Research ArticleAtherosclerosis

Insulin-induced vascular redox dysregulation in human atherosclerosis is ameliorated by dipeptidyl peptidase 4 inhibition

See allHide authors and affiliations

Science Translational Medicine  29 Apr 2020:
Vol. 12, Issue 541, eaav8824
DOI: 10.1126/scitranslmed.aav8824

From stress to sensitivity

Poor glycemic control drives cardiovascular disease, but aggressive blood glucose lowering does not improve cardiovascular risk. To understand the underlying mechanisms, Akoumianakis et al. studied the relationship between the local redox state of blood vessels and cardiovascular outcomes of patients with coronary atherosclerosis. They found that diseased vessels were insulin resistant and had increased oxidative stress and reduced nitric oxide bioavailability, which could be reversed by treatment with an inhibitor of dipeptidyl peptidase 4 (DPP4). Vascular insulin sensitivity was also restored in mice with atherosclerosis upon treatment with an oral DPP4 inhibitor. Results uncover how DPP4 inhibition induces insulin sensitization in the vascular wall and suggest that cotreatment with insulin may be therapeutic for patients with cardiometabolic disease.


Recent clinical trials have revealed that aggressive insulin treatment has a neutral effect on cardiovascular risk in patients with diabetes despite improved glycemic control, which may suggest confounding direct effects of insulin on the human vasculature. We studied 580 patients with coronary atherosclerosis undergoing coronary artery bypass surgery (CABG), finding that high endogenous insulin was associated with reduced nitric oxide (NO) bioavailability ex vivo in vessels obtained during surgery. Ex vivo experiments with human internal mammary arteries and saphenous veins obtained from 94 patients undergoing CABG revealed that both long-acting insulin analogs and human insulin triggered abnormal responses of post–insulin receptor substrate 1 downstream signaling ex vivo, independently of systemic insulin resistance status. These abnormal responses led to reduced NO bioavailability, activation of NADPH oxidases, and uncoupling of endothelial NO synthase. Treatment with an oral dipeptidyl peptidase 4 inhibitor (DPP4i) in vivo or DPP4i administered to vessels ex vivo restored physiological insulin signaling, reversed vascular insulin responses, reduced vascular oxidative stress, and improved endothelial function in humans. The detrimental effects of insulin on vascular redox state and endothelial function as well as the insulin-sensitizing effect of DPP4i were also validated in high-fat diet-fed ApoE−/− mice treated with DPP4i. High plasma DPP4 activity and high insulin were additively related with higher cardiac mortality in patients with coronary atherosclerosis undergoing CABG. These findings may explain the inability of aggressive insulin treatment to improve cardiovascular outcomes, raising the question whether vascular insulin sensitization with DPP4i should precede initiation of insulin treatment and continue as part of a long-term combination therapy.


Type 2 diabetes mellitus is a global health epidemic and an important driver of cardiovascular complications (1). This is believed to result from hyperglycemia, which has crucial and profound effects on the human vasculature (2). Such effects range from vascular protein kinase C activation, glycation of a variety of important proteins such as Akt (3), advanced glycation end-product (AGE) formation, and downstream AGE signaling and dysregulation of vascular redox signaling by affecting the activity of enzymes such as reduced form of nicotinamide adenine dinucleotide phosphate (NADPH) oxidases and the coupling status of endothelial nitric oxide (NO) synthase (eNOS) (2). Despite the important vascular effects of hyperglycemia and the protective effect of standard glycemic control on cardiovascular clinical end points, aggressive glucose lowering with insulin analogs has failed to further improve cardiovascular outcomes in type 2 diabetes, despite achieving optimal glycemic control (46), suggesting that the cardiovascular benefit of glucose lowering in diabetes is limited to changes at the high end of the serum glucose range. The Action to Control Cardiovascular Risk in Diabetes (ACCORD) clinical trial was the first landmark study to display no cardiovascular risk (mortality and nonfatal events such as myocardial infarction) benefit after intensive glycemic control largely by insulin-based treatments (7), suggesting that glycemic control is not sufficient to prevent vascular complications and local vascular parameters should be considered.

The vascular effects of insulin involve downstream activation of insulin receptor substrate 1 (IRS1) (8), leading to potential activation of the phosphatidylinositol 3-kinase/Akt pathway or the mitogen-activated protein kinase pathway (8). Akt signaling has been associated with the ability of insulin to activate eNOS and increase NO bioavailability (9). However, it is unclear how vascular insulin signaling changes in diabetes and human atherosclerosis because mechanistic data available are focused on either in vitro cell culture or in vivo animal models.

The dipeptidyl peptidase 4 (DPP4) protease cleaves proline dipeptides from the N-terminus of polypeptides including glucagon-like peptide 1 (GLP1), a peptide with glucose-lowering abilities (10). It is now evident that DPP4 inhibitors (DPP4is) ameliorate cellular insulin resistance (IR) in experimental models (10), but their effects on insulin signaling in the human vascular wall are unknown.

In this study, we explored the direct effects of human and synthetic insulins on redox signaling and NO bioavailability in human arteries and veins from patients with coronary atherosclerosis. We hypothesized that vascular IR may be responsible for the inability of aggressive insulin treatment to reduce cardiovascular risk in patients with diabetes. We further investigated the potential role of insulin sensitization strategies in restoring physiological insulin signaling in the human vascular wall.


Effects of insulin on vascular redox state in patients with atherosclerosis

We first investigated the association between circulating insulin and endothelial function in nondiabetic patients undergoing coronary artery bypass surgery (CABG) (study 1; Table 1 and table S1). High serum insulin was associated with reduced vasorelaxations of human vessels [saphenous veins (SVs)] in response to acetylcholine (ACh; Fig. 1A) and bradykinin (BK; Fig. 1B), but not to sodium nitroprusside (SNP; Fig. 1C), suggesting an inverse association between serum insulin and NO bioavailability in the human endothelium. To explore whether the inverse association between insulin and endothelial function is causal, we then exposed human vessels from atherosclerosis patients with diabetes and without diabetes or evidence of systemic IR, evidenced by homeostatic model assessment of systemic IR (HOMA-IR<2.9) to exogenous insulin ex vivo, using both long-acting insulin analogs (degludec or M1 metabolite of glargine) and human insulin (study 2; Table 1). All insulin types significantly reduced vasorelaxations in response to ACh but not to SNP in all patient vessels (Fig. 1, D to G for insulin glargine and fig. S1 for insulin degludec and human insulin), suggesting a class effect of insulin on the vascular wall, even in the absence of diabetes or systemic IR.

Table 1 Demographic characteristics of study participants.

BMI, body mass index; HOMA-IR, homeostatic model assessment–insulin resistance; hsCRP, high-sensitivity C-reactive protein; ACEi, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker. Age and BMI are presented as means ± SEM; HOMA-IR is presented as median[25th to 75th percentile]; P values are calculated by Fisher’s exact tests for categorical variables, by unpaired t tests for continuous normally distributed variables (age, BMI, and waist-to-hip ratio), and by Mann-Whitney U tests for continuous non-normally distributed variables (HOMA-IR and hsCRP).

