Lower plasma ApoA1 levels are found in Parkinson’s disease and associate with APOA1 genotype

Abstract

Background

The discovery of novel plasma-based biomarkers could lead to new approaches in the treatment of Parkinson’s disease (PD). Here we explore the role of plasma Apolipoprotein A1 (ApoA1) as a risk marker for PD and evaluate the influence of APOA1 promoter variation on plasma ApoA1 levels.

Methods

Plasma ApoA1 and the single nucleotide polymorphism rs670 were assayed in a discovery cohort (Cohort 1) of 301 PD, 80 normal controls, and 165 subjects with other neurodegenerative diseases, as well as a cohort (Cohort 2) of 158 PD patients from a second clinical site. Additionally, rs670 was genotyped in a third cohort of 1494 PD and 925 normal control subjects from both clinical sites.

Results

Compared with both normal and disease controls, PD patients have lower plasma ApoA1 (p<0.001 for both comparisons). Moreover, in PD patients, plasma ApoA1 levels are correlated with genotype at the APOA1 promoter polymorphism, rs670. Specifically, lower plasma ApoA1 levels were found in rs670 major allele (G) homozygotes in both Cohort 1 (p=0.009) and in a replication cohort (Cohort 2, n=158 PD patients, p=0.024). Finally, evaluating rs670 genotype frequencies in 1930 PD cases vs. 997 normal controls, the rs670 GG genotype shows a trend towards association (OR: 1.1; p=0.10) with PD.

Conclusions

Our results are compatible with a model whereby circulating ApoA1 levels may be useful in risk-stratifying subjects for the development of PD, with higher ApoA1 levels suggesting relative protection. Future studies evaluating modulation of ApoA1 as a novel therapeutic strategy in PD are warranted.

Introduction

Affecting one million people in the US alone, Parkinson’s disease (PD) is a major public health problem, with no treatments available to slow disease progression.¹ Clinical symptoms result from neurodegeneration in susceptible areas of the brain, with dopaminergic neurons of the substantia nigra particularly vulnerable. At the time of clinical diagnosis, 30–50% of substantia nigra dopaminergic neurons may already be lost, highlighting the long prodromal phase.² Thus, markers of PD risk, particularly potentially-modifiable markers of PD risk, would be important in understanding and treating PD.

Apolipoprotein A1 (ApoA1) is a major component of high-density lipoprotein (HDL) and has been extensively studied in the cardiovascular literature, where higher levels have been associated with protection from vascular disease.³ Recently, ApoA1 has been linked to PD in several ways. We previously identified ApoA1 as a candidate biomarker for PD risk, from an unbiased discovery screen of 96 plasma proteins.⁴ Specifically, we found that lower ApoA1 levels were associated with earlier age at disease onset in PD patients. Furthermore, we showed that in asymptomatic individuals at high risk for developing PD, lower plasma ApoA1 correlated with decreased putaminal dopamine transporter binding. In addition, others reported that ApoA1 is lower in the postmortem CSF of PD patients compared to normal controls,⁵ and that differential ApoA1 isoform expression may be observed in the CSF of PD patients vs. controls.⁶ Finally, two large epidemiological studies from the United States⁷ and Taiwan⁸ recently reported a decreased risk of developing PD among individuals taking statin medications. Since statins increase ApoA1 and HDL levels, these data suggest that lower ApoA1 plasma levels may be associated with increased PD risk.

In the cardiovascular literature, ApoA1 levels have been reported to be genetically influenced, with a single nucleotide polymorphism (SNP) at position −75bp (rs670) in the promoter of the APOA1 gene linked with ApoA1 expression.⁹ Specifically, the A allele of rs670 was first reported to correlate with higher circulating ApoA1 and HDL levels, with studies assuming an A-dominant model.^9,10 However, other groups have been unable to replicate this expression quantitative trait locus (eQTL) effect.^11,12 Furthermore, the eQTL effect has been reported to interact with other demographic and/or lifestyle factors^13–15 and studies investigating the regulatory mechanism have yielded conflicting data.^16,17 In the realm of neurological diseases, the rs670 polymorphism has been reported to influence risk for cognitive decline in multiple sclerosis¹⁸ and in Alzheimer's Disease.¹⁹ However, the genotype associated with increased risk for poor outcome has not been consistent^{18, 19} and none of the neurological studies measured circulating ApoA1 levels. Thus, while some data suggests that the A allele at rs670 may associate with increased ApoA1 levels,⁹ and that rs670 genotype may affect neurological outcomes, further evaluation of eQTL effects of rs670 genotype on circulating ApoA1 levels in neurological disease is still needed.

