Multiancestral polygenic risk score for pediatric

asthma
Bahram Namjou, MD,a,b Michael Lape,a,c Edyta Malolepsza, PhD,e Stanley B. DeVore,b,d Matthew T. Weirauch, PhD,a,b,c,f
Ozan Dikilitas, MD,g,h Gail P. Jarvik, MD, PhD,i,j Krzysztof Kiryluk, MD,k Iftikhar J. Kullo, MD,h Cong Liu, PhD,l
Yuan Luo, PhD,m Benjamin A. Satterfield, MD, PhD,h Jordan W. Smoller, MD, ScD,n,o,p Theresa L. Walunas, PhD,q
John Connolly, PhD,r Patrick Sleiman, PhD,r,s Tesfaye B. Mersha, PhD,b,d Frank D. Mentch, PhD,r
Hakon Hakonarson, MD, PhD,r,s Cynthia A. Prows, MSN, APRN,b,t,u Jocelyn M. Biagini, PhD,b,d
Gurjit K. Khurana Hershey, MD, PhD,b,d,v Lisa J. Martin, PhD,b,t and Leah Kottyan, PhD,a,b,v The eMERGE Networkw
Cincinnati, Ohio; Cambridge and Boston, Mass; Rochester, Minn; Seattle, Wash; New York, NY; Chicago, Ill; Philadelphia, Pa; and
Bethesda, Md

Background: Asthma is the most common chronic condition in Methods: This study applied a Bayesian regression framework
children and the third leading cause of hospitalization in method using the Trans-National Asthma Genetic Consortium
pediatrics. The genome-wide association study catalog reports genome-wide association study summary statistics to derive a
140 studies with genome-wide significance. A polygenic risk multiancestral PRS score, used one Electronic Medical Records
score (PRS) with predictive value across ancestries has not been and Genomics (eMERGE) cohort as a training set, used a
evaluated for this important trait. second independent eMERGE cohort to validate the score, and
Objectives: This study aimed to train and validate a PRS relying used the UK Biobank data to replicate the findings. A phenome-
on genetic determinants for asthma to provide predictions for wide association study was performed using the PRS to identify
disease occurrence in pediatric cohorts of diverse ancestries. shared genetic etiology with other phenotypes.

From athe Center for Autoimmune Genomics and Etiology, cthe Division of Biomedical Vanderbilt University Medical Center; U01 HG011166 to Vanderbilt University Med-
Informatics, dthe Division of Asthma Research, fthe Division of Developmental ical Center serving as the Coordinating Center. Phase III: U01 HG8657 to Kaiser Per-
Biology, tthe Division of Human Genetics, uthe Department of Patient Services, and manente Washington/University of Washington; U01 HG8685 to Brigham and
v
the Division of Allergy and Immunology, Cincinnati Children’s Hospital Medical Women’s Hospital; U01 HG8672 to Vanderbilt University Medical Center; U01
Center; bthe Department of Pediatrics, College of Medicine, University of Cincinnati; HG8666 to Cincinnati Children’s Hospital Medical Center; U01 HG6379 to Mayo
o
the Stanley Center for Psychiatric Research, ethe Broad Institute of Massachusetts Clinic; U01 HG8679 to Geisinger Clinic; U01 HG8680 to Columbia University Health
Institute of Technology and Harvard University, Cambridge; gthe Department of Inter- Sciences; U01 HG8684 to Children’s Hospital of Philadelphia; U01 HG8673 to North-
nal Medicine, and hthe Department of Cardiovascular Medicine, Mayo Clinic, Roches- western University; U01 HG8701 to Vanderbilt University Medical Center serving as
ter; ithe Division of Medical Genetics, Department of Medicine, and jthe Department the Coordinating Center; U01 HG8676 to Partners Healthcare/Broad Institute; and
of Genome Sciences, University of Washington Medical Center, Seattle, kthe Division U01 HG8664 to Baylor College of Medicine. Phase II: U01 HG006828 to Cincinnati
of Nephrology, Department of Medicine, College of Physicians and Surgeons, and lthe Children’s Hospital Medical Center/Boston Children’s Hospital; U01 HG006830 to
Department of Biomedical Informatics, Columbia University, New York; mthe Depart- Children’s Hospital of Philadelphia; U01 HG006389 to Essentia Institute of Rural
ment of Preventive Medicine, and qthe Division of General Internal Medicine and Ge- Health, Marshfield Clinic Research Foundation, and Pennsylvania State University;
riatrics, Department of Medicine, Northwestern University Feinberg School of U01 HG006382 to Geisinger Clinic; U01 HG006375 to Group Health Cooperative/
Medicine, Chicago; nthe Psychiatric and Neurodevelopmental Genetics Unit, Center University of Washington; U01 HG006379 to Mayo Clinic; U01 HG006380 to Icahn
for Human Genomic Medicine, Massachusetts General Hospital, and pthe Department School of Medicine at Mount Sinai; U01 HG006388 to Northwestern University; U01
of Psychiatry, Harvard Medical School, Boston; rthe Center for Applied Genomics, HG006378 to Vanderbilt University Medical Center; and U01 HG006385 to Vander-
Department of Pediatrics, Children’s Hospital of Philadelphia; sthe Perelman School bilt University Medical Center serving as the Coordinating Center. Genotyping center
of Medicine at the University of Pennsylvania, Philadelphia; and wthe National Human support U01 HG004438 to Center for Inherited Disease Research and U01 HG004424
Genome Research Institute, National Institutes of Health, Bethesda. to the Broad Institute. Phase I: U01 HG004610 to Group Health Cooperative/Univer-
Supported by grants NIH R01 HG010730, NIH R01 NS099068, NIH R01 GM055479, sity of Washington; U01 HG004608 to Marshfield Clinic Research Foundation and
NIH U01 AI130830, NIH R01 AI141569, and NIH U01 AI150748 to M.T.W.; grants Vanderbilt University Medical Center; U01 HG04599 to Mayo Clinic; U01
NIH R01 DK107502, NIH R01 AI148276, NIH U19 AI070235, NIH U01 HG011172, HG004609 to Northwestern University; U01 HG04603 to Vanderbilt University Med-
and NIH P30 AR070549 to L.C.K.; grants NIH R01 AR073228 and NIH R01 ical Center, also serving as the Administrative Coordinating Center; U01 HG004438 to
AI024717 and Cincinnati Children’s Hospital Medical Center Academic Research Center for Inherited Disease Research; and U01 HG004424 to the Broad Institute
Committee (ARC) Award 53632 to M.T.W. and L.C.K.; grants NIH R01 HG010166, serving as genotyping centers.
NIH R01 HL145422, NIH R25 GM129808, NIH R01 AI127392, NIH UG3 This research has been conducted using data from UK Biobank, a major biomedical
OD023282, NIH U19 AI070235, NIH U54 AI117804, NIH R01 NS096053, NIH database (www.ukbiobank.ac.uk).
R01 DK107502, NIH R01 HD089458, NIH R01 HL132153, NIH R01 AI139126, NIH Disclosure of potential conflict of interest: The authors declare that they have no relevant
R01 HL135114, NIH R01 HD099775, and NIH U01 HG011172 to L.J.M.; and grants conflicts of interest.
NIH R01 HL132344 and NIH R01 HG011411 to T.B.M. Received for publication September 14, 2021; revised March 7, 2022; accepted for pub-
The eMERGE (Electronic Medical Records and Genomics) Network was initiated and lication March 29, 2022.
funded by the National Human Genome Research Institute through the following Corresponding author: Leah Kottyan, PhD, Cincinnati Children’s Hospital Medical Cen-
grants: Phase IV: U01 HG011172 to Cincinnati Children’s Hospital Medical Center; ter, 3333 Burnet Avenue, Cincinnati, OH 45229. E-mail: Leah.Kottyan@cchmc.org.
U01 HG011175 to Children’s Hospital of Philadelphia; U01 HG008680 to Columbia 0091-6749
University; U01 HG011176 to Icahn School of Medicine at Mount Sinai; U01 Ó 2022 The Authors. Published by Elsevier Inc. on behalf of the American Academy of
HG008685 to Mass General Brigham; U01 HG006379 to Mayo Clinic; U01 Allergy, Asthma & Immunology. This is an open access article under the CC BY-NC-
HG011169 to Northwestern University; U01 HG011167 to University of Alabama ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
at Birmingham; U01 HG008657 to University of Washington; U01 HG011181 to https://doi.org/10.1016/j.jaci.2022.03.035