View this table:
Fig. 1 Insulin impairs endothelial function in humans with coronary atherosclerosis.

(A to C) Vasorelaxation curves of phenylephrine precontracted human vessels in response to (A) acetylcholine (ACh; endothelium-dependent, n = 110), (B) bradykinin (BK; endothelium-dependent, n = 38), and (C) sodium nitroprusside (SNP; endothelium-independent, n = 92) per circulating insulin tertiles in study arm 1. (D to G) Serial rings of human vessels were treated with and without insulin M1 (10 μM) before testing vascorelaxation in response to ACh [(D) for patients without diabetes (n = 6 pairs) and (E) for patients with diabetes (n = 6 pairs)] or SNP [(F) for patients without diabetes (n = 6 pairs) and (G) for patients with diabetes (n = 6 pairs)]. *P < 0.05 versus high tertile in panel (A); versus low tertile in panel (B); versus control in panels (D) to (E); NS, nonsignificant versus control (P > 0.05); P values calculated by two-way ANOVA for repeated measures with (ACh dose) × (insulin treatment) interaction; data presented as means ± SEM.

To understand how insulin could cause endothelial dysfunction in vessels from patients with vascular disease, we then explored the interactions between insulin and vascular redox state in patients without diabetes from study 1 (to avoid treatment confounding in diabetic patients). We observed that increased serum insulin was associated with increased NADPH oxidases activity in human vessels [internal mammary arteries (IMAs) and SV], evidenced by increased NADPH-stimulated superoxide (O2·) and particularly Vas2870-inhibitable O2· production from these vessels as measured by lucigenin chemiluminescence (fig. S2). Vas2870 is an inhibitor of all NADPH oxidase isoforms. To examine whether exogenous insulin administration could causally increase oxidative stress in the human vascular wall, we first exposed human IMA and SV (obtained from patients in study 2) to human insulin, insulin glargine M1, and insulin degludec in a screening dose-response experiment. We observed that all insulin types increased vascular basal, NADPH-stimulated, and Vas2870-inhibitable O2· (fig. S3). Human insulin and M1 glargine had that effect at concentrations of ≥10 nM, whereas degludec displayed similar effect at 100 nM (fig. S3). This is in agreement with the described pharmacodynamic properties of these insulin types, considering that M1 glargine is very similar to human insulin and degludec is less potent than the other two (1113).

Further incubations with insulin glargine M1, used as a representative insulin analog and named as “insulin” in the subsequent results sections, ex vivo demonstrated that insulin (10 nM) increased O2· generation in both SV and IMA from patients with diabetes and from patients without diabetes or systemic IR, which is comparable to the in vivo situation [serum insulin median[25th to 75th percentile], 5.5[3.4 to 8.3] nM in study 1; Fig. 2, A to C (for SV) and D to F (for IMA), and fig. S4, A to C (for human insulin)]. This was due to activation of vascular NADPH oxidases as documented by the increase in NADPH-stimulated and Vas2870-inhibitable O2· production. These findings were replicated using dihydroethidium (DHE) staining on intact IMA segments treated with insulin in the presence or absence of Vas2870 (Fig. 2G). Complementary to this, long-term insulin treatment for 8 hours activated several proinflammatory pathways in human IMA, further supporting a potentially detrimental direct effect of insulin on the vascular wall of patients with atherosclerosis (fig. S5).

Fig. 2 Insulin increases NADPH oxidase activity in vessels from patients with coronary atherosclerosis.

(A to C) Effect of exogenous insulin (glargine M1, 10 nM) on basal [(A), n = 7 pairs], NADPH-stimulated [(B), n = 7 pairs], and Vas2870-inhibitable [(C), n = 7 pairs] superoxide (O2·) in saphenous vein (SV) segments. T2DM, type 2 diabetes mellitus. (D to F) Effect of exogenous insulin (glargine M1, 10 nM) on basal [(D), n = 5 pairs], NADPH-stimulated [(E), n = 5 pairs], and Vas2870-inhibitable [(F), n = 5 pairs] O2· in IMA segments. HOMA-IR for patients without diabetes was 1.64[0.87 to 2.73] (median[25th to 75th percentile]). (G) Example dihydroethidium (DHE) staining images for in situ visualization of basal and Vas2870-inhibitable O2· production in response to insulin (glargine M1, 10 nM) in IMA. DHE staining appears as red and autofluorescence as green. (H to J) Effects of ex vivo insulin incubation (glargine, 10 nM) on basal [(H), n = 5 pairs], NADPH-stimulated [(I), n = 5 pairs], and Vas2870-inhibitable [(J), n = 5 pairs] O2· generation in aortic tissue from wild type mice. *P < 0.05 versus control after Bonferroni correction in (A) to (F). P = 0.048 versus control for (H) to (J) by Wilcoxon signed-rank tests. Patients with diabetes receiving an oral dipeptidyl peptidase 4 (DPP4) inhibitor were excluded from these experiments. P values are calculated by Wilcoxon signed-rank test in (A) to (F) and (H) to (J); data are presented as means ± SEM.

To test whether these observations on the effects of insulin on human vessels were associated with the presence of vascular disease, we exposed mouse aortas from healthy wild-type animals to the same protocol of insulin treatment ex vivo. We observed that insulin reduced vascular O2· in the mouse aortas by reducing NADPH oxidase activity by 50 to 70%, providing a positive control for the study intervention (Fig. 2, H to J, and fig. S6, A to C).

DPP4 inhibition restores vascular redox responses to insulin in human atherosclerosis

Our results, so far, suggest that insulin has a class stimulatory effect on vascular NADPH oxidases in human atherosclerosis, independently of the presence of diabetes or systemic IR. To explore whether an insulin-sensitizing intervention such as DPP4i treatment would reverse these effects, we selected patients receiving chronic treatment with oral DPP4i and exposed their IMA to insulin ex vivo. We observed a marked reversal of the effects of insulin on vascular redox state, suppressing NADPH oxidase activity and vascular O2· generation (Fig. 3, A to C). This was not observed with metformin, a common antidiabetic medication with insulin-sensitizing properties (fig. S7) (14), suggesting that DPP4i may have strong vasculature-specific effects. To examine whether DPP4i acts directly on the human arterial wall to reverse the effects of insulin on vascular redox state, we exposed human arteries and veins to insulin ex vivo in the presence or absence of KR62436, a synthetic DPP4i. We found that insulin alone stimulated O2· production in both human IMA and SVs, whereas preincubation of rings from the same vessels with KR62436 reversed their vascular responses to insulin, leading to reduced vascular O2· in response to exogenous insulin administration by suppressing the activity of NADPH oxidases (Fig. 3, D to I, and fig. S8, A to F). DPP4i had no direct antioxidant or O2·-scavenging properties, as evidenced by its neutral effect on xanthine oxidase–derived O2·, which is used as a chemical protocol to evaluate direct O2·-scavenging properties (fig. S9). These findings suggest that DPP4 inhibition in the human vascular wall restores physiological responses of vascular redox signaling to exogenous insulin administration. Crucially, insulin administered to apolipoprotein E (ApoE)−/− mice [fed with high-fat diet (HFD) for 4 weeks to stimulate cardiometabolic disease], increased basal, NADPH-stimulated, and Vas2870-inhibitable O2· measured in aortic tissue, all of which were reversed by 4-week pretreatment with linagliptin (a clinically used DPP4i) during the course of HFD (fig. S10, A to C). This provides an in vivo validation of the proof of concept that cardiometabolic disease is characterized by vascular IR, which can be rescued by oral DPP4i treatment.

