Frequency of Variants in Mendelian Alzheimer’s Disease Genes within the Alzheimer’s Disease Sequencing Project (ADSP)

INTRODUCTION

Mendelian Alzheimer’s disease (AD) makes up less than 1% of all AD cases and is characterized by an early age of onset (<65 years old).¹ Rare mutations in the PSEN1, PSEN2, and APP genes have been previously characterized to show Mendelian AD inheritance patterns within families.¹ Identification and classification of variants in these three genes has aided both in the molecular classification of pathways involved in AD pathogenesis and screening for known AD causing variants within families. Variants within these three genes show an autosomal dominant inheritance pattern with near complete penetrance.¹ Publicly available clinical databases such as ClinVar, Online Mendelian Inheritance of Man (OMIM), and Alzforum have compiled clinically-relevant variants presumed to cause AD and other types of dementia (“clinical variants”). These databases are important resources but also have limitations. For example, variants identified in the 1990s may have only had few controls to determine their clinical relevance. Additionally, variants shown to be Mendelian with assumed near complete penetrance may also be found in late onset familial and sporadic AD as part of more comprehensive screening strategies that have less biased ascertainment than the highly-selected, large pedigrees used originally to identify these genes leading to conflicting interpretations of clinical variants.^2,3

The Alzheimer’s Disease Sequencing Project (ADSP) seeks to identify novel genetic risk factors for AD. The data collected and generated through the ADSP consist of whole genome sequence (WGS) and whole exome sequence (WES) data from family, case-control, and cohort study designs.⁴ Within the ADSP, standardized variant calling and data management pipelines (VCPA), as well as an ADSP Quality Control (QC) protocol have been implemented.^5,6 Bringing together high-quality sequence data from across the AD research community has allowed for an increased sample size, overall increasing statistical power and the ability to search for rare variation associated with AD. Given that ADSP data are biased towards the selection of variants of incomplete penetrance, we aimed to leverage these data to further characterize previously-reported AD clinical variants.

Previously, two studies using a prior release of ADSP data examined variants within APP, PSEN1, PSEN2, and other dementia related genes.^7,8 With a newer release and doubled sample size, we seek to examine the frequency of previously-reported clinically-relevant and conflicting clinically-relevant variants as well as predicted loss-of-function (LoF) and damaging missense (DM) variants in these three Mendelian AD genes within the ADSP WES and WGS datasets. We report on the distribution of these variants by racial and ethnic group to clarify the clinical implications of these previously-reported variants in the context of a history of disparity in representation in most studies of AD. Knowing the distribution of these variants in the ADSP datasets, which are publicly accessible to qualified researchers, may also inform studies focused on gene and variant identification at other loci in the context of AD risk.

METHODS

Statistical analysis

The ADSP phenotype and genotype data were used to identify individuals carrying at least one alternate allele for a clinical variant or a LoF/DM variant and generate descriptive statistics based on carrier status and AD diagnosis. In order to examine the distribution of genetic variants across different populations, we defined population groups based on reported race and ethnicity. We compared the mean age of onset of AD by carrier status using a linear mixed regression model with carrier status as the exposure and age of onset as the outcome, adjusting for sex and the first four principal components of ancestry, accounting for an empirical Balding-Nichols kinship matrix. All statistical analyses were performed using R 4.2.1. A schematic of the analysis is shown in Figure 1.

• Download figure
• Open in new tab

Figure 1. Flow Chart of the Analysis.

Figure 1 represents the schematic of the analysis. ADSP whole genome sequencing data and whole exome sequencing data were combined, and variants with MAF<1% in PSEN2, PENS1, and APP genes were extracted. Subsequently, VEP and dbNFSP annotation was conducted to identify the additional loss-of-function (LoF) and damaging missense (DM) variants in the ADSP participants, followed by the curation of clinical variants from ClinVar, OMIM, and Alzforum databases. We then reported demographic and AD status of the carriers of these variants.

