Participants in the present investigation were elderly individuals recruited across eight community sites within Shenzhen, China, from August to December 2021. Enrollment was governed by the following eligibility requirements: (1) subjects - whether officially registered or otherwise - had to have established residence in Shenzhen for at least six months; (2) they were required to be 60 years of age or older; (3) they had to willingly consent to involvement in the survey and execute the corresponding informed consent form. Regarding participant selection, individuals were omitted from the analysis if they had incomplete Geriatric Depression Scale (GDS) or Hamilton Anxiety Scale (HAMA) records (n = 64), or if the volume of their urine specimens was inadequate to enable OH-PAHs quantification (n = 3). As a result, 413 participants were ultimately included in the final dataset. The research was granted ethical clearance by the Ethics Review Committee of the Shenzhen Nanshan Center for Chronic Disease Control (Approval No. II20220015).

For the measurement of urinary OH-PAHs, samples underwent enzymatic deconjugation followed by solid-phase extraction (SPE), after which the extracts were analyzed on a UHPLC system (ExionLC™, SCIEX) that was operated in tandem with a Triple Quad 6500+ mass spectrometer (SCIEX, USA). As a first step in sample work-up, the urine specimens were allowed to reach 4 °C after thawing, vortexed for 30 s, and an aliquot of 1.5 mL was transferred into a tube containing ammonium acetate buffer (pH 4.5-5.0) and a preparation of β-glucuronidase/aryl sulfatase. The enzymatic hydrolysis proceeded at 37 °C over 12 h, after which a cocktail of internal standards was added. Sample clean-up used a Bond Elut C18 cartridge holding 200 mg of sorbent in a 3 mL format. Before the specimen was applied, the cartridge was primed by passing methanol (MeOH) and subsequently Milli-Q water through it. Once the sample had been loaded, the cartridge was rinsed with Milli-Q water and then with MeOH/water prepared at a 1:1 (v/v) ratio. Acetonitrile (ACN) eluted the compounds of interest, and the effluent was concentrated to near dryness under a gentle nitrogen gas stream. The remaining residue was dissolved in a solvent mixture of ACN, MeOH, and water at a 1:1:8 (v/v/v) ratio, vortexed, and centrifuged. The clear upper layer was then pipetted into a 2 mL autosampler vial ahead of injection. Analyte separation was performed on an ACQUITY UPLC HSS T3 column (Waters, Ireland; 100 mm × 2.1 mm, with a particle size of 1.8 μm). For each run, 2 μL of the extract was injected, with ultrapure water serving as mobile phase A and a 1:1 (v/v) blend of MeOH and ACN as mobile phase B. The analytes were resolved using a gradient over a 16-minute total runtime, organized as follows: B was held at 20% for the opening 0.5 min; raised to 40% in a linear manner from 0.5 to 1 min; brought up to 60% between 1 and 4 min; kept constant at 60% from 4 to 6.5 min; advanced to 70% over the 6.5-12 min window; elevated further to 100% during 12-13 min; maintained at 100% for 13-15 min; and finally returned to 20% during the re-equilibration phase spanning 15.1-16 min^[43,44]. Analyte concentrations were established using the isotope-dilution internal standard methodology, with quantitation anchored to a ten-point standard curve covering 0-100 ng/mL (0, 0.2, 0.5, 1.0, 2.0, 5.0, 10.0, 20.0, 50.0, and 100.0 ng/mL). Across these points, the R² values exceeded 0.999. For linearity assessment, regression was carried out under a 1/x weighting scheme. The limits of detection (LOD) and of quantification (LOQ) were defined as the analyte levels corresponding to signal-to-noise ratios of 3 and 10, respectively (S/N = 3 and S/N = 10). In cases where an individual measurement proved to be lower than the LOQ, the recorded figure was replaced by LOQ/√2^[45].

In order to ensure analytical reliability, quality control was managed through a pooled sample formed by mixing identical volumes drawn from the individual urine specimens; this aggregate served as laboratory-generated external quality-control material and was incorporated into each batch run so that both the effectiveness of the pre-processing steps and the constancy of the instrument could be followed continuously. To check for possible contamination, a solvent blank and a process blank were run alongside the samples at intervals of every 15 injections within a given batch. A process blank prepared from 1.5 mL of Milli-Q water underwent the same analytical sequence as the urine samples, and no OH-PAH analytes were detected in any blank run. To assess recovery, the pooled quality control (QC) material was enriched with OH-PAHs at three graded concentrations: 10 ng/mL (low), 25 ng/mL (intermediate), and 50 ng/mL (high). Recovery percentages were found to vary between 66.55% and 104.07%, while the accompanying relative standard deviations (SDs) ranged from 1.33% to 17.89%.

