Oxidative Stress and Inflammatory Biomarkers for Populations with Occupational Exposure to Nanomaterials: A Systematic Review and Meta-Analysis

Exposure to nanomaterials (NMs) is suggested to have the potential to cause harmful health effects. Activations of oxidative stress and inflammation are assumed as main contributors to NM-induced toxicity. Thus, oxidative stress- and inflammation-related indicators may serve as biomarkers for occupational risk assessment. However, the correlation between NM exposure and these biomarkers remains controversial. This study aimed to perform a meta-analysis to systematically investigate the alterations of various biomarkers after NM exposure. Twenty-eight studies were found eligible by searching PubMed, EMBASE and Cochrane Library databases. The pooled results showed NM exposure was significantly associated with increases in the levels of malonaldehyde (MDA) [standardized mean difference (SMD) = 2.18; 95% confidence interval (CI), 1.50–2.87], 4-hydroxy-2-nonhenal (HNE) (SMD = 2.05; 95% CI, 1.13–2.96), aldehydes C6-12 (SMD = 3.45; 95% CI, 2.80–4.10), 8-hydroxyguanine (8-OHG) (SMD = 2.98; 95% CI, 2.22–3.74), 5-hydroxymethyl uracil (5-OHMeU) (SMD = 1.90; 95% CI, 1.23–2.58), o-tyrosine (o-Tyr) (SMD = 1.81; 95% CI, 1.22–2.41), 3-nitrotyrosine (3-NOTyr) (SMD = 2.63; 95% CI, 1.74–3.52), interleukin (IL)-1β (SMD = 1.76; 95% CI, 0.87–2.66), tumor necrosis factor (TNF)-α (SMD = 1.52; 95% CI, 1.03–2.01), myeloperoxidase (MPO) (SMD = 0.25; 95% CI, 0.16–0.34) and fibrinogen (SMD = 0.11; 95% CI, 0.02–0.21), and decreases in the levels of glutathione peroxidase (GPx) (SMD = −0.31; 95% CI, −0.52–−0.11) and IL-6 soluble receptor (IL-6sR) (SMD = −0.18; 95% CI, −0.28–−0.09). Subgroup analysis indicated oxidative stress biomarkers (MDA, HNE, aldehydes C6-12, 8-OHG, 5-OHMeU, o-Tyr, 3-NOTyr and GPx) in exhaled breath condensate (EBC) and blood samples were strongly changed by NM exposure; inflammatory biomarkers (IL-1β, TNF-α, MPO, fibrinogen and IL-6sR) were all significant in EBC, blood, sputum and nasal lavage samples. In conclusion, our findings suggest that these oxidative stress and inflammatory indicators may be promising biomarkers for the biological monitoring of occupationally NM-exposed workers.

The exclusion criteria were: (a) duplications; (b) non-original studies (e.g., case reports, reviews, conference abstracts, letters and protocols); (c) preclinical studies (in vitro and in vivo); (d) control groups that consisted of individuals who were not occupationally exposed to NMs were unavailable; (e) data of interest could not be extracted from the included articles; and (f) irrelevant topics. Eligible studies were independently selected by two researchers and disagreements were resolved by the third researcher.

Data Extraction
Two reviewers independently completed data extraction. The extracted information included the first author, country, publication year, study design, sample size, study subjects of exposed (including exposure duration in workplaces) and non-exposed groups, the type of exposed NMs and the sample type utilized in the analysis of outcomes and outcome measures. The results were expressed as mean and standard deviation (SD); if 95% confidence interval (CI), interquartile range, and minimum and maximum values were provided, they were converted to SD appropriately. The data in the figures were extracted by using the GetData Graph Digitizer version 2.26 (GetData Pty Ltd., Kogarah, Australia). Any divergences were resolved by discussion with the third author.

Quality Assessment
The quality of included observational studies was assessed by the Newcastle-Ottawa Scale (NOS) [21]. NOS consisted of eight items that were categorized into three domains: selection of study groups, comparability of groups and determination of exposure. If the answer was yes, one star was assigned for each item (except of the comparability group item, which was given two stars). The total NOS scores ranged from 0 to 9 by adding up all the stars, and studies with a NOS score >6 were considered to be of high quality. Quality assessment was performed independently by two reviewers; any disagreements were resolved by a third reviewer.

