Protective Variants in Alzheimer’s Disease

Abstract

Purpose of review:

Over the last decade over 40 loci have been associated with risk of Alzheimer’s disease (AD). However, most studies have either focused on identifying risk loci or performing unbiased screens without a focus on protective variation in AD. Here, we provide a review of known protective variants in AD and their putative mechanisms of action. Additionally, we recommend strategies for finding new protective variants.

Recent findings:

Recent Genome-Wide Association Studies have identified both common and rare protective variants associated with AD. These include variants in or near APP, APOE, PLCG2, MS4A, MAPT-KANSL1, RAB10, ABCA1, CCL11, SORL1, NOCT, SCL24A4-RIN3, CASS4, EPHA1, SPPL2A, and NFIC.

Summary:

There are very few protective variants with functional evidence and a derived allele with a frequency below 20%. Additional fine mapping and multi-omic studies are needed to further validate and characterize known variants as well as specialized genome-wide scans to identify novel variants.

Introduction

An overview of Alzheimer’s Disease

Alzheimer’s disease (AD) is a debilitating neurological disease characterized by progressive cognitive deterioration, eventual loss of independence necessitating full-time care, and ultimately, death [1]. The first neuropathological AD hallmark is aggregation of extracellular amyloid-β (Aβ) peptides into amyloid plaques, followed by cytotoxic intracellular neurofibrillary tau tangles (NFTs). Aβ and Tau aggregation are accompanied by neuroinflammation, gliosis and neurodegeneration [1,2]. Pathology starts in the temporal lobe, including the hippocampus, before spreading to the prefrontal cortex, then the rest of the neocortex [1].

AD is divided into early onset (EOAD) and late-onset Alzheimer’s disease (LOAD): LOAD accounts for at least 95% of AD and is defined by onset later than 60 years of age. Apolipoprotein E (APOE) is the predominant genetic risk factor for LOAD, with three alleles, ε2, ε3, and ε4, contributing to disease risk. The ε4 allele confers the largest known common genetic risk for LOAD, increasing AD risk 2–3 fold (HR = 2.47) for European heterozygotes and 7–11 (HR = 8.74) fold for European homozygotes [3], with similar effects in other populations [4–6]. Conversely, the heterozygous ε2/ε3 allele is associated with approximately halved (HR = 0.62) risk in Europeans [3]. Genome-wide association studies (GWAS) of case/control populations have identified a further 27 loci associated with LOAD (nearest gene: ABCA7, BIN1, CD2AP, CD33, CLU, CR1, EPHA1, MS4A4A, MS4A4E, MS4A6A, PICALM, HLA-DRB5, PTK2B, SORL1, SLC24A4-RIN3, DSG2, INPP5D, MEF2C, NME8, ZCWPW1, CELF1, FERMT2, CASS426, ADAM10, BCKDK-KAT8 and ACE [7–12]), with modest effect sizes ranging from 0.54 – 1.21 OR. These loci are generally named for their closest gene, but intergenic, promoter and intron variants may be enhancers or repressors of other genes, and most loci still require fine-mapping.

EOAD only accounts for between 1–5% of all AD cases with onset ranging from 30–60 years of age, with 13% of cases autosomal dominantly inherited [13]. However, sporadic EOAD and LOAD may share a common architecture [14,15].

Protective variants in AD

As with other complex diseases, understanding both risk and protective factors for AD will be key to understanding, predicting, diagnosing and treating the disease. It is important to recognize that every genetic locus involves multiple alleles, where one is the reference and the others are protective or confer risk. The most common allele in the global population [16] is the reference for the purpose of this paper. This can include single nucleotide polymorphisms (SNPs), insertion/deletions (INDELs), and structural variants (SVs), though SNPs are the most commonly studied variant type. Protective variants are, therefore, by definition, variants where a less common allele is associated with reduced risk of developing a disease or delays onset.

Though some studies do not make this distinction, we also require a derived minor allele because this means the allele developed in humans [17]. As a post-reproductive disorder, AD may not be under active selection, but early symptoms can manifest during reproductive and child-rearing age, and there are several other ways AD genes could be under selection (for a review see [18]).

