Neuroglobin and Alzheimer’s dementia: Genetic association and gene expression changes

1. Introduction

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder characterized by deterioration of memory, cognition, behavior, emotion, and intellect. With the exception of rare early-onset forms with Mendelian inheritance it most commonly affects people over the age of 65. It is a major public health problem, affecting over 5 million Americans today, a number that could range from 11 to 16 million by the year 2050 (Alzheimer’s Association data 2008; www.alz.org). Early-onset forms of AD (about 5% of cases) have an exclusively genetic etiology, with an autosomal dominant mode of inheritance and three identified causative genes PSEN1, PSEN2 and APP (Goate, et al., 1991, Levy-Lahad, et al., 1995, Sherrington, et al., 1995). In late-onset AD variation in the apolipoprotein E (APOE) gene has been shown to be a genetic risk factor (Strittmatter, et al., 1993) but it is believed that many more remain to be identified (Daw, et al., 2000, Jarvik, et al., 1996). A number of pathogenic mechanisms have been implicated in the neurodegenerative process of AD, among the most studied being apoptosis, (Bamberger and Landreth, 2002, Shimohama, 2000, Takuma, et al., 2005) oxidative stress, (Cecchi, et al., 2002, Gibson and Huang, 2005, Nunomura, et al., 2006, Reynolds, et al., 2007, Shi and Gibson, 2007, Zhu, et al., 2007) hypoxia, (L. Li, et al., 2007, Peers, et al., 2007) and inflammation (McGeer, et al., 2006, Weisman, et al., 2006, Wyss-Coray, 2006). Although those mechanisms are closely related to each other, the exact cause – effect relationship that leads to disease remains unknown.

In an effort to find the remaining genetic loci contributing to late-onset AD risk, several linkage studies have been performed (Blacker, et al., 2003, Kehoe, et al., 1999, Pericak-Vance, et al., 2000) and have reported a large number of chromosomal regions potentially harboring risk loci. Although some of these regions appear more consistently in the literature (Bertram and Tanzi, 2004) no gene has been implicated with a certainty similar to that for APOE. In fact, recent genome scans for association have shown that it is unlikely for another locus to exist in the genome bearing a single risk variant with the effect size of APOE (Waring and Rosenberg, 2008). The lack of consistency and the weak results in both linkage and association studies likely reflect an underlying genetic and allelic heterogeneity. In a previous study we addressed heterogeneity through the incorporation of covariates in a genome wide linkage analysis and detected strong linkage with a LOD score of 3.91 on chromosome 14q when the presence or absence of hallucinations was considered (Avramopoulos, et al., 2005). Sequencing of multiple patients excluded the presence of mutations in PSEN1, which is located in the same region (Avramopoulos, et al., 2005). Here we report on a follow-up study of genetic association for selected candidate genes in the 14q region that provides significant evidence for the involvement of the neuroglobin gene (NGB), and we further show that NGB has an RNA expression profile that supports its involvement in AD. Our data, in combination with the previous functional studies, make NGB a very interesting candidate as a genetic determinant of AD risk.

2. Materials and Methods

2.1 Sample Description

Genotyping sample

We initially screened 5 candidate genes in our linkage region using a study design that attempted to reduce heterogeneity by integrating information on the presence of hallucinations. We genotyped 99 patients with comorbid hallucinations, 125 patients without hallucinations and 152 cognitively healthy control subjects aged 58 to 99 years (Table 1). Cases were from the NIMH collection and were assessed for psychotic symptoms as described (Avramopoulos, et al., 2005) while controls were from the collection of the Indiana cell repository (NCRAD). Our follow up study on NGB included 351 cases from the NIMH and the Indiana repositories as well as 289 healthy controls (Table 1) aged 48 to 99 (median =74, mean=73.2 ), 197 from NCRAD and 92 cognitively healthy spouses of the offspring of the NIMH subjects.

The samples used for sequencing NGB were 24 cases from the NIMH families showing the strongest linkage on 14q and 24 healthy controls. Samples used for gene expression analyses were punches from the temporal lobe of 30 deceased patients with confirmed AD pathology and 26 controls with no brain pathology. The time between death and harvest of the brain (Post Mortem Delay; PMD) varied from 2 to 24 hours. Cases were older than controls (83.3±4.6 vs. 75.1±14.3 years mean±SD) and included more females (22 of 30 vs. 13 of 26). Both these variables were found to correlate significantly with the gene expression and were corrected for in our model. PMD was higher in the controls (11.5±5.1 vs. 7.7±4.1 hours mean ±SD) but was not found to correlate with gene expression measurements (p=0.8).

All procedures involving human subjects were in accordance with the Declaration of Helsinki and were approved by the Johns Hopkins Institutional review board.

