Proteomics analysis of genetic liability of abdominal aortic aneurysm identifies plasma neogenin and kit ligand: The Atherosclerosis Risk in Communities Study

Abstract

Background:

Genome-wide association studies have reported 23 gene loci related to abdominal aortic aneurysm (AAA)—a potentially lethal condition characterized by a weakened dilated vessel wall. This study aimed to identify proteomic signatures and pathways related to these risk loci to better characterize AAA genetic susceptibility.

Methods:

Plasma concentrations of 4,870 proteins were determined using a DNA aptamer-based array. Linear regression analysis estimated the associations between the 23 risk alleles and plasma protein levels with adjustments for potential confounders in a race-stratified analysis of 1,671 Black and 7,241 White participants. Significant proteins were then evaluated for their prediction of clinical AAA (454 AAA events in 11,064 individuals), and those significantly associated with AAA were further interrogated using Mendelian randomization (MR) analysis.

Results:

Risk variants proximal to PSRC1-CELSR2-SORT1, PCIF1-ZNF335-MMP9, RP11–136O12.2/TRIB1, ZNF259/APOA5, IL6R, PCSK9, LPA, and APOE were associated with 118 plasma proteins in Whites and 59 were replicated in Black participants. Novel associations with clinical AAA incidence were observed for kit ligand (HR: 0.59; 95% CI: 0.42 – 0.82 for top vs 1^st quintiles) and neogenin (HR: 0.64; 95% CI: 0.46 – 0.88) over a median 21.2-year follow-up; neogenin was also associated with ultrasound-detected asymptomatic AAA (N=4,295; 57 asymptomatic AAA cases). MR inverse variance weighted estimates suggested that AAA risk is promoted by lower levels of kit ligand (OR per SD=0.67; p=1.4×10⁻⁵) and neogenin (OR per SD=0.50; p=0.03).

Conclusions:

Low levels of neogenin and kit ligand may be novel risk factors for AAA development in potentially causal pathways. These findings provide insights and potential targets to reduce AAA susceptibility.

Graphical Abstract

BACKGROUND

Abdominal aortic aneurysm (AAA) is a potentially life-threatening condition typified by a focal vessel enlargement in the abdominal cavity with 80–90% occurring in the infrarenal section ¹. The nascent stages and progression of AAA are characterized by a disruption in extracellular matrix (ECM) integrity of the aortic wall that involves chronic inflammation ^2–4, aberrant immune responses including endothelial infiltration of neutrophils and monocyte-derived macrophages, and chronic matrix metalloproteinase (MMP) activity that catabolizes the ECM ^4–7. Despite our understanding of characteristic AAA pathophysiology, the underlying genetic predispositions to disease development are not as well established.

Apart from smoking, genetics is among the strongest risk factors for AAA ^8,9, and twin studies have reported a 70–77% heritability of the condition ^10,11. Genome-wide association studies (GWASs) have revealed multiple genetic variants associated with AAA ^12–15, and 23 independent gene loci have been discovered to date ^14,15. No comprehensive proteomics study has interrogated these variants with the goal of identifying potential downstream protein architecture through which they may influence risk of AAA manifestation.

To explore potential mechanisms involved in the genetic predisposition of AAA, we assessed the associations of the GWAS-reported AAA risk variants at the 23 loci with an array of 4,870 plasma proteins measured in 8,912 participants of the Atherosclerosis Risk in Communities (ARIC) study. Pathway analyses were conducted to examine potential activation or suppression of pathway networks based on identified SNP-protein associations. We then evaluated whether identified proteins were associated with clinical AAA incidence over a median 21.2-year follow-up in ARIC. Proteins that were associated with clinical AAA were further tested for associations with a separate group of asymptomatic AAA outcomes, ascertained by an abdominal ultrasound screening at ARIC visit 5 conducted between 2011 and 2013. Finally, two-sample Mendelian randomization analysis was conducted to determine whether identified proteins may be causally related to risk of clinical AAA.

MATERIALS AND METHODS

The data that support the findings of this study are available from the ARIC Coordinating Center (arichelp@umn.edu) upon reasonable request. Please see the Major Resources Table in the Supplemental Materials.

Study population

The ARIC prospective cohort study was designed to identify risk factors for cardiovascular disease and atherosclerosis in 15,792 individuals ¹⁶. The Institutional Review Board at each participating institution approved the study protocol and participants provided written informed consent prior to enrollment. Self-reported Black and White male and female participants aged 45–64 were recruited between 1987 and 1989 across four communities in the US (Washington County, Maryland; the northwest suburbs of Minneapolis, Minnesota; Jackson, Mississippi; and Forsyth County, North Carolina). The Jackson center enrolled only Black participants. ARIC participants received annual follow-up calls (semi-annual after 2012) to report cardiovascular events and hospitalizations. Information on cardiovascular risk factors and conditions, including anthropometrics, history of physician-diagnosed cardiovascular conditions, medication use, and behavior risk factors, was collected at baseline and several follow-up examinations. The present study used proteomic biomarker data from plasma collected at ARIC visit 3, which took place from 1993 to 1995.

Genotyping

Information on genotyping, quality control, and imputation procedures has been detailed previously ¹⁷. Briefly, genomic DNA from whole blood was genotyped using the Affymetrix Genome-Wide Human SNP array 6.0 (Affymetrix, Santa Clara, CA, USA). To expand the number of genetic markers beyond this genotype array, ARIC investigators generated race-specific imputation of variant dosages using the TopMed reference panel (freeze 5b) ¹⁸. Prior to imputation, individuals who were first-degree relatives, genetic outliers, or whose array genotypes did not match genotype data from other platforms were excluded. Based on the GWAS data, ARIC generated principal components using EIGENSTRAT ¹⁹ to reflect population substructure or genetic ancestry of ARIC participants.

Assessment of kidney function

Estimated glomerular filtration rate (eGFR) at visit 3 was included in models to control for the potential for confounding of poor kidney function on plasma proteins. eGFR (mL/min/1.73 m²) was calculated using the Chronic Kidney Disease Epidemiology Collaboration combined creatinine-cystatin C equation ²⁰. Details on the measurement of serum creatinine and cystatin C have been described elsewhere ^21,22.

