The Epistasis Project: A Multi-Cohort Study of the Effects of BDNF,DBH,and SORT1 Epistasis on Alzheimer’s Disease Risk

Abstract

Pre-synaptic secretion of brain-derived neurotrophic factor (BDNF) from noradrenergic neurons may protect the Alzheimer’s disease (AD) brain from amyloid pathology. While the BDNF polymorphism (rs6265) is associated with faster cognitive decline and increased hippocampal atrophy, a replicable genetic association of BDNF with AD risk has yet to be demonstrated. This could be due to masking by underlying epistatic interactions between BDNF and other loci that encode proteins involved in moderating BDNF secretion (DBH and Sortilin). We performed a multi-cohort case-control association study of the BDNF, DBH, and SORT1 loci comprising 5,682 controls and 2,454 AD patients from Northern Europe (87% of samples) and Spain (13%). The BDNF locus was associated with increased AD risk (odds ratios; OR = 1.1–1.2, p = 0.005–0.3), an effect size that was consistent in the Northern European (OR = 1.1–1.2, p = 0.002–0.8) but not the smaller Spanish (OR = 0.8–1.6, p = 0.4–1.0) subset. A synergistic interaction between BDNF and sex (synergy factor; SF = 1.3–1.5 p = 0.002–0.02) translated to a greater risk of AD associated with BDNF in women (OR = 1.2–1.3, p = 0.007–0.00008) than men (OR = 0.9–1.0, p = 0.3–0.6). While the DBH polymorphism (rs1611115) was also associated with increased AD risk (OR = 1.1, p = 0.04) the synergistic interaction (SF = 2.2, p = 0.007) between BDNF (rs6265) and DBH (rs1611115) contributed greater AD risk than either gene alone, an effect that was greater in women (SF = 2.4, p = 0.04) than men (SF = 2.0, p = 0.2). These data support a complex genetic interaction at loci encoding proteins implicated in the DBH-BDNF inflammatory pathway that modifies AD risk, particularly in women.

Keywords

Alzheimer’s disease brain-derived neurotrophic factor dopamine beta-hydroxylase epistasis genetics neurotrophins Sortilin

INTRODUCTION

Understanding the genetic factors that modify risk and/or progression of Alzheimer’s disease (AD) is fundamental for the identification and monitoring of at-risk individuals and could lead to novel therapeutic targets and biomarkers. Brain-derived neurotrophic factor (BDNF) is the most abundant neurotrophin in the brain and is a well-characterized protective factor against AD pathology. Specifically, it has been reported that pre-synaptic secretion of BDNF from noradrenergic neurons modulates neural circuitry, synaptic plasticity, neuronal survival and differentiation, and protects neurons from various types of insult (reviewed in [1]). In accordance with these findings, reduced levels of BDNF (mRNA and protein) have been observed in peripheral blood of AD patients [2] and in cortical and subcortical brain regions of individuals with AD at autopsy [3, 4]. In the aged population and in subjects with mild cognitive impairment and AD patients, reduced brain BDNF levels correlate with lower cognitive scores and faster cognitive decline [5 –7]. Moreover, there is some evidence that genetic variation at the BDNF locus may moderate downstream effects of the pathological protein, amyloid-β, in autosomal dominant AD. For example, asymptomatic adults with signs of brain amyloidosis who carry the methionine allele at rs6265 show faster decline in episodic memory, lower hippocampal volume, and increased tau markers in cerebrospinal fluid compared to valine homozygotes [8 –12]. This could be attributed to the association of this allele with altered intracellular trafficking and reduced secretion of BDNF [13 –15]. Additionally, there is evidence that the protective effect of lifetime exposure to cognitively stimulating activities on cognitive performance may be weaker in carriers of the methionine allele [16]. Thus, the methionine allele may be a prognostic biomarker for predicting cognitive decline due to AD. That being said, extensive study in over 30 populations (Caucasian, Asian, African, and mixed) has failed to demonstrate a replicable association of BDNF genetic variants with altered AD risk (http://www.Alzgene.org [17]).

Studies using rodent models have shown that a reduction in BDNF levels may precede the appearance of amyloid plaque pathology [18]. Reduced BDNF levels have been associated with altered synaptic plasticity [19 –21], increased cortical amyloid plaque numbers, and reduced noradrenergic innervation of the hippocampus, frontal cortex, and cerebellum [22] of animal models, effects that can be reversed with BDNF treatment [23 –27].

Taken together, data from animal and human studies strongly support an important neuroprotective effect of BDNF and that an imbalance in BDNF signaling may be an early event in AD pathogenesis. It follows that modulation of BDNF secretion could be critical to this process. In this regard, previous studies using rodent models have reported a feed forward loop between BDNF and the noradrenergic system, whereby noradrenaline (norepinephrine) induces astrocytic and neuronal BDNF secretion, which in turn stimulates noradrenergic signaling as part of an anti-inflammatory mechanism [28 –32]. In support of this, reduced noradrenaline levels have also been reported in aged and AD brains and it has been hypothesized that increasing noradrenergic signaling in the brain could halt the progression of neurodegeneration and cognitive decline (reviewed in [33]). Noradrenaline is synthesized from dopamine, a reaction catalyzed by dopamine beta-hydroxylase (DBH) and is an anti-inflammatory agent that can attenuate the cortical inflammatory response to the Aβ₄₂ peptide [34]. Thus deregulation of DBH-mediated synthesis of noradrenaline could have downstream effects on BDNF secretion. An additional factor in BDNF secretion is the intracellular sorting of BDNF to the secretory pathway, which is regulated by binding of the BDNF pro-domain to the luminal domain of Sortilin [35]. This interaction prevents BDNF degradation, targets BDNF to the secretory pathway [36] and facilitates the release of mature BDNF [37].

The aims of this first study of Stage 2 of the Epistasis Project (the results of Stage 1 [38 –44] are summarized in Table 1), were to further evaluate the contribution of genetic variation at the BDNF locus to AD risk in a large case-control dataset (n = 8,136) and determine whether the association could be masked by genetic epistasis between BDNF variants and sex, age, APOE ɛ4 status, and variants at loci encoding proteins that control BDNF secretion (DBH and Sortilin).

