Dapagliflozin effect on endothelial dysfunction in diabetic patients with atherosclerotic disease: a randomized active-controlled trial

The study was designed to determine whether there is a difference in endothelial function between treatment with dapagliflozin and glibenclamide. The two primary endpoints were (i) the change in rest FMD at 1 min measured at randomization and after 12 weeks of treatment and ii) the change in post-I/R FMD at 1 min measured at randomization and after 12 weeks of treatment. The detailed methods in which FMD was obtained in each situation are described in the Additional file 1. Prespecified secondary parameters included the area under the curve (AUC) of the diameter vs. time scatterplot and the derived parameters of blood flow velocity, arterial diameter and shear rate, plasma viscosity, and vascular and inflammatory marker activity levels.

Sample size calculations have been published elsewhere [18]. Briefly, the two study endpoints equally split an alpha level at the bilateral α = 0.025. If one of them proved significant, the alpha value could then be recycled to test the secondary endpoints based on the index primary endpoint at the bilateral α = 0.05. Baseline continuous and categorical data were compared by Student’s t or Wilcoxon-Mann–Whitney U tests and a two-tailed Fisher’s exact test, respectively. Data are presented as the mean ± standard deviation (SD) for normally distributed data and medians and interquartile ranges (IQRs) for non-normally distributed data. The analyses were conducted according to the intention-to-treat principle. Changes from baseline were compared between treatments for each primary endpoint by analysis of covariance (ANCOVA) adjusted for baseline values. Prerequisites for the ANCOVA models (linearity, distribution normality and equal variance) were checked using histograms, normal probability graphs and residual dispersion. Exploratory intragroup comparisons of pretreatment vs. posttreatment values were performed by a related-samples Wilcoxon signed rank test or Student’s t-test paired for variables with non-normal or normal distribution, respectively. Analyses were performed using SPSS v.22 for Macintosh (IBM Corp., Armonk, NY, USA).

The CONSORT flow diagram describing all enrollment processes is shown in the Additional file 1: Fig. S2. All patients were above the 75th percentile for patient age, sex, and race based on their distribution in the Brazilian population. Eight patients also had been previously diagnosed with CVD and were evenly distributed between the two study arms.

Before randomization, patients remained in the run-in phase to adjust medical treatments for a median of 68 (62) days for those who were lately allocated into the dapagliflozin arm and 89 (91) days for those who were allocated into the glibenclamide arm (p = 0.13). At the end of the run-in, the final metformin dose was similar for patients in the dapagliflozin arm (median 1700 mg/day [50]) and glibenclamide arm (median 1750 mg/day [50]; p = 0.75). The mean final dose of losartan was the same for both groups (median 100 mg/day [50]). To ensure stable clinical control, after reaching the prespecified therapeutic targets, patients remained without new changes in their therapeutic regimens for at least 45 days. After this period, only those patients who remained in the therapeutic targets were randomized.

As one patient requested withdrawal before concluding the study assessments, the analyses were performed on data from 49 patients in the glibenclamide arm and 48 patients in the dapagliflozin arm (Table 1). There were no significant differences in baseline characteristics between the study arms. No patient was an active smoker, and those who were previously smokers (~ 40%) stopped smoking 19 ± 9 years ago.

There was no difference between groups in terms of HbA1c either at randomization or at week 12. Median HbA1c reductions of − 0.8 (0.7)% and − 0.7 (0.95)% for the dapagliflozin and glibenclamide groups, respectively. Similar results were obtained for the plasma glucose levels (median reductions of [− 31(46) mg/dL] and [− 34(54) mg/dL] for the glibenclamide and dapagliflozin groups, respectively).

