Using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines ^{27– 29} , two databases (PubMed and EMBASE) were searched up until 19 ^th August 2016. Two investigators (PCD and IML) independently conducted two identical searches in both databases using appropriate search algorithms (PubMed: Supplementary File 1; EMBASE: Supplementary File 2). The lists of the included articles were reviewed to identify publications that were relevant to the topic under review.

We included studies that met at least one of the following eligibility criteria: a) measurements of PGC-1a (mRNA and/or protein concentrations) in conjunction with measurements of FNDC5; b) measurements of FNDC5, and/or Irisin concentrations and/or UCP1 in WAT, along with the following criteria: c) measurements of physical activity levels and/or exercise interventions, and d) human participant study. No other eligibility criteria were set (e.g., language, date of publication). From the included studies, we retrieved outcomes regarding the effects of physical activity on PGC-1a in conjunction with FNDC5 in muscle, FNDC5 in muscle, Irisin in the bloodstream and UCP1 in WAT. We report the studies’ design, the participants’ characteristics, the Irisin identification and other outcome methods and study outcomes. We have also recorded the secondary associations in the included studies, i.e. associations between FNDC5 and/or circulating Irisin and several health-related phenotypes [e.g. energy expenditure, blood pressure, waist to hip ratio, body mass index (BMI)].

Two independent reviewers (PCD and GSM) evaluated the risk of bias of the studies included in the current review via the “Cochrane Collaboration’s tool for assessing risk of bias” ³⁰ . Conflicts in the risk of bias assessment were resolved by IL and ADF. We also evaluated independently (PCD and GSM) the quality of reporting in the included randomised controlled trials (RCTs), controlled trials (CTs) and single group design studies (SGS) using the Consolidated Standards of Reporting Trials (CONSORT) checklist ³¹ , which is a 25-item checklist and we provided a score for each study included. For CTs and SGS, we used a modified CONSORT checklist comprised of 18 items, given that these studies are not RCTs and therefore, seven out of the 25 items of the CONSORT checklist are not applicable for CTs and SGS (i.e. randomization, blinding). We also evaluated independently (PCD and GSM) the quality of the reporting data of the included cross sectional studies (CSS) using the 22-item checklist of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) and we also provided a score for each study included ³² . Disagreements on studies’ CONSORT and STROBE scores were arbitrated by IL and ADF. JT and PS then reviewed the molecular and genomic content of the review independent of the search process.

The initial searching date was the 14 ^th September 2015 while weekly alerts were received from both databases up until the 19 ^th August 2016. Overall, the searching procedure revealed 51 studies that involved 2474 participants and met the inclusion criteria, and were therefore included in this systematic review. The reference lists of these studies did not result in the identification of additional relevant articles. The searching outcome is presented in a PRISMA flow diagram in Supplementary Figure 1.

The characteristics and the results of the included studies can be found in Table 1. From the 51 eligible studies, 12 (23.5%) were RCTs, of which four were cross-over RCTs, eight (15.7%) were CTs, 23 (45%) were SGS, and eight (15.7%) were CSS. One of the included RCTs ³³ reported the effect of resistance exercise training versus the effects of resistance exercise training combined with Ursolic supplementation, because for the latter group the effects of resistance exercise cannot be isolated, we will report only the results from the resistance exercise training group. Furthermore, one of the CTs ³⁴ will be included in the results of both CTs and CSS because this study consisted of a controlled trial nested within a CSS. Eight of the included studies examined overweight/obese adults and children ^{17, 35– 41} , while 11 studies included a clinical population, including patients with chronic obstructive pulmonary disease (COPD) ^{23, 34, 42} , heart failure ⁴³ , metabolic syndrome ⁴⁴ , haemodialysis ⁴⁵ , osteoporotic ⁴⁶ , anorexia nervosa ^{36, 47} , pre-diabetes ¹⁶ and diabetes type II ⁴⁸ .

