PILRA polymorphism modifies the effect of APOE4 and GM17 on Alzheimer’s disease risk

PILRA (rs1859788 A > G) has been suggested to be a protective variant for Alzheimer’s disease (AD) and is an entry co-receptor for herpes simplex virus-1. We conducted a nested case–control study of 360 1:1-matched AD subjects. Interactions between the PILRA-A allele, APOE risk variants (ε3/ε4 or ε4/ε4) and GM17 for AD risk were modelled. The associations were cross-validated using two independent whole-genome sequencing datasets. We found negative interactions between PILRA-A and GM17 (OR 0.72, 95% CI 0.52–1.00) and between PILRA-A and APOE risk variants (OR 0.56, 95% CI 0.32–0.98) in the discovery dataset. In the replication cohort, a joint effect of PILRA and PILRA × GM 17/17 was observed for the risk of developing AD (p .02). Here, we report a negative effect modification by PILRA on APOE and GM17 high-risk variants for future AD risk in two independent datasets. This highlights the complex genetics of AD.


Results
The descriptive statistics of the 360 AD cases and 360 matched controls from the discovery dataset (i.e. NSHDS) are presented in Table 1. The mean time to event was 9.6 ± 4.1 years (i.e. time between blood collection and AD diagnosis). The mean age of AD diagnosis was 70.8 ± 6.4 years.
The PILRA R78-A allele was not associated with AD in the discovery dataset (crude Odds ratio (OR) 0.94, 95% confidence interval (CI) 0.74-1.21, p = 0.656; Table 2). The interactions terms were modelled using conditional logistic regression and additive coding for PILRA R78G-A and GM17 (see Methods). We found negative interactions between PILRA R78G-A x GM17 and PILRA R78G-A x APOE risk variants (ε3/ε4 or ε4/ε4) for the risk of AD (OR for the interaction 0.72, 95% CI 0.52-1.00 and 0.56, 95% CI 0.32-0.98 respectively; Table 3). The interaction term of PILRA R78G-A x anti-HSV-1 IgG seropositivity was not significant (Table 3). These interaction effects are also visualized in Fig. 1A-C where PILRA R78G is plotted against APOE, GM genotypes, and anti-HSV-1 IgG in separate groups. Table 4 shows the descriptive statistics of subjects with different PILRA R78G genotypes among cases and controls separately. The distribution of PILRA R78G genotype in cases and controls, stratified by APOE, GM17, and anti-HSV-1 IgG status is also presented in Fig. 1A-C. Controls with APOE risk variants, the GM17 allele and anti-HSV-1 IgG antibodies all seemed to have higher frequencies of PILRA A/A genotype compared to their cases ( Fig. 1A-C). In contrast, subjects (cases and controls combined) carrying both PILRA R78G A/A and APOE risk variants had lower frequencies of detectable anti-HSV IgM antibodies compared to subjects with APOE risk variants and non-PILRA R78G A/A genotypes (Fig. 1D).
Next, we sought to assess the main or interaction effects of PILRA R78G in two AD whole-genome sequencing datasets with different study designs: a large family-based AD sample from NIMH and an AD case-control  www.nature.com/scientificreports/  www.nature.com/scientificreports/ dataset from NIA ADSP ( Table 5). The case-control sample from the NIA ADSP contained three subcohorts: a Non-Hispanic White cohort, an African-American cohort and a Hispanic cohort. Using transmission family-based approaches, we saw an association of AD risk with PILRA R78G (p = 0.0495) and APOE rs429358 (ε4, p = 1.78 × 10 −15 ) and rs7412 (ε2, p = 5.01 × 10 −5 ) SNPs, but not with GM17 (rs1071803, p = 0.9). This method is used to evaluate both linkage and association with the phenotype of interest in family pedigrees. When including one of the following interaction terms: PILRA R78G × APOE risk variants or PILRA R78G × GM 17/17, we found that the family-based joint test for the main effect PILRA G78R and the interaction effect PILRA R78G × GM 17/17 was significant (p = 0.02, Table 6). However, none of the interaction terms in each of the two models was significant. Finally, in the non-Hispanic white subpopulation of the NIA ADSP dataset (n = 1669), PILRA R78G was not associated with AD (p = 0.94). The variant rs1071803, which codes for GM17, was missing in NIA ADSP and the interaction term PILRA R78G × APOE risk variants were not significant (p = 0.66 using additive coding and p = 0.27 using recessive coding).   www.nature.com/scientificreports/

