Exercise as a complementary therapy for depression: a systematic review and meta-analysis of randomized controlled trials and bioinformatics exploration gene network model .

Background The effect of pharmacological treatment (PT), exercise treatment (ET), and both in depressive symptoms remains a matter of debate. The present study aimed to clarify (1) the effect of ET as a monotherapy or complementary treatment of pharmacological therapy; (2) changes in the dose-response for different exercise prescription characteristics; and 3) hypothesizing about the gene network model of exercise effects on depression. Methods We sought Randomized Controlled Trials (RCT) addressing the effects of exercise on depressive patients, published in peer-reviewed journals between 2003-2019 in Scopus, Cochrane, Pubmed/Medline, ISI Web of Knowledge and APA PsycNET databases. Standardized mean difference (SMD) was calculated considering the mean difference on depression scales (pre and post-intervention) and pooled standard deviation for each intention-to-treat in each study. For the gene network model of exercise on depression an in silico analyses were used. Results We found 1,165 articles and selected 15 studies to this meta-analysis. RCTs with different ET and PT prescriptions were examined using the delta (pre and post-intervention) of a validated depression scale compared to the control group in different treatment conditions. Standardized condence intervals (CI: 95%), including the assumption of heterogeneity of the studies and their participants. Analyses of forest and funnel plots were performed using the Review Manager (RevMan) Version 5.3 software (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014). The GeneCards database (https://www.genecards.org) was used to explore proteins-related genes associated with exercise and depression.

Major depressive disorder (MDD) is a worldwide mental health concern leading to severe morbidity and increased mortality risk [1]. Patients with depression have a twentyfold greater risk of suicide compared to the general population [2]. Several countries have conducted epidemiological surveys, and the high prevalence and persistence of MDD con rm that this is a worldwide disorder of utmost importance [3].
The years lived with disability (YLDs) per 1000 people with depression disease are more than ve times higher than the existing global mean burden of disease estimates [4].
Regardless of the research growth during the last decade, no major advances have occurred in the treatment of depression [5]. Although pharmacological therapy (PT) is the current gold standard for the treatment of depression, this treatment may need a combination of therapies such as antidepressants, antipsychotics, electroconvulsive therapy and psychotherapy [6]. Thus, more investment is required to search for effective adjuvant therapies to enhance the treatment of depressive disorders. In recent years, systematic reviews and meta-analyses evidenced that physical activity and exercise have been improving depressive symptoms and bene t physical and mental health. Therefore, clinical guidelines have included exercise as a possible rst-line intervention to the treatment of depressive symptoms [7]. Current reviews have been showing different methodologies and results when investigating the evidence that exercise treatment (ET) has a bene cial effect on reducing depressive symptoms [8][9][10]. Ekkekakis [11] suggests a carefully scrutiny on several grounds in these design-studies, hence aiming to eliminate these biases. Due to the potential inappropriate selection criteria applied, results could lead to misleading interpretations. Other reviews were published to set a consensual exercise prescription and recommend essential variables in the treatment of patients with depression [12][13][14].
Although these reviews have been helpful, there is an incomplete understanding of the effect of ET with and without PT in the depression treatment. While the impact of exercise interventions among subjects with depression has been explored, the optimal dose that might prove useful remains non-consensual.
Also, studies with bioinformatics analyses [15] in an attempt to elucidate gene network models and biomarkers involved in the depression treatment with exercise are still lacking. Hypothetically, this analysis could be a reliable and straightforward form to study how exercise could decrease depressive symptoms. Different from the traditional published meta-analyses, the current study was structured with a robust methodological criterion to select and analyse the effect of ET with or without PT and bioinformatics data in depressive symptoms and exercise. Therefore, the current study intended to elucidate (1) the effect of ET with and without PT for depression; (2) changes in the dose-response of different exercise prescriptions, intervention and control group characteristics, and 3) a hypothesis about the gene network model in silico analyses to clarify how exercise could reduce depressive symptoms.

