Glucose-dependent insulinotropic peptide and risk of cardiovascular events and mortality: a prospective study

Aims/hypothesis Evidence that glucose-dependent insulinotropic peptide (GIP) and/or the GIP receptor (GIPR) are involved in cardiovascular biology is emerging. We hypothesised that GIP has untoward effects on cardiovascular biology, in contrast to glucagon-like peptide 1 (GLP-1), and therefore investigated the effects of GIP and GLP-1 concentrations on cardiovascular disease (CVD) and mortality risk. Methods GIP concentrations were successfully measured during OGTTs in two independent populations (Malmö Diet Cancer–Cardiovascular Cohort [MDC-CC] and Prevalence, Prediction and Prevention of Diabetes in Botnia [PPP-Botnia]) in a total of 8044 subjects. GLP-1 (n = 3625) was measured in MDC-CC. The incidence of CVD and mortality was assessed via national/regional registers or questionnaires. Further, a two-sample Mendelian randomisation (2SMR) analysis between the GIP pathway and outcomes (coronary artery disease [CAD] and myocardial infarction) was carried out using a GIP-associated genetic variant, rs1800437, as instrumental variable. An additional reverse 2SMR was performed with CAD as exposure variable and GIP as outcome variable, with the instrumental variables constructed from 114 known genetic risk variants for CAD. Results In meta-analyses, higher fasting levels of GIP were associated with risk of higher total mortality (HR[95% CI] = 1.22 [1.11, 1.35]; p = 4.5 × 10−5) and death from CVD (HR[95% CI] 1.30 [1.11, 1.52]; p = 0.001). In accordance, 2SMR analysis revealed that increasing GIP concentrations were associated with CAD and myocardial infarction, and an additional reverse 2SMR revealed no significant effect of CAD on GIP levels, thus confirming a possible effect solely of GIP on CAD. Conclusions/interpretation In two prospective, community-based studies, elevated levels of GIP were associated with greater risk of all-cause and cardiovascular mortality within 5–9 years of follow-up, whereas GLP-1 levels were not associated with excess risk. Further studies are warranted to determine the cardiovascular effects of GIP per se. Electronic supplementary material The online version of this article (10.1007/s00125-020-05093-9) contains peer-reviewed but unedited supplementary material, which is available to authorised users.


Introduction
The enteroendocrine peptide glucose-dependent insulinotropic polypeptide (GIP) and proglucagon-derived peptides, such as glucagon-like peptide-1 (GLP-1), were classically viewed as regulators of islet function, nutrient absorption, appetite and energy homeostasis [1]. The observation that the G-proteincoupled receptors, through which these regulatory peptides exert their effects, are widely expressed in the cardiovascular system has triggered a lot of interest in their translational relevance beyond metabolic control [2].
Both experimental and clinical data, such as the outcomes from the LEADER, SUSTAIN-6, HARMONY and REWIND trials, support therapeutic benefits of GLP-1 receptor agonists with regards to cardiovascular outcomes in type 2 diabetes [3][4][5][6][7]. Further, a missense variant in the gene encoding the GLP-1 receptor has been associated with protection against heart disease [8]. While the bulk of the studies published so far have focused on GLP-1, GIP has received less attention. Data from our laboratory demonstrated that fasting GIP concentrations were significantly higher in individuals with a history of cardiovascular disease (CVD) than in those without, and that GIP receptor (GIPR) gene mRNA expression is higher in the arterial wall of individuals with symptoms of CVD [9]. Moreover, a common variant in GIPR (rs10423928), which is in complete linkage disequilibrium with rs1800437, associates with increased risk of stroke in individuals with type 2 diabetes and, recently, Ussher et al. demonstrated that reduction in GIPR signalling is linked to ischaemic cardioprotection in mice [10]. Thus, evidence that GIP and/or GIPR are involved in cardiovascular biology is emerging. In light of these findings, we explored whether circulating levels of GIP (and GLP-1) are associated with cardiovascular death and total mortality risk in two large, population-based cohorts. We also performed a two-sample Mendelian randomisation (2SMR) analysis using the GIPR variant rs1800437 previously associated with features of the metabolic syndrome and CVD [11] as an instrumental variable to study the effect of increased GIP levels on coronary artery disease (CAD) and myocardial infarction. Furthermore, a 2SMR analysis in a reverse direction from CAD to GIP was performed, using 114 known genetic risk variants for CAD as instrumental variables.

