Non-Invasive Measurement of Exercise-Induced Oxidative Stress in Response to Physical Activity. A Systematic Review and Meta-Analysis

Physical activity may benefit health by modulating oxidative stress and inflammation. However, the selection of suitable exercise-induced oxidative stress biomarkers is still challenging. This study aimed at systematically summarizing the available evidence on exercise-induced oxidative stress measured in urine and/or saliva. Two meta-analyses including the most frequently quantified biomarkers of oxidative stress, namely, urinary isoprostane and DNA oxidation products, were performed. Three electronic databases (PubMed, EMBASE and Cochrane CENTRAL) were interrogated. Among 4479 records, 43 original articles were included in the systematic review and 11 articles were included in meta-analysis I and II, respectively. We observed a pooled trend of increase of urinary isoprostanes in response to physical activity (+0.95, 95% CI: −0.18; 2.09). In comparison with aerobic exercise, anaerobic training determined a greater induction of isoprostanes (+5.21, 95% CI: 2.76; 7.66, p < 0.0001), which were markedly increased after vigorous physical activity (+6.01, 95% CI: 1.18; 10.84, p < 0.001) and slightly decreased in response to exercise interventions protracted over time (e.g., months) (−1.19, 95% CI: −2.25; −0.12, p < 0.001). We recommend the most integrative approach of oxidative stress multi-marker panels in response to physical activity instead of selecting one preferential biomarker to quantify physical activity-induced oxidative stress in humans.


Introduction
In the last few decades, the scientific interest in physical activity-induced oxidative stress has been fuelled by three complementary concepts. First, physical activity helps prevent chronic diseases and improve health [1]. Second, oxidative stress is involved in the pathogenesis [2] or represents a downstream consequence of several diseases [3]. Third, physical activity influences cellular redox homeostasis and, thus, oxidative status in humans [4]. Since the pioneering discovery that lipid peroxidation biomarkers increase in subjects following acute exercise [5,6], the understanding of the exercise-induced oxidative stress was further extended by the introduction of some key scientific discoveries including (1) the involvement of pro-oxidants species in the production and modulation of muscles force [7]; (2) the dose-response effect of training on primary antioxidant levels in cardiac and skeletal muscle [8]; (3) the contribution of nitric oxide in muscle vasodilatation [9] and its production in contracting muscles [10]. Overall, this body of evidence laid the foundations for applying the theory of hormesis to exercise-induced oxidative stress [11]. In general terms, hormesis has been defined as a "process in which a low dose of a chemical agent or environmental factor that is damaging at high doses, induces an adaptive groups (e.g., sex, age classes, smoking habits, physical activity intensities), all data were extracted. Biomarker variations described throughout this systematic review are in reference to the resting condition (i.e., baseline). If biomarker measurements at different time points were available, only the first time point following physical activity was extracted. Data on urinary isoprostane were generally reported as "isoprostanes" because several articles did not specify if free, total or specific isomers of isoprostane were analysed. Data originally presented by graphs were extracted by the Web plot digitizer software (https://apps.automeris.io/wpd/).

Quality Assessment
The risk of bias (quality) assessment was appraised using three specific tools based on the study design of the included researches, namely: (1) the National Institute of Health (NIH) Quality Assessment Tools for observational, case-series, cross-sectional and beforeafter studies were used for a critical appraisal of the internal validity of the studies [21]; (2) the PEDro scale (available at: https://pedro.org.au/english/resources/pedro-scale/ Accessed on 1 July 2020) to appraise the randomised controlled trials; and (3) the Johanna Briggs Institute checklist [22] was used to check the methodological quality of the quasiexperimental trials (i.e., non-randomised). Since each tool adopts different ratings, we expressed our quality rating as a percentage and the quality score underwent re-coding based on the tertiles (1st tertile = poor quality; 2nd tertile = medium quality; 3rd tertile = high quality).

