Characteristics of the study population at baseline

Table 1 presents the characteristics of the 1,099 participants. The mean age of the adolescent population in this study was 15.41 ± 2.30 years, with male participants accounting for 51.32% of the total, indicating a relatively balanced gender distribution. The analysis of lipid indicators showed that the mean LDL-C was 2.25 ± 0.68 mmol/L, the mean TC was 4.04 ± 0.77 mmol/L; the mean TG was 0.90 ± 0.52 mmol/L; and the mean HDL-C was 1.38 ± 0.32 mmol/L. The mean AIP was − 0.23 ± 0.28.

Table 1 Baseline characteristics of the participantsDetection and correlation of PFAS in the population

In this study, the detection rates for the four types of PFAS were all > 99%; PFOS had the highest detection rate (100%). Of all detected compounds, PFOS exhibited the maximum mean concentration (8.10 ± 7.33 ng/mL), whereas PFNA exhibited the minimum mean concentration (0.94 ± 0.70 ng/mL) (Table S1). Figure S3 shows the correlations between the four types of PFAS; all correlations between the four monomers were significant (p < 0.05). Of these, PFOS exhibited the strongest correlation with PFOA (r = 0.70). The weakest correlation was observed between PFHxS and PFNA (r = 0.26).

Analysis of weighted generalized linear regression models

The weighted generalized linear regression model aimed to evaluate the correlations between each type of PFAS and both lipid levels and AIP (Table 2). With all covariates adjusted, LDL-C only had a statistically significant positive correlation with PFOS (β = 0.07, 95% CI: 0.03–0.12). TC exhibited significant positive correlations with both PFOS (β = 0.10, 95% CI: 0.04–0.15) and PFOA (β = 0.08, 95% CI: 0.005–0.15). TG exhibited significant statistical positive correlations with PFNA (β = 0.07, 95% CI: 0.03–0.11), PFOS (β = 0.06, 95% CI: 0.03–0.10), and PFOA (β = 0.07, 95% CI: 0.02–0.12). Similarly, AIP exhibited statistically significant positive correlations with PFNA (β = 0.04, 95% CI: 0.02–0.06), PFOS (β = 0.04, 95% CI: 0.02–0.05), and PFOA (β = 0.05, 95% CI: 0.02–0.07). However, there is no statistically significant linear correlation was found between HDL-C and any of the different types of PFAS.

Table 2 The association between a single poly- and perfluoroalkyl substances component and lipid parameter as well as atherogenic index of plasmaWQS regression model of PFAS with lipid levels and AIP

After adjusting for all covariates, the WQS index, representing a mixture of PFAS types, exhibited positive associations with LDL-C (β: 0.05, 95% CI: 0.007–0.10), TC (β: 0.07, 95% CI: 0.01–0.12), TG (β: 0.05, 95% CI: 0.01–0.08), and AIP (β: 0.03, 95% CI: 0.006–0.04). It is noteworthy that no significant association was found between the PFAS mixture and HDL-C (β: -0.002, 95% CI: -0.33–0.02) (Table S2). Figure 2A-D visually presents the WQS weight distribution among the components of the PFAS mixture. PFOS had the largest weight contribution to the overall effects of PFAS on LDL-C, TC, and TG, with weights of 0.89, 0.94, and 0.39, respectively. For the association between AIP and PFAS, PFOA had the most prominent weight. In contrast, PFHxS had a relatively smaller weight contribution in the associations between all lipid levels, AIP, and PFAS. To further investigate the associations between PFAS and lipid parameters, including the atherogenic index of plasma (AIP) across gender-specific adolescent subgroups, we conducted gender-stratified analyses. Our findings revealed significant positive correlations between the WQS index of PFAS mixtures and LDL-C (95% CI: 0.02–0.17), TC (95% CI: 0.05–0.22), TG (95% CI: 0.02–0.14), and AIP (95% CI: 0.05–0.22) in the male cohort. In contrast, among females, significant associations were only observed between the WQS index and TG (95% CI: 0.03–0.14) as well as AIP (95% CI: 0.03–0.09) (Table S2). Notably, PFOS and PFOA emerged as the predominant contributors in both gender groups, with weight coefficients exceeding those of other PFAS congeners (Figure S4, Figure S5).

