Background Apart from several established clinical risk factors for atrial fibrillation (AF), a number of biomarkers have also been identified as potential risk factors for AF. None of these have so far been adopted in clinical practice.
Objective To use a novel custom-made proteomics chip to discover new prognostic biomarkers for AF risk.
Methods In two independent community-based cohorts (Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) study (978 participants without AF, mean age 70.1 years, 50% women, median follow-up 10.0 years) and Uppsala Longitudinal Study of Adult Men (ULSAM) (n=725, mean age 77.5 years, median follow-up 7.9 years)), ninety-two plasma proteins were assessed at baseline by a proximity extension assay (PEA) chip. Of those, 85 proteins showed a call rate >70% in both cohorts.
Results Thirteen proteins were related to incident AF in PIVUS (148 events) using a false discovery rate of 5%. Of those, five were replicated in ULSAM at nominal multivariable p value (123 events, N-terminal pro-B-type natriuretic peptide (NT-pro-BNP), fibroblast growth factor 23 (FGF-23), fatty acid-binding protein 4 (FABP4), growth differentiation factor 15 (GDF-15) and interleukin-6 (IL-6)). Of those, NT-pro-BNP and FGF-23 were also associated with AF after adjusting for established AF risk factors. In a prespecified secondary analysis pooling the two data sets, T-cell immunoglobulin and mucin domain 1 (TIM-1) and adrenomedullin (AM) were also significantly related to incident AF in addition to the aforementioned five proteins (Bonferroni-adjustment). The addition of NT-pro-BNP to a model with established risk factors increased the C-statistic from 0.605 to 0.676 (p<0.0001).
Conclusions Using a novel proteomics approach, we confirmed the previously reported association between NT-pro-BNP, FGF-23, GDF-15 and incident AF, and also discovered four proteins (FABP4, IL-6, TIM-1 and AM) that could be of importance in the development of AF.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
The prevalence and incidence of atrial fibrillation (AF) is gradually increasing worldwide, with severe consequences on global public health.1 Recently, a large-scale study with pooled data from three cohorts with replication identified age, race, height, weight, systolic and diastolic blood pressure, current smoking, use of antihypertensive medication, diabetes, and history of myocardial infarction and heart failure as the most important clinical risk factors for AF.2
Apart from these established clinical risk factors, other biomarkers have also been identified as potential risk factors for AF. N-terminal B-type natriuretic peptide (NT-pro-BNP)/BNP,3 ,4 troponins,5 white blood cell count,6 low testosterone levels in men,7 low high-density lipoprotein cholesterol and high serum triglycerides,8 anaemia and renal failure,9 advanced glycation end-products and their receptors,10 mid-regional proatrial natriuretic peptide and C-reactive protein11 have all been identified as biochemical AF risk factors, but none of those have been adopted in clinical practice.
We have recently contributed to the development of a custom-made proteomics chip based on the proximity extension assay (PEA) technology that is capable of simultaneously measuring 92 proteins selected to potentially be of importance for cardiovascular (CV) disease. The PEA is based on the innovation that each protein is evaluated by a complimentary pair of antibodies to which unique partially complementary oligonucleotides are coupled. The binding of the two antibodies with oligonucleotides to the protein enables quantification of the concentration of the proteins by quantitative real-time PCR. This dual antibody/DNA approach minimises the risk of cross-reactivity and, therefore, many proteins could be assessed simultaneously by the chip.
In the present study, we evaluated those proteins as potential risk markers for incident AF in two different community-based cohorts of elderly subjects, the Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) study and Uppsala Longitudinal Study of Adult Men (ULSAM). In our primary analysis, we used a classic discovery (PIVUS)/replication (ULSAM) approach in order to limit the risk of spurious associations. In a predefined, secondary explorative effort, we also performed an analysis based on pooled data from the two studies.
The PIVUS study
Eligible subjects were 70 years of age and living in the city of Uppsala, Sweden. Subjects were chosen at random from the Swedish Total Population Register. In total, 1016 individuals out of 2025 invited took part in the investigation (50.1%).12 The investigation took place between April 2001 and June 2004. Reinvestigations (including ECG) were performed after 5 and 10 years.