Methods
The registration of this systematic review and meta-analysis has been approved in the International Prospective Register of Systematic Reviews (PROSPERO) with the protocol number CRD42019122638. We followed the PICOS strategy [16] and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement [17]. The recommendations of the Cochrane Collaboration [18] were used as a complementary guide.

Types of studies
RCTs conducted between 2003-2019 were selected, whether exercise was bene cial or not in the treatment of depression. To reduce the risk of bias, conference proceedings and unpublished studies were not used [19], and no restrictions on the language of the studies were applied.

Participants
Males and females aged 18 and over (with no upper age limit) with clinical depression as de ned by the Diagnostic or Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV), International Statistical Classi cation of Diseases and Related Health Problems 10th Revision (ICD-10) criteria or other validated depression scale. Depression encompassing a co-morbid illness such as diabetes (insulin-dependent), cancer, multiple sclerosis, cardiovascular disease and mixed psychiatric diagnosis was not included.

Interventions
Intervention group had as ET aerobic or resistance training programs or a combination of both, in addition to PT, or not. In control group, we analysed interventions, such as pharmacotherapy, psychotherapy, another exercise program or no intervention (placebo or wait-list control). Considering exercise parameters, we included studies in accordance with the following data criteria: minutes per session, duration of intervention (week), frequency (days per week), adherence (percentage of nishers ) and the amount of energy expended and measured in kilocalories per week (Kcal/week), or the percentage of maximum heart rate (%HR max ), or the percentage of heart rate reserve (%HRR), or the percentage of maximal oxygen uptake (VO 2max ), or the percentage of maximal oxygen uptake reserve (VO 2reseve ), or metabolic equivalent (METs).

Outcomes
Pre and post-intervention scores of validated depression-rating scales were analysed. Studies that measured outcomes immediately before and after a single exercise session were not included.

Information sources
We conducted searches on PubMed/MedLine, Scopus, ISI Web of Knowledge, Cochrane Trials and APA PsycNET databases. No lters were used to search any of the databases to conduct the broadest search and reduce the risk of bias [19].

Search
Medical Subject Headings (MeSH) and search-indexed descriptors were used to re ne data search [20].
Three thematic word groups with MeSH terms were used to conduct the searches. Within each group, the terms were combined using the Boolean operator OR and interaction between sets using the operator AND to form a phrase. Searches were conducted in April 2019 using the following terms: ("depressive disorder" or "unipolar depression" or "major depressive disorder") AND ("exercise" or "exercise programs") AND ("clinical trial" or "randomized controlled trial").

Study selection and data collection process
Studies were screened and data independently extracted by two researchers. Another researcher was requested to con rm the eligibility of the identi ed studies. Identi ed studies were tabulated on a worksheet (Microsoft Corporation, Redmond, USA) to con rm if they met the eligibility criteria. All their titles and abstracts were screened and the full-text articles were assessed for potential inclusion by the main investigator. Studies were included in the qualitative synthesis after the exclusion of a comprehensive text review. We identi ed all studies that presented a high risk of bias for the nishselection process and nally, we included all quality studies into a quantitative synthesis to perform the meta-analysis.

Appraisal of methodological quality
The quality of the selected studies was appraised using the Delphi-list [21]. We analysed randomized controlled trials (RCT) through: randomization, allocation concealment, baseline comparability, eligibility criteria, blinding, descriptive measures for the primary outcome and intention-to-treat analysis [22]. In this study, except double-blinding, which is not applicable within the framework of trials involving physical exercises, all these features were taken into account by the qualifying examination. Two researchers independently calculated an overall quality score of the Delphi items that scored positive and discussed them to achieve consensus. Studies selected did exhibit weaknesses concerning some criteria and these de ciencies were taken into consideration and explained in the results and discussion sections.

