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

Abstract

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 bypass surgery (CABG), discovering 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 analogues and human insulin triggered abnormal responses of post-insulin receptor substrate 1 (IRS1) downstream signalling 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 signalling, 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 and 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.

Introduction

Type 2 diabetes mellitus is a global health epidemic and an important driver of cardiovascular complications (1). This is believed to result from hyperglycaemia, 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 signalling and dysregulation of vascular redox signalling by affecting the activity of enzymes such as NADPH-oxidases and the coupling status of endothelial nitric oxide synthase (eNOS) (2, 4). Despite the important vascular effects of hyperglycaemia and the protective effect of standard glycemic control on cardiovascular clinical endpoints, aggressive glucose lowering with insulin analogues has failed to further improve cardiovascular outcomes in type 2 diabetes, despite achieving optimal glycemic control (5–7), suggesting that the cardiovascular benefit of glucose lowering in diabetes is limited to changes at the high end of the serum glucose range. The 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 (8), suggesting 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) (9), leading to potential activation of the PI3K/Akt pathway or the MAPK pathway (9). Akt signalling has been associated with the ability of insulin to activate eNOS and increase NO bioavailability (10). However, it is unclear how insulin vascular signalling 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 (11). It is now evident that DPP4 inhibitors ameliorate cellular insulin resistance (IR) in experimental models (11), but their effects on insulin signalling in the human vascular wall is unknown.

In this study, we explored the direct effects of human and synthetic insulins on redox signalling 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 investigated the potential role of insulin sensitization strategies in restoring physiological insulin signalling in the human vascular wall.

Results

Effects of insulin on vascular redox state in patients with atherosclerosis

We first investigated the association between circulating insulin and endothelial function in non-diabetic 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, SV) 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 (HOMA-IR<2.9) to exogenous insulin ex vivo, using both long-acting insulin analogues (M1 metabolite of glargine or degludec), 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; 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.

	Study 1	Study 2	P-value
Participants (n)	580	94
Age (years)	66.8±0.4	68.3±1.0	0.167
Males (%)	81.1	88.6	0.100
Hypertension (%)	72.1	78.2	0.300
Hyperlipidaemia (%)	77.7	93.2	<0.001
Type 2 diabetes mellitus (%)	21.7	31.8	0.042
Smoking			0.344
Active (%)	10.1	6.9
Past (%)	54.0	49.4
BMI (kg/m2)	28.5±0.2	28.5±0.4	0.928
Waist-to-hip ratio	0.98±0.01	0.99±0.01	0.715
HOMA-IR in patients without diabetes	1.16[0.73-1.83]	1.26[0.53-1.82]	0.534
Plasma hsCRP (mg/L)	1.40[0.60-4.10]	1.20[0.50-2.35]	0.549
Medication
Antiplatelet (%)	81.2	83.5	0.678
ACEi/ARBs (%)	61.9	68.3	0.456
Statins (%)	82.0	88.0	0.214
β-blockers (%)	65.9	74.7	0.127
Calcium channel blockers (%)	25.6	31.3	0.287
Insulin (%)	7.4	8.4	0.662
Oral hypoglycemics (%)	15.5	25.3	0.023

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 mean ± standard error of the mean; HOMA-IR is presented as median[25^th-75^th percentile]; P-values are calculated by Fischer’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, hsCRP).

Fig. 1. Insulin impairs endothelial function in humans with coronary atherosclerosis.