ULSAM is a longitudinal population-based study including men born between 1920 and 1924 in Uppsala County, Sweden, being invited for the first time at age 50 years (n=2322) between 1970 and 1974.13 The participating men were thereafter invited to attend examinations at age 60, 70, 77, 82 and 88 years. The present study used the 77 years investigation as baseline (conducted in 1997–2001), since the proteomic chip was analysed in samples collected at this examination cycle. Of 1398 invited men, 838 (60%) participated. Of these, 172 were excluded due to lack of plasma for proteomics analysis. A reinvestigation (including ECG) was performed at the age of 82 years.
The ethics committee of Uppsala University approved the studies, and all subjects gave their informed prior consent.
All samples were collected in the morning, after an overnight fast. Standard laboratory techniques were used to measure lipid variables and fasting blood glucose. Waist circumference was measured at the umbilical level. Blood pressure was measured in the supine position after 15 min rest. A12-lead ECG was performed for analysis of AF. Body mass index (BMI) was calculated as weight divided by the square of height.
Cystatin C-based glomerular filtration rate (GFR) was assessed by a cystatin C immunoassay from Gentian (Moss, Norway) on Architect ci8200 (Abbott Laboratories, Abbott Park, Illinois, USA).
Diabetes was defined as a history of diabetes or a finding of fasting plasma glucose ≥7.0 mmol/L at the examination. Myocardial infarction, stroke and heart failure were assessed by medical records, the Swedish in-hospital registry and the Swedish cause of death registry, in parallel with the collection of data on prevalent and incident AF.
The Olink Proseek Multiplex Cardiovascular 96×96 kit was used to measure proteins in plasma by real-time PCR using the Fluidigm BioMark HD real-time PCR platform as described earlier.14 Of the wells, 1 is a negative control while 3 are positive controls (spiked in interleukin-6 (IL-6), IL-8 and vascular endothelial growth factor (VEGF)-A) resulting in 92 measured proteins. Each sample includes two incubations, one extension and one detection control used to determine the lower detection limit and to normalise the measurements. The resulting relative values obtained are log 2-transformed for subsequent analysis.
Values below detection limit (LOD) were replaced by LOD/20.5. Further details regarding LOD, reproducibility and validations are given at Olink’s webpage (http://www.olink.com/data-you-can-trust/validation/).
In the QC/QA process, we deleted proteins with a call rate of <70%. Six such proteins were found in the PIVUS study (β-nerve growth factor, SIR2-like protein (SIRT2), IL-4, brain natriuretic peptide (BNP), nuclear factor (NF)-κ-B essential modulator (NEMO) and melusin (ITGB1BP2)). In the ULSAM, the corresponding number was 3 (ITGB1BP2, IL-4 and heat shock protein 27). Thus, data on 85 of the proteins were included in the statistical analyses. Seven subjects in the PIVUS study and one in the ULSAM were removed due to a high number of proteins with measurements <LOD. The data were adjusted for plate to remove any influence of drift in measurements between plates.
The analysis of the proteins took place in August 2014.
During the follow-up periods, incident cases of AF (International Code of Diagnosis (ICD)-10 code I48) were collected and validated (by AF at the ECG) by medical records, the Swedish in-hospital registry and the Swedish cause of death registry. Also, cases discovered at the ECG at ages 75 and 80 years in PIVUS and at age 82 years in ULSAM were included as incident cases.
In PIVUS, the censor date for the follow-up was 10 years from the baseline investigation, thus ranging from 2011 to 2014. In ULSAM, the censor date was 31 December 2008.
Kaplan-Meier curves regarding total mortality over the follow-up periods in the two cohorts are given in online supplementary figure S1.
Prevalent cases of AF (ECG diagnosis) at baseline in PIVUS and ULSAM were excluded from the analyses.