Fig. 2

A weighted quantile sum regression model was used to assign weights to each poly- and perfluoroalkyl substances (PFAS) component, illustrating the single contributions of each type of PFAS to the mixed effect. (A) LDL-C; (B) TC; (C) TG; (D) AIP

BKMR model analysis of the association between PFAS and both lipid and AIP

Figure 3A-E visually presents the joint effect relationship between the PFAS mixture and lipid levels as well as the AIP. After adjusting for all covariates, analysis showed that mixed PFAS exposure had a positive joint effect on LDL-C, TC, TG, and AIP, while no significant joint effect was observed with HDL-C. In particular, PFOS had the highest PIP for LDL-C (0.85), TC (0.91), TG (0.75), and AIP (0.90) (Table S3). When other types of PFAS were controlled at P25, P50, and P75, respectively, each quartile increase in PFOS concentration led to an increase in LDL-C by 0.13 mmol/L, 0.15 mmol/L, and 0.18 mmol/L, an increase in TC by 0.19 mmol/L, 0.21 mmol/L, and 0.22 mmol/L, and an increase in AIP by 0.04 in all cases. Furthermore, when other types of PFAS were controlled at P25 and P50, each quartile increase in PFOS concentration resulted in an increase in TG by 0.10 mmol/L and 0.09 mmol/L, respectively (Figure S6A-E). To further investigate the potential non-linear relationship between exposure and effect, Figure S7A-E graphically presents the univariate exposure-effect function estimates for each type of PFAS. With all other types of PFAS held constant at their median levels, PFOS exhibited a clear positive association with LDL-C, whereas PFHxS and PFNA demonstrated a negative association with LDL-C. Similarly, PFOS was positively associated with TC, while PFHxS, PFNA, and PFOA were negatively associated with TC. The associations between TG and PFOS, PFHxS, and PFOA respectively exhibited almost U-shaped associations characterized by an initial drop and a subsequent upturn, while PFNA exhibited a positive association with TG. The associations between AIP and PFOS, PFHxS, and PFOA respectively also showed an almost U-shaped association characterized by an initial drop and a subsequent upturn, while PFNA showed a positive correlation with AIP. In addition, we plotted bivariate exposure-response functions at the 25th, 50th, and 75th percentiles to investigate whether there were interactions between different types of PFAS. Analysis showed that in the bivariate exposure-response models for HDL-C, TG, TC, and AIP, no significant trend changes were found for individual types of PFAS from P25 to P75, indicating no potential interactions between PFAS. However, in the bivariate exposure-response model for LDL-C, an interaction was detected between PFOS and PFOA (Figure S8A-E). Findings from gender-stratified analyses revealed that in the male cohort, the positive associations between PFAS and LDL-C, as well as TC, achieved statistical significance at quantiles exceeding 0.5. The positive correlation between PFAS and TG reached significance at quantiles > 0.35, while the association between PFAS and AIP remained statistically significant across all quantiles (Figure S9). In contrast, although positive trends were observed between PFAS and various lipid parameters in the female cohort, statistically significant positive correlations were only identified for the relationship between PFAS and AIP (Figure S10). PIP analyses further indicated that in males, PFOS exhibited the highest PIP values in the associations between PFAS mixtures and LDL-C (PIP = 0.82), TC (PIP = 0.90), TG (PIP = 0.60), and AIP (PIP = 0.70). In females, PFOS demonstrated the highest PIP in the relationships between PFAS mixtures and LDL-C (PIP = 0.37) and TC (PIP = 0.38), whereas PFNA showed the highest PIP in the associations with TG (PIP = 0.84) and AIP (PIP = 0.65) (Table S3).

Fig. 3

Mixed effects of perfluorooctanoic acid (PFOA), perfluorooctane sulfonate (PFOS), perfluorononanoic acid (PFNA), and perfluorohexane sulfonate (PFHxS) on low density lipoprotein cholesterol (LDL-C) (A), total cholesterol (TC) (B), triglycerides (TG) (C), high-density lipoprotein cholesterol (HDL-C) (D), and atherogenic index of plasma (AIP) (E) with 95% CIs, comparing scenarios where all pollutants were set at specific percentiles (ranging from the 25th to the 75th percentiles) against a baseline scenario where all pollutants were at their median value (50th percentile)

Suppression effect of RBC folate on the relationship between PFAS, lipid levels and AIP