With the preceding data in mind, we hypothesized that lower levels of plasma ApoA1 may not only correlate with PD, but in fact predispose individuals to development of PD. From this model, we developed three predictions and examined them in cohorts from two independent clinical sites. These predictions were: (1) PD patients would have lower levels of ApoA1 than controls, (2) lower ApoA1 levels would be correlated with GG genotype at rs670 in PD patients, and (3) the rs670 GG genotype would be enriched in PD, suggesting that APOA1 genotype is an upstream risk factor for PD, acting through reduction of ApoA1 levels.

Patients and Methods

For additional details, please see Supplement.

Subjects

Subjects in Cohort 1 were recruited from the University of Pennsylvania (UPenn) and consisted of PD patients (n=301), subjects with other neurodegenerative diseases (ND; n=165), and normal controls (NCs; n=80; see Table 1 and Suppl. Tables 1 and 2 for additional data). Plasma, DNA, clinical and demographic information was collected from these subjects. Cohort 2 comprised 158 PD patients from the Pacific Northwest Udall Center (PANUC) for whom we previously reported plasma ApoA1 levels,⁴ now newly evaluated for rs670 genotype.

Table 1.

Demographic information for clinical cohorts of Parkinson’s disease (PD) and normal controls (NC).

Clinical/ Demogra phic Features	Cohort 1			Cohort 2		Cohort 3
	PD	NC	P	PD	NC	PD	NC	P
Number Female/Male (n)	301 116/185	80 45/35	0.04a	158 49/109	N/A	1494 444/1050	925 581/344	<0.001a
Age at DNA Median yrs (IQR)	67.0 (60.0–74.0)	66.0 (61.0–77.0)	nsb	65.0 (59.0–71.3)	N/A	68.0 (60.0–75.0)	66.0 (57.0–74.0)	nsb
Age at plasma Median yrs (IQR)	67.0 (60.0–75.0)	65.50 (59.3–75.0)	nsb	67.43 (62.5–75.5)	N/A	N/A	N/A	-
Age at PD Onset Median yrs (IQR)	59.0 (52.9–67.0)	N/A	-	58.0 (65.0-50.5)	N/A	60.0 (51.0–67.0)	N/A	-
Disease Duration Median yrs (IQR)	7.0 (4.0–11.0)	N/A	-	6.0 (3.0–11.0)	N/A	7.0 (4.0–11.0)	N/A	-
UPDRS Motor Score Median score (IQR)	23.0 (14.0–33.0)	N/A	-	27.0 (19.0–35.0)	N/A	N/A	N/A	-
Modified Hoehn and Yahr stage (IQR)	2.0 (2.0–3.0)	N/A	-	2.5 (2.0–3.0)	N/A	N/A	N/A	-
MoCA (IQR)	25.0 (21.0–27.0)	27.0 (25.0–29.0)	0.001b	25.0 (23.0–27.0)	N/A	N/A	N/A	-

Abbreviations: IQR, interquartile range; NC, Normal Controls; PD, Parkinson’s disease

^aχ²test,

^bMann-Whitney test.

Cohort 3 consisted of an additional set of 1494 PD patients and 925 NCs included only for genotyping – these subjects were from UPenn (416 PD and 123 NC) and from the Parkinson’s Genetic Research (PaGeR) Study based at the Veterans Affairs Puget Sound Health Care System in Seattle (1078 PD and 802 NC).

PD patients in all three cohorts met United Kingdom Parkinson’s Disease Society Brain Bank clinical diagnostic criteria for PD.²⁰

IRB approval for these studies was obtained from each institution.

