1784 Diabetes Care Volume 42, September 2019

Jee-Young Moon,1 Tin L. Louie,2 A Genome-Wide Association Deepti Jain,2 Tamar Sofer,2,3 fi – Claudia Schurmann,4 Jennifer E. Below,5 Study Identi es Blood Disorder Chao-Qiang Lai,6 M. Larissa Aviles-Santa,7 fl Gregory A. Talavera,8 Caren E. Smith,6 Related Variants In uencing Lauren E. Petty,5 Erwin P. Bottinger,9 Yii-Der Ida Chen,10 Kent D. Taylor,10 Hemoglobin A1c With Implications Martha L. Daviglus,11 Jianwen Cai,12 Tao Wang,1 Katherine L. Tucker,13 for Glycemic Status in U.S. JoseM.Ordov´ as,´ 6,14 Craig L. Hanis,5 Ruth J.F. Loos,4 Neil Schneiderman,15 Hispanics/Latinos Jerome I. Rotter,10 Robert C. Kaplan,1,16 1 Diabetes Care 2019;42:1784–1791 | https://doi.org/10.2337/dc19-0168 and Qibin Qi

OBJECTIVE

We aimed to identify hemoglobin A1c (HbA1c)-associated genetic variants and examine their implications for glycemic status evaluated by HbA1c in U.S. Hispanics/ Latinos with diverse genetic ancestries.

RESEARCH DESIGN AND METHODS

We conducted a genome-wide association study (GWAS) of HbA1c in 9,636 U.S. Hispanics/Latinos without diabetes from the Hispanic Community Health Study/ Study of Latinos, followed by a replication among 4,729 U.S. Hispanics/Latinos from three independent studies.

RESULTS Our GWAS and replication analyses showed 10 previously known and novel loci 28 1 associated with HbA1c at genome-wide significance levels (P < 5.0 3 10 ). In Department of Epidemiology and Population particular, two African ancestry–specific variants, HBB-rs334 and G6PD-rs1050828, Health, Albert Einstein College of Medicine, G6PD fi Bronx, NY which are causal mutations for sickle cell disease and de ciency, respectively, 2Department of Biostatistics, University of had ∼10 times larger effect sizes on HbA1c levels (b = 20.31% [23.4 mmol/mol]) Washington, Seattle, WA and 20.35% [23.8 mmol/mol] per minor allele, respectively) compared with other 3Division of Sleep and Circadian Disorders, De- partment of Medicine, Brigham and Women’s HbA1c-associated variants (0.03–0.04% [0.3–0.4 mmol/mol] per allele). A novel – fi HBM Hospital, Boston, MA Amerindian ancestry speci c variant, -rs145546625, was associated with 4The Charles Bronfman Institute for Personalized HbA1c and hematologic traits but not with fasting glucose. The prevalence of Medicine, Icahn School of Medicine at Mount hyperglycemia (prediabetes and diabetes) defined using fasting glucose or oral Sinai, New York, NY 5 glucose tolerance test 2-h glucose was similar between carriers of HBB-rs334 or Human Genetics Center, University of Texas Health Science Center at Houston, Houston, TX

CARDIOVASCULAR AND METABOLIC RISK G6PD -rs1050828 HbA1c-lowering alleles and noncarriers, whereas the prevalence 6Jean Mayer USDA Human Nutrition Research of hyperglycemia defined using HbA1c was significantly lower in carriers than in Center on Aging, Tufts University, Boston, MA 7National Heart, Lung, and Blood Institute, Na- noncarriers (12.2% vs. 28.4%, P < 0.001). After recalibration of the HbA1c level taking HBB-rs334 and G6PD-rs1050828 into account, the prevalence of hypergly- tional Institutes of Health, Bethesda, MD 8Graduate School of Public Health, San Diego cemia in carriers was similar to noncarriers (31.3% vs. 28.4%, P = 0.28). State University, San Diego, CA 9Department of Preventive Medicine, Icahn CONCLUSIONS School of Medicine at Mount Sinai, New York, NY This study in U.S. Hispanics/Latinos found several ancestry-specific alleles asso- 10Institute for Translational Genomics and Pop- ciated with HbA through erythrocyte-related rather than glycemic-related path- ulation Sciences, Harbor-UCLA Medical Center, 1c Torrance, CA ways. The potential influences of these nonglycemic-related variants need to be 11Institute for Minority Health Research, Univer- considered when the HbA1c test is performed. sity of Illinois at Chicago, Chicago, IL care.diabetesjournals.org Moon and Associates 1785

