Search and Study Selection Processes
Fifteen publications [17-31] were included in quality overview, and only 14 [17-25,27-31] were included in the quantitative synthesis (meta-analysis); from one study  it was not possible to contact the authors or obtaining the FS data from the graphs (Figure 1).
Description of the included studies
The summarization of the characteristics of the 15 included studies are shown in Table 1. The publication year ranged from 2008 to 2018. In total, the 15 included studies used 963 rodents; however, one study  with 24 male Sprague-Dawley rats was not included in the meta-analysis as mean FS values were not presented by the authors. Data from Matsuura et al (2010) study  did not take part in the final meta-analysis. Thus, the following description of the results refers to the other 14 included studies in the quantitative synthesis. Considering only the meta-analysis we have performed, 851 rats (90.6%) and 88 mice (9.4%) were used, a population of 939 rodents. Regarding the studies with rats, 11 studies [17,19-27, 29] used Sprague-Dawley (n=785, 92.2%), and one  used Fischer344 (n=66, 7.8%) lineages. All studies with mice [18,30,31] used C57BL/6 lineage (n=72, 81.8%), and one study  used Athymic Swiss Nude (n=16, 18.2%).
With regard to the experimental groups in the meta-analysis, 477 animals (50.8%) were allocated into the sedentary group and 462 (49.2%) to the trained group. Of the studies involving rats, six [17,19,23,24,27,28] used females (n=395, 46.4%) and six [20-22,25,26,29] used male animals (n=384, 53.6%). Of the studies involving mice, only one  used female (n=22, 25%) and two [30,31] used male (n=50, 56.8%) animals. However, we could not retrieve this information from all included studies, as Wang et al. (2018)  was lacking this information. Thus, 16 mice (18.2%) were not classified by gender . Age was mentioned in 10 studies [17,18, 20, 24-31], which ranged from 23 days to 6 months old. Finally, baseline body weight was described in 4 studies [19, 21-23], ranging from 175 g to 350 g.
The cardiomyopathy animal model followed the intraperitoneal administration of DOX protocol in all included studies. The cumulative dose of DOX ranged from 3 mg/Kg to 32 mg/Kg (mean 13.1 mg/Kg). However, two studies [17, 30] used less than 10 mg/Kg, which was the modal value of the cumulative doses. The protocols for anthracyclines-induced cardiomyopathy were highly heterogeneous regarding the fractioning of the cumulative dose, frequency of injections and timing of administration in relation to the exercise protocol. DOX administration was reported to occur prior [17,18, 26], concomitantly [20, 23, 30, 31], and after the end of training protocol [19, 21, 22, 24, 25, 27, 29]. The precise moment when the last cardiac function assessment in relation to the last DOX injection has occurred (in hours or days) was not accurately described among the included studies. Yet, all studies had mentioned if it has occurred after the end of exercise training protocol, after DOX protocol and/or before euthanasia.
Regarding exercise modalities, most of the studies (93.3%) included aerobic exercise training as follows: supervised treadmill running (33.3%) [18, 26, 27, 30, 31], voluntary wheel running (26.7%) [19, 20, 23], supervised swimming (6.7%) or supervised treadmill running and voluntary wheel running (26.7%) [21, 22, 24, 25]. Only one study reported to use strength training . The velocity of supervised treadmill running with a progressive transition was mentioned in 8 studies [18, 21, 22, 24, 25, 27, 30, 31], which ranged from 10 to 30 m/min. The inclination of supervised treadmill was reported in 8 studies [18, 21, 22, 24-27, 31], and variat ed from 0 to 18º. The duration of each exercise session, mostly performed at 5 days/week, was described in 9 studies [18, 21, 22, 24-27, 30, 31], and lasted between 20 to 60 min/session. The duration of the complete exercise protocol ranged from five days to four months, as follows: 5 days , 10 days , 2 weeks [30,31], 4 weeks , 6 weeks , 8 weeks , 10 weeks [20-24], 11 weeks , 12 weeks , 16 weeks . Three studies of supervised treadmill reported to include a period of 5 days for habituation to training [18, 21, 27], and one study mentioned the same habituation protocol to supervised treadmill and voluntary wheel running trained animals . Although none of the studies involving voluntary wheel running reported a period or protocol of acclimatization, all of them referred free voluntary access 24h/day. Non-exercised (sedentary) animals were used as the controls in all studies.
