The effect of dietary anthocyanins on biochemical, physiological, and subjective exercise recovery: a systematic review and meta-analysis

Abstract Anthocyanins (ACN), the sub-class of (poly)phenols responsible for the red-blue-purple pigmentation of fruit and vegetables, have gained considerable interest in sport and exercise research due to their potential to facilitate exercise recovery. A systematic literature search was performed using PubMed, The Cochrane Library, MEDLINE, SPORTDiscus and CINAHL. Thirty nine studies were included and the standardized mean difference (Hedges g) for creatine kinase (CK), anti-oxidative and inflammatory markers, strength, power and delayed onset muscle soreness (DOMS) indices were pooled in separate meta-analyses; meta-regression was also performed on reported ACN dose. Immediately post-exercise there was an increase in antioxidant capacity (g: 0.56) and reduced C reactive protein (g: −0.24) and tumor necrosis factor α (g: −40); p ≤ 0.02. Strength was improved with ACN at all time points (g: 0.45–0.67). DOMS (g: −0.23) was lower 24 hours post-exercise and power was improved 24 hours (g: 0.62) and 48 hours (g: 0.57) post exercise. The CK was lower 48 hours post-exercise (g: −0.31) and there was a trend for a positive association with ACN dose (p = 0.057). This systematic review provides new data showing ACN-rich foods promote functional and subjective recovery likely due to the antioxidant and anti-inflammatory properties of ACN.


Introduction
Physical exercise places a degree of mechanical and metabolic stress on the body, which both contribute to a common pathological response involving oxidative stress and inflammation (Pyne 1994). Exercise induced muscle damage (EIMD) is typically characterized by an initial insult followed by a secondary inflammatory response, which is more prominent following eccentric actions (Bongiovanni et al. 2020). The EIMD has more immediate implications, including delayed onset muscle soreness (DOMS) and impaired muscular strength and power, which have the ability to compromise performance and quality of training (Clarkson, Nosaka, and Braun 1992;MacIntyre, Reid, and McKenzie 1995). Although the mechanisms and time course can differ, both mechanical and metabolic stress can cause an increase in the appearance of intracellular proteins in the blood (e.g., creatine kinase: CK) potentially due to disruptions in calcium homeostasis from a loss of cell membrane integrity (Brancaccio, Maffulli, and Limongelli 2007;Tee, Bosch, and Lambert 2007). In addition, exercise also induces signaling cascades, largely orchestrated by reactive oxygen and nitrogen species (RONS), transcriptional release of pro-inflammatory cytokines (e.g., tumor necrosis factor alpha [TNF-a] and interleukin-6 [IL-6]), and acute phase proteins, e.g., C-reactive protein; CRP (Ebbeling and Clarkson 1989;Pyne 1994) and an increase in these immunological markers are thought to be associated with muscle soreness, loss of muscle function as well as overtraining and fatigue (Gleeson et al. 1995;Hecksteden et al. 2016;MacIntyre, Reid, and McKenzie 1995). Given the potential for physiological stress associated with strenuous exercise and the potential for compromised training and/or competition performance due to loss of strength and power and muscle soreness that can last for several days, there has been a strong emphasis to identify natural recovery strategies (Bongiovanni et al. 2020;Howatson and van Someren 2008).
