Research Letter | Neurology
Accuracy of Imputation for Apolipoprotein E εAlleles in Genome-Wide
Genotyping Data
Eero Vuoksimaa, PhD; Teemu Palviainen, MSc; Noora Lindgren, MSc; Juha O. Rinne, MD, PhD; Jaakko Kaprio, MD, PhD
Introduction
Given the importance of the apolipoprotein E (APOE) gene for risk of Alzheimer disease, determining
this genotype is important in cognitive aging studies. Before the genome-wide genotyping era, the
APOE gene (alleles ε2, ε3, ε4) was directly genotyped and defined by 2 single-nucleotide
polymorphisms (SNPs), rs429358 and rs7412, in chromosome 19. Owing to rapid development of
genotyping technology, the price of genome-wide arrays covering approximately 500000 SNPs has
decreased to less than $100, making it more cost-effective compared with direct genotyping of a
single gene, such as APOE. In addition to genotyped SNPs, the information can be used to impute
common nonmeasured variants such as rs429358 and rs7412 that are not directly genotyped on
many chips. However, imputation accuracy depends on the genome-wide arrays, quality control, and
reference samples.1,2
In this diagnostic study, we evaluated the association of reference panels with imputation
quality of rs429358 and rs7412 by comparing imputation based on 3 different reference panels: 1000
Genomes (1000G),3 Haplotype Reference Consortium (HRC),4 and the Finnish-specific Sequencing
Initiative Suomi (SISu).5
Methods
We used a population-based older Finnish Twin Cohort6 study to examine the correspondence
between rs429358 and rs7412 directly genotyped using a Sequenom (Taqman) and imputed
rs429358 and rs7412 using 1000G Phase III version 5, HRC release 1.1, and SISu reference panels. Raw
genotype data in a larger sample (5343 participants) using 5 array versions (12 v1.0 A, 12 v1.1 A, 24
v1.0 A, 24 v1.1 A, and 24 v1.2 A) of HumanCoreExome (Illumina) were merged before the quality
control phase. We removed variants with call rate less than 97.5%, samples with call rate less than
95%, variants with minor allele frequency less than 1%, and variants with Hardy-Weinberg
equilibrium P < 1.0 × 10−6. We removed samples with heterozygosity test method-of-moments F
coefficient estimate values less than −0.03 or greater than 0.05, multidimensional scaling principal
component analysis outliers, and samples that failed sex check. The number of genotyped autosomal
variants after quality control was 239 894 (5328 participants).
We then performed prephasing using Eagle software version 2.3 (Broad Institute) and
imputationwithMinimac3 software version 2.0.1 (University ofMichigan) (Table 1). The study sample
(1704 participants) with directly genotyped rs429358 and rs7412 was extracted from each imputed
data set. Ethical approval was obtained from the ethical committee of the Hospital District of
Southwest Finland, and participants gave written informed consent.
Results
Participants were of European ancestry (mean [SD] age, 74.2 [4.9] years; 775 [45%] women). For
directly genotyped individuals, 984 (57.7%) had ε3/ε3 genotype, 521 (30.6%) had ε3/ε4 or ε4/ε4, 171
(10%) had ε2/ε2 or ε2/ε3, and 28 (1.6%) had ε2/ε4. Allele frequencies were 0.060 for ε2, 0.765 for
ε3, and 0.175 for ε4.
Author affiliations and article information are
listed at the end of this article.
Open Access. This is an open access article distributed under the terms of the CC-BY License.
JAMA Network Open. 2020;3(1):e1919960. doi:10.1001/jamanetworkopen.2019.19960 (Reprinted) January 24, 2020 1/4
Downloaded From: https://jamanetwork.com/ by a University of Turku User  on 05/25/2020
Based on 1000G, 1701 individuals (99.82%) had correctly classified alleles of both rs429358
and rs7412 (Table 1). Results were similar for HRC and SISu: 1702 individuals (99.88%) and 1703
individuals (99.94%) had correctly classified rs429358 and rs7412, respectively (Table 1).
Using the HRC reference panel, 1702 individuals (99.88%) had correctly classified APOE
genotype and ε4 carrier status (Table 2). Two (0.12%) of the ε4 noncarriers based on directly
genotyped SNPs were incorrectly classified as ε4 carriers.