Prevalence, Prediction and Prevention of Diabetes in Botnia study
The Prevalence, Prediction and Prevention of Diabetes-Botnia (PPP-Botnia) study is a population-based study in western Finland started in 2004 to obtain estimates of prevalence and risk factors for type 2 diabetes, impaired glucose tolerance, impaired fasting glucose and the metabolic syndrome in the adult population. Participants were randomly recruited from the national Finnish Population Registry to represent 6-7% of the population in the 18-75 year age range (mean age 51 ± 17 years) [12]. Altogether, 5208 individuals participated in the study (54.7% of those invited). A follow-up study was conducted between 2011 and 2015, in which 3870 (74.3%) individuals participated. After exclusion of individuals with partially missing data, 4572 individuals remained for analysis of fasting GIP and 4398 for post-challenge GIP (see electronic supplementary material [ESM] Fig. 1). The number of individuals with diabetes included in analysis was 307 at the basal visit and 284 at the re-investigation visit. Diagnosis of diabetes was confirmed from participants' records or based on fasting plasma glucose concentration ≥ 7.0 mmol/l and/or post-challenge glucose ≥ 11.1 mmol/l. The participants gave their written informed consent and the study protocol was approved by the Ethics Committee of Helsinki University Hospital, Finland.

Malmö Diet and Cancer-Cardiovascular Cohort, Sweden
Between 1991 and 1996, a prospective, population-based study, the Malmö Diet and Cancer study, was conducted in the city of Malmö, Sweden, including questionnaires, blood sample donations and anthropometrical measurements at the baseline examination (n = 30,447). All people born in the years 1926-1945 and living in Malmö were invited to participate. To study cardiovascular risk factors, a sample of the study population (n = 6103) was randomised into a substudy, the Malmö Diet and Cancer-Cardiovascular Cohort (MDC-CC) [13]. During 2007-2012, a new clinical examination was performed (n = 3734) within the MDC-CC, with the addition of OGTT [14]. A schematic description of the study population is presented in ESM Fig. 2. Fasting blood samples were collected from 3692 individuals (fasting GIP available in n = 3479). Four-hundred-and-forty-nine individuals did not perform the complete OGTT (386 with previously known diabetes, 63 for various reasons), resulting in postchallenge (2 h) blood samples available in 3243 individuals (post-challenge GIP available in n = 3070). The characteristics of non-attendees at the re-examination have been described elsewhere [13]. The participants gave their written informed consent and the study protocol was approved by the Ethical Review Board, Lund, Sweden.

Genotyping
In both cohorts, information on genotype rs1800437 was obtained from genome-wide association study data performed at the Broad genotyping facility using Illumina OmniExpressExome BeadChip v1.0 B (MDC-CC, n = 3344) or Illumina HumanExome BeadChip v1.0 (PPP-Botnia, n = 4905). The call rate was >99.9% and the SNP was in Hardy-Weinberg equilibrium in both cohorts.

Clinical assessment
PPP-Botnia Two BP recordings were obtained from the right arm of a sitting person after 30 min of rest and their mean value was calculated. If there was more than 5 mmHg difference between the two recordings, the recording was repeated. BMI was calculated as weight (kg) divided by the square of the height (m).
MDC-CC BP was obtained after 10 min of rest in the supine position. BMI was calculated as weight (kg) divided by the square of the height (m).
OGTT In both cohorts, a 75 g OGTT, the most appropriate method for the clinical assessment of glucometabolic status [14], was performed after an overnight fast. The OGTT was performed according to same standardised protocol in both cohorts (individuals with known diabetes did not undergo an OGTT).

Laboratory assays
For both PPP-Botnia and MDC-CC participants, GIP was analysed by the same laboratory using the following procedure: during OGTT, blood samples were drawn in order to analyse GIP at 0 and 120 min. Serum GIP was analysed using Millipore's Human GIP Total ELISA (Merck Millipore, Darmstadt, Germany; no. EZHGIP-54 K; minimum detection level 1.65 pmol/l, intra-and inter-assay CV 1.8-6.1% and 3-8.8%, respectively) [15].

Statistical analysis
All analyses were performed in SPSS v. 22.0 (SPSS, Armonk, NY, USA), except for the Mendelian randomisation (MR) analyses and the meta-analyses, which were performed using R software version 3.5.2 [20]. The 2SMR analyses were built using MendelianRandomization [21] and TwoSampleMR [22] packages. A two-tailed p value < 0.05 was considered significant. Skewed continuous variables were logarithmically transformed. Individuals with missing values on covariates were excluded from respective analysis. Cox regression models were used to calculate HRs for each 1 SD increment of log-transformed fasting and post-challenge GIP and GLP-1 concentrations on mortality from CVD and total mortality risk. Individuals who died from external causes were censored.   Table 2). An additional two SNPs (rs472109, rs4754698) were removed from analysis for being palindromic with intermediate allele frequencies