Data selection
We selected the following characteristics in all studies: 1) total number and age of each group; 2) depression-rating scales used to the diagnosis; 3) type of intervention and other exercise parameters mentioned in the section 2.1.3; 4) pre (M1) and post (M2) intervention depression-rating scales scores (n, means and standard deviation) to calculate the effect size (ES) of intervention and control groups. The mean difference (MD = M2-M1) and the pooled standard deviation (SD pooled = √ (SD M2 2 (n M2 -1) + SD M1 2 (n M1 -1)) / (n M2 + n M1 -2)) of each group in all studies were calculated to ES analyses [23]. Regarding pre-selected articles that did not present the necessary data in the text, values were requested to the authors by e-mail.

Risk of bias
The risk of bias was assessed using qualitative analysis for each included study, and each risk of bias item presented on the Delphi-list. This scale provides a quality assessment of RCT studies, and the high quality is de ned as achieving over 50% of the maximum attainable score, meaning ve or more criteria met on the Delphi-list [21]. To analyse the risk of publication bias, we used funnel plot visual inspection. The risk among studies was assessed using the results of heterogeneity within the forest plot.
Heterogeneity was measured using the T 2 , X 2 , and I 2 tests. In the T 2 test, T 2 > 1 suggests the presence of substantial statistical heterogeneity. If the X 2 value is statistically signi cant (p < 0.05), there is also evidence of heterogeneity. In the I 2 test analysis, the percentage of the variance attributed to the heterogeneity of the study ranges from low (25% < I 2 < 50%) to moderate (50% < I 2 < 75%) to high (I 2 > 75%) [19].

Summary measures
Analyses were performed considering two different groups, experimental and control. The main analysis was related to the additional effect of ET with or without PT for depression. Also, we included the effect of subgroup analyses among different exercise prescriptions, intervention, and control group characteristics encompassed in the selected studies. Control group data were replicated and compared with the different intervention groups in their studies to comparison analysis in this meta-analysis For multiple comparison groups in the same study, control group had the sample divided based on the number of groups that existed for comparison [24]. This was to maintain control group correct sample size and allocate the right weight to groups with more subjects.
Standardized mean difference (SMD) was calculated considering the mean difference on depression scales (pre and post-intervention) and pooled standard deviation for each intention-to-treat in each study. This outcome was reported on different validated scales. Consequently, the SMD in this review was calculated based on the random-effect model with 95% con dence intervals (CI: 95%), including the assumption of heterogeneity of the studies and their participants. Analyses of forest and funnel plots were performed using the Review Manager (RevMan) Version 5.3 software (Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

The Gene Network Model in silico analyses
The GeneCards database (https://www.genecards.org) was used to explore proteins-related genes associated with exercise and depression.
Genes aforementioned were inserted into the String database (https://string-db.org), which permits exploring genes and their encoded protein interactions as a network. String uses data mining to nd published articles which have investigated direct or indirectly targeted protein interactions in different species. According to the interactions found, we can suppose that exercise could result similar interactions, independently of the results from meta-analysis. Our interaction exploration was performed considering Homo Sapiens as the studied specie.

Study selection and characteristics
A total of 1,165 articles was found (PubMed = 235; Cochrane = 334; Scopus = 438; ISI Web of Knowledge = 133; APA PsycNET = 25). After applying the exclusion criteria for eligibility, a total of 17 studies were subjected to a qualitative analysis. After this, 15 studies were included, and two studies were excluded from the meta-analysis. Figure 1 shows the screening ow used in the study selection.
This review shows 17 studies selected that met the quality assessment score based on the Delphi-list (Table 1). Regarding the RCT studies evaluated, scores ranged between 3 and 7 points (mean = 5.5 points). Two studies [25,26] scored 3, indicating unsatisfactory quality, and therefore, were excluded.