Table 1

Summary of the results obtained from Stage 1 of the Epistasis Project. Synergy Factors and p-values are given for interactions between genes studied in Stage 1. All analyses controlled for age, gender, APOE ɛ4, and collection center. ^*p < 0.05

Study [reference]	Interacting genes	Genotypes	Synergy factors (95% CI, p-value)	Population
			All	North Europe	North Spain
Combarros et al., 2009 [38]	IL6×IL10	rs2069837 AA×rs1800871 C+	1.6 (1.1– 2.4, 0.01)^*	1.7 (1.05– 2.6, 0.03)^*	2.0 (0.9– 4.4, 0.09)
Combarros et al., 2010 [39]	DBH×IL1A	rs1611115 T +×rs1800587 TT	1.9 (1.2– 3.1, 0.005)^*	1.7 (1.02– 2.8, 0.04)^*	3.4 (0.9– 12.3, 0.07)
Lehmann et al., 2012 [40]	HFE×TF	rs1799945 CC×rs1130459 AA	1.5 (1.1– 2.1, 0.02)^*	2.0 (1.3– 3.05, 0.002)^*	1.3 (0.7– 2.55, 0.4)
Kölsch et al., 2012 [41]	PPARA×INS	rs1800206 CC×rs689 TT	1.7 (0.9– 3.2, 0.1)	2.5 (1.2– 5.4, 0.02)^*	0.8 (0.2– 3.0, 0.7)
Heun et al., 2012 [42]	PPARA×IL1A	rs4253766 CC×rs3783550 C+	1.6 (1.15– 2.2, 0.005)^*	1.7 (1.2– 2.5, 0.006)^*	1.5 (0.7– 3.0, 0.3)
Bullock et al., 2012 [43]	GSTM3×HHEX/IDE/KIF11	rs7483 A +×rs1111875 GG	2.3 (1.6– 3.2, <0.00001)^a, *	2.2 (1.5– 3.3, 0.0001)^a, *	2.8 (1.2– 6.4, 0.015)^a, *
Medway et al., 2013 [44]	CYP19A1×IL10	rs1062033 GG×rs1800896 A+	1.9 (1.2– 3.25, 0.01)^b, *	1.9 (1.1– 3.3, 0.02)^b, *	1.4 (0.35– 5.55, 0.65)^b

^ain≥75 years (the 3-way interaction with age±75 years gave p = 0.02); ^bin women (the 3-way interaction with gender in North Europe gave p = 0.02).

MATERIALS AND METHODS

Study population

DNA samples were collected from individuals of Caucasian European descent by seven research centers within two geographical regions namely, Northern Europe (Bristol, Nottingham, OPTIMA and Rotterdam) and Spain (Madrid, Oviedo and Santander). All AD cases were diagnosed “definite” or “probable” by CERAD or NINCDS-ADRDA criteria. AD cases were sporadic, i.e., possible autosomal dominant cases were excluded, based on family history. The cohorts in Stage 1 of the Epistasis Project were described in [38]. In this Stage 2, the Bonn cohort was replaced by that from Madrid, giving a better overall balance between Northern Europe and Spain. The Madrid cohort fulfilled our criteria of an all-Caucasian population drawn from a narrow geographical region. The demographics of each sample collection are shown in Table 2. Research ethics approval was obtained by each of the participating groups. Written informed consent was given for donation of blood samples from all participants, or their legal representatives.

Table 2

Demographics of the sample collections used in this study. The number of samples (n) in each dataset along with mean age-at-sample collection (±standard deviation; SD) is given. The sample numbers (n and % of dataset) are also given for the demographic variables included in the analyses, i.e., age-at-collection (participants aged over 75; n > 75 versus < 75), sex (female versus male) and APOE ɛ4 carriers (ɛ4 + versus ɛ4–)

		Age-at-sample-collection (years)			Female		APOEɛ4+
	n	Mean±SD	range	n 75+	n	%	n	%
Controls	5682	82±8	60–106	4762	3198	56	1303	23
Northern Europe	5239	82±7	60–106	4513	2925	56	1231	23
Bristol	88	81±8	62–96	69	39	44	12	14
Nottingham	148	78±10	60–104	96	65	44	32	22
OPTIMA	208	80±7	62–100	154	89	43	49	24
Rotterdam	4795	83±7	60–106	4194	2732	57	1138	24
Spain	443	77±10	60–104	249	273	62	72	16
Madrid	278	76±11	60–104	135	168	60	42	15
Oviedo	59	74±5	63–87	29	36	61	12	20
Santander	106	81±8	62–99	85	69	65	18	17
AD	2454	81±8	60–109	1945	1568	64	1218	50
Northern Europe	1858	83±7	60–101	1589	1196	64	926	50
Bristol	210	80±9	60–99	160	106	50	137	65
Nottingham	253	83±8	61–101	211	140	55	151	60
OPTIMA	267	79±8	61–100	190	123	46	178	67
Rotterdam	1128	84±6	61–101	1028	827	73	460	41
Spain	596	76±8	60–109	356	372	62	292	49
Madrid	232	75±10	60–109	106	147	63	118	51
Oviedo	138	77±5	63–88	92	86	62	53	38
Santander	226	77±7	60–98	158	139	62	121	54

Polymorphism selection

We downloaded Caucasian European (CEU) population data from the HaploView platform (release 27) using gene coordinates (gene±10 kbp flanking regions) extracted from the UCSC MySQL server (hg18) at https://genome.ucsc.edu/. Polymorphisms with a genotype rate <50%, and that were in Hardy-Weinberg equilibrium (p < 0.001) and with a minor allele frequency (MAF) <5% were removed. The remaining polymorphisms were introduced into the tagger function in HaploView. A tagging threshold of r² > 0.8 and log odds (LOD) score >3.0 was used. The “Aggressive Tagging” approach was implemented, which allows a haplotype of up to three polymorphisms to be used as a proxy. We identified 34 polymorphisms (5 in BDNF, 23 in DBH, and 6 in SORT1) that have a minor allele frequency >5% in Caucasian Europeans and that cover 80% of the variation in these genes Table 3. Due to the considerable variation in DBH, coverage of this gene was restricted to rs1611115, which has previously been associated with increased AD risk in men aged <75 [39] and is responsible for 30–50% of the considerable variation in DBH activity [46 –53]. The 12 polymorphisms were taken forward for genotyping.

Table 3

Polymorphisms included in the study. The polymorphisms identified as representative of a linkage disequilibrium block (TagSNP) are listed for each gene. The size (kbp captured), location (region captured) and coverage of the linkage disequilibrium block captured by the TagSNP, known polymorphisms that lie within the LD block (SNPs captured) and the r² value are listed. The overall coverage of the total gene area is given in the Gene column