For the primary analysis, the PIVUS study was used as the discovery sample and ULSAM for replication. For discovery, Cox proportional hazard analyses were performed for each the 85 proteins adjusting for age and sex. The proteins showing a false discovery rate (FDR)<0.05 were taken further to Cox proportional hazard analyses in the replication sample. FDR was calculated according to the original version of Benjamini and Hochberg from 1995. At the replication step, two levels of adjustment were calculated, one with adjustment for age and sex only and another adjusted for multiple risk factors (age, sex, systolic blood pressure, BMI, diabetes mellitus, prior myocardial infarction, prior heart failure and smoking). A nominal p value of <0.05 for the multiple-adjusted analysis was considered as a valid replication in ULSAM.
In the predefined secondary analysis, we pooled the two data sets and Cox proportional hazard analyses were performed for each of the 85 proteins adjusting for age, sex and study. The proteins showing a p value <0.000588 (Bonferroni adjustment for 85 tests) were also analysed following adjustment for multiple risk factors. We also tested the interaction between the protein levels and study (protein*study) regarding incident AF.
In this pooled data set, we also used Harrel's C-statistics analyses in order to investigate whether the addition of the novel risk markers to a model with established risk factors improved the model discrimination.
STATA V.14 was used for calculations (StataCorp LP, College Station, Texas, USA).
In PIVUS, 148 incident events of AF occurred during a median follow-up of 10.0 years (range 0.3–10.9 years) in 978 individuals free from known AF at baseline at age 70 years (incidence rate 16.9/1000 person years at risk).
In ULSAM, 123 incident AF events occurred during 7.9 years of follow-up (range 0.08–11.2) in 725 subjects free of AF at baseline at age 77 years (incidence rate 23.4/1000 person years at risk). Baseline characteristics are given in table 1.
Thirteen of the proteins were related to incident AF in the PIVUS study when adjusted for age and sex using a FDR of 0.05 (see online supplementary figure S2 and table S1 for all proteins). Of those, five were replicated in ULSAM (NT-pro-BNP, fibroblast growth factor 23 (FGF-23), fatty acid-binding protein 4 (FABP4), growth differentiation factor 15 (GDF-15), and IL-6). Of those, NT-pro-BNP and FGF-23 remained associated with AF also after adjustment for traditional CV risk factors (see table 2, online supplementary figure S3 and table S2 for all proteins). Further addition of GFR (mean value 93.1 (standard deviation 22.6) mL/min/1.73 m2) to the model adjusting for CV risk factors did not have any major effect of the HRs for NT-pro-BNP and FGF-23 (HR 1.56, 95% CI 1.33 to 1.83, p<0.0001 for NT-pro-BNP and HR 1.20, 95% CI 1.03 to 1.39, p=0.017 for FGF-23).
In the analysis merging the ULSAM and PIVUS data sets for a meta-analysis based on individual data, the five proteins mentioned above turned out to be the most significant ones and passing the Bonferroni-adjustment level. In this analysis, all five proteins were statistically significant also following adjustment for traditional CV risk factors (Bonferroni corrected). In addition to those five proteins, also T-cell immunoglobulin and mucin domain 1 (TIM-1) and adrenomedullin (AM) were related to incident AF with an age-adjusted and sex-adjusted p value below the Bonferroni-adjusted limit and were also significant following adjustment for traditional CV risk factors (see table 3, online supplementary figure S4 and table S3 for all proteins). Following further adjustment also for NT-pro-BNP, only IL-6 (p=0.0072) and TIM-1 (p=0.010) showed p values <0.05.
We used the merged ULSAM and PIVUS data set to evaluate if the addition of the two proteins still being significantly related to incident AF in the discover/validation approach after adjustment for CV risk factors (NT-pro-BNP and FGF-23) would add predictive power on top of these CV risk factors. The addition of NT-pro-BNP and FGF-23 to CV risk factors improved C-statistics from 0.605 (95% CI 0.567 to 0.643) to 0.676 (95% CI 0.638 to 0.714), p<0.0001. This improvement in C-statistics was almost entirely dependent on NT-pro-BNP (increment in C-statistics 0.071, p=0.0001), since addition of FGF-23 alone to CV risk factors only increased C-statistics by 0.013 (p=0.18). Further addition of the proteins discovered in the merged analysis (TIM-1 and AM) did not add any further predictive power.
