Genome-wide Meta-analysis on the Sense of Smell Among US Older Adults.

Olfactory dysfunction is common among older adults and affects their safety, nutrition, quality of life, and mortality. More importantly, the decreased sense of smell is an early symptom of neurodegenerative diseases such as Parkinson disease (PD) and Alzheimer disease. However, the genetic determinants for the sense of smell have been poorly investigated. We here performed the first genome-wide meta-analysis on the sense of smell among 6252 US older adults of European descent from the Atherosclerosis Risk in Communities (ARIC) study, the Health, Aging, and Body Composition (Health ABC) study, and the Religious Orders Study and the Rush Memory and Aging Project (ROS/MAP). Genome-wide association study analysis was performed first by individual cohorts and then meta-analyzed using fixed-effect models with inverse variance weights. Although no SNPs reached genome-wide statistical significance, we identified 13 loci with suggestive evidence for an association with the sense of smell (Pmeta < 1 × 10). Of these, 2 SNPs at chromosome 17q21.31 (rs199443 in NSF, P = 3.02 × 10; and rs2732614 in KIAA1267-LRRC37A, P = 6.65 × 10) exhibited cis effects on the expression of microtubule-associated protein tau (MAPT, 17q21.31) in 447 frontal-cortex samples obtained postmortem and profiled by RNA-seq (P < 1 × 10). Gene-based and pathway-enrichment analyses further implicated MAPT in regulating the sense of smell in older adults. Similar results were obtained after excluding participants who reported a physician-diagnosed PD or use of PD medications. In conclusion, we provide preliminary evidence that the MAPT locus may play a role in regulating the sense of smell in older adults and therefore offer a potential genetic link between poor sense of smell and major neurodegenerative diseases.


INTRODUCTION
T he human olfactory system provides us a remarkable ability to detect and discriminate odorants, and warns us of spoiled foods or dangerous environments. Olfactory dysfunction may not only result in poorer nutritional status and adverse effects on safety and quality of life, but is associated with higher mortality. 1, 2 The sense of smell decreases with age. Poor sense of smell affects over 20% of US older adults. 3 Previous research suggests that the loss of sense of smell is among the earliest symptoms of neurodegenerative diseases such as Parkinson disease (PD) 4,5 and Alzheimer disease (AD). 6 For PD, this observation is consistent with the Braak hypothesis that Lewy body deposition first affects the olfactory bulb and lower brain stem before spreading into substantia nigra and cortex. 7 Moreover, as the nasal cavity directly interacts with the environment, a dual-hit hypothesis was proposed that environmental neurotoxicants may initiate PD development in the olfactory structures decades before PD clinical diagnosis. 8 Therefore, research on the sense of smell in older adults may lead to a better understanding of the natural history and etiology of PD and potentially for other neurodegenerative diseases.
However, the genetic basis for the loss of the sense of smell is largely unknown. Limited studies have suggested that genetic variations at human olfactory receptors or the ApoE e4 allele may contribute to variable olfactory sensitivity and perception. 9,10 Two earlier studies also suggest potential links to regions on chromosome 4q. 11,12 Here, we report the first ever genome-wide meta-analysis on the sense of smell among 6252 US older adults of European descent from 3 cohorts: the Atherosclerosis Risk in Communities (ARIC) study, the Health, Aging, and Body Composition (Health ABC) study, and the Religious Orders Study and the Rush Memory and Aging Project (ROS/MAP). We further examined suggestive loci with gene-based analyses, pathwayenrichment analyses, as well as expression quantitative trait loci (eQTL) analysis using 447 postmortem prefrontal-cortex tissues. Overall, our results suggest that microtubule-associated protein tau (MAPT), a major susceptibility gene for PD 13,14 and possibly for AD, 15 may also play a role in regulating the sense of smell in older adults.

