated across eight machine learning models. The feature importance rankings and model performance metrics were considered together. Targeted gene networks were derived with the website (https://genemania.org/), and potential functions of the target genes were explored through Gene Ontology analysis, including biological process (BP), molecular function (MF), and cellular component, as well as through Kyoto Encyclopedia of Genes and Genomes (KEGG).

Association of Crucial OS Molecules with Noise and Stroke in UK Biobank Data

Deng et al. constructed a database linking proteomics to disease associations using longitudinal data from the UK Biobank (UKB) public database focus on individuals aged 37–73 at recruitment (https://proteome-phenome-atlas.com/).[32] Based on bioinformatics screening of key OS molecules, we utilized this database to analyze the expression changes of key molecules induced by noise exposure, identifying which of these changes are associated with stroke. Finally, the core molecules were identified.

Mouse Cerebral Ischemia Injury

A middle cerebral artery occlusion (MCAO) model was induced in all the mice by using the intraluminal filament method. Briefly, under isoflurane (MedChemExpress, Shanghai, China) anesthesia (3% induction, 1% maintenance), a midline neck incision was made to expose the right common carotid artery, external carotid artery (ECA), and internal carotid artery (ICA). Following neurovascular dissection, a silicone-coated monofilament (tip diameter: 0.22 ± 0.02 mm; Yaokemei, Wuxi, China) was introduced via ECA arteriotomy and advanced into the ICA until MCA origin occlusion. Filament retention maintained permanent occlusion to establish sustained cerebral ischemia. The core temperature was maintained at 37.0 ± 0.2 °C through the use of a feedback-regulated heating pad (RWD Life Science, Shenzhen, China).

Noise Exposure

Ten mice were randomly allocated into the following two groups: a noise exposure group (n = 5) and a control group (n = 5). The noise-exposed group underwent 2-hour sessions (85 dB sound pressure level; Aihua Instruments, Hangzhou, China) daily for 4 consecutive weeks.[33,34] During noise exposure, the animals were unrestrained in stainless steel wire cages, and the noise levels were continuously monitored during exposure (Aihua Instruments, Hangzhou, China). Both the noise-exposed and control groups were housed under standard laboratory conditions (22 ± 2 °C, 12 hours light/dark cycle, ad libitum access to food and water) in stainless steel wire cages. The control group was maintained under the same conditions without any noise exposure, ensuring that all the environmental factors other than the experimental noise were consistent between the groups.

Neurological Behavior Tests

Neurological function was assessed at 24 hours post-MCAO using the modified neurological severity score (mNSS), a validated 14-point scale that integrates motor, sensory, reflex, and balance assessments, where 0 indicates intact function and 14 represents maximal deficit.[35] The established criteria: scores 1–4, 5–9, and 10–14 correspond to mild, moderate, and severe injury, respectively. All evaluations were conducted by blinded independent investigators.

Humane Endpoints

To minimize animal suffering, predefined humane endpoints were strictly observed throughout the study. Mice would be euthanized immediately if they exhibited any of the following signs: (i) severe and persistent distress or pain, including but not limited to continuous convulsions, vocalization; (ii) inability to access food or water spontaneously, leading to a body weight loss of >20%; (iii) severe motor impairment preventing basic locomotion; or (iiii) any other condition suggesting a moribund state, including but not limited to lethargy unresponsive to gentle stimulation, severe hypothermia, or labored breathing. Fortunately, no mice in this study met these criteria prior to the scheduled endpoint.

2,3,5-Triphenyltetrazolium Chloride (TTC) Staining and Peripheral Blood Collection