Proteomics measurement and quality control

EDTA-plasma was collected during ARIC visit 3 and stored at −80 degrees C. Samples were analyzed using a modified aptamer-based capture array (SomaScan version 4.0, SomaLogic, Inc., Boulder, CO, USA) that has been described previously ^23,24. Briefly, plasma samples were transferred to the SomaLogic laboratory and incubated with proprietary SomaLogic reagents. Protein levels in the plasma samples were measured using single-stranded DNA-based modified aptamers that bind specific protein epitopes. All measures were reported as relative fluorescent units, and validation of specificity and intra- and inter-assay variability have been reported ^25,26.

Protein measurements by SomaScan have been standardized and normalized as previously described ^26,27. Briefly, hybridization control normalization was applied to each sample based on a set of hybridization control sequences to correct for systematic biases during hybridization. Median signal normalization was applied to measures within plates to remove sample or assay biases due to variations in pipetting, reagent concentrations, assay timing, and other sources of systematic variability within single plate runs. Each plate contained calibrators for each aptamer reagent to correct for plate-to-plate variation based on global reference materials. ARIC previously conducted a pilot study of SomaScan v.3 in 42 ARIC participants and reported excellent metrics of assay reproducibility: median coefficient of variance (quartile 1, quartile 3) of 5.0 (4.1, 6.9) and median intraclass correlation (quartile 1, quartile 3) of 0.96 (0.92, 0.98) ²⁷.

For ARIC visit 3 measurements in the whole cohort by SomaScan v.4, we examined protein distributions and applied log base 2 transformation to all protein measures to correct for skewness. Four-hundred twenty-two blind duplicate plasma aliquots were included, and the median inter-assay Bland-Altman coefficient of variation was 6.3%. The median split sample reliability coefficient was 0.85 after applying the following quality control filters on 5,284 available aptamer measurements: Bland-Altman coefficient of variation >50% or a variance of <0.01 on the log scale (n=94 excluded) and non-specific binding to non-proteins (313 excluded). After all quality control measures were completed, 4,870 aptamer measurements remained, corresponding to 4,697 unique human proteins or protein complexes. Supplementary Table S1 shows assay binding characteristics and intra- / inter-assay coefficients of variance for those plasma proteins found to be significant in the main analyses.

AAA Ascertainment

Clinical AAA Incidence

Ascertainment of AAA events in the ARIC cohort have been previously described ^2,3. Briefly, annual or semi-annual telephone calls with ARIC participants were conducted to determine any interim hospitalizations and identified deaths. ARIC also conducted surveillance of local hospitals to identify additional hospitalizations or deaths. Participant identifiers were additionally linked with Medicare data from the Centers for Medicare and 3 Medicaid Services (CMS) for 1991–2011, to find other outpatient or hospital events for those over 65 years of age. Searching the hospitalization, death records, and Medicare data identified a total of 517 incident clinical AAA events (including 195 severe cases consisting of non-mutually exclusive 186 AAA medical procedure interventions and 30 AAA rupture events) through December 31, 2016. Thoracic, thoracoabdominal, or unspecified aortic aneurysms were treated as non-events.

Asymptomatic AAA by ultrasound screening

To identify additional asymptomatic AAAs in the surviving ARIC cohort, an abdominal ultrasound scan was performed at ARIC Visit 5 in 2011–2013. The description of the abdominal ultrasound exam in ARIC can be found elsewhere ^2,3. In brief, cardiac ultrasonographers, trained by a radiologist with special vascular imaging expertise, conducted the abdominal ultrasound scan. The sonographers made anterior–posterior and transverse diameter measurements on transverse images of the abdominal aorta. Vascular imaging physicians overread all images of abdominal aortas with the maximal infrarenal diameter ≥ 2.8 cm or probable aortic pathology, plus a 5% random sample of the remaining cohort. The definition for ultrasound AAA was based on infrarenal diameter ≥ 3.0 cm. The correlation coefficient between the physician readers and ultrasonographers for the maximum infrarenal diameter was 0.92.

Statistical analysis

Analyses included participants who attended visit 3 when plasma samples for protein measurements were collected. Those missing genetic data (n=2737), plasma protein outcomes (n=1041), or other covariates (n=197) were excluded, resulting in a final sample of 1,671 Black and 7,241 White ARIC participants in the primary analysis. Multivariable linear regression analysis estimated the associations between the AAA risk allele of each SNP and the log base 2 transformed protein outcomes. Analysis was race-specific and adjusted for age at visit 3, sex, field center, eGFR at visit 3, and ten principal components of ancestry. For each aptamer measurement, outliers that were outside of 6 standard deviations (SD) from the mean were excluded. Bonferroni correction was carried out to correct for multiple testing: 23 genetic variants*4,870 protein outcomes stipulated a significance threshold of p≤4.46 ×10⁻⁷ in Whites. The analysis in Black participants was treated as a replication sample for significant SNP-protein associations identified in Whites, and Bonferroni correction values were contingent on the numbers of proteins tested in the replication analysis.

Analysis of identified proteins with clinical AAA incidence

Proteins identified in the main analysis were secondarily examined in a prospective analysis of clinical AAA incidence from Visit 3 baseline (1993–95) through December 31, 2016 among ARIC participants who had complete data. Participants reporting prior AAA surgery or aortic angioplasty at Visit 1, and those diagnosed with AAA between Visit 1 and Visit 3 were excluded (n=138) ^2,3. Cox regression estimated hazard ratios between plasma protein quintiles and risk of incident AAA, and a trend test modeled rank-ordered protein quintiles as continuous variables. The analysis adjusted for age, sex, race, field center, smoking status, and eGFR in model 1 (N=11,064; 454 incident clinical AAA events), and further adjusted for BMI, prevalent hypertension, diabetes, CHD, and stroke, total cholesterol, and use of lipid lowering medication in model 2 (N=10,701; 434 incident clinical AAA events). These covariates were adjusted for as potential confounders based on their known associations with AAA. Bonferroni correction was applied to control for multiple testing: 118 significant proteins from the SNP-protein analysis stipulated a significance threshold of p<4.2×10^-4. Modifying effects of race, sex, and age were individually tested by including the interaction term in the respective Cox regression models.