(A-C) Vasorelaxation curves of phenylephrine pre-contracted 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-E) Serial rings of human vessels were treated with and without insulin M1 (10 μM) prior to testing vascorelaxation in response to acetylcholine [D for patients without diabetes (n=6 pairs) and E for patients with diabetes (n=6 pairs)] or sodium nitroprusside-SNP [F for patients without diabetes (n=6 pairs) and G for patients with diabetes (n=6 pairs)]. *P < 0.05 vs high tertile in panel A; vs low tertile in panel B; vs control in panels D-E; NS: non-significant vs control (P > 0.05); P-values calculated by two-way ANOVA for repeated measures with (Ach dose) x (insulin treatment) interaction; data presented as mean ± 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 (IMA) and SV], evidenced by increased NADPH-stimulated O₂^.- and particularly Vas2870-inhibitable O₂^.- production from these vessels as measured by lucigenin chemiluminescence (fig. S2). Vas2870 is a pan-NOX inhibitor of NADPH-oxidases. 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-inhibtable O₂^.- (fig. S3). Human insulin and M1 glargine had that effect at concentrations ≥ 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 (12–14).

Further incubations with insulin glargine M1, used as a representative insulin analogue 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 as well as and from patients without diabetes or systemic IR, which is comparable to the in vivo situation (serum insulin median[25th-75th percentile]: 5.5[3.4-8.3] nM in Study 1; Fig. 2, A to C for SV and D to F for IMA; fig. S4, A to C for human insulin). This was due to activation of vascular NADPH oxidases as documented by the increase of NADPH-stimulated as well as the Vas2870-inhibitable O₂^.- production. These findings were replicated using 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 8h activated several pro-inflammatory 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-oxidases activity in vessels from patients with coronary atherosclerosis.

(A-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. (D-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 internal mammary artery (IMA) segments. HOMA-IR for patients without diabetes was 1.64[0.87-2.73] (median[25^th-75^th 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 auto-fluorescence as green (H-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 vs control after Bonferroni correction in panels A-F. P = 0.048 vs control for panels H-J by Wilcoxon sign rank tests. Patients with diabetes receiving an oral dipeptidyl peptidase 4 (DPP4) inhibitor were excluded from these experiments. P-values are calculated by Wilcoxon sign rank test in panels A-F, H-J; data are presented as mean ± 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-oxidases activity by 50-70%, providing a positive control for the study intervention (Fig. 2, H to J; 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 like DPP4 inhibitor (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 striking 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 (15) (fig. S7), 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 pre-incubation 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; fig. S8, A to F). DPP4i had no direct antioxidant or O₂^.--scavenging properties, as evidenced by its neutral effect on xanthine oxidase (XO)-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 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 pre-treatment 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-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 saphenous vein (SV) segments of a subgroup of patients with diabetes receiving oral DPP4 inhibitor (DPP4i) treatment in vivo. (D-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 internal mammary artery (IMA) with or without ex vivo preincubation with DPP4-i KR62436 (70 μM). (G-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-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 p47^phox subunit (n = 5) O₂^.- in human SV with or without ex vivo DPP4-I preincubation). P = 0.047 by Wilcoxon sign-rank test in panels A-C. *P < 0.05 vs control by Wilcoxon sign rank test in panels D-L followed by Bonferroni correction as appropriate; data presented as mean ± SEM.

To understand the underlying mechanisms by which insulin and DPP4 regulate NADPH-oxidases 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 p47^phox subunit of the enzymes (Fig. 3, K and L). These effects were reversed after pre-treatment of these vessels with DPP4i, in which case insulin led to GTP-Rac1 induction and prevented the membrane translocation of Rac1 and p47^phox (Fig. 3, J to L). DPP4 inhibition had no direct effects on Rac1 activation or Rac1/p47^phox 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 striking increase in 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 endothelial nitric oxide synthase (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 DPP4 inhibitor (DPP4i) pre-incubation (A, n=5-7 per intervention). (B-C) Effect of insulin (glargine M1, 10 nM) on the phosphorylation of eNOS at ser1177 in (B) human saphenous vein (SV) segments (n = 5) versus (C) human umbilical vein endothelial cells (HUVEC) in vitro (n =5) in the presence or absence of DPP4i preincubation. (D-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-H) Effect of ex vivo insulin (glargine M1, 10 nM)/DPP4i incubations on (G) endothelium-dependent acetylcholine (Ach) vasorelaxations (n = 5 – 6 per intervention) and (H) endothelium-independent sodium nitroprusside (SNP) vasorelaxations (n= 5 – 6 per intervention). *P < 0.05 vs control. P-values are calculated by Wilcoxon sign rank tests in panels A-F and by two-way ANOVA for matched observations in panels G-H. Data presented as mean ± SEM.