Fig. 3 DPP4 inhibition modulates the activation of vascular NADPH oxidases in response to insulin in humans.

(A to C) Effect of ex vivo insulin incubation (glargine M1, 10 nM) on basal [(A), n = 5 pairs], NADPH-stimulated [(B), n = 5 pairs], and Vas2870-inhibitable [(C), n = 5 pairs] superoxide (O2·) in SV segments of a subgroup of patients with diabetes receiving oral DPP4i treatment in vivo. (D to F) Effect of ex vivo insulin incubation (glargine M1, 10 nM) on basal [(D), n = 10], NADPH-stimulated [(E), n = 10], and Vas2870-inhibitable [(F), n = 10], O2· in IMA with or without ex vivo preincubation with DPP4i KR62436 (70 μM). (G to I) Effect of ex vivo insulin incubation (glargine M1, 10 nM) on basal [(G), n = 13], NADPH-stimulated [(H), n = 13], and Vas2870-inhibitable [(I), n = 13] O2· in SV with or without ex vivo DPP4i preincubation. (J to L) Effect of ex vivo insulin incubation (glargine M1, 10 nM) on (J) Rac1 GTP activation (n = 5), (K) membrane translocation of active Rac1 (n = 5), and (L) the p47phox subunit (n = 5) O2· in human SV with or without ex vivo DPP4i preincubation). P = 0.047 by Wilcoxon signed-rank test in (A) to (C). *P < 0.05 versus control by Wilcoxon signed-rank test in (D) to (L), followed by Bonferroni correction as appropriate; data presented as means ± SEM.

To understand the underlying mechanisms by which insulin and DPP4i regulate NADPH oxidase activity in human vessels, we explored their direct effects on the regulatory subunit of NOX1 and NOX2 isoforms of NADPH oxidases, Rac1. Insulin activated Rac1 (Fig. 3J), triggering its membrane translocation together with the p47phox subunit of the enzymes (Fig. 3, K and L). These effects were reversed after pretreatment of these vessels with DPP4i, in which case insulin led to guanosine 5′-triphosphate (GTP)–Rac1 reduction and prevented the membrane translocation of Rac1 and p47phox (Fig. 3, J to L). DPP4 inhibition had no direct effects on Rac1 activation or Rac1/p47phox membrane translocation (Fig. 3, J to L). In agreement with these findings, high circulating DPP4 activity was positively associated with vascular O2· generation (basal, NADPH-stimulated, and Vas2870-inhibitable) in human vessels (fig. S11). These findings imply that targeting DPP4 in patients with diabetes may restore physiological vascular insulin signaling, at least in the presence of advanced atherosclerosis.

DPP4 inhibition modulates the effects of exogenous insulin on eNOS in human vessels

To better understand how exogenous insulin controls vascular redox state in human vessels, we next investigated the direct effects of insulin on vascular NO bioavailability and eNOS coupling in vessels from patients with atherosclerosis. Insulin directly induced vascular eNOS uncoupling, documented by a marked increase in L-NG-nitroarginine methyl ester (LNAME)–inhibitable O2· (Fig. 4A). This suggests that insulin turns eNOS from a source of NO to a source of O2·, further dysregulating vascular redox signaling. Treatment of these vessels with DPP4i reversed the effects of insulin on eNOS coupling (Fig. 4A), confirming that insulin treatment together with DPP4i improves vascular redox signaling by restoring eNOS coupling in human atherosclerosis.

Fig. 4 DPP4 regulates the effect of insulin on vascular eNOS coupling in humans.

(A) Effect of ex vivo insulin incubation (glargine M1, 10 nM) on eNOS uncoupling evidenced by the LNAME-induced reduction of vascular superoxide [LNAME-Δ(O2·)] in the presence or absence of DPP4i preincubation [(A), n = 5 to 7 per intervention]. (B and C) Effect of insulin (glargine M1, 10 nM) on the phosphorylation of eNOS at Ser1177 in (B) human SV segments (n = 5) versus (C) human umbilical vein endothelial cells (HUVECs) in vitro (n = 5) in the presence or absence of DPP4i preincubation. (D to F) Effect of ex vivo insulin [glargine M1 (10 nM) on (D) vascular tetrahydrobiopterin (BH4) content (n = 5)], (E) total biopterin content (n = 5), and (F) BH4 bioavailability (n = 5) in the presence or absence of DPP4i. (G to H) Effect of ex vivo insulin (glargine M1, 10 nM)/DPP4i incubations on (G) endothelium-dependent ACh vasorelaxations (n = 5 to 6 per intervention) and (H) endothelium-independent SNP vasorelaxations (n = 5 to 6 per intervention). *P < 0.05 versus control. P values are calculated by Wilcoxon signed-rank tests in (A) to (F) and by two-way ANOVA for matched observations in (G) to (H). Data presented as means ± SEM.

Given that insulin has been shown to affect eNOS activity via Akt-mediated Ser1177 phosphorylation in vitro (9), we next explored the effects of insulin on eNOS phosphorylation status in humans with vascular disease. We found that insulin alone did not induce eNOS phosphorylation at the activation site Ser1177, whereas significant Ser1177 phosphorylation was induced by insulin in the presence of DPP4i (Fig. 4B). In contrast, insulin increased eNOS phosphorylation at Ser1177 in human umbilical vein endothelial cells (HUVECs) used as a biological positive control, whereas DPP4i conveyed no additional benefit in these cells (Fig. 4C). These findings highlight the discrepancy in vascular insulin responses between humans with vascular disease and disease-free in vitro and in vivo models.

To understand how insulin induces eNOS uncoupling, we quantified vascular eNOS cofactor tetrahydrobiopterin (BH4), a key regulator of eNOS coupling. Insulin reduced BH4 bioavailability without affecting total biopterins content [that includes dihydrobiopterin (BH2) and biopterin (B)], resulting in reduced BH4/total biopterins ratio (Fig. 4, D to F). This finding suggests that insulin induces BH4 oxidation without affecting its biosynthesis, leading to eNOS uncoupling by changing the stoichiometry between BH4 and BH2/B (Fig. 4, D to F). Conversely, in the presence of DPP4i, insulin increased vascular BH4 content and the ratio of BH4/total biopterins, improving eNOS coupling (Fig. 4, D to F).

Given that, in the presence of DPP4i, insulin can improve eNOS coupling and activate eNOS, we then hypothesized that it would also improve endothelial function in human vessels. We found that insulin impaired the vasorelaxations of human vessels to ACh, whereas insulin had the opposite effect in the presence of a DPP4i, improving ACh-induced vasorelaxations (Fig. 4G). These effects were endothelium specific and did not affect the endothelium-independent vasorelaxations to SNP (Fig. 4H). DPP4i alone did not affect the relaxations of these vessels to ACh or SNP, confirming its role as a modulator of insulin signaling in human vessels. Insulin also impaired endothelium-dependent ACh vasorelaxations in the aortas of HFD-fed ApoE−/− mice, an effect abolished by oral linagliptin, whereas there were no differences in endothelium-independent vasorelaxations between the aortas of treated versus control mice in the absence of insulin stimulation (fig. S12).