We further examined the associations of kit ligand and neogenin with severe AAA by combining those requiring medical procedure interventions and rupture events. After excluding those with no protein data or missing covariates: n for events=116 for model 1; n for events=110 for model 2; covariate adjustments for models 1 and 2 are the same as for the analysis of all clinical AAA above.

Previous evidence has indicated that IL-6 signaling pathways promote greater risk of AAA ⁵. We therefore treated IL-6 and sIL-6Ra as candidate proteins (significance threshold of p<0.05) and simultaneously evaluated their associations with the risk of incident AAA by including both in the same model.

Analysis of identified proteins with ultrasound-detected AAA

For proteins that were significantly associated with clinical AAA incidence, we further determined whether these proteins also predicted the risk of ultrasound-detected asymptomatic AAA among ARIC participants who underwent an abdominal ultrasound screening at Visit 5, which detected 75 asymptomatic AAAs among 5,911 surviving participants. After exclusions of participants diagnosed with a clinical AAA before Visit 5 or those with missing data, 4,295 participants (57 asymptomatic AAAs) remained in the analysis, with 4 cases detected in Black participants (n=751). We used inverse probability of attrition weighting (IPAW) to adjust for the potential selection bias caused by differential participation in the Visit 5 exam ²⁸ and obtained odds ratios (OR) and 95% confidence intervals from IPAW generalized estimating equation (GEE) models using a logit link and robust standard errors ²⁸. Details on the application of IPAW and GEE to the ARIC Visit 5 ultrasound data can be found elsewhere ^2,3. Due to the small number of these AAA cases, covariate adjustments were limited to the potential confounders from the clinical AAA analysis with the strongest magnitudes of association and significances. The primary and secondary analyses described above were conducted using SAS 9.4 statistical software.

Two-sample Mendelian randomization

Two-sample MR analyses were restricted to two proteins associated with incident, clinical AAA—neogenin and kit ligand. Instrumental variables [i.e., protein quantitative trait loci (pQTLs)] were identified by GWA analyses of protein levels with genetic variant dosages obtained from the population-based deCODE Health study (N=35,369) ²⁹. Clumping was used to prune pQTLs that were in linkage disequilibrium (LD) (r² for LD<0.01), did not reach the significance threshold (p≤5×10⁻⁸) or were within a clumping distance threshold of 5 Mb. Neogenin and kit ligand pQTLs are described in Supplementary Table S2 including pQTL chromosome locations, rsid numbers, allele frequencies, percent variance in protein level explained by each pQTL, and F-statistics. Replication of significant and independent pQTLs was carried out in White ARIC participants. The exposure summary statistics from the deCODE Health study were used in the MR analysis, and linear regression analysis was conducted to estimate the variances in protein levels explained by their respective pQTLs (%) (Supplementary Table S3). For replicated pQTLs, AAA outcome summary statistics were obtained from the most recent and largest GWAS meta-analysis (N=39,221 cases and 1,086,107 controls) ³⁰. In total, five pQTLs were identified for neogenin and seven for kit ligand. Random effects inverse variance weighted (IVW) meta-analyses of Wald ratios were performed for both proteins. MR-Egger and MR-Pleiotropy residual sum and outlier (PRESSO) sensitivity analyses were conducted to determine whether associations between the protein and AAA may have been influenced by pleiotropy. MR-PRESSO outlier tests identified pleiotropic outlier variants with Bonferroni correcting for multiple comparisons. Identified outliers were removed, and IVW analyses for kit ligand and neogenin were re-run to more accurately assess their potential causal effects in AAA. MR statistical analysis was performed using R software (R version 4.1.3 for the main MR analyses and R version 3.6.3 for MR-PRESSO) and R Two Sample MR package version 0.5.6 ³¹.

Pathway analysis for SNP-protein associations

Network pathway analysis was conducted using the Ingenuity Pathway Analysis (QIAGEN Inc. ³²) to identify potential mechanisms and enriched signaling cascades for the top SNP-protein associations. Benjamini-Hochberg false discovery rate q-values were generated from SNP-protein association p-values. For each AAA-related SNP, results of all 4,870 SNP-protein associations were uploaded to the IPA platform as a comparison, including protein identifiers, parameter estimates (in the form of log expression ratios), and q-values. Proteins were considered for pathway analysis where protein-SNP associations met a threshold of q≤0.05. Core analyses were conducted using ARIC data as the reference set, and direct and indirect experimentally confirmed relationships across species were examined.

RESULTS

Sample Characteristics

Multiple tables were generated to inspect the analytic sample. Shown in Table 1, a comparison of demographic, lifestyle, and clinical characteristics between all ARIC participants at Visit 3 (N=12,887) and those in the analytic sample (N=8,912) did not show material differences apart from a nominally lower frequency of Black participants (−4.5%). Characteristics of participants included in the main analysis stratified by race are shown in Supplementary Table S4, and those included in the follow-up analysis stratified by incident AAA status are shown in Supplementary Table S5 (N=11,064; 454 AAA events).

Table 1.

Demographic, lifestyle, and clinical characteristics of all ARIC participants at Visit 3 (N=12,887) and those included in the primary analysis (N=8,912)^*.

