Datasets resulting from the review process aiming to quantify the use of markers of oxidative status in animal welfare studies and in adjacent fields focused on wild animals (ecophysiology, conservation physiology). Datasets used for the metanalyses aiming to examine variation in markers of oxidative status across three conditions associated with a negative valence (social isolation, noise exposure, predation exposure).

Review:

The reviewing process was conducted in journals assumed to be representative of their respective research fields (animal welfare, ecophysiology, conservation physiology) and actively publishing research in the last decade (2013-2022) to reflect current trends. Only research articles or short communications were considered in the reviewing process to reflect active research in each field (as opposed to reviews, commentaries, or perspective articles). Plant articles were also excluded. Because of the overrepresentation of ecophysiological studies relative to other fields and because such studies can often be found in journals also publishing exclusively physiological studies, the reviewing process of ecophysiological studies was limited to journals with a clear ecophysiological section. Biology Letters does not have an ecophysiological section but, because of the strong evolutionary and ecological focus of this journal, its Physiology section was considered as an ecophysiological section. The reviewing process was also limited to the Animal Welfare section of the journal Animals publishing articles in a variety of other disciplines. For general conservation and welfare journals (i.e., not primarily focusing on physiology), only articles including physiological parameters were considered to reflect the fields of conservation physiology and animal welfare physiology.

The advanced search tool on the website of each journal was used to search for the keywords: “oxidative stress”, “oxidative status”, and “antioxidant”.

For each article, the conditions considered to be susceptible to affecting markers of oxidative status were assigned to the following 27 categories: condition (physical/clinical condition), infection (infection, parasitism), activity (physical activity), restraint, density, predictability, predation, noise, temperature, radioactive radiation, hypoxia/reoxygenation, UVB exposure, salinity, pH, chemicals, environment (environment physical properties other than those detailed in the list), time (year, season), location, urbanity/tourism, water deprivation, food availability/supplementation, antioxidant supplementation, reproduction, migration, social aspects, isolation, growth/hatching asynchrony.

For each article, the considered animal species were classified in the following taxa: mammals, birds, reptiles, amphibians, fish, insects, crustaceans, mollusks, helminths and cnidaria.

The number of biomarkers of oxidative status was also quantified in each article and arranged in the following categories:

-        Reactive Oxygen Species (ROS)/pro-oxidants: superoxide anion, hydrogen peroxide (H₂O₂), xanthine oxidase, labile plasma anion,

-        Antioxidants: antioxidant capacity, superoxide dismutase (SOD), catalase (CAT), glutathione peroxidase (GPX), glutathione S-transferase (GST), glutathione reductase (GR), glucose-6-phosphate dehydrogenase (G6PDH), glutathione (GSH)/thiols, uric acid (UA), aryl hydrocarbon receptor (AHR), vitamin E (Vit. E), carotenoids (Car.), omics,

-        Oxidative damage: MDA/lipid peroxidation, hydroperoxides (HP), protein carbonyls (PC), DNA damage.

Some of these markers were not necessarily measured using the same techniques (e.g., antioxidant capacity: FRAP, TEAC, ORAC, BAP, OXY; hydroperoxides: HYDROP, ROM; DNA damage: 8-OHdG, comet assay).

Meta-analysis:

The reviewing process was conducted by searching in Google Scholar for articles on oxidative markers measured in response to three conditions (social isolation, noise exposure, predation) by entering the following search terms:

-        (1) “oxidative stress” “social isolation”

-        (2) “oxidative stress” “noise”

-        (3) “oxidative stress” “predation”

Despite its practicability, Google Scholar may not necessarily be recommended as a sole literature source to conduct meta-analyses. ^[1] However, it provides more outputs than other academic search systems, which reflects the overall existing literature when checking at least the first 200 results provided and checking the reference list of the selected papers. ^[2] This is the strategy that was used here, by checking at least the first 880 Google Scholars outputs for each search, thereby ensuring maximal coverage of the existing literature. To minimize the effects of potential confounding factors resulting in spurious relationships, only experimental studies were included while correlative studies were excluded. Moreover, only oxidative markers measured in given biological material of given animal species and with sufficient sample sizes (≥ 5) ^[3] were retained. Finally, only studies measuring oxidative markers in nervous tissues (thereby potentially reflecting affective states) or in peripheral tissues accessible without sacrificing animals (e.g., plasma) were retained (as these tissues may, in most cases, be the only tissues available to assess animal welfare in the wild). An exception to this rule was made regarding oxidative markers measured in body homogenates in small animals (e.g., tadpoles, insects) exposed to predators (as body homogenates are typically used to measure markers of oxidative status in these taxa). Because of all these limitations, analyses had to be restricted to superoxide dismutase (SOD) activity, catalase (CAT) activity, glutathione (GSH) concentration and malondialdehyde (MDA) concentration, measured

-        (1) in the hippocampus (HP) and prefrontal cortex (PFC) in laboratory rodents (rats and mice) exposed to social isolation,

-        (2) in the hippocampus (HP), brain and blood plasma/serum of laboratory rodents exposed to noise,

-        (3) in body homogenates of tadpoles, crustaceans and insects exposed to predators.

