The Low-Expression Variant of FABP4 Is Associated With Cardiovascular Disease in Type 1 Diabetes

Cardiovascular disease (CVD) impacts health and life span of individuals with type 1 diabetes, who are affected by CVD more frequently and develop more severe CVD at a younger age (1). Individuals with type 1 diabetes carry many risk factors for CVD (2). Diabetes itself is a major risk factor, but in combination with kidney complications, this risk is increased 10-fold (3,4). Furthermore, the recently observed increase in obesity in type 1 diabetes (5,6) might place even more individuals at risk for CVD. The mechanism by which obesity results in increased morbidity and mortality remains unclear. However, obesity is associated with abnormal fatty acid metabolism and secretion of multiple adipokines, which have the potential to increase morbidity and mortality (7,8). One such adipokine is fatty acid binding protein 4 (FABP4) (adipocyte FABP or aP2), a fatty acid carrier protein mainly expressed in adipocytes and macrophages but also expressed in venous and capillary endothelial cells (9), where it is required for fatty acid transport from the circulation to fatty acid–consuming tissues such as the heart (10). In humans, higher circulating FABP4 levels associate with insulin resistance and the metabolic syndrome (11) and predict the development of CVD (12) as well as future cardiovascular morbidity and mortality among those with already manifested CVD (13). Furthermore, the FABP4 gene was recently recognized as a shared risk factor in a joint genome-wide association study (GWAS) on coronary heart disease and type 2 diabetes in ∼500,000 individuals (14).

Studies in preclinical models have demonstrated that inhibition of the FABP4 protein through genetic deletion (15,16) or pharmacological inhibition (17,18) protects against the harmful effects of obesity, insulin resistance, and atherosclerosis. In humans, a functional single nucleotide polymorphism (SNP), rs77878271 (A>G), in the promoter region of the FABP4 gene, has been described to result in a clinical presentation that is strikingly similar to the phenotype of FABP4-deficient mice (19). The minor G allele of this SNP impacts the FABP4 promoter, resulting in reduced FABP4 transcription in the adipose tissue and reduced circulating triglyceride concentrations combined with lower CVD risk and protection from obesity-induced type 2 diabetes (19,20). It also reduces FABP4 transcription in carotid plaques (20) and epicardial fat tissue two- to fourfold (21). In addition to its effects on transcription, FABP4 rs77878271 was identified as the only cis-acting variant affecting circulating FABP4 protein concentrations in a large-scale GWAS of the plasma proteome with 21,758 participants (22).

While individuals with type 1 diabetes have high risk of CVD at an early age, there is increasing evidence that the pathophysiology is partially different from that of the general population (2). Furthermore, the role of FABP4, a novel drug target for CVD, in the development of complications in type 1 diabetes has not yet been evaluated. Therefore, we aimed to examine the role of FABP4 in relation to the development of diabetes complications and mortality in type 1 diabetes, focusing on the low-expression variant of FABP4 rs77878271. To this end, we genotyped rs77878271 and assessed the impact of this SNP on the risk of CVD, end-stage kidney disease (ESKD), and mortality in a cohort of nearly 6,000 individuals with type 1 diabetes.