The panel of quantified urinary OH-PAHs comprised nine analytes, namely 1-hydroxynaphthalene (1-OHNap), 2-hydroxynaphthalene (2-OHNap), 2-hydroxyfluorene (2-OHFlu), 3-hydroxyfluorene (3-OHFlu), 1/9-hydroxyphenanthrene (1/9-OHPhe), 2-hydroxyphenanthrene (2-OHPhe), 3-hydroxyphenanthrene (3-OHPhe), 4-hydroxyphenanthrene (4-OHPhe), and 1-hydroxypyrene (1-OHPyr). As the isomers 1-OHPhe and 9-OHPhe could not be resolved chromatographically, the value reported for 1/9-OHPhe corresponds to the combined concentration of the two. For subsequent analyses, the following molar sums were adopted: ΣOH-NAPs, aggregating 1-OHNap and 2-OHNap; ΣOH-FLUs, comprising 2-OHFlu and 3-OHFlu; and ΣOH-PHEs, summing 1/9-OHPhe, 2-OHPhe, 3-OHPhe, and 4-OHPhe; and ΣOH-PAHs, representing the total sum of all nine OH-PAH metabolites (1-OHNap, 2-OHNap, 2-OHFlu, 3-OHFlu, 1/9-OHPhe, 2-OHPhe, 3-OHPhe, 4-OHPhe, and 1-OHPyr). Before the data were interpreted, individual OH-PAH concentrations were adjusted on the basis of urine specific gravity (SG), a normalization approach that has been documented in earlier investigations^[43]. Detailed procedures are provided in Supporting Information [Supplementary Text 1].

The HAMA was used to assess the anxiety level of older adults, an instrument that has also been applied to measure anxiety symptoms in very recent studies of older adults^[46]. Developed by Hamilton, HAMA is a semi-structured interview tool that evaluates the severity of anxiety symptoms through 14 items, covering both psychological and physiological aspects of anxiety^[47]. Each item is scored from 0 (absent) to 4 (severe), yielding a maximum total score of 56. Higher total scores correspond to greater anxiety severity. In this study, a score of ≥ 14 was defined as indicative of anxiety^[48].

The GDS was employed to assess depressive status in older adults. As a widely used screening instrument for older-adult depressive symptoms^[49], which has been adopted in several recent studies^[50-52], this 15-item scale uses yes/no responses referring to participants’ feelings during the preceding week, with scores ranging from 0 to a maximum total score of 15. In the current study, a GDS score of ≥ 5 indicated significant depressive symptoms^[53,54].

Based on the differences in demographic characteristics of the study population and potential influencing factors of anxiety and depressive symptoms reported in previous literature^[35,55], different adjustment models were adopted, and the following covariates were included: Model 1: crude model without covariate adjustment; Model 2: adjusted for sex and age; Model 3: fully-adjusted model, adjusted for sex, age, marital status, educational level, BMI, exercise, smoking status, and drinking. Age and BMI were included as continuous variables in all regression models; in Table 1, they were additionally presented as categorical groups (age: 60-65 and > 65 years; BMI: non-overweight < 24 kg/m² and overweight ≥ 24 kg/m², according to the Chinese Expert Consensus on Medical Nutrition Treatment of overweight/obesity^[56]). Other covariates were treated as categorical variables in regression models, defined as follows: sex (male, female); marital status (married, unmarried/divorced/widowed); education (< 3 years, 3-6 years, > 6 years); exercise (hardly, < 3 times/week, ≥ 3 times/week); smoking status (never, former, current smoker); drinking (no, yes).