Statistical Methods
Continuous variables were summarised and reported as mean ± standard deviation (SD) or median ± inter quartile range (IQR) or mean ± standard error of the mean (SEM). Methods in [23] were used to approximate the SD from the sample size, median and IQR.
For each study, the mean change from baseline was computed. Since the correlation coefficient r between post score and pre score is needed for computing the standard error, the value r = 0.7 was imputed as suggested by Rosenthal [24]. Then, effect sizes were computed as standardised mean differences based on the Hedges' g method.
To estimate the pooled effect of the physical activity intervention on oxidative stress biomarkers a random-effect meta-analytic model with the DerSimonian-Laird estimator (inverse variance method) was used. The average effect size and a 95% Confidence Interval (CI) were computed by the Jackson method. The heterogeneity among the studies was inspected by the Cochran's Q test and the Higgins I 2 statistics.
Publication bias was assessed by visual inspection of funnel plots and by carrying out the Eggers' test. Meta-regression models were built to check the influence of (i) the quality of the included studies; (ii) types of physical activity, (iii) duration of physical activity (i.e., "acute" or "chronic") and (iv) intensities of physical activity on the relationship between oxidative stress biomarker and physical activity. Results were expressed as regression coefficients (95% CI).
We performed a set of sensitivity analyses to: (i) identify influential studies that resulted in variation, using graphic display of heterogeneity plots, which fit the same meta-analysis model for all the possible study combinations and look for specific patterns performing clustering with k-means, DBSCAN and Gaussian mixed models [25]; (ii) to check for the outliers and the influence of each included article on the overall heterogeneity and (iii) to evaluate the studies that were more contributing to the heterogeneity, previously identified by the Baujat diagnostic and plot. Studies reporting spot urine biomarkers without any normalisations to creatinine and/or data referred to subgroups not relevant to the research question (e.g., smokers versus non-smokers) were excluded from the metaanalysis. All the analyses have been performed using R, version 4.0.2 [26]
research question (e.g., smokers versus non-smokers) were excluded from th ysis. All the analyses have been performed using R, version 4.0.2 [26] 3. Results

Study and Participant Characteristics
Sample sizes ranged from 5 to 98 subjects. Overall, 957 adults aged bet 72 years (39.8 ± 18.2 years) were included in the systematic review and 70% Step-by-step study selection process. Modified from: [20]. Table 1 presents a summary of the included studies. The majority of the studies (47%) were uncontrolled experiments employing a before-after design (n = 20), while a cumulated 49% consisted of controlled before-after, longitudinal and randomised controlled trials (n = 7, respectively). One study used randomised cross-over design and another was self-controlled case-series. Studies were mainly located in Japan (n = 4), Brazil (n = 4), Italy (n = 4), Spain (n = 4) and USA (n = 4) followed by Canada (n = 3), Netherlands (n = 3), Iran (n = 2) and Denmark, Egypt, Germany, Greece, India and Mexico (n = 1, respectively). Nine studies did not clearly state the study location. The studies were published from 1993 to 2019.    Sample sizes ranged from 5 to 98 subjects. Overall, 957 adults aged between 19 and 72 years (39.8 ± 18.2 years) were included in the systematic review and 70% of them were males (n = 671). A total of 60% of subjects were healthy, 10% were not classified, while a cumulated 30% reported a disease among diabetes or obesity (13%), respiratory diseases (9%) and a miscellaneous of other pathologies including arthritis rheumatoid, cancer and periodontitis (8%). Subjects were generally active (41%) or professional athletes (24%), whereas 33% of them reported a sedentary lifestyle and 2% did not provide any details. Only 27% of them performed low-intensity physical activity and 42% were engaged in high-intensity or medium-intensity (25%) physical activity protocols (for 6% of them the physical activity intensity has not been specified). A total of 5% of subjects underwent both aerobic and anaerobic exercises, while the remaining followed aerobic (67%) or anaerobic (27%) protocols. Table 2 summarises key characteristics of biomarkers of oxidative stress. The two most investigated oxidative stress biomarkers were urinary isoprostanes and DNA