Search strategy and study selection

A systematic literature search for relevant studies published in English was conducted in the databases of PubMed, the Cochrane Central Register of Controlled Trials, and Web of Science from their inceptions to January 12nd, 2018. In addition, the reference lists of relevant articles, reviews, and meta-analyses were manually checked for other suitable studies. The words or terms used for searching were linked with “endothelial function”, “diabetes”, and “exercise training” (see Additional file 1: Table S1).

Studies eligible for inclusion in this meta-analysis required to fulfill the following criteria: (i) participants were diagnosed with type 2 diabetes; (ii) the intervention groups received land-based normoxic exercise training programs with durations ≥ 8 weeks, a time-window which is commonly employed to assess the chronic exercise training effects on health outcomes [28]; (iii) the controls received no exercise (or usual care), exercise training programs different from the intervention groups, or non-exercise intervention programs comparable to the intervention groups; (iv) outcomes on endothelial function assessed by FMD had to be reported pre- and post-intervention; and (v) allocation to the intervention or control group should be random. Studies were also considered eligible if they compared the effects of exercise training with durations ≥ 8 weeks on endothelial function assessed by FMD in patients with type 2 diabetes versus non-diabetes controls. Studies were excluded if they were posters or protocols, had a lack of proper controls, or did not report outcomes on FMD.

Two authors (S.Q. and X.C.) reviewed the titles, abstracts, and/or full-texts for each of the articles identified by the literature search after removal of duplicates, aiming to determine the eligibility for this meta-analysis. During the study selection process, discrepancies were resolved by discussion with a third author (U.S. or Z.S).

Data extraction and quality assessment

A standardized data abstraction form was applied to collect the following information from eligible studies: author information, participant characteristics [including sex distribution (proportion of males), duration of diabetes, and the means of age, body mass index (BMI), blood pressure, glycemic control, and cardiorespiratory fitness at baseline], intervention details (including type of exercise, intensity, time for one bout of exercise, frequency, and intervention duration), and study outcomes (including FMD at baseline and post-intervention). If the outcomes of interest were incomplete or could not be imputed, corresponding authors of the original studies were contacted via emails. Moreover, in this meta-analysis FMD was calculated as the percent change in diameters following reactive hyperemia compared with the baseline diameters at rest.

The Cochrane Collaboration “Risk of Bias” tool [29], which includes items on selection bias, performance bias, detection bias, attrition bias, and reporting bias, was applied to evaluate the quality of eligible studies. All the data collection and quality assessment were initially done by one author (S.Q.), and later checked by another author (X.C.). Discrepancies, if occurred, were resolved by referring back to the original studies.

Data synthesis and statistical analysis

For studies reporting standard errors, 95% confidence intervals (CIs), or interquartile ranges, the standard deviations were obtained using the methods described in Cochrane Handbook for Systematic Reviews [29] or reported previously [30]. For studies including two different exercise training interventions, the control group was split into two groups with smaller sample sizes, aiming to provide reasonably independent comparisons and to overcome the unit-of-analysis error [29]. Post-intervention FMD values were primarily chosen for analysis in general, but only the change scores from baseline were selected for assessing the impact of the existence of type 2 diabetes on endothelial function in response to exercise training, which is because the baseline FMD results were not comparable between type 2 diabetes patients and non-diabetes controls.

The weighted mean differences (WMDs) with 95% CIs were calculated using a random-effects model, which seems to better account for between-study heterogeneity and could provide more conservative results than a fixed-effects model [29]. The heterogeneity was evaluated using the I ² statistic, with the value > 50% indicative of significant heterogeneity [29]. Subgroup analysis was conducted to investigate the impact of exercise types on endothelial function, and meta-regression analyses were undertaken to assess the influence of patient and intervention characteristics in moderating changes in endothelial function. Sensitivity analyses were performed to assess the robustness of the findings by restricting the analyses to studies using exercise training as the sole intervention, reporting no or only minor changes in medication use during the intervention periods, or employing the intention-to-treat analysis. Publication bias was evaluated using the Begg’s and Egger’s tests, with the P < 0.10 indicative of significance [29]. All the analyses were conducted using STATA software (Version 12.0, StataCorp LP, College Station, Texas). A 2-sided P < 0.05 was considered statistically significant unless otherwise indicated.