Characterizing abnormal vascular insulin signaling in humans with vascular disease

Given that insulin induces oxidative stress and endothelial dysfunction in vessels from patients with atherosclerosis, we hypothesized that these dysregulated vascular redox responses to insulin could reflect abnormal downstream insulin signaling, representing a default state of vascular IR in these human vessels, even in patients with no evidence of IR or diabetes. To understand the nature of these unexpected responses, we investigated the balance between phosphorylation (activation) of vascular Akt versus extracellular signal–regulated kinase 1 & 2 (Erk1&2), as representative downstream mediators of the two insulin signaling axes dysregulated in IR. We observed that in vessels from our patients, insulin failed to stimulate Akt phosphorylation, whereas it significantly induced phosphorylation of Erk1&2, resulting in an imbalance between the two signaling axes in vessels from nondiabetic (Fig. 5, A to C) and diabetic (Fig. 5, D to F) patients. Aortic rings from healthy wild-type mice were used as a positive control, confirming that insulin significantly induces Akt phosphorylation much more so than Erk1&2 in the vessel wall of these mice (Fig. 5, G to I). In line with our previous findings, insulin increased the phosphorylation of Akt, but not Erk1&2, in vascular segments from patients with diabetes taking oral DPP4i treatment (Fig. 5, J to L). These results confirm that there is a selective dysregulation of downstream insulin signaling in vessels from patients with vascular disease, in favor of Erk1&2 activation (over Akt), indicating the presence of vascular IR. This abnormal vascular insulin signaling can be reversed by pretreatment with a DPP4i.

Fig. 5 The human vascular wall is characterized by vascular IR, and this is reversed by DPP4 inhibition.

(A to C) Effect of ex vivo insulin (glargine M1, 10 nM) on the phosphorylation of (A) Akt (n = 5), (B) Erk1&2 (n = 5), and (C) the balance of the two (n = 5) in the human vascular wall (SV) in patients without diabetes and with coronary atherosclerosis. (D to F) Effect of ex vivo insulin (glargine M1, 10 nM) on the phosphorylation of (D) Akt (n = 5), (E) Erk1&2 (n = 5), and (F) the balance of the two (n = 5) in SV of patients with diabetes and coronary atherosclerosis. (G to I) Effect of ex vivo insulin (glargine M1, 10 nM) on the phosphorylation of (G) Akt (n = 5), (H) Erk1&2 (n = 5), and (I) the balance of the two (n = 5) in atherosclerosis-free wild-type mouse aortic tissue. (J to L) Effect of ex vivo insulin (glargine M1, 10 nM) on the phosphorylation of (J) Akt (n = 5), (K) Erk1&2 (n = 5), and (L) the balance of the two (n = 5) in the human vascular wall (SV) in patients with diabetes on oral DPP4i treatment. *P < 0.05; NS, nonsignificant versus control. P values are calculated by Wilcoxon signed-rank tests in all panels. Data presented as means ± SEM.

Characterizing the insulin-sensitizing properties of DPP4 inhibition

To understand the mechanisms by which DPP4 inhibition regulates downstream insulin and redox signaling in the vascular wall of patients with atherosclerosis, we first examined whether DPP4 inhibition acts directly on the human vascular wall. In the presence of a synthetic DPP4i, insulin increased the activation of Akt over Erk1&2 in both human arteries (Fig. 6, A to C) and veins (Fig. 6, D to F, and fig. S13). To prove that this DPP4i-induced shift of insulin signaling in human vessels is responsible for the beneficial effect of the combined DPP4i/insulin treatment on vascular redox state, we first examined whether insulin activates Rac1 in the presence of an Erk1&2 inhibitor. We found that Erk1&2 inhibition using 3-(2-aminoethyl)-5-((4-ethoxyphenyl)methylene)-2,4-thiazolidinedione abolished the ability of insulin to stimulate Rac1 GTP activation, suggesting that Erk1&2 signaling is responsible for the insulin-mediated activation of NADPH oxidases in human vascular disease (Fig. 6G). On the contrary, the combination of insulin with DPP4i did not induce vascular eNOS phosphorylation in the presence of the Akt inhibitor wortmannin (Fig. 6H), suggesting that the DPP4i-induced effect of insulin on eNOS phosphorylation is dependent on activation of Akt.

Fig. 6 DPP4 inhibition regulates vascular insulin signaling via restoring local insulin sensitivity in an AMPK-dependent manner.

(A to C) Effect of DPP4i on insulin (glargine M1, 10 nM)–stimulated phosphorylation of (A) Akt (n = 5), (B) Erk1&2 (n = 5), and (C) the balance between the two (n = 5) in human IMA segments. (D to F) Effect of DPP4i on insulin (glargine M1, 10 nM)–stimulated phosphorylation of (D) Akt (n = 5 to 7), (E) Erk1&2 (n = 5 to 7), and (F) the balance between the two (defined as the ratio of pAkt/pErk1&2, n = 5 to 7) in human SV segments. (G) Effect of Erk1&2 inhibition using 3-(2-aminoethyl)-5-((4-ethoxyphenyl)methylene)-2,4-thiazolidinedione (70 μM) on insulin (glargine, 10 nM)–stimulated Rac1 activation (n = 5 to 6). (H) Effect of Akt inhibition using wortmannin (100 nM) on the ability of insulin (glargine M1, 10 nM)/DPP4i combination to induce eNOS Ser1177 phosphorylation (n = 5 to 8). (I) Effect of DPP4i on insulin response substrate 1 (IRS1) Ser307 phosphorylation, a site linked with molecular IR (n = 5 pairs). (J) Effect of DPP4i on AMPK Thr172 phosphorylation (n = 5 to 7). (K) Consequence of AMPK preinhibition on the effects of insulin (glargine M1, 10 nM)/DPP4i incubations on vascular NADPH-stimulated O2· (n = 5). (L) Consequence of AMPK preinhibition by compound C on the effect of DPP4i on vascular IRS1 Ser307 phosphorylation (n = 5). *P < 0.05 versus control. P values are calculated by Wilcoxon signed-rank tests in all panels. Data presented as means ± SEM.

IRS1 acts as a hub for insulin’s signaling, and its phosphorylation at Ser307 shifts after receptor signaling toward Erk1&2. We examined whether DPP4 inhibition modifies the responses of human vessels to insulin by targeting IRS1. DPP4i significantly reduced the phosphorylation of IRS1 at Ser307 (Fig. 6I). To understand how DPP4 inhibition controls IRS1 phosphorylation, we then explored the ability of DPP4i to regulate the activation of adenosine 5′-monophosphate–activated kinase α2 (AMPKα2), a molecule with known insulin-sensitizing properties, which has recently been linked with DPP4 signaling in vitro (15). We found that DPP4i directly induced phosphorylation of AMPKα2 at its activation site Thr172 (Fig. 6J). Preincubation of human vessels with compound C, an AMPK inhibitor, rendered DPP4i unable to reverse the stimulatory effects of insulin on vascular NADPH oxidase activity (Fig. 6K). Last, AMPK inhibition abolished the ability of DPP4i to rescue insulin sensitivity at the level of IRS1 (Fig. 6L). These findings suggest that DPP4i restores vascular insulin sensitivity and elicits antioxidant responses to insulin via an AMPK-mediated mechanism.