oxidation products (i.e., 8-oxo-dG or 8-OH-dG). In particular, 44% of the studies focused on DNA oxidation products (n = 19), 40% analysed isoprostanes (n = 17) and the remaining 16% included a large variety of other biomarkers: (i) measured in saliva, such as peroxidase, lipid hydroperoxides, superoxide dismutase, catalase, total antioxidant status or capacity, advanced oxidation protein products, glutathione, vitamin C and Uric Acid (UA); (ii) measured in urine, such as allantoin, hydrogen peroxide and urate; (iii) measured in both urine and saliva such as malondialdehyde (MDA). Table 2. Biomarkers of oxidative stress measured in urine and/or saliva at baseline and after physical activity intervention. Data are presented as originally reported by the studies as follows: * mean (SEM), † median (IQR), § mean (SD).      A total of 84% of the studies quantified oxidative stress biomarkers in urine, either using spot urine (n = 21), either 24 h (n = 11) or 12 h urine (n = 4), while eight articles out of 43 used saliva specimens (19%). The most widely performed analytical technique was ELISA (49% of the studies), followed by HPLC (26%), while the remaining 25% of the studies used radioimmunoassay (n = 2), GC-MS, tandem mass spectrometry and ultra-performance liquid chromatography and flow cytometry (n = 1, respectively) or other analytical techniques generally defined as "colorimetric", "spectrophotometric" or "enzymatic".

The Effect of Physical Activity on Oxidative Stress Biomarkers in Saliva
Salivary biomarkers followed very heterogeneous patterns, increasing or decreasing after physical activity (Figure 2). Refs. [29][30][31]34,45,58,68] reported significant changes due to physical activity. In [29], salivary peroxidase and UA were significantly increased in a group of non-smoker females (n = 12, age = 22.7 ± 2.9) who performed the Bruce protocol treadmill test until exhaustion. Participants were healthy and were abstaining from exercise for three months. Salivary SOD and TAS significantly increased after a six-month intervention of Thai Chi (5 days per week) involving 24 sedentary volunteers, aged between 60-74 years, who were diagnosed with periodontal disease [45]. [30] examined the effect of 1 h of exhaustive treadmill running on CAT and Vitamin C, finding a significant decrease in saliva shortly after exercise in 25 healthy sedentary males (21 ± 3 years). Salivary TBARS significantly increased in [31,68], both involving professional soccer players. [68] measured TBARS in salivary samples provided by eight athletes (males, 27.2 ± 5.5 years) after a 90 min soccer game. [31], instead, investigated TBARS changes in 27 males aged 22.5 ± 4.2, engaged in a supervised anaerobic training protocol. As previously observed for [29], UA has significantly increased also in 11 healthy and well-trained males aged 25.9 ± 2.8 years who completed an experimental resistance exercise protocol, after a rest period of 72 h [34]. In contrast, [58] reported that salivary UA decreased in 32 soccer players (21.2 ± 4.2 years) after the Bangsbo Sprint Test [70].

Quality of the Studies
Out of 43 studies, 25 (58%) were classified as "high quality", comprising mainly beforeafter design (n = 16), controlled before-after (n = 5), RCTs (n =3) and randomised cross over (n = 1). On the contrary, 18 articles were identified as potentially affected by the risk of bias and accordingly classified as "medium quality". They were based on a longitudinal design (n = 7), before-after or RCTs (n = 4, respectively), controlled before-after (n = 2) and case series self-controlled (n = 1). Specific quality domains were identified as potentially biased. In particular, in 90% of the studies with a before-after design, researchers assessing the outcomes were not blinded to the participant's exposures or interventions. In 79% of the before-after studies, outcome measures of interest (i.e., oxidative stress biomarkers) were not taken multiple times before the intervention, whereas several studies measured oxidative stress biomarkers multiple times after the intervention. The majority of the observational studies (88%) did not provide any sample size justifications nor statistical power descriptions. None of the-observational studies examined physical activity at different levels (e.g., subgroups by intensity); however, it is worth mentioning that 63% of them analysed oxidative biomarker changes during competitive races. Subjects' allocation was not concealed in the majority of the RCTs (88%) and 88% of the RCTs, there was no blinding of the subjects neither of the researchers who administered the training protocol.