Statistical Analysis
Extracted data were stored in a Microsoft Excel file and exported to STATA version 15.0 (Stata Corp., College Station, TX, USA) for the statistical analysis. Since all biomarkers were measured as continuous variables, the standardized mean difference (SMD) and 95% CI were used to express the effect size. The significance of the total SMD was examined by the Z-test and two-sided p < 0.05 denoted a statistically significant association of biomarkers with NM exposures. The heterogeneity among studies was assessed by the Cochran-Q test and I-squared (I 2 ) statistics. p-values < 0.1 or I 2 -values > 50% indicated the presence of heterogeneity. A random effects model was used to pool the estimates if significant heterogeneity was observed; otherwise, a fixed effects model was utilized. Subgroup analyses stratified by NM types and sample sources were conducted to explore the effects of the study characteristics (potential sources of the heterogeneity). The likelihood of publication bias (with over-reporting of positive results and under-reporting of negative results in studies) was appraised with the Egger's linear regression test (briefly, the standard normal deviate was regressed against the estimate's precision) [22]. If the publication bias was encountered (indicated by p < 0.05), the trim-and-fill method [23] was applied to adjust the publication bias and further ascertain the influence of the publication bias on the outcomes of the meta-analysis. The stability of the meta-analysis results was also confirmed by a sensitivity analysis with a leave-one-out method [24] (that is, each study was omitted at a time and then the pooled estimates were reassessed. The reassessed results were compared with the original results to judge whether the removed studies would alter the pooled results).

Literature Search
As shown in Figure 1, the electronic database search retrieved 7210 records. After the removal of 4871 duplicates, 2339 studies underwent the title and abstract screening, which resulted in 2295 of them excluded because they were reviews or meta-analyses (n = 130), case reports (n = 5), conference abstracts (n = 70), preclinical studies (n = 662) and irrelevant topics (n = 1428). The remaining 44 studies were entered into a full-text screening to further assess their eligibility. As a result, 16 studies were excluded because of data unavailable (n = 9), without non-exposed controls (n = 6) and unclear exposure order for NMs and filtered air in two groups (n = 1). Eventually, 28 studies with 2636 participants (including 1374 exposed and 1262 non-exposed) were included in the meta-analysis [14][15][16][17][18][19]. bias on the outcomes of the meta-analysis. The stability of the meta-analysis results was also confirmed by a sensitivity analysis with a leave-one-out method [24] (that is, each study was omitted at a time and then the pooled estimates were reassessed. The reassessed results were compared with the original results to judge whether the removed studies would alter the pooled results).

Literature Search
As shown in Figure 1, the electronic database search retrieved 7210 records. After the removal of 4871 duplicates, 2339 studies underwent the title and abstract screening, which resulted in 2295 of them excluded because they were reviews or meta-analyses (n = 130), case reports (n = 5), conference abstracts (n = 70), preclinical studies (n = 662) and irrelevant topics (n = 1428). The remaining 44 studies were entered into a full-text screening to further assess their eligibility. As a result, 16 studies were excluded because of data unavailable (n = 9), without non-exposed controls (n = 6) and unclear exposure order for NMs and filtered air in two groups (n = 1). Eventually, 28 studies with 2636 participants (including 1374 exposed and 1262 non-exposed) were included in the meta-analysis [14][15][16][17][18][19].

Study Characteristics and Quality Assessment
The main characteristics of each included article are summarized in Table 1. The publication years of these 28 studies ranged from 2014 to 2022. Ten studies were conducted in China (including five in mainland and five in Taiwan), nine in Czech, two in Netherlands Antioxidants 2022, 11, 2182 5 of 26 and one in the USA, Italy, Latvia, Russia, Korea, Israel and Australia, respectively. Except of two studies that had a panel design, the other studies used a cross-sectional design. The sample size of all studies was small (n < 100 in 19 studies; n < 200 in six studies and n > 200 in three studies). Most of the participants worked in a workplace producing NMs for more than one year (except some without clear descriptions) and the results may theoretically reflect the long-term exposure effect. Some studies specifically stated the NM type for occupational exposure, such as graphene, silica oxide nanoparticles (SiO 2 NPs), iron oxide nanoparticles (IONPs), titanium dioxide nanoparticles (TiO 2 NPs), indium tin oxide nanoparticles (ITONPs), multi-walled carbon nanotubes (MWCNTs), while others did not provide the detail and were categorized as mixed NMs. All studies attempted to explore non-invasive biomarkers in blood, urine, sputum, nasal lavage or EBC samples. All of the included studies were deemed to be of high quality because the NOS score was 7, 8 or 9 (Table 1).