Protective variants can be important for both prevention and treatment of AD: Knowledge of the genetic basis of successful aging without dementia is critical for understanding the biology of individual resistance or resilience to AD neuropathology. Thus, knowledge of protective variants could refine AD risk assessment in asymptomatic individuals, aiding AD prevention [19]. Furthermore, the identification of genetic variants that confer protection via a loss-of-function or gain-of-function offers potential drug targets [20], especially if they protect against known AD risks. Most drug candidates never reach the clinic, but those with the same mechanism as protective variants have a higher success rate [20], already having been validated in human disease. Genetic modifiers in particular can protect against deleterious effects of other genetic risk factors by nonlinearly interacting to reduce penetrance or effect size [20]. This is of particular importance for AD because genetic modifiers may allow us to protect against the effects of the common APOE ε4 allele or the highly penetrant autosomal dominant AD mutations. Drugs are already being developed for some variants with individual protective effects, and genetic modifiers are promising targets.

Here, we provide a review of the known AD protective variants and their mechanisms of action, and we recommend strategies for finding new protective variants.

Literature Search

To identify AD associated protective variants, we conducted a systematic PubMed database search for all articles published up to October 2018 (inclusive). We used the following search term: (Alzheimer’s OR Dementia OR “amyloid protein” OR “tau protein”) AND (Protect\* AND (Genetic OR SNP OR “Single Nucleotide Polymorphism*” OR Polymorphism, Single Nucleotide OR mutation OR allele)) English [LA].

We used five study inclusion criteria: 1) the study conducted a genetic association or linkage analysis; 2) the effect allele was derived, had a minor allele frequency (MAF) below 20%, and was protective; 3) studies used case/control populations, cohorts or families; 4) the outcomes were AD dementia or other clinically relevant endophenotypes; 5) the study participants were at least 65 years old. Studies were excluded if they were animal studies, case reports, review articles, were conducted in a non-dementia clinical population or reported effects for haplotype, interaction analysis or structural variants.

Article citations and abstracts were imported into Covidence and independently rated against the inclusion/exclusion criteria by SJA and BFH for nomination into the full-text screen. Articles selected for full-text screening were further assessed for inclusion in the final review. The following variables were extracted from articles that were selected for the final review: 1) Study design (i.e. case/control, cohort, or family; candidate gene study, GWAS, rare variant analysis); 2) Sample characteristics (i.e. age, gender, sample size, ethnicity); 3) genetic variation (i.e. loci, rsID, effect size, effect allele). In addition to the above search, we examined reported associations from GWAS [11,12,21–24].

Search Results

The PubMed search returned 1,975 articles. We first removed 1,746 based on the inclusion/exclusion criteria, leaving 229 studies. We then excluded 111 studies for the following reasons: non-significant results, haplotype analyses, interaction analyses, structural variant analysis, and APOE only results. The remaining 124 studies reported protective associations for 183 genetic variants (Supplementary Table 1). 59 variants had a derived effect allele with a frequency below 20%, as reported in the Genome Aggregation Database [16]. Thirteen of the 59 variants were significant at p < 0.05 in GWAS of AD conducted in either European, African-American or Trans-ethnic populations and 7 were rare variants that were not available in GWAS. We also included 11 SNPs from seven GWAS of AD that met our criteria for protective variants [11,12,21–24].

Of the 20 variants identified in the literature search we do not discuss the following candidate genes due to limited evidence of functional association in AD and limited evidence of replication: CETP-rs2303790 [25], CHI3L1-rs4950928 [26], IL16-rs4072111 [27], HFE-rs1799945 [28], LRRK2-rs33949390 [29], CYP19-rs11538071 [30], and SERPINA3-rs11538071 [31]. The variant with the most evidence out of these variants is GRB2 Associated Binding Protein 2 (GAB2-rs4945261), but we are still uncertain of the association. rs4945261 is a marginally protective (P < 0.001) variant in European meta-analysis [32] and IGAP, but appears to be a risk variant in Chinese populations and is not genome-wide significant in GWAS.

Where the same SNP was reported in multiple studies, we chose to discuss the study with the largest sample size, leaving a total of 19 variants which we review below (Table 1).