Characteristic	All ARIC Visit 3 Participants	Analytic Sample a
Age, mean (SD)	60.0 (5.7)	60.1 (5.7)
Sex (female), n (%)	7170 (55.6)	4822 (54.1)
Race group, n (%)
Black	2997 (23.3)	1671 (18.8)
White	9852 (76.5)	7241 (81.3)
Field center, n (%)
Forsyth County, NC	3342 (25.9)	2388 (26.8)
Jackson, MS	2622 (20.4)	1462 (16.4)
Minneapolis, MN	3497 (27.1)	2714 (30.5)
Washington County, MD	3426 (26.6)	2348 (26.4)
Smoking status, n (%)
Current	2283 (17.8)	1605 (18.1)
Former	5286 (41.1)	3750 (42.2)
Never	5277 (41.1)	3537 (39.8)
BMI (kg/m2), mean (SD)	28.5 (5.6)	28.5 (5.5)
Diabetes, n (%)	1988 (15.5)	1377 (15.5)
Hypertension, n (%)	4312 (33.7)	2961 (33.4)
Previous MI, n (%)	747 (5.8)	545 (6.1)
Previous stroke, n (%)	252 (2.0 )	185 (2.1)
Incident AAA	517 (4.0)	370 (4.2)
eGFR (mL/min/1.73 m2), mean (SD)	88.6 (15.5)	88.2 (15.3)

^*ARIC participants at Visit 3 with proteomic, genetic, ancestral, and eGFR data.

Associations of proteins and AAA-related SNPs

A total of 1,671 Black and 7,241 White ARIC participants were included in the SNP-protein analysis. Table 2 presents top risk variants for the 23 AAA loci, their chromosome positions, proximal genes, frequencies, and previously reported odds ratios with AAA. Further information for each SNP and associated phenotypes is presented in Supplementary Table S6. In the main analysis in Whites, a total of 118 proteins were significantly associated with 8 of the 23 AAA-related risk variants, including rs11591147 (proximal to PCSK9), rs118039278 (LPA), rs3827066 (PCIF1-ZNF335-MMP9), rs10808546 (RP11–136O12.2/TRIB1), rs4129267 (IL6R), rs602633 (PSRC1-CELSR2-SORT1), rs964184 (ZNF259/APOA5) and rs429358 (APOE), following Bonferroni correction. Of these 132 associations, 59 were replicated in Black participants, and 101 were directionally consistent between the two groups (Supplementary Tables S7-S10).

Table 2.

AAA-related variants and frequencies among Black and White ARIC participants.

rsID	Chr	Position*	Proximal genes	Ref	Risk†	OR**	Black freq	Black R2	White freq	White R2
rs11591147	1	55039974	PCSK9	T	G	1.58	0.99	0.94	0.98	0.94
rs602633	1	109278889	PSRC1-CELSR2-SORT1	T	G	1.14	0.26	0.99	0.77	0.98
rs4129267	1	154453788	IL6R	T	C	1.14	0.86	1.0	0.60	1.0
rs1795061	1	214235937	SMYD2	C	T	1.13	0.52	0.99	0.31	0.99
rs7255	2	20679060	AC012065.7-LDAH	T	C	1.10	0.41	0.99	0.54	0.98
rs10023907	4	87851383	MEPE	C	T	1.09	0.64	1.0	0.66	1.0
rs3176336	6	36681039	CDKN1A	A	T	1.10	0.79	0.99	0.39	0.99
rs118039278	6	160564494	LPA	G	A	1.28	0.01	0.95	0.07	0.95
rs10808546	8	125483576	RP11–136O12.2-TRIB1	T	C	1.10	0.67	0.99	0.56	1.0
rs10757274	9	22096056	CDKN2BAS1-ANRIL	A	G	1.24	0.21	0.99	0.48	0.97
rs10985349	9	121662964	DAB2IP	C	T	1.17	0.16	1.0	0.18	0.98
rs1412445	10	89243047	LIPA	C	T	1.10	0.39	0.99	0.33	0.99
rs964184	11	116778201	ZNF259-APOA5	C	G	1.18	0.20	0.99	0.14	0.99
rs4936098	11	130410772	ADAMTS8	A	G	1.13	0.78	1.0	0.63	1.0
rs9316871	13	22287782	LINC00540	G	A	1.15	0.83	1.0	0.77	1.0
rs55958997	15	78623530	CHRNA3	C	A	1.12	0.26	0.99	0.37	0.99
rs35254673	16	84885919	CRISPLD2	A	G	1.09	0.07	0.99	0.26	0.98
rs4401144	18	22613705	CTAGE1	C	T	1.11	0.66	1.0	0.49	1.0
rs6511720	19	11091630	LDLR	T	G	1.24	0.87	0.95	0.88	0.91
rs429358	19	44908684	APOE	T	C	1.17	0.21	0.95	0.15	0.99
rs3827066	20	45957384	PCIF1-ZNF335-MMP9	C	T	1.22	0.04	0.99	0.16	0.99
rs73149487	20	63849998	ABHD16B	T	G	1.26	0.99	0.90	0.96	0.96
rs2836411	21	38447907	ERG	C	T	1.11	0.32	1.0	0.34	1.0

^*Based on Genome Reference Consortium Human Build 38

^†Modeled allele related to greater odds of AAA

^**Previously reported by Klarin et al. (14) or Jones et al. (15)

Abbreviations: AAA=abdominal aortic aneurysm; ARIC=Atherosclerosis Risk in Communities; rsID=reference single nucleotide polymorphism cluster identification number; Chr=chromosome; Ref=reference; freq=frequency; R² = estimated imputation accuracy