Continuous variables were expressed as mean ± SD, while categorical variables were presented as n (%). Due to the uneven distribution of urinary OH-PAHs, logarithmic (lg) transformation was applied. Spearman rank correlation was used to associations among urinary OH-PAH concentrations. Logistic regression and multiple linear regression analyses were used to evaluate the relationship between OH-PAH exposure (continuous variable and categorical variable) and depressive and anxiety symptoms and scores, and the median concentration of each quantile group was used as a continuous variable for trend testing. Separate single-exposure regression models were fitted for each exposure metric, with only one exposure term included per model - either an individual OH-PAH metabolite, a single family-level summed index (ΣOH-NAPs, ΣOH-FLUs, ΣOH-PHEs), or the total summed index of all nine OH-PAH metabolites (ΣOH-PAHs). To account for multiple testing, Benjamini–Hochberg false discovery rate (FDR) correction was applied to the P values of these single-exposure models for each of the four outcomes, separately for continuous-exposure models and categorical (tertile; T2 and T3) exposure models, and separately within each of the three adjustment models (Model 1, Model 2, and Model 3).

Restricted cubic splines (RCS) were used to evaluate the nonlinear associations between urinary OH-PAHs and outcomes. Three knots were set in the model, and the knot positions were determined based on the 10th, 50th, and 90th percentiles of OH-PAHs exposure. For the mixture analyses, all nine log-transformed individual OH-PAH monomer metabolites were entered concurrently as distinct exposure components. Accordingly, Bayesian kernel machine regression (BKMR) was used to accurately estimate the joint and individual effects of the mixed OH-PAH components and their potential interactions. For continuous scale outcomes (HAMA and GDS scores), a Gaussian outcome family was specified, whereas a binomial family was used for the binary prevalent anxiety and depressive symptom outcomes. The model was fitted using the native single-chain Markov Chain Monte Carlo (MCMC) implementation of the bkmr R package with 10,000 iterations and a fixed random seed (set.seed = 1000) to ensure full reproducibility. Fifty knots were defined for the kernel-machine term, and variable selection (varsel = TRUE) with estimation of the residual variance (est.h = TRUE) was enabled, using the package’s default prior distributions. Convergence was assessed by visual inspection of the trace plots for key model parameters (β coefficients, sigsq.eps, and r mixture-component parameters). Point estimates and posterior SDs derived from the MCMC posterior samples were used to construct 95% confidence intervals (CIs) for the univariate predictor–response curves and the overall mixture-effect summaries. Meanwhile, to quantify the joint effect of mixed exposure to the nine urinary OH-PAHs on anxiety and depressive symptoms and the relative contribution weight of each exposure component, quantile g-computation (QGC) was further applied. A Gaussian family was specified for continuous HAMA and GDS scores and a binomial family with a logit link for the binary anxiety and depressive symptom outcomes. Exposure concentrations were stratified into quartiles (q = 4), and the bootstrap-based qgcomp.glm.boot function with 100 bootstrap resamples (B = 100) was used to derive confidence intervals for the mixture-level effect estimates.

Considering that the effects of OH-PAH exposure on anxiety and depressive symptoms reported in previous studies may be closely related to BMI and sex, we conducted stratified analyses based on BMI and sex. In addition, studies have shown a clear connection between BMI status and the metabolism of PAHs in the human body. We further calculated the interaction term between OH-PAH exposure and BMI to explore the potential impact of BMI on this association. For statistically significant interaction terms, the Johnson-Neyman method was used to further explore potential moderating effects, aiming to determine the key BMI thresholds characterizing the relationship between OH-PAHs and depressive and anxiety symptoms, HAMA scores, and GDS scores. To reduce bias in the estimation of regression coefficients caused by extreme BMI values, Z-score standardization was performed on BMI: specifically, based on the original BMI data of the study population, BMI was converted to standardized values with a mean of 0 and an SD of 1. Johnson-Neyman analysis was conducted based on standardized BMI, and the obtained critical values were subsequently converted back to the original BMI units via inverse transformation to facilitate practical interpretation of the results and elaboration of their clinical implications. To further identify key PAH metabolites associated with anxiety and depressive symptoms and cross-check the consistency of key pollutant signals with traditional regression results, two tree-based machine-learning algorithms, random forest (RF) and extreme gradient boosting (XGBoost), were used for exploratory feature-importance analysis of urinary OH-PAH metabolites. Stratified ten-fold cross-validation was applied to conduct hyperparameter grid-search tuning and model performance evaluation. Within each cross-validation fold, the Synthetic Minority Oversampling Technique (SMOTE) was implemented to address class imbalance for binary anxiety and depressive outcomes. Regularization constraints including limited tree depth, minimum node-size thresholds, and restricted tree numbers were adopted to mitigate potential overfitting. Model performance metrics, including the area under the receiver operating characteristic (ROC) curve (AUC) and F1-score, were calculated as averaged values across all ten cross-validation folds. Based on cross-validated performance, XGBoost exhibited better AUC and F1-score compared with RF and was selected for subsequent SHapley Additive exPlanations (SHAP) interpretation. Following hyperparameter optimization via cross-validation, the final XGBoost model was refit on the complete dataset. SHAP values were computed from this full-dataset final model to quantify the marginal contribution of each OH-PAH metabolite. The full input feature set for the XGBoost model comprised all individual urinary OH-PAH metabolites as well as pre-defined summed OH-PAH indices (ΣOH-NAPs, ΣOH-FLUs, ΣOH-PHEs, ΣOH-PAHs). Gain-based feature importance quantifies the total splitting gain contributed by each feature during model training; near-zero importance values indicate negligible predictive contribution. Machine-learning outputs served as exploratory feature-ranking results to cross-check findings derived from conventional regression analyses. To evaluate the robustness of the primary results, this study conducted a sensitivity analysis in the population excluding OH-PAHs concentration outliers [defined as individuals exceeding the original distribution interquartile range (IQR) value by more than three times].