Table 1:

Reported protective variants from candidate-gene studies for Alzheimer’s Disease

Study	Gene	CPRA	rsID	EA	NEA	OR (95% CI)	P	AF cases	AF Control	AF gnomAD
Candidate Loci
Bertram 2007	APOE	19:45408564:A:T	rs449647	T	A	0.72 (0.63, 0.82)	2.60E-07	0.14	0.19	0.20
Jiao 2015	TOMM40	19:45396144C:T	rs11556505	T	C	0.48 (0.34, 0.69)	< 0.001		0.14†	0.12
Lv 2008	APP	21:27543963:A:G	rs466433	G	A	0.58 (0.42, 0.79)	0.002	0.15	0.23	0.08
Lv 2008	APP	21:27544041:T:C	rs364048	C	T	0.54 (0.42, 0.81)	0.001	0.14	0.23	0.08
GWAS Loci
Wang 2015	NOCT	4:139930032:A:G	rs13116075	G	A	0.49	7.90E-06		0.15†	0.11
Kunkle 2018	SORL1	11:121435587:T:C	rs11218343	C	T	0.8 (0.75, 0.85)	2.90E-12		0.04†	0.06
Kunkle 2018	SLC24A4	14:92932828:T:C	rs12881735	C	T	0.92 (0.89, 0.94)	7.40E-09		0.21†	0.19
Kunkle 2018	CASS4	20:54997568:G:A	rs6024870	A	G	0.88 (0.85, 0.92)	3.50E-08		0.08†	0.07
Kunkle 2018	EPHA1	7:143107876:C:T	rs11762262	A	C	0.9 (0.88, 0.93)	1.30E-10		0.19†	0.20
Liu 2017	SPPL2A	15:50709337:T:G	rs59685680	G	T	0.92 (0.89, 0.95)	7.32E-10		0.19†	0.18
Rare Variant Analysis
Ridge 2017	RAB10	2:26358156:A:G	rs142787485	G	A	0.58	0.018	0.028	0.04	0.022
Jonsson 2012	APP	21:27269932:C:T	rs63750847	T	C	0.23	<0.001	0.13	0.62	0.0005
Sims 2017	PLCG2	16:81942028:C:T	rs72824905	G	C	0.68	5.38E−10	0.006	0.009	0.005
Ghani 2016	MS4A						0.047	0.043	0.082‡
Lupton 2014	ABCA1						0.032	0.13	0.19‡
Modifier Variants
Lalli 2015	CCL11	17:32669615:A:T	rs9909184	A	G	10*	4.85E-08		0.14†	0.16
Jun 2016	LRRC37A	17:46275856:T:G	rs2732703	G	T	0.73 (0.65, 0.81)	5.80E-10		0.13†	0.08
Jun 2017	NFIC	19:3405594:T:A	rs9749589	A	T	0.73 (0.66, 0.81)	1.50E-10			0.15

^*AAOS of MCI;

^†AF of whole cohort;

^‡frequency of rare variants in cases vs controls;

EA: Effect allele; NEA: Non-effect allele; AF: Allele frequency

APOE ε variants:

Apolipoprotein E (ApoE) is located on chromosome 19q13.32 and encodes a protein involved in lipid homeostasis and cholesterol metabolism. APOE plays a critical role in Aβ homeostasis by modulating both fibrillogenesis and clearance of Aβ. Furthermore, APOE promotes synaptic integrity and facilitates anti-oxidant and anti-inflammatory activity [33]. Mutations in APOE occurred in humans over time, each conferring additional protection against AD. The ancestral ε4 isoform is associated with AD risk compared to the other isoforms. The ε3 mutation from arginine to cysteine at residue 112 occurred 200,000 years ago and confers protection against cytotoxic domain interactions [34]. The ε2 mutation is a second substitution from cysteine to arginine at position 158 that occurred 60–80,000 years ago in ε3 carriers, enhancing APOE stability and resistance to denaturation [35]. Though these derived isoforms may have been selected for, it is unclear whether natural selection against AD, selection for pleiotropic traits or simply genetic drift are responsible.

The APOE ε2 allele is associated with decreased risk and delayed onset of AD, increased longevity, decreased AD-related neuropathology, reduced age-associated cognitive decline and greater cortical thickness (for a review see [33,36]). The ε2 allele is associated with slower Aβ aggregation and more efficient degradation and clearance [33]. However, while the APOE ε2 allele is protective against AD, ε2 carriers are still susceptible, with 0.2% of autopsy-confirmed AD cases homozygous for APOE ε2 [37]. APOE ε2 carriers who develop AD tend to display a milder, non-amnestic clinical phenotype with a higher burden of small vessel disease [38].