Glycated hemoglobin (HbA1c)results Hispanics/Latinos. Understanding genetic from this ongoing study have been de- from nonenzymatic and mostly irrevers- determinants of HbA1 in this population is scribed elsewhere (20). The BioMe Biobank ible chemical modification by glucose of of public health and clinical importance is an ongoing hospital- and outpatient- hemoglobin molecules carried in eryth- because U.S. Hispanics/Latinos, the based population research study that has rocytes. HbA1c reflects the average con- largest minority group in the U.S., are enrolled more than 34,000 participants centration of blood glucose over the disproportionally affected by diabetes since September 2007. This is a electronic average life span of an erythrocyte (;3 (14) and also have poorer diabetes man- medical records–linked biobank that in- months in humans) and indicates glycemic agement compared with non-Hispanic tegrates research data and clinical care status over a longer period compared with whites (15). Moreover, the diverse U.S. information for consented patients at fasting glucose (1). Thus, HbA1c is used both Hispanic/Latino population, admixture of the Mount Sinai Medical Center, which as a measure of glycemic control and as African, European, and Amerindian ances- serves diverse local communities of upper a diagnostic criterion for diabetes (2). tries, may offer opportunities to identify Manhattan (21). fi HbA1c levels are heritable, with a her- novel ancestry-speci c alleles affecting The analysis excluded participants itability of ;50% (3). Previous genome- HbA1c levels (16). Therefore, this study with diabetes (self-reported, on antihy- wide association studies (GWAS) have conducted a GWAS of HbA1c in U.S. perglycemic medications, HbA1c $6.5% identified several diabetes- and glucose Hispanics/Latinos of diverse back- [48 mmol/mol], fasting glucose $200 –related genetic loci associ- grounds using data from the Hispanic mg/dL, or glucose $200 mg/dL after ated with HbA1c (4–7). In addition, GWAS Community Health Study/Study of Lati- oral glucose tolerance test [OGTT]), have identified several erythrocyte- nos (HCHS/SOL) and replicated novel those with self-reported history of ma- related genetic loci that are associated genetic variants associated with HbA1c jor blood abnormalities (self-reported, levels using data from other Hispanic/ if known), and those who had received with HbA1c (4,7–9). HbA1c levels attributed to nonglycemic-related genetic variants Latino cohorts. We also examined the a blood transfusion 3 months before may not reflect glycemic status, and potential implications of nonglycemic- HbA1c measures. The GWAS of HbA1c this implication has been noted in di- related HbA1c variants for glycemic status levels described here included 9,636 abetes screening and diagnosis (4,7,9,10). evaluated using HbA1c. U.S. Hispanics/Latinos from the HCHS/ For example, the common G6PD variant SOL (discovery study) and 4,777 U.S. rs1050828, which affects red blood cell RESEARCH DESIGN AND METHODS Hispanics/Latinos from three replication (RBC) life span, was recently found to be studies, the SCHS (n = 395), the BPRHS Study Participants (n = 832), and the BioMe (n = 3,550). associated with lower HbA1c in African The HCHS/SOL is a population-based Characteristics of study participants Americans, and it was estimated that study of 16,415 Hispanic/Latino adults, are reported in Supplementary Table 1. 650,000 African Americans with diabe- aged 18–74 years, living in four U.S. The study was approved by the Institu- tes would be missed when screened by metropolitan areas (Bronx, NY; Chicago, tional Review Boards at all participating HbA1c if this genetic information was IL; Miami, FL; and San Diego, CA) (17,18). institutions, and all participants gave not taken into account (10). Another A comprehensive battery of interviews written informed consent. study reported that African Americans relating to personal and family character- with sickle cell trait, who are heterozygous istics and health status and behaviors, for an abnormal hemoglobin allele and and a clinical assessment with a blood HbA1c, Plasma Glucose, Liver, and usually have no symptoms of sickle cell draw, were conducted at an in-person Hematologic Measures disease, had lower HbA1c levels compared baseline clinic visit during 2008–2011. The In the HCHS/SOL, HbA1c was measured in with those without sickle cell trait (11). Starr County Health Study (SCHS) is a EDTA whole blood using a Tosoh G7 Prior GWAS of HbA1c levels have been population-based study of 1,980 Mexican- automated high-performance liquid largely conducted in populations of Eu- Americanadults (aged $20 years) in Starr chromatography analyzer (Tosoh Biosci- ropean ancestry (4,5,12) and East Asians County, TX. Survey collection has been ence, San Francisco, CA). In the SCHS, (6,7,13). A recent transethnic genome- previously described (19). The Boston HbA1c was measured in EDTA whole wide meta-analysis included individuals PuertoRicanHealthStudy(BPRHS)isa blood using the DCA Vantage Analyzer from other ethnic groups (e.g., African longitudinal cohort of 1,500 Puerto Rican point of care device (Siemens, Malvern, American, South Asian)(10), but noGWAS adults (aged 45–75 years) living in the PA) following standard protocols. In of HbA1c levels has been conducted in U.S. greater Boston, MA, area. Baseline data the BPRHS, a Tosoh G7 automated

12Department of Biostatistics and Collaborative 15Department of Psychology, University of Mi- This article contains Supplementary Data online Studies Coordinating Center, University of North ami, Miami, FL at http://care.diabetesjournals.org/lookup/suppl/ Carolina, Chapel Hill, NC 16Public Health Sciences Division, Fred Hutchin- doi:10.2337/dc19-0168/-/DC1. 13 Department of Biomedical and Nutritional son Cancer Research Center, Seattle, WA © 2019 by the American Diabetes Association. Sciences, University of Massachusetts Lowell, Corresponding author: Qibin Qi, qibin.qi@ Readers may use this article as long as the work Lowell, MA einstein.yu.edu is properly cited, the use is educational and not 14 IMDEA Food Institute, Campus de Excelencia for profit, and the work is not altered. More infor- Internacional Universidad Autonoma´ de Madrid– Received 24 January 2019 and accepted 24 May 2019 mation is available at http://www.diabetesjournals Consejo Superior de Investigaciones Cient´ıficas, .org/content/license. Madrid, Spain 1786 HbA1c Variants and Glycemic Status in Hispanics Diabetes Care Volume 42, September 2019