We further explored whether DPP4i exerts its insulin-sensitizing effects by increasing the vascular bioavailability of GLP1. In the presence of GLP1 receptor (GLP1R) blockade, insulin still induced O2· generation in human arteries and DPP4i prevented this effect; however, DPP4i failed to lead to an insulin-induced reduction of vascular O2· below the baseline, suggesting that abnormal insulin signaling in the vascular wall is not totally reversed by DPP4i in the presence of GLP1R blockade. This finding demonstrates that the effect of DPP4i on vascular insulin signaling is partly GLP1R-mediated (fig. S14, A to C). Considering that protein kinase C–β (PKCβ) has demonstrated endothelial insulin-sensitizing properties (16), we investigated whether PKCβ could drive the vascular effects of DPP4i. Upon PKCβ-specific inhibition, we observed that the O2·-propagating effects of insulin were attenuated (fig. S14, D to F). DPP4i, on the other hand, maintained its ability to further improve insulin sensitivity, as evidenced by the significant reduction in arterial O2· in response to insulin, even in the presence of PKCβ inhibition (fig. S14, D to F).

We then explored the ability of DPP4i to modify redox-sensitive inflammatory transcriptional pathways in human primary endothelial cells. DPP4i partly prevented nuclear translocation of nuclear factor κB (NFκB) in tumor necrosis factor–α (TNFα)–stimulated HUVECs (fig. S15). Given that NFκB is a redox-sensitive transcriptional pathway, it is likely that it is directly involved in the development of vascular IR in human atherosclerosis.

Clinical implications of the interaction between DPP4 and insulin

To explore the value of systemic DPP4 activity and insulin concentration as biomarkers of vascular redox state in patients with coronary artery disease, we stratified patients in subgroups depending on plasma DPP4 activity and insulin. We observed that patients in the lowest tertile of both DPP4 activity and insulin had markedly lower NADPH oxidase–derived O2· production in their IMA compared to patients in the highest tertile of DPP4 activity and insulin, as evaluated by measuring arterial NADPH-stimulated (Fig. 7A) and Vas2870-inhibitable (Fig. 7B) O2· production, which was independent of the use of statins [known pleiotropic regulator of vascular NADPH oxidase activity (17)] upon multivariate regression analysis (table S2). This confirmed a cumulative effect of high serum insulin and high DPP4 activity on vascular oxidative stress, introducing their potential role as combined biomarkers and therapeutic targets in patients with atherosclerosis.

Fig. 7 Clinical implications of the interactions between systemic DPP4 activity and insulin.

(A and B) Associations of arterial NADPH-stimulated (A) and Vas2870-inhibitable (B) O2· with combined serum insulin and DPP4 activity tertiles in study 1 (n = 580). (C) Association of combined high serum insulin and serum DPP4 activity with relative risk for cardiac death after adjustment for other risk factors (euroSCORE II, hyperlipidemia, hypertension, NYHA class, and circulating hsCRP). P values in (A) and (B) are calculated by Kruskal-Wallis tests. In (C), the hazard ratio and P+ value presented are calculated from Cox regression after adjusting for euroSCORE II, hyperlipidemia, hypertension, NYHA class, and circulating hsCRP. Hazard ratio is presented as HR[95% CI]. Data presented as median[25th to 75th percentile] in (A) and (B).

Diabetes has an inflammatory pathophysiological component, and we have found that it is characterized by elevated plasma inflammatory cytokines such as interleukin-6 (IL-6) and TNFα, as well as high-sensitivity C-reactive protein (hsCRP) (fig. S16). However, the circulating concentrations of these inflammatory biomarkers were not associated with arterial redox state (fig. S16), whereas the positive association of high DPP4 activity/high insulin with arterial NADPH oxidase activity was independent of the circulating concentration of IL-6, TNFα, or hsCRP (table S2).

We next explored the predictive value of circulating DPP4 activity and insulin on cardiovascular and all-cause mortality. In total, we recorded 49 patient deaths, 21 of which classified as cardiac. Patients in the highest tertile for both serum DPP4 activity and serum insulin displayed significantly higher risk for cardiac death compared to the rest of the sample population {hazard ratio (HR) [95% confidence interval (CI)], 3.43[1.02 to 11.54]; P = 0.047}, after adjusting for traditional cardiovascular risk factors such as euroSCORE II, hyperlipidemia, hypertension, active smoking, New York Heart Association (NYHA) class, and circulating hsCRP (as a marker of residual inflammatory risk) (Table 2 and Fig. 7C).

Table 2 Multivariate Cox regression model for survival in study 1.

Asterisk denotes the independently significant predictors; plus symbol denotes adjusted value.

View this table:


This study demonstrates that the presence of vascular IR in humans with advanced atherosclerosis, independently of systemic IR or even diabetes, results in increased vascular oxidative stress and endothelial dysfunction upon treatment with human or synthetic insulins, independently of circulating plasma glucose. This is reversed by DPP4 inhibition, which allows insulin to exert its antioxidant and vasoprotective actions, whereas the circulating DPP4/insulin balance is an independent predictor of cardiac mortality in patients with atherosclerosis (fig. S17).

Aggressive glycemic control has inconsistent effects on cardiovascular outcomes (18). The UK prospective diabetes study (UKPDS) first demonstrated that standard glycemic control (metformin versus sulfonylureas with/without insulin) reduced microvascular but not macrovascular disease risk (19). The Action in Diabetes and Vascular Disease: Preterax and Diamicron MR Controlled Evaluation (ADVANCE) trial further linked intensive glycemic control with reduced composite risk for vascular adverse events, which was, however, driven by reduced nephropathy risk (20), whereas the ACCORD trial showed no benefit of aggressive glycemic control on cardiovascular outcomes, suggesting that the pharmacological means to achieve glycemic control may be as important as the degree of control (7). On the other hand, insulin treatment as a means of glycemic control has been associated with increased risk for acute ischemic events and cardiovascular (21) or all-cause mortality (22). The Outcome Reduction with Initial Glargine Intervention (ORIGIN) and Cardiovascular Safety of Degludec Versus Insulin Glargine in Subjects with Type 2 Diabetes at High Risk of Cardiovascular Events (DEVOTE) trials also found no beneficial effect of insulin glargine or insulin degludec on cardiovascular risk (5, 6), highlighting the need to understand the direct effects of insulin on the vasculature.