Given that insulin has been shown to affect eNOS activity via Akt-mediated ser1177 phosphorylation in vitro (10), 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 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 co-factor tetrahydrobiopterin (BH4), a key regulator of eNOS coupling. Indeed, 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. Indeed, 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 vs 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 vs 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 non-diabetic (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 pre-treatment with a DPP4i.

Fig. 5. The human vascular wall is characterized by vascular insulin resistance (IR), and this is reversed by dipeptidyl peptidase 4 inhibition.

(A-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 (saphenous veins, SV) in patients without diabetes and with coronary atherosclerosis. (D-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-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-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 (saphenous veins, SV) in patients with diabetes on oral DPP4 inhibitor (DPP4i) treatment. *P < 0.05; NS Non significant vs control. P-values are calculated by Wilcoxon sign rank tests in all panels. Data presented as mean ± 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. Indeed, 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; 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 by using 3-(2-Aminoethyl)-5-((4-ethoxyphenyl)methylene)-2,4-thiazolidinedione (Erk-i) 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 upon activation of Akt.

Fig. 6. Dipeptidyl peptidase 4 inhibition regulates vascular insulin signalling via restoring local insulin sensitivity in an AMP-activate kinase (AMPK)-dependent manner.

(A-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 internal mammary artery (IMA) segments. (D-F) Effect of DPP4i on insulin (glargine M1, 10 nM)-stimulated phosphorylation of (D) Akt (n = 5-7), (E) Erk1&2 (n = 5-7), and (F) the balance between the two (defined as the ratio of pAkt/pErk1&2, n = 5-7) in human saphenous vein (SV) segments. (G) Effect of Erk1&2 inhibition using 3-(2-Aminoethyl)-5-((4-ethoxyphenyl)methylene)-2,4-thiazolidinedione (70 µM) on insulin (glargine, 10nM)-stimulated Rac1 activation (n = 5-6). (H) Effect of Akt inhibition using wortmannin (100 nM) on the ability of insulin (glargine M1, 10 nM)/DPP4i combination to induce endothelial nitric oxide synthase (eNOS) ser1177 phosphorylation (n = 5-8). (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 = 5-7). (K) Consequence of AMPK pre-inhibition on the effects of insulin (glargine M1, 10 nM)/DPP4i incubations on vascular NADPH-stimulated O₂^.- (n = 5). (L) Consequence of AMPK pre-inhibition by compound C on the effect of DPP4i on vascular IRS1 Ser307 phosphorylation (n = 5). *P < 0.05 vs control. P-values are calculated by Wilcoxon sign rank tests in all panels. Data presented as mean ± SEM.

Insulin receptor substrate (IRS1) acts as a hub for insulin’s signaling and its phosphorylation at Ser307 shifts post-receptor signaling towards Erk1&2. We examined whether DPP4 inhibition modifies the responses of human vessels to insulin by targeting IRS1. Indeed, 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 AMP-activate kinase (AMPKα2), a molecule with known insulin-sensitizing properties, which has recently been linked with DPP4 signaling in in vitro models (16). We found that DPP4i directly induced phosphorylation of AMPKα2 at its activation site Thr172 (Fig. 6J). Pre-incubation 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). Finally, AMPK inhibition abolished the ability of DPP4i to rescue insulin sensitivity (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 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 beta (PKCβ) has demonstrated endothelial insulin-sensitising properties (17), 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. Indeed, DPP4i partly prevented nuclear translocation of nuclear factor kappa B (NFkB) in TNF-α-stimulated human umbilical vein endothelial cells (HUVEC) (fig. S15). Given that NFkB 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-oxidases activity (18)] 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 as well as therapeutic targets in patients with atherosclerosis.