Gene	Tag SNP	Region Captured
		Kbp	Region	Coverage	Polymorphisms	r ²
BDNF; Coverage = 89%	rs6265(BDNF1)	0.0	intron,ncRNA, missense	4%	rs6265	1.0
	rs12288512 (BDNF2)	2.8	near-gene-5	7%	rs12288512, rs12273363	0.9–1.0
	rs11030102 (BDNF3)	19.8	intron, near-gene-5	7%	rs11030102, rs10835211	0.9–1.0
	rs11030119 (BDNF4)	67.2	intron, Untranslated-3	30%	rs11030119, rs962369, rs7127507, rs1013402, rs11030108, rs7124442, rs1519480, rs925946	0.9–10
	rs11030104 (BDNF5)	32.3	Intron	15%	rs11030104, rs7103411, rs16917237, rs10501087	1.0
	rs6265, rs12288512, rs11030104	69.8	intron, ncRNA, missense	4%	rs6265, rs11030104, rs12288512, rs1491850	0.9–1.0
	rs12288512, rs11030119, rs11030104	5.8	Intron	7%	rs11030119, rs11030104, rs12288512, rs7934165, rs10767665	0.9–1.0
	rs11030102, rs11030119, rs11030104	79.5	intron, Untranslated-5	15%	rs11030119, rs11030104, rs11030102, rs2203877, rs10835210, rs1519479, rs7931247	0.8–1.0
DBH; Coverage = 3%	rs1611115 (DBH1)	0.0	near-gene-5	3%	rs1611115	1.0
SORT1; Coverage = 86%	rs2228604 (SORT1a)	94.8	intron, near-gene-5, coding-synon, untranslated-3	61%	rs4603158, rs2228604, rs3879450, rs7518013, rs10745352, rs10858085, rs444387, rs3853501, rs3768497, rs11142, rs10745354, rs10858086, rs10858092, rs4970752, rs4970751, rs1880670, rs443345	0.9–1.0
	rs17585355(SORT1b)	0.0	Intron	4%	rs12037569	1.0
	rs7536292(SORT1c)	29.0	Intron	7%	rs7536292, rs12037569	0.8–1.0
	rs17646665(SORT1d)	0.0	Intron	4%	rs17585355	1.0
	rs1149175 (SORT1e)	12.4	Intron	7%	rs1149175, rs11581665	0.9–1.0
	rs10745354, rs7536292 (SORT1f)	75.6	Intron	4%	rs10745354, rs7536292, rs464218	0.9–1.0

Genotyping and imputation

Genotyping for all centers except Rotterdam (below) was performed by LGC (Hoddesdon, Hertfordshire, UK), using the KASPar technology by KBioscience http://www.kbioscience.co.uk. Six to eight duplicate samples and one blank (H₂O) sample were included on each plate. All blank samples were negative. The concordance between duplicates was 100% with the exception of one discordant sample for BDNF rs11030104 in the Bristol dataset, which was removed from analyses. Where data were available from previous in-house TaqMan genotyping, LGC genotypes were 100% concordant (BDNF rs6265; n = 483, DBH rs1611115; n = 396). Genotyping in the Rotterdam cohort was done on Version 3 Illumina-Infinium-II HumanHap550 SNP array (Illumina, San Diego, CA, USA). Genotypes for the Rotterdam cohort were imputed from whole genome data using MACH software (http://www.sph.umich.edu/csg/abecasis/MACH/) and with the 1000 genomes project (phase I version 3 reference panel, positive strand) as a reference. The reliability of imputation was estimated for each imputed SNP with the ratio of expected and observed dosage variance (O/E ratio). Only samples with high-quality extracted DNA were genotyped. In order to allow analysis alongside the genotyped cohorts, the imputed probabilities from Rotterdam were converted to genotypes such that a value of 0 was coded as homozygote for allele 2, a value of 1 was coded as heterozygote and a value of 2 was coded as homozygote for allele 1. It should be noted that a dosage of 1 could be heterozygote, but it could also mean that person has a 0.5 probability of being homozygote for allele 1 and a 0.5 probability of being homozygote for allele 2. That being said, it is commonplace in genetics to round these values to 0/1/2 with the implied corresponding genotypes. Supplementary Table 1 shows the genotype counts and allele frequencies as well as genotyping rate for all individual centers. The genotyping rate for each polymorphism ranged from 0.99 to 1.00 in the total dataset. All genotypes were confirmed to be in Hardy-Weinberg equilibrium (α= 0.001) in controls from all individual centers. APOE ɛ4 genotypes were determined according to the genotypes at rs429358 and rs7412. Tests for pair-wise intragenic linkage disequilibrium for the 5 BDNF polymorphisms and the 6 SORT1 polymorphisms were consistent with the Haploview data (r < 0.8) with the exception of rs6265 and rs11030104 in BDNF (r² = 0.9) and rs2228604 and rs10745354 in SORT1 (r² = 1.0). The high degree of co-inheritance of these pairs was replicated in both the Northern European and Spanish controls. The polymorphisms rs11030104 (BDNF) and rs10745354 (SORT1) were therefore removed from the study as they provided no further information.

Statistical analyses

All analyses were performed in R version 3.5.0 [54]. The relative risk (odds ratios and 95% confidence intervals; CI) for AD was calculated for each of the 10 polymorphisms brought forward for analysis using a general linear model for dominant and recessive inheritance models controlling for age-at-extraction, sex, APOE ɛ4 allele, and individual center. The Akaike Information Criterion (AIC) was used to compare goodness-of-fit, with smaller values of AIC corresponding to preferable models. The statistical power for each sample included in the study to detect a significant (α= 0.05) association of a polymorphism with a minor allele frequency 0.2 (mean of polymorphisms included in the study) with AD risk (OR = 1.2) under a dominant model and assuming a prevalence of AD of 10% in the population aged over 65 is as follows: Total dataset 97%, Northern Europe 92% (Bristol 14%, Nottingham 14%, OPTIMA 17%, Rotterdam 78%), Spain 31% (Madrid 18%, Oviedo 11%, Santander 13%). All tests were first performed in the total dataset and subsequently stratifying by geographical region: Northern Europe = Bristol, Nottingham, OPTIMA and Rotterdam; Spain = Madrid, Oviedo and Santander. When an association (p < 0.05) was observed in the Northern European dataset, analyses were subsequently tested in the Rotterdam subset as the only collection center with sufficient statistical power to detect an association. Interactions were assessed using the Synergy Factor (SF) described in [55]. Using this method, the SF = actual OR (joint effect of two factors)/predicted OR (product of the OR for each factor alone). The null hypothesis states that the actual OR is equal to the predicted OR. This method assesses interactions on a multiplicative scale rather than an additive scale. Throughout the study, significance was considered at the α= 0.05 level. However, due to the large number of tests performed (n = 224 in total), the Bonferroni cut-off (α= 0.0002) can be considered as a more stringent cut-off for significance that minimizes the false-positive associations reported. Associations that surpass this cut-off are indicated in the text and figures.

