Biological aging mediates the associations between urinary metals and osteoarthritis among U.S. adults

The NHANES is an ongoing national cross-sectional survey and data is available on the website of the America Centers for Disease Control and Prevention (CDC) (http://​www.​cdc.​gov/​nchs/​nhanes.​htm). The study protocol was approved by the National Center for Health Statistics (NCHS) Research Ethics Review Board. All participants provided written consent at the time of recruitment. In the current study, OA patients were included according to self-reported information (n = 46,736). Next, we excluded individuals whose information on 9 metals was missing (n = 34,147). Further, we excluded individuals with missing creatinine data (n = 5). Collectively, 12,584 participants were enrolled, including 1356 OA patients (Fig. 1).

A study showed an 81% agreement between self-report and clinical confirmation of OA [22]. In NHANES, OA was diagnosed by professionals and the information was collected by questionnaire survey. Briefly, all participants ≥ 20 years of age were asked two arthritis-related questions. First, they were inquired: “Has a doctor or other health professional ever told you that you have arthritis?” Individuals with an affirmative response were asked a follow-up question: “Which type of arthritis was it?” The participants who answered “osteoarthritis” were enrolled in the study.

The detection data of 9 urinary metals were obtained from NHANES 1999–2016. Barium (Ba), Cd, cobalt (Co), cesium (Cs), molybdenum (Mo), Pb, antimony (Sb), thallium (Tl), and uranium (Tu) were primarily measured from spot urine samples by using inductively coupled plasma mass spectrometry (ICP-MS). The limit of detection (LOD) divided by the square root of two was used to replace the values below the LOD. The concentrations of metals were all corrected by urinary creatinine and were expressed as micrograms per gram creatinine.

Biological age and phenotypic age used different calculation methods and incorporated different biomarkers to measure biological aging. Biological age was proposed by Klemera and Doubal [23] based on 8 biomarkers (Ln-C-reactive protein (CRP), serum creatinine, glycosylated hemoglobin, serum albumin, serum total cholesterol, serum urea nitrogen, serum alkaline phosphatase, and systolic blood pressure) [24]. The values j and i represent the number of biomarkers and samples respectively. The values k, q, and s are the regression slope, intercept, and the root means squared error of a biomarker regressed on chronological age, respectively. The value r_j² represents the variance explained by regression of chronological age on biomarkers.

Phenotypic age was calculated by using 9 aging-related variables [25], including chronological age, albumin, creatinine, glucose, CRP, lymphocyte percent, mean cell volume, red blood cell distribution width, alkaline phosphatase, and white blood cell count. where xb =  − 19.907 − 0.0336 × Albumin + 0.0095 × Creatinine + 0.1953 × Glucose + 0.0954 × LnCRP-0.0120 × Lymphocyte Percent + 0.0268 × Mean Cell Volume + 0.3306 × Red Cell Distribution Width + 0.00188 × Alkaline Phosphatase + 0.0554 × White Blood Cell Count + 0.0804 × Chronological Age.

The measurement method of telomere length has been reported elsewhere [26]. Briefly, blood samples were collected and determined by quantitative polymerase chain reaction (qPCR) assays to assess the telomere length relative to standard reference DNA (the T/S ratio). All sample analyses were performed in duplicate strictly following the manufacturer’s instruction and all results were checked to meet the laboratory’s standardized criteria for acceptability before being released for reporting.

All analyses were performed with SAS (version 9.4) or R (version 3.6.3) and accounted for the complex sampling design of the NHANES, and the code for analyses were in Additional file 2. Chi-square tests and t-tests were used to assess participants’ demographic characteristics by OA status. The metal concentrations were Ln-transformed to obtain approximate normal distribution (continuous variables) or categorized into four quartiles (Q1, Q2, Q3, and Q4) as categorical variables. Multivariable logistic regression was applied to estimate odds ratios (OR) and corresponding 95% confidence interval (CI) for the associations of metals and biological aging markers with OA risk. Multivariable linear regression was used to explore the associations of metals with biological aging markers. All analyses adjusted covariates for age (< 60 or ≥ 60 years), sex (male or female), race/ethnicity (Mexican American, other Hispanic, non-Hispanic White, non-Hispanic Black, or others), marital status (married/cohabiting, widowed/divorced/separated, or never married), physical activity (moderate or vigorous), drinking alcohol status (ever or never), body mass index (BMI), the ratio of family income to poverty (PIR), and serum cotinine concentration. Missing data for the covariates were coded as a missing indicator category for categorical variables and were imputed with the median for the continuous variable. We used the false-discovery rate (FDR) correction to adjust for multiple tests in the regression models. The trend test across increasing exposure groups was calculated using integer values (1, 2, 3, and 4). Pearson correlation analysis was used to assess the correlations among Ln-transformed metals. The directed acyclic graphs (R package “dagitty” and “ggdag”) showed the main relationships among exposures, outcomes, covariates, and mediators (Additional file 1: Fig. S1).

Weighted quantile sum (WQS) regression was applied to explore the overall effects of metals on OA as it performed well in characterizing environmental mixtures [27]. R package (“gWQS”) can empirically calculate the WQS index comprised of weighted sums of individual metal concentrations. The WQS index (ranged from 0 to 1) represented the mixed exposure level of metals, and the components of concern were identified by non-negligible weights. The final result was interpreted as the simultaneous effect on OA of a one-quantile increase of mixed metals.