At the experimental endpoint (24 hours post-MCAO), the mice (body weight: 22–25 g) were deeply anesthetized using 3% isoflurane. Under deep anesthesia, the mice were euthanized via terminal eyeball removal for blood collection. Approximately 0.8 mL of peripheral blood was collected into anticoagulant tubes. Death following blood collection was confirmed by the absence of a heartbeat and respiratory arrest, followed by cervical dislocation to ensure complete demise. Immediately thereafter, mouse brains were rapidly extracted, flash-frozen in optimal cutting temperature compound (Beyotime Biotechnology, Shanghai, China), and coronally sectioned into six sequential 2 mm slices for TTC staining. The sections were incubated in 2% TTC (Sigma-Aldrich, St. Louis, MO, USA) at 37 °C for 15 minutes, followed by fixation with 4% paraformaldehyde (Sinopharm Chemical Reagent Co., Ltd, Shanghai, China). Digital images of TTC-stained sections were acquired under a uniform overhead white light source using a digital camera (Canon EOS 800D, Canon Inc., Tokyo, Japan). Infarct areas per slice were quantified using ImageJ version 1.53j (NIH, Bethesda, MD, USA). The total infarction volume percentage was calculated with the following formula: infarct area/whole brain area × 100%. Euthanasia was performed under deep isoflurane anesthesia by terminal cardiac puncture followed by cervical dislocation, which was consistent with approved ethical protocols. Death was confirmed by the absence of heartbeats and respiration.

Enzyme-Linked Immunosorbent Assay (ELISA)

C-C Motif Chemokine Ligand 3 (CCL3) protein levels in the peripheral blood of mice were measured using an ELISA kit (YOBIBIO, Catalog Number: U48-1549E, Shanghai, China). Then, the blood was centrifuged (Eppendorf 5424R, Eppendorf, Hamburg, Germany) for 20 minutes (2795 × g, 4 °C) to obtain plasma to evaluate the level of CCL3 protein.

Statistical Analysis

All analyses were performed in R software (version 4.3.1; R Core Team, Vienna, Austria), with P < 0.05 was considered statistically significant. All experiments were performed with at least three biological replicates. The normality of data distributions was assessed using the Shapiro-Wilk test, while homogeneity of variance was evaluated via Levene’s test. Data were presented as mean ± standard deviation (SD) unless otherwise stated. The NHANES analyses utilized the complex survey design and sample weights, with the sample size (n = 3306) representing the number of human participants. An unpaired student’s t-test was used for continuous variables, and a chi-squared test was used for categorical variables. Logistic and subgroup analyses were used to estimate the odds ratios (ORs) and 95% confidence intervals (CIs) for exploring the relationship between OBS and stroke. Subgroup analyses were conducted to examine potential effect modification by age, sex, race, education, marital status, BMI, and smoking status. This study presented findings from three different models. Model 1 referred to the unadjusted model. Model 2 considered demographic factors (age, sex, race, education, and marital status). Model 3 incorporated demographic factors, smoking habits, BMI, glucose, and blood pressure. The model for each subgroup analysis adjusts for covariates, except the subgroup itself. Restricted cubic spline (RCS) analysis was performed to evaluate potential nonlinear associations between the OBS and stroke outcome. ORs with 95% CIs were estimated across the range of the OBS. For diagnostic application, OBS was categorized into quartiles (Q1 [<10], Q2 [10,11,12,13,14,15,16,17,18], Q3 [18,19,20,21,22,23,24], Q4 [>24].

RESULTS

Candidate Biomarkers Associated with OS in Ischemic Stroke

Transcriptome intersection analysis of OS-related genes revealed 76 genes associated with stroke, of which 27 overlapped with both stroke and OS [Figure 1A]. These were subsequently evaluated across eight machine learning algorithms with good performance [Figure 1D and E], and the discrimination ability of each model was evaluated using ROC analysis. The discriminative performance of the eight machine learning algorithms was evaluated using the area under the receiver operating characteristic curve (AUC). The KNN model achieved the highest predictive performance (AUC = 0.722), followed by the SVM and DT models (both AUC = 0.667). The RF and XGB models showed comparable performance (both AUC = 0.639), whereas the GLM and NNET models yielded lower but acceptable discrimination (both AUC = 0.583). The GBM model demonstrated limited predictive ability (AUC = 0.500) [Figure 1B]. The top five stroke-associated genes from each machine learning model were ranked [Figure 1C]. In the models with an AUC > 0.6, the five most critical genes were compared, and we identified eight genes emerging as critical, appearing in at least two models. They were selected as the key biomarkers because they consistently appeared among the top-ranked features across multiple algorithms and demonstrated stable discriminative performance, indicating robustness to model choice. We then experimentally validated these eight genes in a stroke animal model and compared their expression levels with those in normal control mice [Figure 1F]. The results revealed that the expression of seven of these genes with the exception of Protein Phosphatase 1 Regulatory Subunit 15A (Ppp1r15a), significantly differed in tissue samples from the stroke model (all P < 0.05), suggesting that these genes may be involved in the pathophysiological mechanisms of stroke [Figure 2].