The Finnish Diabetic Nephropathy (FinnDiane) Study is an ongoing multicenter study with the aim of identifying risk factors for late complications in type 1 diabetes. The diagnosis of type 1 diabetes was made by the attending physician at the time of onset according to national evidence-based clinical practice guidelines. Participants took part in the study by visiting their attending physician. During that visit, blood samples were drawn, urine samples collected, anthropometric data gathered by a trained nurse, and questionnaires regarding health, medications, physical activity, and eating habits completed as previously described (23). In brief, serum lipids or lipoproteins were measured centrally as previously described (24) or locally at each center using accredited methods. Blood pressure was measured twice in the sitting position, and the average of these measurements was used in the analyses. BMI was calculated as weight in kilograms divided by the square of height in meters, and overweight was defined as BMI 25.0–29.9 kg/m² and obesity as BMI ≥30.0 kg/m². Central obesity was defined as waist-to-hip ratio >0.85 for women and >0.90 for men. Staging of diabetic nephropathy (DN) was based on urinary albumin excretion rate (AER) or albumin-to-creatinine ratio in two of three timed overnight or 24-h urine collections or morning spot urine samples for albumin-to-creatinine ratio (Supplementary Table 1). ESKD was defined as ongoing dialysis or kidney transplant. “Any diabetic nephropathy” included microalbuminuria, macroalbuminuria, and ESKD, while “advanced diabetic nephropathy” included macroalbuminuria and ESKD. The estimated glomerular filtration rate (eGFR) was calculated with the Chronic Kidney Disease Epidemiology Collaboration formula (25). Furthermore, DNA samples for additional individuals with type 1 diabetes were obtained in collaboration with the Finnish National Institute for Health and Welfare. For these, information was retrieved through careful review of medical records from outpatient visits at the time of inclusion.

For this prospective study, all FinnDiane Study participants with an available DNA sample (n = 5,797) were genotyped for the FABP4 rs77878271 polymorphism with the TaqMan SNP Genotyping Assay (Applied Biosystems, Carlsbad, CA). After removal of duplicate samples (n = 66), 5,731 individuals remained. We excluded individuals with undetermined genotype (n = 306 [5.6%]), age <16 years at baseline (n = 105 [1.8%]), or diabetes onset age >40 years or if there were any indications of another diabetes type including insulin treatment not initiated within 1 year (n = 243 [4.2%]). A total of 5,077 individuals remained for the analyses.

We retrieved information on cardiovascular and cerebrovascular events from the Finnish Care Register for Health Care on 31 December 2014. We defined a coronary artery disease (CAD) event as an ICD-9 or ICD-10 code for myocardial infarction (ICD-10, I21–I23; ICD-9, 410–412) or surgical procedure code for coronary artery bypass surgery or balloon angioplasty. We defined stroke as an ICD code for ischemic or hemorrhagic stroke (ICD-10, I60–I64; ICD-9, 430–434). CVD was defined as incident CAD and/or stroke. Participants were censored at the data retrieval date of the study (31 December 2014) or death. Information on deaths was obtained from Statistics Finland until 31 December 2014 by complete linkage (100% coverage) using personal identification numbers (26). Mortality was defined as any cause of death excluding external causes of death, such as injury, poisoning, violence, accidents, and self-harm (ICD-10 codes S00–T98, V01–Y98).

We sought replication in 852 adult individuals with type 1 diabetes from the outpatient clinic at the Steno Diabetes Center Copenhagen (SDCC) in Denmark (27). Baseline examination took place from 1993 to 2001, as the original cohort was recruited for study of late diabetes complications (28). Follow-up data on end points were retrieved from the Danish National Health Registry on 31 December 2016. Causes of death were obtained from the Danish National Death Registry (available until 31 December 2015) after tracing of the participating individuals recruited at baseline. CAD was defined as nonfatal and fatal myocardial infarction (ICD-10 codes I21–I24) or a procedural code for coronary intervention (percutaneous arterial intervention or coronary bypass grafting) (www.sst.dk). Stroke was defined as a nonfatal or fatal stroke or other manifestations of major cerebrovascular occlusive disease (ICD-10 codes I61–I66). In the current study, CVD was defined as a composite of the SDCC “CAD” and “stroke” instead of the original definition used (27). ESKD was defined as CKD stage 5 (ICD-10 code N18.5), eGFR <15 mL/min/1.73 m², a procedural code for initiation of permanent dialysis or kidney transplantation, or death because of kidney failure from CKD. Genotyping was performed with Illumina HumanCoreExome-12v1 BeadChip and imputation with a 1000 Genomes (phase 3v5) reference panel as previously described (29). SNP rs77878271 was imputed (r² = 0.84), and dosages were converted to the most likely genotypes using a 90% threshold for the genotype posterior probability.