At 12 weeks, the systolic BP was lower (130 ± 17 vs. 137 ± 16 mm Hg; p = 0·035), and the diastolic BP tended to be lower (78 ± 10 vs. 81 ± 9 mm Hg; p = 0·054) in the dapagliflozin group than in the glibenclamide arm. In contrast, the mean systolic BP during the 24-h ABPM was similar in both arms at randomization and at the end of the trial (126 ± 13 vs. 126 ± 12 mmHg; p = 0.98 for dapagliflozin and glibenclamide arms, respectively). Similarly, the mean diastolic BP obtained during 24-h ABPM was equivalent in both arms at the beginning and the end of treatments (77 ± 8 vs. 77 ± 8 mmHg; p = 0.78 for the dapagliflozin and glibenclamide arms, respectively). The average daytime systolic BP (baseline: 126 ± 10 vs. 128 ± 10 mmHg; 12-week: 127 ± 13 vs. 127 ± 12 mmHg; p > 0.05 for both) and diastolic BP (baseline: 77 ± 9 vs. 77 ± 8 mmHg; 12-week: 77 ± 9 vs. 77 ± 8 mmHg; p > 0.05 for both) were also equivalent in both the dapagliflozin and glibenclamide arms.

Whole-blood viscosity increased [0.2 (0.43) cPs; p = 0.001] after dapagliflozin treatment but not after glibenclamide treatment [0.007 (0.43) cPs; p = 0.624]. Patients experienced a weight loss of 1.7 (2.3) kg with dapagliflozin and a weight gain of 1.1 (2.5) kg with glibenclamide (p < 0.001).

The posttreatment change in rest FMD at 1 min (the first coprimary endpoint) was significantly different between dapagliflozin and glibenclamide (Fig. 1c and d; Table 2). For the dapagliflozin group, there was a median 3.3% increase between baseline and 12 weeks, whereas a median 1.2% decrease was measured for the glibenclamide arm. The difference between the study arms in the primary endpoint remained significant when adjusted for office SBP change (p < 0.001), office DBP change (p < 0.001), or pulse pressure change (p < 0.001). No significant interaction between changes in FMD and changes in HbA1c, SBP, DBP, or body weight was found. There was an increase and a decrease in the AUC for arterial diameter continuously measured during the first 5 min after cuff deflation for the dapagliflozin and glibenclamide arms, respectively (p = 0.001; Table 2).

For the intragroup comparison of pretreatment and posttreatment rest FMD, there were significant differences during hyperemia (p < 0.0001) and at 1 min (p < 0.0001), 2 min (p = 0.001), 3 min (p = 0.007), 4 min (p = 0.015), and 5 min (p = 0.004) in the dapagliflozin arm (Fig. 1c). The same analysis identified no significant differences in the glibenclamide arm (hyperemia: p = 0.363; 1-min: p = 0.151; 2-min: p = 0.644; 3-min: p = 0.395; 4-min: p = 0.247; 5-min: p = 0.289; Fig. 1d). Baseline rest FMD at 1 min had a weak negative correlation with baseline HbA1c (r = − 0.15; p = 0.016). However, the change in 1-min FMD after 12 weeks of treatment was not associated with the change in HbA1c (r = 0.08; p = 0.13). No statistically significant interaction was found between prior coronary artery disease and the change in rest FMD at 1 min after the treatments (p = 0.45).

Prespecified secondary parameters were analyzed to verify the effects of the treatments on microvascular resistance. The end diastolic velocity and the peak systolic velocity were both equivalent between groups at baseline (Table 1). The change in both velocities was distinct after 12 weeks of dapagliflozin compared to glibenclamide at both 1 and 5 min during the rest of the FMD test (Table 2). The resistive indices measured at baseline and at the 1-min and 5-min points were equivalent between the two study arms at randomization (Table 1). After 12 weeks of therapy, the change in the resistive index at 1 and 5 min during the rest FMD was significantly different between the dapagliflozin and glibenclamide groups (Table 2). No differences were observed between treatment arms in terms of anterograde or retrograde blood flow, anterograde or retrograde shear rate, or shear stress either at baseline or after 12 weeks of treatment (Additional 1: Table S2).