Figure 1.

Screening and identification of key oxidative stress (OS) biomarkers for stroke using machine learning algorithms. (A) Venn diagram illustrating the intersection between stroke-related differentially expressed genes (DEGs) and known OS-related genes, identifying 27 overlapping candidate biomarkers. (B) Receiver operating characteristic (ROC) curves evaluating the discriminative performance of eight machine learning models in distinguishing stroke samples from controls. (C) Feature importance rankings displaying the top five most critical genes identified by each machine learning algorithm. (D) Reverse cumulative distribution of residuals and (E) Boxplots of residuals representing the error distribution and stability of the eight models. (F) Intersection analysis displaying the frequency of key gene features shared across the top five performing models to identify robust biomarkers. AUC, area under the curve; DEG, differentially expressed gene; DT, decision tree; GBM, gradient boosting machine; GLM, generalized linear model; KNN, K-nearest neighbors; NNET, neural networks; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machine; XGB, XGBoost.

Figure 2.

Experimental validation of key oxidative stress (OS) candidate genes in a mouse model of ischemic stroke with public datasets. Boxplots illustrate the differential mRNA expression levels of eight candidate biomarkers in brain tissue samples from the Stroke group compared to the Control group. The center line in each box represents the median, box limits indicate the 25th and 75th percentiles, whiskers extend to 1.5 times the interquartile range, and individual dots represent biological replicates. Seven genes (Adm, Gadd45a, Egr1, Ccl3, Sgk1, Cxcl1, and Vim) exhibited statistically significant differences in expression (P < 0.05), whereas Ppp1r15a showed no significant difference. P-values indicating the level of significance are displayed above each comparison. Adm, adrenomedullin; Ccl3, C-C Motif Chemokine Ligand 3; Cxcl1, C-X-C Motif Chemokine Ligand 1; Egr1, Early Growth Response 1; Gadd45a, Growth Arrest and DNA Damage Inducible Alpha; OS, Oxidative Stress; Ppp1r15a, Protein Phosphatase 1 Regulatory Subunit 15A; Sgk1, Serum/Glucocorticoid Regulated Kinase 1; Vim, Vimentin.

The Association of OS Biomarkers with Noise and Ischemic Stroke

Through bioinformatics analysis, we identified seven key OS molecules involved in the pathophysiological process of stroke. Further, based on the UKB database, we examined whether these molecules exhibited changes in expression levels due to noise exposure. Among these seven validated genes, four had protein expression data available in UKB. Table 2 shows that with the prolongation of noise exposure, the expression of CCL3 increases (P = 0.038). Elevated CCL3 level was further associated with higher stroke risk in Table 3 (P < 0.001). Data for GADD45A, EGR1, and SGK1 were not available in the UKB database for protein expression analysis, thus preventing the investigation of their roles in noise exposure and stroke.

Table 2.

The proteins associated with noise exposure

Trait	Protein	Sample size	Estimated effect (95% CI)	SE	P-value
Noisy workplace	ADM	17,262	−0.034 [−0.157 to 0.088]	0.063	0.588
Noisy workplace	CCL3	17,081	0.052 [0.002 to 0.101]	0.024	0.038
Noisy workplace	CXCL1	17,051	0.032 [−0.003 to 0.068]	0.018	0.071
Noisy workplace	VIM	16,786	0.004 [−0.025 to 0.033]	0.015	0.775

Note: The association between noise exposure and protein levels was assessed by linear regression. Data are presented as estimated effect with 95% confidence interval (CI). SE, standard error.

Table 3.

The association between CCL3 and stroke

Protein	Disease category	Disease	HR [95% CI]	P-value
CCL3	Circulatory system diseases	Stroke, excluding SAH	1.15 [1.08–1.22]	2.78E-05
CCL3	Circulatory system diseases	Stroke, including SAH	1.15 [1.08–1.23]	1.04E-05

Note: The association between CCL3 and stroke risk was assessed using Cox proportional hazards regression. Data are presented as hazard ratio (HR) with 95% CI. SAH, subarachnoid hemorrhage.