Further replication was attempted in 3,678 Finnish individuals with type 1 diabetes from FinnGen (FinnGen data freeze 7), a large nationwide study of 309,312 individuals combining genomes and registry data. Type 1 diabetes was defined with ICD-10 code (E10.[0–9], ICD-9 250[0–8]B) as a hospital discharge diagnosis or cause of death or ICD-10 E10 for eligibility of reimbursed medications. Details on data generation and genotype imputation were previously published (30). The SNP rs77878271 was imputed (INFO 0.98, batch range 0.97–1.00). All models were adjusted for four principal components of ancestry. Information on cardiovascular and cerebrovascular events by 31 December 2019 was retrieved from the hospital discharge registry, causes of death registry, and medication reimbursement registry, with use of the same ICD-9 or ICD-10 codes used in the FinnDiane Study.

To investigate causal effects of FABP4 on CVD, we applied two-sample Mendelian randomization (MR), which uses SNPs as instrumental variables (IV) for the exposure of interest (FABP4) and summary data from different populations for the IV-exposure association and the IV-outcome association. We considered only the FABP4 rs77878271 as an IV, since it is the only cis-acting variant with strong and independent effect on circulating FABP4 (22). SNP-exposure association was extracted from summary data from the SCALLOP consortium (Systematic and Combined AnaLysis of Olink Proteins) (22), including n = 19,372 individuals of European ancestry. For SNP-outcomes associations, we used available GWAS summary data on CVD outcomes in type 1 diabetes in the GWAS catalog (www.ebi.ac.uk/gwas/). For CAD, two studies with a total of n = 8,426 individuals (31,32), including n = 4,850 from the FinnDiane Study (31), were available. No GWAS on stroke or the broader definition CVD in type 1 diabetes was available. Since we had one IV, we used the Wald ratio implemented in the TwoSampleMR R package. Associations from different studies were meta-analyzed before MR was performed.

The ethics committee of Helsinki and Uusimaa hospital districts approved the FinnDiane Study and FinnGen protocols, and the local ethics committee (Copenhagen, Denmark) approved the SDCC study protocol. FinnDiane Study and SDCC participants gave written informed consent before participation, and FinnGen participants provided informed consent for biobank research, based on the Finnish Biobank Act. Study-specific consents collected before the Finnish Biobank Act came into effect (September 2013) and the start of FinnGen (August 2017) were transferred to the Finnish biobanks after approval by Fimea (Finnish Medicines Agency), National Supervisory Authority for Welfare and Health. Recruitment protocols followed the biobank protocols approved by Fimea. All studies were performed in accordance with the Declaration of Helsinki.

The FinnGen study is approved by the Finnish Institute for Health and Welfare (permit numbers THL/2031/6.02.00/2017, THL/1101/5.05.00/2017, THL/341/6.02.00/2018, THL/2222/6.02.00/2018, THL/283/6.02.00/2019, THL/1721/5.05.00/2019, THL/1524/5.05.00/2020, and THL/2364/14.02/2020), Digital and Population Data Service Agency (permit numbers VRK43431/2017-3, VRK/6909/2018-3, and VRK/4415/2019-3), the Social Insurance Institution (permit numbers KELA 58/522/2017, KELA 131/522/2018, KELA 70/522/2019, KELA 98/522/2019, KELA 138/522/2019, KELA 2/522/2020, and KELA 16/522/2020), Findata (THL/2364/14.02/2020), and Statistics Finland (permit numbers TK-53-1041-17 and TK/143/07.03.00/2020 [earlier TK-53-90-20]). The Biobank Access Decisions for FinnGen samples and data used in FinnGen data freeze 7 are as follows: THL Biobank BB2017_55, BB2017_111, BB2018_19, BB_2018_34, BB_2018_67, BB2018_71, BB2019_7, BB2019_8, BB2019_26, BB2020_1, Finnish Red Cross Blood Service Biobank 7.12.2017, Helsinki Biobank HUS/359/2017, Auria Biobank AB17-5154 and amendment #1 (17 August 2020), Biobank Borealis of Northern Finland_2017_1013, Biobank of Eastern Finland 1186/2018 and amendment 22 §/2020, Finnish Clinical Biobank Tampere MH0004 and amendments (21.02.2020 and 06.10.2020), Central Finland Biobank 1-2017, and Terveystalo Biobank STB 2018001.