APOE promoter variants:

Two studies in our literature review reported evidence of a protective effect for a SNP, rs449647, located in the regulatory region of APOE. Bertram et al [5] conducted a meta-analysis of 36 studies combining 6,286 cases and 6,683 control of mixed ancestry and found that rs449647 was associated with 1.38-fold reduction in risk of AD. Bratosiwicz-Wasik et al 2018 [39] also observed that rs449647 was associated with reduced risk in a small Polish sample (n = 461). While rs449647 is not in strong linkage disequilibrium (LD) with the APOE ε defining SNPs, neither study adjusted for potential confounding effects due to the presence of APOE ε4 or ε2.

TOMM40:

Translocase of outer mitochondrial membrane 40 (TOMM40) is located on chromosome 19q13.32, approximately 15 kb upstream of APOE, and encodes a protein that is localized to the outer membrane of the mitochondria. It is an essential subunit of the Translocase of the Outer Membrane apparatus which is involved in importing protein precursors into the mitochondria [40]. 19q13 contains multiple genes and exhibits moderate to strong LD. As such, the presence of additional genetic variants in this region that exert effects beyond that of APOE has not been fully resolved [41]. Our literature search identified the SNP rs11556505-T as being associated with 2 times reduced risk of AD in a Han Chinese case-control study of 547 individuals [42]. However, in large European GWAS of AD risk, rs11556505-T is associated with increased risk of AD and an earlier age of onset [11,12,43]. Genetic variation in the TOMM40 region has been repeatedly associated with risk of AD. In particular, a highly polymorphic poly-T variant (rs10524523) in intron 6 has been variably associated with risk and protective effects that are likely dependent on the phenotype being evaluated, age, sex and APOE genotype of the subjects [41]. However, further validation of the effects of TOMM40 beyond that of APOE is needed.

SORL1:

Sortilin-Related Receptor 1 (SORL1, also known as SorLA and LR11) is a low-density lipoprotein (LDL) receptor and vacuolar protein sorting 10 (VPS10) domain receptor located on chromosome 11q24.1. In the brain, it is mainly expressed intracellularly [44–46]. In AD pathology, SORL1 binds directly to APOE [47], APP [45] and Aβ [47], and regulates their sorting and endocytosis. A SORL1 SNP (rs11218343) was first associated as a protective variant for AD at genome-wide significance in a meta-analysis between Japanese, Korean and European samples, conferring a 1.3 times reduced risk of AD [48]. It was replicated in Han Chinese [49] and is common in East Asians, with a MAF of 0.298, but is low frequency in Europeans (MAF=0.039). Both IGAP and Marioni et al [11,12,21] further replicate this association in Europeans. Although the common variant in SORL1 is protective, multiple rare variants increase risk for EOAD [50]. Of these EOAD variants, 3 were associated with LOAD in IGAP, including one weakly risk associated variant and two protective variants, and were also associated with SORL1 expression [51]. However, directionality of the eQTLs depended on brain region and SNP, so causality is unclear. These data suggest that higher SORL1 expression is protective against AD, but further validation is needed before the GWAS hit can be determined to be functionally protective.

APP:

Amyloid precursor protein (APP) is located on chromosome 21q21.3 and encodes a transmembrane protein implicated in neurite growth, neuronal adhesion, synaptogenesis and axon pruning, with concomitant involvement in learning and memory, as well as development [52]. APP is cleaved by the ɣ-secretase and β-secretase complexes into Amyloid-β (Aβ) peptides of various lengths and toxicities. Aβ can form toxic dimers or larger oligomers which can then form both extracellular diffuse and neuritic plaques that are one of the hallmarks of Alzheimer’s disease [53]. Fifty-one pathological APP mutations have been identified in 121 AD families [15]. A large association study in the Icelandic population identified a rare variant, rs63750847, that reduced AD risk by more than 4-fold [54]. rs63750847 results in an alanine to threonine substitution at position 2 in the Aβ domain (A673T), immediately downstream of the BACE1 cleavage site. This variant inhibits BACE1 cleavage resulting in decreased Aβ production and aggregation [55,56]. This variant has been observed in other nordic populations but is extremely rare or undetected in other populations [57].

In addition, a study conducted in a Han Chinese population (n = 646) reported two polymorphisms in the promoter region of APP (rs466433 and rs364048) that are associated with reduced risk [58]. These results, however, are not replicated in Caucasian GWAS at genome-wide significance, with the lowest p-value (p = 0.03) obtained in the meta-analysis of UKBB and IGAP.