2.2 Candidate Gene Identification and SNP selection

The 1 LOD interval identified in our previous linkage study spanned a 26 Mb region containing approximately 150 known genes. Through a systematic literature search on each of the genes for co-occurrence in publications with keywords relevant to AD or psychosis (dementia, Alzheimer’s, psychosis, hallucinations, schizophrenia, brain, neuron, hippocampus, presenilin, amyloid, amyloid beta, secretase, apoptosis, inflammation, oxidative stress, aging, cholesterol, mitochondria) we chose five genes for follow up: Dihydrolipoamide S-succinyltransferase (DLST), the hypoxia-inducible factor 1, alpha subunit (HIF1A), neuroglobin (NGB), numb homolog (NUMB), and sphingosine-1-phosphatase (SGPP1).

We downloaded reference genotype data from the HapMap project (Frazer, et al., 2007, The_International_HapMap_Consortium, 2003) genome browser (www.hapmap.org, October 2005 release) in each of the candidate genes and surrounding regions to the extent of linkage disequilibrium (LD) islands according to the Gabriel et al definition (Gabriel, et al., 2002). We then analyzed the SNPs with allele frequency greater than 2% for pairwise LD and removed SNPs that had a partner with r ^{greater than 0.8. We genotyped the remaining 26 tagging SNPs (}Table 2). In our follow up genotyping of 37 SNPs tagging the NGB gene (Table 3) we used the HapMap June 2006 release and extended the region 50 kb 5′ and 3′ of the gene to include potential regulatory sequences. The SNPs used in the follow up are listed in Table 3. The LD structure of the region made it necessary to extend into neighboring genes as noted in Table 3, however each of the SNPs outside NGB showed strong LD with the NGB region and their genotypes are likely to mirror genotypes of other SNPs within the gene.

2.3 Genotyping

Genotyping was performed in a 384 well format using the Taqman® method and “assays by design” from Applied Biosystems (Foster City, CA). Fluorescence end reads were performed on a ABI 7900HT sequence detection system and genotypes were called using SDS 3.1 software (Applied Biosystems). Follow up genotyping of 37 SNPs in NGB was performed as part of a larger genotyping project at the BROAD institute (www.broad.mit.edu/gen_analysis/genotyping) using the Illumina golden gate assay (Fan, et al., 2006). Illumina array calls were analyzed using Bead Studio software.

2.4 Sequencing

We analyzed by nucleotide sequencing all four known coding exons of NGB on 24 affected individuals from the families showing the strongest linkage in the region and 24 healthy controls. PCR reactions were used to amplify each exon and 30–60bp of its flanking sequence. Excess primers and nucleotides were inactivated by incubation with 1 U of shrimp alkaline phosphatase and 0.5 U of Exonuclease 1. The purified products were sequenced using the BigDye terminator 3.1 sequencing kit (Applied Biosystems) and the sequencing products were cleared of excess fluorescent nucleotides by isopropanol precipitation. The products were resuspended in EDTA and loaded on an ABI 3730 genetic analyzer. The sequences were aligned to the RefSeq (Pruitt, et al., 2007) transcript using the CodonCode sequence analysis software (CodonCode Corporation, Dedham, MA).

2.5 Gene Expression

To further explore the potential role of NGB in AD we measured the expression of NGB in 30 pathologically confirmed AD cases and 26 controls without brain pathology. RNA was extracted from the superior temporal lobe of punches from flash frozen brains using TRIzol reagent (Invitrogen) and first strand complementary DNA was generated using the TaqMan reverse transcription kit by Applied Biosystems (cat#N8080234). Primers were designed to quantitatively amplify from the NGB transcript cDNA a 100bp product crossing an exon – exon junction, thus ensuring that potential contaminating genomic DNA is not amplified. The PCR product from the transcript was quantified using SYBR-green real time detection with Applied Biosystems reagents (Foster City, CA, cat#4312704) and an ABI 7900 sequence detection system (Applied Biosystems). The ACTB housekeeping gene transcript was measured in the same manner to control for total RNA levels. Normalized NGB levels (ratios of NGB/ACT) were log transformed to achieve a normal distribution for further analysis, as described below.

2.6 Analytical Methods

Case control tests for genetic association were performed using the analytical tool kit UNPHASED (Dudbridge, 2008) (www.mrc-bsu.cam.ac.uk/personal/frank/software/unphased/). Tests of Hardy-Weinberg equilibrium (HWE) were performed using Haploview (Barrett, et al., 2005) (www.broad.mit.edu/mpg/haploview) which was also used for calculations of LD. Gene expression data was analyzed using generalized linear/non linear models in the STATISTICA version 7.1 software package (StatSoft Inc.; www.statsoft.com).