We evaluated differences in continuous variables between genotypes of rs77878271 using ANOVA (normally distributed variables) or the Kruskal-Wallis test (nonnormally distributed variables) and differences in categorical values using Fisher exact test. We calculated the P value for Hardy-Weinberg equilibrium (HWE) deviations using the exact tests of HWE (33). We evaluated FABP4 rs77878271 associations with end points using Cox proportional hazards models. To estimate “lifetime” effect, we used age as an underlying timescale, which we calculated as years from birth date to event date or death date/data registry retrieval date (Supplementary Table 2). To be able to adjust for confounders, we used follow-up years from the baseline examination as the timescale, which was calculated as years from the baseline examination date until event date or death date/data registry retrieval date. In parallel with the Cox regression analyses, we visualized the results by estimating Kaplan-Meier curves using age as the timescale and grouped by FABP4 rs77878271 genotypes (AA/AG/GG). We used an additive genetic model for rs77878271 (AA = 0, AG = 1, GG = 2, encoded as a continuous variable) unless stated otherwise. We used R software (version 4.0.2) for all analyses. We evaluated time-dependent effects using the survSplit() function (survival R package) to divide the time variable (age) into time groups (tgroup). We meta-analyzed individual results using an inverse-variance fixed-effects model in the R package meta and performed post hoc power calculations using information in Supplementary Table 2 and R package survSNP. P values <0.05 were considered significant.

The FinnGen data may be accessed through Finnish Biobanks’ FinBB portal (www.finbb.fi). Any other data sets generated or analyzed for this study are not publicly available; sharing individual-level phenotypic data is not permitted, in relation to local legislation and study participant consent.

Minor allele frequency (MAF) of FABP4 rs77878271 was 6.0% in the FinnDiane Study. At baseline, there was no difference in sex, age, diabetes duration, diabetes onset age, cardiovascular risk factors, laboratory measures, and microvascular complications between carriers of the FABP4 rs77878271 genotypes AA, AG, and GG (Table 1). There was, however, a difference in the use of angiotensin II receptor blockers (ARBs) (P = 0.0001) and lipid-lowering medication (P = 0.004) among those with FABP4 rs77878271 genotypes; a higher proportion of those with the low-expression GG genotype were on lipid-lowering therapy and ARBs. Since the rs77878271 G allele has been associated with reduced triglyceride concentrations and lower total cholesterol, and as the use of lipid-lowering medication was higher in those with the GG genotype in this study (Table 1), we also tested the association of rs77878271 with all available lipids after adjustment for age, sex, duration, BMI, and lipid-lowering medication but found no significant associations (Supplementary Table 3).

Since presence of DN may impact clinical variables at baseline, we tested the FABP4 rs77878271 association with baseline variables separately in FinnDiane Study participants with DN as well as in those without any sign of DN. In the participants with any DN, the low-expression G allele was associated with a decrease of 0.17 mmol/mL in triglycerides (P = 0.018) (Fig. 1A). On the other hand, among the participants with normal AER, the G allele was also associated with worse glycemic control (HbA_1c) (P = 0.017) (Fig. 1B).