Proteins with the strongest associations with the AAA variants as well as those subsequently found to be related to incident AAA (e.g., neogenin and kit ligand) are shown in Table 3. Among the nineteen plasma proteins related to the rs602633 variant (proximal to PSRC1-CELSR2-SORT1), associations were observed for granulin (p<1.0×10⁻²⁰⁰), complement C1q TNF-related protein 1 (CTRP-1) (p=2.2×10⁻¹⁸⁴), and neogenin (p=5.6×10⁻¹³); results were replicated in Black participants for granulin (p=1.8×10⁻⁸) and CTRP-1 (p=1.8×10⁻⁴) but not for neogenin. Associations were observed between rs11591147 (PCSK9) and plasma PCSK9 (p=4.2×10⁻³⁴) and between rs3827066 (PCIF1-ZNF335-MMP9) and kit ligand (p=2.4×10⁻⁸) among White individuals, and the former was replicated in Black participants (p=0.01). The rs4129267 risk allele (IL6R) was significantly associated with five plasma proteins in Whites. Of these, moderate to strong associations with soluble interleukin-6 receptor subunit alpha (sIL-6Ra) (p<1.0×10⁻²⁰⁰) and C-reactive protein (p=2.2×10⁻⁷) were observed in White individuals with the former replicated in Black participants (p=6.3×10⁻¹¹⁶) (Table 3). The rs429358 risk allele (APOE) was associated with 68 plasma proteins; of these, top associations were observed for neurofilament light chain (p<1.0×10⁻²⁰⁰), leucine-rich neuronal protein-1 (p<1.0×10⁻²⁰⁰), and tubulin specific chaperone A (p<1.0×10⁻²⁰⁰), all of which were replicated in Black participants. Finally, the rs964184 risk allele (ZNF259/APOA5) was associated with 33 plasma proteins; of these, top associations were observed for apolipoprotein A-V (p=6.0×10⁻⁹⁵) and protein phosphatase inhibitor 2 (p=7.7×10⁻⁴⁷) with the former replicated in Black participants (p=3.9×10⁻⁴).

Table 3.

Select^* associations of AAA-related variants with plasma proteins and corresponding standard deviations (SD) in ARIC participants. The proximal gene(s) to each variant are indicated.

		White participants		Protein SD	Black participants		Protein
Variant	Protein	Estimate† (95% CI)	p-value		Estimate† (95% CI)	p-value	SD
rs602633	Granulin	0.184 (0.175, 0.193)	<1.0E-200	0.25	0.080 (0.058, 0.102)	1.8E-08	0.29
PSRC1-CELSR2-SORT1	CTRP-1	0.115 (0.108, 0.123)	2.2E-184	0.21	0.051 (0.03, 0.07)	1.8E-04	0.23
	Neogenin	−0.031 (−0.040, −0.023)	5.6E-13	0.22	−0.011 (−0.030, 0.008)	2.5E-01	0.24
rs4129267	sIL-6Ra	−0.339 (−0.35, −0.33)	<1.0E-200	0.34	−0.348 (−0.38, −0.32)	6.3E-116	0.32
IL6R	C-reactive protein	0.093 (0.06, 0.13)	2.2E-07	1.08	0.073 (−0.179, 0.033)	1.8E-01	1.04
rs3827066	Kit ligand	−0.045 (−0.061, −0.029)	2.4E-08	0.39	−0.016 (−0.076, 0.044)	6.1E-01	0.40
PCIF1-ZNF335-MMP9
rs11591147	PCSK9	0.275 (0.231, 0.319)	4.2E-34	0.33	0.235 (0.053, 0.417)	1.1E-02	0.36
PCSK9
rs429358	NEFL	−0.608 (−0.632, −0.585)	1.0E-200	0.46	−0.65 (−0.694, −0.606)	6.0E-151	0.54
APOE	LLR neuronal protein 1	0.305 (0.292, 0.319)	1.0E-200	0.37	0.339 (0.313, 0.366)	1.0E-118	0.40
	TBS chaperone A	−0.685 (−0.707, −0.664)	1.0E-200	0.60	−0.712 (−0.757, −0.667)	5.2E-167	0.64
rs964184	Apolipoprotein A-V	0.346 (0.313, 0.378)	6.0E-95	0.72	0.103 (0.046, 0.16)	3.9E-04	0.69
ZNF259/APOA5	PPP1R2	0.104 (0.09, 0.118)	7.7E-47	0.34	0.013 (−0.013, 0.04)	3.2E-01	0.33

^*The strongest protein associations for each SNP are shown here with exceptions for proteins subsequently found to be related to incident AAA: kit ligand, neogenin, and C-reactive protein.

^†Change in log2 protein measurement per copy of the AAA at-risk allele

Linear regression model adjusted for age, sex, field center, ten principal components of ancestry, and estimated glomerular filtration rate.

Bonferroni correction for multiple testing, significant associations at p<1.14E-6 in Whites. The analysis in Black participants was treated as replication of associations observed in White participants. Significance thresholds in Black participants were as follows: rs3827066: p≤0.025; rs4129267: p≤0.01; rs11591147: p≤0.05; rs118039278: p≤0.05; rs10808546: p≤0.125.

Abbreviations: AAA=abdominal aortic aneurysm; ARIC=Atherosclerosis Risk in Communities study; sIL-6Ra=soluble interleukin-6 receptor subunit; PCSK9=Proprotein Convertase Subtilisin/Kexin Type 9; NEFL=Neurofilament light polypeptide; LLR=Leucine-rich repeat; TBS=Tubulin-specific; PPP1R2=Protein phosphatase inhibitor 2

Clinical AAA Incidence

Of the 132 SNP-protein associations in the main analysis, 118 unique proteins were identified. Of these, kit ligand and neogenin were significantly and inversely related to the risk of incident AAA over a median of 21.2 (maximum 23.8) years of follow-up (p<4.0×10⁻⁴) after adjustment for age, sex, race, field center, smoking status, and eGFR (Table 4, Model 1). Compared to those in bottom quintiles, participants in the top quintiles of plasma kit ligand and neogenin were found to be at 44% (p=4.7×10⁻⁴; p for trend=1.4×10⁻⁵) and 43% lower risks of AAA (p=3.8×10⁻⁴; p for trend=1.3×10⁻⁴), respectively. Magnitudes of associations were marginally attenuated with additional adjustment for BMI, prevalent status for hypertension, diabetes, CHD, and stroke, total cholesterol, and use of lipid lowering medication (Table 4, Model 2); the association of neogenin became borderline significant based on the Bonferroni corrected threshold of p=4×10^-4. The modifying effects of sex and age (stratified by tertiles) were tested for in the above associations and were found to be null for both kit ligand (p=0.59 for interaction by sex; p=0.36 for interaction by age) and neogenin (p=0.27; p=0.58, respectively). Race interactions were also tested, and results are shown in Supplementary Table S11. No modifying effect of race was observed for associations of kit ligand (p for interaction=0.91) and neogenin (p for interaction=0.59) with risk of incident AAA.