IV, instrumental variable; MI, myocardial infarction
As for analyses of incident non-fatal CVD, the two cohorts were analysed with different methods (Cox regression in MDC-CC, logistic regression in PPP-Botnia), since exact time to event was not known for PPP-Botnia. Because of this, and because endpoints were defined and recorded differently, no meta-analysis was performed. Further, in exploratory, crosssectional analyses for associations between GIP concentration and prevalent subtypes of CVD, logistic regression was used to calculate ORs. Model 1 (adjusted for age and sex) was used for the primary analysis and further adjusted for relevant physio- . Proportional hazard assumptions were tested using Schoenfeld residuals. Fixed-effects meta-analysis of mortality variables was performed in R using the metafor package [23].
A 2SMR was performed with fasting GIP levels as exposure variable, CAD and myocardial infarction were defined as outcome variables, and rs1800437 as the instrumental variable. We applied the Wald ratio method as statistical modelling for the 2SMR analysis with a summary data for the outcomes from CARDIoGRAMplusC4D consortium and UK Biobank. In addition, to further explore the direction of association between GIP and CAD, we carried out a reverse 2SMR analysis from CAD to f a s t i n g G I P. L o c i f r o m a m e t a -a n a l y s i s o f CARDiOGRAMplusC4D and UK Biobank [24] were used for the exposure summary data and constructing the instrumental variables. The summary data for the outcome (fasting GIP) was from the MDC-CC cohort. Out of 147 SNPs in meta-analysis of CARDiOGRAMplusC4D and UK Biobank (ESM Table 1) with p value < 5 × 10 −8 and r 2 measure of linkage disequilibrium <0.2, 116 SNPs were selected with information also in the MDC-CC (ESM Table 2). When running MR analysis, SNPs rs472109 and rs4754698 were removed as their effect alleles were ambiguous. In total, 114 SNPs were utilised to construct the instrumental variables for 2SMR from CAD to fasting GIP. The inverse variance weighted (IVW) method, which is a widely accepted approach for 2SMR analyses with several SNPs as instrumental variables, was used for the main analysis. The sensitivity analyses for the pleiotropy effect was performed using the MR Egger method [25].

Results
Detailed characteristics of the study populations are presented in Tables 1 and 2.

Total and cardiovascular mortality
Fasting GIP In Cox regression analyses, adjusted for sex and age (Model 1), each 1 SD increment of log-transformed fasting GIP concentration was associated with higher total mortality risk in both cohorts. To determine the extent to which this was mediated by known risk factors for CVD we further adjusted the analyses for BMI, FPG, fasting insulin, SBP, LDL-cholesterol, HDL-cholesterol and smoking (Model 2), and the associations remained significant (Table 3). In Model 3, diabetes status, lipid-lowering treatment, BPlowering treatment and educational level were included on top of the covariates in Model 2 ( Table 3). The cumulative incidence of total mortality for each quartile of fasting GIP is shown in Kaplan-Meier plots (Fig. 1a, b).
Increased fasting GIP concentration was also associated with risk of cardiovascular mortality in all models (Table 3).
Post-challenge GIP Increased GIP concentrations after a standard OGTT (post-challenge) were associated with higher risk of total and cardiovascular mortality in a meta-analysis of the two cohorts in all models. However, the post-challenge associations were driven mainly by the MDC-CC cohort (Table 3).
Sensitivity analysis To rule out the possibility that the higher mortality risk was a result of individuals being diabetic and hence contributing to a larger extent to total mortality, analyses were carried out on associations of GIP and risk of mortality, as well as cardiovascular mortality, in both cohorts excluding individuals with prevalent diabetes. The associations between GIP and mortality risk essentially remained unchanged (ESM Results, ESM Tables 3-5).

Incident non-fatal cardiovascular events
Fasting GIP Next, we analysed associations between GIP concentrations and incident, non-fatal CVD. Fasting GIP concentration was associated with incident CVD during a mean follow-up of 8.8 years in all models in PPP-Botnia. In MDC-CC, fasting GIP was not associated with higher risk of incident non-fatal CVD (Table 4).
Post-challenge GIP The post-challenge GIP concentration was associated with non-fatal incident CVD in PPP-Botnia. In MDC-CC, the associations were not significant (Table 4).
A cross-sectional, exploratory analysis of GIP concentrations and CVD prevalence is presented in ESM Results.
Fasting and post-challenge GLP-1 Corresponding analyses were performed for GLP-1 in 3625 subjects but no significant associations were observed for either fasting or post-challenge levels of GLP-1 and mortality risk, or CVD subgroups in the MDC-CC study (ESM Table 6). GLP-1 was not measured at the basal visit in the PPP-Botnia cohort.