Statistical analyses were conducted using R (version 4.3.3), with the “qgcomp”, “bkmr”, and “rms” R packages used to analyze QGC, BKMR, and RCS curves, respectively. A two-tailed P value < 0.05 was considered statistically significant.

This study demonstrated that urinary ΣOH-PHEs and 1-OHPyr were associated with higher odds of prevalent anxiety symptoms in the overall older-adult study population. No statistically significant association between 1-OHPyr and depressive symptoms was observed in the full cohort, although a suggestive marginal trend was present, and a statistically significant association was detected only among female participants. These associations exhibited sex-specific heterogeneity: the associations of 1-OHPyr with depression and 3-OHPhe with anxiety were observed exclusively in females. Additionally, the association between 1-OHPyr and anxiety was observed only in non-overweight participants. BKMR indicated higher odds of prevalent anxiety symptoms at higher percentiles of the OH-PAH mixture, whereas QGC did not yield a statistically significant overall mixture effect for anxiety or depressive symptoms. Despite the non-significant overall mixture effect in QGC, 1-OHPyr stood out as the leading positive contributor to mixture-related effects. To our knowledge, this is the first study of its kind to center on older adults and simultaneously investigate the associations between OH-PAH exposure and both depressive and anxiety symptoms (and their respective scale scores). It circumvents the one-sidedness of single-outcome research, comprehensively captures the multidimensional mental health impacts of OH-PAH exposure in older adults, and addresses the critical gap in existing literature left by isolated investigations into individual symptoms in this population.

Our study demonstrated that elevated urinary 1-OHPyr and ΣOH-PHEs concentrations were associated with higher odds of prevalent anxiety symptoms in older adults. Similar results have been found in other populations. As one illustration, a cross-sectional investigation drew on data collected from 4,026 adults aged 20 or above within the National Health and Nutrition Examination Survey (NHANES) spanning 2007 to 2012, and examined how exposure to OH-PAHs was related to anxiety. The results revealed that 1-OHPyr exhibited a significant nonlinear association with anxiety, while mixed OH-PAH exposure was positively correlated with anxiety - with 1-OHNap and 3-OHFlu as the primary contributors to the joint effect^[55]. Meanwhile, two other studies based on the NHANES database have both established an association between PAH exposure and depression in US adults^[33,35] and reported that the highest quartiles of six PAH metabolites were linked to an elevated risk of depressive symptoms both in the crude model and after adjusting for age and sex^[35]. However, in older adults, there are currently no reports on the association between exposure to PAHs and depressive and anxiety symptoms. In the future, more prospective cohort studies need to be conducted to further explore their potential health impacts.