RESULTS

Genetic variation at the BDNF locus is associated with increased AD risk

We tested for association of the 4 BDNF polymorphisms with altered AD risk in the total population. The model preferred (lowest AIC) for BDNF1 was the recessive inheritance model, whereas the preferred models for BDNF2–4 were the dominant models. The OR and 95% CI for the preferred models of each polymorphism are plotted in Fig. 1. The BDNF1 polymorphism, previously shown to regulate BDNF trafficking, was not associated (at the α= 0.05 level) with altered AD risk in the total (rs6265 MM; OR = 1.2, p = 0.3), Northern European (OR = 1.0, p = 0.8), or Spanish (OR = 1.6, p = 0.1) datasets (Fig. 1A) However, the OR in the total dataset was of a similar yet greater magnitude to those for BDNF2 (Fig. 1B; rs12288512 A allele OR = 1.1, p = 0.04) and BDNF3 (Fig. 1C; (rs113030102G allele OR = 1.1, p = 0.005), associations which surpassed our cut-off for significance. Both BDNF2 (OR = 1.2, p = 0.01) and BDNF3 (OR = 1.2, p = 0.005) were associated with increased AD risk in the Northern European but not Spanish (OR < 1.0, p > 0.3) populations. The associations of BDNF2–4 were even stronger in the Rotterdam subset (BDNF2; OR = 1.2, p = 0.002, BDNF3; OR = 1.3, p = 0.0008). Another polymorphism, BDNF4 (Fig. 1D), that was not significant in the total dataset (p = 0.08) despite a similar OR (rs11030119 A+; OR = 1.1), was associated with increased AD risk in the Northern European (OR = 1.2, p = 0.01) and Rotterdam (OR = 1.2, p = 0.002) but not Spanish (OR = 0.8, p = 0.2) population. It should be noted that none of these associations surpassed the Bonferroni cut-off (p < 0.0002) for this study (best p = 0.0008 for BDNF3 in the Rotterdam dataset).

Fig.1

BDNF main effects on AD risk. The odds ratios (OR) and 95% confidence intervals (CI) are plotted for the main effects of the BDNF polymorphisms on AD risk in A–D. The x-axis is plotted on log–10 scale. p-values <0.05 are labelled next to the corresponding OR.

To determine whether interactions between BDNF polymorphisms could explain the contradictory findings for rs6265 (BDNF1) across multiple populations, we sequentially added pair-wise interactions between the 4 BDNF polymorphisms into the model. No interaction between BDNF polymorphisms was associated with altered AD risk (p > 0.5).

Women showed increased susceptibility to AD risk-modifying polymorphisms at the BDNF locus

We next tested whether interactions between BDNF polymorphisms and age-at-sampling, sex and possession of the APOE ɛ4 allele could contribute to AD risk. We found no association of the BDNF1*sex interaction with AD risk (SF = 0.95, p = 0.9). However, inclusion of BDNF2*sex, BDNF3*sex or BDNF4*sex interactions improved the AIC for the model and the synergy factors (SF) for these interactions are plotted in Fig. 2A–C. We observed a positive synergistic interaction between BDNF2, BDNF3, and BDNF4 and sex in the total dataset; sex*BDNF2 (SF = 1.5, p = 0.002), sex*BDNF3 (SF = 1.4, p = 0.002), and sex*BDNF4 (SF = 1.3, p = 0.02). The interactions were observed in the Northern European dataset (SF = 1.4, p = 0.02; SF = 1.4, p = 0.007; SF = 1.3, p = 0.02, respectively) and Rotterdam (SF = 1.4, p = 0.02, SF = 1.5; p = 0.01, SF = 1.4, p = 0.04, respectively) datasets and the sex*BDNF2 interaction was also observed in the Spanish population (SF = 1.94, p = 0.03). To explore these interactions further, BDNF2–4 were tested for association with AD in women and men separately in the datasets that showed a synergistic interaction (Fig. 2D–F). While BDNF2 was associated with reduced AD risk in Spanish men (OR = 0.6, p = 0.03), all three polymorphisms were associated with increased risk in women from the total (BDNF2; OR = 1.3, p = 0.0006, BDNF3; OR = 1.3, p = 0.00008, BDNF4; OR = 1.2, p = 0.007, respectively) and Northern European (OR = 1.3, p = 0.0007, OR = 1.4, p = 0.00005, OR = 1.2, p = 0.001, respectively) and Rotterdam (OR = 1.4, p = 0.0001, OR = 1.4, p = 0.00003, OR = 1.4, p = 0.0002, respectively) datasets, associations that surpassed the Bonferroni cut-off for this study (p < 0.0002) in at least one of the datasets.

We next tested for association of interactions between BDNF variants and the APOE ɛ4 allele. The BDNF3*ɛ4 interaction (SF = 0.47, p = 0.02) was associated with altered AD risk in the Spanish dataset and BDNF3 was nominally associated with decreased AD risk in Spanish ɛ4 carriers (OR = 0.58, p = 0.05) but not in non-carriers (OR = 1.2, p = 0.2). That being said, these interactions may be of less importance since including them in the models did not improve the AIC. Interactions between other BDNF variants and ɛ4 were not associated with AD risk (p > 0.06) and neither were interactions between any BDNF variant and age (p > 0.2).

Fig.2

BDNF interactions with sex. The synergy factors (SF) and 95% confidence intervals (CI) for the interactions between BDNF polymorphisms and sex are plotted (A–C). The odds ratios (OR) and 95% CI for the main effects of BDNF2, BDNF3 and BDNF4 on AD risk in men and women are plotted (D–F). The x-axis is plotted on a log–10 scale. p-values <0.05 are labelled next to the corresponding SF/OR.

BDNF and DBH polymorphisms act synergistically to increase AD risk in Northern European women

We have previously identified an association of the DBH polymorphism, rs1611115, with increased AD risk in men aged <75 [39] in a sample set that overlaps with that included in this study. Therefore, unsurprisingly, when testing for an association of DBH with AD risk in the total population, we found that the DBH polymorphism was associated with increased risk of AD in the total dataset (rs1611115T + OR = 1.12, p = 0.04), albeit that the association was not apparent in either of the regional subsets (OR = 1.1, p = 0.1 and OR = 1.3, p = 0.1). Addition of interactions between the DBH polymorphism and age-at-sampling, sex and possession of the APOE ɛ4 allele did not improve the AIC for the model. Since DBH and BDNF have both been implicated in regulating noradrenergic neurotransmission, we sought to determine whether polymorphisms at the BDNF and DBH loci could interact to alter AD risk. The addition of interactions between DBH and all 4 BDNF polymorphisms improved the AIC for the model. The SF for the interactions are plotted in Fig. 3A–D. A synergistic BDNF1*DBH interaction contributed to AD risk in the total (SF = 2.2, p = 0.007) and Northern European (SF = 2.2, p = 0.02) and Rotterdam (SF = 2.2, p = 0.03) datasets. While the SF (SF = 1.4) was comparable in the Spanish population (p = 0.3), the interaction did not reach our cut-off for significance. Since the BDNF polymorphisms have different effects in men and women, we next tested the BDNF*DBH interactions in male and female subgroups separately. Due to testing rs6265 under a recessive inheritance model, the low number of male MM (rs6265), T + (rs1611115; dominant model) carriers in the Spanish dataset (n = 7), meant that the BDNF*DBH1 interaction could only be tested in the Northern European dataset. In that dataset, the BDNF*DBH1 interaction contributed more to AD risk in women (SF = 2.4, p = 0.04) than men (SF = 2.0, p = 0.2). The remaining BDNF*DBH interactions did not contribute to the AD risk (p > 0.1) and it should be noted that none of the BDNF1*DBH interactions surpassed the Bonferroni cut-off (p < 0.0002) for this study.