MR analyses
We performed a 2SMR analysis by Wald ratio method between GIP levels as exposure and CAD (n = 184,305; 60,801 cases, 123,504 controls) and myocardial infarction (n = 171,875; 43,676 cases, 128,199 controls) as outcome variables in CARDIoGRAMplusC4D. The same procedure was applied using data from UK Biobank (CAD: n = 296,525; 34,541 cases, 261,984 controls) [24]. For the exposure (GIP), the initial sample was 3344 individuals of Swedish ancestry; 4905 individuals of Finnish ancestry were used as the replication sample [26]. We utilised rs1800437 as the instrumental variable (see ESM Table 7 for more details). The results show a significant association between fasting GIP and both CAD (p = 0.002) and myocardial infarction (p = 0.013), as presented in Table 5 using CARDIoGRAMplusC4D data, and significant associations between fasting GIP and CAD using UK Biobank data (p = 0.001). Further, a reverse 2SMR analysis was carried out with CAD as exposure and GIP as outcome variable (Table 5; detailed analysis in ESM Table 8). The nonsignificant 2SMR result using the IVW method (p = 0.148) shows that there is no directional association from CAD to GIP. There was no evidence of pleiotropy found through the MR Egger method (Table 5; p = 0.595). The single SNP MR estimates using each of the 114 SNPs used in the reverse MR from CAD to GIP can be found in ESM Table 9. The bidirectional MR analysis confirmed the possible direction solely from GIP to CAD.

Discussion
This observational study demonstrates that high plasma concentration of fasting GIP is associated with higher risk of total and cardiovascular mortality in two general populations. Further, using a 2SMR, we demonstrated an association between increased GIP levels and CAD.
The results from the two studied populations were generally comparable, with the most consistent effect being the association between fasting GIP concentration and mortality risk. The discrepancies found may be explained by the mean age difference between the two populations (72 years for MDC-CC vs 50 years for PPP-Botnia), resulting in fewer outcomes (mortality and cardiovascular mortality) for PPP-Botnia and there may be differences in the underlying pathology of CVD at different ages. Other potential reasons are differences in population and lifestyle, and the sources and definitions of CVD in the two studies (self-reported for PPP and register-derived for MDC-CC). However, the associations between increased GIP levels and higher risk of CAD/ myocardial infarction were confirmed using the large CARDIoGRAMplusC4D data in 2SMR analysis.
Recently, Ussher et al. showed that genetic elimination of GIPR improved survival rate and reduced adverse cardiac remodelling following experimental myocardial infarction in mice [10]. Furthermore, epidemiological studies have shown that fasting GIP concentrations are significantly higher in individuals with a history of CVD and GIPR mRNA expression is higher in the arterial wall of individuals with symptoms of CVD [9]. A suggested mediator of the possible cardiovascular detrimental effects of GIP is osteopontin (OPN) [9,27,28]. OPN regulates synthesis of extracellular matrix and the proliferation and migration of endothelial and vascular smooth muscle cells during repair and remodelling of blood vessels. OPN also promotes inflammation and recruitment of leucocytes to the vessel wall [29]. Accordingly, plasma OPN has been associated with the presence and severity of CAD in humans [30]. Notably, GIP stimulation increases OPN expression in mouse arteries and individuals with symptomatic CVD have higher plaque expression of GIPR and OPN (also known as SPP1) mRNA. Further, GIP infusion increases plasma concentration of OPN in humans and this effect is strongest in carriers of the minor allele of the GIPR rs10423928 locus [9]. Interestingly, there is also a known CAD locus (rs46522) in the UBE2Z gene, suggested to be mediated by a nonsynonymous coding SNP (rs2291725) in the GIP gene, but the effect of this locus on GIP function and expression is still poorly understood [31,32]. While recent experimental data do not actually support a direct damaging effect of GIP on cardiac cells [10,33], recent clinical observations in obese individuals with hyperglycaemia and insulin resistance show an association of increased circulating GIP levels with biomarkers of chronic low-grade inflammation (this, in turn, might facilitate CVD) [34].
The MR associations between GIP and CAD shown in our study indicate a direct role for GIPR signalling in the pathways leading to these endpoints, although we cannot conclude that all of the risk increase observed in this study is due to direct effects of GIP on the cardiovascular system. The risk increase for death may, as an example, be mediated by unhealthy fat distribution, independent of insulin levels, that is associated with higher GIP release, or by promotion of obesity [35,36]. However, our analyses were adjusted for BMI, implicating other pathways. Another possibility is that the association is mediated by effects on glucose homeostasis and risk of diabetes. This was addressed by adjusting for fasting glucose and insulin values in Model 2 and addition of diabetes status in Model 3. We also did a set of analyses wherein we excluded all diabetic individuals (prevalent and incident diabetes cases) in the MDC-CC cohort (ESM Tables 3-5), with associations between GIP and mortality risk essentially unchanged. In the PPP-Botnia cohort, the diabetes status of individuals who did not attend the follow-up visit could not be determined. Instead, we analysed risk of incident and total CVD in individuals who were normoglycaemic both at baseline and at follow-up and found that all associations remained in this smaller subset (ESM Table 4). A third possibility is that part of the associations could be due to unmeasured covariates.
The LEADER, SUSTAIN-6, HARMONY and REWIND studies [3][4][5][6][7] found lower rates of cardiovascular events among high-risk individuals with type 2 diabetes treated with the GLP-1 analogues liraglutide, semaglutide, albiglutide and dulaglutide, respectively, vs placebo. In our study, neither fasting nor post-challenge GLP-1 concentrations were associated with the risk of CVD or death, nor did we find any protective effects of GLP-1 on mortality and CVD risk. This discrepancy could be due either to the different populations studied (e.g. the PPP-Botnia and MDC-CC are population cohorts consisting of only 5.9% and 4.4% individuals with diabetes, respectively, in contrast to the LEADER and SUSTAIN trial in which only diabetic individuals were studied) or, even more likely, to different concentrations of GLP-1 as the cardioprotective effects of GLP-1 agonists/analogues demonstrated earlier are attributed to pharmacologically induced, supraphysiological levels of GLP-1 in contrast to the normal, physiological GLP-1 levels in our study.