Participating Studies
Details of participating cohorts were published elsewhere. [16][17][18][19] Briefly, the ARIC study is an ongoing longitudinal study that was established in 1987 to 1989 to investigate risk factors for cardiovascular diseases. 16 The sense of smell was measured as part of the ARIC Neuro-Cognitive Study (NCS) in 2011 to 2013. Of the 6523 participants of ARIC NCS, 4066 were eligible after excluding non-Whites (n ¼ 1563), and participants without valid data on the smell identification test (n ¼ 279) or genome-wide association study (GWAS) genotyping (n ¼ 615). The Health ABC is a prospective study established in 1997 to 1998 to investigate risk factors for disability and functional decline among older adults. 17 The sense of smell was tested among 2537 participants in 1999 to 2000. After excluding self-reported non-Whites (n ¼ 976), participants without valid data on the smell identification test (n ¼ 106) or GWAS (n ¼ 111), 1344 were eligible for analysis. The ROS and MAP studies are longitudinal cohorts established in mid-1990s to investigate aging and AD incidence and progression among adults 65 years or older. 18,19 The sense of smell was measured as part of its clinical evaluation. Our analysis included 1065 Whites after excluding participants without data on genotyping (n ¼ 661) or the sense of smell (n ¼ 644).

The Smell Identification Tests
All 3 cohorts used validated smell identification tests to examine the sense of smell: the 12-item Sniffin' Sticks screening test (Burghart, Wedel, Germany) 20 in ARIC and the 12-item Brief Smell Identification Test (B-SIT, Sensonics, Haddon Heights, NJ) 21 in the Health ABC and ROS/MAP. Both tests assess participant's sense of smell to correctly identify 12 daily odorants, although the exact odorants were somewhat different. Briefly, participants were instructed to smell each odorant and choose the correct odorant from 4 possible answers in a multiple-choice format. One point was given for each correct answer with a total score from 0 to 12. Both tests have been validated and widely used in clinical and epidemiological studies. 20,21 All cohorts assessed global cognitive function during the same study visit in which the sense of smell was evaluated. The ARIC and ROS/MAP studies used the Mini-Mental State Examination (MMSE) that has a maximal score of 30, whereas the Health ABC study used the Modified Mini-Mental Status Examination (MMMSE) with a maximal score of 100. Each cohort also determined ApoE genotypes by genotyping 2 ApoE variants at codons 130 and 176 (formerly 112 and 158) separately.
Genotyping, Imputation, and Statistical Analysis GWAS genotyping was performed using the Affymetrix Genome-wide Human SNP array 6.0 in ARIC and ROS/MAP, and Illumina Human1M-Duo BeadChip in Health ABC. Imputation was performed in each cohort using MACH based on HapMap Phase II CEU build 36 reference panel. Prespecified criteria were used for quality control of each cohort.
Each cohort performed its own GWAS analysis following a standardized procedure with ProbABEL 22 or PLINK, 23 using a linear regression model with analysis of allele dosages. The outcome was the natural log of the sense of smell score plus 1 to account for the few participants who had a score of 0. Potential confounders were adjusted for in 2 separate models: first for age, gender, study sites, cognitive function, and the first 2 principal components (PCAs) in Model 1, and then additionally for ApoE e4 allele in Model 2. We further adjusted for ApoE e4 because it is the most important genetic risk factor for AD and predicts poor sense of smell in all participating cohorts.
We performed meta-analysis using a fixed-effect model with inverse variance weights in METAL, 24 after filtering out SNPs with MAF < 1% [25][26][27] or low imputation quality (r 2 < 0.3). We applied genomic control for each dataset before meta-analysis (e-Methods). We considered P value < 5 Â 10 À8 as genome-wide significant and P value between 1 Â 10 À5 and 5 Â 10 À8 as suggestive evidence for an association.
We used VEGAS 28 (Versatile Gene-based Association Study) for gene-based analysis and calculated individual gene-wise P values for each gene in relation to the sense of smell, and used binomial test to identity enriched ingenuity canonical pathways (http://www.ingenuity.com; IPA, Ingenuity 1 Systems) based on the results of meta-analysis in Model 2, the enrichment test P values were derived from 10,000 rounds of permutation tests (e-Methods).