RESULTS

Aggregating variants in ClinVar, OMIM, and Alzforum that met our criteria yielded a list of 293 clinical variants and 51 conflicting clinical variants, totaling 344 variants: 29 in PSEN2, 251 in PSEN1, and 64 in APP (Supplementary Table e-2). After removing genetically identical individuals, we had a total of 31,490 individuals consisting of 13,825 AD cases, 14,715 controls, and 2,950 with unknown AD status (Table 1). We identified 78 individuals that carried at least one clinical variant, 550 individuals that carried at least one conflicting clinical variant, 12 individuals that carried at least one additional LoF variant, and 843 individuals that carried at least one additional DM variant (Table 1).

Among the 293 variants from the aggregated list of clinical variants, 31 had alternate alleles present within the ADSP data: 4 in PSEN2, 25 in PSEN1, and 2 in APP (Table 2). All 31 variants seen within the ADSP participants are coding variants and were rare within the ADSP (MAF <1%). Overall, 78 ADSP participants (0.3%) carry at least one alternate allele of the variants curated from the clinical databases (Table 2). The 78 clinical variant carriers consisted of 5 LoF variant carriers and 73 DM variant carriers. PSEN1 variant carriers constituted the greatest proportion (0.22%) of the clinical variant carriers, followed by PSEN2 variant carriers (0.02%) and APP variant carriers (0.01%). One pair of related individuals have a PSEN1 (rs63750082, p.Gly206Ala) variant, and we inferred that they are siblings based on kinship coefficient (0.24) and age (57 and 64 years old, respectively) and are a part of the ADSP Family study. Within the group of 78 individuals carrying at least one clinical variant, as expected, most were AD cases (N=62, 79%) or those with unknown AD status (N=13, 17%); very few were controls (N=3, 4%). Non-Hispanic White participants have the highest proportion of clinical variant carriers (59%) and are the largest proportion of the data (67%), whereas Non-Hispanic Blacks constitute 16.8% of the data, but are only 3.9% of the clinical variant carriers (Table 1).

The higher number of clinical variant carriers in cases compared to controls was consistent across the three genes. We observed the highest proportion of clinical variant carriers among cases for PSEN1, with 0.4% of the AD cases carrying a clinical variant as compared to 0.01% of the controls (Table 3). Among PSEN2 carriers, 0.03% of those who were diagnosed with AD carry an alternate allele in PSEN2 variants compared to 0.01% of controls. And APP carrier rate is 0.01% in the AD cases as compared to 0% in the controls. Noteworthy, the higher proportion of PSEN1 carriers among the cases are mainly due to two variants: rs63749824 (p.Ala79Val) and rs63750082 (p.Gly206Ala) (Table 4). When restricting controls to more advanced ages (>65 or > 85), we observed similar carrier rates between cases and elder controls for clinical variant carriers of all 3 genes (Table 3).

We observed 38 out of 51 conflicting clinical variants, which consist of 13 PSEN2 variants, 9 PSEN1 variants, and 16 APP variants among the ADSP participants (Table 2). There were 550 individuals (1.75%) carrying at least one alternate allele of a conflicting clinical variant within the ADSP datasets, of which 240 are PSEN2 variant carriers (0.76%), 88 are PSEN1 variant carriers (0.28%), and 222 are APP variant carriers (0.70%) (Table 2). There were 57 related individuals among the conflicting clinical variant carriers. Among the conflicting clinical variant carriers, we observed a higher percentage of controls (N=270, 49.09%) than cases (N=224, 40.73%). Non-Hispanic White individuals (65.45%) constituted the largest proportion of the conflicting clinical variant carriers, followed by Non-Hispanic Black (17.27%), Non-Hispanic Other/Unknown (0%), Hispanic White (1.09%), Hispanic Black (0.18%), and Hispanic Other/Unknown individuals (16%) (Table 1).

We observed that the rates of conflicting clinical variant carriers are higher in controls for PSEN2 and APP variants, with 0.80% and 0.76% of the controls carrying at least one conflicting clinical variant as compared to 0.69% and 0.64% of the cases, respectively. The rates for PSEN1 conflicting clinical variant carriers were 0.27% of the controls were PSEN1 variant carriers in contrast to 0.29% of the cases. And the rates of conflicting variant carriers increased when looking at elder controls except for PSEN1 variant carriers (Table 3). We observed 23 variants that were seen more often in controls rather than cases: 11 in PSEN2, 5 in PSEN1, and 7 in APP, of which 5 variants were only seen in controls (rs201269325, p.Gly709Ser; rs63750831 p.Val94Met; rs115760359, p. Pro218Pro; rs63750227 p.Ala409Thr; rs866044092 p.Val150Met) (Supplementary Table e-3). Out of the 38 conflicting clinical variants, 27 of the conflicting clinical variants were all observed in both cases and controls, diminishing support for these variants (Supplement Table e-3).