Amongst the available strategies, dietary interventions, particularly fruit, have gained considerable attention when it comes to improving recovery following EIMD (Doma et al. 2021;Naderi et al. 2018). Fruit could influence the recovery process because they contain (poly)phenols which could interact with the secondary cascade associated with EIMD, via their antioxidant and anti-inflammatory properties (Bowtell and Kelly 2019;Pereira Panza, Diefenthaeler, and da Silva 2015). However, not all fruit is equal in terms of (poly)phenolic content and abundance and certain fruit might be more beneficial for exercise recovery. For example, in a simple meta-analysis of 25 studies of fruit on recovery from EIMD, berries were reported to have the greatest overall effect (Doma et al. 2021). Berries are rich in anthocyanins (ACN), a subclass of (poly)phenols, responsible for the red-blue-purple pigmentation in fruits (Manach et al. 2004;P erez-Jim enez et al. 2010). These compounds have gained much research interest due to their propensity to maintain the balance between the oxidative and anti-oxidative systems and reduce inflammatory cytokines (Wang et al. 1999). In vitro ACN have been shown to act as more potent antioxidants as compared to some other (poly)phenols (Pojer et al. 2013) and to exhibit anti-inflammatory actions similar or superior to nonsteroidal and other anti-inflammatory drugs (Pereira Panza, Diefenthaeler, and da Silva 2015; Seeram et al. 2001). Additionally, ACN might supress pain at both an enzymatic (e.g., cyclooxygenase) and transcriptional (e.g., nuclear factor kappa beta) level (Pojer et al. 2013;Seeram et al. 2001). Moreover, cyanidin-3-glucoside, a common ACN found in berries (Sandoval-Ram ırez et al., 2020), has been shown to up-regulate the expression of transcriptional pathways related to muscle function and reduce fatigue in rodent models (Hu et al. 2020;Matsukawa et al. 2017). Other benefits include the potential for ACN to enhance blood flow that might aid the removal of waste products and muscle metabolites (Keane et al. 2016b;Rodriguez-Mateos et al. 2019). Recently, Bloedon et al. (2019) reported that ACN-rich whole foods reduce exercise-induced oxidative stress and inflammation, but the effect on muscle soreness and functional (e.g., strength and power) recovery, which are arguably more important measures of exercise recovery (Byrne, Twist, and Eston 2004;Damas et al. 2016;Torres et al. 2012), were not reported. Therefore whilst there is some evidence that ACN might be beneficial in facilitating recovery, the narrative nature of these reviews or lack of attention to the potential active compounds or other aspects of recovery means the conclusions are not quantitively derived or based on a complete picture of the available work (Bloedon et al. 2019;Doma et al. 2021;Harty et al. 2019;Naderi et al. 2018;Owens et al. 2019;Vitale, Hueglin, and Broad 2017). Therefore, the aim of this study was to synthesize and evaluate the effects of anthocyanin-rich foods on biochemical, physiological, and subjective indices of exercise recovery in human trials.

Search strategy
A systematic literature search of the following electronic bibliographic databases: PubMed and The Cochrane Library as well as searching MEDLINE, SPORTDiscus, and CINAHL via EBSCOhost was carried out from inception until August 2020. The search strategy (Supplementary material) was conducted using Medical Subject Heading (MeSH) and Boolean operations devised using two key concepts (1) anthocyaninrich foods and (2) exercise recovery. Furthermore, the reference list of retrieved literature reviews was hand-searched to find potential articles that could be included in the systematic review.
Exclusion criteria were non-adult, smoker or diseased participants, animal and in vitro studies. Studies were also excluded if anthocyanins are given alongside another intervention (i.e., pharmacological agent or dietary supplement; other juice or fruit to increase palatability could be included as long as anthocyanin content was reported) and no appropriate control or reference groups could be identified. Titles and abstracts were independently reviewed by two researchers (RK and CH) to evaluate their eligibility for inclusion in this review. Only full texts that were published in English or had an existing translation were retrieved and examined.

Data extraction and quality assessment
The study data was extracted into pre-piloted forms by the main reviewer (RK) and checked for accuracy by a second reviewer (KJ). Any discrepancies were resolved by reviewing the original article. The following data was extracted from each study: the first author's last name(s), publication date, funding source, participants characteristics, sample size, supplement type, ACN content, dosing strategy and duration, any dietary restrictions, wash out period and type of exercise (metabolic, mechanical or combined], outcome time points, outcome measures, mean ± SD of the outcomes specified above were also extracted. Where necessary data was extrapolated from figures and graphs and authors were contacted to provide missing data (Abbott et al. 2020;Hurst et al. 2019;Hurst et al. 2020;McCormick et al. 2016;Morehen et al. 2021), if they did not respond within 1 week a follow up was sent, those who did not reply within a month were excluded (e.g., Beals et al. 2017;Lamb et al. 2019) for variables where data could not be obtained. A modified PEDro scale (de Morton 2009) was used to assess the methodological quality of the selected studies. One point could be awarded for each the original 11 items as well as additional items thought to be relevant to the study design.