Plasma ApoA1

Blood was collected and processed for plasma as previously described.²¹ Plasma levels of ApoA1 were measured by enzyme-linked immunosorbent assay (ELISA, Abnova, Taiwan, Catalog #0462) as previously described.⁴ Samples were run in duplicate (<1% of samples had CV>0.2). All samples were processed on one lot of ELISAs, on the same day, by the same operators, if possible. Where not possible to assay all samples on one day, control and disease samples were randomly allocated to plates, and operators blinded to disease status.

ApoA1 Genotyping

DNA was extracted from EDTA plasma as previously described.²² The SNP rs670 in the APOA1 promoter was genotyped using TaqMan allelic discrimination assays (Applied Biosystems, Foster City, CA) and a custom PANDORA (PAn Neurodegenerative Disease Oriented Risk Allele) panel for multiple parallel PCR-based genotyping reactions (Sequenom, San Diego), as previously described.²³

Determination of linkage disequilibrium structure and analysis of existing PD genome-wide association data

Linkage disequilibrium (LD) and haplotype structures around rs670 were analyzed with Haploview, using data from the CEU population from the 1000 Genomes Project.²⁴

We then interrogated whether any SNPs genotyped in the NINDS PD genome-wide association study (GWAS)²⁵ were strongly linked with rs670. We interrogated the 100 kb region surrounding rs670 for SNPs genotyped in publicly available PD GWAS data,²⁵ estimating LD between each SNP and rs670, from the 1000 Genomes project in the CEU population.²⁴

Statistical analyses

Linear regression analyses evaluated the influence of rs670 genotype and/or disease category on plasma ApoA1 levels, with age and gender as covariates. Logistic regressions assessed whether plasma ApoA1 levels were predictive of PD status, with and without adjustment for rs670 genotype and other covariates, as described in the text.

For genetic association analyses, Cochran-Armitage tests of trend were used for the co-dominant model, and Chi-square analyses were used for the minor-allele-dominant model.

All statistical analyses were performed in Graphpad Prism (San Diego, CA), or R statistical software (www.r-project.org). R scripts are available upon request. Because directionality was hypothesized for all major analyses, one-tailed p-values are reported.

Results

Plasma levels of ApoA1 are lower in PD patients

Plasma ApoA1 from 301 PD patients, 80 NC, and 165 individuals with other ND was measured by ELISA as previously described (Cohort 1, see Table 1, Suppl. Table 1)⁴

As predicted, PD patients had lower levels of ApoA1 vs. NC or individuals with other ND (p<0.001, Fig. 1a, Table 2, Suppl. Fig. 1). Women had higher ApoA1 levels than men across all groups, as previously reported (Suppl. Fig. 2).²⁶ Adjusting for age and sex, either alone or together, however, did not change our result (p<0.001 for difference between PD and other groups in ApoA1 levels with or without age and sex as covariates, Table 2, Suppl. Fig. 1).

Figure 1. Plasma levels of ApoA1 differ by disease category and by rs670 genotype.

In Cohort 1, plasma ApoA1was lower in PD vs. other subgroups (a); area under the curve (AUC) for plasma Apoa1 was 0.68 (95% CI 0.65–0.75, p<0.001), with ROC classifying PD vs. normal shown in red (b); and rs670 GG carriers had lower plasma ApoA1 levels (c). In Cohort 2 (PD only), plasma ApoA1 levels were lower in rs670 GG carriers (d). Means +/− SEM are indicated for each group. *p<0.05. **p<0.01. ***p<0.001.

Table 2.

Parkinson’s Disease patients have lower plasma apolipoprotein A1 levels.

Outcome	Predictor(s)a	P-value		Direction
Plasma ApoA1	PD	PD	p<0.001	−
	PD + Age + Sex	PD Age Male Sex	p<0.001 p<0.001 p<0.001	− + −
Outcome	Predictor(s)a	P-value	Direction
PD vs. Normal	Plasma ApoA1	ApoA1	p=0.002	Lower in PD
	Plasma ApoA1 + Age + Sex	ApoA1 Age Male Sex	p<0.001 p<0.001 p=0.326	Lower in PD Higher in PD n/a
	Plasma ApoA1 + Age + Sex + rs670	ApoA1 Age Male Sex rs670 GG	p<0.001 p<0.001 p=0.329 p=0.207	Lower in PD Higher in PD n/a n/a

Abbreviations: ApoA1, Apolipoprotein A1; n/a, not applicable; PD, Parkinson’s disease.