The estimated risk of bias assessment results can be found in Table 2, and a summary is displayed in Supplementary Figure 2. Five RCTs ^{44, 49– 52} , and all the included CTs and CSS, as well as 22 of the 23 SGS, displayed a high risk of bias due to inadequate generation of a randomised sequence, while four RCTs ^{23, 33, 42, 53} showed low risk of bias, and three RCTs ^{41, 54, 55} , as well as one SGS ⁵⁶ , showed unclear risk of bias because there was no description of the method used for allocation (even though the participants were said to be “randomly” assigned). Six RCTs ^{23, 42, 49, 50, 52, 53} displayed low risk of bias for “allocation concealment”, while two ^{44, 54} showed unclear risk of bias because of the lack of description of the randomization allocation. Also, four RCTs ^{33, 41, 51, 55} , and all the included CTs and SGS, as well as CSS, showed high risk of bias due to the lack of concealment of allocations before assignment. In “blinding of participants and personnel”, all RCTs, CTs, SGS and CSS displayed high risk of bias because the exercise interventions could not be blinded to the participants.

In “blinding of outcome assessment”, three RCTs displayed low risk of bias ^{23, 53, 54} , while five RCTs ^{42, 44, 49, 50, 52} and one CT ¹⁶ showed unclear risk of bias because of the lack of information regarding the blinding of assessments. Also, four RCTs ^{33, 41, 51, 55} , the remaining seven CTs, and all the included SGS and CSS showed high risk of bias due to the knowledge of the allocated interventions by the assessors. Seven RCTs ^{23, 33, 41, 42, 51, 53, 54} , one CT ³⁷ , five SGS ^{39, 56– 59} and one CSS ⁶⁰ displayed low risk of bias, while five RCTs ^{44, 49, 50, 52, 55} , the remaining seven CTs, the remaining 18 SGS and the remaining eight CSS showed unclear risk of bias for “incomplete outcome data” because of the lack of information on the participants who dropped out or exclusions in the analysis. All the included studies showed low risk of bias of “selective reporting” because they reported all the outcomes measured, and all the included studies displayed low risk of bias in “other bias”.

The results of our evaluation in the quality of the reporting data showed a mean score of 13.6 out of 25 (54.4%) for the included RCTs, 10.56 out of 18 (58.68%) for the included CTs and 10.52 out of 18 (58.44%) for the included SGS ( Table 3). The CSS displayed a mean score of 13.37 out of 22 (60.8%) ( Table 4). The score represents the number of items (with percentage of items) on the checklist that were reported satisfactorily in each study. Therefore, a high score represents a high adherence to reporting guidelines, while a low score represents low adherence to reporting guidelines.

The link between PGG-1a and FNDC5 in muscle in response to physical activity/exercise

Acute effects of exercise

Five studies ^{16, 17, 50, 61, 62} investigating the link between PGC-1α with FNDC5 in muscle in response to acute exercise showed an increase of the PGC-1α mRNA in muscle; however, only two studies ^{16, 62} also found an increase in muscle FNDC5 mRNA, while one study ⁴³ detected a positive association of PGC-1α with FNDC5 in muscle. More specifically, a study found that an aerobic (2.1±0.8-fold over baseline, p=0.05) and a resistance (3.5±0.9-fold over baseline, p=0.01) training session increased PGC-1α splice variant1 but it did not change FNDC5 mRNA in the muscle of healthy adults ⁵⁰ . Similarly, a resistance training session increased PGC-1α splice variant1 four hours post exercise (200%, over baseline and over control, p<0.05), but it did not change FNDC5 mRNA in the muscle of healthy adults ⁶¹ . A 45-minute endurance exercise session increased Exon 11 of PGC-1α mRNA in muscle (7.4-fold over baseline, p<0.05), but it did not change FNDC5 mRNA in muscle in both healthy and pre-diabetic adults, while a positive association between PGC-1α and FNDC5 mRNA was found at baseline (r=0.82, p<0.01) when data of the two groups were combined ¹⁶ . Furthermore, PGC-1α mRNA in muscle increased (>6-fold over baseline, p<0.05) in response to acute exercise; however, FNDC5 mRNA in muscle was not altered in sedentary overweight and obese adults ¹⁷ . Also, a resistance exercise session increased Exon 11 of PGC-1α mRNA in muscle of both young (4-fold over baseline, p<0.05) and older (2-fold over baseline, p<0.05) healthy adults, while it increased FNDC5 mRNA in muscle only in young (1.4-fold over baseline, 95% Confidence Interval=0.3–2.2, p<0.05) healthy adults ⁶² . Finally, PGC-1α mRNA in muscle was positively associated with FNDC5 mRNA in muscle (r=0.56, p<0.05) in a sub-set of 24 patients with heart failure ⁴³ ; stratification was ad hoc.

