Collagen Biomarkers and Incidence of New Onset of Atrial Fibrillation in Subjects with No Overt Cardiovascular Disease at Baseline: The Multi-Ethnic Study of Atherosclerosis

Abstract

Background:

Atrial fibrosis is a hallmark of structural remodeling in atrial fibrillation (AF). Plasma Procollagen Type III N-Terminal Propeptide (PIIINP) reflects collagen synthesis and degradation while Collagen Type I Carboxy-Terminal Telopeptide (ICTP) reflects collagen degradation. We aimed to study baseline plasma PIIINP and ICTP and their associations with incident AF in participants initially free of overt cardiovascular disease.

Methods:

In a stratified sample of the Multi-Ethnic Study of Atherosclerosis (MESA), initially aged 45–84 years, 3071 participants had both PIIINP and ICTP measured at baseline. Incident AF in 10 year follow-up was based on a hospital ICD code for AF or atrial flutter, in- or out-patient Medicare claims through 2011 (primarily in those aged 65–84 years), or ECG 10 years after baseline (n=357). The associations of PIIINP and ICTP with incident AF were estimated using Poisson regression with follow-up time offset.

Results:

Baseline PIIINP (5.50±1.55 μg/L) and ICTP (mean ± SD, 3.41±1.37 μg/L) were positively related (both p<0.0001) to incident AF in a model adjusting for age, race/ethnicity, and sex, with an apparent threshold (relative incidence density 2.81 (1.94–4.08) for PIIINP ≥8.5 μg/L (3.5% of the sample) and 3.46 (2.36–5.07) for ICTP ≥7 μg/L (1.7% of the sample). Findings were attenuated but remained statistically significant after further adjustment for systolic blood pressure, height, body mass index, smoking, and renal function. Additional adjustment for other risk factors and biomarkers of inflammation did not alter conclusions.

Conclusions:

Plasma collagen biomarkers, particularly at elevated levels, were associated with excess risk for atrial fibrillation.

Graphical Abstract

In 2010 the US prevalence of AF was 6.1 million and is expected to rise to at least 12.1 million by 2050.^1,2 AF is associated with significant morbidity and mortality.^3,4,5,6 Cardiac fibrosis is thought to play a central role in the pathogenesis of AF.⁷ Development and progression of atrial fibrosis are hallmarks of structural remodeling in AF and considered the substrate of AF perpetuation.⁸ Type I and III collagens are the major collagens in both normal and diseased tissue. Two major biomarkers of collagen metabolism, Procollagen Type III N-Terminal Propeptide (PIIINP) and Collagen Type I Carboxy-Terminal Telopeptide (ICTP), reflect collagen synthesis and degradation and degradation only, respectively.⁹ AF is common in subjects with symptomatic and clinically recognizable inflammation conditions. Low-grade, asymptomatic inflammation may also contribute to the pathogenesis of AF.^10,11,12,13 Processes of inflammation and pathologic fibrosis are often intertwined.¹⁴

Retrospective studies have suggested that circulating fibrosis biomarkers are associated with AF, but prospective studies are limited. In the Cardiovascular Health Study, a prospective cohort study of elderly men and women in which levels of plasma PIIINP were assessed in 2,935 participants, 767 developed AF during 9 years follow-up. Risk of incident AF generally increased with PIIINP, with a non-significant tendency to flatten at higher PIIINP levels.¹⁵ ICTP was not studied in CHS. In the Atherosclerosis Risk in Communities study the relationships between matrix metalloproteinases 1, 2 and 9, tissue inhibitor matrix metalloproteinase 1, 2 and C-terminal propeptide of collagen type-I with incident AF were examined after adjusting for confounders.¹⁶ In models adjusted for age, sex and race, all biomarkers were associated with AF, but only the relationship between plasma matrix metalloproteinase-9 remained significant in the fully-adjusted model.