1
2 NAMJOU ET AL J ALLERGY CLIN IMMUNOL
nnn 2022

Results: The multiancestral asthma PRS was associated with

asthma in the 2 pediatric validation datasets. Overall, the Abbreviations used
multiancestral asthma PRS has an area under the curve (AUC) AUC: Area under the curve
of 0.70 (95% CI, 0.69-0.72) in the pediatric validation 1 and eMERGE: Electronic Medical Records and Genomics (consortium)
AUC of 0.66 (0.65-0.66) in the pediatric validation 2 datasets. FDR: False discovery rate
GWAS: Genome-wide association study
We found significant discrimination across pediatric subcohorts
ICD: International Classification of Diseases
of European (AUC, 95% CI, 0.60 and 0.66), African (AUC, 95% LD: Linkage disequilibrium
CI, 0.61 and 0.66), admixed American (AUC, 0.64 and 0.70), OR: Odds ratio
Southeast Asian (AUC, 0.65), and East Asian (AUC, 0.73) PCA: Principal component analysis
ancestry. Pediatric participants with the top 5% PRS had 2.80 PheWAS: Phenome-wide association study
to 5.82 increased odds of asthma compared to the bottom 5% PRS: Polygenic risk score
across the training, validation 1, and validation 2 cohorts when TAGC: Trans-National Asthma Genetic Consortium
adjusted for ancestry. Phenome-wide association study analysis
confirmed the strong association of the identified PRS with
asthma (odds ratio, 2.71, PFDR 5 3.71 3 10265) and related
phenotypes. although pediatric- and adult-onset asthma have numerous shared
Conclusions: A multiancestral PRS for asthma based on Bayesian risk loci, genetic studies have also identified distinct risk loci in
posterior genomic effect sizes identifies increased odds of children.14 Furthermore, the risk of asthma is different across
pediatric asthma. (J Allergy Clin Immunol 2022;nnn:nnn-nnn.) ancestral groups, with children of African descent having higher
frequencies than children of European descent do,15-20 so it is
Key words: Genetics, asthma, GWAS, polygenic risk score, PRS, important that for a PRS to be maximally clinically informative,
PheWAS the PRS should be developed and validated using training datasets
that are similar to the populations to whom the PRS will be
applied. Thus, a PRS developed using primarily adult studies or
Asthma is an inflammatory disease of the airway with symptoms based on limited ancestral diversity may not be optimal to predict
including coughing, wheezing, chest tightness, and shortness of pediatric asthma. To address the current limitations, our goal was
breath caused by airflow obstruction and hyperresponsiveness.1,2 to develop a high-quality PRS for pediatric asthma using data
Asthma affects 7 million children across the United States, yielding from a diverse cohort representing multiple ancestries at all stages
a prevalence equaling 10% of children, but this prevalence can vary of the process. To address this goal, we used an existing large-
by sex, ancestry, and age.1 Patients with asthma experience scale genome-wide association derived from diverse populations
increased morbidity with respiratory infection, days absent from and a Bayesian framework to derive a multiancestral PRS and
school and work, emergency department visits, hospitalizations, tested and validated this score in multiple cohorts and phenotype
and even death. While asthma is a significant economic and health definitions. With the development of this pediatric asthma PRS,
burden, proactive treatment and specific interventions can prevent we are now poised to test the clinical utility.
severe disease requiring emergency or inpatient management.3,4 Our intent is to evaluate the clinical utility of this PRS in a
Given this, a major focus is primary prevention.5 There have prospective study that will include individuals across ancestral
been substantial efforts to develop asthma prediction models based groups. It is routine in clinical settings to ask about an individual’s
on clinical factors,6,7 but these tools rely on early life phenotypes race and ethnicity. Yet, how an individual self-identifies is not
that may not have been assessed. Thus, development of additional equivocal to their genetic ancestral group. Furthermore, many
predictive tools are warranted. individuals have admixed ancestry. Thus, we developed a multi-
The etiology of asthma is multifactorial with contributions from ancestral PRS that could be used for all individuals.
environmental and genetic factors.4,8,9 Twin studies estimate a her-
itability of 60% to 80% of asthma susceptibility attributable to ge-
netic factors while also highlighting the etiological contributions of
shared environment.10 At the time of development of this multian- METHODS
cestral asthma polygenic risk score (PRS), there were >140 studies Study overview
in the genome-wide association study (GWAS) catalog with 2167 The study design is presented in Fig 1. We divided the combined Electronic
Medical Records and Genomics (eMERGE) I, II, and III imputed dataset from
variants reported with genome-wide significant results; 50 of the re-
over 105,000 samples (https://emerge-network.org/genomics) into 2 nonover-
ported studies resulted from studies of participants not of full Euro- lapping independent datasets (eMERGE training dataset and eMERGE valida-
pean ancestry.11 These GWASs identify genetic loci that increase tion 1 dataset 2) based on date of participation in eMERGE. The UK Biobank
risk for asthma with a common genetic variant-based narrow sense cohort was used as a validation 2 dataset. For the training eMERGE dataset,
heritability of 14.9%;12 however, any single risk locus does not pro- patients with asthma and controls were identified based on a validated algo-
vide sufficient risk discrimination to be of practical clinical use. By rithm using structured data (International Classification of Diseases [ICD]
accounting for genotypes at risk variants in proportion to the effect version 9 [ICD-9 493.xx] or version 10 [ICD-10 J45, J46] codes) for asthma
size of each genetic locus, PRSs have the potential to be tools for as well as unstructured data as previously described.21 Participants in the
incorporating genetic risk into clinical decision support.13 eMERGE validation dataset were classified as cases by having > _2 ICD-9/10
A PRS for asthma has been developed; however, the data used codes for asthma and as controls by having no history of asthma or atopy.
The minor differences in case definitions were based on improvements in clin-
to generate the PRS were limited to individuals of European
ical data extraction over the past 15 years of eMERGE. For the validation 2
descent. The limited diversity of participants used to generate the dataset in the UK Biobank, we used a previously established algorithm that
PRS is a problem because different demographic groups may was developed by UK Biobank team (Resource ID 4124) to identify partici-
have different underlying genetic etiologies. For example, pants with asthma.
J ALLERGY CLIN IMMUNOL NAMJOU ET AL 3
VOLUME nnn, NUMBER nn

FIG 1. Study design. A, Summary statistics from 985,837 genetic variants with minor allele frequencies
(MAF) >1% in the overall (cases and controls combined) TAGC discovery GWAS were used to develop
the multiancestral asthma PRS in the context of the LD of the 1000 Genomes reference panel and the
training dataset. B, The eMERGE dataset was split into 2 independent datasets (training and validation 1)
and the UK Biobank was used as a validation 2 dataset. Subjects with confirmed pediatric-onset asthma
and controls were used for PRS training, validation 1, and validation 2 cohorts. C, Individual genotypes
from each subject in each dataset was used to perform a PCA. For each individual, genotypes at each of
the 985,837 variants included in the multiancestral asthma PRS were used to calculate a PCA-adjusted
PRS. The adjusted multiancestral PRS was applied to each individual in the training, validation 1, and vali-
dation 2 cohorts in preparation for it to be similarly applied to individuals recruited into an institutional re-
view board–approved prospective intervention study. Figure created in BioRender.