To obtain age-specific cumulative incidences for each end point, we evaluated the effect of FABP4 rs77878271 on the risk of CVD, stroke, ESKD, and death, using age as the timescale. Kaplan-Meier curves for carriers of genotypes AA, AG, and GG demonstrated a difference in cumulative incidence of CVD (P = 0.0071) (Fig. 2) and CAD alone (P = 0.023) (Supplementary Fig. 1) but not of stroke (P = 0.097) (Supplementary Fig. 1). The probability of a CAD event by age 60 years among participants with GG genotype was 41.1% (95% CI 0.0–65.7) vs. 26.9% (95% CI 21.1–32.2%) for AG and 24.9% for AA (95% CI 22.9–26.8%), respectively. The probability of a CVD event for participants with rs77878271 GG, AG, and AA by age 60 years was 54.0% (95% CI 6.8–77.3%), 36.0% (95% CI 29.8–41.7%), and 31.1% (95% CI 29.0–33.1%). There were no differences in ESKD and mortality (Supplementary Fig. 1). We then calculated hazard ratios (HRs) for all outcomes (Table 2), using age as a timescale. Each copy of the FABP4 rs77878271 G allele additively increased the risk of stroke by 30% (P = 0.038), CAD by 27% (P = 0.014), CVD by 29% (P = 0.003), and mortality by 22% (P = 0.039) in the sex- and diabetes onset age–adjusted model. The FABP4 rs77878271 was not associated with increased risk of ESKD.

To adjust for other confounders, we evaluated the effect of FABP4 rs77878271 on the risk of incident stroke, CAD, CVD, ESKD, and mortality, using follow-up time from the baseline visit as the timescale in a Cox proportional hazards model. After we adjusted for sex, age, and diabetes duration at baseline, the FABP4 rs77878271 G allele additively increased risk of stroke by 37% (HR 1.37, P = 0.048) and CVD by 31% (HR 1.31, P = 0.02) (Fig. 3, model 1). We adjusted the model further for use of lipid-lowering medication and ARBs (Fig. 3, model 2), since these differed among FABP4 rs77878271 carriers at baseline (Table 1), after which similar but somewhat higher risk estimates were observed for stroke (HR 1.48, P = 0.01), CVD (HR 1.36, P = 0.006), and mortality (HR 1.28, P = 0.02) but not for CAD or ESKD (Fig. 3, model 2). The associations with stroke and CVD remained even after addition of baseline nephropathy stage (Fig. 3, model 3) to model 1.

We sought replication of our findings regarding stroke, CAD, CVD, and mortality in 852 adult Danish individuals with type 1 diabetes from the SDCC cohort (see Supplementary Table 4 for clinical characteristics). Further replication was sought in 3,678 Finnish individuals with type 1 diabetes from FinnGen. The MAF of FABP4 rs77878271 was 2.4% in SDCC and 6.1% in FinnGen, and the SNP was in HWE in both cohorts (Supplementary Table 2). FABP4 rs77878271 was not associated with any of the individual outcomes in SDCC (Supplementary Table 5) or in FinnGen (Fig. 4). Nevertheless, the magnitude and direction of the results for rs77878271 were mostly consistent with those from the FinnDiane Study. Furthermore, in the meta-analysis of SDCC and FinnDiane Study associations, the low-expression G allele of rs77878271 increased the risk of stroke (P = 0.04), CAD (P = 0.006), and CVD (P = 0.003) by 26%, 26%, and 17%, respectively, without significant heterogeneity (I² = 0–51%, P > 0.05) (Fig. 4).

We investigated the causal effect of FABP4 on CAD using two-sample MR. In line with our observational findings, the MR analysis revealed that one SD unit decrease in FABP4 causally increased the odds of CAD 2.4-fold (causal odds ratio 2.40, 95% CI 1.18–5.08, P = 0.016).

To understand the background behind our contradictory finding, we performed additional sensitivity analyses. As the events started to increase at ∼40 years of age, based on the Kaplan-Meier curves (Fig. 1 and Supplementary Fig. 1), we evaluated the age-dependent effects by dividing the follow-up time (age) into two different tgroups (greater than and less than 40 years) and calculating the HRs for each tgroup. These time-dependent analyses revealed that the rs77878271 low-expression G allele increased the risk of all end points except ESKD exclusively when followed to ages >40 years (Fig. 5) but not when follow-up time was limited to ages <40 years.

