The electronic search returned 2513 potentially relevant studies. An additional 121 studies were identified from the end reference lists of the included studies and conference proceedings. Following the removal of duplicates and elimination of ineligible studies, eight studies were retained for review (Fig. 1). All eight studies reported on the effect of dietary interventions (i.e. no studies were retrieved meeting the eligibility criteria that reported on physical activity/exercise alone). These studies were seven RCTs [29,30,31,32,33,34,35] and one single-group study . Four of the studies were conducted in Europe (Denmark, Spain, Italy and the UK) and four in Asia (India, Malaysia, Japan and the Republic of Korea). All eight studies used dietary modulation alone (Table 1).
Characteristics of participants with type 2 diabetes
The total sample size across the included studies was 395, including at least 225 men and 135 women; one study failed to report the sex of participants (n = 35) . Participants were aged 33–77 years (58 ± 4 years [mean ± SD]) with a BMI range of 25–45 kg/m2. Six studies recruited only individuals diagnosed with type 2 diabetes mellitus [29, 30, 32,33,34,35]. One study compared participants with type 2 diabetes mellitus and control participants at baseline, prior to the introduction of any intervention, however, these 13 participants were not included in the analyses reported here . Two studies [31, 36] reported on a sample including individuals with and without type 2 diabetes mellitus (Table 1). Only one study reported the ethnicity of participants and none of the included studies reported the time since diagnosis of type 2 diabetes mellitus. Five studies reported treatment regimens: insulin injection (n = 11), insulin secretagogue (n = 34), metformin (n = 52), sulfonylurea (n = 12), α-glucosidase inhibitor (n = 41), sitagliptin (n = 7), thiazolidinedione (n = 1), peroxisome proliferator-activated receptor γ (PPAR-γ) antagonist (n = 15) or statins (n = 5) [30, 31, 34]. One study recruited individuals who managed their type 2 diabetes mellitus through diet only , and two studies used eligibility criteria that restricted the use of insulin but allowed hypoglycaemic drugs and dietary interventions, but provided no specific details [32, 33]. Two studies [35, 36] did not report any treatment regimens (Table 1).
The duration of interventions ranged from 21 days to 6 months. All included studies involved some form of dietary intervention, including dietary manipulation and/or supplements. One RCT supplemented participants with a synbiotic , one single-group study instructed participants to follow a strict vegetarian diet , one RCT  instructed participants to follow a strict Ma-Pi diet  and a further RCT asked participants to follow a diet recommended for type 2 diabetes mellitus incorporating an increased intake of sardines . The last two of these studies included information on nutrient intake, although Balfegó et al  did not provide any information on what constituted the type 2 diabetic diet. Two RCTs provided participants with probiotics [31, 33], but only one recorded nutrient intake and provided guidelines on dietary intake based on type 2 diabetes mellitus guidelines . One RCT provided a prebiotic, detailed in Vulevic et al , with no dietary advice, but recorded nutrient intake ; and one RCT provided participants with a digestive supplement and recorded nutrient intake at baseline .
All eight studies reported on glucose control, including HbA1c, fasting blood glucose, 2 hfs OGTT, HOMA-IR or insulin, or a combination of these, and the gut microbiota (using various techniques; ESM Table 1).
Methodological quality assessment
All included studies were assessed for methodological quality using the Cochrane Collaboration risk of bias tool (Table 2) . Seven studies explicitly reported their hypotheses, objectives, statistical testing procedures and main findings. Four studies reported a power calculation including details of whether sample size was obtained [30, 31, 33, 34]. Only one study retained the sample size at follow-up . One study provided a power calculation based on changes in TNF-α upon request . However, it is unclear from the information reported in the article whether this was the primary outcome, making it difficult to establish whether the sample size was adequate. All but one study  reported attrition rates, and one study reported using an intention-to-treat analysis .