RAB10:

Ras-Related GTP-Binding Protein 10 (RAB10) is located on chromosome 2p23.3 and plays a broad role across cell types in endocytosis and mediates trafficking from the trans-Golgi network to the plasma membrane [59]. In neurons, RAB10 also regulates vesicle trafficking to mediate neurite outgrowth and may also be involved in neurotransmitter release [59]. A novel linkage approach was used to identify resilience alleles in elderly cognitively normal APOE e4 carriers within densely affected AD families [60]. This study identified a linkage region on chromosome 2, with a candidate SNP, rs142787485, in the 3’ UTR of RAB10, associated with 1.7-fold reduced risk of AD, which was replicated in two independent cohorts. Inhibition of RAB10 expression resulted in decreased Aβ42 levels and Aβ42/Aβ40 ratio in cultured cell lines. RAB10 may affect APP processing via a direct interaction with APP. Other members of the RAB GTPase family mediate trafficking and sorting of APP and β- and ɣ-secretases into intracellular compartments [61].

NOCT:

Nocturnin (NOCT, also known as CCRN4L), located on chromosome 4q31.1, contains a CCR4-like deadenylase catalytic domain suggesting that NOCT may be involved in degradation of mRNAs [62,63]. There is limited data regarding NOCT function and regulation in humans, though mouse knockout studies indicate a role in lipid metabolism, adipogenesis and osteogenesis [62,63]. A GWAS of AD progression identified four suggestive loci (p < 1E-05), including rs13116075 upstream of NOCT, which was associated with a two-fold reduction in rate of progression [64]. rs13116075 was only marginally associated (p < 0.05) with AD risk [12] and not associated with age of onset [43]. Although NOCT is the closest gene there is no direct functional data linking this variant to NOCT.

SCL24A4-RIN3:

Solute Carrier Family 24 (Sodium/Potassium/Calcium Exchanger), Member 4 (SCL24A4) is abundantly expressed in the brain [65] and terminates calcium signaling in an ATP-dependent manner [66]. An SCL24A4 intronic variant (rs10498633) is associated with 1.1-fold reduced risk of AD [11,21]. rs10498633 is also near RIN3 (Ras and Rab Interactor 3), an endocytosis regulator that interacts with BIN1 [67] and is associated with risk of EOAD, suggesting that the functional AD SNP may lie within RIN3 [68].

CASS4:

Cas Scaffolding Protein Family Member 4 (CASS4), located on chromosome 20q13.31, encodes a poorly characterized scaffolding protein that has been implicated in focal adhesion, cellular adhesion, cell migration, and motility [69]. Three SNPs, rs7274581, rs6024870 and rs6069736, within the CASS4 locus have been associated with a reduced risk of AD in GWAS conducted in IGAP case-control studies and in a meta-analysis of IGAP and Proxy case-controls in the UK Biobank [11,12,21]. Furthermore, CASS4 variants have also been associated with slower progression in AD patients, reduced NFT burden and reduced neuritic plaque burden and longitudinal change in amyloid burden [64,70,71]. In a Drosophila model, upregulation of the CASS4 Drosophila ortholog, p130CAS, enhanced Tau toxicity, implicating CASS4 as a Tau toxicity modulator [72].

EPHA1:

EPH receptor A1 (EPHA1) is located on chromosome 7q34 and belongs to the ephrin family of tyrosine kinase receptors. EPHA1 binds to membrane-bound ephrin-A ligands on adjacent cells, allowing contact-dependent bidirectional signaling between adjunct cells and as such plays a role in synaptic formation and plasticity, axonal guidance, and brain development [73–76]. The SNP rs11762262 was associated with a 1.11-fold reduced risk of AD in a case-control cohort of 89,769 European participants [21]. Other variants in the EPHA1 locus that are in high LD with rs11762262 have been associated with reduced brain amyloidosis [77], hippocampal atrophy and greater cerebral metabolic rate in MCI [78] and regulate expression of EPHA1 in human whole blood but not in brain tissues [79].

SPPL2A:

An intronic variant (rs59685680) in Signal Peptide Peptidase Like 2A (SPPL2A) was associated with 1.08-fold reduced risk of AD in a meta-analysis of 114,564 proxy-cases and controls from the UK Biobank and IGAP [22]. SPPL2A is located on chromosome 15q21.2 and is a presenilin homolog that encodes a protease that localizes to the late endosome/lysosome and regulates B cell homeostasis [80]. The role of this locus in Alzheimer’s disease is currently unknown.