Summary of studies
The 15 studies included in the meta-analysis provided a total sample size of 1,532 individuals, with a mean age 44 years, which ranged from 21 to 75 years. The mean percentual of females' sample was 66.5% and varied between 30 and 100%. Investigations occurred in different countries, expanding different personal characteristics in the present study. Co-morbidities were not found among the participants. In all selected articles, interventions with aerobic, resistance training, or a combination of both, and validated depression-scales were used, and other characteristics of the studies are described in Table 2.  Intensity: HRmax = Maximum heart rate. HRR = Heart rate reserve. Kcal/kg/week = Energy expended and measured in kilocalories per week. VO2max = Maximal oxygen consumption.
Adherence: percentage of finishers.

Standardized mean difference (SMD)
One study [30] presented four different intervention groups relative to the intensity and volume of exercise prescribed, which were included separately in the comparisons; two studies [32,37], which showed two different intervention groups associated with the presence of supervision or not and to the type of exercise, have also been included separately to ES analyses. SMD analyses in studies with more than one intervention group had the number of participants in the control group divided by the number of groups that existed for comparison in these studies. However, twelve studies [27-29, 31, 33-36, 38-41] showed just one intervention group to compare . To interpret SMD we used the scale suggested by Cohen [23]: > 0.30 small; > 0.50 moderate; > 80 large.

Main analyses
The forest plot of different treatments for depression in Figure 2 shows  Figure 2 shows a symmetrical plot in the absence of publication bias.

Subgroup analyses
Forest and funnel plot of different characteristic analyses are displayed in the supplementary data, and Table 3 Table 3 shows that in favour of ET, with or without PT, the subgroup analysis on the treatment for depression has displayed : a small ES for two and three sessions (p < 0.00001), intervention time between one to nine weeks (p = 0.006), 50 to 70% of intervention adherence (p = 0.0002), different age-range groups intervention at 30 to 60 years old (p = 0.002) and 20 to 50 years old (p = 0.02), when compared to an active control group with stretching and light exercise (p = 0.03); a moderate ES on resistance training (p = 0.02), moderate exercise intensity (60 to 80%) controlled by MHR (p = 0.0008), different exercise duration per session at 45 to 60 min (p < 0.0001) and 20 to 30 min (p < 0.0001), intervention time for more than 20 weeks (p < 0.00001), when compared to a control group with usual treatment use (p = 0.0007); a large ES for seven sessions (p = 0.008), and intervention with people aged 60 years or more (p < 0.0001). The heterogeneity between studies in subgroup analysis was not exhibited, except on the moderate intensity exercise (60 to 80%), exercise duration time for 45 to 60 min, intervention time between one to nine weeks, and intervention adherence between 50 to 70%. HRmax = Maximum heart rate; HRR = Heart rate reserve; Kcal / kg / week = Energy expended and measured in kilocalories per week; VO 2 max = Maximal oxygen consumption; RCTs = Randomized clinical trials; SMD = Standard mean deviation; 95% CI = confidence interval; Heterogeneity: T 2 = Tau > 1; X 2 = Chi-value (p < 0.05); I 2 = percentage of the variance; NA = Not applied, heterogeneity analyses with one subgroup selected;

The gene network model in silico analyses
Speci c and non-speci c (unknown) signi cant interactions (PPI Enrichment p-value <0.01) [15] among all genes were found in the network (Figure 4). Interactions were explored when speci ed edges were clearly shown, and the combined score reached 0.4 or above, determined as medium con dence (see String Statistics Section on https://string-db.org). Two genes (TPH2 and SLC6A4) have not met the criteria above and were excluded from the analysis. Therefore, the following genes remained signi cant (PPI Enrichment p-value <0.03) in the model: BDNF, HTR2A, APOE, IL6, INS, and TNF ( Figure 5). The combined score of each interaction is shown in Table 4.