We investigated the relationship between LDL-C and TC with all individual types of PFAS and total PFAS. No significant indirect effects of RBC folate were detected; we only detected significant direct and total effects for PFOA, PFOS, and total PFAS on LDL-C and TC (Table S4). However, mediation analysis of TG with PFOA, PFOS, and PFNA, showed that RBC folate exerted suppression effects of 25%, 37.5%, and 12.5%, respectively. This means that the presence of RBC folate reduced the strength of the relationship between PFAS and TG, resulting in a direct effect of PFAS on TG that was greater than the total effect. This suppression effect was also significant in the relationship between TG and total PFAS, where RBC folate contributed to a 33.33% suppression effect (Fig. 4A-D). Furthermore, mediation analysis of AIP with PFOA, PFOS and PFNA, revealed that the suppression effects mediated by RBC folate were 16%, 25%, and 10%, respectively. This effect remained significant in the relationship between AIP and total PFAS, where RBC folate exhibited 20% suppression (Fig. 4E-H). However, no significant direct, mediation/ suppression, or total effects were observed for the relationship between HDL-C and all individual types of PFAS as well as total PFAS (Table S4). It is worth noting that no mediation or suppression effects of total serum folate were observed for the relationship between individual PFAS, total PFAS, and lipid levels as well as AIP (Table S5). In gender-stratified mediation analyses, we observed several trends. Among males, RBC folate exerted suppressive effects accounting for 40%, 33%, and 32% of the total associations between PFOS and TG, PFOS and AIP, and total PFAS and TG, respectively (Figure S11, Table S6). Among females, RBC folate demonstrated suppression effects contributing 57%, 55%, 68%, 47%, 34%, 58%, 58%, 63%, and 42% of the total associations between PFOS and LDL-C, PFOS and TC, PFOS and TG, PFOS and AIP, PFOA and TG, PFNA and TG, total PFAS and LDL-C, total PFAS and TC, and total PFAS and AIP, respectively (Figure S12, Table S7). Notably, the suppression effects of RBC folate appeared more pronounced in females. Conversely, serum folate exhibited no mediation/suppression effects on the associations between any individual PFAS or total PFAS and lipid parameters or AIP in either gender group (Table S8, Table S9).

Fig. 4

Suppression effects of red blood cell folate on the associations of perfluorooctanoic acid (PFOA) (A), perfluorooctane sulfonate (PFOS) (B), perfluorononanoic acid (PFNA) (C) and total poly- and perfluoroalkyl substances (PFAS) (D) with triglycerides (TG) as well as the associations of PFOA (E), PFOS (F), PFNA (G) and total PFAS (H) with atherogenic index of plasma (AIP)

Network toxicology identified the underlying mechanisms by which PFAS may influence atherosclerosis or hyperlipidemia

The two-dimensional structures of the four PFAS are shown in Figure S13A-D. The toxic endpoints shared by the four PFAS were included skin sensitization, eye corrosion/irritation, respiratory toxicity, and blood-brain barrier disruption, with PFHxS, PFNA, and PFOA additionally linked to carcinogenicity, liver injury, nephrotoxicity (Figure S14A-D and Table S10).

Venn analysis identified 14 PFAS-hyperlipidemia intersection targets (Fig. 5A), 16 PFAS-atherosclerosis intersection targets (Fig. 5B), 29 intersection targets between folate and hyperlipidemia (Figure S15A), and 47 intersection targets between folate and atherosclerosis (Figure S15B).

Fig. 5

Venn diagrams, protein-protein interaction (PPI) networks, and “components-targets-disease” networks showing poly- and perfluoroalkyl substances (PFAS) intersecting with hyperlipidemia and atherosclerosis. Venn diagram of PFAS and hyperlipidemia targets (A). Venn diagram of PFAS and atherosclerosis targets (B). Venn diagram of the intersection targets of PFAS, folate, and atherosclerosis (C). PPI network of the potential targets of PFAS affecting hyperlipidemia (D). PPI network of the potential targets of PFAS affecting atherosclerosis (E). PPI network of the core targets of PFAS affecting atherosclerosis (F). A “PFAS-targets-hyperlipidemia” network. The yellow circle nodes represent PFAS compounds, the blue round rectangle nodes represent predicted intersection targets, the purple hexagon represents hyperlipidemia, and the gray edges indicate compound-target interactions (G). A “PFAS-targets-atherosclerosis” network. The yellow circle nodes represent PFAS compounds, the blue round rectangle nodes represent predicted intersection targets, the purple hexagon represents atherosclerosis, and the gray edges indicate compound-target interactions (H)

In the PPI network, the larger and darker the color of the target node in the network graph, the more important the target in the network. Core PPI network targets included albumin (ALB), peroxisome proliferator activated receptor γ (PPAR-γ), PPAR-δ, nuclear receptor subfamily 1 group H member 4 (NR1H4), and interleukin-10 (IL-10) for PFAS-hyperlipidemia intersection targets (Fig. 5D); ALB, PPAR-γ, NR1H4, IL-10, and IL-4 for PFAS-atherosclerosis intersection targets (Fig. 5E-F); tumor necrosis factor (TNF), caspase-3, sarcoma (SRC), kinase insert domain receptor (KDR), and lactate dehydrogenase A (LDHA) for folate-hyperlipidemia intersection targets (Figure S15C-D); and TNF, caspase-3, SRC, KDR, and caspase-1 for folate- atherosclerosis intersection targets (Figure S15E-F), visualized in component-target-disease networks (Fig. 5G-H, Figure S15G-H).