Table 1. Cross Tabulation of Directly Genotyped and Imputed Single-Nucleotide Polymorphism of rs429358
and rs7412 for 1704 Individualsa
Imputation Based Genotyped, No. (%) Total, No.
1000 Genomes
rs429358 TT CT CC
TT 1152 (99.74) 3 (0.26) 0 1155
CT 0 503 (100) 0 503
CC 0 0 46 (100) 46
rs7412 CC CT TT
CC 1505 (100) 0 0 1505
CT 2 (1.03) 192 (98.97) 0 194
TT 0 1 (20.0) 4 (80.0) 5
Haplotype Reference Consortium
rs429358 TT CT CC
TT 1153 (99.83) 2 (0.17) 0 1155
CT 0 503 (100) 0 503
CC 0 0 46 (100) 46
rs7412 CC CT TT
CC 1505 (100) 0 0 1505
CT 0 194 (100) 0 194
TT 0 1 (20.0) 4 (80.0) 5
Sequencing Initiative Suomi
rs429358 TT CT CC
TT 1153 (99.83) 2 (0.17) 0 1155
CT 0 503 (100) 0 503
CC 0 0 46 (100) 46
rs7412 CC CT TT
CC 1505 (100) 0 0 1505
CT 0 194 (100) 0 194
TT 0 1 (20.0) 4 (80.0) 5
a Genotypes were imputed to 1000 Genomes Phase
III version 5,3 Haplotype Reference Consortium
release 1.1,4 and Sequencing Initiative Suomi Finnish-
only reference panels.5 Imputation to 1000
Genomes and Haplotype Reference Consortium
reference panels was done using the University of
Michigan Imputation Server. The Sequencing
Initiative Suomi reference panel consists of
16 962023 variants from 3775 high-pass whole-
genome (depth up to 30×) sequences.
Table 2. Cross Tabulation of Directly Genotyped andHaplotype Reference ConsortiumReference Panel Imputation–BasedAPOE Status in 1704 Individuals
APOE Status (% Individuals)
APOE Status Based on Imputed Single-Nucleotide Polymorphisms, No.
Totalε2/ε2 ε2/ε3 ε2/ε4 ε3/ε3 ε3/ε4 ε4/ε4
ε2/ε2 (0.29%) 4 0 1 0 0 0 5
ε2/ε3 (9.74%) 0 166 0 0 0 0 166
ε2/ε4 (1.64%) 0 0 28 0 0 0 28
ε3/ε3 (57.75%) 0 0 0 983 1 0 984
ε3/ε4 (27.88%) 0 0 0 0 475 0 475
ε4/ε4 (2.70%) 0 0 0 0 0 46 46
Abbreviation: APOE, apolipoprotein E.
JAMANetworkOpen | Neurology Accuracy of Imputation for Apolipoprotein E ε Alleles in Genome-Wide Data
JAMA Network Open. 2020;3(1):e1919960. doi:10.1001/jamanetworkopen.2019.19960 (Reprinted) January 24, 2020 2/4
Downloaded From: https://jamanetwork.com/ by a University of Turku User  on 05/25/2020
Discussion
This study found that by using arrays described in theMethods section, imputation to all 3 reference
panels, 1000G, HRC, and SISu, yielded high imputation accuracy of rs429358 and rs7412, 2 SNPs
needed to determine polymorphic APOE ε alleles. The number of Finnish samples do vary in different
reference panels: 99 in 1000G Phase III, approximately 1900 in the HRC, and 3800 in the Finnish-
only SISu. Considering 1000G Phase III yielded improved accuracy compared with Phase I.1,2 Our
results also suggest that determination of APOE can be reached equally well with most recent freely
available cosmopolitan reference panels comparedwith a population-specific reference panel. Still,
all Finnish samples and inclusion of only 1 brand of arrays were also limitations, and these results
should be confirmed in people with different ancestry.
ARTICLE INFORMATION
Accepted for Publication:December 2, 2019.
Published: January 24, 2020. doi:10.1001/jamanetworkopen.2019.19960
Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2020 Vuoksimaa
E et al. JAMA Network Open.
