Effects of hyperoxia on vascular tone in animal models: systematic review and meta-analysis

To assess the risk of bias in studies, we planned to use the SYRCLE risk of bias tool. This tool has several items that relate to studies in which separate intervention and control groups are used. In such experiments, animal housing and animal selection may have a significant impact on outcomes. However, the studies included in this review (both in vivo and ex vivo) were all short-term pretest-posttest designs, performed in the same animal/vessels (e.g. no control groups). As a consequence, the SYRCLE risk of bias tool was not suitable for this review and was substituted by a customized quality checklist (see “ Quality assessment” section). Due to the large variation in the size of the arteries investigated in the in vivo experiments, we added artery size post hoc as a possible source of heterogeneity (see “ Data analysis”).

We searched the databases Pubmed and Embase for articles published from inception up to 22 November 2017. The search strategy included various terms for vasoconstriction, vasodilation and hyperoxia. These categories were then combined with the AND operator and filtered for animal studies with the SYRCLE animal filter [ 22, 23]. We did not apply any language or date restrictions. Additional file 1 shows the search strategies used. We also checked the reference lists of included studies and relevant reviews for additional articles on hyperoxia that were not identified in the initial search.

Studies identified by the initial search were subjected to 3-phase screening, performed by two investigators (BS and AS). In the first phase, titles were screened to exclude studies not related to blood vessels or hyperoxia. In the second phase, titles and abstracts of articles were screened for eligibility based on the following predefined inclusion criteria: (1) controlled studies with repeated measures involving (2) adult healthy animals or isolated arteries or arterioles from adult healthy animals, (3) short-term normobaric, normocapnic hyperoxia in comparison with a normoxic control (see “ Oxygenation definitions” for an explanation), (4) vessel diameter or tension measurements reported. In case of doubt, the full text of the publication was evaluated. Disagreements were resolved by discussion (BS and AS). In phase 3, full-text articles were assessed for final inclusion in more detail, based on the following exclusion criteria: (1) case reports, reviews and studies without an intervention, (2) disease models or models focusing on the vascular beds of the lung, brain or retina, the fetal vasculature or the ductus arteriosus, (3) studies without a focus on vasoreactivity, (4) vessels preconditioned by hypoxia, an intervention in the experimental design other than oxygen (e.g. addition of CO ₂, endothelium removal) or long term hyperoxia, (5) none of our predefined outcome measures were reported. In particular, the brain was excluded from analysis because it is known to possess extensive mechanisms for the regulation of its perfusion, therefore making it unsuitable for inclusion in meta-analysis of other more comparable vasculatures. Similarly, we excluded studies in non-healthy animals, to reduce anticipated heterogeneity because of altered regulation of perfusion during states of disease.

For the in vivo studies, when the intervention was applied systemically, the measurements while the animal was breathing a gas with 20–21% O ₂ were considered as normoxic. Consequently, measurements during the inhalation of a gas with an oxygen content > 21% were recorded as hyperoxic. For in vivo studies where hyperoxia was established locally through superfusion, normoxia was defined as the state where tissue was superfused with a physiologic salt solution (PSS) equilibrated with 0–5% O ₂. With these concentrations, supply of oxygen to tissue by the PSS is limited as much as possible and occurs predominantly through the microcirculation. In these studies, hyperoxia was defined as superfusion with a PSS with an oxygen concentration of > 5%.

In ex vivo studies, partial pressure of oxygen (PO ₂) is directly comparable to partial pressure of oxygen in arterial blood (P _aO ₂). Hence, the lower limit of normoxia was defined as PSS equilibrated with at least 7% O ₂, which yields a PO ₂ of ~ 55 mmHg. Because ex vivo oxygen exposures are often much higher than attainable in vivo (e.g. 100% O ₂ ex vivo gives a PO ₂ of ~ 760 mmHg, while inhalation of pure oxygen gives a P _aO ₂ of approximately 350 mmHg), we used measurements at the lowest non-hypoxic oxygen exposure as the “normoxic” state and any exposures above as “hyperoxic”.

