Background
There is a long-standing, inverse relationship between the prevalence of Alzheimer's disease (AD) and of rheumatoid arthritis (RA). Jenkinson and colleagues first described the decreased prevalence of RA in patients suffering from senile dementia of the Alzheimer's type as compared to cognitively intact individuals [
1]. Further retrospective studies of clinical and autopsy data revealed that patients with RA exhibit a reduced prevalence of AD [
2]. A study by Myllykangas-Luosujarvi and colleagues evaluating AD pathology in patients with and without RA revealed that AD-associated neuropathology occurred four times less often in patients with RA as compared to the general population [
3].
The basis of this inverse relationship is unclear but may include both genetic and environmental factors. RA and AD each have a strong genetic component, i.e., 50% of RA risk and 60% of AD risk is attributable to genetic factors, supporting the original hypothesis of Jenkinson and colleagues that genetics might explain the relationship between AD and RA [
4,
5]. Alternatively, anti-inflammatory medications used therapeutically for the treatment of RA could decrease AD risk by reducing AD-associated inflammation or via other mechanisms, .e.g., modulation of APP processing [
6,
7]. Supporting this possibility, an initial double-blind, placebo-controlled study by Rogers
et al. provided evidence that indomethacin slowed cognitive decline in patients with AD relative to placebo [
8]. These findings were further supported by Breitner and colleagues who found that multiple anti-inflammatory medications slow disease progression and delay disease onset [
9]. However, there has been little success replicating these findings in larger, randomized clinical trials [
10‐
13]. Hence, whether anti-inflammatory agents delay the onset of AD remains unclear.
The recent advent of RA genome wide association studies (GWAS) has identified single nucleotide polymorphisms (SNP)s associated with RA that provide a foundation for evaluating the initial hypothesis of Jenkinson
et al. that genetic variants that increase the risk of RA also decrease the risk of AD. To this end, we tested whether seventeen RA-associated SNPs with genome-wide significance were associated with AD in a two-stage analysis using separate AD case-control populations. We found that none of the seventeen alleles associated with increased RA risk were also associated with reduced AD risk. Rather, we found three RA-associated SNPs that were nominally associated with AD (p < 0.05). One of these SNPs, rs2837960, was found to be significantly associated with AD in a combined analysis of our Stage 1 and Stage 2 populations when the Stage 2 population was restricted to individuals of similar age as Stage 1. The gene closest to rs2837960 is
BACE2, the product of which has been implicated in amyloid protein precursor (APP) processing [
14,
15]. When we evaluated the expression of
BACE2 isoforms as a function of rs2837960, we found a trend for
BACE2 expression with rs2837960. In summary, genetic variants that increase RA risk do not decrease AD risk. The inverse relationship between RA and AD may thus be better explained by environmental factors such as the use of anti-inflammatory medications. Further functional investigation of rs2837960 is needed to elucidate the mechanism by which this SNP may modulate AD and RA.
Discussion
The primary finding of this investigation is that the majority of seventeen SNPs that exhibit a genome-wide significant association with RA are not associated with AD. Furthermore, the minor allele of rs2837960, which was found to be significantly associated with AD risk after combined analysis of Stage 1 and age-matched Stage 2 data, was associated with an increased risk of both RA and AD. Hence, these results contest the hypothesis that genetics underlie the inverse relationship between RA and AD, i.e. that alleles associated with an increased risk of RA are protective against AD. A secondary finding is that we have pursued the role of rs2837960 in its possible regulation of the nearby
BACE2 gene. We report the presence of multiple
BACE2 isoforms in human brain and that rs2837960 shows a trend for association with
BACE2tot and
BACE2d7, which represent total BACE2 and functional BACE2, respectively [
14]. In summary, the genetic underpinnings of RA have negligible overlap with AD with the exception of rs2837960, which is associated with both RA and AD, possibly through its effects on
BACE2 expression.