Assessment of Study Quality
In order to minimize poor reporting problems in preclinical studies and to classify most of the items of the risk of bias tools as “unclear”, we decided to use three different tools, SYRCLE, RevMan and CAMARADES. SYRCLE risk of bias regarding randomization and allocation concealment were inadequate to judge for all studies, which implied in unclear risk. Baseline characteristics were considered equal between control and intervention groups, and consequently assessed as low risk of bias. Two studies reported to have animal cages not randomly placed [24,29]. Most of the studies did not reported blinding strategies for caregivers and/or investigators regarding the intervention each animal received, and only one study  provided information on blinding of the examiner of the main outcome. Most of the studies reported to not have conflict of interest regarding protocols and funders. However, for those studies which did not have an explicit conflict of interest statement [21, 23, 26] the risk of bias was considered unclear (Supplementary Table 2).
According to RevMan quality assessment (Figure 2), the included studies were classified as having unclear or high risk of bias according to allocation concealment or blinding of participants and personnel. Only one study  was classified with low risk of bias as the authors clearly described blinding strategy for the assessment of outcome, and all studies presented low risk of bias regarding selective reporting. Finally, random sequence generation and presence of other bias remained unclear for all studies as they all lack this information.
CAMARADES mean score was 6.33 ± 1.04. Four studies [21,23,25,26] were underscored, mostly due to the absence of description of blinded assessment of outcomes, monitoring of body weight parameters and sample size calculation. None of the studies received a maximum score, but two [18, 31] were rated with 8 points, as only sample size calculation was not provided (Supplementary Table 3).
Efficacy of exercise training on the cardiac function of preclinical models of DOX-induced cardiomyopathy
Fourteen studies [17-25, 27-31] investigating the effect of exercise training on FS in animals with DOX-induced cardiomyopathy were included in the meta-analysis. Overall, pooled analysis of individual effect sizes of all exercise training protocols showed that this intervention favors DOX-treated animals with regard to reduction of FS in comparison to sedentary state [n=947, MD=7.46%, 95% CI 5.58 to 9.34, p<0.001] (Figure 3). The heterogeneity between the studies was considered moderate (I2 = 59%) , reflecting the possible anticipated differences between training strategies, timing of outcome assessment and features of animal model.
Subsequently, in order to identify the possible factors which might have influenced on exercise-mediated preservation/improvement of cardiac function following DOX treatment, we conducted an exploratory sensitivity analysis (Figure 4). We retrieved the mean difference in FS (%) between sedentary and exercised DOX-treated animals for each variable of interest. The timing when exercise was performed in relation to DOX exposure (before vs. concomitantly and/or after; p=0.0187) was identified as the only possible factor influencing on exercise-mediated protective efficacy on cardiac function in DOX-treated animals.
Subgroup analyses revealed that exercise training has greater efficacy on attenuating DOX-induced FS impairment if executed before DOX exposure (MD=8.20%, 95% CI 6.27 to 10.13, p=0.010) (Supplementary Figure 1A). However, modality of the exercise (supervised treadmill or voluntary wheel running) with MD=7.03%, 95% CI 5.45 to 8.61, p=0.060) (Supplementary Figure 1B) and total duration of the exercise program (MD=7.13%, 95% CI 5.44 to 8.81, p=0.330) had no influence on the main outcome (Supplementary Figure 1C).