Using VEP to annotate rare variants (MAF <1%) observed in the ADSP participants, we identified 12 additional LoF variants in the 3 genes that had not been previously identified as a clinical or conflicting clinical variant: 5 in PSEN2, 1 in PSEN1, and 6 in APP (Supplement Table e-4). Twelve participants were identified as additional LoF variant carriers: 6 are PSEN2 variant carriers, 1 is PSEN1 variant carrier, and 5 are APP variant carriers (Table 2). One additional LoF variant carrier has two LoF variants in APP (rs1568829407, p. Asp360LeufsTer50 and rs1568829384, p. Val362ThrfsTer10). Additionally, 197 rare variants (MAF < 1%) within ADSP were predicted to be additional DM variants (Supplement Table e-5). Eighty-three out of the 197 additional DM variants were found in PSEN2 with another 45 in PSEN1 and 69 in APP (Table 2). The frequency of DM variant carriers is much higher than that of clinical, conflicting clinical, or LoF variant carriers, with 843 participants carrying at least one alternate allele of the predicted DM variants. Specifically, 382 participants carry DM variants in PSEN2, 74 carry DM variants in PSEN1, and 387 carry DM variants in APP. And 100 individuals of all the additional DM variant carriers were first-degree relatives evaluated via the kinship coefficients. Among the 12 the additional LOF variant carriers, 5 individuals were AD cases, 3 were controls, and another 4 with AD unknown status. This difference was reversed in the additional DM variant carriers since there were more controls than cases (43.30% vs 48.40%, Table 1).

The rate of carrying additional LoF variants in the APP gene was similar among cases and controls. Four AD cases were identified as carrying an additional LoF variant in the PSEN2 gene while only one control had a PSEN2 additional LoF variant (Table 3), suggesting these may be additional functional PSEN2 variants. The frequency of additional DM variants in controls indicates that this class of variation is too diverse for identifying additional functional DM variants. Overall, the low clinical/LoF/DM variant carrier rates within ADSP participants suggest that it’s extremely unlikely for controls who are greater than 60 years old to be a carrier of a pathogenic Mendelian AD variant (Table 3).

The carrier rate varied by NIAGADS defined study sites (Supplementary Table e-6), with highest carrier rates in a family study and a sample of individuals selected based on family history (the “enriched” sample within the ADSP Discovery case/control set). Because family history of AD is not available across the ADSP study, we were not able to evaluate the carrier rate by presence or absence of family history.

The ADSP samples included in this analysis are predominantly Non-Hispanic White (66.9%), followed by Non-Hispanic Black (16.85%), Hispanic White (1.02%), Hispanic Black (0.23%), Hispanic Other/Unknown (14.68%), and Non-Hispanic Other/Unknown (0.32%). None of the clinical variant carriers were of Non-Hispanic Other/Unknown population, and the lowest observed carrier rate was found among individuals reported as Hispanic Black (0.003%) compared to Non-Hispanic White (0.15%), Non-Hispanic Black (0.01%), Hispanic White age(0.04%), or Hispanic Other/Unknown (0.05%) (Supplementary Table e-7). Within each population, cases have a higher clinical variant carrier rate compared to the controls (Supplementary Table e-7).

We examined whether carrying an alternate allele of one of the curated variants is associated with the age of onset of AD. Among participants with AD and available age of onset data, we observed 62 clinical variant carriers, 5 additional LoF variant carriers, 224 conflicting clinical variant carriers, 365 additional DM carriers, and 13,169 non-carriers. The mean age of onset of AD among clinical variant carriers was significantly lower (55.6 ± 10.9 range 34-76 years) than the mean age of onset in the non-carriers (75.6 ± 9.5 range 39-90 years) when accounting for relatedness among participants (score test p-value=7.8×10^-57, Figure 2). Since we only observed 5 additional LoF variant carriers with age of onset, we collapsed the clinical variant carriers with additional LoF variant carriers and tested the difference in the mean age of onset between the clinical+LoF variant carriers and the non-carriers, and we observed a significant difference (score test p-value= 1.1×10^-54, Figure 2). However, interpretation of these results must be placed in the context of the age-based risk score used for selection of a portion of the subjects included in the ADSP dataset.