Meta-Analysis Results
The number of experimental datasets for meta-analysis was larger than the actual number of included articles because multiple NM types, detection time points and sample sources were included for some studies. The detailed data that were extracted for each variable are presented in Table S1.

Association between Occupational NM Exposure and Inflammatory Biomarker Levels
Only the effects of NM exposure on the levels of IL-1β (Figures 8 and 9), TNF-α (Figures 10 and 11) and ICAM were still significant in the analyses of subgroups (regardless of NM types and sample sources) with at least two datasets (Table S3). LT-B4 and LT-E4 in the EBC samples (but not urine) were found to be significantly increased by NM exposure. FENO was only observed to be significantly higher in the TiO 2 NP-exposed workers compared with the non-exposed controls (Table S3). NM types (mixed) and sample sources (serum) were the same in all datasets for the analysis of MPO, fibrinogen and IL-6sR. The number of datasets (only two) was small for MIP-1β and MCP-1. Thus, the subgroup analysis was not performed for them.
oxidants 2022, 11, x FOR PEER REVIEW 17 of exposure. FENO was only observed to be significantly higher in the TiO2NP-expos workers compared with the non-exposed controls (Table S3). NM types (mixed) and sa ple sources (serum) were the same in all datasets for the analysis of MPO, fibrinogen a IL-6sR. The number of datasets (only two) was small for MIP-1β and MCP-1. Thus, subgroup analysis was not performed for them. Figure 8. Forest plots assessing the effects of exposure to different NMs on the level of IL-1β co pared with the non-exposed control group. NMs, nanomaterials; NPs, nanoparticles; TiO2NPs, t nium dioxide nanoparticles; MWCNTs, multi-walled carbon nanotubes; IL-1β, interleukin-1β; SM standardized mean difference; CI, confidence interval [17,19,39,40]. compared with the non-exposed control group. NMs, nanomaterials; NPs, nanoparticles; TiO 2 NPs, titanium dioxide nanoparticles; MWCNTs, multi-walled carbon nanotubes; IL-1β, interleukin-1β; SMD, standardized mean difference; CI, confidence interval [17,19,39,40].

Publication Bias and Sensitivity Analysis
Egger's test showed the publication bias was present for analyses of MDA (p = 0.002), HNE (p < 0.  . Forest plots assessing the effects of NM exposure on the level of TNF-α in different samples compared with the non-exposed control group. NMs, nanomaterials; EBC, exhaled breath condensate; TNF-α, tumor necrosis factor-α; SMD, standardized mean difference; CI, confidence interval [17,19,25,36,39,40]. The negative effects on the levels of IL-5, LT-C4, LT-D4 and neutrophils were maintained after correction. Sensitivity analyses showed that pooled estimates remained in the same directions when the studies were omitted one by one, suggesting the stability and reliability of this meta-analysis and the results were not influenced by any one study ( Figure 12). same directions when the studies were omitted one by one, suggesting the stability and reliability of this meta-analysis and the results were not influenced by any one study (Figure 12). Sensitivity analysis for TNF-α. The vertical axis shows the omitted study. The horizontal axis represents the corresponding pooled estimate when the study is excluded. Every circle indicates the pooled SMD (all of them are close to 1.52 obtained in the overall meta-analysis). The two ends of every broken line represent the respective 95% CI. TNF-α, tumor necrosis factor-α; SMD, standardized mean difference; CI, confidence interval [17,19,25,36,39,40].