Several sensitivity analyses were also performed. First, given the potential non-linearity and non-additive relations among metals, Bayesian kernel machine regression (BKMR) was used to assess the joint effect of all metals (“BKMR” packages) and the dose–response relationship between single metal and OA risk when fixing other metal concentrations. Second, we further adjusted for the occupation of participants (industry and agriculture, service and transportation, and others). Third, we further adjusted for age-related diseases including diabetes (yes or no), hypertension (yes or no), cardiovascular disease (yes or no) and cancer (yes or no), and OA-related medicine use (e.g., meloxicam, celecoxib) (yes or no). Fourth, we further adjusted for the survey cycle (1, 2, 3, 4, 5, 6, 7, 8, and 9). Fifth, participants with abnormal urinary creatinine were excluded. Finally, pregnant participants were excluded.

The potential mediating effects of biological aging markers on the associations of single and mixed metals (shown as WQS index) with OA risk were estimated by two mediation models, including parallel mediation (R package “mediation”) and serial mediation analyses (R package “bruceR”). Parallel mediation models used individual indicators as a mediator, while serial mediation models used a pathway as a mediator. Mediation analyses used the quasi-Bayesian Monte Carlo method with 1000 simulations based on normal approximation. The direct effect (DE) represented the effects of metal exposure on OA without a mediator. The indirect effect (IE) represented the effects of metal exposure on OA through the mediator. The proportion of mediation was calculated by using IE divided by TE (total effect).

Among the 12,584 adults, 1356 were diagnosed with OA. The demographic characteristics of the study participants with or without OA are listed in Table 1. Overall, age, sex, ethnicity, marital status, physical activity, drinking alcohol status, family PIR, BMI, serum cotinine, biological age, phenotypic age, and telomere length were statistically significant between OA and non-OA participants.

The distribution of metal concentrations is shown in Additional file 1: Table S1. The detection rates of metals were greater than 75%. Pearson’s coefficient between Ln-transformed metals showed moderate correlations between Cs and Tl (r = 0.58), Ba and Co (r = 0.41), and Cd and Pb (r = 0.40), while other correlations were relatively poor (Additional file 1: Fig. S2).

Figure 2 displayed the associations between creatinine-adjusted metal concentrations and OA risk by the survey-weighted logistic regression models. The highest exposure quantile of Cd (OR 1.64, 95%CI 1.20 to 2.23), Co (OR 1.59, 95%CI 1.20 to 2.10), and Cs (OR 1.48, 95%CI 1.13 to 1.93) increased the risk of OA compared to quantile 1 (all P for trend < 0.05). These associations were also presented in Ln-transformed metals and OA risk (all P < 0.05).

The WQS index of mixed metals was positively associated with OA risk (OR 1.23, 95%CI 1.10 to 1.37) (Fig. 2). Additionally, the highest weighted metals in WQS models were Cd (54.45%), Co (27.14%), and Cs (9.23%) (Additional file 1: Fig. S3).

In the sensitivity analyses, the BKMR model also showed a significant and positive association of mixed metals with OA risk (Additional file 1: Fig. S4). After further adjusting for the occupation of the participants, adjusting for diabetes, hypertension, cardiovascular disease, cancer, and OA-related medicine use, adjusting for the survey cycle, excluding participants with abnormal urinary creatinine, or excluding pregnant participants, the results were not materially changed (Additional file 1: Table S2-6).

Figure 3 showed the associations of metals with biological aging markers based on the linear regression. We found the highest quartile of Cd, Co, Cs, Pb, and Tl (versus quartile 1) were associated with a growing of biological age (all P for trend < 0.001). With increasing quantiles of Cd, Co, Cs, and Pb, phenotypic age was increased (all P for trend < 0.001). Cd and Cs were negatively associated with telomere length (all P for trend < 0.05). Moreover, mixed metals had positive associations with biological age (β 4.91, 95%CI 4.52 to 5.31) and phenotypic age (β 5.90, 95%CI 5.45 to 6.34) and had a negative association with telomere length (β − 0.04, 95%CI − 0.05 to − 0.02).

Table 2 shows the associations of aging markers with OA risk based on the logistic regression. Each 1-year increase in biological age increased the risk of OA by 6% (95%CI 1.05 to 1.07). Similarly, each 1-year increase in phenotypic age was associated with an increased OA risk (OR 1.04, 95%CI 1.04 to 1.05). Additionally, for each unit increase in telomere length (mean T/S ratio), OR for OA decreased by 74% (95%CI 0.07 to 0.97), which was consistent with the quantile analysis (Q4 vs. Q1: OR 0.40, 95%CI 0.17 to 0.90).

Furthermore, parallel mediation analyses were performed to evaluate the potential mediation effects of biological aging on the associations of metals with OA risk. Biological age had significant mediated effects on the associations of Cd, Co, and Cs with OA risk, and the proportion of mediation was 69.27%, 18.43%, and 30.50% respectively (all P < 0.05). The mediated proportion of phenotypic age was 61.01%, 31.83%, and 17.00% on the associations of Cd, Co, and Cs with OA risk respectively. Telomere length also mediated the association between Cs and OA risk and the proportion of mediation was 9.81% (Additional file 1: Table S7). Moreover, biological age, phenotypic age, and telomere length parallelly mediated the associations of mixed metals with OA risk with 57.60%, 48.30%, and 9.50% proportion of mediation respectively (all P < 0.05) (Fig. 4).

As cell senescence is a leading cause of whole-body aging, we further explored the potential paths of associations between metals and OA risk by the serial mediation models. Additional file 1: Table S8 shows the association between Cs and OA risk serially mediated by telomere length-biological age path and telomere length-phenotypic age path, with 4.55% and 4.17% proportion of mediation respectively. Besides, the serial mediated effect of telomere length-biological age path and telomere length-phenotypic age path on the association of mixed metals with OA risk were 0.70% and 0.60% respectively. And the proportion of mediation was 11.67% and 9.84%, respectively (Fig. 4).