high-performance liquid chromatography the Illumina OmniExpressExome (BioMe- have two copies of the A allele. A survey analyzer was used to measure HbA1c.In OMNI) (n = 1,867) or the Illumina Multi- logistic regression was used to compare the BioMe,dataonHbA1c were extracted Ethnic Global BeadChip (BioMe-MEGA) the prevalence of hyperglycemia defined from electronic medical records of par- array (n = 1,683) and imputed to the using fasting glucose, 2-h glucose, or HbA1c ticipants at the time of enrollment. In 1000G phase 3 reference data. between carriers and noncarriers of HBB- the HCHS/SOL, plasma glucose (fasting rs334 T allele or G6PD-rs1050828 A allele. and 2 h OGTT glucose) was measured Statistical Analysis We also constructed an unweighted using a hexokinase enzymatic method Linear mixed-effects regressions were genetic risk score (GRS) based on five (Roche Diagnostics Corporation, Indian- used to test genome-wide SNP-HbA1c SNPs (ANK1-rs4737010, HK1-rs72805692, apolis, IN). Hemogram and platelet associations among 9,636 individuals TMPRSS6-rs855791, HBM-rs145546625, count were measured using a Sysmex without diabetes using data from and TMC6-rs2748424) that might be as- XE-2100 instrument (Sysmex America, HCHS/SOL. Correlations between individ- sociated with HbA1c through the eryth- Mundelein, IL). Serum iron and unsatu- uals were accounted for by incorporating rocytic pathway according to a previous rated iron binding capacity (UIBC) were covariance matrices corresponding to ge- GWAS of HbA1c (10), by summing the measured on a Roche Modular P chem- netic relatedness (kinship), household, HbA1c-raising alleles. The associations of istry analyzer using a Fe reagent kit and and census block group as random effects. this GRS with HbA1c level and prevalence a UIBC reagent kit, respectively (Roche The model also included field center, of hyperglycemia were examined, taking Diagnostics), and was measured age,sex,thefirst five principal components into account the complex design of the inserumwithRochereagents on a Cobas to adjust for ancestry (16), and sampling HCHS/SOL study. 6000 Analyzer (Roche Diagnostics) us- weights (24). Similar linear regression All analyses were performed using R ing a particle-enhanced immunoturbidi- models were used to test associations 3.3.2 (R Foundation for Statistical Com- metric assay. Total iron binding capacity between 18 potentially novel SNPs and puting, Vienna, Austria) or SAS 9.4 (SAS 25 was calculated as the sum of serum iron HbA1c (P , 1 3 10 ), adjusting for age, Institute, Cary, NC) software. and UIBC, and saturation was sex, top principal components, center, calculated as the percentage of serum and relatedness (if appropriate), using iron in total iron binding capacity. Alanine data from the three replication cohorts. RESULTS fi aminotransferase(ALT)andaspartateami- Inverse variance xed-effect meta-analyses GWAS of HbA1c notransferase (AST) were measured in were performed to combine the results of In the discovery GWAS in the HCHS/SOL, serum on a Roche Modular P Chemistry replicationstudiesandtocombinethese we identified 12 genome-wide significant 28 Analyzer using an a-ketoglutaratic enzy- with results from HCHS/SOL. associations withHbA1c (P, 5.03 10 ), matic method (Roche Diagnostics). Sus- We examined the potential implica- 6 of which were previously known loci pected nonalcoholic fatty liver disease tions of two variants, HBB-rs334 and (G6PC2, GCK, ANK1, HK1, FN3K, and was defined by the elevated aminotrans- G6PD-rs1050828, on the screening of TMPRSS6) (Fig. 1 and Table 1). We ferase levels as AST .31 IU/mL or ALT .40 hyperglycemia (prediabetes and diabe- then sought replication of 18 potentially IU/mL for men and AST or ALT .31 IU/mL tes) using HbA1c among 10,470 HCHS/ novel independent SNPs with suggestive 25 for women (22). SOL participants without diagnosed di- associations with HbA1c (P , 1.0 3 10 ) abetes (those with self-reported diabe- in three independent studies of Hispanics/ Genotyping and Imputation tes or antidiabetic medication use were Latinos in theU.S. (Supplementary Tables In the HCHS/SOL, genotyping was per- not included in this analysis). We reesti- 2 and 3). Of note, G6PD and HBB variants formed with an Illumina custom array mated the effects of HBB-rs334 and (10,11) were not known to be associated (15041502 B3), which consists of the G6PD-rs1050828 on HbA1c levels, respec- with HbA1c when we selected SNPs for Illumina Omni 2.5M array (HumanOmni tively, because the additional samples replication in this study. Four SNPs, in- 240 2.5-8v1-1) plus ;150,000 custom single included the natural right-end of HbA1c cluding rs334 at HBB (P = 2.7 3 10 nucleotide polymorphisms (SNPs), with distribution, instead of cutoff at 6.5% in HCHS/SOL), rs145546625 at HBM the quality control performed at the (48 mmol/mol) in our GWAS analysis, but (P =8.23 1029), rs145546625 at HCHS/SOL Genetic Analysis Center (16). still without prominent influences by self- TMC6 (P =2.03 10215), and rs1050828 Genome-wide imputation was performed awareness or medication. Then, measured at G6PD (P =9.63 102132), were robustly with 1000 Genomes Project (1000G) phase HbA1c levels were adjusted to account for associated with HbA1c in the combined rep- 1 worldwide reference panel (v3, released genetic variants using the re-estimated lication analyses with nominal significance March 2012) using SHAPEIT2 and IMPUTE2 effects in the following equation: (all P # 0.05) (Table 1 and Supplementary software, as described previously (16). In the Table 2). No significant heterogeneity was ð%Þ¼ SCHS, genome-wide SNPs were genotyped HbA1cadjusted observed across cohorts. Regional plots HbA1c ð%Þþ 0:33 using the Affymetrix Genome-Wide Human measured for these four novel HbA1c loci are shown SNP Array 6.0 and imputed to the 1000G 3 # of T-alleles in HBB-rs334 þ 0:35 in Supplementary Fig. 1. The effect sizes of phase 1 reference data (19). In the BPRHS, 3 # of A-alleles in G6PD-rs1050828 HBB rs334 (b = 20.31% [23.4 mmol/mol] genome-wide SNPs were genotyped using Eq. 1 per minor allele) and G6PD rs1050828 the Affymetrix Axiom Genome-Wide LAT (b = 20.35% [23.8 mmol/mol] per minor Array and imputed to the 1000G phase allele) on HbA1c were ;10 times larger 1 reference data (23). In the BioMe, where a hemizygous AO of rs1050828 in compared with other HbA1c-related var- genome-wide SNPs were genotyped with the X for men was coded to iants (Table 1). care.diabetesjournals.org Moon and Associates 1787

Figure 1—Manhattan plot for GWAS of HbA1c in the HCHS/SOL. A total of 11,510,031 SNPs with MAF .0.01 were tested in 9,636 U.S. Hispanics/Latinos without diabetes from the HCHS/SOL. Six previously known loci (green), four newly identified and replicated loci (red), and two newly identified but not 28 replicated loci (pink) associated with HbA1c at the genome-wide significance level. P , 5.0 3 10 (dashed line).