Discussion
There have been studies stating the potential relationships of oxidative stress biomarkers [47,48] and FENO [49] with occupational exposure to NMs, but all of them only reviewed the results of known individual articles. No meta-analyses have been conducted to synthesize all data from each study to overcome the low statistical power and achieve a comprehensive and reliable conclusion. In the present study, we, for the first time, included 28 epidemiological studies with 2636 participants and performed a meta-analysis to examine the effects of NM exposure on oxidative stress and inflammatory biomarkers. After overall analysis, subgroup meta-analyses, trim-and-fill adjusted estimates and the sensitivity analysis, we found occupational NM exposure was significantly associated with increases in the levels of MDA, HNE, aldehydes C6-12, 8-OHG, 5-OHMeU, o-Tyr, 3-NOTyr, IL-1β, TNF-α, MPO, fibrinogen, and decreases in the levels of GPx and IL-6sR.
Existing evidence from in vitro and in vivo studies supported that NM inhalation exposure can induce excessive production of reactive oxygen (ROS, including super oxides, superoxide radicals, hydroxyl radicals and hydrogen peroxide) and reactive nitrogen species (RNS, including peroxynitrite anion and nitric oxide) [50,51]. The ROS could react with the chain polyunsaturated fatty acids in the membrane to trigger the process of lipid peroxidation, resulting in the release of reactive, toxic aldehydes, including MDA, HNE and aldehydes C6-C12 [52]. ROS can attack DNA and RNA to induce oxidative modifications on guanine or thymine bases, leading to the generation of 8-OHdG, 5-OHMeU (both were DNA damage biomarkers) and 8-OHG (RNA damage biomarker) [53]. Hydroxyl radicals can oxidize phenylalanine into o-Tyr, and RNS can mediate the nitration of p- Figure 12. Sensitivity analysis for TNF-α. The vertical axis shows the omitted study. The horizontal axis represents the corresponding pooled estimate when the study is excluded. Every circle indicates the pooled SMD (all of them are close to 1.52 obtained in the overall meta-analysis). The two ends of every broken line represent the respective 95% CI. TNF-α, tumor necrosis factor-α; SMD, standardized mean difference; CI, confidence interval [17,19,25,36,39,40].