Discussion
This study aimed to review the effect of additional ET, with or without PT, on the depression treatment. Moreover, its objective was to investigate subgroups in the selected studies to compare different exercise prescription, intervention, and control group characteristics, which could in uence ET response. Finally, explore the hypothesis about the gene network model in silico analyses to explain how exercise could reduce depressive symptoms. After these analyses, results presented a moderate SMD on the additional effect of ET on PT for depression. Subgroup analyses showed that the moderate intensity, in 20 to 60minute sessions preferably every day, could be the ideal exercise prescription in depressive treatment.
Interventions prescribed 20 weeks or more with 50 to 70% of adherence may show a better result. When analysing based age-range studies, we found that ET can trigger a better result in older adults. Our gene network model also indicates that ET can be useful in decreasing depressive symptoms due to BDNF and in ammation interaction pathway, and regulation of HTR2A and APOE.

Meta-analysis evidence
This is the rst review that con rms that exercise provides a vital add-on effect on PT for depression, according to the Gourgouvelis, Yielder [42] trial. It is not new that exercise training programs can be considered a treatment for depression and that PT is the gold-standard. However, antidepressants also promote a more rapid initial therapeutic response than exercise [43]. Considering this, it is adequate that ET and PT working together show a better result in the depression treatment.
There are preferences and growing evidence for using aerobic exercise in moderate-intensity to treat depression, as recent systematic reviews have shown [8,12,14]. Although, it is already clear that resistance training can produce a similar positive neurophysiological alteration by reducing depressive symptoms [44]. Even though one study, our review showed that resistance training could be a good alternative for the treatment. Based on the overload principle of training (progression), the exercise under a minimum intensity or threshold will not induce su ciently the body to result in important neurophysiological adaptations for depression treatment [45]. In the initial program stage, a self-selected comfortable exercise intensity seems to be the better option to induce intrinsic exercise motivation and adherence [10]. We sustain that exercise intensity should be pleasurable, but it should also be gradually increased to guarantee a minimum level required to induce a training effect. Despite the diversity of tools to assess prescribed intensity (i.e., HR max , HRR, Kcal/kg/week, VO2 max , and the Borg scale), it is yet premature to provide recommendations concerning what is the better to control exercise intensity. In relation to the volume of prescribed exercise , other reviews [12][13][14] argue a reduced volume in the session duration and frequency when compared with our results. Utilizing the principle of progression again, we understand that the target in the volume prescription needs to follow our ndings, which show a better response for both 20 to 30 min and 45 to 60 min every day. In the depression treatment, it does not matter if the patients are training for 20 or 60 minutes if they do it [46].
Regarding ET duration , past reviews reported positive outcomes after at least 8 or 9 weeks [12,14].
However, based on the chronical neurobiological effects of exercise in depression, a regular exercise intervention is critical for satisfactory treatment response [47]. Corroborating that, our results show a better response for treatment of interventions equal to or more than 20 weeks. Intensity control contributes to the adherence in exercise intervention [10]. Notwithstanding , in the study by Callaghan, Khalil [48], the preferred intensity group only showed 66% of adherence. As well, when we observed data from studies where the preferred intensity was not prescribed but showed a signi cant reduction in depressive symptomatology, three studies [49][50][51] presented high rates of adherence (80-100%). To our understanding, adherence rate is an essential factor for ET, although it resulted in a small effect in the present review. Adherence could be achieved not only by the most convenient intensity, but also with other strategies of intervention (e.g., knowledge of subjects' daily and leisure activities, exercise supervised and preferences about the context) [46].
It is essential to report that the effectiveness of pharmacological treatment in depression is not substantially affected by age [52]. According to a recent meta-analysis [53], exercise improves the response and can be considered as a routine tool in the management of depression in older adults.
Noneless, Mura and Carta [54] have noticed the di culty to establish the real effectiveness of exercise on depressive symptoms in elderly . However, the same author suggested that physical activity combined with antidepressants might be a bene cial strategy in treating late-life depression. The large ES showed in the older adult group in the current review can be related to the bene ts that exercise promotes in other chronic diseases usually common in the elderly. Decreasing the symptoms of these comorbidities can result in a better socio-economic situation and positively impact on the quality of life of these depressive older adults [55].
Clinical trials that used an active control group, as well as usual treatment, must be carefully analysed because sometimes the results do not show reduced depressive symptoms when both groups are compared [11]. Researches [30,56] with stretching or yoga and light exercise in an active control group showed a signi cant reduction in the depressive symptomatology. It is not surprising since a metaanalysis concluded that yoga was an effective adjunct treatment of major psychiatric disorders, particularly depression and anxiety [57]. However, most often, there are improvements when the differences between baseline and post-intervention in each group are assessed, as in our study. When an exercise group is compared to an active control group instead of a passive control group, the differences between the e cacy of exercise and active placebos are much less pronounced. It is expected that doing something is likely to induce higher expectations for improvements than doing nothing [58].
Corroborating our viewpoint, a meta-analysis that investigated the predictors in control groups of exercise RCTs among adults with depression demonstrated that control group responses could negatively in uence antidepressant e cacy [59].