The predictive value of collagen biomarkers for AF is largely unknown in subjects free of clinically recognized cardiovascular disease (CVD). Studying such persons affords the opportunity to examine the contribution of fibrosis to AF before the fibrosis can be the result of clinically overt CVD. It also offers advantages for determining the directionality of the fibrosisAF relationship, because people free of CVD would be less likely to have clinically unrecognized AF at baseline, which could itself drive atrial mechanical remodeling and consequent fibrosis. Therefore our hypothesis was that PIIINP and ICTP are positively associated with incidence of AF in middle-aged to older adults free of prevalent, clinically apparent CVD. Understanding associations between these collagen biomarkers and incident AF will suggest potential mechanisms of disease and may help in development of prevention strategies.

Methods

Study Population

The Multi-Ethnic Study of Atherosclerosis (MESA) was initiated to investigate the prevalence, correlates, and progression of subclinical CVD in people initially free of overt clinical CVD, radiation or chemotherapy-treated cancer, other serious major illness, or cognitive impairment in the judgment of the screening interviewer prior to baseline.¹⁷ Between 2000 and 2002, 6,814 men and women of white, black, Hispanic, or Chinese race/ethnicity, aged 45–84 years were enrolled. The institutional review boards at all participating centers approved the study, and all participants signed informed consent. Following standard MESA procedures for data availability, the data, analytic methods, and study materials will be made available to other researchers for purposes of reproducing the results or replicating the procedure after submission of a request to the MESA Publications and Presentations Committee (https://mesa-nhlbi.org/Publications.aspx) and obtaining that committee’s approval.

In the present stratified design, we excluded 1000 randomly selected MESA participants whose blood sample inventory had been depleted in earlier studies, as well as those who were missing baseline continuous blood pressure waveform measurement (the latter based on the original purpose of the design). We included 3071 participants, namely, all with any adjudicated CVD event (coronary heart disease, stroke, peripheral vascular disease or heart failure) prior to 2011 plus a 56% random sample of all remaining participants (who had not had a CVD event through 2011). With random sampling and case weighting, our data analyses are representative of all of MESA, although with somewhat expanded standard errors of our estimates. All included participants had assessment of both PIIINP and ICTP at baseline.

Height and weight were measured in the MESA clinic and body mass index (BMI) was defined as weight/height² (kg/m²). Questionnaires were administered to ascertain sex, race/ethnicity, medical history, medication use, and smoking status. Sitting rest blood pressure was determined oscillometrically; the average of the second and third measurements was used. Diabetes was defined as fasting plasma glucose (determined from a venous blood sample) ≥126 mg/dL or taking antidiabetic medication. Blood lipids were measured at the Collaborative Studies Clinical Laboratory (Fairview-University Medical Center, Minneapolis, MN) using CDC/NHLBI standards. High density lipoprotein cholesterol (HDL-C) was measured (Roche Diagnostics, Indianapolis, IN) after precipitation of non-HDL-C with magnesium/dextran. Low density lipoprotein cholesterol was calculated using the Friedewald equation.

Collagen Biomarkers

Blood was drawn following a 12 hour fast, and EDTA plasma was stored at −70°C. EDTA plasma PIIINP (μg/L) and ICTP (μg/L) were selected as measures reflecting type III and type I collagen turnover, respectively. Assays were performed in the Molecular Epidemiology and Biomarker Research Laboratory (University of Minnesota, under direction of Myron Gross, PhD) using competitive radioimmunoassay kits (UNIQ #06099, ICTP and UniQ #06098 PIIINP, Orion Diagnostica, Espoo, Finland)¹⁸. In the PIIINP assay, a known amount of labelled PIIINP and an unknown amount of unlabeled PIIINP in the sample compete for the limited number of high affinity binding sites of the antibody. After separating the free antigen, the amount of labeled PIIINP in the sample tube is inversely proportional to the amount of PIIINP in the sample. The description of the ICTP assay is parallel to that for PIIINP, using an ICTP specific antibody. The concentrations in unknown samples are obtained from a calibration curve. Coefficients of variation for internal quality control samples during the main runs were 9.3% for PIIINP high control, 16.5% for PIIINP low control, 6.3% for ICTP high control, and 8.8% for ICTP low control.