The biological mechanisms linking PAH exposure to depressive and anxiety symptoms remain incompletely understood. Toxicological evidence suggests that PAHs may affect mental health through multiple interrelated pathways^[57,58]. For instance, chronic PAH exposure can disrupt cerebral metabolism, particularly lipid and fatty acid metabolism, leading to anxiety- and depression-like behaviors in mouse models^[8]. Exposure to PAHs can also induce oxidative stress, characterized by increased production of reactive oxygen species and impaired antioxidant defense systems, which may result in DNA damage and contribute to the development of depression^[58-60]. Alterations in neurotransmitter regulation have also been implicated in PAH-induced neurotoxicity^[9]. Furthermore, PAH exposure can trigger neuroinflammatory responses, including the upregulation of pro-inflammatory cytokines in the brain, and inflammation is increasingly recognized as a key mechanism underlying depressive and anxiety symptoms^[57,61-64].

Stratified logistic regression and multiple linear regression analyses showed that in female participants, elevated 1-OHPyr levels were significantly associated with higher odds of prevalent depressive symptoms, while high 3-OHPhe concentrations were significantly correlated with higher odds of prevalent anxiety symptoms. Existing studies have confirmed that women are more susceptible to depression and anxiety than men, which can be attributed to physiological and social factors. Physiologically, women show stronger amygdala activation in response to negative emotional stimuli, resulting in poorer emotional regulation^[65,66]. In addition, estrogen modulates GABAergic neurons in the medial preoptic area, indirectly increasing women’s vulnerability to anxiety and depression^[67]. Socially, women suffer more from work-family conflict and cumulative disadvantages^[68], while lacking sufficient decision-making power in families and the labor market^[69], which further impairs their mental health. Therefore, women may be more sensitive to the adverse effects of PAH exposure. This explains our subgroup analysis finding that 1-OHPyr was significantly associated with depressive symptoms only in females.

Our study findings indicate significant BMI-exposure interactions for several phenanthrene-derived OH-PAH metabolites with respect to HAMA scores; these interaction findings should be regarded as exploratory. Among these metabolites, 2-OHPhe and 3-OHPhe were further subjected to Johnson-Neyman analysis given their statistically significant continuous-scale interaction terms. Note that the higher BMI critical thresholds, above which the model points toward a protective-direction effect, should be interpreted cautiously, as very few participants fall within those high-BMI ranges. In the BMI-stratified analyses, elevated 1-OHPyr levels were significantly associated with higher odds of prevalent anxiety symptoms with a notable linear trend; this represents an exploratory stratified observation and should not be interpreted as evidence that BMI modulates the 1-OHPyr–anxiety association. Jin et al. reported that when investigating the association between PAH exposure and asthma, only in subjects with BMI ≥ 30 were 1-OHNap, 2-OHNap, and 1-OHPyr significantly associated with asthma^[70]. Meanwhile, Peris-Camarasa et al. found that high BMI was associated with increased urinary PAH concentrations^[71]. Peng et al. provide a plausible explanation for these observations, asserting that obesity may influence the urinary concentrations of PAH biomarkers due to the lipophilic nature of PAHs^[72]. In contrast, Chen et al. demonstrated that higher BMI was associated with reduced urinary 1-OHPyr excretion in highly exposed Chinese coke oven workers, which may be explained by increased pyrene storage in adipose tissue, leading to less pyrene being metabolized to 1-OHPyr^[73]. Thus, a plausible explanation for our findings is as follows: First, PAHs are typical lipophilic environmental pollutants with low water solubility and high affinity for adipose tissue^[74]. After entering the body, lipophilic PAHs preferentially accumulate in adipose tissue, reducing their free concentrations in blood; because the adipose tissue compartment is larger in obese individuals, this in vivo dilution is more pronounced and further lowers their effective exposure levels^[75-77], thereby alleviating neurotoxicity to brain regions related to emotional regulation. Second, obesity-associated changes in renal hemodynamics (e.g., glomerular hyperfiltration) can alter the urinary excretion kinetics of PAHs, affecting the urinary concentrations of PAH metabolites and potentially introducing bias into the correlation analysis^[78,79]. Third, obesity can be characterized as a chronic, low-grade inflammatory condition. In this state, adipose tissue releases pro-inflammatory cytokines such as tumor necrosis factor-alpha (TNF-α) and interleukin-6 (IL-6), which are capable of crossing the blood–brain barrier and thereby contributing to the onset or worsening of anxiety and depression^[80-82], which may confound the association between PAH exposure and psychological symptoms. Collectively, the regulatory effect of BMI on the association between PAH exposure and depressive or anxiety symptoms in older adults is attributed to the combined effects of lipophilic accumulation, dilution and altered excretion kinetics, and obesity-related low-grade inflammation.