Next, we performed interaction analyses to evaluate whether the association between FABP4 rs77878271 and the outcomes is potentially modified by other factors, such as DN, obesity, diabetes onset age, diabetes duration, and HbA_1c, by adding these variables as an interaction term to model 2 and using time from baseline as the timescale. As it has been suggested that circulating FABP4 is cleared through the kidneys (34,35), and since renal clearance can be reduced in kidney disease, we first examined whether the FABP4 rs77878271 associations with clinical end points were modified by DN. A significant interaction of FABP4 rs77878271 with advanced nephropathy stage was identified for CAD (P_interaction = 0.02). With stratification by advanced nephropathy versus normal AER, the FABP4 rs77878271 G allele was associated with increased risk of CAD only among individuals with normal AER (P = 0.009) (Fig. 6). Since obesity is suggested to interact with the FABP4 rs77878271 SNP, we tested whether obesity (BMI ≥30.0 kg/m²) or overweight/obesity (BMI ≥25.0 kg/m²) modified the associations with any of the end points. There was no interaction between rs77878271 and obesity, between the variant and overweight/obesity (Supplementary Fig. 2), or between the variant and BMI analyzed as a continuous variable (Supplementary Table 6). However, for individuals without obesity (BMI <30 kg/m²), the G allele increased the risk of CAD and CVD, whereas no association was present for those with obesity (BMI >30 kg/m²). The association between rs77878271 and CAD was modified by HbA_1c, and the association between the variant and ESKD was modified by diabetes duration and HbA_1c (Supplementary Fig. 3).

We found that the low-expression G allele of FABP4 rs77878271 is robustly associated with increased risk of CVD, suggesting that genetically determined low levels of FABP4 increase the risk of CVD in individuals with type 1 diabetes of both Finnish and Danish descent. Furthermore, MR, including data from cohorts of other descent as well, demonstrated that lower circulating FABP4 causally increases the odds of CAD 2.4-fold. These findings are in striking contrast to findings in the general population, where the low-expression G allele confers protection against CVD (19,20). To understand the background of this unexpected association, we performed several sensitivity analyses, including interaction analyses, where we tested the modifying effect of several CVD risk factors and analyzed rs77878271 as a time-dependent variable. However, the paradoxical effect remained. Notably, a recent study in mice with streptozotocin-induced diabetes and genetic Fabp4 deficiency showed, in contrast to prior expectations, that genetic blockage of Fabp4 aggravates cardiac contractile dysfunction instead of alleviating it (36). In nondiabetic mice, genetic Fabp4 deficiency reduces the uptake of free fatty acids in the heart in combination with a compensatory, beneficial increase in glucose uptake to meet energy demands of the heart (10). In diabetes, such compensatory glucose uptake also occurs, but the energy yield from glucose is reduced as a consequence of insulin deficiency (36). This results in detrimental energy insufficiency in the diabetic heart because of reduced energy yield from glucose in combination with genetic FABP4 deficiency, which reduces the energy yield from free fatty acids (36).

Since the first description of FABP4 as a circulating protein (37), studies regarding serum FABP4 have emerged (38). Although serum FABP4 studies in type 1 diabetes have been scarce, higher FABP4 serum levels have been associated with elevated preeclampsia risk (39,40), higher BMI (41), and worse glycemic control (42) and have been suggested to be involved in autoimmune destruction of β-cells in type 1 diabetes (43). In this study, we did not find any association between the FABP4 low-expression G allele and BMI, but in participants with normal AER, the low-expression G allele was associated with higher HbA_1c (Fig. 1). In type 2 diabetes, higher serum FABP4 levels have been linked to clinical outcomes related to worse cardiometabolic health (44). In addition, serum FABP4 has been proposed as a novel biomarker for kidney complications in type 2 diabetes due to its independent associations with DN stage (45) and with kidney function decline (46). Whether lowering serum FABP4 concentrations in individuals with diabetes would result in lower risk of kidney complications is difficult to determine based on observational data. As the kidney is the key organ for the clearance of the circulating FABP4 protein, impairment of kidney function will naturally result in markedly increased serum FABP4 levels (35), complicating any conclusions regarding causality. In this study, using genetics, we did not detect an association between FABP4, proxied by the low-expression G allele, and ESKD in type 1 diabetes. However, our interaction analyses with diabetes duration revealed that the effect of the low-expression G allele on ESKD is evident only after long duration of diabetes (Supplementary Fig. 3).