Discussion
There have been studies stating the potential relationships of oxidative stress biomarkers [47,48] and FENO [49] with occupational exposure to NMs, but all of them only reviewed the results of known individual articles. No meta-analyses have been conducted to synthesize all data from each study to overcome the low statistical power and achieve a comprehensive and reliable conclusion. In the present study, we, for the first time, included 28 epidemiological studies with 2636 participants and performed a meta-analysis to examine the effects of NM exposure on oxidative stress and inflammatory biomarkers. After overall analysis, subgroup meta-analyses, trim-and-fill adjusted estimates and the sensitivity analysis, we found occupational NM exposure was significantly associated with increases in the levels of MDA, HNE, aldehydes C6-12, 8-OHG, 5-OHMeU, o-Tyr, 3-NOTyr, IL-1β, TNF-α, MPO, fibrinogen, and decreases in the levels of GPx and IL-6sR.
Existing evidence from in vitro and in vivo studies supported that NM inhalation exposure can induce excessive production of reactive oxygen (ROS, including super oxides, superoxide radicals, hydroxyl radicals and hydrogen peroxide) and reactive nitrogen species (RNS, including peroxynitrite anion and nitric oxide) [50,51]. The ROS could react with the chain polyunsaturated fatty acids in the membrane to trigger the process of lipid peroxidation, resulting in the release of reactive, toxic aldehydes, including MDA, HNE and aldehydes C6-C12 [52]. ROS can attack DNA and RNA to induce oxidative modifications on guanine or thymine bases, leading to the generation of 8-OHdG, 5-OHMeU (both were DNA damage biomarkers) and 8-OHG (RNA damage biomarker) [53]. Hydroxyl radicals can oxidize phenylalanine into o-Tyr, and RNS can mediate the nitration of p-tyrosine to form 3-NOTyr [54]. Thus, the increases in these ROS/RNS-related indicators were speculated to be associated with the oxidative stress damages induced by NM exposure. This hypothesis had been demonstrated by some authors. For example, the summary analysis of 49 data by An et al. showed that the level of MDA in model rats or mice was increased by 5.52-fold after TiO 2 NP exposure compared with the controls [12]. A short-time exposure of silver NPs was reported to induce the formation of 4-HNE-protein adducts in SUM159 cells to drive cell death [55]. Compared to non-exposed cells, cells exposed to palladium NPs were observed to have an increased accumulation of nuclear acid damage biomarkers, especially 8-OHG (the level of which seemed to be higher than that of 8-OHdG, indicating RNA damage may be more severe) [56]. Consistent with these model studies, we also identified that the levels of MDA, HNE and 8-OHG were significantly increased in occupationally NM-exposed workers. Also, the increase fold of 8-OHG (SMD = 2.98) was higher than 8-OHdG (SMD = 1, which was even non-significant in the subgroup analysis) and 5-OHMeU (SMD = 1.9). Aldehydes C6-C12, 3-NOTyr and o-Tyr were only measured in NM-exposed workers [15,16,29,44], not in cell and animal studies. However, in line with the predicted theory, our meta-analysis also confirmed positive correlations of these three biomarkers with NM exposure. GPx is an enzymatic free radical scavenger that could protect the body from ROS-induced damages. The accumulation of ROS and RNS after NM exposure was attributed to a reduced level and inactivation of GPx [57,58]. Similar to these cell studies, our meta-analysis identified a disturbance of GPx in NM-exposed workers.
Other than oxidative stress, activation of inflammation is one of the key mechanisms involved in NM-related hazardous effects [13]. NM-induced inflammatory response, on one hand, may be a result mediated by excessive ROS via activation of the mitogen-activated protein kinase-NF-κB signaling pathway [59,60]; on the other hand, it may be associated with activation of ROS-independent hypoxia-inducible factor signaling pathways [61]. Main biomarkers reflecting the inflammatory response are various cytokines released by immune cells, such as IL-1β, TNF-α and IL-6sR. It has been reported that cells or rats exposed to NMs exhibited upregulations of IL-1β and TNF-α; inhibitions of these two biomarkers reversed NM exposure-induced cell death and pathological changes in tissues [62,63]. In agreement with these studies, we also detected statistically significant increased levels of IL-1β and TNF-α in NM-exposed employees. IL-6sR was proved to bind with soluble gp130 to form a complex that inhibited IL-6 trans-signaling-mediated pro-inflammatory effects [64]. The level of IL-6sR was shown to be lower in inflammatory diabetes patients compared with healthy subjects [65]. Theoretically, IL-6sR was downregulated in NM-exposed workers, which was confirmed in our meta-analysis. Although MPO and fibrinogen are not cytokines, accumulated studies have demonstrated their links with inflammation-related diseases, including NM exposure [66,67]. MPO, a heme enzyme expressed in neutrophils, monocytes and macrophages, was suggested to drive inflammation due to its roles in catalyzing the oxidation reaction to generate ROS [68]. Fibrinogen was shown to directly stimulate the production of cytokines by the mitogen-activated protein kinase-NF-κB signaling pathway [69]. Similar to these findings, we also verified the levels of MPO and fibrinogen were significantly higher in NM-exposed workers than those in non-exposed populations.