Study population of eMERGE and UK Biobank and Multicenter Research Ethics Committee and Patient Information Advisory
phenotype ascertainment Group approved the UK Biobank study. All analyses were conducted using
The eMERGE Network is a National Institutes of Health–organized and deidentified data.
–funded consortium of US medical research institutions (https://www.genome. Prior to analyses, participant-level quality control was employed. Self-
gov/Funded-Programs-Projects/Electronic-Medical-Records-and-Genomics- reported race and ethnicity was not used to identify ancestry. We used
Network-eMERGE). Postimputation whole genome genotyping data for partic- genetically defined ancestry for ancestry-specific analyses. Principal compo-
ipants from the eMERGE Network were made available for this study (dbGAP nent analysis (PCA) of genome-wide genetic variants was used to establish
[phs000888.v1.p1]). The associated BAM, xml, and vcf files are available on ancestry (Table I). The FRAPOSA software package was used to perform PCA
the eMERGE Commons web portal, accessible to sites as well as outside inves- and assign all individuals into 5 major superpopulations (European, African,
tigators who apply for access (see eMERGE Network in Web Resources). The Admixed American, East Asian, and South Asian).26 We used the phase 3
imputation process and genotype quality control in eMERGE followed guide- release of the 1000 Genomes data as a reference that consists of 2504 individ-
lines that have been published previously.22 Briefly, all subjects and variants uals from 5 superpopulations as shown in Table E1 in this article’s Online Re-
had missingness <2%. Individual level genotype data were derived from 78 array pository available at www.jacionline.org.27 The major steps in the FRAPOSA
batches across 12 academic medical centers. Each batch was imputed using the algorithm include computing principal components of the reference dataset
Michigan Imputation Server, which provides a missing single-nucleotide variant using the matched variants only and projecting computed principal compo-
genotype imputation service using the minimac3 imputation algorithm with the nents to the target data using an optimized implementation of the Online
Haplotype Reference Consortium genotype reference set. Only 1 genetic dataset Augmentation, Decomposition, and Procrustes (OADP) transformation.
was retained for each participant.22 Next, the algorithm predicts the ancestry membership by using the K20-near-
The UK Biobank is a large long-term biobank study in the United Kingdom est-neighbor method.19 The pairwise correlation between self-reported race
developed to support the investigation of the respective contributions of and genetic ancestry was 99% in UK Biobank and 97% in the eMERGE data-
genetic predisposition and environmental exposure to the development of set, if we exclude those who self-identified as having a mixed or Hispanic race/
diseases23 including asthma.24,25 The UK Biobank postimputed data were ob- ethnicity. All participants with sex inconsistencies were removed; additionally
tained through application ID: 47377. the dataset was pruned to remove participants to prevent duplicated individ-
All participants in the eMERGE and UK Biobank cohorts provided written uals, twins, and first-degree relatives using PLINK’s implementation of
informed consent prior to study inclusion. The institutional review board of KING robust kinship coefficients.28 In the KING pipeline, after the kinship
each contributing institution approved the eMERGE study. The North West (relationship) matrix is calculated using high-quality markers for all individ-
uals, kinship-based pruning of samples is performed in which the program by
4 NAMJOU ET AL J ALLERGY CLIN IMMUNOL
nnn 2022

TABLE I. Pediatric study population

Cohort Ancestry Case/control Female/male Age, mean (SD) (y)

Training All 1,324/3,174 2,058/2,440 11.83 (5.32)

European 322/1,736 941/1,117 11.82 (5.20)
African 768/727 697/798 12.55 (5.17)
Admixed 203/642 367/478 12.14 (5.15)
Eastern Asian* 21/40 34/27 11.68 (5.11)
Southern Asian* 10/29 19/20 10.95 (6.01)
Validation 1 All 1,255/7,710 3,988/4,977 10.37 (5.97)
European 386/3,982 1,883/2,485 10.90 (5.75)
African 588/1,352 905/1,035 10.04 (6.25)
Admixed 270/2,180 1,105/1,345 11.13 (5.50)
Eastern Asian* 9/114 62/61 9.69 (6.17)
Southern Asian* 2/82 33/51 10.07 (6.18)
Validation 2 All 16,462/398,808 219,728/195,542 8.40 (4.69)
European 15,586/376,234 207,785/184,035 8.10 (4.82) 
African 310/7,508 4,385/3,433 8.21 (4.70) 
Admixed 205/5,020 2,717/2,508 8.16 (5.06) 
Eastern Asian 49/1,622 877/794 7.96 (4.12) 
Southern Asian 312/8,424 3,964/4,772 9.54 (4.74) 
A total of 70,290 cases and 467,247 controls across eMERGE and the UK Biobank in 5 superpopulations (European, Admixed American, African, East Asians, South Asian) were
evaluated. Pediatric onset was defined as diagnosis before age 18 years. Please see Table E4 for a list of all participants, including those without pediatric-onset asthma.
*Due to low sample size, subgroups identified with asterisks only included in combined adjusted PRS analysis; ancestry-specific results from these subgroups are not presented.
 Mean age of onset for validation 2 cases at the time of diagnosis (see Methods).

default, randomly excludes 1 member of each pair of samples and prints all summary statistics in the context of linkage disequilibrium between variants as
independent individuals for downstream analysis. assessed on an external reference panel (ie, the phase 3 release of the 1000
Results were evaluated in all individuals as well as participants who were Genomes data).30 The GWAS on which the PRS is derived is a multiancestral
enrolled in eMERGE as a child (age < _ 18 years). It was not possible to identify metanalysis, and the effect size for all genetic variants in the study are inverse-
asthma age of onset for all subjects in eMERGE who were over 18. Because variance weighted with fixed effects accounting for all ancestral populations.
the UK Biobank enrolled only adult participants, pediatric-onset asthma was Continuous shrinkage priors that were implemented in this pipeline allowed
identified through an assessment of the date of diagnosis in the context of the for marker-specific adaptive shrinkage: the amount of shrinkage applied to
subject’s current age (UK Biobank field identifiers 21003, 22147, 3786). For each genetic marker is adaptive to the strength of its association signal in
both eMERGE and the UK Biobank, we are confident in the identification of the GWAS. The pipeline can accommodate diverse underlying genetic archi-
subjects with pediatric-onset asthma, while the true adult-onset asthma with tectures. Linkage disequilibrium (LD) and an LD matrix were determined and
no past medical history of asthma in childhood was unable to be accurately built based on the highest number of ancestry representation in the discovery
determined using electronic medical records. set, which in our case was European. The training process (Fig 1) used the dis-
covery GWAS summary statistics (multiancestral TAGC discovery GWAS),
the reference population (individual-level 1000 Genomes genotype data),
Discovery GWAS for identifying variants to include and the individual-level genotype and phenotype data of target population
in PRS analysis (training dataset in order to tune the hyperparameters of the prediction model
The PRS was built using the GWAS results from a 2018 study published by using CS [auto mode]) so that the pipeline automatically learned the sparse-
the Trans-National Asthma Genetic Consortium (TAGC), which assessed ness of the genetic architecture from data and adjusted for the LD structure
23,948 patients with adult- and pediatric-onset asthma and 118,538 controls.29 accordingly.24
Table E2 in this article’s Online Repository at www.jacionline.org describes We adjusted for confounding effects due to population stratification with a
the studies included in TAGC. The summary statistics from 2,001,280 auto- linear regression model using the 10 principal components of ancestry in all
somal genetic variants that passed quality control filters were accessed through participants.31 After calculating a principal component–adjusted PRS, age and
the GWAS catalog (https://www.ebi.ac.uk/gwas/home, assessed on January 8, sex were used as covariates using a logistic regression fitting model imple-
2021) and were used to calculate the PRS. A total of 985,837 autosomal ge- mented in R version 4.1.0.19 The residuals from this model were used to create
netic variants with minor allele frequencies >1% in the combined training an ancestry-corrected PRS distribution. The distribution of unadjusted
and validation datasets (with cases and controls combined) were identified. compared to the ancestry-adjusted PRSs across the 5 ancestral groups in the
These 985,837 common markers were found in the 1000 Genome reference training and validation cohorts are presented in Fig E1 in this article’s Online
panel, the TAGC, and in the training and validation datasets with genotyping Repository available at www.jacionline.org.
rates of 99.9% in the training and validation 1 datasets. A total of 983,520 of The PRS prediction accuracy and performance was assessed by using
these markers (99.7%) were present in the validation 2 dataset with a total gen- receiver-operating characteristic area under the curve (AUC), odds ratio (OR)
otyping rate of 98%. The complete list of the selected variants, the effect al- per 1 SD, and by variance explained (R2 based on the Pseudo R2 calculation
leles, allele frequencies across ancestral groups, and posterior effect sizes based on the McFadden method as applied in Stata) in logistic regression after
(see below) are included in Table E3 in this article’s Online Repository at accounting for covariates (10 principal components, age, and sex). The me-
www.jacionline.org. dian of the adjusted percentile distribution between cases and controls after
ancestry standardization (ie, mean PRS of 0 and SD of 1 in each group) was
assessed. As our long-term goal is to evaluate this PRS clinically, we selected
Polygenic risk scores the top 5% as a threshold for high risk. This threshold was selected to identify
PRSs were calculated using PRS-CS, a Bayesian polygenic prediction those at highest genetic risk while minimizing the number of people receiving
method that infers posterior effect sizes of genetic variants using GWAS a ‘‘high risk’’ result who would not develop asthma (Fig E2 in this article’s
J ALLERGY CLIN IMMUNOL NAMJOU ET AL 5
VOLUME nnn, NUMBER nn