RA and AD each have a strong genetic component that accounts for approximately 50% and 60% of their risk, respectively [
4,
5]. The remainder of RA and AD risk is likely derived from environmental influences. The vast majority of RA-associated SNPs implicate gene products involved in immune system processes. Chronic inflammation of the brain is a common feature of AD pathology, raising the possibility that RA-associated SNPs that influence immune system function could influence AD risk [
25‐
27]. It is well established that some of the most strongly AD-associated genes, including
CLU,
CR1,
TNF and
CCR2, exhibit ontological association with immune system processes [
28‐
36]. Hence, the impetus for pursuing genetic overlap between RA and AD is greater than that provided by their epidemiologic relationship alone. However, our results indicate that RA-associated SNPs, which pertain largely to gene products involved in immune system processes, are not associated with AD.
There are several possible interpretations of our primary findings. The lack of overlap between RA-associated SNPs and AD could be due in part to the tissue-specific expression of DNA and RNA binding proteins required to interact with these SNPs to manifest effects on gene expression [
37]. However, if any of the seventeen RA-associated SNPs included in this study are capable of modulating peripheral immune system activity, either alone or in combination with each other, then it is probable that their peripheral effects on the immune system would indirectly affect immune system activity within the CNS. Evidence supporting the ability of peripheral inflammation to modulate CNS inflammation has been reported previously [
38]. Therefore, if RA-associated SNPs are only functional in the periphery then their effects on immune system function and inflammation should manifest in the CNS, even if the same SNPs do not modulate endogenous immune system function within the brain.
What is yet unclear is whether RA-associated alleles actually propagate inflammation and, if so, why they would not be expected to increase, rather than decrease, AD risk. In fact, the results of our study suggest that alleles that increase RA risk may likewise increase AD risk, i.e., rs2738960_G increases risk of both RA and AD, while rs3761847_G and rs2002842_A show a similar trend. If these observations are replicated in future studies, alleles that are pro-inflammatory may emerge as risk factors for both RA and AD. More explicitly, considering the role of genetics and environment in RA and AD, these results suggest that RA genetics alone may enhance rather than reduce AD risk. Hence, the inverse epidemiologic relationship between RA and AD is likely explained by an environmental RA-associated influence. In this regard, McGeer
et al. postulated that the reduced prevalence of AD in RA patients is related to the use of anti-inflammatory drugs for the treatment of RA [
2]. Multiple studies of anti-inflammatory agents have since been performed to test for their ability to modify AD risk and cognitive decline in AD patients, yielding mixed results [
8,
39‐
42]. To some extent, variability in study outcome may be explained by the additional ability of a subset of anti-inflammatory medications to reduce production of the neurotoxic Aβ
1-42 peptide [
6]. Further investigation is required to clarify the functional genetics of RA- and AD-associated SNPs and the role of anti-inflammatory medications in AD.
In pursuit of the functional genetics of rs2837960, which is associated with an increased risk of RA and AD, we investigated its association with the expression of
BACE2 isoforms in human brain [
43]. Thus, our secondary finding is that the minor allele of rs2837960 showed a strong trend for association with increased expression of
BACE2tot and
BACE2d7, the latter of which may represent the majority of functional BACE2 in human brain [
14].
BACE2 encodes a transmembrane aspartic protease and is ~75% homologous with
BACE1 with regard to amino acid sequence [
20]. Although the function of BACE2 is disputed, it appears to possess both β-secretase and α-secretase-like activities [
15]. Data obtained from the study of
BACE1/BACE2 double-knockout mice suggest that
BACE2 expressed in glia contributes significantly to Aβ production [
44]. This glial-specific expression is likely due to the more distal of the two distinct
BACE2 promoters, neither of which share similarity with the
BACE1 promoter [
18‐
20].