The change in post-I/R FMD dilation at 1 min increased a median of 2(9)% in the dapagliflozin arm and decreased a median of 0.4(11)% in the glibenclamide arm. Nevertheless, this difference did not reach statistical significance (p = 0.258) (Table 2). In accordance with the hierarchical statistical analysis strategy for the secondary parameters, we did not investigate the factors related to microvascular and macrovascular functions based on the diameter and velocity data from the post-I/R FMD.

At randomization, no statistically significant differences between study arms were found in nitrite or nitrate levels at any of the three sampling times. After 12 weeks of treatment, there was a 10% increase in the 1-min nitrite level in the dapagliflozin group and a 4% decrease in the glibenclamide group. There were significant differences between study arms in terms of absolute value (p = 0.028) and the change from baseline to 1 min (p = 0.016). The AUCs for the three time points for nitrite level measurements were equivalent at baseline but increased by 24% (p = 0.004) after dapagliflozin and decreased by 14% after glibenclamide treatment (p = 0.24). No statistically significant changes were found in plasma nitrate or endothelin-1 levels between the study arms either at pretreatment or after 12 weeks of treatment (Additional file 1: Table S1).

No statistically significant differences were found between dapagliflozin and glibenclamide in terms of the pretreatment or posttreatment levels of systemic inflammatory activity markers (C-reactive protein, interleukin-2 (IL-2), interleukin-6 (IL-6), and tumor necrosis factor alpha (TNF-α), vascular inflammation markers (soluble vascular cell adhesion molecule-1 (sVCAM), soluble intercellular adhesion molecule-1 (sICAM), or systemic oxidative stress marker (free 8-isoprostane) (Additional file 1: Table S1).

According to the intervention pill counts, the adherence rates at the first, second, and third months were 99 ± 3%, 97 ± 9%, 99 ± 4% in the dapagliflozin group and 98 ± 4%, 97 ± 6% and 98 ± 4% in the glibenclamide group. The alanine and aspartate aminotransferase levels did not differ between groups and did not change between randomization and week 12 (Additional file 1: Table S1). One adverse event was recorded in a patient in the glibenclamide arm who presented with transient acute confusion unrelated to hypoglycemia or clinical or radiological alterations characteristic of cerebral ischemia. A diagnostic hypothesis of progressive dementia was made, and the patient was referred for neurological investigation. The patient undertook the final study evaluation and was included in analyses predefined as intention-to-treat. However, the analyses performed after the patient’s data were excluded remained unchanged. Last, there were no adverse effects noted in the dapagliflozin arm.

Three previous studies in T2DM reported inconsistent dapagliflozin effects on FMD [22‐24]. The first study showed improvement in FMD after two days of treatment [23]; however, the authors reported no effect after 4 weeks of treatment [25]. In a second trial, although FMD was not significantly changed in the intention-to-treat analysis, a post hoc analysis indicated that an improvement was detected in a subgroup of patients with HbA1c > 7.0% at baseline. In the last study, 12-week treatment with dapagliflozin was compared with placebo, and although the change in FMD was inversely proportional to the change in HbA1c (r = − 0.4; p = 0.017), there was no significant improvement in FMD after dapagliflozin [24]. The main limitations of these trials were the selection of patients with FMD close to normal values as well as the suboptimal and unbalanced reduction in blood glucose in the study arms [24]. Hence, we carefully considered the equipoise of glycemic control and the presence of artery wall disease and dysfunction as requirements for designing the present trial. The pretreatment FMD values were ~ 1/2–1/3 of the mean values reported in the aforementioned studies [22, 23], suggesting that the patients in this study have worse vasomotor function than prior trials. The posttreatment HbA1c was 7.1% in both arms of the study. In light of this, our current study demonstrated that dapagliflozin leads to an improvement in FMD in individuals with severe arterial wall dysfunction when compared to glibenclamide as well as, in an intragroup exploratory analysis, when compared with baseline values. Notably, this difference occurred with equivalent glucose lowering between the two groups.