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

(A-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, hyperlipidaemia, hypertension, NYHA class and circulating hsCRP). P-values in panels A-B are calculated by Kruskal Wallis tests. In panel C, the hazard ratio and P⁺ value presented are calculated from Cox regression after adjusting for EuroSCORE II, hyperlipidaemia, hypertension, NYHA class and circulating hsCRP. Hazzard radio is presented as HR[95% confidence intervals]. Data presented as median[25^th-75^th percentile] in panels A-B.

Diabetes has an inflammatory pathophysiological component, and we have found that it is characterized by elevated plasma inflammatory cytokines such as interleukin 6 (IL6) and tumor necrosis factor alpha (TNFα), as well as high sensitivity C-reactive protein (hsCRP). 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-oxidases activity was independent of the circulating concentration of IL6, TNFα, or hsCRP (table S2).

We next explored the predictive value of 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 (HR[95%CI]=3.431.02-11.54], P=0.047), after adjusting for traditional cardiovascular risk factors such as euroSCORE II, hyperlipidemia, hypertension, active smoking, NYHA class, and circulating hsCRP (as a marker of residual inflammatory risk) (Table 2, Fig. 7C).

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

	Predictors	HR[95% CI], P+
Cardiac mortality
	High serum DPP4/High insulin*	3.30[1.01-10.77], P=0.048
	EuroSCORE II (per SD)*	1.70[1.44-2.00], P<0.001
	Smoking	1.99[0.24-16.66], P=0.526
	Hypertension	0.93[0.29-2.93], P=0.894
	Hyperlipidaemia	1.16[0.43-3.12], P=0.769
	hsCRP (per SD)	1.03[0.99-1.06], P=0.158

DPP4: Dipeptidyl peptidase 4; hsCRP: high sensitivity C-reactive protein; HR: Hazard ratio

^*denotes the independently significant predictors

⁺adjusted value

Discussion

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 when treated with human or synthetic insulins, independently of circulating plasma glucose. Importantly, 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 (19). The UKPDS trial first demonstrated that standard glycemic control (metformin vs sulfonylureas with/without insulin) reduced microvascular but not macrovascular disease risk (20). The ADVANCE trial further linked intensive glycemic control with reduced composite risk for vascular adverse events, which was, however, driven by reduced nephropathy risk (21), whereas the ACCORD trial showed no benefit of aggressive glycemic control on cardiovascular outcomes, suggesting the pharmacological means to achieve glycemic control may be as important as the degree of control (8). On the other hand, insulin treatment as a means of glycemic control has been associated with increased risk for acute ischemic events and cardiovascular (22) or all-cause mortality (23). The ORIGIN and DEVOTE trials also found no beneficial effect of insulin glargine or insulin degludec on cardiovascular risk (6, 7), highlighting the need to understand the direct effects of insulin on the vasculature.

Oxidative stress is a key feature of atherogenesis (24) and of vascular complications in diabetes and IR (4), and it has been proposed to play a role in endothelial IR (25). 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 (10, 26). On the other hand, endothelium-specific IR has been associated with endothelial dysfunction in mouse studies (27). 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 (28). 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 signalling in the human vascular cells of patients with atherosclerosis, which may compromise the vascular benefits of 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 ageing 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 BARI 2D trial, where insulin-sensitizing approaches were associated with a favorable cardiovascular outcome in atherosclerosis patients (29). DPP4 is a glycoprotein that cleaves N-terminal dipeptides from proteins such as GLP1 (30), promoting IR in obesity and diabetes (31) whilst its pharmacological inhibition is a therapeutic target in diabetes and a potential insulin sensitizer (16). However, the vascular implications of DPP4i in humans and its interactions with vascular insulin signalling are unknown.