Inflammatory Markers

High Sensitivity C-reactive protein (hs-CRP) was measured using the BNII nephelometer (Dade Behring Inc, Deerfield, IL). Intra-assay and interassay analytic coefficient of variance (CV) ranged from 2.3 to 4.4% and 2.1 to 5.7%, respectively. Interleukin 6 (IL-6) was measured by ultrasensitive ELISA (Quantikine HS Human IL-6 Immunoassay, R&D Systems, Minneapolis, MN). The laboratory analytic CV for this assay was 6.3%. D-dimer was measured using an immunoturbidimetric assay on a Sta-R analyzer (Liatest D-DI, Diagnostica Stago, Parsippany, NJ). The lower limit of detection was 0.01μg/mL. Intra-assay CV for D-dimer was 12.2% and inter-assay CV range, 5–14%. The NMR measurement of GlycA and the lipoprotein particles in the MESA study have been described in our previous publication.^19,20 NMR spectra used were those acquired for NMR LipoProfile® testing at the CLIA-certified LipoScience (now LabCorp) clinical laboratory in Raleigh, NC. GlycA is a composite of alpha glycoprotein inflammatory markers; its chemical origins, and its analytic and biological variability have been described in detail.¹⁹ Reported coefficients of variation for intra- and interassay precision were 1.9% and 2.6%, respectively. The sum of small plus medium high-density lipoprotein particles (HSM-P), which we have shown to have anti-inflammatory properties, had coefficients of variation of 9% (small HDL particles) and 14% (medium HDL particles).¹⁹

Outcome Events: Atrial Fibrillation

AF was identified based on assignment of an ICD-9 diagnosis code for AF (427.31) or atrial flutter (427.32) in any position at hospital discharge or (for those enrolled in fee-for-service Medicare) in inpatient or outpatient Medicare claims through 2011; or in a 12-lead electrocardiogram at Exam 5 (10 years after baseline). It is recognized both that age is a strong risk factor for AF and that inclusion of Medicare outpatient claims provides more complete AF case-finding in ages ≥65 years than in younger people; baseline age-specific rates were 4.4% (72 AF cases) in ages 45–64, vs 19.68% (285 AF cases) in ages 65–84 years.

Statistical analysis

All analyses were weighted inversely to the sampling weight (1 or 0.56, in effect treating each participant who did not have a CVD event before 2011 as representing1/0.56 = 1.79 participants). Mean and standard deviation (SD) or counts and percentages were calculated for description. We initially performed Poisson regression with follow-up time offset predicting new onset of AF from PIIINP and ICTP, both as continuous variables. We also examined the data across equal concentration interval categories of PIIINP and ICTP (categories specified in the tables, lowest and highest categories are open-ended) as a way to display the shapes of the associations with the outcomes. Because the analyses of the equal interval categories strongly suggested threshold associations of the collagen biomarkers with incident AF, we report the associations in categories, with the lowest collagen biomarker levels as the reference, plus the p-value for linear trend from the continuous models. The cutpoints were selected based on boundaries of the equal interval biomarker category variable. We also examined spline plots (data not shown), which agree with our visual judgments. In this way we regard the continuous models only as general tests of upward trend and do not report linear slopes. The minimal model included age, race/ethnicity, and sex as covariates. The risk factor model added covariates height, heart rate, systolic and diastolic blood pressure (BP), BP lowering medications, body mass index (BMI), former and current smoking, diabetes, total cholesterol, HDL-C, triglycerides, cholesterol-lowering medication, clinic, estimated glomerular filtration rate (eGFR) and low eGFR (< 60 mL/min/1.73m²), hsCRP, IL-6, D-dimer, GlycA, and small plus medium HDL particles. (The propriety of inclusion of height and BMI in the same model is discussed in the Supplemental Methodological Note.) After backwards elimination, several variables were not associated with AF and the risk factor model shown in the tables included only systolic blood pressure, height, BMI, current smoking, and low eGFR. We recognize that age, renal status, and some risk variables may be causally linked to fibrosis as indicated by collagen biomarkers. To the extent that they are causally linked to fibrosis, we may be adjusting for mediating factors, and therefore underestimating the strength of association. In sensitivity analyses, we employed the random sample of cohort participants without weighting and/or using Cox proportional hazards regression. These analyses yielded similar conclusions, with the estimated associations in the Cox models slightly stronger than in the Poisson models (data not shown). The advantage of the Poisson models is that they give direct estimates of the absolute incidence densities. At the suggestion of a reviewer, we examined AF with or without incident CVD during followup, according to whether the incident CVD occurred before, with (30 days before or after), or after AF diagnosis. In this exploratory analysis, we identified AF without incident CVD, at least until 30 days after AF diagnosis and compared that to AF that was accompanied by CVD before or with AF diagnosis. For the AF without incident CVD, we computed continuous net reclassification improvement (NRI) to compare models with covariates only to models that included a collagen biomarker. We considered P<0.05 to be noteworthy in general screening of findings. Analyses were performed using PC-SAS version 9.4 (SAS institute, Cary, North Carolina).