Chronic effects of exercise

Of the eleven eligible studies ^{2, 16– 19, 40, 62– 67} that examined the link between PGC-1α with FNDC5 in muscle in response to chronic exercise, only two ^{16, 66} showed that chronic exercise increased PGC-1α and FNDC5 mRNA in muscle, while four studies ^{18, 19, 62, 63} showed no effect of chronic exercise on PGC-1α and FNDC5 mRNA in muscle. In the five studies that only measured FNDC5 in muscle, one study ² found increased and four ^{17, 40, 64, 65} showed no effect of chronic exercise on FNDC5 mRNA in muscle.

A 12-week of endurance and resistance combined exercise training increased Exon 11 of PGC-1α mRNA in muscle (1.2-fold in healthy and 1.6-fold in pre-diabetic adults over baseline, p<0.05) and FNDC5 mRNA in muscle (1.4-fold in healthy and 2-fold in pre-diabetic adults over baseline, p<0.05) ¹⁶ . Furthermore, an 8-week sprints exercise program increased PGC-1a and FNDC5 mRNA in muscle (p<0.05) in healthy adults ⁶⁶ . Finally, Bostrom et al. (2012) showed that in eight older participants selected from a larger group of 27 participants, chronic exercise increased FNDC5 mRNA in muscle (p<0.05) ² .

A 21-week endurance and resistance combined exercise program in healthy adults did not alter PGC-1α and FNDC5 mRNA in muscle ⁶² . One of the included studies ¹⁹ found no effect of chronic exercise on PGC1a or FNDC5 mRNA in younger adults (despite detecting significant changes in ~1,000 other mRNAs and finding mitochondrial enzyme activity was increased in ~25%) ⁶⁸ . Similarly, an 8-week resistance exercise program did not alter PGC-1α or FNDC5 mRNA in muscle of young healthy adults ⁶³ . In addition, 12 weeks of resistance training did not alter PGC-1α splice variant1 mRNA, and it did not change the FNDC5 mRNA in muscle in untrained young females ¹⁸ . Also, a 12-week aerobic and resistance exercise combined program ¹⁷ and an 8-week aerobic exercise program ⁴⁰ did not alter FNDC5 mRNA in muscle of sedentary obese adults, while chronic exercise had no effect on FNDC5 mRNA in muscle of healthy adults ⁶⁴ . Finally, a 3-week sprint interval training program did not alter FNDC5 mRNA in muscle of healthy adults ⁶⁵ .

The effects of physical activity/exercise on Irisin

Acute effects of exercise

Studies using enzyme-linked immunosorbent assays (ELISA)

Eighteen of the included studies ¹² , ¹⁷ , ²¹ , ²² , ³⁴ , ³⁵ , ³⁸ , ⁴⁴ , ⁴⁹ , ⁵⁰ , ⁵² , ⁵⁶ , ⁵⁸ , ⁶² , ⁶⁶ , ^{69– 71} examined the effects of acute exercise on circulating Irisin, and a further seven studies ^{34, 36, 46– 48, 60, 72} investigated the association of circulating Irisin with physical activity levels using commercial ELISA kits. Thirteen studies ^{12, 21, 22, 38, 44, 49, 50, 52, 58, 66, 69– 71} showed that acute exercise increased circulating Irisin in healthy individuals, while five studies ^{17, 34, 35, 56, 62} showed no effect of acute exercise on circulating Irisin. Also, three studies ^{34, 48, 60} showed a positive association of circulating Irisin with physical activity levels in healthy and COPD patients, while four studies ^{36, 46, 47, 72} showed no association or a negative association of circulating Irisin with physical activity levels in both healthy and clinical populations.