Online Repository at www.jacionline.org). To measure the discrimination of each decile in the training, validation 1, and validation 2 cohorts
the multiancestral asthma PRS, we report the top 5% of this distribution as are included in Fig E2.
a high polygenic score and report the increased odds of asthma by comparing To confirm that using multiancestral priors did not reduce the
the top 5% compared to both the bottom 5% and bottom 95%. After fitting the performance of the PRS, we calculated PRS for European cohorts
regression model, the marginal effect of sex and ancestry were also evaluated
using posterior effect sizes after training using only the European
using the delta method implemented in Stata. These marginal effects measure
the impact of unit change in a single variable on the prediction of asthma while studies in TAGC (Table E6 in this article’s Online Repository at
all other variables of the adjusted multiancestral asthma PRS are constant. www.jacionline.org) with trends toward better performance
Our primary outcome is predictive value in the pediatric cohorts based on with multiancestral priors. As shown in Table III and Table E7
the plan to use this PRS in a prospective study focused on children. In the in this article’s Online Repository at www.jacionline.org, the
tables in the Online Repository, we also report outcomes in the overall cohort multiancestral PRS performance was consistent across all ances-
because it has additional power and allows us to compare performance in the tries, with better performance in the validation pediatric cohorts.
subset of pediatric individuals. The pediatric-only training dataset demonstrated significant
We also benchmarked the performance of the multiancestral PRS using 2 discrimination using the covariate-adjusted multiancestral PRS
previously published PRS algorithms.32,33 Notably, the number of genetic var- (European: AUC, 0.67 [95% CI, 0.64-0.70], OR per SD, 1.27; Af-
iants included and the populations used in the previously published PRS are
rican: AUC, 0.57 [95% CI, 0.54-0.60], OR per SD, 1.13; and Ad-
different.
mixed American: AUC, 0.68 [95% CI,0.64-0.72], OR per SD,
1.61). These results were replicated in the pediatric validation 1
PheWAS analyses cohorts (European: AUC, 0.60 [95% CI, 0.57-0.63], OR per
To evaluate pleiotropic effects of the multiancestral PRS for asthma against SD, 1.20; African: AUC, 0.61 [95% CI, 0.58-0.63], OR per SD,
other traits, a phenome-wide association study (PheWAS) was performed 1.27; and Admixed American: AUC, 0.64 [95% CI, 0.61-0.68],
using the R PheWAS software package in the training and validation cohorts.34 OR per SD, 1.25) (Table III [pediatric], Table E7 [overall]). The
Briefly, ICD-9 codes were translated into PheWAS codes according to Phe- pediatric validation 2 cohort was used to further replicate the mul-
WAS map.34 Cases and controls were identified based on > _2 occurrences of tiancestral PRS and provided the opportunity to measure the per-
the PheWAS code on different days in the cases and no instances in the con- formance of the multiancestral asthma PRS in Eastern and
trols.34 For each PheWAS code, the asthma PRS was included in a logistic
Southern Asian cohorts (European: AUC, 0.66 [95% CI, 0.65-
regression model adjusted for age, sex, and the 10 principal components.
The OR is based on regression analyses using each phenotype as the dependent 0.67], OR per SD, 1.57; African: AUC, 0.66 [95% CI, 0.63-
variable and adjusted PRS (a quantitative value calculated for each individual) 0.69], OR per SD, 1.43; Admixed American: AUC, 0.70 [95%
as an independent variable. A false discovery rate (FDR) of 0.05 using the CI, 0.67-0.74], OR per SD, 1.63; Eastern Asian: AUC, 0.73
Benjamini–Hochberg procedure was implemented to account for multiple [95% CI, 0.66-0.80], OR per SD, 1.32; and Southern Asian:
testing. AUC, 0.65 [95% CI, 0.62-0.68], OR per SD, 1.32).
We compared the multiancestral asthma PRS from this study to
the 2 previously published PRSs32,33 in our 2 European training
and validation datasets. The number of genetic variants in these
RESULTS PRSs were limited15,22 and were developed based on genetic
The number of participants in this study in each ancestral group studies of European ancestry while the current PRS was devel-
that passed quality control steps (see Methods) are broken down oped based on a multiancestral GWAS. The AUCs in the full,
by age and sex and presented in Table I. In total, we analyzed covariate-adjusted models were lower for the Belsky et al32 study
70,290 participants with asthma and 467,247 controls across the (AUC, 0.59; 95% CI, 0.57-0.61) and Dijk et al33 study (AUC,
multiancestral training, validation 1, and validation 2 datasets. 0.60; 95% CI, 0.59-0.62) compared to the multiancestral PRSs
Each dataset includes participants from each of the 5 superpopu- in our training and validation European cohorts. Furthermore,
lations as defined by PCA of independent genetic variants (Table I the multiancestral PRS outperformed the previous European-
and Table E4 in this article’s Online Repository at www. derived PRS in non-European ancestries (Fig E3 in this article’s
jacionline.org). Online Repository at www.jacionline.org).
An ancestry-specific asthma PRS for asthma is not optimal for In the full model logistic regression analyses of the PRSs,
clinical implementation, because individuals may align with female sex was associated with reduced odds of asthma in
multiple ancestries. Thus, we evaluated a single multiancestral pediatric cohorts across all ancestries (training-pediatric: OR,
asthma PRS that accounted for the underlying ancestral differ- 0.74; 95% CI, 0.65-0.86; P < .0001; validation 1–pediatric: OR,
ences in the PRSs (Fig 1). The ancestry harmonization of the PRS 0.69; 95% CI, 0.61-0.79; P < .0001, and validation 2–pediatric:
distribution was performed to account for the density and range of OR, 0.67; 95% CI, 0.65-0.69; P < .0001). This finding is consis-
each ancestry-specific distribution (Fig E1). After adjustment for tent with pediatric-onset asthma being more common in males
ancestry, the multiancestral AUC for the validation pediatric co- than in females.35,36 To assess its confounding effect, we calcu-
horts was 0.70 (95% CI, 0.69-0.72) and 0.66 (95% CI, 0.65-0.66) lated the marginal effects of sex for prediction probability of
in the pediatric validation 2 cohort (Fig 2, Table II, and Table E4). asthma after fitting the logistic regression in the validation 2
The discrimination of the PRS between the top 5th percentile to cohort as a combined cohort and in each ancestry separately
the bottom 5th percentile was measured in the training (OR, (Fig E4 in this article’s Online Repository at www.jacionline.
2.80; 95% CI, 1.87-4.12), validation 1 (OR, 3.31; 95% CI, org). Indeed, the better predictive probability of asthma from
2.29-4.78) and validation 2 (OR, 5.82; 95% CI, 5.19-6.53), data- the overall PRS model in males compared with in females is
sets as shown in Table II for pediatric cohorts and Table E5 in this consistent with the regression analysis. Similarly, we evaluated
article’s Online Repository at www.jacionline.org for the full da- the marginal effects of ancestry on the multiancestral PRS and
tasets. A comparison of the PRS percentile distribution between found that there was substantial overlap consistent with ancestral
cases and controls is presented in Fig 3. The risk prediction per normalization.
6 NAMJOU ET AL J ALLERGY CLIN IMMUNOL
nnn 2022

FIG 2. Multiancestral PRS performance. Overall adjusted PRS performance are shown for the training
pediatric cohort (A), validation 1 pediatric cohort (B), and the validation 2 pediatric cohort (C). The AUC and
95% CI are shown adjusted for age, sex, and 10 ancestral principal components.