Fig.3

BDNF interactions with DBH. The synergy factors (SF) and 95% confidence intervals (CI) for the interactions between BDNF1 (A), BDNF2 (B), BDNF3 (C), and BDNF4 (D) with DBH1 are plotted. The x-axis is plotted on a log–10 scale. p-values <0.05 are labelled next to the corresponding SF.

SORT1 genetic variation is not associated with AD risk

Finally, we tested for an association of SORT1 polymorphisms with AD risk. We found no association in the total, Northern European, or Spanish populations (p > 0.05). We tested for an association of interactions between the SORT1 polymorphisms and age-at-sampling, sex and possession of the APOE ɛ4 allele with AD risk. We observed an antagonistic interaction between SORT1e (rs1149175 GG) and sex in the Spanish dataset (SORT1e OR = 2.2, sex OR = 2.1, SF = 0.4, p = 0.007) and SORT1e was associated with increased AD risk in Spanish men (OR = 2.27, p = 0.002). That being said, these interactions may be of less importance since including them in the models did not improve the AIC. None of the interactions with the BDNF variants contributed to AD risk in the total or regional subsets (p > 0.05).

DISCUSSION

This multi-cohort study of 5,682 controls and 2,454 AD patients (recruited from seven research centers from Northern Europe and Spain) represents a comprehensive exploration of the contribution of epistatic interactions between genes encoding proteins implicated in BDNF secretion and noradrenergic innervation to AD risk. The main finding of this study is the synergistic interaction between BDNF risk-modifying variants (BDNF2–4) and sex (SF = 1.3 to 1.5) such that the BDNF-associated risk was observed in women (OR = 1.2 to 1.3) but not men (OR = 0.9 to 1.0). We also show evidence of a synergistic interaction between the BDNF and DBH loci such that carriers of the minor allele at DBH (rs1611115) who are also homozygous for the minor allele at BDNF1 (rs6265) are at increased risk of AD (SF = 2.2) than carriers of either allele alone (BDNF OR = 1.1 to 1.2, DBH OR = 1.1), an effect that was greater in women (SF = 2.4) than in men (SF = 2.0).

In support of the meta-analyses performed by Alzgene, we report only a nominal association of rs6265 (BDNF1) with AD risk (OR = 1.2, p = 0.3). On the other hand, we report three novel risk-modifying BDNF polymorphisms that were associated with increased AD risk in the total or Northern European populations (OR = 1.1, p < 0.01). It should be noted that the OR for rs6265 was similar to that reported for the associations reported for the other 3 polymorphisms. We therefore cannot rule out the possibility that rs6265 has a similar frequency to the other BDNF variants in AD patients, but that the significance of this increased frequency is attenuated by other covariates in the model, namely age-at-collection, sex, APOE ɛ4 or collection center. Nevertheless, since none of these associations were significant when accounting for multiple testing (lowest p = 0.0008, Bonferroni cut-off = 0.0002), these findings warrant replication in large independent cohorts.

The strongest finding of this study is the interaction between BDNF variants 2–4 (rs12288512, rs113030102, rs11030119) and sex (SF = 1.3 to 1.5) such that women carriers from Northern Europe were at greater risk of AD (OR = 1.3 to 1.4) than men (OR = 0.9 to 1.0), associations that were significant even after accounting for multiple testing in at least one dataset. The evidence for the BDNF2*sex interaction reported here is particularly strong, as it was observed in both the Northern European (increased risk in women) and Spanish (decreased in men) subsets as well as the total dataset. This is not the first study to identify sex differences related to the BDNF locus in AD, albeit that a previous study reported an association of BDNF1 (rs6265) with increased AD risk in women from the Han Chinese population [56], a variant that did not interact with sex in this study (p = 0.9). Moreover, the gender effect may not be limited to AD; a greater neuroprotective effect against methamphetamine-induced toxicity (i.e., greater preservation of corpus striatal dopamine levels) has been reported in male compared to female mice overexpressing Bdnf [57]. In another study, modulation of Bdnf expression in response to acute stress was impaired in female rats [58]. The mechanism underlying the gender difference observed in this and other studies has yet to be fully explored. However, a direct interaction between estrogen receptors and BDNF transcription has been shown to alter hippocampal physiology during development in the rat, suggesting that hormonal variation during a critical time window may contribute to sex-specific effects in BDNF transcription [59]. Whether the BDNF variants reported here also affect BDNF transcription has yet to be explored. If a functional effect is demonstrated, a hypothesis arises whereby BDNF expression may be natively lower in adult women, which, when combined with carrying a genetic variant that may also affect brain BDNF levels, could have a multiplicative effect on susceptibility to AD pathology, thereby increasing risk for AD in female carriers compared to male non-carriers. This is an intriguing avenue worth pursuing in future functional studies.

It is possible that BDNF variants could differentially influence a range of brain regions and networks. Therefore, functional studies to determine how these variants may influence noradrenergic innervation of the hippocampus, frontal cortex, and cerebellum could also be informative. Moreover, whether functional effects can be directly attributed to the polymorphisms included in the study, or rather to other polymorphisms within the linkage disequilibrium blocks that they capture, also warrants further investigation.

We also report a synergistic interaction (SF = 2.2) between BDNF1 (rs6265) and DBH1 (rs1611115) that contributed more to AD risk in Northern European women (SF = 2.4) than men (SF = 2), an association that was not apparent for the other BDNF variants. This result is particularly intriguing since the rs6265 variant alone showed the weakest association with AD risk. As the interaction did not survive adjustment for multiple testing (p = 0.007, Bonferroni cut-off = 0.0002), validation in an independent study is necessary. If validated, this interaction could indicate a combinatorial functional effect between BDNF and DBH variants that is specific to the rs6265 polymorphism, thus highlighting the importance of considering epistatic interactions in genetic studies to unmask otherwise hidden associations with AD. We have previously reported a synergistic interaction between DBH (rs1611115) and pro-inflammatory cytokines [39, 45] that contributed to increased risk for AD. As BDNF and DBH are involved in a feed-forward anti-inflammatory mechanism, these studies could point toward a complex interaction between genes that encode proteins involved in neuroinflammatory pathways and furthermore that these interactions may differ according to sex.

A limitation of this study is the use of AD diagnosis as a single end-point, which does not account for the possibility of disease-modifying effects of these variants on brain atrophy or cognitive decline, particularly if these effects are only evident at early stages. The ORs reported in this study, although modest (1.1 to 1.6), suggest a mild effect on AD risk, as would be expected from a complex genetic disorder such as AD.

Overall, these data suggest a complex interplay between BDNF and sex that warrants further investigation in large, independent studies.