Four studies provided enough information to confirm the use of adequate sequence generation [29, 30, 33, 34], two studies provided sufficient detail on the methods used to conceal allocation sequences [30, 33] and three studies provided explicit detail on blinding of the research team [30, 31, 33]. Seven studies provided detail on outcome assessors [29,30,31,32,33,34, 36] and five studies provided sufficient detail to establish the likely absence of selective outcome reporting [29, 32,33,34,35]. Overall, seven of the studies were considered to be low risk of bias [29,30,31,32,33,34, 36] and the risk of bias of one  was unclear (Table 2).
Effects of lifestyle modulation on gut microbiota
Additional data on the gut microbiota were requested from all authors of the included studies. Two studies comparing supplementation with a prebiotic vs control for 12 weeks  and the Ma-Pi diet vs a control diet for 21 days  provided sufficient data for meta-analyses to be conducted. Changes in the relative abundance of bacteria in the genera Bifidobacterium (SMD 1.29% [95% CI −4.45, 7.03], I2 = 33%), Roseburia (SMD −0.85% [95% CI −2.91, 1.21], I2 = 79%) and Lactobacillus (SMD 0.04% [95% CI −0.01, 0.09], I2 = 0%) (ESM Fig. 1) were reported, however, these changes were not significantly different when comparing dietary interventions with controls.
Candela et al  compared a fibre-rich macrobiotic diet with a control diet. The authors reported a significant change in weighted UniFrac following receipt of the intervention. Levels of Faecalibacterium were significantly negatively correlated with fasting blood glucose; Akkermansia and Bacteroides both showed a positive significant relationship with LDL-cholesterol; and Ruminococcus was significantly positively correlated with fasting blood glucose. Significant increases in the relative abundance of Peptostreptococcaceae and Leuconostocaceae were also reported, and both of these genera were positively correlated with dietary components (fermented products).
Balfegó et al  compared a type 2 diabetes diet, one enriched with 100 g of sardines and one without. The authors reported a decrease in Firmicutes and an increase in Escherichia coli in both groups between baseline and study completion. In the sardine enriched group there was also a decrease in the Firmicutes:Bacteroidetes ratio and an increase in Bacteroides-Prevotella when compared with baseline. Kim et al  reported a significant increase in the relative abundance of Bacteroidetes and a correlation between weighted and unweighted UniFrac and a strict vegetarian diet. However, these analyses involved the whole sample of participants, including those without a confirmed diagnosis of type 2 diabetes. Two RCTs [31, 33] supplemented participants with probiotics and reported significant changes in bacterial composition: Andreasen et al  reported a significant increase in the presence of Lactobacillus acidophilus from near non-detectable levels to 6.4 colony-forming units; and similarly, Firouzi et al  reported significant increases of 4.5- and twofold in Bifidobacterium and Lactobacillus spp., respectively.
Pedersen et al  conducted an RCT in which participants were supplemented with a prebiotic compared with placebo. No between-group differences were shown; however, an increase in α diversity was reported within the prebiotic group. Furthermore, correlations between bowel permeability, metabolic profile, inflammatory markers and bacteria were reported (ESM Table 2). Sheth et al  supplemented participants with a synbiotic (two species of Lactobacillus and Bifidobacterium each, one species of Streptococcus and yeast, and 300 mg oligosaccharide), although the dietary intake provided alongside the supplements was unclear. Increases in both Lactobacillus and Bifidobacterium were reported following the intervention. Sasaki et al  reported significant changes in the Firmicutes:Bacteroidetes ratio between baseline and 12 weeks following supplementation with 300 and 900 mg/day of transglucosidase.
Effects of dietary intervention modulation on glucose control
Four studies comparing dietary interventions including prebiotics, probiotics and Ma-Pi diet vs controlled diets for a duration of between 21 to 84 days reported or provided sufficient data to enable meta-analyses of fasting blood glucose, HbA1c, fasting insulin and HOMA-IR (ESM Fig. 2) [30, 32,33,34]. Reductions were shown in all glucose control variables. HbA1c was significantly reduced (standardised mean difference [SMD] −2.31 mmol/mol [95% CI −2.76, −1.85] [0.21%; 95% CI −0.26, −0.16]; I2 = 0%, p < 0.01), however, fasting blood glucose (SMD −0.25 mmol/l [95% CI −0.85, 0.35], I2 = 87%, p > 0.05), fasting insulin (SMD −1.82 pmol/l [95% CI −7.23, 3.60], I2 = 54%, p > 0.05) and HOMA-IR (SMD −0.15 [95% CI −0.63, 0.32], I2 = 69%, p > 0.05) were not significantly reduced when comparing dietary interventions with comparator groups.