^aIn linear regression (top) and logistic regression (bottom) models, plasma apolipoprotein A1 (ApoA1) levels were significantly associated with PD state, with lower levels seen in PD patients. This association persisted after adjustment for age and sex, as well as genotype at rs670. For logistic regressions, only two categories (PD vs. normal) were evaluated as outcomes, under a binomial regression model. Results shown are from Cohort 1, since Cohort 2 does not have control individuals for inter-group comparisons.

When used as a classifier with no additional data, plasma ApoA1 levels demonstrate an area under the curve (AUC) of 0.68 (95% CI 0.65–0.75, p<0.001) for differentiating PD from NC, with a maximum sensitivity of 71% and specificity of 60% (Fig. 1b). Evaluating males and females separately resulted in AUC values of 0.69 for females and 0.62 for males, suggesting this difference is stronger in females, but present in males as well (Suppl. Fig. 3).

Plasma levels of ApoA1 are associated with rs670 genotype

We next genotyped the APOA1 promoter polymorphism rs003670 in Cohort 1 individuals, as well as an additional 158 PD patients in whom we had previously measured and reported ApoA1 plasma levels (Cohort 2, Table 1), using a TaqMan-based genotyping assay.

In Cohort 1 overall, and in PD patient subgroups in both cohorts, GG genotype carriers had significantly lower plasma ApoA1 levels (p=0.012 for Cohort 1 GG vs. other genotypes, p=0.019 for Cohort 2 GG vs. other genotypes, Fig.1c and 1d, Suppl. Fig. 4). This difference strengthened after adjusting for age and sex (rs670 GG genotype vs. other genotypes, Cohort 1 p=0.009, Cohort 2 p=0.024).

Genotypes at rs670 show a trend towards association with PD

Having demonstrated that rs670 genotype is associated with plasma ApoA1 levels in PD, and that ApoA1 levels are lower in PD, we next asked whether rs670 genotype is associated with PD.

We genotyped rs670 in an additional 1494 PD cases and 925 normal controls for whom we had DNA only (Cohort 3), for a total of 1953 PD cases and 1005 normal controls. Of these, the vast majority (1930 PD, 997 controls) were of White, non-Latino ancestry, and these individuals were used for association analyses.

As shown in Table 3, we found a non-significant trend towards enrichment of the GG genotype in PD (odds ratio 1.11, p=0.101), with slightly weaker results if a co-dominant model was substituted for the A-dominant model. Results were not affected in models adjusting for age (p=0.094 for GG genotype association with PD) or sex (p=0.101 for GG genotype association with PD).

Table 3.

Genotypic distribution of rs670 in Parkinson’s disease patients and controls.

	Na	MAF rs670
PD	1930	0.168
Normal Control	997	0.180

A-dominant model
	AA/AG (%)	GG (%)	Chi-square
PD	579 (30.0)	1351 (70.0)	X2 = 1.63
Normal Control	322 (32.3)	675 (67.7)	P = 0.101

Co-dominant model
	AA (%)	AG (%)	GG (%)	C-A test
PD	70 (3.6)	509 (26.4)	1351 (70.0)	X2 = 1.15
Normal Control	36 (3.6)	286 (28.8)	675 (67.7)	P = 0.142

Abbreviations: C-A trend test, Cochran-Armitage trend test; MAF, minor allele frequency

^a2927 white, non-Latino subjects were genotyped at rs670, and genotypic frequencies were compared between PD and neurologically normal controls. P-values are one-tailed, given an expected direction of association.

Recombination is common in the APOA1 gene region surrounding rs670

The APOA1 gene is located on chromosome 11, and we evaluated the APOA1 region for linkage disequilibrium structure and functional elements. The APOA1 promoter region containing rs670 overlaps with a DNase hypersensitivity site as well as multiple transcription factor binding sites, offering potential mechanisms for the observed eQTL effect. Moreover, while a haplotype spanning several apolipoprotein genes in the region has been described,²⁷ recombination is common in the region (Fig. 2; Suppl. Fig. 5).²⁴ As a consequence, analysis of data from the NINDS-sponsored, publicly-available PD GWAS²⁵ demonstrates that rs670 was neither directly genotyped nor well represented by any of the SNPs genotyped, with a maximum r² of 0.53 for the SNP (rs11216162) genotyped in this study (Fig. 2), which is most tightly correlated with rs670. Thus, imputation of rs670 genotype from existing datasets is difficult, and it is possible that a genetic signal associating with PD may have been missed in prior array-based genotyping studies.