Footnotes

ACKNOWLEDGMENTS

The authors are indebted to all the donors and their families for their participation in this study. We would also like to thank the Alzheimer’s Research UK and South West Dementia Brain Bank (SWDBB) for providing brain tissue for DNA extraction included in this study. Stage 2 of the Epistasis Project was funded by the Jamuna Trust and the Herbertpur Trust. The SWDBB is part of the Brains for Dementia Research programme, jointly funded by Alzheimer’s Research UK and Alzheimer’s Society and is supported by BRACE (Bristol Research into Alzheimer’s and Care of the Elderly) and the Medical Research Council. This research benefitted from funding awarded to the NIHR Great Ormond Street Hospital Biomedical Research Centre. OB, P S-J and ER were supported by grants from, IDIVAL, Instituto de Salud Carlos III (Fondo de Investigación Sanitario) and European Funds for Regional Development (FEDER); PI15/00058, CP13/00091, PI08/0139, PI12/02288, PI16/01652, PI13/01008), JPND (DEMTEST PI11/03028 and the CIBERNED program. The Rotterdam Study is funded by Erasmus Medical Center and Erasmus University, Rotterdam, Netherlands Organization for the Health Research and Development (ZonMw), the Research Institute for Diseases in the Elderly (RIDE1 and 2), the Ministry of Education, Culture and Science, the Ministry for Health, Welfare and Sports, the European Commission (DG XII), and the Municipality genotype data for the Rotterdam Study is supported by the Netherlands Organisation of Scientific Research NWO Investments (nr. 175.010.2005.011, 911-03-012).

Authors’ disclosures available online ().

The supplementary material is available in the electronic version of this article: .

References

Lima Giacobbo

, Doorduin

, Klein

, Dierckx

, Bromberg

, de Vries

EFJ

(2018) Brain-derived neurotrophic factor in brain disorders: Focus on neuroinflammation. Mol Neurobiol. doi: 10.1007/s12035-018-1283-6

Qin

, Cao

, Cawley

, Liu

, Yuan

, Loh

, Cheng

(2017) Decreased peripheral brain-derived neurotrophic factor levels in Alzheimer’s disease: A meta-analysis study (N=7277). Mol Psychiatry 22, 312–320.

Ferrer

, Marin

, Rey

, Ribalta

, Goutan

, Blanco

, Tolosa

, Marti

(1999) BDNF and full-length and truncated TrkB expression in Alzheimer disease. Implications in therapeutic strategies. J Neuropathol Exp Neurol 58, 729–739.

, Wu

, Qin

, Cao

, Liu

, Cao

, Cheng

(2018) Postmortem brain, cerebrospinal fluid, and blood neurotrophic factor levels in Alzheimer’s disease: A systematic review and meta-analysis. J Mol Neurosci 65, 289–300.

Peng

, Wuu

, Mufson

, Fahnestock

(2005) Precursor form of brain-derived neurotrophic factor and mature brain-derived neurotrophic factor are decreased in the pre-clinical stages of Alzheimer’s disease. J Neurochem 93, 1412–1421.

Michalski

, Corrada

, Kawas

, Fahnestock

(2015) Brain-derived neurotrophic factor and TrkB expression in the “oldest-old,” the 90+Study: Correlation with cognitive status and levels of soluble amyloid-beta. Neurobiol Aging 36, 3130–3139.

Buchman

, Yu

, Boyle

, Schneider

, De Jager

, Bennett

(2016) Higher brain BDNF gene expression is associated with slower cognitive decline in older adults. Neurology 86, 735–741.

Boots

, Schultz

, Clark

, Racine

, Darst

, Koscik

, Carlsson

, Gallagher

, Hogan

, Bendlin

, Asthana

, Sager

, Hermann

, Christian

, Dubal

, Engelman

, Johnson

, Okonkwo

(2017) BDNF Val66Met predicts cognitive decline in the Wisconsin Registry for Alzheimer’s Prevention. Neurology 88, 2098–2106.

Lim

, Villemagne

, Laws

, Ames

, Pietrzak

, Ellis

, Harrington

, Bourgeat

, Salvado

, Darby

, Snyder

, Bush

, Martins

, Masters

, Rowe

, Nathan

, Maruff

(2013) BDNF Val66Met, Abeta amyloid, and cognitive decline in preclinical Alzheimer’s disease. Neurobiol Aging 34, 2457–2464.

10.

Lim

, Villemagne

, Laws

, Ames

, Pietrzak

, Ellis

, Harrington

, Bourgeat

, Bush

, Martins

, Masters

, Rowe

, Maruff

(2014) Effect of BDNF Val66Met on memory decline and hippocampal atrophy in prodromal Alzheimer’s disease: A preliminary study. PLoS One 9, e86498.

11.

Lim

, Hassenstab

, Cruchaga

, Goate

, Fagan

, Benzinger

, Maruff

, Snyder

, Masters

, Allegri

, Chhatwal

, Farlow

, Graff-Radford

, Laske

, Levin

, McDade

, Ringman

, Rossor

, Salloway

, Schofield

, Holtzman

, Morris

, Bateman

(2016) BDNF Val66Met moderates memory impairment, hippocampal function and tau in preclinical autosomal dominant Alzheimer’s disease. Brain 139, 2766–2777.

12.

Lim

, Hassenstab

, Goate

, Fagan

, Benzinger

TLS

, Cruchaga

, McDade

, Chhatwal

, Levin

, Farlow

, Graff-Radford

, Laske

, Masters

, Salloway

, Schofield

, Morris

, Maruff

, Bateman

(2018) Effect of BDNFVal66Met on disease markers in dominantly inherited Alzheimer’s disease. Ann Neurol 84, 424–435.

13.

Egan

, Kojima

, Callicott

, Goldberg

, Kolachana

, Bertolino

, Zaitsev

, Gold

, Goldman

, Dean

, Lu

, Weinberger

(2003) The BDNF val66met polymorphism affects activity-dependent secretion of BDNF and human memory and hippocampal function. Cell 112, 257–269.

14.

Chen

, Patel

, Sant

, Meng

, Teng

, Hempstead

, Lee

(2004) Variant brain-derived neurotrophic factor (BDNF) (Met66) alters the intracellular trafficking and activity-dependent secretion of wild-type BDNF in neurosecretory cells and cortical neurons. J Neurosci 24, 4401–4411.

15.

Chiaruttini

, Vicario

, Li

, Baj

, Braiuca

, Wu

, Lee

, Gardossi

, Baraban

, Tongiorgi

(2009) Dendritic trafficking of BDNF mRNA is mediated by translin and blocked by the G196A (Val66Met) mutation. Proc Natl Acad Sci U S A 106, 16481–16486.

16.

Ward

, Andel

, Saunders

, Thow

, Klekociuk

, Bindoff

, Vickers

(2017) The BDNF Val66Met polymorphism moderates the effect of cognitive reserve on 36-month cognitive change in healthy older adults. Alzheimers Dement (N Y) 3, 323–331.

17.