Results

Table 1 and Supplemental Table 1 summarize the baseline characteristics according to PIIINP and to ICTP equal concentration interval categories. The mean age increased monotonically across all categories of ICTP and from the 1^st category of PIIINP through the 7 μg/L category. There were small differences in ethnic distribution among the PIIINP categories, while Chinese were less likely to have high ICTP. While there was no significant difference for systolic and diastolic BP among the categories, there was an increase in height, heart rate, BMI, and antihypertensive medication use with higher PIIINP and ICTP levels. Total cholesterol and HDL-C were lower in the higher PIIINP and ICTP levels. There was no difference in diabetes. Renal function was lower with higher PIIINP and ICTP levels. Hs-CRP, IL-6 and D-dimer were higher with higher PIIINP and ICTP levels. Small and medium HDL particles were lower with higher PIIINP and ICTP. Both current smoking and GlycA were inversely associated with PIIINP, although smoking was unrelated to ICTP and GlycA had a nomonotonic, but generally positive association with ICTP. ICTP and PIIINP had a correlation coefficient of 0.4.

Table 1:

Baseline characteristics according to PIIINP and to ICTP 1 μg/L interval categories

	Procollagen Type III N- Terminal Propeptide (PIIINP) category						p trend
Minimum - Maximum	1.32–4.49	4.5–5.49	5.5–6.49	6.5–7.49	7.5–8.49	8.5–27.71
Sample size	N=735	N=1050	N=721	N=332	N=126	N=107
Mean (μg/L)	3.93	5.00	5.94	6.92	7.89	10.3
ICTP (μg/L)	2.8±1.11	3.18±1.11	3.5±1.47	4.03±1.95	4.13±2.53	5.25±3.8	<0.0001
Age (years)	61.7±12.7	62.1±13.3	62.9±13.3	63.9±13.8	62.0±12.8	63.0±13.2	0.006
Male (%)	44.6	50.5	49.7	49.1	46.2	52	0.53
Height (cm)	165±12.9	166.5±12.9	167.1±12.9	167.2±13.2	166.3±13.6	167.1±13.1	0.02
Heart rate (beats/min)	62.6±11.1	62.9±12.1	63±12.2	62.8±12.8	64.8±13.7	64.6±15.2	0.03
Systolic pressure (mmHg)	125.8±25.6	125.7±27.5	126.7±27.3	129.4±29.3	127.8±29.1	126±27	0.32
Diastolic pressure (mmHg)	71.8±12.5	71.9±12.8	72.1±13.1	72.7±13	73.6±13.9	70.8±14.1	0.79
Antihypertensive Rx (%)	29.9	31.5	33.8	39.4	46.1	38.6	0.01
Body Mass Index (kg/m2)	26.9±6	28±6.7	28.7±6.9	29.7±7.3	30.4±7.3	30.3±8.8	<0.0001
Former smoker (%)	35	37	38.2	36.2	39.8	38.7	0.79
Current smoker (%)	15.4	12.5	9.4	9	9.9	12.5	0.003
eGFR* <60 ml/min/1.73 m2 (%)	2.6	4	5.2	8.1	8.2	10.5	<0.0001
	Collagen Type I Carboxy-Terminal Telopeptide (ICTP) categories						P trend
Minimum - Maximum	1.12–2.99	3–3.99	4–4.99	5–5.99	6–6.98	7.01–23.45
Sample size	N=1329	N=1080	N=430	N=132	N=45	N=55
Mean (μg/L)	2.44	3.45	4.4	5.41	6.43	9.55
PIIINP (μg/L)	4.95±1.46	5.52±1.55	6.21±2.55	6.57±2.21	7.27±2.65	7.75±3.18	<0.0001
Age (years)	59.9±12.1	63.2±13.3	65.7±13.4	65.9±13.6	68.8±12.2	70.6±12.6	<0.0001
Male (%)	46.1	50.4	50.7	50.8	50.7	54.3	0.07
Height (cm)	166±13	166.6±13	166.6±13.4	167±12.5	168.1±10.9	167.4±12.9	<0.0001
Heart rate (beats/min)	62.4±11.4	63.2±12.3	63.5±13	64.5±14.3	64.1±14.7	64.3±13.9	0.0002
Systolic pressure (mmHg)	124.2±26.3	127.3±27	129.4±27.6	125.2±27.2	134.8±37.4	137.6±32.9	0.06
Diastolic pressure (mmHg)	72.1±13.2	72.3±12.5	71.7±12.2	70.2±13.1	71.1±14.1	73.4±15.4	0.86
Antihypertensive Rx (%)	27.9	32.6	42.8	50.1	46.4	60.8	<0.0001
Body Mass Index (kg/m2)	27.4±6.4	28.5±6.8	29.4±7.4	30.6±7.8	31.4±7.4	29.3±7.3	<0.0001
Former smoker (%)	36.7	36.5	37.5	37.1	41.7	38.3	0.36
Current smoker (%)	12.4	11.6	12.7	10	11.1	11.6	0.29
eGFR* <60 ml/min/1.73 m2 (%)	1.3	3	10.8	10.4	19	56.6	<0.0001