Literature search and study characteristics

The literature search result and study selection process are shown in Fig. 1. Of the 3136 unique articles identified, 77 were searched for full-text assessment after screening of titles and/or abstracts, with 12 considered eligible for inclusion [16–24, 31–33]. Since two articles had two different exercise training protocols with a non-exercise control group, providing three independent comparisons for each [17, 18], a total of 16 studies (databases) were finally included.

Of the 16 studies included, the durations of exercise interventions ranged from 8 to 26 weeks, with the majority being 12 weeks; and the frequency of exercise varied from 3 to 5 times/week. The intensity for aerobic exercise was moderate to vigorous in general based on the position statement on physical activity and exercise intensity by Norton and colleagues [34]. The time for one session of aerobic exercise ranged from 20 to 60 min. For resistance exercise, it was composed of 2–3 sets ranging from 10 to 15 repetitions; however, its intensity was generally not well defined except three reported to be 50–70% of one repetition maximum. There was one study adding a moderate energy-restricted dietary program [19] and another one adding standard care [22] to both the intervention and control groups.

The approaches for assessing FMD were well described in each individual study except the one by Hollekim-Strand and colleagues [24]. All participants were required to be fasted for measuring FMD. The cuff was frequently placed at the upper-arm in FMD measurement procedures, with inflated cuff pressure ranging from 30 to 50 mmHg above the systolic blood pressure or from 200 to 250 mmHg to occlude the brachial artery for 4.5–5 min (Table 1). The images of post-deflation diameter were continuously recorded within a time-window that logged from the last 30 s of occlusion through 180 s of hyperemia. Of the included studies, four reported the approaches for randomization, and three utilized per-protocol analyses (see Additional file 1: Table S2). The dropout rates of included studies were low in general except two around 20%.

Exercise training and endothelial function in type 2 diabetes

Compared with non-exercise controls

Ten RCTs from eight studies enrolling 377 patients with type 2 diabetes were included [16–23]. Meta-analysis showed that exercise training led to an overall improvement in FMD by 1.77% (95% CI 0.94–2.59%), with no evidence of heterogeneity (I ²  = 35.3%; Fig. 2a). When taking exercise types into consideration, results showed that the FMD was increased by 1.21% (95% CI 0.23–2.19%, I ²  = 13.8%) for aerobic exercise (five RCTs with 168 patients), 2.49% (95% CI 1.17–3.81%, I ²  = 35.3%) for combined aerobic and resistance exercise (four RCTs with 190 patients), and 1.60% (95% CI − 0.52 to 3.72%) for resistance exercise (one RCT with 19 patients).

Univariate meta-regression analyses suggested that changes in FMD related to exercise training or specifically aerobic or combined exercise was not significantly associated with any of the baseline values or changes scores of BMI, blood pressure, glycemic control, or peak oxygen consumption (all P > 0.05; Table 2). Sensitivity analyses indicated that the improvements in FMD associated with exercise training or specifically aerobic or combined exercise still remained significant and were not significantly altered after excluding studies having non-exercise interventions, reporting substantial changes in medication use, or utilizing the per-protocol analyses (Table 3). No evidence of publication bias was observed for FMD results associated with exercise training or specifically aerobic or combined exercise (All P > 0.80 for both Begg’s and Egger’s tests).

Variables	All exercise			Aerobic exercise			Combined exercise
	N	β	P	N	β	P	N	β	P
Age (log-transformed)a	10	− 0.53	0.72	5	20.1	0.17	4	− 0.06	0.97
Disease durationa	6	− 0.03	0.82	3	− 0.08	0.69	2	NA	NA
Training duration	10	− 0.10	0.26	5	NA	NA	4	− 0.13	0.17
Proportion of malesa	10	0.01	0.50	5	− 0.04	0.20	4	0.06	0.28
Baseline dataa
FMD at baseline	9	0.43	0.23	5	1.33	0.29	3	1.16	0.48
Body mass index	10	− 0.16	0.34	5	− 0.30	0.19	4	− 0.32	0.34
Systolic blood pressure	6	0.03	0.70	4	− 0.10	0.34	2	NA	NA
Diastolic blood pressure	6	0.01	0.95	4	0.41	0.30	2	NA	NA
Mean arterial pressure	7	0.23	0.05	4	− 0.19	0.76	3	0.20	0.28
Fasting blood glucose	6	0.17	0.73	4	− 0.71	0.29	2	NA	NA
Hemoglobin A1c	10	1.11	0.09	5	− 1.97	0.44	4	1.21	0.17
Peak oxygen consumption	8	− 0.01	0.95	3	− 3.27	0.61	4	− 0.03	0.95
Change scores post-exercise interventionb
Body mass index	6	− 2.10	0.55	3	− 4.64	0.57	3	3.60	0.59
Systolic blood pressure	6	− 0.16	0.30	4	− 0.20	0.28	2	NA	NA
Diastolic blood pressure	6	− 0.10	0.55	4	− 0.22	0.35	2	NA	NA
Fasting blood glucose	5	1.44	0.62	4	− 4.66	0.34	1	NA	NA
Hemoglobin A1c	9	− 2.52	0.24	5	− 5.02	0.12	3	1.96	0.73
Peak oxygen consumption	6	− 0.38	0.74	3	1.14	0.55	3	− 1.40	0.52