To explore the functional relevance of top SNPs from the GWAS analysis, we applied the publicly available data from the RegulomeDB (http://regulomedb.org/) 29 (e-Methods). We then generated RNA expression data using frozen postmortem dorsolateral prefrontal cortex (DLPFC) brain tissue from 447 ROS/ MAP participants to define the biological consequence of the risk variants of identified SNPs (e-Methods).

Ethical Aspects
Individual study protocols were approved by relevant institutional review boards and all study participants provided written consent. Table 1 presents characteristics of study participants. As expected, the sense of smell decreased with age and was lower in men than women. Further, higher cognitive score was associated with better sense of smell, whereas the ApoE e4 allele was related to poorer sense of smell. These population characteristics were similarly associated with the sense of smell across cohorts (Table S1, http://links.lww.com/MD/A526). Although these studies used different screening tests of smell identification, the distributions of the sense of smell scores were comparable across cohorts ( Figure S1, http://links.lww.com/ MD/A526).
We conducted gene-based analysis using VEGAS 28 to identify genes that harbor multiple association signals based on P meta values from the model that adjusted for ApoE. The SNPs were mapped to 17,693 genes. No genes were statistically significant after Bonferroni correction (P < 2.8 Â 10 À6 ¼ 0.05/ 17,693). Genes with P < 1.0 Â 10 À3 were listed in Table S5, http://links.lww.com/MD/A526. Interestingly, MAPT (P gene ¼ 9.4 Â 10 À5 ) was among the top genes identified in this analysis along with nearby LRRC37A (P gene ¼ 2.9 Â 10 À5 ) and NSF (P gene ¼ 1.4 Â 10 À5 ), all located on chromosome 17. LRRC37A and NSF each harbor 1 of the 2 annotated eQTLs (rs2732614 and rs199443, respectively) whereas the expression of MAPT is regulated by both SNPs in brain tissues. 30 We further conducted 2 pathway enrichment analyses (top 10 identified pathways listed in Table S6, http://links.lww.com/ MD/A526). The first analysis was based on results from Model 2 and we observed 21 significantly enriched canonical pathways with the permutation P < 0.05. These include the CDK5 and p38 MAPK signaling pathways both of which are implicated in neurodegeneration and MAPT is a member of both pathways. The second analysis were based on the 164 genes with P gene < 0.01 identified by VEGAS. This analysis identified 113 enriched canonical pathways with the permutation P < 0.05. The CDK5 signaling pathway was again identified as 1 of the top ranked pathways.
We further examined MAPT expression using RNA-seq data of 447 postmortem frontal-cortex tissues from the ROS/ MAP study in order to define the biological consequence of the risk variants of rs2732614 and rs199443. We found strong correlations between genotypes of the risk allele (C allele for both rs2732614 and rs199443) and higher expression level of MAPT (P < 10 À15 , Figure 2); however, we observed no significant correlation between these 2 SNPs and the expression levels of NSF or LRRC37A (rs2732614: P ¼ 0.62 for NSF and 0.54 for LRRC37A; rs199443: P ¼ 0.79 for NSF and 0.11 for LRRC37A).
As MAPT is 1 of the most important susceptibility genes for PD, we further performed a sensitivity analysis for the 13 SNPs

DISCUSSION
To the best of our knowledge, this is the first genome-wide meta-analysis on the sense of smell among older adults. The study population includes 3 well-characterized aging cohorts with objective measurements for smell identification. Although no SNP reached genome-wide significance, the preponderance of the evidence suggests MAPT as a potential susceptibility locus for poor sense of smell. This is particular interesting because MAPT is 1 of the most important susceptibility genes for late-onset sporadic PD 13,14 and possibly for AD. 15 Smell impairment is common among older adults, particularly in patients with neurodegenerative diseases. More importantly, recent data suggest that decreased sense of smell may precede the clinical onset of both PD 4,5 and AD. 6 For example, in the Honolulu Asia Aging Study, individuals in the lowest quartile of the sense of smell score were 5 times more likely to develop PD in the next 4 years than those in the highest quartile 5 ; further poor sense of smell was associated with higher prevalence of incidental Lewy bodies in postmortem brains, a condition that may represent prodromal PD. 32 Similar observations have also been made for both mild cognitive impairment and AD. 6,33 Therefore, the declining sense of smell in older adults may indicate underlying neurodegenerative processes. Characterizing the genetic and environmental determinants for the sense of smell in the context of aging may provide targets for intervention with enormous potential translational impact.