Table 4.

Cox regression analysis of plasma proteins and incident, clinical AAA among ARIC participants over a median 21.2-year follow-up.

Plasma protein Related variant; proximal genes		Plasma protein quintile [Hazard ratio (95% confidence interval), p-value]
	1	2	3	4	5	p for trend
Kit ligand rs3827066; PCIF1-ZNF335-MMP9
Model 1	ref	1.09 (0.84, 1.40) 0.53	0.82 (0.62, 1.08) 0.15	0.71 (0.52, 0.95) 0.02	0.56 (0.40, 0.78) 4.7E-4	1.4E-5
Model 2	ref	1.08 (0.83, 1.40) 0.57	0.79 (0.59, 1.06) 0.11	0.73 (0.53, 1.00) 0.05	0.59 (0.42, 0.82) 0.002	1.3E-4
Neogenin rs602633; PSRC1-CELSR2-SORT1
Model 1	ref	0.87 (0.65, 1.15) 0.31	0.82 (0.62, 1.08) 0.16	0.69 (0.52, 0.93) 0.01	0.57 (0.42, 0.78) 3.8E-4	1.3E-4
Model 2	ref	0.85 (0.64, 1.14) 0.28	0.83 (0.62, 1.10) 0.19	0.73 (0.55, 0.99) 0.04	0.64 (0.46, 0.88) 0.006	0.004

Model 1: adjusted for age, sex, race, field center, smoking status, and estimated glomerular filtration rate (N=11,064; 454 AAA events). Bonferroni correction stipulated a significance threshold of p<4.2E-4 for 118 tests.

Model 2: model 1 + BMI, prevalent status for diabetes, hypertension, CHD, and stroke, total cholesterol levels, and use of lipid lowering medication (N=10,701; 434 AAA events).

Abbreviations: ARIC = Atherosclerosis Risk in Communities; AAA=abdominal aortic aneurysm.

Associations of severe AAA incidence with kit ligand and neogenin (per SD given the small n of AAA cases) are shown in Supplementary Table S12. The direction of associations between these two proteins and severe AAA is consistent with that for all incident clinical AAA, with the association between kit ligand and severe AAA reaching statistical significance in both models 1 (p=0.002) and model 2 (p<0.001).

C-reactive protein (CRP) showed a borderline and positive association with the risk of AAA (p for trend =0.001 and 0.003, data not shown)—a well-established association that has been previously reported in ARIC using an immunoassay method ³. The remaining proteins were not significantly associated with incident AAA after correction for multiple comparisons testing (associations shown in Supplementary Table S13).

Plasma levels of IL-6 and sIL-6Ra were both differentially associated with clinical AAA incidence (Table 5). Higher levels of sIL-6Ra were associated with lower risk of AAA but did not reach significance (p for trend=0.12). By contrast, plasma IL-6 levels were confirmed to be significantly related to greater risk of incident AAA ⁵. Mutual adjustment did not materially change the associations for either protein, and levels of these two proteins were not correlated (Pearson correlation=0.008, p=0.41).

Table 5.

Cox regression analysis of IL-6 and sIL-6Ra with incident AAA among ARIC participants over a median 21.2-year follow-up.

Plasma protein	Plasma protein quintile [Hazard ratio (95% confidence interval), p-value]
	1	2	3	4	5	p for trend
sIL-6Ra
Model 1	ref	1.10 (0.82, 1.49) 0.40	1.05 (0.77, 1.42) 0.27	0.84 (0.61, 1.15) 0.77	0.87 (0.64, 1.20) 0.53	0.12
Model 1 + IL-6	ref	1.10 (0.81, 1.48) 0.54	1.04 (0.77, 1.41) 0.81	0.83 (0.61, 1.14) 0.26	0.87 (0.63, 1.19) 0.37	0.10
IL-6
Model 1	ref	0.96 (0.69, 1.33) 0.79	1.21 (0.88, 1.65) 0.24	1.29 (0.94, 1.76) 0.11	1.36 (1.00, 1.85) 0.05	0.01
Model 1 + sIL-6Ra	ref	0.97 (0.70, 1.34) 0.84	1.22 (0.89, 1.67) 0.22	1.30 (0.95, 1.78) 0.10	1.37 (1.01, 1.87) 0.04	0.008

Model 1: adjusted for age, sex, race, field center, smoking status, estimated glomerular filtration rate, BMI, prevalent status for diabetes, hypertension, CHD, and stroke, total cholesterol levels, and use of lipid lowering medication (N=10,701; AAA events = 434); significance threshold of p<0.05

Abbreviations: ARIC = Atherosclerosis Risk in Communities; AAA=abdominal aortic aneurysm; sIL-6Ra=soluble interleukin-6 receptor subunit; IL-6=interleukin-6

Ultrasound-detected asymptomatic AAA

Of the two novel proteins found to be significantly associated with risk of clinical AAA incidence, neogenin was observed to be inversely and significantly associated with risk of ultrasound-detected asymptomatic AAA (OR=0.70; 95% CI: 0.53–0.94; p=0.02), independent of age, sex, race, cigarette smoking status, estimated glomerular filtration rate, and prevalent coronary heart disease (Supplementary Figure S1). Kit ligand was not associated with ultrasound-detected AAA (OR=0.98; 95% CI: 0.67–1.43).