TABLE II. Adjusted multiancestral PRS performance in 3 independent multiancestral pediatric cohorts
Transancestral PRS performance
in pediatric cohorts AUC OR per SD* Pseudo R2 ORy top 5% vs 95% ORz top 5% vs bottom 5%

Training 0.73 (0.71-0.74) 1.21 (1.12-1.30) 0.11 1.84 (1.42-2.39) 2.80 (1.87-4.12)
Validation 1 0.70 (0.69-0.72) 1.22 (1.15-1.30) 0.08 2.16 (1.74-2.67) 3.31 (2.29-4.78)
Validation 2 0.66 (0.65-0.66) 1.59 (1.57-1.62) 0.04 2.37 (2.25-2.49) 5.82 (5.19-6.53)

The overall PRS multiancestry performance with 95% CIs after covariate and ancestry adjustment is presented. The OR of having asthma in patients within the top 5th percentile of
the PRS distribution compared to the remaining 95% are shown with 95% CIs.
*The OR (95% CI) per SD (P < .0001).
 The OR (95% CI) when comparing the top 5% of standardized adjusted PRS distribution against remaining 95% (P < .0001).
àThe OR (95% CI) when comparing the top 5% of standardized adjusted PRS distribution against bottom 5% (P < .0001).

A PheWAS was performed in the combined full training and and white blood cell disorders (Fig 4, Table IV, Table E8 in this
validation cohorts to evaluate potential pleiotropic effects of the article’s Online Repository at www.jacionline.org). Notably, the
multiancestral asthma PRS in this study with other traits. As asthma PRS was more strongly associated with asthma than
expected, this approach confirmed the strong association of the with the related phenotype allergic rhinitis (OR, 1.24; 95% CI,
multiancestral asthma PRS with asthma (OR, 2.71; 95% CI, 2.04- 1.10-1.40; PFDR, 3.21 3 1024) (Tables E8 and E9 in this article’s
3.03; FDR-corrected P [PFDR], 3.71 3 10265) (Table IV). This Online Repository at www.jacionline.org), supporting the
exploratory analysis also identified >300 pleiotropic association phenotype-specificity of the asthma PRS. This approach also de-
effects (PFDR < .05) including positive associations with asthma tected novel negative associations with traits such as hyperlipid-
severity and exacerbation, emphysema, and pulmonary insuffi- emia and hypercholesterolemia (OR, 0.62; 95% CI, 0.55-0.69;
ciency, as well as diabetes, eosinophilic esophagitis, food allergy, PFDR, 6.55 3 10217) (Fig 4, Table IV, Table E8).
J ALLERGY CLIN IMMUNOL NAMJOU ET AL 7
VOLUME nnn, NUMBER nn

participants assembled, we were able to evaluate the performance

of the PRS overall as well as in documented pediatric cases. This
evaluation revealed that our asthma PRS performed better in
children than in the overall cohort of subjects with combined
pediatric- and adult-onset asthma. These findings are consistent
with previous genetic variant-based heritability estimates sup-
porting a larger genetic contribution in children compared with in
adults.14 While we confidently identified individuals with
pediatric-onset asthma, a limitation of this study was our inability
to identify individuals with adult-onset asthma (ie, some adults
with asthma could have developed disease as a child).
Notably, our multiancestral asthma PRS performed better
based on AUCs than prior asthma PRSs, especially when
considering non-European populations. Based on assessment
of the PGS catalog (https://www.pgscatalog.org/) in August
2021, there are currently 2 published studies focusing on PRS
development for asthma alone.32,33 These studies were limited
primarily to European ancestral groups and PRS development
was based on P-value thresholding. In complex diseases with
many modest genetic effects such as asthma, the P-value thresh-
olding methodology can underperform due to the omission of
many variants with weaker phenotype association.30 In contrast,
the Bayesian approach used in this study incorporates genome-
wide variants after considering the underlying LD population
substructure and generates a posterior effect size for all variants
included in the study. The benefits of the Bayesian strategy
relative to other PRS approaches that use effect-size weighted
additive model include30 (1) all association data from a
GWAS are used—including information that is usually not
included in approaches that start with robust genetic associa-
tions; (2) the approach incorporates the differences in LD and
genetic architecture between ancestral groups; and (3) the use
of continuous shrinkage priors allows the model to consider
robust genetic signals with large effect sizes as well as small
effects with less significant association signals. This suggests
that the Bayesian methodology is superior for the development
FIG 3. PRS percentile comparison between pediatric patients with asthma
and controls. The violin plots of the median ancestry standardized PRS
of a PRS. However, because we compared different methodol-
distributions between cases and controls in pediatric training (A), validation ogy and different populations (multiancestral), a broader
1 (B), and validation 2 cohorts. In the boxplot inset, the dot within each box comparison of these methods for the development of multian-
indicates the median score. In the training cohort (A), the median standard- cestral pediatric asthma PRSs is justified.
ized score of cases was 60% versus 47% in controls (P < .0001). In the vali-
The improved performance of our PRS may also be due to both
dation 1 cohort (B), the median standardized score of cases was 58% versus
48% in controls (P < .0001). In the validation 2 cohort (C), the median stan- the approaches used—Bayesian and multiancestral. Because
dardized score of cases was 67% in cases versus 49% in controls (P < .0001). asthma risk varies by ancestry, including individuals from diverse
The top and bottom of the boxes indicate the 75th and 25th percentiles (in- ancestral groups will be essential to ensure the clinical use of
terquartile range), respectively. PRSs does not exacerbate health disparities.37 Other approaches
to develop transancestral and multiancestral PRSs differ by how
they select and weigh genetic risk variants and how they integrate
DISCUSSION genetic data with other clinical and environmental data.38-42
Pediatric asthma affects ;10% of the children in the United While we used PRS-CS, other models use best linear unbiased
States. There is no cure, underscoring the importance of preven- prediction (BLUP) and least absolute shrinkage and section oper-
tion and early identification. In this study, we developed a PRS for ator approaches (LASSO) to estimate genetic effect sizes in joint
asthma using an ancestrally diverse group of individuals and models of multiple variants, and predictions are performed simul-
trained and validated the PRS’s performance using multiple inde- taneously.43-46 There are also numerous methods that use
pendent cohorts, which included pediatric onset as well as any age different approaches to account for differences in LD in individ-
of onset. We demonstrate that our asthma PRS has good discrim- uals of different ancestry.47,48 As statistical methods are devel-
inatory performance in people of diverse ancestries and especially oped to improve multiancestral PRS, multiancestral asthma
children, is more discriminating than prior scores, and reveals po- PRS should be continuously refined with future publications of
tential pleiotropic effects. Taken together, these results support genetic association studies of asthma in larger, admixed popula-
the value of our multiancestral PRS. tions. It is possible that similarly powered ancestry-specific ana-
The PRS performed well across 3 independent datasets and in 5 lyses would identify additional loci with more impactful effect
different ancestral groups. Because of the large number of sizes; however, the results of the multiancestral study might prove
8 NAMJOU ET AL J ALLERGY CLIN IMMUNOL
nnn 2022

TABLE III. Multiancestral asthma PRS performance in 3 independent cohorts

Pediatric cohorts AUC (full model)* Pseudo R2 OR per SDy

European ancestry
Training 0.67 (0.64-0.70) 0.05 1.27 (1.12-1.44)
Validation 1 0.60 (0.57-0.63) 0.02 1.20 (1.08-1.34)
Validation 2 0.66 (0.65-0.67) 0.04 1.57 (1.55-1.60)
African ancestry
Training 0.57 (0.54-0.60) 0.01 1.13 (1.01-1.26)à
Validation 1 0.61 (0.58-0.63) 0.02 1.27 (1.14-1.42)
Validation 2 0.66 (0.63-0.69) 0.03 1.43 (1.24-1.65)
Admixed American ancestry
Training 0.68 (0.64-0.72) 0.07 1.61 (1.28-2.02)
Validation 1 0.64 (0.61-0.68) 0.03 1.25 (1.06-1.47)
Validation 2 0.70 (0.67-0.74) 0.07 1.63 (1.36-1.95)
Eastern Asian ancestry
Validation 2 0.73 (0.66-0.80) 0.08 1.32 (0.99-1.77)à
Southern Asian ancestry
Validation 2 0.65 (0.62-0.68) 0.03 1.32 (1.18-1.48)
The standardized PRS distribution was evaluated in a logistic regression model adjusted for age and sex as well as the 10 principal components that informed the 5
superpopulations. The AUC and 95% CI, OR per SD, and the total variation explained (pseudo R2) are shown.
*AUC full model includes age, sex, and 10 principal components.
 The OR per SD of PRS distribution (logistic regression P < .0001).
àFor the African Ancestry eMERGE datasets 1-pediatric and Eastern Asian UK Biobank-pediatric cohorts, P 5 .03 and .07, respectively.