The human MAPT locus has long been associated with the risk for PD. 13,14 A recent meta-analysis of GWAS data further implicated MAPT as a susceptibility locus for late onset sporadic AD. 15 The human MAPT encodes the microtubule-associated protein tau. It is notable that tau-related pathology was found in the bulbar component of the anterior olfactory nucleus only in patients of neurodegenerative disorders with marked loss of the sense of smell, such as AD, PD, and Lewy body dementia. 34,35 Experimental studies further suggest that overexpression of tau may contribute to olfactory dysfunction in vivo. 36,37 These findings indicate that overexpression or abnormal aggregation of tau may represent a common mechanism between olfactory dysfunction and several neurodegenerative diseases.
The genomic architecture in the region spanning MAPT is highly complex. It covers $1.8 Mb block of LD encompassing not only MAPT but also several other genes that have been linked to neurodegenerative diseases, such as STH (Saitohin) and NSF (N-ethylmaleimide sensitive factor). We, however, in the RNA-seq data of 447 postmortem frontal-cortex tissues, found that rs2732614 and rs199443 had statistically significant cis effects on the expression of MAPT (P < 10 À15 ), but not on NSF or LRRC37A. Additional gene-based and pathway-enrichment analyses also showed the potential of MAPT as a candidate susceptibility gene for poor sense of smell. Although our study provides fairly consistent evidence that MAPT may be a candidate susceptibility locus for the sense of smell in older adults of European descent, fine-scale mapping of this LD block with next generation sequencing will be necessary to identify potential causal locus.
ApoE e4 is the most important genetic risk factor for lateonset sporadic AD 38 and has been linked to olfactory dysfunction in several studies. 9 The present study confirms that the ApoE e4 allele is a risk factor for poor smell identification even after accounting for cognitive function. In our initial analysis, we also identified an intronic SNP in TOMM40 (rs2075650) in association with the sense of smell; but this signal was extinguished after adjustment for the ApoE e4 allele. Although the mechanisms are yet unknown, preliminary studies showed that ApoE was expressed at high levels in various types of cells in the olfactory epithelium and its underlying lamina propria, and that ApoE might play a critical role in olfactory nerve regeneration in mice. 39,40 Furthermore, experimental studies suggest that in contrast to the ApoE e2 and e3 alleles, the ApoE e4 fails to promote neurite outgrowth in olfactory epithelium cultures, which may lead to olfactory dysfunction in AD. 41 The main limitation of the current study was the lack of independent confirmation of GWAS signals. GWAS analysis often requires very large sample size to obtain adequate power to detect modest effects of underlying variants in complex diseases. However, few large population-based studies of older adults have both the sense of smell and GWAS data. We were able to identify data from 3 well-established cohorts with similar population characteristics and data collection. We chose to combine our samples for SNP discovery rather than set up the conventional 2-stage design to maximize power to detect signals. We did, however, follow up suggestive signals with functional and expression analyses. Finally, the determinants for the sense of smell are likely complex 42,43 and our study was conducted among old White adults, therefore the results should be interpreted in this context. Further studies are needed to confirm our findings and to examine whether they could be generalized to other populations.
In conclusion, we report the results of the first comprehensive GWAS analysis to understand the genetic basis for the sense of smell in older adults of European ancestry. We provide  novel evidence that the MAPT locus may regulate the sense of smell in older adults. This finding warrants independent confirmation and further mechanistic investigations into how MAPT region contributes to the loss of the sense of smell and the development of related neurodegenerative diseases.