Bioinformatics analysis evidence
Interesting molecular interactions among genes related to BDNF, serotonin receptor (5HTR2A), IL-6, INS, APOE, and TNF were identi ed in our protein-protein network model associated exercise. Literature has shown molecular interactions among these proteins in different conditions. BDNF and 5HT2A/2C agonists interact differently in the brain. While 5HT2A/2C agonists decrease BDNF on the hippocampus, they increase it dramatically in rats' neocortex and parietal cortex [60]. BDNF is also mediated by APOE isoforms. Sen, Nelson [61] demonstrated in a recent study that cultured human astrocytes responded differently when treated with APOE2, APOE3, and APOE4. BDNF was 19% overexpressed in the astrocytes treated with APOE2 and 3, while APOE 4 decreased it 21% on average. Regarding in ammatory markers, IL-6 and TNF are pro-in ammatory cytokines that in uence cell insulin uptake, causing insulin resistance [62,63]. IL-6 and TNF-a are overexpressed in psychiatric disorders [64]. Decreased BDNF, serotonin de cits, APOE4 overexpression, increased pro-in ammatory cytokines and insulin resistance were already investigated as biomarkers of depression [63][64][65][66][67]. Furthermore, depressive patients show an atrophied hippocampus [68], which may be affected by decreased BDNF and increased cortisol. However, all of these biomarkers are modulated by exercise. For instance, IL-6 is reduced in older adults who performed regular exercise (2-5 times, weekly) [69]. In addition, neurotransmitter synthesis and trophic factors secretion (e.g. BDNF) can improve serotoninergic, noradrenergic and dopaminergic circuitry through exercise [47,70]. Hence, results from our gene network model suggest a clinical modulation of depressive symptoms by protein interactions through exercise.

Limitations
This systematic review has some limitations that must be considered. We included studies only since 2003, and this might have reduced the extent of the search scope. Concerning the missing data, selected studies in this current review do not report some essential data. Therefore, measures should be taken to improve the methodological standard of these RCT studies. Concerning heterogeneity, the random-effect model results in broader con dence intervals around the point estimates. However, it is a more conservative choice for the analysis [18]. Future research studies should evaluate exercise prescription reported in trials according to the application of the principles of exercise training (speci city, progression, overload, reversibility, diminishing returns). Also, studies should evaluate feelings of autonomous motivation in subjects with depression, as they are associated with the adherence and response to the treatment.

Conclusion
This study corroborated the additional effect of ET on PT for depression. We found that a long program with regular exercise has more effect on the treatment-response than the adherence percentage, and that ET can bene ciate older adults than adults and adolescents. Finally, the gene network model analyses involving molecular interactions in depression treatment showed biomarkers modulated by exercise, but this needs to be further explored. Availability of data and materials Data access statements, also known as data availability statements, are used in publications to describe where the data directly underpinning the publication can be found and under what conditions they can be accessed.

Competing Interests
The authors declare that they have no competing interests.