Additional criteria were as follows (1) the study acknowledged whether or not they received funding; (2) compliance to the intervention was reported; (3) ACN content was reported in the supplement either according to the manufacturer's nutritional label or confirmed by analysis in the study; participants refrained from taking antioxidant and anti-inflammatory drugs and supplements (4) before and/or (5) during study (6) sample size calculation was included. In studies where a cross-over design an additional item was included "(7) a minimum 7-day washout between trial treatments." Thus, a total of 17 points could be awarded for parallel studies and 18 for crossover studies. For parallel studies a score of <7, 7-10, 11-14, and 15-17 and for crossover studies <8, 8-11, 12-15, and 16-18 was poor, fair, good and excellent, respectively (Doma et al. 2021). Risk of bias was also assessed according to Cochrane Collaboration guidelines and is represented graphically to indicate the overall quality of all studies (Higgins and Green 2011).

Statistical analysis
Standardized mean differences (SMD) were calculated using Hedge's g using independent groups and for parallel studies and paired groups for crossover studies (Borenstein et al. 2019). To calculate the standard deviation within groups for crossover studies the correlation between pairs of observations (r; which was calculated from studies where individual data was provided (Hurst et al. 2019;McCormick et al. 2016) and assumed to be 0.5 (Amiri et al. 2019;Doma et al. 2021;Higgins and Green 2011)) was included. Both study designs were included in an inverse random effects metaanalysis (due to study design heterogeneity) using Stata v.16.0 (StataCorp, College Station, Texas, USA) sub-grouped by study design to determine whether the inclusion of crossover designs influenced the SMD (Supplemental information (Higgins and Green 2011)). Where there were sufficient studies (Jackson and Turner 2017) separate meta-analyses were conducted for immediately post ( 2 h), 24 hours post and 48 hours post exercise. For studies that reported measures over several time points, the data were only analyzed for the most recent in that time interval. Hurst et al. (2019) reported different doses so these were pooled to get an overall ES before inclusion in the meta-analysis (Higgins and Green 2011). If DOMS was measured at different sites, the largest ES was included in the meta-analysis.
Secondly, sensitivity analysis was performed by omitting one study at a time to evaluate the potential bias and robustness of the overall SMD. Heterogeneity between studies was determined by the I 2 statistic. For the I 2 statistic, I 2 values 25%, 50%, 75%, and >75% indicated no, little, moderate, and significant heterogeneity, respectively. To identify potential sources of heterogeneity, moderator analysis was performed using sub-group analysis for categorical variables including training status, exercise type and study duration. In addition, where ACN content was reported a meta-regression was conducted on the most reported variables (MVC, DOMS and CK). Potential publication bias for each outcome was evaluated by Egger's test (p < 0.10) and visual inspection of funnel plots (Begg and Mazumdar 1994). Where publication bias was detected, trim and fill analysis was conducted (Steichen 2010). The SMD were interpreted as small (>0.2), moderate (>0.5) and (!0.8) large (Sullivan and Feinn 2012) and a Z effect p < 0.05 was considered significant.

Literature search and study characteristics
The search results are presented in Figure 1, following full search and exclusion of irrelevant articles 39 articles were included in this review. A total of 27 independent group studies and 12 crossover studies with 767 participants were included in this review (Table 1). Of the interventions used tart cherry was the most common (18 studies). Other studies used blackcurrant (6 studies), grape (6 studies), blueberry (3 studies), chokeberry (3 studies), bilberry (1 study), plum (1 study) and one a mixed anthocyanin cocktail. The duration of the studies varied greatly with some investigating the acute influence (1-2 h before), most investigating the shortterm influence (2-10 days) and some the longer-term influence (20 days-8 weeks) of ACN. Most studies were in trained individuals and the median age was 24 (range 18-48) years. The median ACN content, where reported, was 80 (range 8-3600) mg/day. The quality of studies was rated as poor (n ¼ 1), fair (n ¼ 8), good (n ¼ 19), excellent (n ¼ 11; Table 1). The risk of bias is presented in Figure 2, which showed the percentage of studies with low, medium and high risk of bias for each domain. The main potential sources of bias came from allocation, blinding of the intervention or did not acknowledged whether they received funding (other bias).