In this study we demonstrated that pre-treatment of patients with advanced atherosclerosis with an oral DPP4i in vivo as well as 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 signalling pathways in in vitro and mouse models (16), in an AMPKa2-mediated manner. Metformin, another insulin sensitizer that acts via AMPK signalling (15), did not have similar effects, which suggests that DPP4i may have more important pleiotropic effects via affecting GLP1R signalling as well as other pathways such as that of interleukin 10 (11).

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 (17). We also show that, further to its short-term effects, DPP4i blocks proinflammatory NFkB signalling, which has been linked with molecular IR in endothelial cells (32), and this could have implications for chronic DPP4i treatment in vivo.

Clinical trials such as SAVOR-TIMI 53 and EXAMINE, have shown no benefit of DPP4i add-on antidiabetic treatment on cardiovascular complications of diabetes (33). 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 (34). In the recent CARMELINA trial examining the effect of linagliptin on cardiovascular outcomes, almost 60% of participants received inulin treatment (35). 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 non-insulin-treated patients). Furthermore, significantly fewer patients in the linagliptin group initiated or increased doses of pre-existing insulin therapy (35), 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. Indeed, there are some borderline demographics differences between studies 1 and 2 and patients with or without diabetes within the studies which, although justified based on 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 was 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 DPP4 inhibitors restores “physiological” insulin signalling 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 DPP4-i on cardiovascular outcomes in patients with diabetes and atherosclerosis, which could help expand the clinical benefits associated with glycemic control.

Methods

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 signalling 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 signalling 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 prospective follow up period. In study 2, we further explored the mechanisms by which exogenous insulin treatment and DPP4 inhibition affect vascular redox signalling, by 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 high fat diet (HFD)-fed ApoE-/- mice (treated with linagliptin vs vehicle followed by insulin stimulation)

Study 1 was powered against vascular superoxide in human mammary arteries. We estimated that we would need 161 patients to detect a 6% difference between the bottom and top tertiles of arterial Log(O₂^.-) with power 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 power 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 power 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 power 0.8 to detect a hazard ratio of 3.3 with event probability of 0.07 and non-inferiority 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 5 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, power 0.9 and SD for a difference in the response of the pairs of 0.25. Similarly, with 5 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 power 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 EC50, with power 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 NHS Trust. During surgery, SV and IMA segments were obtained and transferred to the lab within 20 minutes from harvesting, and used for superoxide (O₂^.-) measurements and vasomotor studies. Blood samples were also collected prior to surgery and processed within 20 minutes. 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 Trust. Vascular segments and ThAT from the mediastinal region were collected and incubated with insulin with or without pre-incubation with a DPP4-i as explained in the online supplemental material. The incubated samples were then used for O₂^.- measurements, vasomotor studies, western immunoblotting and other signalling experiments.

Study 2 included 94 patients undergoing CABG surgery at the John Radcliffe hospital, Oxford University Hospitals NHS Trust. Vascular segments and ThAT from the mediastinal region were collected and incubated with insulin with or without pre-incubation with a DPP4-i as explained in the online supplemental material. The incubated samples were then used for O₂^.- measurements, vasomotor studies, western immunoblotting and other signalling 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 hypercholesterolaemia, since it was specifically designed to allow comparison of the responsiveness of human vessels from patients with and without diabetes to the study interventions, unlike study arm 1 that included unselected, consecutive patients undergoing coronary bypass surgery. The use of human vessels from patients with atherosclerosis with multiple risk factors, taking standard medication for stable 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; 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 non-steroidal 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 because of 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 National Health Service (NHS) Digital, a nation-wide 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 was first stored in a secured network and then link-anonymized and analysed. 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 UK Office for National Statistics, 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 that 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 (December 15, 2017) or date of death. Adjudication of the cause of death was defined by 3 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 due to proximate cardiac causes (chronic ischemic heart disease), corresponding to ICD-10 codes of I20-I25 (ischemic heart diseases) and I30-53 (other forms of heart disease) (36).