^*eGFR: Estimated glomerular filtration rate ml/min/1.73 m² based on CKD-EPI formula using serum creatinine.

Results are given either in percentage (%) or as mean ± standard deviation. Equal concentration interval categories of PIIINP and ICTP were used (note that lowest and highest categories are open-ended).

Table 2 shows the mean ICTP and PIIINP concentrations about to AF without or without accompanying CVD. The incident disease subgroups with consistently high collagen markers were the 166 people with incident AF without any incident CVD event and the 61 people with incident AF who did subsequently have an incident CVD event, but not until at least 30 days after the diagnosis of AF.

Table 2.

Collagen biomarker mean values according to presence or absence of atrial fibrillation (AF) with or without CVD

		PIIINP (μg/L)	ICTP (μg/L)
	N	Mean±SE	Mean±SE
No AF
Neither incident CVD nor incident AF	2217	5.41±0.03	3.28±0.03
Incident CVD, no incident AF	497	5.58±0.07	3.44±0.06
AF without other CVD
No incident CVD, any incident AF	166	5.89±0.13	3.88±0.14
Incident CVD after incident AF	61	5.90±0.26	4.48±0.23
AF accompanied by other CVD
Incident CVD within 30 days of incident AF	63	5.52±0.15	3.63±0.14
Incident CVD before incident AF	67	5.39±0.18	3.97±0.33

Incident AF during the 10 year follow-up was identified in 357 participants. Figure 1 summarizes the AF incidence density for each category of each collagen biomarker. Positive linear trend was significant in each model (Table 3). Threshold relationships were suggested visually in Figure 1, with small gradual increase across the lower categories, particularly in the demographic adjusted model. The apparent threshold for PIIINP was at 8.5 μg/L (3.5% of the sample), with relative incidence density (RID) 2.81 (1.94–4.08) in the demographic adjusted model and 2.29 (1.57–3.34) in the more adjusted model. Correspondingly for ICTP the apparent threshold was at 7 μg/L (1.7% of the sample), with RID 3.46 (2.36–5.07) in the demographic adjusted model and 2.44 (1.60–3.72) in the more adjusted model. Associations of ICTP and PIIINP with incident AF not accompanied by incident CVD event (N=227) were stronger than those for total AF. NRI is about 0.14 (p<0.003) for both markers. We also note that the adjusted incidence density rates in those for whom the addition of the collagen markers suggested a higher risk vs a lower risk were 7.8 vs 5.8 per 1000 person years (p-0.003) for ICTP and 8.2 vs 5.6 per hundred person years (p<0.0001) for PIIINP. The collagen markers were not related to incident stroke (data not shown).