Oxidative stress is a key feature of atherogenesis (23) and of vascular complications in diabetes and IR (2), and it has been proposed to play a role in endothelial IR (24). In vitro and animal studies have demonstrated that, under physiological conditions, insulin exerts antioxidant and vasodilatory effects via Akt-mediated increase in NO bioavailability in the vasculature (9, 25). On the other hand, endothelium-specific IR has been associated with endothelial dysfunction in mouse studies (26). Furthermore, hyperinsulinemia such as that observed in insulin-resistance states has been shown to cause endothelial dysfunction in vivo in humans, which was reversed by vitamin C, an antioxidant, suggesting an underlying role of oxidative stress in this effect of insulin (27). Our work strengthens this body of evidence by demonstrating that exogenous insulin treatment has a class effect characterized by NADPH oxidase activation, eNOS uncoupling, endothelial dysfunction, and inflammatory pathway activation in human vessels from patients with atherosclerosis. This abnormal vascular response to insulin results from a default activation of Erk1&2 rather than Akt insulin signaling in the human vascular cells of patients with atherosclerosis, which may compromise the vascular benefits of insulin treatment with regard to systemic serum glucose lowering. The association between cardiometabolic disease and resulting features of vascular IR was further confirmed in healthy versus HFD-fed ApoE−/− mice. The underlying causes may involve nutrient overload, low-grade inflammation, or aging and warrant further investigation.

We next explored the proof of concept that vascular insulin sensitization could reverse vascular insulin responses, as implied previously by the Bypass Angioplasty Revascularization Investigation 2 Diabetes (BARI-2D) trial, where insulin-sensitizing approaches were associated with a favorable cardiovascular outcome in patients with atherosclerosis (28). DPP4 is a glycoprotein that cleaves N-terminal dipeptides from proteins such as GLP1 (29), promoting IR in obesity and diabetes (30), whereas its pharmacological inhibition is a therapeutic target in diabetes and a potential insulin sensitizer (15). However, the vascular implications of DPP4i in humans and its interactions with vascular insulin signaling are unknown.

In this study, we demonstrated that pretreatment of patients with advanced atherosclerosis with an oral DPP4i in vivo and incubation of human vessels with DPP4i ex vivo reversed vascular responses to exogenous insulin treatment, resulting in an insulin-induced improvement of vascular redox state. This is due to the ability of DPP4i to reduce IRS1 Ser307 phosphorylation, which is known to regulate the switch between the two post–insulin receptor signaling pathways in in vitro and mouse models (15), in an AMPKα2-mediated manner. Metformin, another insulin sensitizer that acts via AMPK signaling (14), did not have similar effects, which suggests that DPP4i may have more important pleiotropic effects via affecting GLP1R signaling and other pathways such as that of IL-10 (10).

Our work suggests that the effects of DPP4i are dependent on AMPKα2, partly mediated by GLP1R signaling, but independent of PKCβ inhibition, another means of insulin sensitization (16). We also show that, further to its short-term effects, DPP4i blocks proinflammatory NFκB signaling, which has been linked with molecular IR in endothelial cells (31), and this could have implications for chronic DPP4i treatment in vivo.

Clinical trials such as Saxagliptin Assessment of Vascular Outcomes Recorded in Patients with Diabetes Mellitus-Thrombolysis in Myocardial Infarction 53 (SAVOR-TIMI 53) and Cardiovascular Outcomes Study of Alogliptin in Patients with Type 2 Diabetes and Acute Coronary Syndrome (EXAMINE) have shown no benefit of DPP4i add-on antidiabetic treatment on cardiovascular complications of diabetes (32). On the other hand, a recent clinical trial showed that DPP4i administration on top of insulin treatment reduces the risk for stroke in patients with diabetes (33). In the recent Cardiovascular and Renal Microvascular Outcome Study with Linagliptin in Patients with Type 2 Diabetes Mellitus (CARMELINA) trial examining the effect of linagliptin on cardiovascular outcomes, almost 60% of participants received inulin treatment (34). However, examining the interaction of DPP4i with insulin treatment in this case would be confounded because insulin was administered in cases where glycemic control was challenging as per clinical guidelines (thus being a surrogate of more advanced diabetes states than noninsulin-treated patients). Furthermore, significantly fewer patients in the linagliptin group initiated or increased doses of preexisting insulin therapy (34), suggesting an inverse confounding association of insulin treatment and linagliptin treatment.

Our study has some potential limitations due to the nature of our clinical research population. There are some borderline demographics differences between studies 1 and 2 and patients with or without diabetes within the studies, which, although justified on the basis of the individual study objectives, may introduce background statistical noise. However, we have applied careful statistical adjustments in our observational analyses and careful matching/paired design in our mechanistic experiments, which have been further validated in cell culture and animal models. In addition, there were a relatively small number of cardiac adverse events, which is a limitation of the outcome arm of our study.

In conclusion, we show that IR is present in the vasculature of patients with coronary atherosclerosis even in the absence of diabetes or markers of systemic IR. This results in vascular oxidative stress and endothelial dysfunction in response to insulin. Pharmacological treatment with DPP4i restores “physiological” insulin signaling in human vessels, allowing insulin to improve vascular redox state and endothelial function. These results strengthen the proof of concept for the importance of vascular sensitization in diabetes and call for appropriately designed randomized clinical trials to explore the effect of combined treatment with insulin and DPP4i on cardiovascular outcomes in patients with diabetes and atherosclerosis, which could help expand the clinical benefits associated with glycemic control.


Study design

In this study, we explored the direct effects of insulin on vascular redox state and endothelial function in patients with atherosclerosis to understand why intensive glucose lowering fails to prevent the macrovascular effects of diabetes. We also investigated the ability of DPP4 inhibition to modify vascular insulin signaling in this population. In study 1, we used a cohort of 580 consecutively enrolled patients with advanced atherosclerosis, undergoing cardiac surgery, to explore the links between endogenous circulating insulin/plasma DPP4 activity and vascular redox signaling studied directly in human vessels obtained during surgery. We then explored the value of endogenous plasma insulin/DPP4 activity in predicting cardiac mortality during the 3.9 years of prospective follow-up period. In study 2, we further explored the mechanisms by which exogenous insulin treatment and DPP4 inhibition affect vascular redox signaling, using ex vivo models of human vessels (arteries and veins obtained from 94 consecutively enrolled patients undergoing cardiac surgery), as described below. These models provide unique insights to the underlying mechanisms of vascular redox regulation in humans, despite inherent limitations of in vivo translation. The mechanisms behind the findings were further explored using human primary endothelial cell culture, and causality was tested in vivo with HFD-fed ApoE−/− mice (treated with linagliptin versus vehicle, followed by insulin stimulation).

Study 1 was powered against vascular superoxide in human IMAs. We estimated that we would need 161 patients to detect a 6% difference between the bottom and top tertiles of arterial log(O2·) with a power of 0.9 and log(O2·) SD of 0.36. Similarly, we would need 157 patients to detect a 13% difference between the top and bottom tertiles of log(NADPH-stimulated O2·) with a power of 0.9 and log(NADPH-stimulated O2·) SD of 0.71. In addition, 160 patients would be needed to detect a 21% difference in log(Vas2870-inhibitable O2·) between the bottom and top tertiles with a power of 0.9 and log(Vas2870-inhibitable O2·) SD of 0.69. For the outcome analysis, power calculations suggested that with 21 cardiac deaths in 580 patients during prospective follow-up, we would have a power of 0.8 to detect an HR of 3.3 with an event probability of 0.07 and a noninferiority margin of 0.1.