Approximately 30% of all patients included in the study had paradoxical arterial constriction after cuff deflation during the FMD protocol. This abnormal arterial response improved after treatment with dapagliflozin but not after glibenclamide treatment. We also observed a decrease in office systolic BP after dapagliflozin, but no difference in the measurements obtained at 24-h ABPM. Both findings suggest attenuation of the vasoconstrictor response to ischemic stimulation and of the white-coat effect after dapagliflozin [26]. The lack of effect of dapagliflozin on BP measured by 24-h ABPM is likely due to the careful clinical management performed in the run-in phase, which aimed to obtain groups that were comparable and suitable for the investigation of endothelial function. This strategy and finding strengthens the quality of the FMD analyses obtained in this trial.

Constriction of the brachial artery during the FMD protocol has already been described, as has its association with subclinical atherosclerotic disease [27]. There is scarce mechanistic information that was mostly obtained from studies with intra-arterial acetylcholine infusion, which has a strong correlation with FMD [28, 29]. In the normal functioning endothelium, acetylcholine promotes a rapid release of NO and, thus, vasodilation [30]. However, in the dysfunctional endothelium, the effect of acetylcholine that predominates is the contraction of smooth muscle cells causing vasoconstriction [30]. By inference, it is possible that the reduction in paradoxical vasoconstriction was favored by the improvement in endothelial function. Nevertheless, other nontested mechanisms may also contribute to this change, such as the direct action of dapagliflozin on vascular smooth muscle.

To assess the entire artery wall response during the FMD protocol, we extended the capture period of the brachial artery diameter and flow velocity to 5 min after cuff deflation. In this manner, we intended to discern any prolonged effects of dapagliflozin on FMD, particularly on microvascular function. A prior study reported that blocking endothelial nitric oxide synthase (eNOS) with NG-monomethyl-L-arginine (L-NMMA) did not alter flow velocity at the peak of hyperemia but reduced the duration of hyperemia [31]. Our hypothesis was that if dapagliflozin acted on NO bioavailability, this effect could be better detected at the end of 5 min. In fact, in dapagliflozin-treated patients, we found a more pronounced change in both the peak systolic velocity and the end diastolic velocity between the study arms, mostly at 5 min after cuff deflation, suggesting a reduction in microvascular resistance. In fact, while in patients treated with dapagliflozin, there was a decline in the resistive index, in those treated with glibenclamide, this index increased.

As demonstrated in several studies [32‐35], microvascular tone is mainly determined by hyperemic flow during FMD, which is particularly reflected in diastolic velocities during hyperemia [33]. As described above, the resistive index equation reflects resistance to blood flow, which can be greatly determined by microcirculation, particularly after cuff release. Thus, it is of special importance that, in this phase, our results show a significant difference. A similar result was reported in the resistive index measured in the renal artery following dapagliflozin treatment [23], corroborating this finding and indicating that this improvement in microvascular function may occur in multiple arterial beds. In the microcirculation, there is a nonlinear relationship between the diameter and the viscosity of the vessel, with the relative effective viscosity increasing six to seven times in vessels that are the size of capillaries [36]. The increase in blood viscosity shifts the curve of the association between the relative effective viscosity and diameter of the vessel upwards [37]. In the ADDENDA-BHS2 trial, blood viscosity increased as a result of treatment with dapagliflozin. Thus, the effect of SGLT2i in decreasing microvascular resistance is possibly underestimated by the concomitant increase in blood viscosity.

Elevated plasma nitrite levels are indicative of increased NO production, and nitrate levels indicate the total available NO in the blood [38]. Dapagliflozin treatment increased nitrite production during FMD. In contrast, neither drug therapy altered nitrate levels. As commented above, SGLT2i activates eNOS and increases NO production [16], which may in part explain the improvement in both macrovascular and microvascular function [17]. Indeed, FMD is abolished after eNOS inhibition [31], and patients with impaired dilatory brachial artery responses do not present with any increase in nitrite levels [38].