Subgroup analysis, publication bias and meta-regression
Subgroup analysis is presented in Table 2. Immediately post-exercise ACN increased TAC and decreased SOD, only in metabolically biased exercise. IL-6 was also reduced by metabolic-type exercise immediately post and 48 hours post exercise. MVC was improved immediately post-exercise with mechanically damaging exercise, whereas 24 hours post and 48 hours post there was a large effect for exercise that had a combined mechanical and metabolic component. CRP was lower immediately post exercise, DOMS at 24 hours post exercise and CK at 48 hours post-exercise with combined-type exercise. MVC was improved immediately post-exercise and 24 hours post-exercise in trained and untrained individuals, but only 48 hours post-exercise in trained participants. CRP was lower immediately post-exercise in trained individuals, whereas in the untrained participants IL-6 was lower. Twenty-four hours post-exercise TNFa and IL-6 was lower in untrained, whereas trained participants had lower SOD and DOMS at 24 hours and CK at 48 hours.
Studies of a shorter duration decreased CRP immediately post-exercise and improved MVC and DOMS 24 hours post and CK and CMJ 48 hours post. Studies of a longer duration Increased TAC and reduced SOD, TNFa and CK immediately post. TAC remained higher at 48 hours and TBARS and TNFa were reduced in the longer duration studies. There was evidence of publication bias for TBARS (p ¼ 0.001) and DOMS (p ¼ 0.009) immediately post-exercise. After screening the titles and abstracts, 62 of the records were deemed potentially eligible for inclusion and full texts were retrieved for further evaluation. Twenty-five articles were excluded and a further 2 found from hand searching, leaving 39 included studies. So too were TAC, CMJ and MVC (p < 0.058) 24 and 48 hours post exercise and IL-6 (p ¼ 0.004) 48 hours post-exercise. No other publication bias was detected. Trim and fill analysis was done for the above variables resulting in lower DOMS immediately post-exercise (SMD: À0.33; 95% CI: À0.60, À0.06) higher CMJ at 24 hours (SMD: 0.47; 95% CI: 0.12, 0.81) and increased TAC at 48 hours post-exercise (SMD: 0.35; 95% CI: 0.04, 0.67). With trim and fill analysis CMJ was no longer significantly higher at 48 hours (SMD: 0.18; 95% CI: À0.10, 0.47), but there was no other materially different SMDs. Meta-regression suggested a trend for a weak positive association with ACN dose and CK (Figure 4) at 48 hours (p ¼ 0.057; I 2 ¼0%), however there was no relationship with MVC or DOMS or CK at any other time point.

Discussion
The present study represents the most comprehensive picture that synthesizes and evaluates the effects of dietary ACN on exercise recovery from all the available literature including additional analyses that consider ACN dose, exercise type, training status, and study duration. These new data showed a beneficial effect for ACN on biochemical, physiological, and subjective recovery following exercise up to and including 48 hours post-exercise. Dietary intake of ACN resulted in an increase in total antioxidant capacity/status immediately post exercise, which was mirrored by a reduction in SOD at the same point which was still reduced 24 hours post exercise, suggesting less reliance on these defence systems over time due to the ability of the ACN to scavenge free radicals (Skarpa nska-Stejnborn, Basta, and Pilaczy nska-Szcze sniak 2006). Dietary ACN have antioxidant potential due to the ability for hydrogen (electron) donation and the positively charged oxygen in the flavonoid molecule (Bi et al. 2014). Moreover, the time course aligns with plasma maximum concentrations of ACN and their metabolites, which typically occurs 1-2 hours after ingestion (Hurst et al. 2019;Keane et al. 2016a). In accordance, a number of studies included in this analysis that measured these indices gave an acute dose preexercise that would coincide with the peak plasma concentrations ( de Lima Tavares Toscano et al., 2020;Hurst et al. 2019;Hurst et al. 2020;Lyall et al. 2009;McAnulty et al. 2014;Silvestre et al. 2014). Interestingly, the antioxidant effects of ACN were predominantly seen in exercise with a major metabolic component, which might be attributable to greater exercise-induced oxidative stress owing to higher oxygen consumption during the exercise, whereas a delayed and prolonged generation of RONS after mechanically strenuous eccentric exercise is likely (Fisher-Wellman and Bloomer 2009), because of the secondary inflammatory-mediated damage that occurs after exercise (Howatson and van Someren 2008;Owens et al. 2019;Bongiovanni et al. 2020).