The oxidative stress and inflammatory biomarkers were measured in multiple specimen types, including EBC, blood, urine, sputum and nasal lavage. Subgroup analysis indicated relative to urinalysis (negative results or small SMD), oxidative stress biomarkers (MDA, HNE, aldehydes C6-12, 8-OHG, 5-OHMeU, o-Tyr, 3-NOTyr and GPx) in EBC and blood samples were significantly or strongly changed by NM exposure. IL-1β, TNF-α, MPO, fibrinogen and IL-6sR were all detected in EBC, blood, sputum and nasal lavage samples and their results were all significant. These findings reflected the fact that NMs may enter into the respiratory tract, blood, other inner organs and urine successively after long-term exposure [70]. The NMs may be deposited in the human body, but not cleared and then released into the urine, or eliminated mainly by mucociliary clearance and ingested [71], ultimately contributing to slight or non-significant changes of these indicators in the urine samples. Furthermore, the negative results of oxidative stress biomarkers in urinary samples may also be associated with the following reasons: (1) it has been reported that there is a time window to detect the responses of urinary biomarkers after exposure. If the sample is not collected in the sensitive time windows, the indicators may be seldomly changed after exposure [29,72]. For example, Zhang et al. detected slower generation and/or urinary excretion kinetics of urinary 8-OHG and HNE in NM-exposed workers. These two biomarkers were only significantly increased at 36 h (an acute model) or three weeks (a chronic model) post-exposure. o-Tyr and 5-OHMeU were not statistically elevated in the longest sampling time points [29]; (2) the elimination half-lives of oxidative stress biomarkers in plasma were also found to be longer than that in urine [73], which may result in their low levels in urinary samples even if the same sampling time points were set as the blood samples; (3) some oxidative stress biomarkers (e.g., HNE and other aldehydes) are intermediary oxidation products which can be fed by a precursor/parent molecule, but consumed by subsequent oxidation reactions or adduct formation. Hereby, the overall concentrations for themselves in urine may be not high; and (4) liquid chromatography electrospray ionization-mass spectrometry/mass spectrometry was used for analyses of urinary oxidative stress biomarkers in all included studies. Although being sensitive relative to the enzyme-linked immunosorbent assay [74,75], it may be still not be the optimal technique for analysis of urinary samples and more new generations of mass spectrometers [76,77] should be applied to further confirm the results in urinary samples to avoid technique-derived deviations. More interestingly, there were published literatures that suggested urinary oxidative stress indicators were excellent biomarkers to identify persons exposed to toxic elements (diagnostic accuracy of 3-NOTyr = 0.753) [78] and predict the hazard effects [79]. The levels of urinary and circulating oxidative stress markers (e.g., 8-OHdG [32], MDA [80]) in occupationally exposed workers were also observed to be significantly correlated. Thus, the biomarker roles of oxidative stress indicators in urine samples should not be discounted and need to be confirmed by designing better experimental protocols in the future (including considering the sampling time windows and the analytical methods).
Several limitations should be addressed. First, the number of available publications and the sample size in each study were small; thus, the results of some indicators may be still inconclusive (such as MCP-1 and MIP-1β in overall analysis as well as FENO in the subgroup analysis). Second, this meta-analysis only preliminarily estimated the association of NM exposure with the oxidative stress and inflammatory biomarkers. The specific threshold values that distinguish NM-exposed populations and normal controls (or pre-and post-exposure) and predict the poor outcomes remain unclear for these biomarkers [78,81]. Which one (or which combination) [82] is the optimal biomarker remains under-investigated. These two issues need to be resolved by the receiver operating characteristic curve analysis and multivariate regression analysis [78,79,81,82]. The reference ranges of significant oxidative stress and inflammatory biomarkers we identified also should be calculated to better explain the biomarker roles of them in different disease settings [74,75,83,84]. Third, meta-analysis was performed based on the mean and SD data of exposed and non-exposed groups collected from each study, not the adjusted results for potential confounders (such as sex, age, smoking or drinking) because they were only provided by some studies and the confounders were different across the studies. Fourth, we excluded the self-control studies (that is, comparisons between before and after exposure) because the exposure time was relatively short (only some hours, which may be meaningless for assessment of long-term exposure effects) and the indicators were different among them (leading to only few results that could be combined). Accordingly, to further confirm biomonitoring effects of our identified biomarkers, more studies that prospectively include workers (just into the factory, previously not exposed to NMs) and assign them to exposed and non-exposed workplaces followed by detecting the levels of biomarkers in multiple follow-up time points are needed.

Conclusions
Our meta-analysis suggests that oxidative stress indicators (MDA, HNE, aldehydes C6-12, 8-OHG, 5-OHMeU, o-Tyr, 3-NOTyr and GPx) in EBC and blood samples, as well as inflammatory mediators (IL-1β, TNF-α, MPO, fibrinogen and IL-6sR) in EBC, blood, sputum and nasal lavage samples were significantly associated with NM exposure. They may represent potential biomarkers for the biological monitoring of the population exposed to NMs at the workplace. The biomarker roles of oxidative stress indicators in urinary samples need to be confirmed by designing experiments with different sampling time points and new analytical methods.