In study 2, the ex vivo experiments were also powered against vascular O2· generation. We estimated that with a minimum of five pairs of samples (serial rings from the same vessel), we would be able to identify a change of the desired readout in response to an intervention [i.e., a change in log(O2·)] by 0.48 with α = 0.05, a power of 0.9, and SD for a difference in the response of the pairs of 0.25. Similarly, with five patients per group, we would be able to detect a 47% change in NADPH-stimulated O2· and a 44% change in Vas2870-inhibitable O2· with a power of 0.9 and α = 0.5. For the vasomotor studies, we estimated that with n = 5 sets of serial rings, we would be able to detect 35% change in maximum vasorelaxation and 40% change in median effective concentration, with a power of 0.9 and α = 0.05.

Study population

Study 1 included 580 prospectively recruited patients undergoing elective cardiac surgery at the John Radcliffe Hospital, Oxford University Hospitals National Health Service (NHS) Foundation Trust. During surgery, SV and IMA segments were obtained and transferred to the laboratory within 20 min from harvesting and used for superoxide (O2·) measurements and vasomotor studies. Blood samples were also collected before surgery and processed within 20 min. Patients were followed up for a mean of 3.9 ± 0.4 years, and mortality was recorded.

Study 2 included 94 patients undergoing CABG surgery at the John Radcliffe Hospital, Oxford University Hospitals NHS Foundation Trust. Vascular segments were collected and incubated with insulin with or without preincubation with a DPP4i as explained in the online supplemental material. The incubated samples were then used for O2· measurements, vasomotor studies, Western immunoblotting, and other signaling experiments.

The demographic characteristics of study 1 and 2 participants can be found in Table 1 and table S1. Study 2 had higher prevalence of diabetes and hypercholesterolemia because it was specifically designed to allow comparison of the responsiveness of human vessels from patients with and without diabetes to the study interventions (with comparable representation of patients with and without diabetes), unlike study arm 1 that included unselected, consecutive patients undergoing CABG (hence the prevalance of hypercholesterolemia and diabetes in study 1 reflect those of the general population). The use of human vessels from patients with atherosclerosis with multiple risk factors, taking standard medication for stable coronary artery disease (CAD), is a strength of this work, as any finding is directly translatable to the typical patient with atherosclerosis.

Participants in any of the two studies should satisfy all of the following inclusion criteria: (i) ability to give informed consent for participation in the study and willingness to comply with all study requirements; (ii) male or female volunteers, aged 18 years or above; and (iii) patients undergoing cardiac surgery. Exclusion criteria included any inflammatory (idiopathic or autoimmune) or infective disease (viral or bacterial disease), renal failure (on dialysis) or liver failure, active malignancy, active use of nonsteroidal or anti-inflammatory drugs, and any other significant disease or disorder, which, in the opinion of the Investigator, may either put the volunteer at risk due to participation in the study or may influence the result of the study or the volunteer’s ability to participate in the study. The protocols of the studies complied with the Declaration of Helsinki, and all patients provided informed written consent.

Follow-up for clinical outcomes

All patients were prospectively recruited in the Oxford Heart Vessels and Fat (ox-HVF) cohort that collects mortality and outcome data by linking the Office for National Statistics (ONS) data with NHS Digital, a nationwide service that collects all data from the electronic patient records available in every NHS hospital in England. Patients had provided consent, and the collected data were first stored in a secured network and then link-anonymized and analyzed. Events were recorded by the clinical care team, being the formal diagnosis for hospitalization or formal primary cause of death, given by the respective NHS hospitals for every hospital admission or outpatient visit. NHS digital is also connected with the U.K. ONS, which offers further cross-check of the mortality data and cause of death. Patients were followed up after surgery until the date of NHS Digital data collection or death. Right censoring was applied for patients who were alive at the data collection time. Follow-up time was defined as the number of days between surgery and the date of data collection (15 December 2017) or date of death. Adjudication of the cause of death was defined by three independent study investigators in a blinded way, based on the International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD-10) codes. Cardiac mortality was defined as any death because of proximate cardiac causes (chronic ischemic heart disease), corresponding to ICD-10 codes of I20 to I25 (ischemic heart diseases) and I30 to 53 (other forms of heart disease) (35).

Risk factor definition

Traditional cardiovascular risk factors were defined according to clinical guidelines and following an interview with each study participant and careful review of their medical notes. Hypertension was defined on the basis of the presence of a documented diagnosis or treatment with an antihypertensive regimen (36). Similar criteria were used for the definition of hypercholesterolemia and diabetes mellitus (37, 38). Smoking history was also assessed, and patients were grouped as never-smokers, ex-smokers (quit >1 week ago), or active smokers.


Wild-type C57BL/6 mice (strain C57BL/6, Envigo Laboratories, UK) were used for aortic tissue ex vivo insulin incubations as biological atherosclerosis-free controls. To test the in vivo ability of DPP4i to regulate vascular insulin responses in the context of cardiometabolic disease, adult (8 to 10 weeks) male C57BL6/ApoE−/− mice were fed a HFD (SDS829108 Western RD diet) and treated with either linagliptin (a DPP4i, Cayman Chemicals; 10 mg/kg in 0.5% carboxymethyl cellulose in sterile distilled water) or control (0.5% carboxymethyl cellulose in sterile distilled water) by oral gavage once daily (between 9 and 10 a.m.) for 28 days. Previous animal and human in vivo studies have established that DPP4i successfully reduce abnormally high glucose without inducing hypoglycemia; hence, continuous glucose monitoring of the circulating glucose levels was not performed. Mice were then culled by exsanguination under terminal anesthetic (isoflurane of >4% in 95% O2/% CO2), where depth of anesthesia was monitored by respiration rate and withdrawal reflexes. Aortic tissue was harvested and used for measuring vascular O2· and its sources, as well as for vasomotor myograph studies after ex vivo insulin incubations.

All animal studies were conducted with ethical approval from the Local Ethical Review Committee and in accordance with the U.K. Home Office regulations (Guidance on the Operation of Animals, Scientific Procedures Act, 1986) and were approved by the Local Ethical Review Committee. Mice were housed in a specific pathogen-free environment, in Tecniplast Sealsafe individually ventilated cages (floor area, 542 cm2) with a maximum of six other mice. Mice were kept in a 12-hour light/dark cycle and in controlled temperatures (20° to 22°C). Water and food were available ad libitum.

Blood sampling and circulating biomarker measurements

Venous blood was collected before surgery after 8 hours of fasting, and serum or plasma was isolated by centrifugation at 3000g for 15 min at 4°C. Serum glucose, insulin, and plasma hsCRP were measured as described previously (39). HOMA-IR was calculated by the formula (glucose × insulin)/405 (glucose measured in milligrams per deciliter and insulin in milliunits per liter). Serum DPP4 activity was measured by a commercial kit (BioVision) according to the manufacturer’s instructions. Serum IL-6 and TNFα were measured by the Quantikine HS ELISA (enzyme-linked immunosorbent) Human IL-6 Immunoassay (order ID: HS600C) and Quantikine HS ELISA Human TNFα Immunoassay (order ID: HSTA00E) from R&D Systems Europe Ltd., according to the manufacturer’s instructions.