The Function of Identified Crucial OS Biomarkers for NAS

Gene network analysis revealed potential molecular interaction involving CCL3 [Figure 3A]. Through GO and KEGG enrichment analysis [Figure 3B] showed that CCL3 functions primarily in BPs, particularly associated with response to chemokine, leukocyte chemotaxis, regulation of chemotaxis, and positive regulation of inflammatory response. In terms of MFs, CCL3 is associated with cytokine activity, chemokine activity, calcium-dependent protein kinase C activity, and protein kinase C activity. Lastly, KEGG analysis revealed that CCL3 is enriched in the cytokine–cytokine receptor interaction and chemokine signaling pathways. These results suggest that noise exposure upregulates CCL3, which were associated with inflammation and OS, regulating the OS levels in the nervous system, thereby contributing to increased stroke risk.

Figure 3.

Functional analysis of CCL3 and the impact of noise exposure on ischemic stroke severity in mice. (A) Gene–gene interaction network centered on CCL3, identifying potential molecular partners. (B) Functional enrichment analysis (Gene Ontology and KEGG) of CCL3, highlighting its primary involvement in chemokine signaling, leukocyte chemotaxis, and inflammatory regulation. (C) Representative images of 2,3,5-triphenyltetrazolium chloride (TTC) staining of coronal brain sections from Control (stroke only) and Noise (stroke + noise exposure) groups. Pale areas indicate ischemic infarction, while red areas indicate viable tissue. (D) Quantification of cerebral infarct volume (percentage of whole brain area). Noise exposure significantly increased infarct size compared to controls (P < 0.05). (E) ELISA quantification of CCL3 protein levels in peripheral blood, showing significantly elevated expression in the Noise group (P < 0.05). (F) Assessment of neurological function using the modified Neurological Severity Score (mNSS). Data are presented as mean ± standard deviation (SD). BP, biological process; CCL3, C-C Motif Chemokine Ligand 3; ELISA, Enzyme-Linked Immunosorbent Assay; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function; mNSS, modified Neurological Severity Score; ns, not significant; TTC, 2,3,5-triphenyltetrazolium chloride.

Baseline Characteristics of the Study Participants with Noise Exposure

A total of 3306 participants were included in the analysis, representing 26,377,818 Americans. The overall prevalence of stroke in this population with occupational noise exposure was 5.20%, and the weighted median OBS was 14.75, with higher values observed among non-stroke population (18.42). Compared to non-stroke individuals, stroke patients tended to be older, less educated, more likely to smoke, and exhibited higher levels of systolic blood pressure and blood glucose levels, along with lower levels of diastolic blood pressure (all P < 0.05). Detailed information about the baseline characteristics of all participants grouped by stroke status is illustrated in Table 4.

Table 4.

Baseline characteristics of participants

Variable	Control	Stroke	P-value
Age, year, n (%)			<0.001
Less 40	1021 (35.90)	6 (5.58)
40–60	1010 (39.60)	45 (36.97)
More 60	1103 (24.50)	121 (57.45)
Sex, n (%)			0.541
Female	907 (29.80)	48 (32.96)
Male	2227 (70.20)	124 (67.04)
Race, n (%)			0.167
Other race	612 (11.98)	23 (15.75)
Black people	675 (10.32)	45 (15.06)
Mexican American	465 (8.92)	18 (7.09)
White people	1382 (68.79)	86 (62.09)
Education, n (%)			0.012
Below high school	851 (19.01)	69 (31.32)
High school	832 (25.58)	42 (25.20)
Above high school	1451 (55.42)	61 (43.49)
Marital, n (%)			0.134
Never married	499 (16.08)	17 (8.34)
Divorced/separated/widowed	724 (19.45)	50 (24.20)
Married/like married	1911 (64.47)	105 (67.46)
BMI, kg/m2, n (%)			0.761
Less than or equal to 25	573 (18.23)	29 (16.68)
More than 25	2561 (81.77)	143 (83.32)
Smoker, n (%)			0.013
Former	952 (30.13)	67 (32.59)
Now	796 (25.76)	57 (38.43)
Never	1386(44.11)	48(28.97)
OBS	18.42 (0.29)	14.75 (0.83)	<0.001
Blood glucose, mmol/L	5.65 (0.05)	6.15 (0.17)	0.016
Systolic blood pressure, mmHg	123.39 (0.50)	130.22 (2.00)	<0.001
Diastolic blood pressure, mmHg	71.32 (0.33)	68.58 (0.89)	0.010