TABLE IV. Selected results of PheWAS applying multiancestral asthma PRS to training and validation 1 cohorts
Description* PheWAS-code Case Control ORy 95% CI PFDR

Positively associated with asthma PRS

Asthma 495 12,963 70,020 2.71 2.41-3.02 3.17 3 10265
Asthma with exacerbation 495.2 3,043 70,020 4.72 3.74-5.95 3.39 3 10239
Emphysema 508 8,276 73,217 1.78 1.56-2.05 1.22 3 10216
Chronic obstructive asthma 495.1 1,433 70,020 3.23 2.35-4.45 4.90 3 10213
Type 1 diabetes 250.1 4,283 69,292 1.91 1.58-2.30 1.16 3 10211
Chronic airway obstruction 496 8,056 70,020 1.56 1.36-1.80 5.57 3 10210
Respiratory failure 509 6,029 73,217 1.61 1.37-1.88 3.05 3 1029
Wheezing 512.1 2,234 47,495 2.17 1.68-2.82 4.37 3 1029
Diabetes mellitus 250 19,764 69,292 1.34 1.21-1.48 9.36 3 1029
Eosinophilic esophagitis 530.15 607 62,044 3.85 2.38-6.20 3.40 3 1028
Type 2 diabetes 250.2 19,137 69,292 1.32 1.19-1.46 9.78 3 1028
Diseases of white blood cells 288 4,694 77,026 1.61 1.35-1.91 1.14 3 1027
Allergic reaction to food 930 1,688 65,842 2.25 1.66-3.04 1.49 3 1027
Negatively associated with asthma PRS
Postmenopausal disorders 627 11,313 75,725 0.55 0.48-0.63 8.00 3 10218
Peripheral enthesopathies 726 16,894 66,608 0.64 0.57-1.71 6.31 3 10217
Hypercholesterolemia 272.11 19,899 51,696 0.62 0.55-0.69 6.55 3 10217
Hematuria 593 7,632 68,213 0.55 0.48-0.64 2.50 3 10216
Benign neoplasm of skin 216 11,233 79,947 0.63 0.56-0.71 1.86 3 10214
Disorders of lipid metabolism 272 41,493 51,696 0.73 0.67-0.80 1.04 3 10211
Hyperlipidemia 272.1 41,299 51,696 0.73 0.67-0.80 1.11 3 10211
Disorders of synovium 727 9,291 66,608 0.64 0.56-1.73 3.65 3 10211
Cataract 366 14,306 81,833 0.67 0.60-0.76 5.14 3 10211
Carbohydrate transport disorder 271 1,410 97,301 0.36 0.27-0.49 1.34 3 10210
Disaccharide malabsorption 271.3 1,302 97,301 0.35 0.25-0.48 1.40 3 10210
Elevated prostate specific antigen 796 2,746 84,778 0.53 0.41-0.67 2.92 3 1027

*Selected results at PFDR < .05. The complete lists of traits are included in Table E5.
 OR < 1 indicates negative association of trait with asthma PRS.

to be more broadly useful when applied to a heterogenous popu- While TAGC might be the largest and most diverse meta-
lation, such as the type of people who go to a primary care setting. analysis from the perspective of genetic ancestry, there are several
Furthermore, a major strength of our multiancestral PRS is that a limitations to consider in the context of the multiancestral asthma
single PRS is applicable to all ancestral groups. Thus, clinicians PRS. Specifically, if there are differences in the genetic etiology
and researchers will not be required to a priori assign an individ- by age of onset of asthma, the TAGC is not composed of a
ual to an ancestral group. majority of pediatric cases. Furthermore, the degree to which
J ALLERGY CLIN IMMUNOL NAMJOU ET AL 9
VOLUME nnn, NUMBER nn

FIG 4. A plot of PheWAS analysis of the asthma PRS within the combined training and validation cohorts.
Manhattan plots of phenome-wide association analyses with phecodes (x-axis) and FDR-corrected PheMap
phenotype probability (y-axis). The red line indicates the Bonferroni level of significance (5.0 3 1025).

pediatric-onset asthma cases are represented in the meta-analysis that asthma is associated with worse lipid profiles at a phenotypic
differs greatly based on the race and ethnicity of the component level.54 One possible explanation for this discrepancy is that the
studies. Thus, while our PRS performed well for pediatric asthma, use of inhaled corticosteroids (a primary treatment of asthma) is
continued refinement of the PRS is warranted with special focus associated with a worse lipid profile in adults.55 It is also possible
on pediatric cases and capturing more ancestral diversity in the that environmental factors, such as pollution, are driving
discovery data. increased risks for asthma as well as poor lipid profiles.56-58
To understand how underlying asthma risk may relate to a While the PheWAS analysis suggests potential pleiotropic effects
variety of conditions, we performed a PheWAS analysis to (both increased and decreased risk of other diseases), additional
examine positive or negative association of other conditions work is required to clarify these relationships.
with the asthma PRS. Notably, we measured a stronger effect size This study is an initial step toward developing a multiancestral
for asthma with exacerbation (PheWAS code 495.2; OR, 4.72) PRS to be used in the eMERGE IV network (https://www.
and chronic asthma (PheWAS code 495.1; OR, 3.23) than asthma genome.gov/Funded-Programs-Projects/Electronic-Medical-
(PheWAS code 495; OR, 2.71) (Table IV). In the case of this Phe- Records-and-Genomics-Network-eMERGE) prospective inter-
WAS, the OR is based on regression analyses using each pheno- vention cohort study beginning in 2022. The eMERGE IV study
type as the dependent variable and adjusted PRS as an will enroll 5000 children (underrepresented, non-European
independent variable. These findings suggest that the asthma preferred) across the 10 eMERGE clinical sites. Multiancestral
PRS may be useful not only for prevention but also to help clini- PRS for 4 phenotypes (asthma, obesity, type I diabetes, type II
cians select treatment strategies for children already diagnosed diabetes) will be calculated and returned to participants’ parents
with asthma. These findings require replication, as we do not and primary care providers. Parents and primary care providers
have sufficiently uniform measurements in the subjects in the of children with a high-risk asthma PRS (top 5th percentile)
training, validation 1, and validation 2 cohorts to identify whether will also receive guideline-informed health recommenda-
PRS in patients is associated with disease severity. However, these tions.59,60 We seek to understand how primary care providers, pa-
findings provide rationale for a controlled prospective study to tients, and patient families change their behavior in reaction to a
test the association of asthma severity with PRS. Not surprisingly, top 5th percentile asthma PRS. The prospective study will collect
other atopic conditions such as eosinophilic esophagitis and food family history information and clinical factors to display along
allergy were positively associated with the asthma PRS, as these with an asthma high-risk PRS that providers can use to calculate
conditions have been noted to have increased rates in patients the Pediatric Asthma Risk Score. Recent studies have validated
with asthma.49,50 However, we also found positive associations the Pediatric Asthma Risk Score as a tool to predict asthma devel-
with type I and type II diabetes. Intriguingly, several studies opment in young children based on family history, eczema before
have reported a higher-than-expected co-occurrence of asthma age 3 years, wheezing apart from colds before age 3, African
and type I diabetes, supporting a partially shared genetic etiol- American ancestry, and sensitization to > _2 food or aero
ogy.51-53 We also found a surprising negative association between allergens.7
our asthma PRS and hypercholesterolemia/hyperlipidemia. In A previous group suggested that their pediatric asthma PRS did
contrast to our findings, previous meta-analyses have reported not provide any discriminatory value above clinical risk factors.33
10 NAMJOU ET AL J ALLERGY CLIN IMMUNOL
nnn 2022