Several factors are consistent with the possibility that rs2837960, or SNPs in tight linkage with rs2837960 (LD of r
2 >0.8), are functional in modulating
BACE2 expression. This evidence includes the observation that (i) rs2837960 resides within a haplotypic block that spans the region containing both the proximal and distal
BACE2 promoters as well as the 5'UTR and first exon of
BACE2, (ii) the region surrounding rs2837960 and its proxy SNPs (r
2 = 1.0, ~4 kb window) is well conserved in primates per rVISTA analysis (data not shown), and (iii) the alleles of rs2837960 and its proxy SNPs are predicted to differentially affect transcription factor binding per PROMO 3.0 analysis of the TRANSFAC database (data not shown) [
45‐
47].
Other studies that examined the association between
BACE2 polymorphisms and AD risk have yielded mixed results [
48‐
54]. These studies differ with our study in that (i) they have focused on SNPs much more proximal to
BACE2 that are not in strong linkage disequilibrium with rs2837960 and (ii) they generally utilized smaller populations than those utilized in our present study. Future analyses of the association between
BACE2 SNPs and AD should therefore take into account SNPs that are more distal to
BACE2, such as rs2837960, as well as utilize larger population sizes that are sufficiently powered to detect associations with AD. Thus, in future studies rs2837960 may emerge as a risk factor for both RA and AD that functionally modulates
BACE2 expression. Elucidation of the precise mechanism by which rs2837960, or a SNP that is proxy to it, modulates
BACE2 expression may contribute to a better understanding of the role of
BACE2 in both AD and RA pathology.
Materials and methods
SNP Selection
The Human Genome (HuGE) Navigator (
http://www.hugenavigator.net) was queried using the search term "rheumatoid arthritis" to identify RA-associated SNPs of genome-wide significance [
55]. Six available studies utilizing individuals of European decent were chosen to mimic the AD MAYO GWAS demographics (Table
7). Sample sizes ranged from ~1,600 (810 RA, 794 non-RA) to ~25,500 (7,322 RA, 18,262 non-RA). Thus, we identified twenty-eight candidate SNPs for study from the literature. SNPs which appeared more than once or that were in tight linkage disequilibrium with each other, i.e. r
2 >0.8 (according to the CEU HapMap population), were considered to be redundant and only those with the lowest RA-associated p-value were retained for further analysis [
17]. This effort reduced the number of candidate SNPs to twenty-two. If a candidate RA-associated SNP was not available within the Mayo Clinic AD GWAS, an appropriate proxy SNP (LD of r
2 >0.8) was selected by using the HapMap-based SNAP proxy search (
http://www.broadinstitute.org/mpg/snap/) [
56]. Ultimately, seventeen of the candidate SNPs or their proxies were present in our AD GWAS dataset.
Table 7
RA GWAS reports identifying RA genetic risk factors
Gregersen et al.,
Nat Genet
, 2009
| 19503088 | 5 |
Raychaudhuri et al.,
Nat Genet
, 2008
| 18794853 | 9 |
Julia et al.,
Arthritis Rheum
, 2008
| 18668548 | 2 |
WTCCC,
Nature
, 2007
| 17554300 | 7 |
Plenge et al.,
N Engl J Med
, 2007
| 17804836 | 3 |
Plenge et al.,
Nat Genet
, 2007
| 17982456 | 2 |
Case and Control Samples
The Mayo Clinic case-control samples used for the Stage 1 analysis have been described in detail in a prior GWAS publication [
57]. The Mayo Clinic case-control series used for the Stage 2 study have also been previously described [
58]. Briefly, clinical diagnoses of probable AD were made according to NINCDS-ADRDA criteria for samples from Jacksonville, FL (JS) and Rochester, MN (RS); age-matched controls had a score of 0 on the Clinical Dementia Rating scale. Additional samples were obtained from the Mayo Clinic brain bank (AUT); autopsy-confirmed diagnosis of AD (NINCDS-ADRDA, Braak score >4.0) was utilized for AD samples while non-AD samples exhibited limited AD pathology (Braak <2.5, not including other unrelated pathology).