In analyses stratified by Hispanic/ three SNPs at the same locus (rs2748427, near suggestive significance (P = 6.2 3 Latino background, associations be- rs761772, and rs2073285) were identi- 1025). Expanding theconditional analysis tween these four SNPs and HbA1c fied in recent HbA1c GWAS (10,12,13), to ;500 kb of our lead SNP, we found a were consistent across groups with no reaching genome-wide significance in suggestive but not genome-wide signif- significant heterogeneity (all P for our study as well (Supplementary Table 6 icant secondary signal at rs80149164 27 heterogeneity $0.14) (Supplementary and Supplementary Fig. 2). In the (P =5.33 10 ) (Supplementary Fig. 2). Table 4). Given that SNP rs1050828 is joint analyses of our lead SNP with located at the X chromosome, we further rs2748427 or rs761772, which are in HbA -Related SNPs and Hematologic examined the association by sex, and moderate-to-high linkage disequilibrium 1c results were consistent between men 2 Traits (LD) with the lead SNP (r = 0.70, 0.53), We then examined associations of and women. In addition, a sensitivity our lead SNP showed attenuated but these newly identified four HbA1c variants analysis excluding participants with still suggestively significant signals (P = fi with diabetes-related traits, hematologic iron de ciency showed similar associa- 2.8 31025, P =3.13 1025), whereas tions between these SNPs and HbA traits (25,26), iron traits (27), and 1c the signals of the other SNPs became null (Supplementary Table 5). liver-related traits in the HCHS/SOL to (P = 0.59, P = 0.78) (Supplementary Table explore potential mechanisms underly- 6). The joint analysis with rs2073285 in ing observed SNP-HbA1c relationships. 2 Conditional Analysis weak LD (r = 0.10) showed that the lead These newly identified HbA1c SNPs Although our lead SNP rs145546625 at SNP was still at genome-wide significance were significantly associated with hema- TMC6 has not been reported before, (P = 1.1 3 10211), and rs2073285 was at tologic traits rather than glycemic traits

Table 1—Summary of genetic variants associated with HbA1c in U.S. Hispanics/Latinos Allele HCHS/SOL SNP Nearest Chr Position Effect Other EAF b (SE) P value Replication P value* Previously known loci before 2017 rs557462 G6PC2/ 2 169777595 C T 0.182 20.03 (0.01) 5.7 3 1029 ABCB11 rs2971670 GCK 7 44226101 T C 0.202 0.04 (0.01) 2.9 3 10211 rs4737010 ANK1 8 41630447 A G 0.263 0.03 (0.01) 5.1 3 10210 rs72805692 HK1 10 71099109 G A 0.066 20.10 (0.01) 6.1 3 10226 rs2256339 FN3K 17 80693281 T A 0.475 20.03(0.01) 1.9 3 10211 rs855791 TMPRSS6 22 37462936 A G 0.438 0.03 (0.01) 8.7 3 10210 Additional loci rs334 HBB 11 5248232 A T 0.013 20.31 (0.02) 2.7 3 10240 0.004 rs145546625 HBM 16 220583 T C 0.066 0.06 (0.01) 8.2 3 1029 0.050 rs2748424 TMC6 17 76124865 G C 0.179 0.05 (0.01) 2.0 3 10215 0.008 rs1050828 G6PD X 153764217 A G 0.020 20.35 (0.01) 9.6 3 102132 0.003

b, effect size for each effect allele of SNP on HbA1c (%); Chr, chromosome; EAF, effect allele frequency; Position, in GRCh37/hg19. *Fixed effect meta- analysis of SCHS, BPRHS, BioMe-Omni, and BioMe-MEGA. 1788 HbA1c Variants and Glycemic Status in Hispanics Diabetes Care Volume 42, September 2019