The cohorts in the current study differ from the general population in several ways. By default, the prevalence of kidney disease is considerably higher among patients with type 1 diabetes than among the general population (47). Still, even among participants with normal AER, who are more comparable with the general population, the rs77878271 G allele was related to increased risk of CAD. The cohorts also differ in other ways, besides the prevalence of kidney disease. Lack of endogenous insulin production, which is the case in type 1 diabetes, could complicate the relationship between FABP4 and the risk of complications in diabetes. Furthermore, even though the prevalence of obesity in type 1 diabetes has dramatically risen during recent decades (6), the prevalence of obesity in the FinnDiane Study was still lower than that in the general population (9.4%) (Supplementary Table 4), and the interaction analysis with obesity, defined as a BMI >30 kg/m², is likely underpowered due to the low prevalence of obesity and the low MAF of the variant. As FABP4 is an adipokine implicated in obesity-induced disorders and is also causally related to BMI (22), it is possible that the deleterious effects of FABP4 on cardiometabolic health would only be detectable in obese participants, of whom there was a limited number in the current study.

Despite strong experimental evidence linking the functional promoter polymorphism in the FABP4 gene to FABP4 expression (19,20,21), it has also been suggested that the circulating FABP4 levels would not be genetically determined (48). However, in studies with individuals of Finnish descent (20), including this one, the rs77878271 G allele frequency is significantly higher (6.0%), which may explain why the effect of this variant is more easily detectable. Furthermore, in a recent large-scale GWAS of the plasma proteome (22), the FABP4 promoter rs77878271 G allele was associated with lower levels of circulating FABP4 protein homogenously across cohorts (P = 6.4 × 10⁻¹⁴, I²= 0%, P_{heterogeneity} = 0.49). Although the GWAS also included participants with diabetes, it is not a given that the same association for this SNP and circulating FABP4 is present in individuals with type 1 diabetes.

Despite significant findings for this variant in the meta-analysis, without evidence of heterogeneity, the lack of replication remains a limitation. As we were not able to find statistical significance in the individual cohorts, we calculated the post hoc power for the cohorts, which revealed that the power to detect the HR observed in the FinnDiane Study (HR 1.29) was 18.2% in the SDCC study due to the lower MAF (2%) and the smaller sample size (Supplementary Fig. 4). Although for FinnGen there was a larger sample and higher MAF (6%), the power to detect an HR of 1.29 was only slightly higher (34.3%) because of the lower CVD incidence rate (6.7%). Of note, replication was attained in the SDCC study for the original CAD definition without procedural codes (27) (HR 2.00, 95% CI 1.07–3.73, P = 0.03), but after phenotype harmonization, the finding was no longer significant (P = 0.17). However, the direction of effect was similar in all cohorts, and the MR, which also included samples not part of discovery or replication cohorts, provided further support for our findings. The need for additional confirmation in other cohorts is still needed. Finally, this study included only Nordic populations; thus, conclusions regarding other populations cannot be drawn. The strengths of this study are the carefully characterized large cohorts of type 1 diabetes and use of genetics to examine the relationship between FABP4 and outcomes. Genetically proxied FAPB4 levels represent a lifetime effect, and, as genes cannot be changed by confounders such as social class, age, or kidney function, their impact on the outcome will be independent of these factors.

In conclusion, our results suggest that a certain level of FABP4 expression seems to be required to maintain cardiovascular health in individuals with type 1 diabetes, suggesting roles as yet unexplored for this hormone in regulating the risk of diabetes cardiovascular complications.