• Download figure
• Open in new tab

Figure 2. Age of AD Onset Among Carriers and Non-Carriers.

Boxplot showing the age of AD onset in years among the carriers and non-carriers of clinical, conflicting clinical, additional LoF, or additional DM variants. The mean and standard deviation (SD) of age of AD onset are 55.98 ± 10.88 years for clinical variant carriers, 70.60 ± 13.78 years for additional LoF variant carriers, 76.30 ± 9.30 for conflicting clinical variant carriers, 75.28 ± 9.73 for additional DM variant carriers, and 75.56 ± 9.48 for non-carriers. Linear mixed models adjusted for sex and the first four principal components, accounting for an empirical Balding-Nichols kinship matrix were used to contrast the mean age of AD onset between: 1) clinical variant carriers and non-carriers; 2) clinical variant carriers + additional LoF variant carriers and non-carriers. The exact p-values for the two comparisons are 7.79e^-57 and 1.10e^-54. *** denotes p-value < 0.001.

When looking at the APOE allele distribution in clinical variant carriers and non-carriers, we found 1 control and 20 cases who carry at least one APOE ε4 allele among the clinical variant carriers. Six clinical variant carriers who were diagnosed with AD carry one APOE ε2 allele whereas none of the control clinical variant carriers were APOE ε2 carriers indicating they may harbor another protective variant. The APOE ε3/ε3 genotype is the most prevalent genotype both in cases and controls among the clinical variant carriers. There is a higher proportion of APOE ε2 allele carriers in the controls (15.8%) as compared to cases (8.1%) among the clinical variant non-carriers. Conversely, more cases (50.3%) were seen compared to controls (26.5%) among the APOE ε4 allele carriers (Supplementary Table e-8). APOE genotype was part of the case selection criterion for the WES case-control study, which can lead to an inverted association of APOE ε4 in some sets of individuals (i.e., cases selected without known genetic risk and controls who avoided AD despite known genetic risk). We didn’t observe any statistical evidence for potential protective or synergic effect by the APOE ε alleles with the clinical variant carriers of the 3 Mendelian genes, however, this might be due to limited clinical carrier counts hence we lack the power to formally test the gene by gene interaction on AD.

DISCUSSION

Two studies using a prior release of ADSP data examined variants within APP, PSEN1, PSEN2, and other dementia related genes.^7,8 Blue et al examined 578 WGS and 10,836 WES samples and Fernandez et al examined 143 WGS and 10,280 WES samples from the ADSP. The current study, using a more recent data release, doubles the sample size over the prior studies and increases the diversity among the study participants.

We identified variants in the PSEN2, PSEN1, and APP genes previously implicated in Mendelian AD within the ADSP WES and WGS data sets. Among clinical variant carriers overall, a higher proportion were cases compared to controls. However, some previously reported clinical variants were observed in control individuals. This may reflect the age of the carrier. Clinical variant carriers who were controls had ages of 73, 74, and 90 and some of these individuals may manifest AD in the future. Our observation of clinical variants in controls may also reflect reduced penetrance or the misclassification of a variant as clinically relevant for AD. Within the ADSP, AD cases were more likely to be carrying clinical variants in PSEN2, PSEN1 or APP (0.03%, 0.41% and 0.01%, respectively) compared to controls (Table 3). However, we observed higher proportions of controls carrying conflicting clinical variants in PSEN2 and APP (Table 3). This could be explained by varying penetrance of the variants or that these variants are potentially not causal for AD.