The increase in the bioavailability of NO can also result from the reduction of its degradation, particularly by attenuating oxidative stress [39]. This possibility was considered in the study design based on a previous report indicating that dapagliflozin can mitigate oxidative stress as assessed by plasma 8-hydroxy-2′-deoxyguanosin, which is a biomarker of oxidative DNA damage [22]. We did not observe any change in the plasma levels of isoprostane, which is a marker of peroxidation of essential fatty acids. The contrasting results between these two studies may be due to differences in the two biomarkers of oxidative stress or differences in the studied populations. It is important to recognize that the lack of a significant change in isoprostane levels does not exclude the possibility of attenuation in oxidative stress and thus the reduction of NO degradation.

We found no significant difference in endothelin-1 levels between treatments. However, other mediators may have changed both the macrovascular and microvascular responses. Activation of the K_ATP channel is of particular relevance here, as glibenclamide was the active control group. Certain studies have shown that inhibition of K_ATP channels with sulfonylureas modestly attenuates hyperemia without affecting peak flow velocity [40]. Nevertheless, contradictory findings have been reported elsewhere [41]. We did not detect changes in nitrite level or systolic or diastolic flow velocity, but we found significant worsening of FMD at 1 min after glibenclamide treatment. In addition, the presence of arterial dilation during the FMD procedure decreased by almost half in the glibenclamide group and increased by almost a third in patients treated with dapagliflozin. This finding increased the differences between experimental arms and the significance of the primary outcome. However, the pretreatment vs. posttreatment analysis demonstrated that dapagliflozin was associated with significant increases in FMD at every time point throughout the 5 min of recording. Thus, dapagliflozin therapy had a direct effect on the measured endpoints.

Endothelial dysfunction after I/R is a ubiquitous finding in several tissues and in a variety of species [42], and it has been associated with the extension of myocardial injury after an acute coronary event [43]. The reintroduction of molecular oxygen at reperfusion generates reactive oxygen species, which can directly inactivate NO. In line with this concept, we found that the levels of nitrate and nitrite both before and after FMD decreased substantially after the period of I/R in both study arms. Such attenuation of FMD after I/R undermined the statistical power for the difference between arms. Despite this, we noticed a slight increase and a minor decrease in post-I/R FMD in the dapagliflozin and glibenclamide arms, respectively, but these differences did not reach statistical significance. We speculate that the drop in FMD after I/R surpassed the endothelial improvement obtained with SGLT2i, particularly in individuals with established atherosclerotic disease. In agreement with that, no changes in oxidative stress, estimated by serum isoprostane, or vascular and systemic inflammatory markers were detected in either study arm. This experimental approach is unprecedented, and there is much room for clarification. For example, what is the expected range of the post-I/R FMD in the population with or without T2D, is it reversible and how long must the treatment time be to mitigate it. The present study was not able to exclude or confirm the SGLT2i effect on this endothelial function marker, but it generated data so that new studies can be planned with greater accuracy.

As mentioned above, current evidence and guidelines support the concept that SGLT2i can reduce the risk of myocardial infarction in individuals with prior manifestations of coronary events but not in those with uneventful clinical histories [4‐6]. Clinical trials with follow-up long enough to verify the effect of SGLT2i on atherothrombotic events are unlikely to occur due to ethical limitations; i.e., the early reduction in hospitalizations for heart failure and cardiovascular mortality (1 month or less) [44] would interrupt the study before the effect of SGLT2i on arterial disease could be properly tested. This limitation is particularly important in individuals with a low annual incidence of cardiovascular events such as those in primary prevention. In this scenario, the present study is an alternative to indicate that among individuals in primary prevention, those with subclinical atherosclerotic disease may potentially benefit from SGLT2i from macro- and microvascular points of view.

The results of this study would be closer to the ideal if all enrolled patients had been using statins for a long period of time but statin effect on FMD reaches stability at long and although half of the patients did not use statins at admission, we did not include this therapy to avoid distortions in the magnitude of the effect of experimental treatments. However, statin use was balanced between the arms, and we found no interaction between the use of statins and the change in the primary outcome.