The present study used a targeted proteomics approach to discover new risk markers for incident AF. Using a conservative discovery-validation approach in two separate samples of elderly subjects, NT-pro-BNP and FGF-23 were robustly associated with future AF following adjustment for traditional AF risk factors. The addition of NT-pro-BNP to a model with the established risk factors increased discrimination for AF. Furthermore, in a predefined meta-analysis based on individual data from the two studies, also FABP4, GDF-15, IL-6, TIM-1 and AM were disclosed as potential risk markers for AF.
In the present study, we used a targeted proteomics approach to explore potential new biomarkers for AF. These 92 proteins have been selected as previously being associated with atherosclerosis or other kinds of CV disease. Since a large number of tests were performed, we first used a conservative discovery-validation approach in two separate samples of elderly subject to avoid false-positive results. However, as the numbers of cases in each of the two cohorts are limited and we therefore might have a problem with false-negative findings, we also performed a prespecified secondary explorative analysis merging the two data sets. In this secondary analysis, we used a strict Bonferroni adjustment to limit the risk of type I error. It should, however, be pointed out that the findings from this secondary analysis have to be replicated in other cohorts to be considered valid.
A recent meta-analysis identified nine clinical variables as being of importance for AF risk.2 We used those clinical variables as confounders in the analysis to ensure that any impact of potentially new protein biomarkers was due to covariation with these clinical risk factors. We did, however, use BMI instead of height and weight in order to not lose too many degrees of freedom, but substituting BMI by height and weight in additional analyses did only change our major findings marginally.
Of the analysed proteins, NT-pro-BNP was by far the biomarker most closely related to incident AF. BNP is a vasodilatory peptide secreted from the myocardial ventricles in response to stress. The more stable precursor NT-pro-BNP is frequently used clinically for the detection and monitoring of patients with heart failure, but we found NT-pro-BNP to be related to AF even when prevalent heart failure was included as a confounder in the analysis. NT-pro-BNP has previously been associated with future AF in the general population,3 ,4 so in that respect our finding is a validation of previous studies. Of the seven proteins found to be of interest in the analyses using both samples, only IL-6 and TIM-1 were still significant when adjusting for NT-pro-BNP (and the traditional CV risk factors).
FGF-23 is a protein identified to have a role in mineral metabolism with primary effects on phosphate reabsorption in the kidneys.15 As such, FGF-23 levels have been found to associate with risk of future all-course mortality and CV disease in patients with chronic renal disease,16 but recent studies have pointed out that FGF-23 is of great importance for the CV system also in the general population.17 Our data are in accordance with a previous community-based study reporting the association between higher FGF23 and higher risk of AF.18 On the other hand, a recent large-scale study found no such association.19 The reason for this discrepancy is unclear.
FABPs are a family of proteins involved in lipid trafficking and metabolism in a variety of cells. Different isoforms of FABPs operate in different tissues. In adipose tissue, FABP4 is the most prominent. FABP4 is highly expressed in adipocytes and transcriptionally controlled by peroxisome proliferator-activated receptor (PPAR) γ agonists, fatty acids and insulin. FABP4 is also expressed by macrophages, and disruption of FABP4 in the atherosclerosis-prone ApoE knock-out mice induced a reduction in the atherosclerotic burden.20 Macrophage expression of FABP4 has, furthermore, been found in unstable human atherosclerotic plaques. A role for FABP4 in the development of AF has not been suggested previously.
AM is a potent vasodilatory peptide found in a variety of tissues, like the myocardium, lung, kidney and in the vasculature. Receptors for this protein are mainly found in the vasculature. Apart from the vasodilatory action, a number of other properties have been described for AM, like antiproliferative actions, antiapoptosis, reduction of oxidative stress, attenuation of the renin–angiotensin system and endothelin action, as reviewed by Wong et al.21 Elevated circulating levels of AM have been found in a number of CV diseases, like hypertension, heart failure and acute myocardial infarction.22 Furthermore, AM expression is upregulated in human unstable atherosclerotic plaques. A recent cross-sectional study showed that AM levels are increased in subjects with prevalent AF.23 In the present study, increased AM levels are also associated with incident AF, but this finding has to be replicated in other prospective studies.