Yet, assessing clinically predictive factors can be challenging due 12. Ferreira MA, Vonk JM, Baurecht H, Marenholz I, Tian C, Hoffman JD, et al.
to the lack of consistent capturing of such data in cohorts with suf- Shared genetic origin of asthma, hay fever and eczema elucidates allergic disease
biology. Nat Genet 2017;49(12):1752-7.
ficient statistical power. Even if the PRS and the clinical risk pre- 13. Torkamani A, Wineinger NE, Topol EJ. The personal and clinical utility of poly-
diction substantially overlap in who is identified at risk, the genic risk scores. Nat Rev Genet 2018;19(9):581-90.
development of such a PRS would still be of value. This is 14. Pividori M, Schoettler N, Nicolae DL, Ober C, Im HK. Shared and distinct genetic
because not all children undergo allergic sensitization testing risk factors for childhood-onset and adult-onset asthma: genome-wide and
transcriptome-wide studies. Lancet Respir Med 2019;7(6):509-22.
before age 3 years, and clinical presentations such as eczema 15. Vergara C, Murray T, Rafaels N, Lewis R, Campbell M, Foster C, et al. African
and wheezing without a cold may not be recognized by parents. ancestry is a risk factor for asthma and high total IgE levels in African admixed
Thus, alternative strategies for risk stratification are needed. populations. Genet Epidemiol 2013;37(4):393-401.
Notably, there are already preventive measures that can be prior- 16. Flores C, Ma SF, Pino-Yanes M, Wade MS, Perez-Mendez L, Kittles RA, et al. Af-
itized if a child is identified as high risk.61 For example, once a rican ancestry is associated with asthma risk in African Americans. PLoS One
2012;7(1):e26807.
child is identified as high risk for asthma, families can be coun- 17. Brim SN, Rudd RA, Funk RH, Callahan DB. Asthma prevalence among US chil-
seled to limit smoke exposure, identify and avoid known aller- dren in underrepresented minority populations: American Indian/Alaska Native,
gens, prevent viral infection, and limit dust and mold Chinese, Filipino, and Asian Indian. Pediatrics 2008;122(1):e217-22.
exposure.4,62,63 However, we recognize that the use of genetics 18. Pino-Yanes M, Thakur N, Gignoux CR, Galanter JM, Roth LA, Eng C, et al. Ge-
netic ancestry influences asthma susceptibility and lung function among Latinos.
alone to predict asthma has inherent limitations, because both J Allergy Clin Immunol 2015;135(1):228-35.
genes and environment contribute to asthma risk. 19. Fishe J, Zheng Y, Lyu T, Bian J, Hu H. Environmental effects on acute exacerba-
A long-term goal beyond eMERGE IV will be to create and tions of respiratory diseases: a real-world big data study. Sci Total Environ 2022;
validate a combined/integrated predictive model that includes 806(Pt 1):150352.
20. Kaur S, Rosenstreich D, Cleven KL, Spivack S, Grizzanti J, Reznik M, et al. Severe
genetic, family history, clinical, and environmental risk factors.
asthma in adult, inner-city predominantly African-American and Latinx popula-
Data collected during the eMERGE IV prospective study will tion: demographic, clinical and phenotypic characteristics. J Asthma 2021 Dec;2:
provide essential elements toward our long-term goal. In addition 1-11, [Online ahead of print.]
to genotype data, family history information, and relevant clinical 21. Almoguera B, Vazquez L, Mentch F, Connolly J, Pacheco JA, Sundaresan AS, et al.
factors, we will also have geocodes to develop a combined/ Identification of four novel loci in asthma in European American and African
American populations. Am J Respir Crit Care Med 2017;195(4):456-63.
integrated predictive model. 22. Stanaway IB, Hall TO, Rosenthal EA, Palmer M, Naranbhai V, Knevel R, et al. The
In conclusion, we present the development and validation of a eMERGE genotype set of 83,717 subjects imputed to ;40 million variants genome
pediatric asthma PRS that performs effectively across ancestries wide and association with the herpes zoster medical record phenotype. Genet Epi-
in 3 independent cohorts and identifies novel pleiotropic relation- demiol 2019;43(1):63-81.
23. Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. UK Biobank: an
ships. In the future, this PRS will be used in the context of
open access resource for identifying the causes of a wide range of complex diseases
additional demographic and clinical risk factors as part of a of middle and old age. PLoS Med 2015;12(3):e1001779.
genome-informed risk assessment to help families of children at 24. Zhu Z, Lee PH, Chaffin MD, Chung W, Loh PR, Lu Q, et al. A genome-wide cross-
high risk for asthma take preventive steps to avoid disease. trait analysis from UK Biobank highlights the shared genetic architecture of
asthma and allergic diseases. Nat Genet 2018;50(6):857-64.
25. Zhu Z, Zhu X, Liu CL, Shi H, Shen S, Yang Y, et al. Shared genetics of asthma and
Clinical implications: This PRS will be used to identify children mental health disorders: a large-scale genome-wide cross-trait analysis. Eur Respir
at increased risk for asthma across a multisite multiancestral J 2019;54(6):1901507.
prospective intervention cohort study. 26. Zhang D, Dey R, Lee S. Fast and robust ancestry prediction using principal compo-
nent analysis. Bioinformatics 2020;36(11):3439-46.
27. 1000 Genomes Project Consortium, Auton A, Brooks LD, Durbin RM, Garrison
REFERENCES EP, Kang HM, et al. A global reference for human genetic variation. Nature
1. Holgate ST, Wenzel S, Postma DS, Weiss ST, Renz H, Sly PD. Asthma. Nat Rev 2015;526(7571):68-74.
Dis Primers 2015;1:15025. 28. Chang CC, Chow CC, Tellier LC, Vattikuti S, Purcell SM, Lee JJ. Second-gener-
2. Kuruvilla ME, Lee FE, Lee GB. Understanding asthma phenotypes, endotypes, and ation PLINK: rising to the challenge of larger and richer datasets. Gigascience
mechanisms of disease. Clin Rev Allergy Immunol 2019;56(2):219-33. 2015;4:7.
3. Ramsahai JM, Hansbro PM, Wark PAB. Mechanisms and management of asthma 29. Demenais F, Margaritte-Jeannin P, Barnes KC, Cookson WOC, Altmuller J, Ang
exacerbations. Am J Respir Crit Care Med 2019;199(4):423-32. W, et al. Multiancestry association study identifies new asthma risk loci that coloc-
4. Castillo JR, Peters SP, Busse WW. Asthma exacerbations: pathogenesis, preven- alize with immune-cell enhancer marks. Nat Genet 2018;50(1):42-53.
tion, and treatment. J Allergy Clin Immunol Pract 2017;5(4):918-27. 30. Ge T, Chen CY, Ni Y, Feng YA, Smoller JW. Polygenic prediction via Bayesian
5. Wiksten J, Toppila-Salmi S, Makela M. Primary prevention of airway allergy. Curr regression and continuous shrinkage priors. Nat Commun 2019;10(1):1776.
Treat Options Allergy 2018;5(4):347-55. 31. Khera AV, Chaffin M, Zekavat SM, Collins RL, Roselli C, Natarajan P, et al.
6. Castro-Rodriguez JA, Holberg CJ, Wright AL, Martinez FD. A clinical index to Whole-genome sequencing to characterize monogenic and polygenic contributions
define risk of asthma in young children with recurrent wheezing. Am J Respir in patients hospitalized with early-onset myocardial infarction. Circulation 2019;
Crit Care Med 2000;162(4 Pt 1):1403-6. 139(13):1593-602.
7. Biagini Myers JM, Schauberger E, He H, Martin LJ, Kroner J, Hill GM, et al. A 32. Belsky DW, Sears MR, Hancox RJ, Harrington H, Houts R, Moffitt TE, et al. Poly-
Pediatric Asthma Risk Score to better predict asthma development in young chil- genic risk and the development and course of asthma: an analysis of data from a
dren. J Allergy Clin Immunol 2019;143(5):1803-10.e2. four-decade longitudinal study. Lancet Respir Med 2013;1(6):453-61.
8. Schoettler N, Rodriguez E, Weidinger S, Ober C. Advances in asthma and allergic 33. Dijk FN, Folkersma C, Gruzieva O, Kumar A, Wijga AH, Gehring U, et al. Genetic
disease genetics: Is bigger always better? J Allergy Clin Immunol 2019;144(6): risk scores do not improve asthma prediction in childhood. J Allergy Clin Immunol
1495-506. 2019;144(3):857-60.e7.
9. Sheth KK, Lemanske RF Jr. Pathogenesis of asthma. Pediatrician 1991;18(4): 34. Carroll RJ, Bastarache L, Denny JC. R PheWAS: data analysis and plotting tools
257-68. for phenome-wide association studies in the R environment. Bioinformatics
10. Thomsen SF. Exploring the origins of asthma: lessons from twin studies. Eur Clin 2014;30(16):2375-6.
Respir J 2014;1(suppl 1); https://doi.org/10.3402/ecrj.v1.25535. 35. Frohlich M, Pinart M, Keller T, Reich A, Cabieses B, Hohmann C, et al. Is there a
11. Buniello A, MacArthur JAL, Cerezo M, Harris LW, Hayhurst J, Malangone C, sex-shift in prevalence of allergic rhinitis and comorbid asthma from childhood to
et al. The NHGRI-EBI GWAS catalog of published genome-wide association adulthood? A meta-analysis. Clin Transl Allergy 2017;7:44.
studies, targeted arrays and summary statistics 2019. Nucleic Acids Res 2019; 36. Pinart M, Keller T, Reich A, Frohlich M, Cabieses B, Hohmann C, et al.
47(D1):D1005-12. Sex-related allergic rhinitis prevalence switch from childhood to adulthood: a
J ALLERGY CLIN IMMUNOL NAMJOU ET AL 11
VOLUME nnn, NUMBER nn