The mean values and standard deviations of the control group and the treatment group(s) were extracted from each study (from the content of the article, from the related supplementary material, from the figures using Graphreader V2 (version 0.9, Jan 2023, Kristian P. Larsen www.graphreader.com) or by contacting the authors). Standardized Mean Differences (SMDs, Hedges’ g) were then calculated for the difference between experimental and control groups included in each study, using the escalc function in the metafor package ^[4] in R version 4.3.1 ^[5] using R Studio (version 2023.06.0). Results were not sensitive to the use of SMD or SMDH (correcting SMD for heteroskedacity). SMD effect sizes were then used to conduct meta-analyses with the rma.mv function of metafor. Because in most analyses, some studies provided more than one data point, study ID was added as a random factor in those models. In order to be able to more easily compare with selection models (described below), all other models were fitted with effect size ID (one unique ID per effect size) as a random effect. For models where study ID was also used as a random effect, effect size ID captures the additional variation between effect sizes within studies, while the use of effect size ID in models with only one effect size per study is identical to study ID, so only effect size ID was kept in those models. All models were fitted with restricted maximum likelihood (REML) estimation. The t-value was used as a test statistic to construct confidence intervals and p-values. ^[6] This first meta-analytical approach provided a general effect size for each of the four oxidative markers (SOD, CAT, GSH, MDA) in response to the three considered conditions (social isolation, noise exposure, predation exposure) (when the number of data points ≥ 5), resulting in a total of 11 models. For the instances where sample size < 5 (e.g., GSH ~ predation model), models were not run. In a second step, meta-regressions were conducted by adding moderators in the model to assess differences in effect size between tissues (hippocampus and prefrontal cortex in the social isolation models; plasma, whole brain and hippocampus for noise exposure models) or taxa (crustaceans, amphibians and insects for predation exposure models), resulting in a total of nine models. Finally, in a third step, further meta-analyses were conducted for each oxidative marker within each moderator to calculate effect sizes within each biological tissue or taxon (when the number of data points ≥ 5). Phylogenetic information was not included in any of the models, as they included only laboratory rodents (mostly rats and some mice; social isolation, noise exposure) or because “taxon” was a variable of interest in the model (predation exposure).

Because the visual examination of funnel plots (and related methods, such as the Egger’s test), does not accurately capture publication bias in meta-analyses, publication bias was assessed here using selection models (using the selmodel function from the package metafor after using the rma function in R) to adjust for studies with given P-values being more likely to be published. ^[7] Because most studies included in the meta-analyses reported results for more than one marker of oxidative status regardless of the significance of the outcomes and of the variation direction (increase or decrease), two-tailed selection models were preferred over one-tailed selection models (note, however, that both model types provided qualitatively similar estimates). These models were run for each oxidative marker on the whole dataset of each of the three considered conditions (total of 11 models with a number of data points between 12 and 58) as described in ^[8]. The cutoff P-value, initially set at 0.05, was in some instances adjusted to lower values to ensure a sufficient sample size (N ≥ 5) within each P-value interval (see Table S2 for the chosen cut-off points). In five cases, evidence of publication bias was detected. In those cases, effect sizes estimated by the function selmodel are provided instead of effect sizes estimated by the rma.mv function. Because it was not possible to adjust for publication bias in the corresponding metaregression analyses (impossible to use the selmodel function after using the rma.mv function necessary for metaregressions), the coefficients and confidence intervals provided should be interpreted with caution.

References

1. Gusenbauer, M., & Haddaway, N. R. (2020). Which academic search systems are suitable for systematic reviews or meta‐analyses? Evaluating retrieval qualities of Google Scholar, PubMed, and 26 other resources. Research Synthesis Methods, 11(2), 181–217. https://doi.org/10.1002/jrsm.1378

2. Ungar, O. J., Muhanna, N., & Chaushu, H. (2023). Google Scholar as a sole literature source for meta-analyses in otolaryngology head and neck surgery. Indian Journal of Otolaryngology and Head & Neck Surgery, 75(2), 864–870. https://doi.org/10.1007/s12070-023-03532-8

3. Curtis, M. J., Bond, R. A., Spina, D., Ahluwalia, A., Alexander, S. P. A., Giembycz, M. A., Gilchrist, A., Hoyer, D., Insel, P. A., Izzo, A. A., Lawrence, A. J., MacEwan, D. J., Moon, L. D. F., Wonnacott, S., Weston, A. H., & McGrath, J. C. (2015). Experimental design and analysis and their reporting: New guidance for publication in BJP: Editorial. British Journal of Pharmacology, 172(14), 3461–3471. https://doi.org/10.1111/bph.12856

4. Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package. Journal of Statistical Software, 36(3). https://doi.org/10.18637/jss.v036.i03

5. R Core Team. (2023). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria [Computer software]. https://www.R-project.org/

6. Assink, M., & Wibbelink, C. J. M. (2016). Fitting three-level meta-analytic models in R: A step-by-step tutorial. The Quantitative Methods for Psychology, 12(3), 154–174. https://doi.org/10.20982/tqmp.12.3.p154

7. Maier, M., VanderWeele, T. J., & Mathur, M. B. (2022). Using selection models to assess sensitivity to publication bias: A tutorial and call for more routine use. Campbell Systematic Reviews, 18(3). https://doi.org/10.1002/cl2.1256

8. Harrer, M., Cuijpers, P., Furukawa, T. A., & Ebert, D. D. (2021). Publication Bias. In Doing Meta-Analysis with R: A Hands-On Guide (Chapman&Hall). CRC Press. https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/