Figure 1:

Incidence densities per 1000 person years for atrial fibrillation according to 1 μg/L interval categories, with open-ended extreme categories, of Plasma Procollagen Type III N-Terminal Propeptide (PIIINP, A) and Plasma Collagen Type I CarboxyTerminal Telopeptide (ICTP, B).

Blue line: adjusted for demographics (age, race/ethnicity, and sex)

Red line: adjusted for demographics and for systolic blood pressure (SBP), height (Ht), body mass index (BMI), current smoker, and estimated glomerular filtration rate (eGFR)<60 ml/min/1.73 m². Covariates are set to their overall means in the plotted points.

Further adjustment for heart rate, diastolic blood pressure, blood pressure lowering medications, former smoking, diabetes, total cholesterol, HDL-cholesterol, triglycerides, cholesterol-lowering medication, hsCRP, IL6, D-dimer, GlycA, and small plus medium HDL particles), did not substantially alter the estimates and is not shown.

Table 3:

Incident Atrial Fibrillation and Collagen Biomarkers (PIIINP and ICTP) in adjusted Poisson models Relative risks across 1 μg/L interval categories of each collagen biomarker

	PIIINP category						p trend
Mean (μg/L)	3.93	5.00	5.94	6.92	7.89	10.3
Sample size	N=735	N=1050	N=721	N=332	N=126	N=107
All incident atrial fibrillation (N=357)
Model: adjusted for age, race/ethnicity, and sex
Relative Risk (CI)	1 (ref)	0.92(0.71–1.18)	1.02(0.78–1.33)	1.25(0.92–1.69)	1.4(0.89–2.18)	2.81(1.94–4.08)	<0.0001
Model: further adjusted for SBP, Ht, BMI, smoker, eGFR<60
Relative Risk (CI)	1 (ref)	0.87(0.68–1.12)	0.91(0.7–1.18)	1.03(0.76–1.41)	1.12(0.71–1.76)	2.29(1.57–3.34)	0.003
Incident atrial fibrillation without other CVD (N=227)
Model: adjusted for age, race/ethnicity, and sex
Relative Risk (CI)	1 (ref)	0.82(0.61–1.11)	1.02(0.75–1.38)	1.28(0.9–1.82)	1.39(0.83–2.33)	3.48(2.33–5.19)	<0.0001
Model: further adjusted for SBP, Ht, BMI, smoker, eGFR<60
Relative Risk (CI)	1 (ref)	0.78(0.58–1.06)	0.91(0.67–1.24)	1.07(0.74–1.53)	1.10(0.65–1.86)	2.75(1.83–4.15)	0.006
Incident atrial fibrillation accompanied by other CVD (N=130, 59 of which involved heart failure)
Model: adjusted for age, race/ethnicity, and sex
Relative Risk (CI)	1 (ref)	1.2(0.74–1.93)	1.03(0.61–1.74)	1.14(0.62–2.12)	1.39(0.57–3.38)	0.85(0.26–2.82)	0.36
Model: further adjusted for SBP, Ht, BMI, smoker, eGFR<60
Relative Risk (CI)	1 (ref)	1.14(0.7–1.84)	0.92(0.54–1.57)	0.95(0.5–1.79)	1.17(0.47–2.88)	0.75(0.22–2.51)	0.09
	ICTP category						p trend
Mean (μg/L)	2.44	3.45	4.4	5.41	6.43	9.55
Sample size	N=1329	N=1080	N=430	N=132	N=45	N=55
All incident atrial fibrillation (N=357)
Model: adjusted for age, race/ethnicity, and sex
Relative Risk (CI)	1 (ref)	1.21(0.97–1.5)	1.18(0.89–1.55)	2.02(1.41–2.9)	1.49(0.85–2.61)	3.46(2.36–5.07)	<0.0001
Model: further adjusted for SBP, Ht, BMI, smoker, eGFR<60
Relative Risk (CI)	1 (ref)	1.08(0.86–1.34)	0.97(0.73–1.28)	1.69(1.17–2.45)	1.04(0.59–1.84)	2.44(1.60–3.72)	0.0005
Incident atrial fibrillation without other CVD (N=227)
Model: adjusted for age, race/ethnicity, and sex
Relative Risk (CI)	1 (ref)	1.38(1.07–1.79)	1.22(0.88–1.7)	2.08(1.34–3.23)	1.92(1.04–3.54)	4.80(3.17–7.25)	<0.0001
Model: further adjusted for SBP, Ht, BMI, smoker, eGFR<60
Relative Risk (CI)	1 (ref)	1.24(0.95–1.61)	1.02(0.73–1.44)	1.77(1.12–2.78)	1.39(0.74–2.61)	3.67(2.31–5.83)	0.001
Incident atrial fibrillation accompanied by other CVD (N=130, 59 of which involved heart failure)
Model: adjusted for age, race/ethnicity, and sex
Relative Risk (CI)	1 (ref)	0.83(0.54–1.27)	1.07(0.65–1.75)	1.85(0.97–3.52)	0.63(0.15–2.62)	0.84(0.26–2.74)	0.36
Model: further adjusted for SBP, Ht, BMI, smoker, eGFR<60
Relative Risk (CI)	1 (ref)	0.75(0.49–1.15)	0.88(0.53–1.49)	1.58(0.8–3.1)	0.42(0.1–1.78)	0.49(0.14–1.73)	0.89