systematic review and meta-analysis. Int Arch Allergy Immunol 2017;172(4): 51. Metsala J, Lundqvist A, Virta LJ, Kaila M, Gissler M, Virtanen SM, et al. The as-
224-35. sociation between asthma and type 1 diabetes: a paediatric case-cohort study in
37. Oh SS, Galanter J, Thakur N, Pino-Yanes M, Barcelo NE, White MJ, et al. Diver- Finland, years 1981-2009. Int J Epidemiol 2018;47(2):409-16.
sity in clinical and biomedical research: a promise yet to be fulfilled. PLoS Med 52. Hsiao YT, Cheng WC, Liao WC, Lin CL, Shen TC, Chen WC, et al. Type 1 dia-
2015;12(12):e1001918. betes and increased risk of subsequent asthma: a nationwide population-based
38. Rudolph A, Song M, Brook MN, Milne RL, Mavaddat N, Michailidou K, et al. cohort study. Medicine (Baltimore) 2015;94(36):e1466.
Joint associations of a polygenic risk score and environmental risk factors for 53. Smew AI, Lundholm C, Savendahl L, Lichtenstein P, Almqvist C. Familial coag-
breast cancer in the Breast Cancer Association Consortium. Int J Epidemiol gregation of asthma and type 1 diabetes in children. JAMA Netw Open 2020;3(3):
2018;47(2):526-36. e200834.
39. Khera AV, Chaffin M, Aragam KG, Haas ME, Roselli C, Choi SH, et al. Genome- 54. Peng J, Huang Y. Meta-analysis of the association between asthma and serum
wide polygenic scores for common diseases identify individuals with risk equiva- levels of high-density lipoprotein cholesterol and low-density lipoprotein choles-
lent to monogenic mutations. Nat Genet 2018;50(9):1219-24. terol. Ann Allergy Asthma Immunol 2017;118(1):61-5.
40. Mars N, Koskela JT, Ripatti P, Kiiskinen TTJ, Havulinna AS, Lindbohm JV, et al. 55. Fessler MB, Massing MW, Spruell B, Jaramillo R, Draper DW, Madenspacher JH,
Polygenic and clinical risk scores and their impact on age at onset and prediction of et al. Novel relationship of serum cholesterol with asthma and wheeze in the
cardiometabolic diseases and common cancers. Nat Med 2020;26(4):549-57. United States. J Allergy Clin Immunol 2009;124(5):967-74.e1-15.
41. Mak TSH, Porsch RM, Choi SW, Zhou X, Sham PC. Polygenic scores via penal- 56. McGuinn LA, Schneider A, McGarrah RW, Ward-Caviness C, Neas LM, Di
ized regression on summary statistics. Genet Epidemiol 2017;41(6):469-80. Q, et al. Association of long-term PM2.5 exposure with traditional and novel
42. Vilhjalmsson BJ, Yang J, Finucane HK, Gusev A, Lindstrom S, Ripke S, et al. lipid measures related to cardiovascular disease risk. Environ Int 2019;122:
Modeling linkage disequilibrium increases accuracy of polygenic risk scores. 193-200.
Am J Hum Genet 2015;97(4):576-92. 57. Keet CA, Keller JP, Peng RD. Long-term coarse particulate matter exposure is
43. Speed D, Balding DJ. MultiBLUP: improved SNP-based prediction for complex associated with asthma among children in Medicaid. Am J Respir Crit Care Med
traits. Genome Res 2014;24(9):1550-7. 2018;197(6):737-46.
44. Zhou X, Carbonetto P, Stephens M. Polygenic modeling with Bayesian sparse 58. Brunst KJ, Ryan PH, Brokamp C, Bernstein D, Reponen T, Lockey J, et al. Timing
linear mixed models. PLoS Genet 2013;9(2):e1003264. and duration of traffic-related air pollution exposure and the risk for childhood
45. Shi J, Park JH, Duan J, Berndt ST, Moy W, Yu K, et al. Winner’s curse correction wheeze and asthma. Am J Respir Crit Care Med 2015;192(4):421-7.
and variable thresholding improve performance of polygenic risk modeling based 59. Expert Panel Working Group of the National Heart, Lung, and Blood Institute
on genome-wide association study summary-level data. PLoS Genet 2016;12(12): administered and coordinated National Asthma Education and Prevention Program
e1006493. Coordinating Committee, Cloutier MM, Baptist AP, Blake KV, Brooks EG, Bryant-
46. Lello L, Avery SG, Tellier L, Vazquez AI, de Los Campos G, Hsu SDH. Accurate Stephens T, et al. 2020 focused updates to the asthma management guidelines: a
genomic prediction of human height. Genetics 2018;210(2):477-97. report from the National Asthma Education and Prevention Program Coordinating
47. Amariuta T, Ishigaki K, Sugishita H, Ohta T, Koido M, Dey KK, et al. Committee Expert Panel Working Group. J Allergy Clin Immunol 2020;146(6):
Improving the trans-ancestry portability of polygenic risk scores by prioritizing 1217-70.
variants in predicted cell-type-specific regulatory elements. Nat Genet 2020; 60. Cloutier MM, Teach SJ, Lemanske RF Jr, Blake KV. The 2020 focused updates to
52(12):1346-54. the NIH asthma management guidelines: key points for pediatricians. Pediatrics
48. Marquez-Luna C, Loh PR, South Asian Type 2 Diabetes Consortium, SIGMA Type 2021;147(6):e2021050286.
2 Diabetes Consortium, Price AL. Multiethnic polygenic risk scores improve risk 61. Tackett AP, Farrow M, Kopel SJ, Coutinho MT, Koinis-Mitchell D, McQuaid EL.
prediction in diverse populations. Genet Epidemiol 2017;41(8):811-23. Racial/ethnic differences in pediatric asthma management: the importance of
49. Gonzalez-Cervera J, Arias A, Redondo-Gonzalez O, Cano-Mollinedo MM, Terree- asthma knowledge, symptom assessment, and family-provider collaboration.
horst I, Lucendo AJ. Association between atopic manifestations and eosinophilic J Asthma 2021;58(10):1395-406.
esophagitis: a systematic review and meta-analysis. Ann Allergy Asthma Immunol 62. Elenius V, Jartti T. Vaccines: could asthma in young children be a preventable dis-
2017;118(5):582-90.e2. ease? Pediatr Allergy Immunol 2016;27(7):682-6.
50. Foong RX, du Toit G, Fox AT. Asthma, food allergy, and how they relate to each 63. Szefler SJ. Advances in pediatric asthma in 2012: moving toward asthma preven-
other. Front Pediatr 2017;5:89. tion. J Allergy Clin Immunol 2013;131(1):36-46.