PLCG2:

1-Phosphatidylinositol-4,5-bisphosphate phosphodiesterase gamma-2 (PLCG2) is a transmembrane signaling enzyme that cleaves the membrane phospholipid PIP2 (1-phosphatidyl- 1d-myoinositol 4,5-bisphosphate) to secondary messengers 1D-myoinositol 1,4,5-trisphosphate (IP3) and diacylglycerol (DAG) using calcium as a cofactor. In brain tissue, PLCG2 is primarily expressed in microglia and secondarily in the granule cells in the dentate gyrus [81]. PLCG2 is located on chromosome 16q23.3 and contains a rare coding variant (rs72824905) that is associated with 1.5-fold reduced risk of AD in a three-stage case-control study of 83,133 subjects of European ancestry [82]. This variant is located in a regulatory domain and confers a small hypermorphic effect on enzyme activity, suggesting that a weak lifelong activation of the PLCG2 pathway may confer protection against AD [81]. The full mechanism through which this occurs is currently unknown, however it could interact through Ca+ signaling with multiple other AD risk genes [83].

MS4A:

The Membrane-Spanning 4 Domain Subfamily A (MS4A) gene cluster on chromosome 11q12.2 includes 18 genes spanning approximately 600kb that have similar polypeptide sequences and topographical structures within cells [84]. The exact function of these proteins is not well characterized, however, they have been implicated in the regulation of calcium homeostasis, immune response, cell activation, growth and development [85]. MS4A proteins are predominantly expressed in cells of the monocyte/microglia lineage indicating that they likely have a role in the immune system [86]. Next-generation sequencing mutation analysis of the MS4A gene cluster in 210 AD cases and 233 controls has found potentially damaging missense substitutions and loss-of-function variants which were twice as frequent in controls compared to cases [87]. It is possible that rare variants in the MS4A cluster could contribute to or explain the protective effects of minor alleles for GWAS significant SNPs [11,12,21]. Elevated expression of MS4A6A is associated increased risk of AD and with more advanced Braak tangle and plaque scores in AD brains [88,89], suggesting that rare variants that disrupt MS4A6A function may reduce MS4A6A expression and result in its observed protective effects. Additionally a common variant in the MS4A cluster, rs7930318, was associated with delayed AAO and with lower expression of MS4A4A and MS4A6A [43].

ABCA1:

ATP-binding cassette A1 (ABCA1) - located on 9q31.1 - is responsible for lipidating apolipoproteins, including APOE, and is necessary for the formation of high-density lipoproteins (HDL) [90]. A single mutation on ABCA1 is responsible for both autosomal dominant and recessive diseases relating to HDL deficiencies [91]. ABCA1 deficiency strongly reduces the release of both lipidated APOE and APOJ from both microglia and astrocytes in mouse models [90], reducing Aβ aggregation in an APOE dependent manner [92]. Microglial expression of ABCA1 via LXR activation also reduces inflammation and cell death and increases phagocytosis in mouse models [93]. In AD, evidence for ABCA1 has been mixed. GWAS have not found significant associations for ABCA1, nor have several candidate genes association studies [94–98]. However, some association studies have found risk [99–103] and protective [99,104,105] variants or haplotypes. Variants are also associated with AD biomarkers [99,102]. A rare variant candidate gene study found that 1.7 times as many controls had at least one rare variant (MAF < 0.05) in ABCA1 at 19% of controls vs 13% of AD cases [106]. Across sources, higher ABCA1 function appears to be protective. A small-molecule drug has been developed that specifically induces ABCA1 through the LXR pathway and may be a treatment for AD [107] and other drugs are being developed that target the pathway [108–116].

Genetic modifiers

CCL11:

C-C motif chemokine ligand 11 (CCL11) located on chromosome 17q12 is one of several chemokine genes clustered in this genomic locus that are involved in immunoregulatory and inflammatory processes. CCL11 plays a role in allergic conditions and inflammatory diseases of the gastrointestinal tract [117] and CCL11 levels increase with aging and are correlated with reduced neurogenesis [118]. Lalli et al [119] conducted a GWAS using whole genome-sequencing data from 72 individuals with early-onset familial Alzheimer’s disease due to the PSEN1 E280A mutation to identify variants that modified age of onset of mild cognitive impairment. Polymorphisms in the CCL gene cluster were associated with age of onset, with rs9909184-A delaying age at onset of MCI by 10 years. A missense variant, rs1129844, located in the signal peptide cleavage site of CCL11 is in high LD with the lead SNP and enhances secretion of CCL11 in HEK-293T cells. In a second cohort of LOAD cases, increased CCL11 plasma levels were associated with an increased AAO, however, in non-carriers CCL11 levels were not observed to increase with age [119]. This suggests that lower levels of CCL11 are associated with neuroprotective response, however, higher levels promote neurodegeneration and memory impairment, potentially by suppressing neurogenesis [120].