ntioxidants 2021, 10, x FOR PEER REVIEW Figure 3. Pooled effect of physical activity interventions and oxidative stress measu 8-OH-dG or 8-oxo-dG. Note: Data presented as sub-groups "A" and "B" refer to mod physical activity and high-intensity physical activity, respectively. Data presented "A", "B" and "C" refer to running on a treadmill until exhaustion, cycling until e running for 20 km, respectively. Pooled effect of physical activity interventions and oxidative stress measured by urinary 8-OH-dG or 8-oxo-dG. Note: Data presented as sub-groups "A" and "B" refer to moderate-intensity physical activity and high-intensity physical activity, respectively. Data presented as sub-groups "A", "B" and "C" refer to running on a treadmill until exhaustion, cycling until exhaustion and running for 20 km, respectively. Figure 3. Pooled effect of physical activity interventions and oxidative stress measu 8-OH-dG or 8-oxo-dG. Note: Data presented as sub-groups "A" and "B" refer to mod physical activity and high-intensity physical activity, respectively. Data presented "A", "B" and "C" refer to running on a treadmill until exhaustion, cycling until e running for 20 km, respectively. Substantial heterogeneity was detected in both meta-analyses (I 2 = 97% 98%, p < 0.001, respectively). Refs. [36,52] were identified as outliers and r DNA oxidation products meta-analysis and, although the heterogeneity r high (I 2 = 95%, p < 0.001), physical activity was significantly associated with of 8-oxo-dG or 8-OH-dG (−0.68, 95% CI from −1.37 to 0.00). We explored the of each study to the overall heterogeneity, detecting that [27,53,54] contrib in the meta-analysis on DNA oxidation products and [42] in the meta-analys tanes. However, the subsequent leave-one-out analysis did not show any ductions in heterogeneity, even after removing those studies identified as gr utors in terms of heterogeneity. No publication bias was found in any meta 0.88 and p = 0.44, respectively) (Figures S1 and S2 in the supplementary mat several meta-regressions were carried out accounting for the potential effec dictors. Substantial heterogeneity was detected in both meta-analyses (I 2 = 97%, p < 0.001; I 2 = 98%, p < 0.001, respectively). Refs. [36,52] were identified as outliers and removed from DNA oxidation products meta-analysis and, although the heterogeneity remained very high (I 2 = 95%, p < 0.001), physical activity was significantly associated with the decrease of 8-oxo-dG or 8-OH-dG (−0.68, 95% CI from −1.37 to 0.00). We explored the contribution of each study to the overall heterogeneity, detecting that [27,53,54] contributed the most in the meta-analysis on DNA oxidation products and [42] in the meta-analysis on isoprostanes. However, the subsequent leave-one-out analysis did not show any significant reductions in heterogeneity, even after removing those studies identified as greater contributors in terms of heterogeneity. No publication bias was found in any meta-analyses (p = 0.88 and p = 0.44, respectively) ( Figures S1 and S2 in the supplementary material). Finally, several meta-regressions were carried out accounting for the potential effect of other predictors.

Discussion
Physical activity can prevent several non-communicable diseases [71] and contribute to ameliorating the quality of life [72]. These peculiar hallmarks strengthen the role of physical activity in public health and continue stimulating research efforts. Although modern redox biology has done great strides, the understanding of the modulation of oxidative stress by physical activity and exercise is still incomplete. Epidemiologic approaches and large-scale studies could support further research in this field; however, a general hint on one or more preferential biomarkers to quantify exercise-induced oxidative stress in non-invasive media is still lacking.