From each article we extracted data on the species, strain, age, weight and sex of the animals used, type of anesthetic used, method and level of oxygen exposure, type of preparation/measurement setup, buffers used, descriptions of variability in preparations and oxygen dose-response relations. Reports on a change in vascular tone upon exposure to hyperoxia were extracted as “experiments” (e). Sometimes an experiment was not supported by data (e.g. data not shown). Experiments with data were extracted as “data sets” (k). These data sets include the number of animals/vessels used, mean and standard deviation or standard error of the vessel diameter (or equivalent, e.g. tension), or change from baseline data, during normoxic and hyperoxic exposures. Data not reported numerically were extracted from graphs with a digital ruler if possible. Because of incomplete outcome data in five studies, we contacted the corresponding author by email to request additional information. We received a response from one author, who was unable to supply the missing data. If a study reported multiple responses from similar vessels in separate groups (e.g. different unique control groups), then only the data from the group with the largest number of paired vessels were used in this meta-analysis. If there was uncertainty about the number of observations (e.g. the number of observations was reported as a range), the lowest value was used to calculate a conservative estimate of the standard error. All data were extracted by one author (BS) and then randomly checked by a second (AS).

The quality of studies was assessed using a tailor-made checklist containing ten items regarding blinding, preparation quality, verification of intervention efficacy, selective outcome reporting, description of animals and arteries, power analysis, ethical approval, conflicts of interest and peer review. For a list of the criteria used, see Additional file 2.

All calculations were made using RStudio v1.1.383 and the “metafor” and “dosresmeta” package (Integrated Development for R. RStudio, Inc., Boston, USA). Graphs were made using GraphPad Prism 7.0 (GraphPad Software, Inc., La Jolla, USA) or RStudio. Effect sizes are presented as the standardized mean difference (SMD) for series of experiments involving paired comparisons as proposed by Gibbons [ 24]. All statistics are reported with their 95% confidence interval. An estimation of the variance of the SMD was done for a one group design with repeated measures [ 24]. Missing SD _difference values were calculated assuming correlation between measurements of 0.7 [ 25]. A sensitivity analysis using higher and lower correlation coefficients was conducted to test the robustness of our results. Effect estimates of the maximal hyperoxic exposure were pooled using a random effects model. Heterogeneity was assessed by the I ² statistic and is reported with its 95% confidence interval [ 26].

As a priori sources of heterogeneity we considered animal species, sex, vascular bed and method of hyperoxia exposure for in vivo studies. For ex vivo studies, we considered species, sex, tissue origin of the vessel, vessel type (systemic or resistance) and the presence of flow. We added baseline artery diameter post hoc as a possible source of heterogeneity for in vivo studies. The relation between effect size and artery diameter was explored through restricted cubic spline modeling, because it does not require any assumption on the type of relation between the two variables ( e.g linear, sigmoidal etc.) [ 27]. The Akaike information criteria (AIC) was used to determine the optimal number of knots and their position [ 28]. For each subgroup a minimum of five data sets, from three unique studies, had to be present. Sources of heterogeneity were investigated using meta-regression, by first performing an overall test for interaction, and if the P value was <0.05, pairwise comparisons between subgroups were made to detect further subgroup interactions. To correct for multiple testing, P values were adjusted with the Holms-Bonferroni method.

The likelihood of publication bias was assessed using the trim and fill method [ 29]. Because SE-based precision estimates cause distortion of SMD funnel plots we used 1/√n as the precision estimate in the trim and fill analysis [ 30].

A flow chart of the study selection process is shown in Fig.  1. After exclusion based on title and abstract, we identified 319 articles investigating the relation between oxygen and vascular tone: 261 studies were excluded based on predefined exclusion criteria. From the 60 included studies, 42 were performed in live animals and 18 studies used isolated arteries and arterioles. One study performed both eligible in vivo and ex vivo experiments [ 31]. Studies were performed between 1974 and 2017 and the majority (64%) was published between 1980 and 2000. Characteristics of the included studies are presented in Tables  1 (in vivo studies) and 2 (ex vivo studies).