All eight studies reported on glucose control; however, only four provided sufficient data to calculate overall effect sizes [30, 32,33,34] (ESM Fig. 2). Kim et al  and Sasaki et al  reported positive changes in fasting blood glucose, 2 hfs OGTT, fasting insulin and HbA1c, although these were not statistically significant. Andreasen et al  reported baseline fasting blood glucose, HbA1c, 2 hfs OGTT and HOMA-IR, but did not report data for these variables at the 4 week follow-up point. A significant between-group improvement in insulin sensitivity was reported, but post-hoc t tests revealed no significant within-group changes. Unfortunately, analyses were not reported for participants with type 2 diabetes mellitus only, making it difficult to extrapolate the effects of the intervention.
Two RCTs showed significant improvements in glucose control [29, 32]. Candela et al  reported reductions in fasting blood glucose (−2.3 vs −1.9 mmol/l), postprandial blood glucose (−4.0 vs −4.3 mmol/l), HbA1c (−5.5 vs −2.2 mmol/mol [−0.5% vs −0.2%] and HOMA-IR (−1.9 vs −1.5). Balfegó et al  reported significant reductions in fasting insulin (−35% vs −23%) and HOMA-IR (−49% vs −22%) for treatment and comparator groups, respectively. Three RCTs supplemented participants with a prebiotic, probiotic or synbiotic. Firouzi et al  reported significant reductions in insulin at 6 and 12 weeks (−16 and −20 pmol/l, respectively) and HbA1c at 12 weeks (−1.1 mmol/mol [−0.1%]) between the probiotic and control group. Sheth et al  reported reductions in fasting blood glucose, postprandial blood glucose and HbA1c; however, it is unclear from the reported findings whether differences between groups were assessed, as data were not provided. Pedersen et al  reported no statistical improvements in glucose control following supplementation with prebiotics.
Effects of dietary intervention modulation on inflammation
Four studies reported no significant differences in the inflammatory markers TNF-α, IL-6, IL-1RA or CRP between groups [29, 31, 33, 34]. Kim et al  reported significant reductions in faecal lipocalin 2; however, these analyses included participants without type 2 diabetes mellitus, making it difficult to extrapolate results for the subgroup of participants with type 2 diabetes. One RCT reported significant reductions in TNF-α, IL-6 and CRP following 21 days consuming the Ma-Pi 2 diet .
Effects of dietary intervention on short-chain fatty acids
Kim et al  reported a significant increase in butyrate and a reduction in total short-chain fatty acids (SCFAs), acetate, propionate and butyrate concentrations. However, these analyses were conducted on all individuals recruited, including those without a confirmed diagnosis of type 2 diabetes. Sheth et al  also reported increased concentrations in butyrate and propionate, although it is unclear what statistical analyses were conducted.
Effects of dietary intervention on anthropometrics
Seven studies reported BMI at baseline [29,30,31,32,33,34, 36] and six of these studies reported BMI postintervention [29, 30, 32,33,34, 36]. Five studies reported no significant changes in anthropometrics, including weight, BMI and hip and waist circumferences [29, 30, 32,33,34]. Kim et al  provided BMI data upon request showing a postintervention reduction; however, it is unclear from the published article or the additional information provided whether this was statistically significant.
Changes in nutrition
Four studies recorded nutritional intake at baseline and the postintervention follow-up, including energy intake and the consumption of fat, protein, carbohydrates and fibre. Although Candela et al  reported dietary intake, no statistical analyses were reported. Firouzi et al  reported a statistically significant 9% reduction in fat intake and Balfegó et al  reported an 11% reduction in energy intake. Pedersen et al  reported a statistically significant 1.1% increase in protein intake in the control group compared with the prebiotic group.