BMI: body mass index; eGFR: estimated glomerular filtration rate; ICTP: Collagen Type I Carboxy-Terminal Telopeptide; Ht: height; PIIINP: Plasma Procollagen Type III N-Terminal Propeptide; SBP: systolic blood pressure

Other demographic variables and risk factors in the more adjusted model were associated with incident AF, including age (RID per yr 1.10, 95% CI 1.08–1.11), black and Hispanic race/ethnicity (lower risk than whites or Chinese), height (RID per cm 1.04, 1.02–1.05), systolic BP (RID per 10 mmHg 1.08, 1.04–1.13), BMI (RID per kg/m² 1.05, 1.03–1.07), and current smoking (RID vs never smoker 1.46, 1.07–1.98). Low eGFR<60 ml/min/1.73 m² was correlated with incident AF when adjusted for PIIINP (RID vs higher eGFR 1.48, 1.13–1.94), but was not significant when adjusted for ICTP. Other risk factor and inflammatory variables were at most weakly associated with future AF.

Discussion

Our main and novel findings were that in a large multi-ethnic and middle-aged sample, initially free of clinically recognized CVD, the collagen biomarkers PIIINP and ICTP predicted new onset of AF during a median follow-up of approximately 10 years. The shape of relationship was generally increasing, with apparent thresholds above 8.5 μg/L for PIIINP and above 7 μg/L for ICTP. The strong effect sizes in the highest categories of PIIINP and ICTP fit well with two earlier reports. Collagen biomarkers were strongly related to total death and chronic inflammatory-related severe hospitalization and death.¹⁸ For the less frequently occurring outcome heart failure with preserved ejection fraction, a threshold association was found in the relationship of PIIINP,²¹ as we report here for AF. Circulating collagen biomarkers should be interpreted with consideration of their meaning in relation to collagen molecules systemically. Collagen biomarkers that are found in the circulation are fragments of collagen molecules that arise from collagen activity throughout the body, including both normal tissue maintenance and tissue repair. Therefore circulating collagen biomarkers are an imperfect measure of tissue repair. At any one time, several organs may be undergoing tissue repair with attendant collagen molecule production. In this case, epidemiologic disease associations would be consistent over broad ranges of PIIINP and ICTP. If only one organ were undergoing tissue repair at the time of plasma collagen biomarker measurement, only a small portion of the variation in plasma collagen biomarker would be expected from that organ repair. Then, epidemiologically, a threshold would be expected.