A resistance training session did not change FNDC5 mRNA in the muscle of healthy adults and circulating Irisin increased (p<0.001) over the following 24-hour ⁵⁰ , indicating no short-term association between FNDC5 and Irisin. Furthermore, an aerobic exercise session increased circulating Irisin (p=0.04) and Irisin concentrations were measured at ~355–459 ng/ml ⁵⁰ , greater than recent mass spectrometry measurements ⁷³ . Similarly, a running exercise session in healthy individuals ⁴⁹ and an aerobic exercise session, as well as a resistance exercise session, in healthy individuals and in metabolic syndrome patients ⁴⁴ increased circulating Irisin (p<0.05). In the latter studies, Irisin concentrations measured at ~99–175 ng/ml ⁴⁹ and ~80–94.6 ng/ml ⁴⁴ , respectively, which is greater than recent mass spectrometry measurements ⁷³ . Also, an acute resistance exercise session increased circulating Irisin (p<0.05) as oppose to aerobic and combined (aerobic and resistance) sessions that did not alter circulating Irisin in healthy males (Irisin concentrations ~18–151 ng/ml) ⁵² . Furthermore, a 90-minute aerobic exercise session increased circulating Irisin during (54 ^th minute) the exercise session (20.4% compared to baseline, F _(3,36)=5.28, p=0.004), but circulating Irisin decreased after the exercise session (p=0.021) in healthy male adults ⁶⁹ . In the latter study, the aerobic exercise session also increased circulating Irisin during (54 ^th minute) the exercise session (F _(3,24)=5.03, p=0.01) in healthy female adults ⁶⁹ . Eight out of the 23 included SGS showed that acute exercise increased circulating Irisin in healthy populations ^{12, 21, 22, 38, 58, 66, 70, 71} , while a resistance exercise session increased FNDC5 mRNA in muscle only in young healthy adults and it did not alter circulating Irisin of both young and older healthy adults ⁶² . In addition, 45 minutes of running did not alter circulating Irisin in obese healthy adults ³⁵ . Similarly, an acute cycling session did not alter circulating Irisin in COPD patients ³⁴ , while an acute exercise session did not alter FNDC5 mRNA in muscle or circulating Irisin in sedentary overweight and obese adults ¹⁷ . Finally, an acute exercise session of both low and high intensity resistance training did not alter circulating Irisin (p>0.05) in sedentary young healthy females (Irisin concentrations ~69–87 ng/ml) ⁵⁶ .

Physical activity levels were positively associated with circulating Irisin in healthy adults (r=0.20, p=0.03), but not in patients with diabetes type II ⁴⁸ , and they were not associated with circulating Irisin in osteoporotic women ⁴⁶ and in anorexic women ⁴⁷ . Furthermore, circulating Irisin concentrations were higher in physically active (Irisin concentrations 128.55±78.71 ng/ml) than in sedentary individuals (Irisin concentrations 105.66±60.2 ng/ml) (p=0.006) ⁷² . However, physical activity levels were negatively associated with circulating Irisin (r=−0.22, p=0.001) in groups of anorexic, obese and healthy women ³⁶ , while they were positively associated with circulating Irisin in both COPD patients (r=0.83, p<0.01) and healthy individuals (r=0.79, p<0.001) ³⁴ . Finally, circulating Irisin was positively correlated with physical activity levels in individuals who demonstrated high weekly physical activity energy expenditure (2050–3840 kcal/week) (Irisin concentrations ~32–261 ng/ml, p=0.04).

Studies using mass spectrometry and western blotting

Only one included study used both western blotting and mass spectrometry to detect circulating Irisin in response to acute exercise. This study showed that submaximal acute aerobic exercise increased circulating Irisin (3.1-fold over baseline, p<0.05), whereas maximal acute aerobic exercise did not alter circulating Irisin, even though tended to be significant (p=0.07), in two healthy volunteered adults ²⁴ .

Chronic effects of exercise

Studies using ELISA

Twenty three included studies ^{12, 16, 18, 22, 23, 33, 34, 37, 39, 41, 42, 45, 51, 53– 57, 59, 62, 65, 70, 74} in the current review examined the effects of chronic exercise on circulating Irisin using commercial ELISA kits, while the populations examined showed large heterogeneity. Nine studies ^{23, 34, 39, 41, 51, 54, 59, 65, 74} showed that chronic exercise increased circulating Irisin, while 12 studies ^{12, 16, 18, 22, 33, 37, 42, 45, 53, 57, 62, 70} showed no effects of chronic exercise on circulating Irisin, and two studies showed that chronic exercise decreased circulating Irisin ^{55, 56} , in both healthy and clinical populations.