Discussion
The key finding of our study is that the PILRA R78G-A allele negatively modifies the effect of APOE and GM17 high-risk variants on AD risk (OR for the GM17 interaction 0.72, 95% CI 0.52-1.00 and OR for the APOE interaction 0.56, 95% CI 0.31-0.98; Table 3 in the discovery cohort). The effect modification seems to be of increased strength in APOEε4 and GM17 homozygotes (Fig. 1A, B), revealing a potential dose-dependent pattern. Similarly, we found a significant joint effect of PILRA R78G and PILRA R78G × GM 17/17 for AD in the replication cohort. While having the PILRA R78G-A allele was associated with reduced risk of AD in the family cohort, this association was not replicated in the other two samples. Previous epidemiological studies have shown that HSV-1 is associated with increased AD risk in genetically predisposed individuals carrying the APOEε4 allele or other AD risk genes [7][8][9][10]23 . The finding that the PILRA R78G-A allele might modify the risk of AD in APOEε4 and GM17 carriers (Table 3) might further enhance our understanding of the complex gene-gene and gene-environment interactions for HSV1-associated AD risk.
PILRA R78G has previously been linked to both HSV-1 and AD 11,15 . The A allele of PILRA R78G causes a conformational change in its sialic binding pocket, which leads to impaired binding capacity for HSV-1 and other ligands 11 . This could make target cells less susceptible to HSV-1 infection through reduced HSV-1 cell fusion, and limit viral entry into neurons in the brain, thus offering some protection against HSV-1-associated AD. The effect of PILRA could also possibly be explained by fewer latently infected neurons in the periphery, which correlate with lower reactivation rates of HSV-1 34 . Importantly, PILRA also function as an inhibitory regulator of microglia activation 35 , and reduced PILRA signaling in R78G-A allelic variants could result in the enhancement of microglial activity 11 . It is therefore possible that the decrease in AD risk associated with having the PILRA R78G-A allele might be attributed to more properly regulated microglia and possibly improved amyloid-β clearance 36 . However, the exact role of microglia in AD initiation and progression remains to be fully elucidated, and it might vary during the course of the disease.
In the discovery cohort, we observed a potential modifying effect of PILRA R78G A/A on the risk of having anti-HSV IgM antibodies (a marker of recent HSV reactivation) among carriers of APOE risk variants (Fig. 1D). Notably, we have previously shown that having APOE risk variants were associated with a higher prevalence of anti-HSV IgM antibodies in the NSHDS sample 6 , thus an association that seems to be negatively modified by PILRA. Figure 1C illustrates that PILRA R78G A/A homozygosity also could have a protective impact on the HSV-1 associated AD risk, although not statistically significant (Table 3). Herein, HSV-1 seropositive controls had a higher frequency of PILRA R78G A/A genotypes compared to HSV-1 seronegative controls.
The primary strength of this study is that controls, sampled from the same population, were closely matched on possible confounding and demographic variables. Another major strength is the prospective design, where blood specimens were obtained several years prior to the disease onset, making it possible to estimate future disease risk. Limitations include the observational nature of our study, as potential unaccounted confounding factors could influence the associations and that the AD diagnoses were clinical and not based on evidence of amyloid deposition or pathologic tau. A further limitation noticed was that only 5.3% of AD cases and 7.3% of controls were PILRA R78G A/A homozygotes ( Table 1), suggesting that this genotype is not common in the studied population. The allele frequency of PILRA rs1859788 seems to vary globally, and is higher in the East Asian population 37 . This variation in allele frequency could possibly explain the lack of association between AD and PILRA R78G in the NSHDS and NIA ADSP material, which was indicated by another study 11 and the family-based NIMH dataset.

Conclusion
Here, we report a negative effect modification by the PILRA R78G-A allele on APOE and GM17 risk variants for future AD risk in two independent datasets. This observation might provide further insight into the complex genetics of HSV1-associated AD.