IL-6 is a proinflammatory cytokine mainly secreted from immune-competent cells, and also from several CV cell types and adipose tissue. IL-6 stimulates the formation of a number of other proinflammatory mediators, such as C-reactive protein (CRP).
IL-6 levels have previously been found to be elevated in patients with AF in cross-sectional studies.24 Several studies have also shown high IL-6 levels to be associated with a poor outcome regarding AF in patients treated with cardioversion, ablation and post open heart surgery.25 However, this appears to be the first study to suggest that high IL-6 levels are associated with future AF in the community.
The TIM family of genes includes three members (Tim-1, Tim-3 and Tim-4) located at the human chromosome 5q33.2. TIM proteins are active in the regulation of T helper (Th) cell immune responses.26 As recently reviewed by Libby, Hansson and colleagues, adaptive immunity is a prominent feature in human atherosclerosis. T-cells of human lesions exhibit a T helper-1 (Th1) cell-associated cytokine secretion pattern, including interferon-γ and tumour necrosis factor.27 Also, AF is associated with an infiltration of immune cells and proteins that mediate an inflammatory response in cardiac tissue.28 How this links to elevated levels of TIM-1 as seen in the present study is not known, but the present finding might give further evidence for a role of the immune system in the development of AF.
GDF-15 is a member of the transforming growth factor-β cytokine superfamily, and its expression is increased in myocardial and vascular cells upon oxidative stress and inflammation. Our data are in accordance with a previous study where the association between GDF-15 and incident AF was statistically significant in age-adjusted and sex-adjusted models, but attenuated after further adjustment for CV risk factors.5 Interestingly, GDF-15 has been put forward as a promising risk marker for adverse events in patients with prevalent AF.29
Given the rising global incidence of AF, it will be increasingly important to better identify those at high risk for AF from those at low risk. In the aforementioned meta-analysis based on diverse populations from the USA and Europe,2 a risk model including nine variables readily available in primary care settings predicted AF with a precision that is comparable to that of the prediction of atherosclerotic CV disease. In the present study, we show that the addition of NT-pro-BNP to the previously described prediction model substantially improved the discrimination of the model. In routine clinical care, NT-pro-BNP is generally used as a diagnostic tool in heart failure patients, but if our finding is replicated in younger individuals, it is possible that NT-pro-BNP could also have utility as an AF risk marker.
Strengths of the present study are the use of a novel targeted proteomics chip being able to measure 85 proteins in two independent cohorts of elderly subjects with a fair number of incident cases of AF. A strength of the proteomics method is that a great number of selected proteins could be analysed at the same time in a limited plasma sample for a recent price per protein. It should, however, be acknowledged that not all proteins are analysed, so it is not an untargeted proteomics approach. Limitations with the technique include a lack of absolute values for the analytes, and another limitation is that the present findings have to be validated in other age and ethnic groups. We are aware that not all important CV proteins, like CRP or Troponin I or T, are included on the chip for technical or patent reasons.
In conclusion, we confirmed the previously reported association between NT-pro-BNP, FGF-23 and GDF-15 and incident AF, and also discovered four new proteins (FABP4, IL-6, TIM-1 and AM) that could be of importance in the development of AF. Our data suggest that an untargeted proteomics approach may be a valuable tool in the discovery of new biomarkers for AF risk.
What is already known on this subject?
A number of risk factors have been identified for atrial fibrillation (AF), being very similar to risk factors for other cardiovascular (CV) disorders.
What might this study add?
By the use of a novel proteomics chip, we identified proteins that were related to incident AF.
How might this impact on clinical practice?
This is a way to identify novel pathophysiological pathways for AF that might have an impact on future therapy for AF. Furthermore, N-terminal pro-B-type natriuretic peptide (NT-pro-BNP) was added to the power to predict AF.
Contributors LL take the full responsibility for the study. LL, JS, EH and JÄ collected data for the study. LL and JÄ drafted the manuscript. MS performed statistical analysis. All authors made critical revisions of the manuscript.
Funding Funding for this study was provided by Uppsala University Hospital.
Competing interests None declared.
Ethics approval The Ethics Committee of Uppsala University.
Provenance and peer review Not commissioned; externally peer reviewed.