The consumption of ACN resulted in reduced CK at 48 hours post, and inflammation (TNFa and CRP) to be reduced immediately post-exercise. As there is an inherent interplay between these markers and RONS (Baird et al. 2012;Lee and Clarkson 2003), the early antioxidant actions    of the ACN represent one potential mechanism that might supress the efflux of CK (through reduced cell membrane disruption) and inflammatory indices. The anti-inflammatory properties of ACN are well documented (Fallah et al. 2020;Speer et al. 2020), and might relate to their ability to interact with cellular enzymes and signaling pathways (Li et al. 2017). For example, ACN have been shown to reduce inflammatory enzymes such as cyclooxygenase and lipoxygenase (Kirakosyan et al. 2018;Wang et al. 1999), which might be mediated by their ability to inhibit mitogen-activated protein kinase and nuclear factor kappa beta pathways (Pojer et al. 2013). Dependent on exercise modality, intensity and duration CK has been shown to peak 24-72 hours after exercise (Baird et al. 2012). Whereas an acute inflammatory response due to immunological activation typically occurs more rapidly (1-4 hours) and a second wave of inflammation is detectable in a similar timeframe to peak CK (Peake et al. 2017). A lower peak in the CK and inflammatory indices might reflect a reduction in muscle damage and also indicate a faster recovery after exercise with ACN compared to a control. While this might relate to the antioxidant capacity of the ACN it could also be because of improved blood flow and clearance (Baird et al. 2012;Rodriguez-Mateos et al. 2019). Other meta-analyses have not suggested an effect of fruit or (poly)phenols on CK (Doma et al. 2021;Hill et al. 2021), it should be acknowledged there are several criticisms of CK as a marker for muscle damage especially owing to its high inter and intra-individual variability and its meaningfulness as a recovery index (Brancaccio, Maffulli, and Limongelli 2007;Hill et al. 2021;Warren, Lowe, and Armstrong 1999). However, some included studies in the current review found a benefit of ACN-rich foods (Carvalho et al. 2018;(Lyall et al. 2009) and an ACN rich cocktail on exercise-induced CK (Lima et al. 2019) it therefore might be that some (poly)phenols such as ACN are more beneficial than others. Moreover, the large number of pooled studies at 48 hours post exercise might account for some of the variability, where the participant numbers amounted to 244.
The aforementioned supports the notion that ACN improves biomarkers related to exercise recovery. However, the influence on symptoms such as functional (i.e., strength and power) and muscle soreness indices perhaps are better representations of recovery facilitation and EIMD (Byrne, Twist, and Eston 2004;Damas et al. 2016;Torres et al. 2012). There was an effect of ACN on reducing DOMS at 24 hours and recovery of strength loss 0, 24 and 48 hours post exercise, whereas power was only increased 24 and 48 hours post-exercise. Reduced strength loss, and recovery of strength, was greater with ACN, initially for eccentrically biased exercise, but CMJ and MVC were improved 24 hours and 48 hours post-exercise after combined metabolic and mechanically strenuous exercise. Both mechanical and metabolic exercise increase RONS due to mitochondrial oxygen consumption, the increased circulating catecholamines, elevated participation of eccentric muscle contraction-induced damage, inflammatory response and/or the intermittent and repeated sprint actions that can cause temporary ischemicreperfusion in the skeletal muscle (Ascensão et al. 2008;Leeuwenburgh and Heinecke 2001). Strength loss after exercise has been proposed to be related to oxidative stress (Çakir-Atabek, Dokumaci, and Ayg€ un 2019), whereas loss in muscle power might be more synonymous with DOMS and the inflammatory response (Byrne, Twist, and Eston 2004). Speculatively, the early increase in antioxidant capacity with ACN might help to reduce strength loss, whereas the recovery in power coincides with the reduced DOMS at 24 hours post-exercise. These data are of great interest because therapeutic recovery interventions (e.g., massage, cold water immersion and compression garments) have shown some benefits in recovery of DOMS, strength and power, but there are limited data to suggest that all facets of recovery can be affected in a positive manner (Brown et al. 2017;Davis, Alabed, and Chico 2020;Leeder et al. 2012). Whereas, in this review, ACN-rich foods are shown to improve physiological and subjective recovery following strenuous exercise and hence should be an integral consideration for practitioners and exercisers to consider in their diet.