FMD flow-mediated dilation, NA not applicable

^a They were the baseline mean values of intervention and control groups

^b They were the average change cores from baseline between intervention and control groups except the study by Maiorana and colleague [21]

Study characteristics	All exercise			Aerobic exercise			Combined exercise
	N	WMD (%)	95% CI	N	WMD (%)	95% CI	N	WMD (%)	95% CI
Exercise training as the sole intervention	8	2.00	1.23–2.77%	4	1.64	0.79–2.49%	3	2.47	0.70–4.23%
No or minor changes in medication use	8	2.16	1.48–2.85%	4	1.64	0.79–2.49%	3	3.20	2.10–4.30%
Using intention-to-treat analysis	8	2.39	1.64–3.15%	3	2.27	0.94–3.61%	4	2.49	1.17–3.81%

FMD flow-mediated dilation, WMD weighted mean difference, CI confidence interval

Exercise training and endothelial function in type 2 diabetics vs non-diabetics

Three controlled studies with 36 type 2 diabetics and 37 non-diabetics were included in this meta-analysis [31–33]. In these studies, increases in FMD were observed among type 2 diabetics (from 0.5 to 0.82%) and non-diabetics (from 1.46 to 2.1%) following 8–12 weeks of exercise training. Pooled results indicated that the magnitude of the improvements in FMD was smaller in type 2 diabetics than non-diabetics after exercise training (WMD − 0.72%, 95% CI − 1.36 to − 0.08%; I ²  < 1%; Fig. 2c) or specifically, after aerobic exercise (two studies, WMD − 0.65%, 95% CI − 1.31 to 0.01%; I ²  = 36.7%).

Summary of the main findings

Our meta-analysis revealed that exercise training, in particular aerobic and combined aerobic and resistance exercise, significantly improved endothelial function in patients with type 2 diabetes, as indicated by increased FMD; and this manner seemed to be independent of changes in traditional cardiometabolic markers including BMI, blood pressure, glycemic control, or cardiorespiratory fitness in relation to exercise training. However, our meta-analysis did not provide adequate evidence that high-intensity interval aerobic exercise was superior to moderate-intensity continuous aerobic exercise in improving endothelial function. Noteworthy, the increases in FMD in response to exercise training in type 2 diabetics were lower than that in non-diabetics, indicating that the presence of diabetes may weaken the exercise training effects on endothelial function.

Interpretations

Our study showed that exercise training, a non-pharmacological therapy, led to an overall improvement in FMD by 1.77%. This is of clinical importance for patients with type 2 diabetes, since on the one hand, every 1% increase in FMD was correlated with an estimated 13% risk reduction of cardiovascular events based on the report from van Sloten and colleagues [35]; and on the other hand, such an increase in FMD is even larger than or comparable to those from pharmacological interventions like statin therapy [36] or phosphodiesterase inhibitors use [37], which result in an improvement in FMD by 0.94% (95% CI 0.38–1.5%) and 2.19% (95% CI 0.48–3.90%), respectively. Although not fully understood, it is speculated that the observed beneficial effects of exercise training on endothelial function might be underlined by the following mechanisms. Firstly, exercise training causes an increase in blood flow, which augments shear stresses on the endothelium, leading to increased nitric oxide synthesis and bioavailability [1]. Secondly, exercise training reduces oxidative stress and the expression of pro-inflammatory molecules [38], both of which are considered initiating factors for endothelial dysfunction. Thirdly, exercise training may help to restore the function of endothelial progenitor cells, promoting endothelial repair and facilitating vascular angiogenesis subsequently [39].