Strength and limitations
The use of well-characterised, prospective cohorts with many participants and a relatively long follow-up time is a significant strength of the current study. Further, we used nationwide registers with 100% coverage and high accuracy. We could not completely exclude confounding effects of unmeasured covariates linked to GIP levels but tried to minimise confounders by adjusting for relevant risk factors. Another strength of this study is the demonstration of an effect of a functional genetic variant in GIPR on CAD/myocardial infarction using the MR approach, suggesting an involvement of the GIP signalling pathway in the pathogenesis of CAD.
We acknowledge that the MR analysis has limitations such as horizontal pleiotropy. To improve the reliability of our GIP to CAD/myocardial infarction MR analysis, we considered the possible confounding phenotypes and tested for their association with our instrumental variable rs1800437, which is associated with insulin secretion [26,28,37], BMI and other related phenotypes (ESM Tables 10 and 11). These phenotypes are likely to mediate at least some of the association between the genetic variant in GIP and CAD (vertical pleiotropy). Because of this, and because the genetic variant affects both the concentration of GIP and the expression and function of its receptor (horizontal pleiotropy), the MR effect size estimates should be interpreted with caution. However, using a genetic variant in GIPR greatly strengthens the evidence that the association is due to GIP signalling, since there are no known alternative ligands for the GIPR. For the reverse MR, there was no evidence of a pleiotropic effect based on MR Egger analyses (Table 5). Furthermore, the two cohorts (PPP-Botnia and MDC-CC) differ regarding the mean age of the participants (those in PPP-Botnia were younger), eventrate and how endpoints were collected, possibly explaining the discrepancies in the results presented. Finally, our data was collected in two Nordic regions, which limits the applicability to other populations.

Conclusion
In two prospective, community-based studies, elevated levels of GIP were associated with greater risk of all-cause and cardiovascular mortality within 5-9 years of follow-up, whereas GLP-1 levels were not associated with excess risk. Further studies are needed to determine the cardiovascular effects of GIP per se.
Acknowledgements Open access funding provided by Lund University. The Knut and Alice Wallenberg foundation is acknowledged for generous support.
Data availability The data that support the findings of this study are available upon request from the Steering Committee of the Malmö Diet and C a n c e r s t u d y b y c o n t a c t i n g i t s c h a i r , O . M e l a n d e r (olle.melander@med.lu.se). Restrictions apply to the availability of these data, which were used under licence for the current study, and so they are not publicly available due to ethical and legal restrictions related to the Swedish Biobanks in Medical Care Act . The funders of the study had no role in study design, data collection, data analysis, data interpretation or writing of the report. AJ, MM, EA and LG had full access to the data in the study and had final responsibility for the decision to submit for publication.

Duality of interest
The authors declare that there is no duality of interest associated with this manuscript.