Effects of animal model
The features of the animal model are decisive to the results interpretation and extrapolation of these data to clinical landscape. We were able to subgroup the studies according to four animal model-related variables: animal species, gender, cumulative DOX dose and timing of cardiac function assessment. Of those, exercise training exerted cardioprotective effects irrespective to all animal species (MD=7.13%, 95% CI 5.44 to 8.81, p=0.080) (Supplementary Figure 2A), gender (MD=7.35%, 95% CI 5.63 to 9.06, p=0.770) (Supplementary Figure 2B), and cumulative DOX dose (MD=7.13%, 95% CI 5.44 to 8.81, p=0.750) (Supplementary Figure 2C). Conversely, the timing of final cardiac function was assessed (MD=7.13%, 95% CI 5.44 to 8.81, p=0.030) was associated with a greater effect size regarding the preservation of cardiac function when considering the studies which had used rats or that had the echocardiography analysis performed up to 10 days after the completion of all protocols (Supplementary Figure 2D).
Effects of study quality
We also investigated the influence of the study quality (according to the items on the CAMARADES checklist) on the main outcome (Supplementary Figure 3), and we identified that the cardioprotective effects of exercise occurred independently on the study quality score (MD=7.13%, 95% CI 5.44 to 8.81, p=0.780).
Mechanisms of exercise-mediated cardioprotection in rodents with DOX-induced cardiomyopathy
The summarization of the main findings and the possible mechanisms implicated on exercise-mediated cardioprotection reported by each included study are presented in Supplementary Table 4. All studies described the exercise-mediated systemic, functional or molecular mechanisms possibly implicated in attenuation, preservation or improvement of FS in DOX-treated animals. Exercise-mediated reduction of DOX-accumulation was the most reported mechanism at cellular level [19, 24, 27, 28, 31], followed by preservation of myosin heavy chain [21-23, 29]. Four studies [17, 18, 22, 25] evaluated the association between exercise training and calcium homeostasis [17, 22, 25], with a particular focus on the preservation of myocardial contraction and global cardiac function in exercised DOX-treated animals. Two studies [17, 20] mentioned that the exercise was not able to protect against DOX-induced impairments on body composition and weight gain. One study  stated that exercise did not reverse the cardiotoxic effects of DOX, although it was associated with upregulation of myocardial protein kinase B activity, which is involved with cell survival.
It is noteworthy that one study  reported exacerbation of DOX-induced myocardial damage following exercise training. In fact, the authors intended to explore the role of a maximum physical effort as a tool to detect early DOX-induced myocardial injuries rather than to evaluate exercise training as a rehabilitative/therapeutic approach for cardiotoxicity.
Analysis of publication bias was found to be influenced by the presence of small study effects. Symmetric funnel plots indicated the absence of publication bias for FS (Egger’s regression test, 2-tailed, p=0.861) (Supplementary Figure 4).
Based on our results, we performed a sample size calculation for future preclinical studies of exercise training as intervention for DOX-induced cardiotoxicity. To obtain a power of at least 90% in a two-sided two-sample t-test with an alpha of 0.05, 11 animals needed to be included in each group to detect a significant mean difference in FS of 7.69% in rats and 4.83% in mice.
Supplementary Table 5 summarizes all registered clinical trials (http://clinicaltrials.gov/) which aim to evaluate the role of exercise training in cardiac function in oncological patients undergoing anthracyclines-based regimens. For this search, we used the terms “cardiotoxicity” and “exercise” (including the term “physical activity”), which returned 23 registered studies between 2008 and 2020. Of these, only 15 studies has been designed to evaluate exercise in an interventional fashion, i.e., investigating the preventive/prophylactic effects of this practice against the possible harmful effects of antineoplastic agents on cardiovascular function. To date, only 3 of these studies have been completed, and however, none of these has published results. Furthermore, these studies are significantly distinct regarding the employed exercise protocols, which means that the chosen modalities, duration, intensity and adherence to program are highly heterogeneous. It highlights the existence of an unmet need for a high quality body of evidence to support the decision-making about the ideal training protocol for cancer patients, including the definition to what extent exercise prescription may protect the cardiac function of patients with a history of sedentary behavior.