AD Association Testing
Association testing of RA-associated SNPs for AD risk was carried out in two stages by using PLINK software (
http://pngu.mgh.harvard.edu/purcell/plink/) [
59]. All genotyped samples were subject to strict quality control including elimination of samples with call rates <90%, MAF <0.01, HW p < 0.001, discrepancy between reported and genotyped sex, cryptic relatedness and discordant genotype clustering. Stage 1 consisted of 1264 non-AD and 843 AD subjects with average ages of 74.3 ± 4.5 (age at last assessment, mean ± SD) and 72.4 ± 4.6 years (age at diagnosis), respectively. The non-AD and AD groups in this series consisted of 51.7% and 57.5% female individuals, respectively. Stage 1 samples were genotyped by using HumanHap300-Duo Genotyping BeadChips processed with an Illumina BeadLab station (Illumina, San Diego, CA) at the Mayo Clinic Genotyping Shared Resource center (Rochester, MN).
We proceeded to test for an association between the seventeen RA-associated SNPs and AD in this Stage 1 case-control population. Stage 1 association testing was performed by using PLINK to generate allelic models that included odds ratios (OR), 95% confidence intervals (CI)s and uncorrected p-values. Logistic regression was also performed using the covariates age, sex and APOE genotype. With regards to multiple testing we expected to obtain approximately one false positive result given α = 0.05 (seventeen unique SNPs; 17 tests × 0.05 = 0.85). Bonferroni correction for multiple testing was also applied to data generated using allelic models.
Stage 2 samples were genotyped by using SEQUENOM MassARRAY iPLEX Platform (Sequenom, San Diego, CA). Overall, Stage 2 consisted of 2677 non-AD and 1102 AD subjects with average ages of 81.0 ± 6.2 and 83.5 ± 6.6 years of age, respectively. The non-AD and AD groups were composed of 55.0% and 64.0% female individuals, respectively. Stage 2 AD-SNP association testing was performed using only the three SNPs identified in Stage 1 as being associated with both RA and AD. PLINK software was used to generate odds ratios, 95% CIs and p-values per allelic modeling. Logistic regression including the covariates age, sex and APOE genotype was also performed. To evaluate the overall significance of Stage 1 and 2 data, they were combined and examined collectively.
Due to the considerable difference in mean age between Stage 1 and Stage 2 individuals, and due to our interest in focusing on genetic, rather than environmental factors, we also chose to examine only Stage 2 individuals between 60 and 80 years of age. Hence, when Stage 2 was limited to individuals between 60 and 80 years of age, our analysis included 912 non-AD and 186 AD subjects with average ages of 73.9 ± 3.8 and 72.8 ± 5.1 years. The non-AD and AD groups consisted of 49.9% and 57.0% female individuals, respectively. Similar to our analysis of our initial Stage 2 population, logistic regression of this modified Stage 2 population was also performed to test for an association between the three AD-associated SNPs from Stage 1. Furthermore, we evaluated the overall significance of RA-associated SNP associations with AD in combined Stage 1 and Stage 2 individuals between 60 and 80 years of age.
Human Tissue
Human anterior cingulate brain specimens were generously provided by the Sanders-Brown AD Center Neuropathology Core and have been described elsewhere [
60]. The samples were from deceased individuals with an average age at death for females of 82 ± 7 years (mean ± SD, n = 29) and for males of 81 ± 8 (n = 24). The average postmortem interval (PMI) for females and males was 3.2 ± 0.8 h and 3.0 ± 0.8 h, respectively.