A 6-month resistance training program increased circulating Irisin in healthy controls (p<0.01), but not in the exercisers ⁵⁴ , while an 8-day vibration exercise increased circulating Irisin in COPD patients (p=0.01) ²³ . Notably, the Irisin concentrations in the latter study ²³ were ~785–1196 ng/ml, a lot greater than recent mass spectrometry based detection of Irisin concentrations ⁷³ . Furthermore, a 12-week resistance exercise increased circulating Irisin in elderly healthy females (Irisin concentrations ~61–83 ng/ml, p<0.05,) ⁵¹ . In addition, a 12-week of endurance and resistance combined exercise training in both healthy and pre-diabetic adults increased FNDC5 mRNA in muscle, while it decreased circulating Irisin (p<0.05) when the data of both healthy and pre-diabetic groups were combined ¹⁶ . In the latter study, Irisin concentrations were detected at 160 ng/ml at baseline and 143 ng/ml after the exercise program, a lot greater than recent mass spectrometry based detection of Irisin concentrations ⁷³ . In addition, an 8-week endurance training program increased circulating Irisin only in middle-aged and not in young healthy adults (Irisin concentrations ~140–168 ng/ml, p<0.05) ⁷⁴ , while an 8-week chronic exercise program in COPD patients increased circulating Irisin (p<0.05) ³⁴ . Finally, a 12-month physical activity intervention increased circulating Irisin by ~12% (p=0.001) in obese children ³⁹ . Notably, in the latter study, Irisin concentrations were 111 ng/ml, a lot greater than recent mass spectrometry based detection of Irisin concentrations ⁷³ .

A 3-week sprint interval training program did not alter FNDC5 mRNA in muscle and showed a gender difference in circulating Irisin, which was decreased in healthy males and increased in healthy females (p<0.05) ⁶⁵ . An 8-week resistance exercise training program increased circulating Irisin compared to control group (p<0.05), while the Irisin concentrations were ~700–850 ng/ml ⁴¹ . Similarly, 3-month cross-fit training increased circulating Irisin (Irisin concentrations ~300–850 ng/ml, p<0.05) only in females ⁵⁹ . On the other hand, a 4-week sprint exercise training decreased circulating Irisin (Irisin concentrations ~200-340 ng/ml, p<0.05) in healthy males ⁵⁵ . Three months of both non-individualized training and individualized training did not alter circulating Irisin (Irisin concentrations ~123–131 ng/ml, p>0.05) in COPD patients ⁴² . Finally, an 8-week low intensity resistance training program did not alter circulating Irisin, while an 8-week high intensity resistance training program reduced circulating Irisin (Irisin concentrations ~51–87 ng/ml, p=0.03) ⁵⁶ .

An 8-week resistance training program in healthy adults did not alter circulating Irisin ³³ and a 26-week aerobic exercise program revealed no changes in circulating Irisin of healthy adults ⁵³ . A 21-week endurance and resistance combined exercise program in healthy adults did not alter FNDC5 mRNA in muscle and circulating Irisin ⁶² . Similarly, a 16-week resistance exercise program in elderly women did not increase circulating Irisin ³⁷ and 12 weeks of resistance training did not alter FNDC5 mRNA in muscle or circulating Irisin ¹⁸ . However, circulating Irisin was positively correlated with FNDC5 mRNA in muscle (r=0.65, 95% Confidence Interval=0.12–0.89, p<0.05) in the latter study ¹⁸ . Finally, five SGS showed that chronic exercise did not alter circulating Irisin in healthy individuals ^{12, 22, 57, 70} and haemodialysis patients ⁴⁵ .

Studies using mass spectrometry and western blotting

Only two included studies used alternative methods than commercial ELISA kits to detect human circulating Irisin in response to chronic exercise. Initially, Bostrom et al. (2012) showed via western blotting that in eight older participants selected from a larger group of 27 participants ⁶⁷ chronic exercise increased FNDC5 mRNA in muscle (p<0.05) and circulating Irisin (2-fold over baseline, p<0.05) ² . Finally, one study contrasted plasma Irisin concentrations in six younger individuals following 12 weeks high intensity aerobic exercise with those found in a separate group of four individuals (no pre-training samples were presented) ⁷³ . This study used mass spectrometry and detected circulating Irisin at 3.6 ng/ml in controls and 4.3 ng/ml in exercisers, which was significantly different between the two groups (p=0.04). No details regarding training or control of hydration in the training group were reported ⁷³ .

We located only one study that examined the effects of exercise on UCP1 mRNA in subcutaneous WAT in humans. This study found that a 12-week intervention of endurance and resistance combined exercise in both healthy and pre-diabetic adults had no significant effect on UCP1 mRNA in subcutaneous WAT, even though UCP1 mRNA was increased (1.82-fold over baseline, p<0.05) when data from both groups were combined ¹⁶ . Also, UCP1 mRNA did not associate with FNDC5 mRNA in muscle (r=0.28, p=0.18) and circulating Irisin (r=-0.11, p=0.60) ¹⁶ .