Note: Continuous variables (age, OBS, blood glucose, systolic and diastolic blood pressure) were compared using the independent samples Student’s t-test and are presented as mean. Categorical variables were compared using the Chi-squared test and are presented as n (percentage). BMI, body mass index.

Association of OBS with Stroke among the Entire Population and Subgroups

Multivariate logistic regression analysis demonstrated an association between OBS and stroke, observing a lower OBS was correlated with stroke [Table 5]. After adjustment of potential confounders, the adjusted ORs with 95% CIs for stroke were 0.97 (0.94–0.99). After categorizing OBS into four groups based on quartiles and adjusting for covariates, Table 5 showed that compared with the OBS >24 group, the participants with OBS <10 had positively associated with stroke risk (OR = 2.40, 95% CI: 1.09–5.27, P = 0.03). Table 6 revealed that this association was consistent across most strata. Strong inverse associations were in participants with higher stroke prevalence in some specific populations, such as greater than 60 years or age, male, White people, higher educated, married/like marital status, and larger BMI index. However, significant effect modification of the association between OBS and stroke was observed only for education (P for interaction <0.001). This emphasizes the importance of health education and lifestyle interventions in stroke prevention, particularly in populations with lower levels of education. RCS analysis exhibited a non-linear relationship between OBS and stroke risk (P for nonlinear <0.05). Further increases in OBS are associated with a more significant reduction in stroke risk above a threshold of 18. This may potentially serve as an effective clinical evaluation threshold.

Table 5.

Weighted logistic regression on the association between OBS and stroke

OBS	Model 1		Model 2		Model 3
	OR (95% CI)	P-value	OR (95% CI)	P-value	OR (95% CI)	P-value
Overall	0.96 (0.94, 0.98)	<0.001	0.95 (0.93, 0.98)	<0.001	0.97 (0.94, 0.99)	0.011
Q4 (>24)	reference		reference		reference
Q3 (18–24)	2.29 (1.13, 4.66)	0.020	2.09 (1.01, 4.33)	0.052	1.94 (0.97, 3.89)	0.062
Q2 (10–18)	3.60 (1.67, 7.76)	0.001	3.03 (1.33, 6.91)	0.014	2.77 (1.21, 6.34)	0.024
Q1 (<10)	3.06 (1.53, 6.13)	0.002	2.76 (1.23, 6.16)	0.017	2.40 (1.09, 5.27)	0.033

Note: The association between the oxidative balance score (OBS) and stroke was assessed using multivariable logistic regression. Data are presented as odds ratio (OR) with 95% CI. Model 1 is a univariate analysis; Model 2 is adjusted for age, sex, race, education, and marital; Model 3 is adjusted for age, sex, race, education, marital, BMI, smoke, glucose, blood pressure. CI, confidence interval; OR, odds ratio.

Table 6.

Subgroup analysis for the association between OBS and stroke

Variable	Sample size	OR (95% CI)	P-value	P for interaction
Age, year				0.301
Less 40	973	0.88 (0.78, 1.00)	0.05
40 to 60	1109	0.99 (0.95, 1.03)	0.54
More 60	1224	0.95 (0.92, 0.98)	0.003
Sex				0.550
Male	2351	0.96 (0.93, 0.99)	0.01
Female	955	0.98 (0.94, 1.02)	0.28
Race				0.283
Other Race	635	1.01 (0.92, 1.11)	0.88
Black people	720	0.96 (0.91, 1.02)	0.18
Mexican American	483	0.98 (0.91, 1.05)	0.48
White people	1468	0.95 (0.92, 0.98)	0.003
Education				<0.001
Less than high school	920	1.04 (0.99, 1.10)	0.09
High school	874	0.94 (0.89, 0.99)	0.02
More than high school	1512	0.94 (0.91, 0.97)	<0.001
Marital				0.224
Never married	516	0.95 (0.90, 1.01)	0.11
Divorced/separated/widowed	774	0.98 (0.94, 1.03)	0.43
Married/like married	2016	0.96 (0.92, 1.00)	0.03
BMI, kg/m2				0.945
Less than or equal to 25	602	0.95 (0.89, 1.02)	0.15
More than 25	2704	0.97 (0.94, 1.00)	0.02
Smoke				0.708
Former	1019	0.98(0.95, 1.00)	0.11
Now	853	0.96(0.90, 1.02)	0.21
Never	1434	0.96(0.91, 1.01)	0.1