Corresponding Author: Eero Vuoksimaa, PhD, Institute for Molecular Medicine Finland (FIMM), PO Box 20
(Tukholmankatu 8), 00014 University of Helsinki, Helsinki, Finland (eero.vuoksimaa@helsinki.fi).
Author Affiliations: Institute for Molecular Medicine Finland (FIMM), University of Helsinki, Helsinki, Finland
(Vuoksimaa, Palviainen, Kaprio); Turku PET Centre, University of Turku, Turku, Finland (Lindgren, Rinne); Division
of Clinical Neurosciences, Turku University Hospital, Turku, Finland (Rinne); Department of Public Health,
Clinicum, University of Helsinki, Helsinki, Finland (Kaprio).
Author Contributions:Dr Vuoksimaa andMr Palviainen had full access to all of the data in the study and take
responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Vuoksimaa, Palviainen, Kaprio.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Vuoksimaa, Palviainen.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Palviainen.
Obtained funding: Vuoksimaa, Rinne, Kaprio.
Administrative, technical, or material support: Kaprio.
Conflict of Interest Disclosures:Dr Rinne reported serving as a neurology consultant for CRST Oy (Clinical
Research Services Turku). No other disclosures were reported.
Funding/Support: This study was supported by the Academy of Finland (grants 314639 and 320109 to Dr
Vuoksimaa, grant 312073 to Dr Kaprio, and grant 310962 to Dr Rinne) and a Sigrid Juselius Foundation grant to Dr
Rinne. Ms Lindgren was supported by the Finnish Cultural Foundation, Päivikki and Sakari Sohlberg Foundation,
Yrjö Jahnsson Foundation, Turku University Foundation, Finnish Brain Foundation, and Finnish State
Research Funding.
Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection,
management, analysis, and interpretation of the data; preparation, review, or approval of themanuscript; and
decision to submit themanuscript for publication.
Additional Contributions:We thank the participants of the older Finnish Twin Cohort study.
REFERENCES
1. Oldmeadow C, Holliday EG, McEvoyM, et al. Concordance between direct and imputed APOE genotypes using
1000 Genomes data. J Alzheimers Dis. 2014;42(2):391-393. doi:10.3233/JAD-140846
2. LuptonMK, Medland SE, Gordon SD, et al. Accuracy of inferred APOE genotypes for a range of genotyping
arrays and imputation reference panels. J Alzheimers Dis. 2018;64(1):49-54. doi:10.3233/JAD-171104
3. Auton A, Brooks LD, Durbin RM, et al; 1000Genomes Project Consortium. A global reference for human genetic
variation. Nature. 2015;526(7571):68-74. doi:10.1038/nature15393
JAMANetworkOpen | Neurology Accuracy of Imputation for Apolipoprotein E ε Alleles in Genome-Wide Data
JAMA Network Open. 2020;3(1):e1919960. doi:10.1001/jamanetworkopen.2019.19960 (Reprinted) January 24, 2020 3/4
Downloaded From: https://jamanetwork.com/ by a University of Turku User  on 05/25/2020
4. McCarthy S, Das S, KretzschmarW, et al; Haplotype Reference Consortium. A reference panel of 64,976
haplotypes for genotype imputation. Nat Genet. 2016;48(10):1279-1283. doi:10.1038/ng.3643
5. Lim ET,Würtz P, Havulinna AS, et al; Sequencing Initiative Suomi (SISu) Project. Distribution andmedical impact
of loss-of-function variants in the Finnish founder population. PLoS Genet. 2014;10(7):e1004494. doi:10.1371/
journal.pgen.1004494
6. Vuoksimaa E, Rinne JO, Lindgren N, Heikkilä K, KoskenvuoM, Kaprio J. Middle age self-report risk score
predicts cognitive functioning and dementia in 20-40 years. Alzheimers Dement (Amst). 2016;4:118-125. doi:10.
1016/j.dadm.2016.08.003
JAMANetworkOpen | Neurology Accuracy of Imputation for Apolipoprotein E ε Alleles in Genome-Wide Data
JAMA Network Open. 2020;3(1):e1919960. doi:10.1001/jamanetworkopen.2019.19960 (Reprinted) January 24, 2020 4/4
Downloaded From: https://jamanetwork.com/ by a University of Turku User  on 05/25/2020