The secondary results of the included studies can be found in Table 1. In 118 muscle profiles, FNDC5 mRNA was modestly and positively correlated with BMI (r ²=0.1, p=0.004), while FDNC5 mRNA was not related to fasting glucose or glycaemic control ¹⁹ . Furthermore, circulating Irisin was not associated with inflammatory indices ³⁹ , blood glucose ^{62, 65} , homeostatic model assessment (HOMA) ^{62, 65, 71} , insulin ^{62, 65, 71} , leptin ⁷¹ , lean body mass ^{46, 57} , fat mass ^{18, 46, 57} , waist to hip ratio ⁷¹ , energy expenditure ^{21, 54} , BMI ⁷¹ , and pulmonary function ³⁴ .

Additional secondary results show that circulating Irisin was positively associated with BMI ^{59, 70, 72, 75} , triglycerides ^{70, 72} , fat mass ^{36, 59} , HOMA ⁷² , insulin ⁷² , blood glucose ⁷¹ and leptin ⁷⁵ , and negatively with high density lipoprotein cholesterol ⁷⁰ , all of which indicate unfavourable effects of Irisin on human health. Nevertheless, some secondary evidence suggests that circulating Irisin was positively associated with fat free mass ^{36, 70} , muscle mass ⁴¹ and energy expenditure ³⁶ , and Irisin that was incubated within white adipocytes in vitro increased glucose and fatty acids uptake ⁶⁶ . Furthermore, circulating Irisin after a maximal workload was significantly greater in individuals with higher VO ₂max than individuals with lower VO ₂max ²¹ . However, circulating Irisin was not associated with VO ₂peak before and post exercise in healthy females ⁵⁷ and sedentary overweight and obese individuals, while it was inversely correlated with VO ₂peak (p<0.05) in healthy males ⁶⁰ .

We were unable to find strong evidence that links PGC-1α and FNDC5 mRNA in muscle in response to exercise training or increased physical activity levels. Notably, we located only one study that examined the effects of exercise on UCP1 in WAT, and this found no effect ¹⁶ . Despite PGC-1α being firmly placed as a central regulator of adaptation to exercise in mice and humans, numerous aspects of the literature are contradictory or incomplete. For example, previous evidence indicates that PGC-1α mRNA accumulates with endurance training, while studies of PGC-1α protein reflect various antibodies that measure distinct molecular entities ranging from 70 to >110 kDa ^{76– 78} . Furthermore, mice lacking PGC-1α adapt normally to endurance exercise training, and in humans the PGC-1α regulated gene network does not correlate with aerobic adaptation ⁶⁸ . Thus any argument that places Irisin as part of the core PGC-1α regulated exercise adaptation program needs to reflect, on both technical and theoretical grounds, that there is great uncertainty of the nature and importance of PGC-1α in exercise and health ⁷⁹ .

When PGC-1α protein content is measured (albeit with uncertainty over protein identities) exercise training increases PGC-1α protein in skeletal muscle or causes nuclear translocation of protein ^{80– 83} . However, the studies included in the current review only relied on measuring PGC-1α mRNA to determine the effects of exercise on PGC-1α, and the time-course of mRNA and protein responses to exercise are distinct. Thus, the link between PGC-1α and FNDC5 in skeletal muscle may reflect measurement of mRNA dynamics and this may explain inconsistent findings for PGC-1α. Also, the proposed mechanism by Bostrom et al. (2012) indicates that induction of PGC-1α mRNA and then protein would activate the transcription of FNDC5, and hence, if this theory was correct, it would be expected that a strong correlation between PGC-1α mRNA and FNDC5 mRNA would exist. However, previous evidence showed that FNDC5 mRNA in muscle is not regularly increased by exercise or differently regulated between those with and without insulin resistance ¹⁹ , and was only modestly increased in a subset of older people following chronic exercise training ¹⁹ . If we focus on more reliable mRNA measures of PGC-1α and FDNC5, then the variable findings may be explained by the different characteristics of the populations examined and the different exercise protocols used.