Bertram

, McQueen

, Mullin

, Blacker

, Tanzi

(2007) Systematic meta-analyses of Alzheimer disease genetic association studies: The AlzGene database. Nat Genet 39, 17–23.

18.

Iulita

, Bistue Millon

, Pentz

, Aguilar

, Do Carmo

, Allard

, Michalski

, Wilson

, Ducatenzeiler

, Bruno

, Fahnestock

, Cuello

(2017) Differential deregulation of NGF and BDNF neurotrophins in a transgenic rat model of Alzheimer’s disease. Neurobiol Dis 108, 307–323.

19.

Linnarsson

, Bjorklund

, Ernfors

(1997) Learning deficit in BDNF mutant mice. Eur J Neurosci 9, 2581–2587.

20.

, Wang

, Wu

, Wei

, Lee

(1998) Brain-derived neurotrophic factor antisense oligonucleotide impairs memory retention and inhibits long-term potentiation in rats. Neuroscience 82, 957–967.

21.

, Li

, Yao

, Zhou

(1999) Deprivation of endogenous brain-derived neurotrophic factor results in impairment of spatial learning and memory in adult rats. Brain Res 835, 259–265.

22.

Braun

, Kalinin

, Feinstein

(2017) Conditional depletion of hippocampal brain-derived neurotrophic factor exacerbates neuropathology in a mouse model of Alzheimer’s disease. ASN Neuro 9, 1759091417696161.

23.

Nagahara

, Merrill

, Coppola

, Tsukada

, Schroeder

, Shaked

, Wang

, Blesch

, Kim

, Conner

, Rockenstein

, Chao

, Koo

, Geschwind

, Masliah

, Chiba

, Tuszynski

(2009) Neuroprotective effects of brain-derived neurotrophic factor in rodent and primate models of Alzheimer’s disease. Nat Med 15, 331–337.

24.

Blurton-Jones

, Kitazawa

, Martinez-Coria

, Castello

, Muller

, Loring

, Yamasaki

, Poon

, Green

, LaFerla

(2009) Neural stem cells improve cognition via BDNF in a transgenic model of Alzheimer disease. Proc Natl Acad Sci U S A 106, 13594–13599.

25.

Kimura

, Takahashi

, Tashiro

, Terao

(2006) Amyloid beta up-regulates brain-derived neurotrophic factor production from astrocytes: Rescue from amyloid beta-related neuritic degeneration. J Neurosci Res 84, 782–789.

26.

Nagahara

, Mateling

, Kovacs

, Wang

, Eggert

, Rockenstein

, Koo

, Masliah

, Tuszynski

(2013) Early BDNF treatment ameliorates cell loss in the entorhinal cortex of APP transgenic mice. J Neurosci 33, 15596–15602.

27.

Zeng

, Zhao

, Xie

(2010) Neurotrophins enhance CaMKII activity and rescue amyloid-beta-induced deficits in hippocampal synaptic plasticity. J Alzheimers Dis 21, 823–831.

28.

Zafra

, Lindholm

, Castren

, Hartikka

, Thoenen

(1992) Regulation of brain-derived neurotrophic factor and nerve growth factor mRNA in primary cultures of hippocampal neurons and astrocytes. J Neurosci 12, 4793–4799.

29.

Juric

, Miklic

, Carman-Krzan

(2006) Monoaminergic neuronal activity up-regulates BDNF synthesis in cultured neonatal rat astrocytes. Brain Res 1108, 54–62.

30.

Ivy

, Rodriguez

, Garcia

, Chen

, Russo-Neustadt

(2003) Noradrenergic and serotonergic blockade inhibits BDNF mRNA activation following exercise and antidepressant. Pharmacol Biochem Behav 75, 81–88.

31.

Cirelli

, Tononi

(2000) Differential expression of plasticity-related genes in waking and sleep and their regulation by the noradrenergic system. J Neurosci 20, 9187–9194.

32.

Chen

, Nguyen

, Pike

, Russo-Neustadt

(2007) Norepinephrine induces BDNF and activates the PI-3K and MAPK cascades in embryonic hippocampal neurons. Cell Signal 19, 114–128.

33.

Leanza

, Gulino

, Zorec

(2018) Noradrenergic hypothesis linking neurodegeneration-based cognitive decline and astroglia. Front Mol Neurosci 11, 254.

34.

Feinstein

, Heneka

, Gavrilyuk

, Dello Russo

, Weinberg

, Galea

(2002) Noradrenergic regulation of inflammatory gene expression in brain. Neurochem Int 41, 357–365.

35.

Chen

, Ieraci

, Teng

, Dall

, Meng

, Herrera

, Nykjaer

, Hempstead

, Lee

(2005) Sortilin controls intracellular sorting of brain-derived neurotrophic factor to the regulated secretory pathway. J Neurosci 25, 6156–6166.

36.

Evans

, Irmady

, Ostrow

, Kim

, Nykjaer

, Saftig

, Blobel

, Hempstead

(2011) Neuronal brain-derived neurotrophic factor is synthesized in excess, with levels regulated by sortilin-mediated trafficking and lysosomal degradation. J Biol Chem 286, 29556–29567.

37.

Yang

, Lim

, Li

, Zhong

, Zhou

(2011) Precursor of brain-derived neurotrophic factor (proBDNF) forms a complex with Huntingtin-associated protein-1 (HAP1) and sortilin that modulates proBDNF trafficking, degradation, and processing. J Biol Chem 286, 16272–16284.

38.

Combarros

, van Duijn

, Hammond

, Belbin

, Arias-Vasquez

, Cortina-Borja

, Lehmann

, Aulchenko

, Schuur

, Kolsch

, Heun

, Wilcock

, Brown

, Kehoe

, Harrison

, Coto

, Alvarez

, Deloukas

, Mateo

, Gwilliam

, Morgan

, Warden

, Smith

, Lehmann

(2009) Replication by the Epistasis Project of the interaction between the genes for IL-6 and IL-10 in the risk of Alzheimer’s disease. J Neuroinflammation 6, 22.

39.

Combarros

, Warden

, Hammond

, Cortina-Borja

, Belbin

, Lehmann

, Wilcock

, Brown

, Kehoe

, Barber

, Coto

, Alvarez

, Deloukas

, Gwilliam

, Heun

, Kolsch

, Mateo

, Oulhaj

, Arias-Vasquez

, Schuur

, Aulchenko

, Ikram

, Breteler

, van Duijn

, Morgan

, Smith

, Lehmann

(2010) The dopamine beta-hydroxylase – 1021C/T polymorphism is associated with the risk of Alzheimer’s disease in the Epistasis Project. BMC Med Genet 11, 162.

40.

Lehmann

, Schuur

, Warden

, Hammond

, Belbin

, Kolsch

, Lehmann

, Wilcock

, Brown

, Kehoe

, Morris

, Barker

, Coto

, Alvarez

, Deloukas

, Mateo

, Gwilliam

, Combarros

, Arias-Vasquez

, Aulchenko

, Ikram

, Breteler

, van Duijn

, Oulhaj

, Heun

, Cortina-Borja

, Morgan

, Robson

, Smith

(2012) Transferrin and HFE genes interact in Alzheimer’s disease risk: The Epistasis Project. Neurobiol Aging 33, 202e201–213.