Note: All analyses were adjusted for all covariants except themselves. BMI, Body Mass Index.

Validation of the OS Biomarker in the Ischemic Stroke Mice Model with Noise

To assess the impairing effects of noise in mice with ischemic brain injury, we analyzed cerebral infarct volume and neurological function scores. Compared with control mice, noise-exposed mice exhibited significantly greater infarct volumes in Figure 3C and D (P = 0.011). The mNSS was used to assess the severity of neurological deficits [Figure 3F]. Compared with the control group, the noise exposure group had no significant reduction in neurological deficits (P = 0.074). CCL3 is a crucial OS biomarker for patients with NAS. Therefore, we examined the levels of CCL3 in the peripheral blood of mice [Figure 3E]. CCL3 expression was significantly greater in noise-exposed mice than in control mice (P = 0.014).

DISCUSSION

Noise is ubiquitous, especially in manufacturing industries where noise levels remain significantly high and effective noise elimination measures or devices are lacking, making it a critical concern for public health and occupational safety. Long-term noise exposure has been established as a clear risk factor for stroke.[36] In this context, this study identified the role of CCL3 in predicting NAS risk among noise-exposed populations and highlighted OBS as a novel early intervention approach. These findings advance our understanding of the mechanisms mediated by OS and the inflammatory factor CCL3.

The increase in noise-induced OS is a well-recognized and well-established mechanism.[37,38] OBS reflects an individual’s oxidative stress status and has been widely applied in population-based studies to evaluate its associations with diverse clinical outcomes,[39,40] its role in stroke has rarely been examined.[41,42] Critically, no prior research has addressed its application in populations exposed to occupational noise. We found that individuals exposed to lower OBS noise were at higher risk of stroke, suggesting that OBS may serve as an important intervention indicator to prevent the occurrence of stroke exogenously. These findings indicate that changes in OS levels may serve as key contributors to stroke pathogenesis in populations exposed to occupational noise, offering a valuable reference for early clinical intervention.

Among the various mechanisms involved in ischemic stroke, OS plays a pivotal role in disease progression.[43] Ischemia deprives cells of oxygen and glucose, resulting in the production of ROS, which further exacerbates cellular damage through lipid peroxidation, protein oxidation, and DNA damage.[44] These oxidative events impair cellular functions and ultimately contribute to inflammation, neuronal apoptosis, and blood–brain barrier disruption, thereby worsening stroke outcomes.[45]

Our results revealed that CCL3 is a key inflammatory mediator associated with noise and stroke. CCL3 is upregulated in response to OS and serves as a key chemokine involved in the pathophysiological process of OS.[46,47] It plays a crucial role in inflammatory responses and chemokine signaling pathways, which are directly linked to OS and the regulation of ROS levels.[48] CCL3 is involved in leukocyte chemotaxis, promoting the recruitment of macrophages and neutrophils to injury sites,[49] thereby amplifying the inflammatory response, oxidative damage, and tissue injury.[50] In addition, CCL3 regulates various signaling pathways associated with neuronal death and tissue damage following ischemia.[51,52] Given its dual roles in OS and inflammation, CCL3 represents a potential therapeutic target for ischemic stroke. However, the precise molecular mechanisms associated with ROS and downstream phenotypic outcomes remain incompletely understood. Further studies are needed to clarify the precise mechanisms through which CCL3 modulates stroke pathophysiology under noise exposure. Exploration of its potential for targeted therapeutic intervention could provide novel insights into targeted therapeutic strategies for mitigating the adverse effects of environmental stressors including noise on cerebrovascular health. Additionally, when occupational noise exposure is unavoidable or temporary removal from environmental noise is not feasible, dietary and lifestyle modifications that reduce OS levels may offer complementary strategies to mitigate risk.