Partly in agreement with our finding, the prior meta-analysis by Montero and colleagues as well as the review paper by Miele and colleagues had also observed the beneficial effects of exercise training on endothelial function assessed by FMD in patients with type 2 diabetes [13, 26]. However, our study, which included more RCTs and introduced subgroup and meta-regression analyses, extended their findings by providing in-depth analyses of the training effects from the specific exercise type on FMD and exploring the potential moderators. In our study despite a non-significant increase in FMD after resistance exercise, both aerobic and combined exercise were effective in improving FMD, which corresponds well with the results from the meta-analysis conducted in a heterogeneous adult population [40]. Yet inconsistent with our finding, Way and colleagues argued that aerobic exercise may not be able to improve FMD in patients with type 2 diabetes [41], which might be due to their limited number of studies included.

In addition, the largest increase in FMD observed in combined exercise suggests that this mixed form might be superior to aerobic or resistance exercise in improving endothelial function based on our subgroup analyses across exercise types with indirect comparisons. However, it is noteworthy that they may not control for energy expenditure or training duration in every section, and that there was only a single study with a small sample size that explored the influence of resistance exercise on endothelial function [17], which might affect the outcomes of interest (e.g., may underestimate the effects of resistance exercise). Future studies are in need to determine which exercise type might be the best one in increasing FMD using head-to-head designs with the energy expenditure- and/or training time-matched for each section across different exercise types.

In recent years Ramos and colleagues reported that high-intensity interval aerobic exercise, which acts in a time-saving manner [42, 43], produces a greater positive influence on endothelial function versus moderate-intensity continuous aerobic exercise in a mixed adult population [44]. However, our meta-analysis in patients with type 2 diabetes did not provide adequate evidence in support of this notion, which might be largely attributable to the differences in the target populations as well as the small number of studies included. Moreover, Ramos and colleagues pointed out that the improvement in endothelial function associated with high-intensity interval aerobic exercise over moderate-intensity continuous exercise might be owing to its superiority in increasing cardiorespiratory fitness, improving glycemic control, and lowering blood pressure, suggesting a potential positive linkage between endothelial function and cardiometabolic markers [44]. Yet our meta-regression analyses, based on the averages of participant characteristics for each study did not support such an assumption, which could be also evidenced by the findings from the individual study by Gibbs and colleagues [20] and the cross-sectional observation [45]. It seems likely that the benefits of exercise training on endothelial function are independent of improvements in cardiometabolic health among patients with type 2 diabetes, which, albeit, still requires further investigations using the individual participant data.

In addition to suggesting a positive influence of exercise training on endothelial function in patients with type 2 diabetes, our study showed that the improvement in endothelial function in response to chronic exercise training was weakened in patients with type 2 diabetes compared with those without. Although the exact mechanism is not well understood, it is speculated that the presence of diabetes may compromise the ability of the endothelium to endogenously increase vascular nitric oxide bioavailability following exercise training [31], possibly because of the persistent hyperglycemia and the increased levels of circulating advanced glycation end products or reactive oxygen species [46, 47]. Moreover, the blunted potential of exercise to stimulate and to mobilize endothelial progenitor cells observed in diabetic patients compared with non-diabetic controls might also contribute to the weakened improvement in endothelial function in response to exercise among type 2 diabetes [46]. These may indicate that more interventions apart from promoting physical exercise might be needed for patients with type 2 diabetes to obtain comparable health benefits like healthy or non-diabetic controls.

Limitations

Despite a comprehensive exploration of the exercise training effects on endothelial function among patients with type 2 diabetes, this meta-analysis has some limitations. Firstly, some studies reported changes in medication use in the intervention period or used co-interventions like dietary programs [48], which may influence the endothelial function and contributed to heterogeneity. However, our sensitivity analyses indicated that they are unlikely to have important impacts on the main findings. Secondly, most of the included studies implemented exercise interventions within 12 weeks, which largely represents a relatively short-term effect on endothelial function, although our meta-regression analysis suggested that intervention periods were not likely to affect the changes in endothelial function associated with exercise training. More studies are therefore warranted to explore the long-term (e.g., over 6 months) effects of exercise training on endothelial function, just as suggested by Lenasi and colleagues [46]. Thirdly, some variances existed in the measurement methods for FMD. However, these methods were well defined in general. Although it is reported that the measurement of peak dilation at 60 s after cuff release may result in an underestimation of true FMD by up to 40% [49], this factor seems to have had a minor influence on our outcomes on FMD since all studies except the one by Hollekim-Strand and colleagues [24] had already adopted a continuous measurement of post-deflation diameter within the time-window as aforementioned. Finally, our study did not search for unpublished studies and had a language restriction, which may cause some selection bias.