A total of 107 entries were identified by the GO enrichment analysis of PFAS mixtures and hyperlipidemia. The main entries were intracellular receptor signaling pathway and nuclear receptor activity. A total of 109 entries were obtained from the GO enrichment analysis of PFAS mixtures and atherosclerosis. The main entries were intracellular receptor signaling pathway, chromatin, and nuclear receptor activity, respectively. The most significant seven entries were selected to plot a histogram (Fig. 6A-B). A total of 165 entries were obtained from the GO enrichment analysis of folate and hyperlipidemia. The main entries were leukocyte tethering or rolling, receptor complex, and protein tyrosine kinase activity. A total of 219 entries were obtained from the GO enrichment analysis of folate and atherosclerosis. The main entries were one-carbon metabolic process, cytosol, and carbonate dehydratase activity. The most significant 10 entries were selected and plotted as a histogram (Figure S16A-B).

Fig. 6

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses and a “components-targets-pathways” network showing PFAS intersecting with hyperlipidemia and atherosclerosis. GO enrichment analysis of PFAS and hyperlipidemia. Green, orange, and blue bars represent the top seven biological process (BP), cellular component (CC), and molecular function (MF) terms, respectively (A). GO enrichment analysis of poly- and perfluoroalkyl substances (PFAS) and atherosclerosis. Green, orange, and blue bars represent the top seven BP, CC, MF terms, respectively (B). A bubble chart of the top 10 pathways based on the KEGG enrichment analysis of PFAS and hyperlipidemia (C). The bubble chart of the top 10 pathways based on the KEGG enrichment analysis of PFAS and atherosclerosis (D). A “PFAS-targets-pathways” network. The blue round rectangle nodes represent the top 10 KEGG pathways of the common targets for PFAS and hyperlipidemia, the purple circle nodes represent KEGG pathways, and the yellow diamond nodes represent PFAS components (E). A “PFAS-targets-pathways” network. The blue round rectangle nodes represent the top 10 KEGG pathways of common targets for PFAS and atherosclerosis, the purple circle nodes represent KEGG pathways, and the yellow diamond nodes represent PFAS components (F)

KEGG enrichment analysis revealed that the PFAS mixture mainly affected blood lipids through the autoimmune thyroid disease, PPAR signaling pathway, and efferocytosis pathways, and affected atherosclerosis via efferocytosis, autoimmune thyroid disease, and PPAR signaling pathways. Folate mainly affected blood lipids through the TNF signaling pathway, the mitogen-activated protein kinase (MAPK) signaling pathway, and lipid and atherosclerosis pathways, and affects atherosclerosis via lipids and atherosclerosis, TNF signaling pathway, and apoptosis pathways. The most significant 10 pathways and corresponding genes were selected to generate KEGG enrichment bubble charts (Fig. 6C-D, Figure S16C-D) and “component-target-pathway” network diagram (Fig. 6E-F, Figure S16E-F).

An intersection was identified between the targets of PFAS mixtures affecting atherosclerosis and the targets of folate affecting atherosclerosis (caspase-1 and carbonic anhydrase 2 (CA2)) (Fig. 5C). No intersection was detected between the targets of PFAS mixtures affecting hyperlipidemia and the targets of folate affecting hyperlipidemia.

PFOS, PFOA, PFNA and PFHxS were selected as ligands, and the four targets possessing the maximum degree values in the PPI network of the intersection targets of hyperlipidemia and atherosclerosis were selected as receptors (Table S11) for molecular docking. The binding energies for each docked complex were all < -6 kcal/mol (Table S12), indicating good affinity. Four high-affinity complexes (docking binding energy was all less than< -8.5 kcal/mol) visualized in Fig. 7, including PFNA-PPARG (single hydrogen bond with the amino acid residue CME-285), PFOA-PPARG (one hydrogen bond with the amino acid residue ARG-288 and two additional hydrogen bonds with LEU-228), PFOS-ALB (one hydrogen bond with the amino acid residue THR-236) and PFOS-PPARG (dual hydrogen bonds with the amino acid residue ARG-288).

Fig. 7

Molecular docking results of top four complexes with the lowest binding energy. perfluorononanoic acid (PFNA) and peroxisome proliferator activated receptor gamma (PPARG) (A). Perfluorooctanoic acid (PFOA) and PPARG (B). Perfluorooctane sulfonate (PFOS) and albumin (ALB) (C). PFOS and PPARG (D)