We observed that oxidation products and antioxidant species were the most frequently used physical activity-induced oxidative stress biomarkers in urine and saliva. The first group was preponderant in urine samples, covering a substantial percentage of the totality of the articles (more than 86%), while antioxidant species were predominantly quantified in saliva. Although saliva may represent an optimal non-invasive media, especially when dealing with a large population and/or uncooperative subjects, literature is scarce and no conclusive evidence can be drawn about salivary oxidative stress biomarkers quantification. Salivary biomarkers have shown an extremely heterogeneous pattern indicating augmentation, drop, or unchanged levels after physical activity even for the same biomarker (i.e., UA, TBARS) in different studies.
Conversely, there is plenty of literature on biomarkers quantified in urine. Urinary isoprostanes and DNA oxidation products were the two most frequently quantified biomarkers of oxidative stress. The association between DNA oxidation products and physical activity did not reach the significance level, discouraging any conclusive interpretations on 8-oxo-dG or 8-OH-dG modulation by exercise. Isoprostanes showed a general trend of the increase due to physical activity interventions, suggesting that changes in urinary isoprostanes might be successfully detected in urine after exercise; hence, isoprostanes might represent useful urinary biomarkers, although the weakness of the association and lack of homogeneity indicate that further research is needed.
Previous literature acknowledged physical activity as a potent inducer of lipid peroxidation, both in humans and in animals [73][74][75][76]. In a previous literature review [77], F2-IsoP, measured in plasma and skeletal muscles, increased after acute exercise, whereas urinary levels were generally increased but this trend required further confirmation. Similarly, [78] summarised the evidence from two studies reporting that isoprostane levels were increased after acute and intense exercise, either in skeletal muscles or in plasma, respectively. In another review [79], Sacheck and colleagues observed that plasma isoprostane levels were increased in horses after a treadmill test, were unaffected by 8 weeks of moderate/low exercise in subjects with type 2 diabetes, or even decreased in urine from trained rats after eccentric muscle exercise. On the contrary, a recent systematic review [71] stated that plasmatic isoprostane is generally reduced after an exercise-training period in both elderly and young subjects and the same result was observed for urinary isoprostane, but only when accompanied by relatively marked gains in aerobic fitness.
Our findings highlighted that anaerobic physical activity induced a greater increase of urinary isoprostanes than aerobic exercise. Although, out of 11 studies, only [57] and [42] applied anaerobic interventions with medium and high intensity, respectively. Noticeably, also aerobic interventions determined an oxidative stress augmentation, which was smaller than that observed after anaerobic exercise and exclusively related to high-intensity aerobic exercise protocols.
In terms of physical activity intensity, our findings suggest that urinary isoprostanes were markedly reduced after both low and moderate intensity of exercise compared to strenuous physical activity. Previous literature reviews suggest similar results: high intense and prolonged aerobic exercise [80] and even anaerobic exercise [81][82][83], have been associated with greater ROS production, thus, oxidative stress [84].
Urinary isoprostanes were generally reduced after physical activity interventions protracted over time, suggesting that regular exercise training could act as an antioxidant at least against isoprostane formation in vivo. On the contrary, we observed that isoprostanes augmented after an acute bout of physical activity, supporting the hypothesis that physical activity could act also as a stressor. These findings are in line with the review published by Nikolaidis in 2011, who reported for the first the comparison between acute and chronic exercise on isoprostanes measured in three different specimens: plasma, urine and skeletal muscle [77]. The underlying mechanism invokes the in vivo upregulation of antioxidant levels, which can be enhanced by regular exercise training that acts as an adaptive stimulus [75,85].
The physiological mechanisms supporting the general finding that oxidative stress can increase after physical activity are fairly well-accepted and understood. During physical activity, several pathways are involved in ROS overproduction: (1) aerobic metabolism and electron chain, which increases the leakage of superoxide radicals due to the increased oxygen consumption while exercising; (2) anaerobic exercise, in turn, can activate different pathways including NADPH oxidase, xanthine oxidase, ischemia-reperfusion, purine oxidation and catecholamine auto-oxidation [81,86].