or liver-related traits (Supplementary sickle cell disease by rs334 hereinafter, comparison between these two groups Table 7). the A-to-T mutation of HBB-rs334 low- was comparable to that between carriers Specifically, HBB-rs334 (A-to-T, Glu7Val) ered HbA1c by 0.33% (3.6 mmol/mol). (;5% of the study population) and non- is a causal mutation for sickle cell The G-to-A mutation of G6PD-rs1050828 carriers (;95% of the study population) trait (heterozygous of the T allele) lowered HbA1c by 0.35% (3.8 mmol/mol), of HBB or G6PD variants. Individuals with and sickle cell disease (homozygous with no significant difference by sex. the bottom 5% of the GRS had lower of the T allele), producing abnormal We then classified individuals into two HbA1c levels compared with those with b-globin in hemoglobin. In line with groups, noncarriers and carriers of the higher GRS (mean 5.38% [35 mmol/mol] this, the minor T allele (HbA1c-lowering HBB-rs334 T allele or G6PD-rs1050828 vs. 5.50% [37 mmol/mol]; P =1.23 27 allele) was associated with lower hemat- A allele. Carriers of HbA1c-lowering al- 10 ).Theprevalenceofhyperglycemia 210 ocrit (P = 1.3 3 10 ), mean corpuscular leles tended to have lower HbA1c levels defined by fasting glucose or 2-h glucose volume (MCV) (P = 1.1 3 10222), and (P , 0.001) as expected, but fasting was comparable between the two groups, mean corpuscular hemoglobin (MCH) glucose and 2-h glucose levels did not whereas the prevalence of hyperglycemia 25 (P = 1.3 3 10 ), and higher mean differ significantly between groups (P = defined by HbA1c was lower in the lower corpuscular hemoglobin concentration 0.14 and P = 0.50, respectively) (Supple- GRS group than in the higher GRS group (P = 3.6 3 10216), which are hematologic mentary Table 9). At the same fasting (21.9% vs. 28.1%) (Supplementary Table 11). characteristics for sickle cell trait and glucose levels, carriers tended to have CONCLUSIONS disease. G6PD-rs1050828 (G-to-A, Val98- lower measured HbA1c than noncarriers, Met) is a causal mutation for glucose- whereas after genetic recalibration by This study, the first GWAS of HbA1c 6-phosphate dehydrogenase (G6PD) Eq. 1, HbA1c levels in respect to fasting to date among Hispanics/Latinos in the deficiency, resulting in the premature glucose became comparable between U.S., showed multiple previously known breakdown of RBCs. The minor A allele carriers and noncarriers (Fig. 2 and Sup- as well as novel loci associated with was associated with lower RBC count (P = plementary Fig. 3). HbA1c at genome-wide significance level 1.4 3 10219), higher MCV (P = 3.7 3 Carriers and noncarriers had compa- (P , 5.0 3 1028). In particular, two blood 10213), higher MCH (P = 4.9 3 10211), rable prevalence of hyperglycemia (pre- disorder–related variants, HBB-rs334 lower RBC distribution width (P = 3.7 3 diabetes and undiagnosed diabetes) and G6PD-rs1050828, showed ;10-fold 10229), higher iron (P = 3.0 3 1025), and defined by fasting glucose $100 mg/dL larger effect sizes (0.3–0.4% [3.3–4.4 higher transferrin saturation (P = 1.1 3 or 2-h glucose $140 mg/dL (carriers vs. mmol/mol] per allele) on HbA1c com- 1026), which are hematologic character- noncarriers: 21.2% vs. 25.4%, 18.6% vs. pared with other variants (0.03–0.04% istics for G6PD deficiency and subsequent 21.8%; P = 0.10, P = 0.17, respectively), [0.3–0.4 mmol/mol] per allele). The iron overload from chronic anemia. In whereas carriers had lower prevalence of G6PD variant was recently identified in a addition, the minor T allele of HBM- hyperglycemia defined by HbA1c $5.7% GWAS of HbA1c in African Americans (10), rs145546625 was associated with lower (39 mmol/mol [12.2% vs. 28.4%], P , and sickle cell trait (determined by the MCV and MCH. 0.001) compared with noncarriers (Table HBB-rs334) was reported to be associated 2). After recalibration, there was no with HbA1c in African Americans (11). fi HbA1c-Related SNPs and Screening for signi cant difference in the prevalence We found that the HbA1c level was Diabetes and Prediabetes of hyperglycemia defined by genetically significantly lower by 0.33% (3.6 Given the much larger effect sizes of HBB- adjusted HbA1c between carriers and mmol/mol; 95% CI 0.24–0.42 [2.6–4.6 rs334 and G6PD-rs1050828 on HbA1c noncarriers (31.3% vs. 28.4%, P = mmol/mol]) per A allele of HBB-rs334 compared with other HbA1c-related var- 0.28). Results were similar but not sig- in Hispanics/Latinos, comparable to the iants and their nonglycemic-related nificant when we examined the potential reported difference in HbA1c between features (strong associations with he- implications of these genetic variants in African Americans with and without matologic traits rather than glycemic the diabetes screen using HbA1c levels, sickle cell trait (0.29% [3.2 mmol/mol]; traits), we then examined the influen- which might be due to the small num- 95% CI 0.23–0.35 [2.5–3.8 mmol/mol]) ces of these two variants on hypergly- ber of individuals with undiagnosed di- (11). The missense A allele of rs334 is cemia screening using HbA1c. abetes among carriers (Supplementary observed in African ancestry at a fre- In 10,470 participants without diag- Table 10). quency of 0.1, is very rare in Amerindian nosed diabetes (self-reported diabetes Although other nonglycemic-related ancestry at a frequency of 0.01, and does or antidiabetic medication use), 224 par- genetic variants have much smaller ef- not appear in other ancestries from ticipants were carriers of the HBB-rs334 T fects on HbA1c compared with HBB-rs334 1000G (25,28), which comaps with ma- allele, of which 222 individuals were and G6PD-rs1050828, a combined effect laria risk. The homozygotes of A allele heterozygous (sickle cell trait) and may have a considerable impact on have sickle cell anemia, a lifelong disease two individuals were homozygous of T HbA1c for the hyperglycemia screening. by chronic hemolytic anemia, caused allele(sickle celldisease)(Supplementary Hence, we constructed an unweighted by the rigid and sickling form of RBCs Table 8). Among 309 carriers of the GRS of five potentially erythrocytic ge- formed by polymerization of hemoglobin G6PD-rs1050828 A allele in X chromo- netic variants. As expected, the GRS S especially at low oxygen concentra- some, 91 individuals were homozygous showed an additive effect on HbA1c levels tions (29). Although the heterozygotes of AA for women and AO for men, and (Supplementary Fig. 4). We classified in- (AT; sickle cell trait) are asymptom- 218 women were heterozygous. With dividuals into two groups by a cutoff of atic, they have increased complications, exclusion of the two individuals with 5th percentile of the GRS. Thus, the such as urinary tract infection, splenic care.diabetesjournals.org Moon and Associates 1789