To gain insights into the functionality of rs2837960 we tested for an association between rs2837960 and
BACE2 expression in human brain. We first screened anterior cingulate samples for the presence of
BACE2 and its known alternatively spliced isoforms that lack exons 7 and 8, respectively. Total RNA and genomic DNA were prepared from human tissue samples; the RNA was reverse transcribed as we have reported elsewhere [
61,
62]. Conventional PCR using Platinum Taq (Invitrogen, Carlsbad, CA) was used to amplify the region of
BACE2 spanning exons 6-9 (Table
8). Thermal cycling conditions consisted of denaturation at 95°C for 5 min followed by 32 cycles of 95°C for 30 s, 60°C for 30 s, 72°C for 1 min and a final extension at 72°C for 2 min. PCR products were separated using 8% TBE-PAGE gel electrophoresis and visualized using SYBR-gold fluorescent stain (Invitrogen) and a fluorescence imager (FLA-2000, Fuji). To confirm the identities of the
BACE2 splice variants, bands were excised, purified and directly sequenced (Davis Sequencing, Davis, CA).
Table 8
Primers used for analyses of BACE2 isoform expression.
BACE2 Exon 6 Forward: | ATAACGCAGACAAGGCCATC |
BACE2 Exon 9 Reverse: | GGACACAGTTGCTGGCTACA |
BACE2
isoform specific RT-PCR primers
| |
BACE2 Exon 6 Forward: | GCCCCAGAAGGTGTTTGAT |
BACE2 Exon 6-8 Junction Reverse: | GGCTGAATGTAAAGCAGAG |
BACE2 Exon 5 Forward: | TGGGTGGAATTGAACCAAGT |
BACE2 Exon 6 Reverse: | GATGGCCTTGTCTGCGTTAT |
To quantify total
BACE2 expression (
BACE2tot) and expression of the
BACE2 isoform lacking exon 7 (
BACE2d7) we designed separate primer sets.
BACE2tot expression was measured by amplification of a product spanning a non-alternatively spliced region of
BACE2 (exons 5-6). Isoform-specific primers designed to amplify
BACE2d7 consisted of a forward primer specific to
BACE2 exon 6 and a reverse primer specific to the junction of exons 6-8 (Table
8). DNA samples were genotyped using a TaqMan SNP Genotyping Assay (ID # C_2688271_10; ABI, Carlsbad, CA).
Quantitative real-time PCR reactions contained ~20 ng of sample cDNA together with 10 μl of PerfeCTA SYBR green SuperMix (Quanta Biosciences, Gaithersburg, MD), 10 μl of ddH2O and 20 pmol of forward and reverse primers. Cycling conditions included a 3 minute denaturation step at 95°C followed by 40 cycles of denaturation for 15 seconds at 95°C and annealing/extension for 45 seconds at 60°C using an MJ Opticon 4 thermal cycler (Biorad, Hercules, CA). A melting curve was generated following cycling to assess the purity of amplification product. Fidelity of amplification was also assessed via visual inspection of PCR products on 8% TBE-PAGE gel stained with SYBR gold. Standard curves were generated from purified PCR products that were quantified by A260/280 spectrophotometric analysis. Standard curves were then used to calculate the copy number for each BACE2 isoform measured.
Hypoxanthine-guanine phosphoribosyltransferase (
HPRT) and ribosomal protein L32 (
RPL32) were used as housekeeping genes per the analysis of geNorm software as described previously [
63‐
65]. Expression levels of each of these genes were measured by using quantitative real-time PCR and gene specific primers under conditions identical to cycling conditions for
BACE2. Standard curves were used to generate exact copy numbers, which in turn were used to calculate the sample-specific geometric mean of
HPRT and
RPL32 expression. The geometric mean was in turn used to normalize subsequent
BACE2 expression data. Analysis of the association between
BACE2 isoforms and rs2837960 genotype was performed using non-parametric Jonckheere-Terpstra testing (PASW Statistics, v.18, IBM, Somers, NY).
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
CRS and SE contributed to experimental design and wrote the manuscript. CRS performed RT-PCR of BACE2 isoforms, rs2837960 genotyping of human brain tissue and association testing between BACE2 isoform expression and rs2837960. FZ performed SEQUENOM genotyping of samples for Stage 2 analyses. SGY, FZ and CRS performed statistical analysis of RA-associated SNPs and AD GWAS data. All authors have read and approved the final manuscript.