3. Results

Based on our literature search five candidate genes were chosen as functional and positional candidates and tested for association with AD. DLST (Dihydrolipoamide S-succinyltransferase) encodes a subunit of the alpha-ketoglutarate dehydrogenase complex, a mitochondrial respiratory component known to be defective in AD patients. It was chosen on the basis of several reports of associations between polymorphisms in DLST with AD (Ma, et al., 2001, Nakano, et al., 1997, Sheu, et al., 1999), although many others fail to identify an association (Brown, et al., 2004, Kunugi, et al., 1998, Matsushita, et al., 2001, Prince, et al., 2001). HIF1A (Hypoxia-inducible factor 1, alpha subunit) was chosen because it is one of the two subunits that comprise a heterodimeric transcription factor induced by hypoxia. It plays a role in neuroprotection by triggering genes involved in erythroporesis, angiogenesis, glucose transport, and glycolysis (Soucek, et al., 2003). Similarly NGB (Neuroglobin) was chosen as an oxygen binding protein expressed mainly in the nervous system (Burmester, et al., 2000) and playing a protective role against hypoxia (Sun, et al., 2001, Sun, et al., 2003). NUMB (Numb homolog) was chosen because it functions in neural cell proliferation; (Li, et al., 2003, Petersen, et al., 2002) and SGPP1 (Sphingosine-1-phosphatase) because its functions mediate cell growth and apoptosis (Cuvillier, 2002).

We successfully genotyped all SNPs but one (rs2883990), which was dropped from the analysis because it had a very low call rate (~50%). For the successful SNPs we achieved an average call rate of 98.4%, ranging from 93.6% (rs177378) to 100% (rs17750684). All SNPs were found to be in HWE in controls except for rs798847 in the HIF1A gene, which was only nominally (p=0.049) deviant from HWE and was kept for analysis. Analysis of the genotype data of each of the two case groups (psychotic, non-psychotic) against the controls for association failed to detect significant allelic associations, except for SNPs in NGB where many associations were detected in both comparisons as shown in Table 2. Among psychotic patients rs972725 and rs2216089 each showed nominally significant associations and are not strongly correlated with each other (r ^{= 0.3). Among non-psychotic patients rs10133981 showed a nominal association (p=0.0141) while rs972725 again showed an association trend (p=0.0615).}

We decided to further investigate NGB by genotyping additional markers to cover an extended region and to include new markers from the latest HapMap release. We included 37 SNPs in and around NGB in a larger study of 349 AD cases and 289 controls; this study did not target psychotic symptoms, but we decided that the prior association in both the psychotic and the non-psychotic cases justified the analysis regardless of psychotic status. There were 29 patients known to have psychotic symptoms and 53 AD patients known to be free of psychotic symptoms. There was a partial sample overlap with the previous analysis (82 cases, 136 controls). When comparing all AD cases with all controls, many SNPs showed a significant association, including rs733416 (p=0.0012) and rs888059 (p=0.009), which showed the highest significance (Table 3). These results did not change (p=0.0013 and p= 0.006) when excluding 41 younger controls between the age of 48 and 60. For SNP rs733416 the risk allele was the rare allele (16% in controls − 23% in cases) with a relative risk of 1.6. The SNPs rs733416 and rs888059 are not correlated with rs972725 (r^{=0.01 and r=0.06) which had previously shown association in both psychotic and non-psychotic patients and are somewhat correlated with each other (r=0.47). In order to assess the study-wide significance of our result we determined the number of independent comparisons performed. First we calculated the r between all pairs of SNPs. When two SNPs have for example an r of 0.7, 70% of the variance of one SNP is accounted for by the other and they represent 1.3 independent tests. For each SNP we determined its highest r from any pairwise SNP comparison, removed it from the set, and added (1- r) to the number of comparisons continuing until all SNPs were removed. We thus found that we had performed 36.4 independent comparisons, setting the desired p-value for study-wide significance to} p=0.00137, which was exceeded by rs733416. Removing the 29 cases with psychosis increased the effect size for the rare allele of rs733416 to a relative risk of 1.64 and the significance improved to p= 7×10^.

Published data strongly support a neuroprotective role for NGB. This along with our association results prompted us to test whether reduced expression of the gene might increase the risk for AD, and whether the AD risk-associated variant is also associated with reduced NGB expression. Table 4 summarizes our expression results while raw data are provided on Supplementary Table 1. In a control-only analysis we found that NGB expression was strongly negatively correlated with age (2% reduction every year; p=0.005) and lower by 24% in females (p<0.004), while PMD did not have any effect. When modeling NGB expression as a function of age, sex and disease status we found that it increased by 23% (p=0.007) in AD patients compared to controls after correction for age and sex, suggesting up-regulation by the disease process. Age and sex effects remained highly significant (2.2%/year p=0.001 and 21% less in females p= 0.0008). Inclusion of rs733416 genotype showed that NGB transcript expression was 13% lower in the carriers of the risk allele and reduced both in cases and in controls, however this difference did not reach statistical significance.