Our findings linking CCL3 and OBS can be biologically contextualized through oxidative stress–inflammation interplay. A lower OBS, reflecting reduced antioxidant capacity from diet and lifestyle, may promote ROS accumulation, which in turn enhances the production of proinflammatory cytokines, including CCL3, amplifying neuroinflammatory responses.[53,54] Conversely, a higher OBS could mitigate ROS accumulation and suppress downstream inflammatory signaling, including CCL3-mediated pathways, providing a potential protective mechanism against noise-related cerebrovascular injury. This mechanistic connection suggests that improvements in OBS may attenuate CCL3-driven inflammation, offering a plausible biological rationale for the observed protective effect of increased antioxidant balance on stroke risk and highlighting the translational potential of lifestyle-based interventions.

The integration of both NHANES (cross-sectional) and UKB (longitudinal) data strengthens the validity of our findings. The NHANES provides a large, representative sample of the United States population, while the UKB enables the examination of temporal relationships and causal pathways between noise-induced molecular expression changes and stroke risk over time. This dual data source improves the robustness and generalizability of the conclusions.

However, this study has several limitations. First, a key limitation is the reliance on self-reported occupational noise exposure, which lacks quantitative data on sound pressure levels (dB(A)) or cumulative exposure duration. While self-reported measures can introduce potential misclassification bias, validation studies have demonstrated moderate to good concordance between self-reported and objectively measured noise exposure.[55,56] Detailed industry classification data were also unavailable (e.g., NAICS [North American Industry Classification System] codes 31–33) in the datasets, precluding the sensitivity analysis. Importantly, any resulting exposure misclassification is likely nondifferential, biasing associations toward the null and thus underestimating the true effect of noise on stroke risk. Second, although the extensive adjustment for covariates, potential confounding factors cannot be excluded. Third, while CCL3 has been identified as a key OS biomarker for patients with NAS, the specific regulatory mechanisms through which this classic inflammatory factor affects OS remain unclear. Further mechanistic explorations using cell or animal models are required to explore its effects on ROS and its impact on phenotypic expression. Moreover, stroke outcomes in the NHANES were based on self-reported physician-diagnosed stroke without subtype information. A previous study suggested that self-reported stroke has a sensitivity of 80–95% and specificity of 96–99%, indicating generally good validity.[57] Finally, occupational noise often co-occurs with other hazards such as organic solvents,[58] which may jointly affect stroke risk; our study could not account for such co-exposures, highlighting the need for multi-exposure research.

CONCLUSION

This study identifies CCL3 as a key OS-derived biomarker for predicting NAS. In settings where occupational noise exposure is unavoidable, exogenous modulation of systemic OS status is associated with a lower risk of NAS. The OBS provides a potential quantifiable metric for assessing intervention efficacy, with maintenance above 18. These mechanistic and translational insights lay a strong foundation for precision prevention initiatives aimed at cerebrovascular health in high-risk occupational populations.

Availability of Data and Materials

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

Author Contributions

Jia Huang conceived and designed the study and critically revised the manuscript. Xiao Yu and CongCong Wu conducted the analysis. Tian Lv and Congcong Wu drafted the manuscript and contributed to the revision of the manuscript. Xiao Yu and Congcong Wu contributed equally to this work. All authors have read and approved the final manuscript.

Ethics Approval and Consent to Participate

The animal experiments were approved by the Shanghai Jiao Tong University Ethics Review Committee (permit number A2024333-001). The experimental procedures followed the ARRIVE guidelines (https://arrivalguidelines.org). The data involving humans approved by the National Center for Health Statistics Ethics Review Committee was obtained from National Center for Health Statistics. This organization has made all its data freely available. The participants provided their written informed consent to participate in this study. This study did not require additional ethical approval or informed consent, nor was it necessary to apply to the institution for data access or usage permission.

Conflicts of Interest

The authors declare no conflicts of interest.

Funding Statement

This work was supported by grants from the Scientific Research Project of Jiangsu Provincial Health Commission (K2024084, JSZJ20241204).