Notwithstanding, there are several limitations within the included studies that warrant discussion. Firstly, studies with a crossover study design were included in the meta-analysis and these could be influenced by the repeated bout effect (RBE) between experimental trials. The RBE refers to the protective effect afforded by a single bout of eccentric-biased muscle actions that provide a protective effect on subsequent bout of exercise (even if this is performed on the contralateral limb) and hence could mask any treatment effect (Howatson and van Someren 2007). However, including crossover studies did not appear to add to heterogeneity to the results (Figure 3). Secondly, some studies which investigated the effects on functional and subjective recovery after "real" game play (Abbott et al. 2020;Kupusarevic, McShane, and Clifford 2019;Morehen et al., 2021); while these arguably have good application they are heavily confounded by the RBE as well as other recovery practices that might be conducted concurrently. Conversely, some studies used    dietary restrictions (Table 1) to reduce phenolic intake. This might lead to an overestimate in the effect, as removal of natural antioxidants from the diet might conceptually impair the natural recovery process; therefore ACN might only restore antioxidant capacity whereas the placebo remains in a depleted state. The balance between reducing background noise and ecological validity needs careful consideration in research designs (Bowtell and Kelly 2019). Thirdly, there was a large difference between ACN content of the interventions and it is not possible to distinguish between different types of ACN, which could have different bioactivities (Rechner and Kroner 2005). Notwithstanding, ACN content is often reported as cyanidin equivalents (Bell et al. 2016;Brown, Stevenson, and Howatson 2019;Hutchison et al. 2016;O'Connor et al. 2013) and this compound is an established biomarker of berries (Sandoval-Ram ırez et al. 2020) and tart cherries (Seymour et al. 2014) suggesting at least some commonality between the interventions. Moreover, this is the first review to comprehensively study ACN on exercise recovery, including a meta-regression of ACN dose. Nonetheless, future studies should try to distinguish the optimum type and dosage of anthocyanins for recovery, an important factor highlighted in a recent review (Sabou et al. 2021). Lastly, blinding of studies was a major source of bias, although it is acknowledged that this is an inherent challenge with studies involving functional foods (Brown et al. 2018). Therefore, results from this meta-analysis have to be interpreted in light of limitations of the literature highlighted above. Nonetheless, the meta-analytical technique is currently the best method to systematically consolidate evidence from previous work (Haidich 2010), but it should be conducted with a forensic eye of the literature in order to interpret the information with insight. To summarize, ACN were shown to have an overall beneficial effect on reducing CK, muscle soreness, strength loss and improving power after exercise. This was accompanied by attenuated inflammation and increased antioxidant capacity/status following the intake of ACN, suggesting a potential causal link. The information provided by subgroup analyses suggested the most beneficial effect on the biomarkers are following metabolically biased exercise and longer-term interventions; whereas shorter duration interventions saw most benefit on physiological variables, which can collectively help inform research designs and application of ACN in exercise recovery. These data provide new information to support the use of ACN-rich foods in promoting recovery following strenuous exercise that can inform exercisers and practitioners.

Acknowledgments
The Authors would like to acknowledge Chelsey Hart who helped with the screening process.