Figure 2. Linkage disequilibrium structure and functional elements of the APOA1 gene region.

2kb haplotype block containing rs670 is outlined by the black triangle in (a) and (b). The best SNP proxy for rs670 from the NINDS-PD GWAS²⁵ (rs11216162) is only moderately correlated with rs670 (r²=0.53), and rs670 overlaps a DNase hypersensitivity site.⁴³ Black blocks indicate r²=1, white indicates r²=0, gray indicates intermediate r² values.

Plasma levels of ApoA1 are associated with age at PD onset, even after adjusting for rs670 genotype

As we previously reported an association between ApoA1 levels and age at PD onset, we repeated this analysis in Cohort 1, which consisted of PD subjects recruited subsequent to our prior study. As expected, age at PD onset was positively correlated with plasma ApoA1 levels (p=0.038), after adjusting for age at plasma collection, sex, and the interaction between plasma ApoA1 levels and age. Importantly, adjusting for rs670 genotype, the association between plasma ApoA1 levels and age at PD onset persisted (p=0.040). In a logistic regression model applied to this cohort (Table 2), lower plasma ApoA1 levels were predictive of PD vs. control status, adjusting for rs670 genotype, sex, and age at plasma (p<0.001 for association of PD status with ApoA1 levels in models with or without correction for rs670 genotype). This relationship persisted in models omitting correction for age at plasma sampling and sex (Table 2).

Discussion

In the present study, we evaluated differences in plasma levels of ApoA1 in a cohort of PD patients, normal controls, and patients with other neurodegenerative diseases, finding that plasma levels of ApoA1 are lowest in PD patients. Genotypic analysis of the APOA1 promoter SNP rs670 revealed an association between the GG genotype and lower plasma ApoA1 levels. Furthermore, rs670 GG genotype showed a trend towards association with PD. Finally, ApoA1 levels were predictive of age at PD onset, even after adjusting for rs670 genotype. These data are consistent with our hypothesis that higher ApoA1 levels may be protective against the development of PD.

Previously, we nominated plasma ApoA1 as a potential biomarker for PD risk, based on an association between circulating ApoA1 levels and age at PD onset, observed in a screen of 96 plasma proteins.⁴ Here, we show for the first time that plasma ApoA1 is significantly lower in PD patients than controls, and that this difference persists after adjusting for age and sex. Moreover, the finding that ApoA1 is also lower in PD patients compared to patients with ALS and with FTD (Supplemental Figure 1) suggests that this decrease is not due to general effects of neurodegeneration. To date, few blood-based proteins have been demonstrated to differ in PD patients vs. controls. For example, candidate PD biomarkers nominated from CSF studies for evaluation in the blood include alpha-synuclein and DJ-1. Both proteins are promising as CSF markers,^28,29 but as blood-based markers, data are conflicting, with the largest study to date concluding there were no significant differences in the plasma of PD patients compared to controls.³⁰ In the blood, the only marker that has consistently been shown to differ in levels comparing PD patients with controls is serum urate.³¹ Thus, our results demonstrating that plasma ApoA1 levels are significantly lower in PD compared to both neurologically normal and diseased controls adds an important, and easily assayed, potential biomarker for PD to an otherwise sparse landscape. Indeed, the ROC characteristics for plasma ApoA1 as a PD vs. control classifier (71% sensitivity, 60% specificity), while modest, are comparable to those described (71% sensitivity, 53% specificity) for CSF alpha-synuclein,²⁹ arguably one of the most mature diagnostic biochemical biomarkers at present in PD.