KANSL1-MAPT:

Jun et al [23] conducted a stratified GWAS in APOE ε4+ (10 352 cases and 9207 controls) and APOE ε4-(7184 cases and 26 968 controls) subgroups to identify novel loci whose effects may be masked by APOE. An intronic SNP, rs2732703, in LRRC37A on chromosome 17q21.31 was found to be associated with 1.3-fold reduced risk of AD in APOE ε4-participants. Additionally, eQTL analysis indicated that rs113986870, which was also genome-wide significant and in high LD with rs2732703, is an eQTL for expression of the first translated exon of KANSL1 and third exon of MAPT. KANSL1 encodes a nuclear protein that is subunit of the KAT8 regulatory NSL complex which is involved in histone acetylation, with mutations associated with developmental delay and intellectual disability [121]. MAPT encodes the tau protein, with the accumulation of hyperphosphorylated tau into Neurofibrillary tangles one of the pathological hallmarks of AD. Increased expression of MAPT exon 3 has been associated with a protective effect against neurodegeneration [122].

NFIC:

Nuclear Factor I-C (NFIC) is located on chromosome 19p13.3 and is a transcription factor with unknown function in AD. It is most associated with tooth root development, regulated by sonic hedgehog and TGF-β [123]. TGF-β is a key signaling molecule in innate immunity and has been linked to APOE pathogenesis of AD [124,125]. NFIC is also linked to TGF-β in wound healing and liver regeneration, acting primarily as a repressor of TGF-β activated gene expression. Unlike other NFI genes, NFIC does not appear to be necessary for proper brain development [126]. An NFIC variant - rs9749589 - was associated with a 1.36 times reduced risk of AD, against the APOE-ε4 allele in a transethnic GWAS in European Americans, African Americans and Japanese [24]. However, it was a slight risk allele in ε4 negative subjects and did not replicate in stage 2 of ADGC and in IGAP. This interesting modifier gene requires further study to understand the biological effects of the variant.

Conclusions

Known Protective Variants

Very few protective variants in the AD literature have strong evidence of association, are the derived allele and have a frequency below 20 percent. Almost all of the variants with strong evidence of a protective effect are either GWAS significant or are from candidate gene studies based on GWAS results. Beyond that, even fewer have strong evidence that they modify the expression or function of a particular gene. This is important because without strong functional evidence, putatively protective variants may just tag other variants that could be risk associated. We suggest that future association studies are performed in an unbiased, genome-wide manner or attempt to fine-map already identified loci in order not to exclude protective variation, and that all known associations are fine-mapped and combined with gene expression data to transition from tagging to true protective and risk variants. Furthermore, the interactions between known associated loci should be interrogated to find protective and synergistically risky interactions between variants. With this approach druggable targets and appropriate prognostic criteria can be developed to improve the prognosis for AD.

Of the variants we looked at in detail, only 5 had strong evidence of a functional protective effect, instead of tagging variation where it is unclear what variant is actually responsible. APOE ε2 has long been associated with a reduction in AD risk, is directly characterized in its effect on APOE function, and has a direct protein consequence. Similarly, APP and PLCG2 associations are coding variants with various levels of functional characterization. The MS4A gene cluster and ABCA1 show a higher burden of rare coding and missense variation in controls than in AD cases. Furthermore, common variation in the MS4A locus is protective and affects expression. An additional three variants had moderate evidence of functional protective effect. The RAB10 variant is located in the 3’UTR and is likely a regulatory variant, and gene expression is associated with AD endophenotype. LRRC37A and CCL11 variants were unlikely to be disease-causing but tagged SNPs that were an eQTL and missense coding variant respectively.

The remaining variants we explored are noncoding and have no evidence of a direct effect on expression of their associated genes. There are various levels of evidence that their putative genes are causal in AD. Expression of both SORL1 and CASS4 are directly associated with AD pathology. There is less evidence regarding any possible causal role for SCL24A4 or RIN3, NOCT, EPHA1, and SPPL2A in AD pathogenesis.