Mendelian randomization

Two-sample Mendelian randomization analysis was conducted to assess the potential for causal relationships between kit ligand or neogenin and AAA. Estimates and 95% CIs are shown in Figure 1 for individual pQTLs as well as results of IVW and MR-Egger analyses. For kit ligand, the IVW estimate indicated a 33% lower odds of AAA per SD increase in log-base 2 protein levels (p=1.4×10⁻⁴), suggesting a causal effect. The MR-Egger-derived estimate showed a greater magnitude of association compared to the IVW result; however, the association was statistically borderline for the more conservative MR-Egger analysis (OR per SD=0.54; p=0.055). Results for neogenin were non-significant for both IVW (OR per SD=0.79; p=0.23) and MR-Egger (OR per SD=0.92; p=0.86) but were directionally consistent with the prospective analysis of neogenin and incident AAA. MR-Egger pleiotropy tests for both kit ligand and neogenin were null (p>0.05), indicating no detectable directional pleiotropy; however, MR-PRESSO global tests indicated the presence of horizontal pleiotropy for both kit ligand (p=0.004) and neogenin (p=0.009). MR-PRESSO outlier tests identified pleiotropic outlier variants for kit ligand (rs247616, p<0.007) and neogenin (rs13379875, p<0.005; rs7528419, p<0.005). Once these outlier variants were removed, corrected-IVW causal estimates were generated. The IVW estimate for kit ligand was attenuated but remained significant (OR per SD=0.72; p=3.3×10⁻⁴), and the estimate for neogenin reached significance (OR per SD=0.50; p=0.03) (Figure 1).

Figure 1.

Two-sample Mendelian randomization examining odds ratio of abdominal aortic aneurysm per SD change in A) kit ligand; and B) neogenin. For the corrected inverse variance weighted (IVW) estimates, rs13379875 and rs7528419 outliers were removed for neogenin and rs247616 for kit ligand based on MR-PRESSO outlier tests.

Pathway analysis

Pathway analyses were conducted based on risk allele-protein associations. Associations between plasma proteins and the rs602633 (PSRC1-CELSR2-SORT1) risk allele provided evidence of atherosclerosis signaling, glutathione-mediated detoxification, pregnane X receptor/retinoid X receptor activation, and eicosanoid signaling. For the rs4129267 (IL6) risk allele, associations were indicative of signaling involving IL-6, the acute phase response, neuroinflammation, and IL-17. For the rs3827066 (PCIF1-ZNF335-MMP9) risk allele, associations were indicative of hematopoiesis of multi- and pluripotent stem cells, melanocyte development, and acute myeloid leukemia signaling pathways. For the above pathways, IPA-derived z-score values did not reach thresholds to indicate either activation (z-score ≥ 2) or inhibition (z-score ≤ −2) and therefore pathway status could not be discerned definitively.

By contrast, protein associations with the rs964184 (ZNF259/APOA5) risk allele revealed LXR/RXR activation (z-score=2.0; p-value=0.0001) and a net upregulation in hepatic fibrosis signaling (z-score=2.0; p-value=0.009). Protein associations for the rs429358 (APOE) risk allele suggested upstream inhibition of N32 (z-score=−2.0; p-value=4.7×10⁻⁶) and N34 (z-score=−2.0; p-value=5.8×10⁻⁶) elovanoid (ELV) signaling. Associations of pathway components with corresponding risk alleles are shown in Supplementary Table S14.

DISCUSSION

To the best of our knowledge, this is the first agnostic, large-scale proteomics examination of genetic risk of AAA. Of the 23 risk variants previously reported and confirmed from large AAA GWASs ^14,15, eight variants were associated with 132 proteins among White ARIC participants. Of these, 59 variant-protein associations were replicated in Black participants, and 101 associations were directionally consistent. In follow-up analyses, we found that higher baseline levels of kit ligand and neogenin were moderately associated with lower risk of incident clinical AAA over a median of 21.2 years of follow-up. Neogenin was also significantly and inversely associated with the risk of ultrasound-detected asymptomatic AAA. MR IVW analyses following removal of outlier SNPs indicated that lower levels of kit ligand and neogenin may be causally related to AAA risk. The associations of circulating kit ligand and neogenin with clinical AAA have not been previously reported in human studies.

Novel AAA risk factors kit ligand and neogenin

Kit ligand and neogenin may represent independent risk factors for AAA, and their known functions appear to coincide with currently known mechanisms of AAA pathogenesis. Kit ligand was identified through its association with rs3827066 (proximal to PCIF1-ZNF335-MMP9)—a variant previously associated with coronary artery disease ³³. Functionally, kit ligand binds the protein-tyrosine kinase receptor, c-kit, which is upregulated in AAA tissue ³⁴ and has been shown to mediate cell survival and proliferation, endothelial permeability ³⁵, and inflammation ³⁶. In the context of vascular disease, kit ligand has been reported to have a protective role and may suppress atherosclerosis in hyperlipidemia through c-kit ³⁷, while its deficiency promotes a pro-inflammatory phenotype in smooth muscle cells that is ameliorated by restoring its expression ³⁶. While this evidence is not definitive, lower levels of kit ligand appear to promote inflammation and disruption of ECM integrity—two characteristic features of AAA. This aligns with its association with lower risk of incident AAA found here and is further supported by the significant MR IVW estimates.

Like kit ligand, plasma neogenin levels were inversely related to clinical AAA incidence but also showed a significant and uniform association with ultrasound-detected asymptomatic AAA in an analysis of 57 total cases. Neogenin was identified through its association with rs602633 (proximal to PSRC1-CELSR2-SORT1), which has been associated with both total cholesterol ³⁸ and coronary artery disease ³³—known risk factors or confounders for AAA. And yet, our prospective analysis of neogenin with incident AAA was independent of coronary heart disease, total cholesterol, and lipid lowering medication use. From a physiological standpoint, neogenin is a cell surface receptor that regulates cell adhesion and binds multiple ligands including macrophage-derived netrin-1. Paradoxically, and in the context of AAA, netrin-1 and neogenin activate a signaling cascade that results in MMP-3 expression in vascular smooth muscle ³⁹. MMP-3 has been shown to be a necessary component in perpetuating inflammation and erosion of the ECM in experimental models of AAA and has also been detected in isolated AAA tissue in humans ³⁹. The inverse relationship of circulating neogenin with both clinical and subclinical AAA is counter to this experimental evidence. While the initial MR IVW and MR-Egger findings indicated that neogenin is not a causal factor in AAA risk, removing outlier SNPs that were identified through MR-PRESSO resulted in a significant IVW after this correction. While these findings should be interpreted with caution, this is the first study to show that low plasma neogenin levels are associated with greater risk of AAA.