levels, regardless of blood glucose lev- els (10). Further, the regulation via the erythrocytic pathway is supported by the significant associations with RBC count, red cell distribution width, MCH concentration, MCV, and higher iron levels, potentially from chronic hemo- lysis (33), but no association with other glycemic traits in our study. TMC6-rs2748424 in the upstream of transmembrane channel-like 6 (TMC6, also known as epidermodysplasia ver- ruciformis 1 [EVER1]) is a common var- iant observed across all ancestries — Figure 2 Scatterplots of measured HbA1c and genetically adjusted HbA1c against fasting glucose in (0.19–0.46 of minor allele frequency) carriers and noncarriers of HBB-rs334 or G6PD-rs1050828 minor alleles. Zoomed-in plots among the individuals without diagnosed diabetes. Red points indicate carriers and blue points indicate (28). Three SNPs (rs2748427, rs761772, noncarriers of HBB-rs334 or G6PD-rs1050828 minor alleles. and rs2073285) near our lead SNP rs2748424 were previously identified in infarction, or sudden death, when exposed with a relatively large effect on HbA1c Japanese, European, and non-Hispanic/ to strenuous exercise, high altitudes, de- in our study of U.S. Hispanics/Latinos Latino transethnic populations (10,13). hydration, or low oxygen levels. Lacy et al. (20.35% [23.8 mmol/mol] per A allele) All of these three previously reported (11) hypothesized that lower HbA1c in as well as in a previous study of African SNPs were at genome-wide significance people with sickle cell trait might be Americans (20.40% [24.4 mmol/mol] in our Hispanic/Latino populations. due to the shorter life span of the RBCs per A allele for men and 20.34% SNPs rs2748427 and rs761772 showed and accordingly less opportunity for gly- [23.7 mmol/mol] for women) (10). The moderate-to-high LD with our lead SNP, cation of hemoglobin. In line with this, our missense A allele in rs1050828 is African- whereas rs2073285 showed weak LD data in the HCHS/SOL demonstrated that ancestry specific (frequency 0.13), is rarely (r2 = 0.1) with our lead SNP. Of note, this variant was not associated with fasting foundinAmerindianancestry(frequency only our lead SNP rs2748424 remained at glucose but was associated with small RBCs 0.01), and is not seen in other ancestries genome-wide significance level in the (microcytes), characterized by lower MCV (28). G6PD (G6PD) catalyzes the reaction conditional analysis, suggesting a poten- and MCH. Because microcytes may be more of the pentose phosphate pathway and tial causal signal represented by this SNP. susceptible to oxidative stress, the life span in turn generates the reduced form of In addition, our lead SNP and two other of RBCs might be shortened (30). These data glutathione as an antioxidant, which pro- SNPs (rs2748427 and rs761772) in LD suggest that HBB-rs334 may influence tects RBCs against oxidative stress (32). were related to hematologic traits, in- HbA1c through hematologic mechanisms G6PD deficiency reduces G6PD activity, cluding the immature fraction of retic- independent of blood glucose. On the results in a premature breakdown of RBCs, ulocytes (34), RBCs, MCH, and MCV (35) contrary, it has also been postulated that and aggravates to hemolytic anemia when rather than glycemic markers such as lower HbA1c in individuals with sickle cell triggered by certain foods (fava beans), fructosamine and glycated albumin trait might be due to assay interference by drugs, or infection. This explains the (13,36), although there were no signifi- hemoglobin S compared with those with- survival advantage of G6PD deficiency cant associations with hematologic traits in out sickle cell trait (11,31). against malaria because it uses RBCs our study. It has been hypothesized that The X chromosome–linked G6PD- as ahostcell(32).However,theshortened TMC6 genetic variants may affect HbA1c, rs1050828isanother erythrocyticvariant life span of RBCs may result in lower HbA1c independent of blood glucose levels,

Table 2—Prevalence of hyperglycemia estimated based on fasting glucose, 2-h glucose, and HbA1c in carriers and noncarriers of HBB-rs334 or G6PD-rs1050828 minor alleles without diagnosed diabetes Prevalence of hyperglycemia, % (95% CI) Carrier Noncarrier P value Fasting glucose $100 mg/dL 21.2 (16.6–25.8) 25.4 (24.0–26.8) 0.10 Post-OGTT glucose $140 mg/dL 18.6 (14.2–23.0) 21.8 (20.5–23.2) 0.17

Measured HbA1c $5.7% (39 mmol/mol) 12.2 (8.9–15.4) 28.4 (27.1–29.7) ,0.001 Fasting glucose $100 mg/dL, post-OGTT glucose $140 mg/

dL, or HbA1c $5.7% (39 mmol/mol) 31.9 (26.5–37.2) 46.6 (45.0–48.2) ,0.001

Genetically adjusted HbA1c $5.7% (39 mmol/mol)* 31.3 (26.0–36.5) 28.4 (27.1–29.7) 0.28 Fasting glucose $100 mg/dL, post-OGTT glucose $140 mg/

dL, or genetically adjusted HbA1c $5.7% (39 mmol/mol)* 42.0 (36.0–48.0) 46.6 (45.0–48.2) 0.13 Prevalence estimates were computed while accounting for the complex study design of HCHS/SOL. *Hyperglycemia defined with genetically adjusted HbA1c. 1790 HbA1c Variants and Glycemic Status in Hispanics Diabetes Care Volume 42, September 2019