41.

Kölsch

, Lehmann

, Ibrahim-Verbaas

, Combarros

, van Duijn

, Hammond

, Belbin

, Cortina-Borja

, Lehmann

, Aulchenko

, Schuur

, Breteler

, Wilcock

, Brown

, Kehoe

, Barber

, Coto

, Alvarez

, Deloukas

, Mateo

, Maier

, Morgan

, Warden

, Smith

, Heun

(2012) Interaction of insulin and PPAR-alpha genes in Alzheimer’s disease: The Epistasis Project. J Neural Transm (Vienna) 119, 473–479.

42.

Heun

, Kolsch

, Ibrahim-Verbaas

, Combarros

, Aulchenko

, Breteler

, Schuur

, van Duijn

, Hammond

, Belbin

, Cortina-Borja

, Wilcock

, Brown

, Barber

, Kehoe

, Coto

, Alvarez

, Lehmann

, Deloukas

, Mateo

, Morgan

, Warden

, Smith

, Lehmann

(2012) Interactions between PPAR-alpha and inflammation-related cytokine genes on the development of Alzheimer’s disease, observed by the Epistasis Project. Int J Mol Epidemiol Genet 3, 39–47.

43.

Bullock

, Medway

, Cortina-Borja

, Turton

, Prince

, Ibrahim-Verbaas

, Schuur

, Breteler

, van Duijn

, Kehoe

, Barber

, Coto

, Alvarez

, Deloukas

, Hammond

, Combarros

, Mateo

, Warden

, Lehmann

, Belbin

, Brown

, Wilcock

, Heun

, Kolsch

, Smith

, Lehmann

, Morgan

(2013) Discovery by the Epistasis Project of an epistatic interaction between the GSTM3 gene and the HHEX/IDE/KIF11 locus in the risk of Alzheimer’s disease. Neurobiol Aging 34, 1309e1301–1307.

44.

Medway

, Combarros

, Cortina-Borja

, Butler

, Ibrahim-Verbaas

, de Bruijn

, Koudstaal

, van Duijn

, Ikram

, Mateo

, Sanchez-Juan

, Lehmann

, Heun

, Kolsch

, Deloukas

, Hammond

, Coto

, Alvarez

, Kehoe

, Barber

, Wilcock

, Brown

, Belbin

, Warden

, Smith

, Morgan

, Lehmann

(2014) The sex-specific associations of the aromatase gene with Alzheimer’s disease and its interaction with IL10 in the Epistasis Project. Eur J Hum Genet 22, 216–220.

45.

Mateo

, Infante

, Rodriguez

, Berciano

, Combarros

, Llorca

(2006) Interaction between dopamine beta-hydroxylase and interleukin genes increases Alzheimer’s disease risk. J Neurol Neurosurg Psychiatry 77, 278–279.

46.

Zabetian

, Anderson

, Buxbaum

, Elston

, Ichinose

, Nagatsu

, Kim

, Malison

, Gelernter

, Cubells

(2001) A quantitative-trait analysis of human plasma-dopamine beta-hydroxylase activity: Evidence for a major functional polymorphism at the DBH locus. Am J Hum Genet 68, 515–522.

47.

Zabetian

, Buxbaum

, Elston

, Kohnke

, Anderson

, Gelernter

, Cubells

(2003) The structure of linkage disequilibrium at the DBH locus strongly influences the magnitude of association between diallelic markers and plasma dopamine beta-hydroxylase activity. Am J Hum Genet 72, 1389–1400.

48.

Tang

, Epstein

, Anderson

, Zabetian

, Cubells

(2007) Genotypic and haplotypic associations of the DBH gene with plasma dopamine beta-hydroxylase activity in African Americans. Eur J Hum Genet 15, 878–883.

49.

Tang

, Anderson

, Zabetian

, Kohnke

, Cubells

(2005) Haplotype-controlled analysis of the association of a non-synonymous single nucleotide polymorphism at DBH (+1603C–>T) with plasma dopamine beta-hydroxylase activity. Am J Med Genet B Neuropsychiatr Genet 139B, 88–90.

50.

Kohnke

, Zabetian

, Anderson

, Kolb

, Gaertner

, Buchkremer

, Vonthein

, Schick

, Lutz

, Kohnke

, Cubells

(2002) A genotype-controlled analysis of plasma dopamine beta-hydroxylase in healthy and alcoholic subjects: Evidence for alcohol-related differences in noradrenergic function. Biol Psychiatry 52, 1151–1158.

51.

Chen

, Wen

, Rao

, Zhang

, Wang

, Rodriguez-Flores

, Sanchez

, Mahata

, Taupenot

, Sun

, Mahata

, Tayo

, Schork

, Ziegler

, Hamilton

, O’Connor

(2010) Human dopamine beta-hydroxylase (DBH) regulatory polymorphism that influences enzymatic activity, autonomic function, and blood pressure. J Hypertens 28, 76–86.

52.

Bhaduri

, Mukhopadhyay

(2008) Correlation of plasma dopamine beta-hydroxylase activity with polymorphisms in DBH gene: A study on Eastern Indian population. Cell Mol Neurobiol 28, 343–350.

53.

Desai

, Nebes

, DeKosky

, Kamboh

(2005) Investigation of the effect of brain-derived neurotrophic factor (BDNF) polymorphisms on the risk of late-onset Alzheimer’s disease (AD) and quantitative measures of AD progression. Neurosci Lett 379, 229–234.

54.

R-Core-Team (2018) R Foundation for Statistical Computing, Vienna, Austria.

55.

Cortina-Borja

, Smith

, Combarros

, Lehmann

(2009) The synergy factor: A statistic to measure interactions in complex diseases. BMC Res Notes 2, 105.

56.

, Bi

, Zhang

, Xu

, Luo

, Wang

, Fang

, Li

, Zhang

, Yao

(2017) Female-specific effect of the BDNF gene on Alzheimer’s disease. Neurobiol Aging 53, 192e111–192e119.

57.

Dluzen

(2004) The effect of gender and the neurotrophin, BDNF, upon methamphetamine-induced neurotoxicity of the nigrostriatal dopaminergic system in mice. Neurosci Lett 359, 135–138.

58.

Luoni

, Berry

, Raggi

, Bellisario

, Cirulli

, Riva

(2016) Sex-specific effects of prenatal stress on Bdnf expression in response to an acute challenge in rats: A role for Gadd45beta. Mol Neurobiol 53, 7037–7047.

59.

Solum

, Handa

(2002) Estrogen regulates the development of brain-derived neurotrophic factor mRNA and protein in the rat hippocampus. J Neurosci 22, 2650–2659.