An interesting aspect brought forward in the included studies showed that the start codon of the FNDC5 gene displays a variation in humans due to the non-ATG start codon ⁶⁴ . In humans, ATG is usually the first codon to lead to efficient protein production, and therefore, the latter may suggest that Irisin, if produced, would be done so in an inefficient manner ⁶⁴ . However, this notion has been questioned by a subsequent study, which supports that human Irisin is mainly translated from its non-ATG start codon, while the molecular weight of the protein is similar to that of important proteins in human body, such as insulin, leptin and resistin ⁷³ , indicating a biological role of Irisin.

The various commercially available antibodies used in the ELISA kits of the studies included in the current systematic review, yield a protein concentration that appears to be ~5–278 times greater than a more recent mass spectrometry data (data that may require independent validation), and still far above what others have found ⁸⁴ . Furthermore, Kurdiova et al. (2014) reported poor agreement between ELISA kit RK-067-16 and EK-067-29 (Phoenix Pharmaceuticals) ¹⁷ . Similarly, no correlation was found between EK-067-52 and ELISA of Adipogen that were used in the same samples ²⁶ . Finally, Montes-Nieto et al. (2016) analysed human Irisin using two different lots (604824 and 605835) of the ELISA kit EK-067-29 (Phoenix Pharmaceuticals) and also found a poor agreement (r=0.226) between them ⁸⁵ . These technical considerations may explain part or all of the equivocal results of the included studies in this current review regarding circulating Irisin.

According to the results of the current systematic review, two studies have measured circulating Irisin via mass spectrometry in response to exercise in humans. In the study by Jedrychowski et al. (2015), blood samples for Irisin identification were collected only after the exercise program from a small number of participants who were sedentary (n=4) or aerobic exercisers (n=6) ⁷³ . In the study by Lee et al. (2014), Irisin was measured only pre and post-acute exercise without a control situation, and the sample size was only two participants ²⁴ . Also, in the latter study a ~3-fold increase of Irisin was reported only after submaximal and not maximal exercise. These studies display methodological limitations and a small number of participants, which indicates that future longitudinal studies of changes in Irisin will clarify if the mass spectrometry measures reflect exercise-induced changes. Furthermore, Bostrom et al. (2012) and Lee et al. (2014) used an antibody that is discontinued for Irisin identification, given that it recognises a peptide of FNDC5 that is not part of the sequence of the secreted Irisin as this identified by mass spectrometry ⁷³ , while Jedrychowski et al. (2015) used an antibody by Adipogen. This may explain the discrepancy in the molecular weight of Irisin between those analysed by Bostrom et al. (2012) and Lee et al. (2014) (~22 kDa) and those analysed by Jedrychowski et al. (2015) (~12 kDa). While the studies that utilised mass spectrometry do not agree ^{24, 73} , reflecting issues of sensitivity and methodology, the latest identification and analysis of Irisin ^{24, 73} indicates that Irisin may circulate in blood and probably has a similar or identical structure to the mouse structure; however, whether it has genuine biological activity remains to be elucidated.

Based on the studies selected for the purposes of the current review, we cannot reach precise conclusions regarding the effects of acute and chronic exercise on PGC-1α in conjunction with FNDC5 mRNA in muscle; this is mainly due to the inconsistency of the findings and the different population characteristics examined. Most of the RCTs ^{33, 44, 49– 52} display high risk of bias, due to inadequate generation of a randomised sequence and a lack of concealment of allocations before assignment, while all the RCTs exhibit high risk of bias since the exercise interventions could not be blinded to the participants. In addition, four RCTs ^{44, 49, 50, 52} display unclear risk of bias because of the lack of information regarding the blinding procedures. Therefore, the risk of bias assessment of the included RCTs indicates that they may provide imprecise results ( Table 2). In addition, the CTs and SGS display a high risk of bias due to the absence of generation of a randomised sequence, inadequate concealment of allocations before assignment and knowledge of the allocated interventions by the outcome assessors. They also display unclear risk of bias due to knowledge of the allocated interventions by the investigators during the study ( Table 2). Finally, the included CSS display high risk of bias due to inadequate generation of a randomised sequence, lack of concealment of allocations before assignment and knowledge of the allocated interventions by the assessors, while they display unclear risk of bias for “incomplete outcome data” because of the lack of information of the participants who were excluded from the analysis. This evidence indicates that the CTs, SGS and CSS may also provide imprecise results. Furthermore, quality of reporting, as expressed through the adherence guidelines (i.e. CONSORT and STROBE), showed low scores of the required results that should have been reported (54.4% for RCTs, 58.68% for CTs, 58.44% for SGS and 60.8% for CSS) by the included studies in the current review. This shows inadequate reporting of the results of the included studies that may not aid the critical appraisal and interpretation of their outcomes.