Transcriptomic profiling of human vessels

The comprehensive measurement of protein coding and long intergenic noncoding RNA transcripts in patients’ samples was performed using the Human Gene-2.1 ST Array and the GeneTitan System (Affymetrix). Microarray data analyses and identification of the differentially expressed genes (DEGs) were performed using the GeneChip Expression Analysis software (version 4, Affymetrix). Pathway enrichment analysis was carried out in ConsensusPathDB-human (

Statistical analysis

Study 1 is a prospective cohort study (the ox-HVF cohort, see of consecutive patients undergoing cardiac surgery, used to test associations between endogenous insulin/DPP4 activity and parameters of vascular function. Study 2 was a mechanistic study involving cases where vascular tissue samples were harvested and used for mechanistic ex vivo experiments. The ex vivo effects of insulin and other interventions on vascular function were tested in serial vascular rings from the same patients, leading to reduced background variability and need for fewer patients in each individual experiment due to the paired design. This cohort was enriched for patients with type 2 diabetes mellitus to allow for safe mechanistic conclusions, which were consistent in both patients with and without diabetes, and this is a strength of study 2. Continuous variables were tested for normal distribution using the Kolmogorov-Smirnov test. Non-normally distributed variables were log-transformed for analysis.

In the clinical studies, continuous variables between three groups were compared using one-way analysis of variance (ANOVA), followed by Bonferroni post hoc test for individual comparisons. For the organ bath experiments, the effect of “serum insulin tertile” on vasorelaxations in response to ACh and BK was evaluated using two-way ANOVA for repeated measures (examining the effect of “Ache, BK, or SNP concentration” × serum insulin tertile interaction on “vasorelaxations”), in a full factorial model.

Sample size calculations were based on previous data from our laboratory. For the ex vivo experiments, sample size calculations were performed on the basis of our previous experience on this model (17), and we estimated that with a minimum of five pairs of samples (serial rings from the same vessel), we would be able to identify a change of log(O2·) by 0.48 with α = 0.05, a power of 0.9, and SD for a difference in the response of the pairs of 0.25. Furthermore, n = 5 sets of independent experiments are also supported by recent guidelines for biological experiments (40). Analysis of paired ex vivo mechanistic experiments was performed by Wilcoxon signed-rank tests, whereas Bonferroni post hoc corrections for individual comparisons were used as appropriate. For the ex vivo organ bath experiments using serial rings from the same vessel incubated with multiple interventions, we performed repeated measures ANOVA and paired t tests for individual comparisons, followed by Bonferroni post hoc correction for multiple testing as appropriate.

To test the cumulative association of serum DPP4 activity and serum insulin with mortality rates, we created a dichotomous categorical variable by splitting the population of study 1 into two groups, one with patients in the high tertile for both serum DPP4 activity and serum insulin and one in the intermediate/low tertiles for those two biomarkers. The combined effect of serum insulin and DPP4 activity on all-cause and cardiac mortality was then examined by multivariate Cox regression survival analysis after adjusting for all traditional cardiovascular risk factors (euroSCORE II, hyperlipidemia, hypertension, active smoking, and plasma hsCRP). We only corrected for risk factors and not the medication associated with them, to avoid overfitting collinearity errors, as per standard practice in this type of clinical studies.

With regard to microarray data processing, normalization, quality control, and differential gene expression analysis was performed with the Affymetrix Transcriptome Analysis Console (TAC 4.0) software. The statistical comparisons between treatments were performed after a repeated measures model for the individual patients. Insulin pathway enrichment analysis was carried out in ConsensusPathDB-human with DEGs (insulin-treated versus untreated controls) that displayed fold change (linear) of >1 or <−1 and P < 0.05. Gene Ontology database was used to functionally annotate DEGs. Raw data are provided in data file S1.


Materials and Methods

Fig. S1. Direct effects of human insulin and insulin analogs on endothelial function ex vivo in humans.

Fig. S2. Association between serum insulin and NADPH oxidase activity in humans.

Fig. S3. Dose-response effects of human insulin and insulin analogs on human vascular redox state.

Fig. S4. Effects of human insulin on vascular redox state in humans.

Fig. S5. Effects of insulin on proinflammatory transcriptional pathways in human arteries.

Fig. S6. Effects of human insulin on vascular redox state in wild-type mice.

Fig. S7. Effects of insulin on vascular redox state in patients with diabetes on metformin.

Fig. S8. DPP4 inhibition regulates the effects of human insulin on human vascular redox state.

Fig. S9. DPP4 inhibition has no direct superoxide (O2·)–scavenging properties.

Fig. S10. Linagliptin, a DPP4i, reverses the prooxidant effects of insulin on the vasculature of HFD-fed ApoE−/− mice.

Fig. S11. Association between circulating DPP4 activity and vascular redox state in humans.

Fig. S12. Linagliptin reverses the effect of insulin on endothelial function in HFD-fed ApoE−/− mice.

Fig. S13. DPP4 inhibition regulates the downstream signaling balance in response to human insulin in humans.

Fig. S14. The role of GLP1R and PKCβ signaling in the vascular insulin-sensitizing properties of DPP4 inhibition.

Fig. S15. Effect of DPP4 inhibition on NFκB nuclear translocation.

Fig. S16. Proinflammatory cytokines, diabetes, and arterial redox state in humans with atherosclerosis.

Fig. S17. Summary and proposed mechanism.

Table S1. Demographic characteristics of patients with and without diabetes.

Table S2. Multivariate regression analysis testing the interaction between insulin/DPP4 activity, use of statins, and plasma inflammatory biomarkers in predicting Vas2870-inhibitable superoxide (O2·) in human IMA.

Data file S1. Raw data.

References (41, 42)


Funding: This study was funded by Sanofi Aventis Deutschland GmbH, the British Heart Foundation (FS/16/15/32047 and Oxford British Heart Foundation Centre of Research Excellence), the National Institute for Health Research (NIHR), and the Oxford Biomedical Research Centre (BRC). I.A. acknowledges funding support by the Alexandros S. Onassis Public Benefit Foundation. Author contributions: I.A. conceived and performed the experiments, performed the data collection and analysis, and wrote the manuscript. I.B., G.D., S.C., and U.S. performed the experiments. L.H. contributed to patient recruitment and data analysis. N.A., E.K.O., C.P., and N.G. contributed to data analysis. M.M. contributed to data collection and analysis. A.S.A. contributed to patient recruitment and data analysis. D.T. reviewed the manuscript. A.K. contributed to manuscript review. R.S., G.K., and M.P. contributed to surgical specimen collection. M.P. contributed to surgical specimen collection. P.W. and N.T. provided the scientific expertise and experimental design support. K.M.C. was involved in the design of the study secured funding and provided the scientific support. C.A. conceived the study, secured the funding, and reviewed the manuscript. Competing interests: This study has been funded by Sanofi Aventis. C.A and K.M.C. are founders, shareholders, and directors of Caristo Diagnostics, an image analysis company. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. The microarray data are accessible on a public depository (accession number GSE147598, which can be found on the following link:

Stay Connected to Science Translational Medicine

Navigate This Article