through erythrocyte life span, iron in the opportunity of glycation, hemo- grant AI-085014, NIDDK grants DK-020595, handling, or glucose concentration dif- globin levels, and glycation rate DK-073541, and DK-085501, and NHLBI grant ference across the erythrocyte mem- (27,33,38,39). On the other hand, the in- HL-102830 from the National Institutes of Health and funds from the University of Texas Health brane (13). creased glycated from high glu- Science Center at Houston. Genotyping services Lastly, to the best of our knowledge, cose concentrations may increase the were provided by the Center for Inherited Dis- HBM-rs145546625 was discovered and oxidative stress or alter erythrocyte con- ease Research, which is funded through a federal replicated in this study of Hispanics/ ditions (40). More investigations are contract from the National Institutes of Health to fi The Johns Hopkins University, contract number Latinos for the rst time. HBM- needed to evaluate the potential use HHSN268200782096C. BRPHS was funded rs145546625 originated from Amerindian of genetic adjustment with HbA1c levels by NHLBI grant P50-HL-105185 and National ancestry (frequency of T allele 0.08) (28). among individuals without symptoms Institute on Aging grant P01-AG-023394. The This Amerindian-specific variant is located but at high risk of these blood abnormities Andrea and Charles Bronfman Philanthropies sup- upstream of hemoglobin-alpha2 (HBA2) (e.g., those with African ancestry and ported the BioMe Biobank program. Q.Q. is supported by NHLBI grants K01-HL-129892, R01- and downstream of hemoglobin-mu family history of sickle cell disease/trait). HL-060712, and R01-HL-140976, and by NIDDK 2 (HBM), in high LD (r . 0.99) with a In summary, this study found multiple grants R01-DK-119268 and R01-DK-120870. splice-site variant rs148323035 in HBM genetic variants associated with HbA1c The funding sources had no role in the study (25). There is evidence suggesting that levels in U.S. Hispanics/Latinos, particularly design or execution, data analysis, manuscript – fi writing, or manuscript submission. these Amerindian ancestral variants may two African ancestry speci cvariantsat fl fl Duality of Interest. No potential con icts of in uence HbA1c levels through blood cell HBB and G6PD and an Amerindian interest relevant to this article were reported. biology (25), but more study is warranted to ancestry–specific variant at HBM.Further Author Contributions. J.-Y.M. and Q.Q. per- elucidate how this variant may influence analysis suggested that these variants may formed the literature search, analyzed and in- terpreted data, and drafted the manuscript. HbA1c through the erythrocyte pathway. influence HbA1c through erythrocyte- The findings of the current study pro- related pathways rather than glycemic- T.L.L., D.J., T.S., and T.W. performed data anal- yses in the HCHS/SOL. C.S., J.E.B., C.-Q.L., C.E.S., vide evidence supporting the significant related pathways. Our findings expand the L.E.P., E.P.B., K.L.T., J.M.O., C.L.H., and R.J.F.L. influences of nonglycemic-related ge- understandings of erythrocyte-related ge- contributed to replication data. M.L.A.-S., G.A.T., netic variants on diabetes screening using netic determinants of HbA1c and their im- Y.-D.I.C., K.D.T., M.L.D., J.C., N.S., J.I.R., and R.C.K. contributed to data generation in the HCHS/SOL. HbA1c tests, consistent with two previous plications for evaluation of HbA1c tests of fi Q.Q. designed the study. All authors contributed studies (10,11). Speci cally, individuals in glycemic status in U.S. Hispanics/Latinos. to data interpretation and editing and review- this study carrying the HbA1c-lowering ing the manuscript. Q.Q. is the guarantor of this alleles of HBB-rs334 or G6PD-rs1050828 work and, as such, has full access to all the data in the study and takes responsibility for the had a lower prevalence of hyperglycemia Acknowledgments. The authors thank the integrity of the data and the accuracy of the (prediabetes and diabetes) defined by participants and staffs of HCHS/SOL, SCHS, Bos- data analysis. ton Puerto Rican Health Study, and BioMe Bio- HbA1c levels compared with noncarriers, Prior Presentation. Parts of this study were bank for their important contributions. whereas there were no differences in the presented in abstract and poster form at the Funding. HCHS/SOL was performed as a collab- prevalence of hyperglycemia defined by 77th Scientific Sessions of the American Di- orative study supported by contracts from the abetes Association, San Diego, CA, 9–13 June fasting glucose or post-OGTT glucose be- National Heart, Lung, and Blood Institute (NHLBI) 2017. tween carriers and noncarriers. In addi- to the University of North Carolina (N01-HC- tion to findings in previous studies (10,11), 65233), University of Miami (N01-HC-65234), our current analysis using genetically ad- Albert Einstein College of Medicine (N01-HC- References 65235), Northwestern University (N01-HC- 1. Mortensen HB, Christophersen C. Glucosyla- justed HbA1c levels demonstrated that 65236), and San Diego State University tion of human haemoglobin a in red blood cells this underestimation could be recali- (N01-HC-65237). The following institutes/centers/ studied in vitro. Kinetics of the formation and brated when genetic information was offices contributed to the HCHS/SOL first funding dissociation of haemoglobin A1c. Clin Chim Acta taken into account. Although our study period through a transfer of funds to the NHLBI: 1983;134:317–326 did not test influences of nonglycemic- the National Institute on Minority Health and 2. The International Expert Committee. Inter- Health Disparities, the National Institute on national Expert Committee report on the role of related genetic variants on the prediction Deafness and Other Communication Disorders, the A1C assay in the diagnosis of diabetes. of diabetes and related complications the National Institute of Dental and Craniofacial Diabetes Care 2009;32:1327–1334 such as stroke and coronary artery dis- Research, the National Institute of Diabetes 3. Pilia G, Chen WM, Scuteri A, et al. Heritability ease, these potential implications have and Digestive and Kidney Diseases (NIDDK), of cardiovascular and personality traits in 6,148 been reported previously (37). the National Institute of Neurological Disor- Sardinians. PLoS Genet 2006;2:e132 fi ders and Stroke, and the National Institutes of 4. Soranzo N, Sanna S, Wheeler E, et al.; WTCCC. Taken together, these ndings support Health Office of Dietary Supplements. The Ge- Common variants at 10 genomic loci influence the current American Diabetes Associa- netic Analysis Center at the University of hemoglobin A(1)(C) levels via glycemic and non- tion recommendations that hemoglobin Washington was supported by NHLBI contracts glycemic pathways [published correction appears variants need to be considered when (HHSN268201300005C AM03) and National in Diabetes 2011;60:1050–1051].Diabetes2010;59: – evaluating the HbA tests (38). However, Institute of Dental and Craniofacial Research 3229 3239 1c contracts (HHSN268201300005C MOD03). Gen- 5. Franklin CS, Aulchenko YS, Huffman JE, et al. it should be noted that the clinical in- otyping efforts were supported by NHLBI HSN The TCF7L2 diabetes risk variant is associated terpretation of HbA1c may need to con- 26220/20054C, National Center for Advancing with HbA(1)(C) levels: a genome-wide associa- sider conditions such as shortened Translational Sciences Clinical and Translational tion meta-analysis. Ann Hum Genet 2010;74: erythrocyte life span, hemolytic anemia, Science Institute grant UL1-TR-000123, and 471–478 fi NIDDK Diabetes Research Center (DRC) grant 6. Ryu J, Lee C. Association of glycosylated microcytosis, iron de ciency, iron over- DK-063491. SCHS was supported in part by hemoglobin with the gene encoding CDKAL1 load, vitamin B12, and folate, which may Division of Intramural Research, National in the Korean Association Resource (KARE) study. impact the HbA1c levels from the changes Institute of Allergy and Infectious Diseases Hum Mutat 2012;33:655–659 care.diabetesjournals.org Moon and Associates 1791