The funnel plot for the in vivo studies shows an indication of publication bias, favoring studies that reported constriction of arterioles upon exposure to hyperoxia (Additional file 3). Balancing the results with the trim and fill method reduced the overall effect size ( P = 0.03) from − 1.42 (− 1.65 to − 1.19) to − 1.03 (− 1.30 to − 0.76). The funnel plot for the ex vivo studies did not reveal evidence of publication bias (Additional file 3).

For all analyses shown, we used a conservative correlation coefficient of 0.7 to impute missing standard deviations of paired data. Missing standard deviations had to be calculated for 21/67 in vivo data sets and 7/18 ex vivo data sets. Using a higher or lower correlation coefficient did not affect the results for the in vivo analysis (Additional file 4). For the ex vivo analysis, the differences were marginal, but when a low correlation coefficient of 0.5 was used, the confidence interval of the overall effect crossed zero.

There are several limitations to this meta-analysis. There was considerable heterogeneity between the in vivo studies, which could only in part be explained by differences in the region of interest and the size of the vessels investigated. For the ex vivo studies we did not find an explanation for the heterogeneity, suggesting the importance of a yet unidentified methodological, systemic or metabolic factor. Second, the funnel plots indicated publication bias for in vivo studies, favoring studies finding vasoconstriction in response to hyperoxia. However, correction with the trim and fill method did not significantly alter the conclusions. Third, due to the relatively small number of studies, some collinearity between the investigated subgroups may be present. Fourth, for some studies we had to make an assumption on the correlation coefficient to impute missing standard deviations. Our conservative estimate may have underestimated the effect, although our sensitivity analysis shows that using a higher correlation coefficient did not significantly alter the results. Fifth, all studies were performed with healthy animals; studies in disease models may give different results. Similarly, the use of live animals required the application of anesthetics in the majority of studies. Most of these drugs have at least some vasodilatory effects, which may have influenced the results. However, because of the pre-post design of the included studies, the effect of the anesthetic is anticipated to be the same under each oxygen condition. Sixth, we excluded studies on the lung, retinal and brain vasculature.

Finally, as reflected by our quality assessment of the included studies, the overall quality of the included studies was poor, which seriously hampers drawing reliable conclusions from the included animal studies. Baseline characteristics of animals were insufficiently described. Similarly, only 14 of the 60 studies properly described artery characteristics, such as vessel diameter and selection of vessels in the arterial tree. Reporting these data is important to increase reproducibility between studies. Similarly, measurement of PO ₂, pH, PCO ₂ and temperature was not (sufficiently) performed in the majority of studies. These measurements are crucial because of the dose response relationship between PO ₂ and hyperoxic vasoconstriction, and the effect of pH, PCO ₂ and temperature on vascular tone. Without measuring PO ₂ the true intervention applied to the vasculature is unknown, which will contribute to heterogeneity. The lack of blinding in most studies may have led to an overestimation of the true effect. Another flaw in most studies is the absence of power calculations. Especially for studies aimed at discovering mechanisms responsible for oxygen-related changes in vessel diameter, a minimum predefined number of vessels is necessary to be able to draw objective statistical conclusions. The majority of the included studies were conducted before the institution of more rigorous legislation on the use of animals in scientific procedures known today. The overall poor quality of studies is therefore likely related to the absence of an ethical board reviewing the studies prior to their execution. Similarly, the same papers were published before editors put certain requirements on the reporting of study data. We have noted that the more recent papers were of considerably higher quality, which shows that the change in procedures concerning animal research, which are now becoming very similar to human trials in terms of ethical board review and data transparency, is for the better.

The majority of the in vivo studies were performed on thin, externalizable skeletal muscle. Studies in other critical visceral organs (e.g. liver, kidney) are currently lacking. We propose that future, high-quality studies focus on visceral organs such as the kidney and liver. These organs are highly relevant for patient care and the results will contribute to the rationale behind selective induction or avoidance of hyperoxia in certain types of patients.