We further demonstrate that circulating ApoA1 levels are significantly correlated with rs670 genotype in patients with Parkinson’s disease. While meta-analyses of combined data from ~3000 individuals in the cardiovascular literature suggest this eQTL effect,³² this is the first demonstration of a SNP-to-plasma level correlation for ApoA1 in patients with neurological disease. Since APOA1 rs670 genotypes have been reported to associate with various outcomes in several neurological diseases (multiple sclerosis¹⁸ and Alzheimer's disease¹⁹), the finding that rs670 genotype affects circulating ApoA1 levels may shed light on mechanism. Indeed, in this regard, it is worth noting that the major source of ApoA1 is the gastro-intestinal tract,³³ and ApoA1 not only crosses the blood-brain barrier, but has actually been proposed as a carrier to help target drugs to the brain.³⁴ Thus, circulating levels of ApoA1 may be proportional to levels of ApoA1 in the brain, although this remains to be formally tested.

We find a non-significant trend towards association of rs670 genotype with PD. The APOA1 locus is complex, with minor allele frequencies (MAFs) for rs670 varying by population. To control for this effect, we limited our genetic association analysis to White, non-Latino subjects, in whom our observed MAF of 0.180 for normal controls is in-line with previously published reports.^13,35 However, it is possible that hidden confounders remain, as we were not able to account for European genetic substructure within our sample. Indeed, allele frequencies for rs670 are known to vary within Europe, with considerable differences in MAF across normal controls from different countries.²⁴ In the context of our biomarker-driven targeted genetic approach, leading to our analysis of a specific SNP (rs670) and PD, the observed trend towards association might be a true signal obscured by cryptic population structure.

Environmental factors may also be obscuring the relationships between rs670 genotype, ApoA1 levels, and PD disease state. Specifically, diet can affect ApoA1 and HDL levels, thus modulating any genetic effects.^13,14 Additionally, smoking³⁶ and exercise³⁷ are two lifestyle factors reported to interact with rs670 genotype and influence plasma HDL/ApoA1. A comprehensive treatment of these environmental factors is beyond the scope of the current study; however, it is intriguing that exercise³⁸ and smoking³⁹ have long been known to correlate with decreased risk of PD.

Understanding the underlying cause of the observed association between plasma ApoA1 and PD is particularly important because plasma ApoA1 levels are modifiable, with relatively-safe, FDA-approved drugs. Thus, if accumulating data suggests that ApoA1 levels are not only correlated with PD, but also have a neuroprotective effect in PD, the path to a human trial is relatively short. In this regard, we note that a potential neuroprotective effect for high ApoA1 expression in animal models of neurological disease has been previously reported. Specifically, in mouse models of Alzheimer's disease (AD), over-expression of ApoA1 had a beneficial effect on cognition,⁴⁰ while ApoA1 depletion worsened cognitive performance.⁴¹ Specifically, overexpression of ApoA1 reduced neuroinflammation by attenuating astrogliosis and pro-inflammatory cytokines,⁴⁰ suggesting a potential mechanism of action. Future studies evaluating the therapeutic benefit of increasing ApoA1 expression in animal models of PD would be a valuable addition to the data presented here.

Our study has two major limitations. First, in an analysis of rs670 genotype and a potential predisposition to PD, genomewide genetic data allowing one to account for population stratification in the analysis would be a valuable addition. In such an analysis, if a bona fide association between rs670 genotype and PD were seen, this would strongly implicate ApoA1 levels in the development of PD. Second, given the potential links between statin medications, lifestyle factors, ApoA1, and PD, a prospective study gathering these data in a longitudinally followed cohort of PD subjects would be helpful in understanding the directionality of potential causal relationships. Because the cohorts in this study were collected prior to our understanding that ApoA1 might be a biomarker of interest, statin and lifestyle data were not systematically ascertained, precluding such an analysis here.

Conclusions

We show that plasma ApoA1 levels are correlated with rs670 genotype in PD patients, with GG genotype associated with lower ApoA1 levels. Compared to controls, PD patients have lower ApoA1, and PD patients may also be enriched in the GG genotype. Finally, even after correcting for rs670 genotype, lower plasma ApoA1 levels are predictive of PD status.

Our results suggest that plasma ApoA1 may be a modifiable biochemical correlate for PD risk. In a disease affecting over 5 million individuals worldwide, with an estimated increase to over 9 million people by 2030,⁴² the impact of any potential risk modification would be substantial.