Other potential protein risk factors

We observed a strong association of CTRP1 with the rs602633 variant (p<2.2×10⁻¹⁸⁴). CTRP1 has been described in inflammation and cardiovascular disease ^40,41; in our study, a nominal association was observed with incident AAA independent of cardiovascular and AAA risk factors including total cholesterol and lipid lowering medication use (p=0.03 for trend). Follow-up research of CTRP1 and AAA risk is warranted. Finally, the strong inverse association of rs4129267 (IL6R) with sIL-6Ra (−0.34; p<1×10⁻²⁰⁰) and its positive association with CRP (0.093; p=2.2×10⁻⁷) confirm previous findings of this and other IL6R gene variants ⁴². Mendelian randomization ^43,44 and experimental studies ^45,46 have indicated that IL-6 signaling promotes AAA pathogenesis, and pharmacologic inhibition of IL-6 reduces STAT3 activation, MMP expression, AAA development and severity ⁴⁵. These phenomena coincide with known IL-6 mechanisms of vascular pathophysiology involving inflammation, ECM remodeling, as well as the opposing associations of IL-6 and sIL-Ra with incident AAA observed here.

Pathway Analysis

Results of pathway analysis were largely inconclusive or biologically dubious, but some require additional scrutiny. Proteins associated with ZNF259/APOA5 variant indicated activation of hepatic fibrosis signaling, which has been reported in ascending thoracic aortic aneurysm tissue ⁴⁷. It has further been suggested that the fibrotic remodeling observed in aortic aneurysms may share characteristics with that of hepatic fibrosis ⁴⁸. For the APOE risk variant, protein associations indicated inhibition of the recently identified N32 and N34 elovanoids (ELVs). ELVs are lipid mediators generated by the ELOVL4 enzyme from omega-3 fish oil fatty acids and have been shown to suppress expressions of MMPs and IL-6—well-described AAA risk markers. And yet, the localization of ELVs does not appear to overlap with that of AAA-related tissue compartments at the present time ^49–51. Follow-up research in tissues directly affected by AAA is warranted.

Strengths and Limitations

This study was composed of a large community-based cohort of Black and White study participants. Proteins were measured using a modified DNA-based aptamer array that has undergone rigorous quality control and reproducibility testing ^23–27. To limit the probability of reverse causation, proteins identified in the main analysis were examined prospectively to determine their associations with the risk of incident AAA. MR analyses were performed to obtain evidence of causal relationships between the proteins and AAA risk. Finally, conservative multiple testing corrections were applied in statistical analyses to limit type I errors.

A number of study limitations must be acknowledged. First, AAA genetic variants have relatively small magnitudes of association with odds of AAA ^14,15. While convergence among their pathways appeared plausible and would result in stronger AAA variant-protein associations, we found that each variant was associated with largely non-overlapping proteomic signatures—indicating limited convergence. AAA risk variants were selected from a meta-analysis of GWAS comprised of ancestrally European individuals. These variants may not as effectively capture the genetic risk of AAA in Black individuals as in Whites. Moreover, the smaller sample size and ancestry-based difference in linkage disequilibrium in the ARIC Black sample may account for some of the non-replicated findings. It must also be recognized that we can only infer that plasma protein levels are correlated with local tissue levels, and follow-up histological studies are warranted to determine whether proteins like kit ligand and neogenin are localized to AAA tissue in nascent or later stages of disease.

Limitations in the MR analysis must also be acknowledged. First, no cis-acting IVs were identified for kit ligand, and only one was identified for neogenin. The inclusion of trans-acting IVs increases the likelihood of horizontal pleiotropy, and this is indicated by the MR-PRESSO global test. By conducting the MR-PRESSO outlier tests and removing SNPs identified by the outlier tests, the possibility of horizontal pleiotropy was likely reduced. Second, there was also limited statistical power to identify pQTLs in Black participants, and we were unable to perform MR analysis in this group. Third, associations between the exposure IVs and AAA outcomes were identified using the AAAgen consortium, which is comprised of multiple studies including ARIC and the deCODE Health study. Since we also used the latter cohort to identify protein-IV associations ²⁹—which were then replicated in the ARIC cohort—the overlap between the pQTL and AAAgen samples can bias the exposure-outcome association away from the null ⁵². However, the contributions of AGES and ARIC to AAAgen were moderate (AIRC: n=8,962 and deCode=267,066 out of 1,125,328 in AAAgen), and therefore the risk of bias was deemed acceptable.

Conclusions

The present study identified over one-hundred proteomic signatures associated with AAA risk variants that coincide with disease pathogenesis including inflammation, endothelial dysfunction, and ECM and fibrotic remodeling. The potential causal roles of low kit ligand and neogenin levels in AAA are novel observations. These observations may link genetic predisposition to disease manifestation, benefit the prevention or treatment of AAA through improved risk stratification, or provide new therapeutic targets. Given our data-driven approach, further studies of the identified proteins and their potential roles in the pathogenesis AAA development are warranted.

Highlights.

• Using a data-driven approach, the proteomics of abdominal aortic aneurysm (AAA) genetic liability was examined among 1,671 Black and 7,241 White participants of the Atherosclerosis Risk in Communities Study.

• Over one-hundred proteomic signatures associated with AAA risk variants were identified and coincided with known mechanisms of AAA pathogenesis including inflammation, endothelial dysfunction, and extracellular matrix and fibrotic remodeling.

• Potential causal roles of low kit ligand and neogenin levels in AAA were identified through two-sample Mendelian randomization analysis and are novel observations that may link genetic predisposition to disease manifestation.

Non-Standard Abbreviations

AAA

Abdominal Aortic Aneurysm

ECM

Extracellular Matrix

MMP

Matrix Metalloproteinase

ARIC

Atherosclerosis Risk in Communities

GFR

Glomerular Filtration Rate

CRP

C-Reactive Protein

CTRP1

Complement C1q TNF-Related Protein-1

sIL-6Ra

Soluble Interleukin-6 Receptor Alpha

MR

Mendelian randomization

PRESSO

Pleiotropy residual sum and outlier