A general remark should be outlined as each oxidative stress biomarker presents advantages and disadvantages [75] and can be affected by a multitude of other factors, including physical activity type and duration [75] as well as the timing of sampling [87] and inter-individual variability [88]. All these aspects have been conducted to a general and well-supported indication mainly referring to the assessment of a set of biomarkers instead of a preferential one. Therefore, although our results slightly support the quantification of isoprostanes in urine when dealing with exercise-induced oxidative stress investigations, we recommend the most integrative approach of multi-marker panels. Urine remains a suitable non-invasive medium with low metal and organic content and whose collection is easy and cost-effective [18,19,89]; however, biomarkers quantification deserves further harmonisation in terms of sampling timing, type of specimen (spot, 12 h and 24 h urine), normalisation to creatinine, analytical method and statistical reporting of results.
To reduce possible drawbacks during the sampling and handling of urinary samples we suggest the following recommendations. First, each subject's clinical status must be investigated, since renal and/or bladder impairments together with other pathologies able to indirectly affect renal/bladder functionality (e.g., diabetes), can contribute to local biomarker formation, altering the total amount attributable to other sites. Second, creatinine needs to be quantified to normalise spot urines, to evaluate kidney efficiency and to reduce intra and inter variability among subjects. A third point should be extended to all biological samples, which all requires a careful control of the temperature during sampling, handling and long-term storage to avoid auto-oxidation. In addition, comparative studies, focusing on different analytical methods, could provide specific corrective factors to support the harmonisation between different analytical techniques. Since, beyond the specimen characteristics, the paramount diversity in analytical techniques and nomenclature systems adopted by different laboratories hamper comparisons among studies. Although chemical techniques are considered superior to immunological techniques based on their higher sensitivity and specificity [90], they are time-consuming and expensive. Therefore, ELISA are becoming more popular, as confirmed from this systematic review. Irrespective from ELISA's poor correlation with chemical assays and their lower specificity, they are cost-effective and may support epidemiological and large-scale studies.
Future development in exercise-induced oxidative stress evaluation in non-invasive media should take into account: (1) emerging media such as saliva, which is a promising non-invasive medium that still requires protocols standardisation [91] in terms of sampling timing and analysis protocols and techniques; (2) physical activity duration, intensity and type as well as sample collection timing, handling and storage can considerably affect as ROS induction as oxidative stress quantification [19]; (3) the intra-individual variability imposes multiple time points assessment of oxidative stress biomarkers, especially at baseline (i.e., before physical activity intervention); (4) low sample sizes, non-blinded researchers and non-concealed allocation of the participants enrolled in randomised trials should be avoided and (5) the quantification of exercise-induced oxidative stress biomarkers should be always complemented with the anti-oxidant counterparts, which is intimately linked to oxidative status and strongly affected by physical activity itself.
As a main strength, we acknowledge that the present study systematically reviews and meta-analyses, for the first time, exercise-induced oxidative stress biomarkers in noninvasive media. The included studies were generally scored highly in terms of quality; thus, the overall risk of bias was low. One of the main limitations is the substantial heterogeneity that has been observed in the meta-analyses. Although we performed a set of sensitivity analyses, none of them were able to elucidate such a heterogeneity; thus, meta-analysis results should be taken with cautiousness. Therefore, we recommend an integrative approach that involves multi-marker panels of oxidative stress to accurately assess the effect of physical activity on oxidative stress levels.

Conclusions
Despite the wide heterogeneity among a large set of oxidative stress biomarkers quantified in both urine and saliva, the present meta-analysis concluded that urinary isoprostanes seem more prone to physical activity modulation. Indeed, we observed that physical activity could elicit an increase of urinary isoprostanes that is greater after anaerobic exercise compared to aerobic one. In addition, low to moderate physical activity seems to evoke a reduction of urinary isoprostanes compared to strenuous exercise as observed for long-lasting training versus single acute bouts. Conversely, no conclusive results have been observed nor for DNA oxidation products quantified in urine after physical activity, for other oxidative stress biomarkers quantified both in urine and in saliva.