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 based on the presence of a documented diagnosis or treatment with an antihypertensive regimen (37). Similar criteria were used for the definition of hypercholesterolemia and diabetes mellitus (38, 39). Smoking history was also assessed, and patients were grouped as never-smokers, ex-smokers (quit >1 week ago) or active smokers.

Mice

Wild-type C57BL/6 mice (strain C57BL/6, ENVIGO labs, 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 (8-10 weeks) male C57BL6/ApoE-/- mice were fed a HFD (SDS829108 Western RD diet) and treated with either linagliptin (a DPP4i, Cayman Chemicals; 10mg/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 9-10am) for 28 days. Previous animal and human in vivo studies have established that DPP4 inhibitors 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 anaesthetic (isoflurane >4% in 95% O2/% CO2), where depth of anaesthesia 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 UK 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 IVC cages (floor area 542 cm²) with a maximum of six other mice. Mice were kept in a 12 h light/dark cycle and in controlled temperatures (20–22°C). Water and food were available ab libitum.

Blood sampling and circulating biomarker measurements

Venous blood was collected prior to surgery after 8h of fasting and serum or plasma was isolated by centrifugation at 3,000g for 15min at 4°C. Serum glucose, insulin and plasma hsCRP were measured as described previously (40). Homeostatic model assessment of systemic IR (HOMA-IR) was calculated by the formula (glucose x insulin)/405 (glucose measured in mg/dL and insulin in mU/L). Serum DPP4 activity was measured by a commercial kit (Biovision) according to the manufacturer’s instructions. Serum IL6 and TNFα were measured by the Quantikine HS ELISA Human IL-6 Immunoassay (order ID: HS600C) and Quantikine HS ELISA Human TNFalpha 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 non-coding 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 were performed using the GeneChip Expression Analysis Software (version 4, Affymetrix). Pathway enrichment analysis was carried out in ConsensusPathDB-human (http://cpdb.molgen.mpg.de/).

Statistical analysis

Study 1 is a prospective cohort study (the Oxford Heart Vessels and Fat Cohort or ox-HVF cohort, see www.oxhvf.com) 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 T2DM patients 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 3 groups were compared by using one-way ANOVA followed by Bonferoni post-hoc test for individual comparisons. For the organ bath experiments, the effect of “serum insulin tertile” on vasorelaxations in response to ACh & BK was evaluated by using two-way ANOVA for repeated measures (examining the effect of “Ach, BK or SNP concentration” x “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 based on our previous experience on this model (18), and we estimated that with a minimum of 5 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, power 0.9 and SD for a difference in the response of the pairs of 0.25. Furthermore, n = 5 sets of independent experiments is also supported by recent guidelines for biological experiments (43). Analysis of paired ex vivo mechanistic experiments was performed by Wilcoxon’s sign rank tests, while Bonferroni post-hoc corrections for individual comparisons were employed 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 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, hyperlipidaemia, hypertension, active smoking) and plasma hsCRP). We only corrected for risk factors and not the medication associated with them, in order to avoid overfitting collinearity errors, as per standard practice in this type if clinical studies.

With regards to microarray data processing, normalisation, quality control and differential gene expression analysis was performed with the Affymetrix Transcriptome Analysis Console (TAC 4.0) Software. The statistical comparisons between treatments was done following a repeated measures model for the individual patients. Insulin pathway enrichment analysis was carried out in ConsensusPathDB-human with differentially expressed genes (DEGs) (Insulin-treated vs. untreated-controls) that displayed fold change (linear)>1 or <-1 and a p-value < 0.05. Gene Ontology database was used to functionally annotate DEGs. Raw data are provided in data file S1.

Single-sentence summary.

Insulin causes vascular oxidative stress in human atherosclerosis that is reversed by restoring vascular insulin sensitivity using a dipeptidyl-peptidase-4 inhibitor.