Study design. Discovery dataset NSHDS.
We used a nested case-control study design, where 360 subjects clinically diagnosed with AD were identified from the population-based Northern Sweden Health and Disease study (NSHDS) 38 . The NSHDS consists of three subcohorts: the Västerbotten Intervention Programme (VIP), the Mammography Screening Project (MA), and The Northern Sweden Monica Project (MO). Blood samples were previously drawn and stored in the Medical Biobank in Umeå, extracted for analysis on average 9.6 years before the AD diagnosis. Controls without neurodegenerative disorders were randomly selected from the NSH-DS cohort and matched 1:1 by age, sampling dates, sex, and subcohort. The diagnostic procedure and selection of subjects have been described in a previous publication 25 .
NIMH family-based dataset and ADSP case-control dataset. The results were cross-validated using two independent whole-genome sequencing datasets, a family-based AD cohort from NIMH and an AD case-control sample from the NIA (ADSP).
Genotyping in NSHDS. Samples were genotyped for APOE (rs429358 and rs7412) and PILRA R78G (rs1859788) using Illumina genome-wide array Human-OmniExpress24 (deCODE genetics, Reykjavik, Iceland) 9 . QPCR-based genotyping assays 11,39 were employed for confirmation of inconclusive sequences. A custom design TaqMan genotyping assay was employed for genotyping of the GM3 and17 alleles (i.e. to determine GM 3/3, GM 3/17 and GM 17/17 genotypes) 6  www.nature.com/scientificreports/ WGS analysis in NIMH and ADSP. Whole genome sequencing in the National Institute of Mental Health (NIMH) AD cohort and AD diagnoses are described elsewhere 40,41 . Variant calls in vcf format for the National Institute of Aging's (NIA) Alzheimer's disease sequencing project (ADSP) cohort were obtained from the National Institute on Aging Genetics of Alzheimer's Disease Data Storage Site (NIAGADS) under accession number: NG00067. The NIA ADSP dataset was divided into three subcohorts: Non-Hispanic White, African-American and Hispanic based on derived principal components. In order to derive more recent admixture principal components were calculated based on 100,000 rare variants using a modified genetic relationship matrix based on the Jaccard index 42 . Outliers based on principal components were excluded.
Serology-NSHDS. Enzyme-linked immunosorbent assays were used for the detection of anti-HSV IgG, anti-HSV-1 IgG, and anti-HSV IgM as previously described 25 .
The APOE variable was dichotomized as having high risk variants (ε3/ε4 or ε4/ε4) compared to ε3/ε4 and ε4/ε4 non-carriers. The rationale for dichotomizing APOE is that the effect of APOEε4 on AD risk is not additive, and the APOE locus is not bi-allelic.
APOE, GM, PILRA R78G, HSV-1, and the risk of AD in NSHDS. Associations between the risk of AD and the PILRA R78G-A allele were assessed by conditional logistic regression models. Interaction models were fitted for PILRA R78G-A and AD with interaction terms for PILRA R78G-A x APOE risk variants, PILRA R78G-A x GM17 and PILRA R78G-A x anti-HSV-1 IgG seropositivity. Each interaction term was modeled separately to estimate the effect modification by the PILRA R78G-A allele on AD risk per these factors. The gene variables contained missing data ranging from n = 3 to 10 (APOE: n = 3 cases and n = 7 controls, PILRA R78G: n = 8 cases and n = 6 controls, GM: n = 10 cases and n = 8 controls). Subjects with missing values were omitted from the statistical analyses. This strategy was chosen since data can be assumed to be missing completely at random due to their blood samples containing insufficient amounts of DNA.
Statistical analyses were performed using R version 4.1.3. A two-tailed p-value < 0.05 was considered significant. The codes are available as supplementary files (Supplementary file 1: discovery cohort and Supplementary file 2: replication cohorts).
APOE, GM, PILRA R78G, and the risk of AD in NIMH and ADSP. PLINK2 43 (www. cog-genom ics. org/ plink/2. 0/) was used to pre-process and extract variants of interest. In the NIMH cohort, we used a robust gene-byenvironment test 44 , which is based on the family-based association test (FBAT) 45 , a generalization of the transmission disequilibrium test. We used the function "fbatge" from the "fbati" package in R. In the case-control cohort, we used PLINK2 and R to perform logistic regression with covariates (Age, Sex, Sequencing center, and first 5 principal components to adjust for the population structure) and the corresponding interaction term. If not mentioned otherwise, we considered an additive model for PILRA G78R and considered the following interaction terms: PILRA R78G A/A x APOE risk variants, PILRA R78G A/A x GM 17/17, PILRA R78G A/A x APOE, or GM 17 risk variants. Information on anti-HSV-1 IgG seropositivity was not available in WGS cohorts.
Ethical approval. The study was performed in accordance with the Declaration of Helsinki and was approved by the Regional Ethical Review Board in Umeå, Sweden (diary no. 09-190 M and 2017/18-31). All participants provided informed consent for long-term storage of blood specimens and for research on the stored samples.

Data availability
Discovery cohort, NSHDS: The dataset generated and analyzed during the current study is uploaded as a supplementary file. Additional information is available from the authors upon reasonable request, and after review and with permission from The Biobank Research Unit at Umeå University. Replication cohorts: The NIMH WGS dataset analyzed during the current study was funded by Cure Alzheimer's Fund, a non-profit organization, and is available from the authors on reasonable request. The NIA ADSP WGS dataset is available from DSS NIAGADS (https:// dss. niaga ds. org/) under accession number: NG00067. Data used in preparation of this article were in part obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http:// adni. loni. usc. edu/ wp-conte nt/ uploa ds/ how_ to_ apply/ ADNI_ Ackno wledg ement_ List. pdf. www.nature.com/scientificreports/