Due to our definition of a protective variant - a derived allele with a frequency of less 20% - our putative protective variant list should not be considered exhaustive. A more liberal MAF cut-off would identify more protective variants (see [127]). For example the GWAS AD associated locus at 7q21 [11,21,128] has been linked to a common missense (MAF = 0.35) variant in PILRA that likely protects against AD via reduced inhibitory signaling in microglia [129]. Similarly, variants in SPI1 are associated with decreased AD risk (rs3740688 [21]) and delayed AAO (rs1057233 [43]), with rs1057233 also associated SPI1 expression levels [43], likely by regulating other AD associated genes expressed in microglia and other myeloid cells [43]. However, both variants are common, with an allele frequency above 30% and the effect allele is the ancestral allele. In contrast, by using a functional definition where the protective variant results in a loss- or gain-of-function only APP and PLCG2 would be considered protective variants [20]. Furthermore, we have only reviewed SNPs and have not considered how structural variation, such as copy number variants, may affect AD risk [130]. Finally, while our literature search did identify a large number of studies conducted in non-European populations, the majority had a small sample size and were underpowered. As such the loci we report here are predominantly the result of association studies conducted in European populations. It is likely that rare protective variants are ancestry-specific and that further evaluation of AD loci in non-European populations could identify novel loci.

Finding and Validating Protective Variants

For initial screens to find protective variants, we recommend GWAS or whole genome rare variant burden/kernel methods. The majority of our identified variants do not tag genes previously implicated in AD pathogenesis, so unbiased methods are most likely to find further protective variation. Beyond that, specialized designs such as AAOS association and APOE stratified analyses may be better powered and designed to find protective variation, and may yield variation that is more definitively protective. As the cost of AD pathological characterization for brain Aβ and phospho-tau becomes more reasonable, it may also make sense to perform large studies to find variants conferring resilience to AD pathology, which may show mechanistic differences from clinical datasets where resilience to AD pathology and delayed pathology both manifest as late age at onset or lower risk for disease. It may also be worthwhile to design studies to find modifiers of other know AD risk factors, particularly in LOAD.

Once protective loci have been found, it is also important to perform fine-mapping and functional assays with expression, chromatin, and protein quantitative trait loci (eQTLs, chQTLs, and pQTLs) within appropriate cell types in order to validate variants as risk or protective instead of just tagging susceptibility.

Concluding statements

In our review we discussed protective variants with a frequency below 20% and a non-ancestral minor allele associated with better AD-related outcomes. These and other more common protective variants may aid in understanding AD pathogenesis, improving Alzheimer’s disease prediction in the context of precision medicine, and facilitating drug development [19,20], but more fine-mapping is required to validate these signals.

Table 2:

Study characteristics

			N
Study	Method	Phenotype	Total	Cases	Controls	Population
Bertram 2007	Candidate Loci - Meta-analysis	AD	12969	6286	6683	EUR + EAS
Jiang 2017	Candidate Gene - Meta-analysis	AD	11478	3621	7857	EUR + EAS
Jiao 2015	Candidate Gene	AD	547	229	318	EAS
Lv 2008	Candidate Gene	AD	646	209	437	EAS
Lv 2008	Candidate Gene	AD	646	209	437	EAS
Wang 2015a	GWAS	AD Progression	680	307	373	EUR
Kunkle 2018	GWAS	AD	82771	21982	41944	EUR
Kunkle 2018	GWAS	AD	82771	21982	41944	EUR
Kunkle 2018	GWAS	AD	82771	21982	41944	EUR
Kunkle 2018	GWAS	AD	82771	21982	41944	EUR
Liu 2017	GWAX	AD	114564	14482	100082	EUR
Jun 2017	GWAS	AD	33269			EUR + AFR + EAS
Ridge 2017	Linkage	AD Resilience	4083	523	3560	EUR
Jonsson 2012	Rare Variant Screen	AD	1795			EUR
Sims 2017	Rare Variant Screen	AD	84905	48402	37022	EUR
Ghani 2016	Rare Variant Screen	AD	443	210	233	EUR
Lupton 2014	Rare Variant Screen	AD	671	311	360	EUR
Lalli 2015	GWAS	AAOS	72	72	0	AMR
Jun 2016	GWAS	AD	34152	7184	26968	EUR

1000 Genomes super-populations: EUR: European; EAS: East Asian; AFR: African; AMR: Admixed American