We explored subsets of AF. We found that the subset of AF which was not associated with incident CVD at all during MESA followup or until at least 30 days after AF diagnosis was strongly associated with the collagen biomarkers, in contrast with AF which was associated with CVD either diagnosed before or concurrent with AF. These findings are suggestive that fibrosis is a cause of AF when CVD is not present, but AF may also be responsive to other features of CVD in those who have AF accompanied by CVD.

There is very limited information in the literature regarding the predictive value of PIIINP and ICTP for AF in study participants without clinically recognized CVD at baseline. In the Cardiovascular Health Study, PIIINP was measured in 2935 participants and the risk of incident AF was studied over a median follow-up of 8.8 years.¹⁵ AF was identified based on an annual resting EKG through the 9^th year of follow-up and discharge diagnoses for all hospitalizations based on the ICD codes. The Cardiovascular Health Study sample was older (mean age range 77–78 years) than the MESA study sample. More than 50% of the participants were treated for hypertension. In contrast to our study some participants in the Cardiovascular Health Study had a history of heart failure, myocardial infarction or stroke. In that study, they found a linear relationship between the risk of AF and PIIINP approximately up to the median value both before and after adjustment for demographics, CVD risk factors and history of heart failure and stroke. Above the median value of PIIINP the association between PIIINP and AF risk was not significant, while in our MESA study only the highest category of PIINP and ICTP levels were predictive for AF.

Fibrosis is a process of normal repair that is activated in response to injury to maintain the original tissue architecture and functional integrity. Prolonged chronic injurious stimuli may cause deregulation of normal processes and result in an excess deposition of extracellular matrix and fibrosis. Fibrosis involves complex multistage inflammatory processes with inflammatory cell and extracellular matrix expansion where numerous factors including cytokines/chemokines, growth factors, adhesion molecules, and signaling processes play a role.^10,14,22,23

Several hypotheses for our findings may be formulated.^22,23,24 In the presence of risk factors like obesity, obstructive sleep apnea, and diabetes, inflammation and an increased sympathetic tone may play a pathological role for AF. Moreover, an important concept is that inflammation contributes to AF before there is consolidation of the inflammatory process leading to structural changes and fibrosis²⁵. However, we reported here that five inflammatory markers did not predict future AF; perhaps such inflammatory factors become important more proximally to the emergence of AF. The molecular basis for the heritability of AF has been explored in depth over the past decade. Rare variants have been identified in ion channels, transcription factors and a wide range of other genes.²⁶

As with other epidemiological studies, the ascertainment and classification of AF is a potential limitation of the present study. Because we had access to hospital discharge diagnoses through MESA cardiovascular events surveillance, plus Medicare outpatient and inpatient claims, our ascertainment may be relatively good.²⁷ However, we are aware that paroxysmal AF may go unrecognized by the patient and not occur at a clinic visit. Therefore, the true AF prevalence is likely to be higher. The type of AF, whether symptomatic or asymptomatic (paroxysmal, persistent, or chronic), cannot be discriminated in this study. A cause-effect relationship cannot be firmly established. It is possible that residual confounding might explain the prediction of new cases of AF by low level collagen biomarkers. Statistically, detecting and describing nonlinear relationships is inherently much more challenging than finding linear associations.

In conclusion, our results add further insight regarding collagen biomarkers and AF. Collagen biomarkers were associated with incident AF in an increasing fashion, with an apparent threshold in asymptomatic subjects free of overt CVD.

What is Known?

• Collagen biomarkers in the circulation are fragments of collagen molecules that arise from collagen activity throughout the body, including both normal tissue maintenance and repair.

• Some atrial fibrillation is known to be associated with fibrosis in the atria.

What the Study Adds?

• In a large sample of men and women free of clinical cardiovascular disease at baseline, atrial fibrillation that was not associated with cardiovascular disease was strongly associated with circulating Type I and Type III collagen during 10 years followup.

• Atrial fibrillation with concurrent or prior cardiovascular disease was not associated with either circulating Type I or Type III collagen.

• The associations with collagen markers showed a pronounced threshold.