Four conflicting clinical variants were labeled as pathogenic/likely pathogenic in either OMIM or Alzforum but not in ClinVar (rs63750197 PSEN2 p.Ser130Leu, rs63750831 PSEN1 p.Val94Met, rs63750847 APP p.Ala673Thr, rs63750363 APP p. Glu665Asp) (Supplementary Table e-3). Arboleda-Velasquez et al reported the protective effect of the APOE3ch mutation (rs121918393.p.Arg154Ser) against a PSEN1 variant (rs63750231.p. Glu280Ala).¹⁸ However, both the APOE3ch and PSEN1 variants were not present in our ADSP sample. In a recent report, Xian et al re-evaluated 452 pathogenic variants in PSEN1, PSEN2, and APP genes from PubMed and Alzforum according to the ACMG-AMP guidelines and found 5.09% of the variants were of uncertain significance.¹⁹ Altogether, these studies support the hypothesis that there is misclassification or reduced penetrance of the pathogenic/likely pathogenic variants from ClinVar, OMIM, and Alzforum, and that our data can aide in reclassifying these variants.

We observed that the mean age of onset for clinical variant carriers was significantly earlier than for non-carriers and individuals with an additional LoF variant had a mean age of onset lower than non-carriers, suggesting that the additional LoF variants may be of clinical relevance. However, this result should be interpreted with caution given distinct ascertainment schemes in different phases of the ADSP, some of which included exclusion thresholds for age at onset. Onset in the 4^th and 5^th decade of life was seen in individuals who did not carry an alternate allele at an identified clinical variant, suggesting that additional variants, potentially in other genes, driving early onset AD have yet to be identified.

Although the ADSP participants used in this analysis are predominantly Non-Hispanic White, a substantial proportion (33.1%) of the subjects are reported as other races/ethnicities, primarily Non-Hispanic Black and Hispanic. The carrier rates for the clinical variants were not similar among defined populations. The highest carrier rate was present among Non-Hispanic White individuals and the lowest carrier rate among participants from the Non-Hispanic Other/Unknown population (0.15% and 0%, respectively) (Supplementary Table e-7). Previous studies have shown that disparities in inclusion in genomic sequencing may result in misclassification or a higher rate of variants of uncertain significance for patients in underrepresented populations.²⁰ A study conducted on assessing the underlying genetic risk for hypertrophic cardiomyopathy showed that variants that had been classified as pathogenic were actually common variants in the African American population.²¹ These variants were misclassified as pathogenic due to the lack of African American controls in previous studies, emphasizing the importance of including diverse ancestry groups in order to avoid health disparities as a result of misdiagnosis.²¹ Another study showed that the sample sizes of most non-White populations within The Cancer Genome Atlas were not sufficient to detect common variants across several cancer types specific to different racial and ethnic groups.²² These studies demonstrate the importance of including diverse populations in genetic studies in order to aid in accurate disease diagnosis and treatment. Further investigation is needed to determine if the differences in clinical variant carrier rates among populations within the ADSP is due to a true difference in the frequency of AD causal PSEN2, PSEN1, APP variants across populations, reflects study design including ascertainment criteria, higher propensity for clinical variant screening for these three genes in White participants, or limitations due to populations historically included in genetic studies of AD. The emphasis of the ADSP on increased diversity of study participants will help to answer some of these questions.

There were 2,950 individuals with unknown AD status in our study sample, out of which we identified 13 clinical variant carriers, 56 conflicting clinical variant carriers, 4 additional LoF variant carriers, and 70 additional DM variant carriers (Supplementary Table e-9). 16.7% of the clinical variant carriers had unknown AD status compared to 3.85% being as controls, indicating a mixture of potential AD cases and controls among the AD unknown individuals (Table 1).

Upon further examination of the phenotypic data, 775 individuals out of the 2,950 do not have any phenotypic data, 313 individuals from the ADNI study had mild cognitive impairment (MCI), 549 individuals were diagnosed with progressive supranuclear palsy (PSP) while 332 individuals had corticobasal degeneration (CBD), an additional 202 individuals were from family studies, with another 779 individuals enrolled from ADSP case-control studies (Supplementary Table e-10).

In conclusion, we identified variants in the PSEN2, PSEN1, APP genes within the ADSP WES and WGS datasets. These data confirm previously-reported clinical variants, identifies additional LoF variants implicated in AD within these three Mendelian AD genes, and suggests conflicting clinical variants that may not be causal for AD. The ADSP generates WES and WGS data sets that are jointly called and QC’ed across studies. These genomic files, along with phenotype files that include AD status, are available to qualified researchers to facilitate the discovery of variants that are protective for or confer risk for AD. Understanding the presence of known Mendelian AD variants within these data may inform analyses of variants at other loci in relation to AD risk.

Appendix 1