To the best of our knowledge, this is the first systematic review that examines the effects of physical activity on the link between PGC-1α and FNDC5 in muscle, circulating Irisin and on UCP1 of WAT in humans. We compared our results with a recent meta-analysis that aimed to identify the effects of exercise on circulating Irisin ⁸⁶ . This meta-analysis concluded that chronic exercise may decrease circulating Irisin in the RCTs while the non-RCTs cannot form any conclusion. However, the latter meta-analysis did not take into consideration the issues raised regarding the validity of the methods used for Irisin identification ²⁶ . In contrast, while we considered the methods used for Irisin identification in the studies included in the current review, our review had a different aim, to systematically identify the effects of physical activity on the link between PGC-1α and FNDC5 in muscle, circulating Irisin and find evidence for regulation of UCP1 in WAT in humans. Regarding circulating Irisin, we also report that we cannot form any firm conclusion of the effects of exercise on circulating Irisin. Our review highlights previous evidence showing that circulating Irisin may only be detected in humans via mass spectrometry ^{25, 26, 73} , while we suggest that the previous available data coming from methods that have not been previously validated for circulating Irisin identification should not be used. This is because recent evidence questioned the antibodies used in the commercial ELISA kits given the polyclonal nature of these antibodies that may attract cross-reacting proteins ²⁶ . However, publications that use commercial ELISA that have not been previously validated to detect human Irisin continue at an alarming rate. Therefore, our review indicates to consider using only valid methods for human circulating Irisin identification in the future. Furthermore, our results are in accordance with a previous review that showed equivocal results among studies examining circulating Irisin due to the methodological variations for Irisin detection ⁸⁷ . In this critical review, the authors examined the commercial antibodies and ELISA used to measure circulating Irisin and concluded that the currently available antibodies should be tested for cross-reacting antigens detection ⁸⁷ . Additionally, another recent review showed that the previous measurements for circulating Irisin identification differs greatly, given that they displayed a molecular weight of the protein between 0.01 ng/ml and 2000 ng/ml ⁸⁸ . The latter critical review concluded that it is necessary to establish accurate methods for irisin measurements. Our systematic review analysis, agrees with the latter conclusion given that the Irisin measurements in the included studies via commercial ELISA kits, displayed a molecular weight of the protein ranging between 22 ng/ml to 1196 ng/ml.

Initially, Irisin was proposed to have a therapeutic effect given the potential to cause a browning formation of WAT that may have anti-obesity and antidiabetic effects ² . This was mainly suggested when Irisin administered in obese mice improved glucose homeostasis and caused weight loss ² . Also, the browning formation that Irisin may cause could lead to reduced weight gain, up-regulated insulin sensitivity, reduced risk of diabetes type II and other metabolic disorders as animal studies indicate ^{89– 93} , as well as increase daily resting energy expenditure in humans ^{94, 95} . The available evidence from the included studies in the current review revealed that the available commercial ELISA kits for Irisin identification either were found to be invalid ^{26, 87} or they should be tested for validity ⁸⁷ . Thus, we cannot confirm a favourable effect of Irisin on human metabolism. Finally, none of the included studies in the current review examined associations of circulating Irisin with indices indicate a therapeutic role of the protein using western blotting and/or mass spectrometry methods.

The current review has a number of strengths. For instance, we used the PubMed and the EMBASE databases using appropriate algorithms with standardized indexing terms. Standardized indexing terms can retrieve records that may use different words to describe the same concept and information beyond that may be contained in the words of the title and abstract ⁹⁶ . Furthermore, the current review used a systematic manner to identify articles according to previous methodology ^{27– 29} , and we used well-established tools ^{30– 32} to evaluate the included studies. To reduce bias, two investigators worked independently on the screening of the included studies for eligibility, risk of bias assessment, and in the provision of CONSORT and STROBE scores. Also, we have not excluded studies based on language. However, a limitation of the current review includes the use of only published literature; we did not include grey literature searching. In this light, there is a potential of publication bias in the current review. Nevertheless, the inclusion of grey literature may itself introduce bias and one reason to include grey literature would be the absence of peer-review sources ⁹⁶ .