Acknowledgments. The authors acknowledge the physicians and nurses at each FinnDiane Study center who participated in patient recruitment (Supplemental Table 8). The following biobanks are acknowledged for delivering biobank samples to FinnGen: Auria Biobank (www.auria.fi/biopankki), THL Biobank (www.thl.fi/biobank), Helsinki Biobank (www.helsinginbiopankki.fi), Biobank Borealis of Northern Finland (https://www.ppshp.fi/Tutkimus-ja-opetus/Biopankki/Pages/Biobank-Borealis-briefly-in-English.aspx), Finnish Clinical Biobank Tampere (www.tays.fi/en-US/Research_and_development/Finnish_Clinical_Biobank_Tampere), Biobank of Eastern Finland (www.ita-suomenbiopankki.fi/en), Central Finland Biobank (www.ksshp.fi/fi-FI/Potilaalle/Biopankki), Finnish Red Cross Blood Service Biobank (www.veripalvelu.fi/verenluovutus/biopankkitoiminta), and Terveystalo Biobank (www.terveystalo.com/fi/Yritystietoa/Terveystalo-Biopankki/Biopankki/). All Finnish Biobanks are members of BBMRI.fi infrastructure (www.bbmri.fi), and FinBB cooperative (https://finbb.fi/) is the coordinator of the BBMRI-ERIC operations in Finland, covering all Finnish biobanks.

Funding. This work was supported by grants from Folkhälsan Research Foundation, Wilhelm and Else Stockmann Foundation, Liv och Hälsa Society, Helsinki University Hospital Research Funds (EVO), Finnish Diabetes Research Foundation, Finnish Foundation for Cardiovascular Research, the Novo Nordisk Foundation (NNF OC0013659), Dorothea Olivia, Karl Walther och Jarl Walther Perklén Foundation, Medical Society of Finland (Finska Läkaresällskapet), and Academy of Finland (299200, 316664, and, for N.M., 331671), EFSD award supported by EFSD/Sanofi European Diabetes Research Programme in Macrovascular Complications, Waldemar von Frenckell Foundation, Jane and Aatos Erkko Foundation, Sigrid Jusélius Foundation, Nylands Nation, and the Otto Malm Foundation.

Duality of Interest. P.-H.G. reports receiving lecture honoraria from Astellas, AstraZeneca, Bayer, Boehringer Ingelheim, Eli Lilly, ELO Water, Genzyme, Medscape, Merck Sharp & Dohme (MSD), Mundipharma, Novartis, Novo Nordisk, PeerVoice, Sanofi, and Sciarc and being an advisory board member of Astellas, AstraZeneca, Bayer, Boehringer Ingelheim, Eli Lilly, Janssen, Medscape, MSD, Mundipharma, Novartis, Novo Nordisk, and Sanofi. P.R. has received honoraria to SDCC for lectures and consultancy from AstraZeneca, Astellas, Boehringer Ingelheim, Bayer, Eli Lilly, Gilead Sciences, MSD, Mundipharma, Novo Nordisk, Sanofi, and Vifor Pharma. The FinnGen project is funded by two grants from Business Finland (HUS 4685/31/2016 and UH 4386/31/2016) and the following industry partners: AbbVie, AstraZeneca UK, Biogen, Bristol-Myers Squibb, Genentech, MSD, Pfizer, GlaxoSmithKline Intellectual Property Development, Sanofi US Services, Maze Therapeutics, Janssen Biotech, and Novartis Pharma AG. No other potential conflicts of interest relevant to this article were reported.

The funders had no role in study design, in the collection, analysis, or interpretation of data, in the writing of the report, or in the decision to submit the manuscript for publication.

Author Contributions. E.H.D., J.S., C.F., L.M.T., P.J.L., N.S., and P.-H.G. took part in the conception and design of the work. E.H.D, C.F., N.M., V.H., and T.S.A. acquired the data, and E.H.D. conducted the analysis for FinnDiane Study participants, N.U. for SDCC participants, and N.M. for FinnGen participants. E.H.D., J.S., C.F., N.U., N.M., L.M.T., V.H, P.R., T.S.A., P.J.L., N.S., and P.-H.G. interpreted data and contributed to the discussion. E.H.D. drafted the manuscript, which all authors revised critically for important intellectual content. All authors accepted the final version. P.-H.G. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.