Sequencing of the four known exons of the NGB gene in 24 AD cases selected from families that showed the strongest evidence of linkage to 14q in our previous study and 24 older cognitively healthy controls revealed no exonic variants in either group.

4. Discussion

During our screening and follow up of positional and functional candidate genes for AD susceptibility we found a study-wide significant association between AD risk and a variant in the NGB gene. Gene expression analysis provided further evidence. Groups at higher risk like older subjects, females (Devi, et al., 2000, Gao, et al., 1998, Miech, et al., 2002, Seshadri, et al., 1997) and carriers of the risk allele, showed lower NGB expression. The presence of AD pathology led to an increase of NGB transcript consistent it’s known neuroprotective response to hypoxia (Sun, et al., 2001). Our data is therefore consistent with a model where women, older individuals and carriers of specific genotypes have lower levels of NGB leading to an increased risk for AD. It is likely that NGB up-regulation by the disease process in individuals with those risk factors is sometimes not sufficient to prevent neurodegeneration.

Neuroglobin is a small globin first identified by Burmester et al (Burmester, et al., 2000) as a member of the globin family after hemoglobin and myoglobin (and later cytoglobin), a member primarily expressed in the brain. It is a highly conserved protein, with only 6% of amino acids variant between mouse and humans (Burmester, et al., 2000) and, as we showed by sequencing, has no common coding variants in humans. It is a cytosolic protein localized near the mitochondria (Schmidt et al 2003), induced by neural hypoxia and cerebral ischemia and protecting neurons from hypoxic and ischemic injury (Greenberg, et al., 2008, Khan, et al., 2006, Sun, et al., 2001, Sun, et al., 2003). Despite its affinity to O₂NGB is unlikely to function as a delivery system as it does not have sufficient concentration to account for changes in intracellular O₂ levels (Brunori and Vallone, 2007) and it does not increase rate of oxygen consumption (Sun et al 2001). Instead, it may be involved in scavenging reactive oxygen and nitrogen species (NO and peroxynitrite) generated in response to brain hypoxia (Herold, et al., 2004). Extensive research on neuroglobin by Khan et al (Greenberg, et al., 2008, Jin, et al., 2008, Khan, et al., 2007a, Khan, et al., 2008, Khan, et al., 2007b, Khan, et al., 2006) shows that neuroglobin’s neuroprotection likely takes place at transduction of the death signal (Khan, et al., 2008). Most relevant to AD, neuroglobin attenuates amyloid beta neurotoxicity in vitro and the AD phenotype of transgenic mice (Khan, et al., 2007a), consistent with previous reports that it protects PC12 cells against amyloid beta induced cell injury (R.C. Li, et al., 2007). In agreement with our results, NGB levels have been shown to decline with in multiple rat brain regions which has been proposed to increase susceptibility to age related neurodegenerative disorders (Sun, et al., 2005). Our study is the first to examine genetic variation around NGB as a risk factor for AD in a human population. Our results are consistent with the literature, strongly supporting a neuroprotective role for neuroglobin through genetic association and gene expression.

According to the genome database’s (genome.ucsc.edu) microRNA target prediction track (Krek, et al., 2005), NGB contains multiple predicted binding sites for hs-miR-214 in its 3′UTR. This is of particular interest in view of our expression results, as microRNAs have a direct effect on both translation and transcript degradation. hs-miR-214 is located within and presumably transcribed with DNM3, a gene highly expressed in the brain. Although we did not find any DNA variation at the microRNA recognition sites and despite the fact that according to the target prediction data of the microrna.org bioinformatics resource hs-miR-214 has more than 4,000 predicted targets in the genome, it would be of interest to determine the expression profile of hs-miR-214 in the healthy and AD affected brain.

NGB is very close in proximity to its neighboring genes, within an area of strong LD thus we cannot exclude that the genetic association we observe reflects their involvement. In fact, our strongest associated SNP is located within a neighboring gene TMEM63C, a transmembrane protein of unknown function. Although we feel it is important to acknowledge this gene, it is also important to note that we have no reason to believe that rs733416 is the actual functional DNA variant increasing risk for AD. According to the LD structure that variant is equally likely to be located within NGB. In fact, Hapmap lists 3 SNPs within and in the 3′ flanking region of NGB that are in r^{≥ 0.7 with rs733416, and others are likely to exist.}

In conclusion our results, especially when combined with the literature, very strongly suggest that NGB warrants further attention for its likely involvement in neurological disorders. Further work on this gene and its product will provide important insight and possibly intervention targets for a multitude of very common conditions, including stroke and AD.

Supplementary Material

01

02

Acknowledgments