7. Chen P, Takeuchi F, Lee JY, et al. Multiple Hispanic Community Health Study/Study of La- 28. Auton A, Brooks LD, Durbin RM, et al.; 1000 nonglycemic genomic loci are newly associated tinos. Ann Epidemiol 2010;20:642–649 Genomes Project Consortium. A global reference with blood level of glycated hemoglobin in East 18. Sorlie PD, Aviles-Santa LM, Wassertheil- for human genetic variation. Nature 2015;526: Asians. Diabetes 2014;63:2551–2562 Smoller S, et al. Design and implementation of 68–74 8. Pare G, Chasman DI, Parker AN, et al. Novel the Hispanic Community Health Study/Study of 29. Kato GJ, Piel FB, Reid CD, et al. Sickle cell association of HK1 with glycated hemoglobin in a Latinos. Ann Epidemiol 2010;20:629–641 disease. Nat Rev Dis Primers 2018;4:18010 non-diabetic population: a genome-wide evalu- 19. Below JE, Gamazon ER, Morrison JV, et al. 30. Vives Corrons JL, Miguel-Garcia A, Pujades ation of 14,618 participants in the Women’s Genome-wide association and meta-analysis in MA, et al. Increased susceptibility of microcytic Genome Health Study. PLoS Genet 2008;4: populationsfromStarrCounty,Texas,andMexico red blood cells to in vitro oxidative stress. Eur J e1000312 City identify type 2 diabetes susceptibility loci and Haematol 1995;55:327–331 9. Syreeni A, Sandholm N, Cao J, et al.; DCCT/ enrichment for expression quantitative trait loci 31. Rohlfing C, Hanson S, Little RR. Measure- EDIC Research Group; FinnDiane Study Group. in top signals. Diabetologia 2011;54:2047–2055 ment of hemoglobin A1c in patients with sickle Genetic determinants of glycated hemoglobin in 20. Tucker KL, Mattei J, Noel SE, et al. The Boston cell trait. JAMA 2017;317:2237 type 1 diabetes. Diabetes 2019;68:858–867 Puerto Rican Health Study, a longitudinal cohort 32. Guindo A, Fairhurst RM, Doumbo OK, fi 10. Wheeler E, Leong A, Liu CT, et al.; EPIC-CVD study on health disparities in Puerto Rican adults: Wellems TE, Diallo DA. X-linked G6PD de ciency Consortium; EPIC-InterACT Consortium; Lifelines challenges and opportunities. BMC Public Health protects hemizygous males but not heterozygous Cohort Study. Impact of common genetic deter- 2010;10:107 females against severe malaria. PLoS Med 2007; minants of Hemoglobin A1c on type 2 diabetes 21. Gottesman O, Kuivaniemi H, Tromp G, et al. 4:e66 risk and diagnosis in ancestrally diverse popula- The Electronic Medical Records and Genomics 33. Fibach E, Rachmilewitz EA. Iron overload in tions: a transethnic genome-wide meta-analysis. (eMERGE) Network: past, present, and future. hematological disorders. Presse Med 2017;46: e296–e305 PLoS Med 2017;14:e1002383 Genet Med 2013;15:761–771 34. Astle WJ, Elding H, Jiang T, et al. The allelic 11. Lacy ME, Wellenius GA, Sumner AE, et al. 22. Clark JM, Brancati FL, Diehl AM. The prev- landscape of human blood cell trait variation and Association of sickle cell trait with hemoglobin alence and etiology of elevated aminotransferase links to common complex disease. Cell 2016;167: A1c in African Americans. JAMA 2017;317:507– levels in the United States. Am J Gastroenterol 1415–1429.e19 515 2003;98:960–967 35. Kanai M, Akiyama M, Takahashi A, et al. 12. Prins BP, Kuchenbaecker KB, Bao Y, et al. 23. Blanco-Rojo R, Delgado-Lista J, Lee YC, et al. Genetic analysis of quantitative traits in the Genome-wide analysis of health-related bio- Interaction of an S100A9 gene variant with Japanese population links cell types to com- markers in the UK Household Longitudinal Study saturated fat and carbohydrates to modulate plex human diseases. Nat Genet 2018;50:390– insulin resistance in 3 populations of different reveals novel associations. Sci Rep 2017;7:11008 400 – 13. Hachiya T, Komaki S, Hasegawa Y, et al. ancestries. Am J Clin Nutr 2016;104:508 517 36. Loomis SJ, Li M, Maruthur NM, et al. Ge- Genome-wide meta-analysis in Japanese popu- 24. Pfefferman D. The role of sampling weights nome-wide association study of serum fructos- fi lations identi es novel variants at the TMC6- when modeling survey data. Int Stat Rev 1993; amine and glycated albumin in adults without – TMC8 and SIX3-SIX2 loci associated with HbA1c. 61:317 337 diagnosed diabetes: results from the Atheroscle- Sci Rep 2017;7:16147 25. Hodonsky CJ, Jain D, Schick UM, et al. Ge- rosis Risk in Communities Study. Diabetes 2018; 14. Menke A, Casagrande S, Geiss L, Cowie CC. nome-wide association study of red blood cell 67:1684–1696 Prevalence of and trends in diabetes among traits in Hispanics/Latinos: the Hispanic Commu- 37. Au Yeung SL, Luo S, Schooling CM. The adults in the United States, 1988-2012. JAMA nity Health Study/Study of Latinos. PLoS Genet impact of glycated hemoglobin (HbA1c) on car- 2015;314:1021–1029 2017;13:e1006760 diovascular disease risk: a Mendelian random- 15. Stark Casagrande S, Fradkin JE, Saydah SH, 26. Schick UM, Jain D, Hodonsky CJ, et al. Ge- ization study using UK Biobank. Diabetes Care Rust KF, Cowie CC. The prevalence of meeting nome-wide association study of platelet count 2018;41:1991–1997 A1C, blood pressure, and LDL goals among identifies ancestry-specific loci in Hispanic/ 38. American Diabetes Association. 6. Glycemic people with diabetes, 1988-2010. Diabetes Latino Americans. Am J Hum Genet 2016;98: targets: Standards of Medical Care in Diabetesd Care 2013;36:2271–2279 229–242 2018. Diabetes Care 2018;41(Suppl. 1):S55–S64 16. Conomos MP, Laurie CA, Stilp AM, et al. 27. Raffield LM, Louie T, Sofer T, et al. Ge- 39. Cohen RM, Franco RS, Khera PK, et al. Red Genetic diversity and association studies in US nome-wide association study of iron traits and cell life span heterogeneity in hematologically Hispanic/Latino populations: applications in the relation to diabetes in the Hispanic Commu- normal people is sufficient to alter HbA1c. Blood Hispanic Community Health Study/Study of La- nity Health Study/Study of Latinos (HCHS/ 2008;112:4284–4291 tinos. Am J Hum Genet 2016;98:165–184 SOL): potential genomic intersection of iron 40. Singh VP, Bali A, Singh N, Jaggi AS. Advanced 17. Lavange LM, Kalsbeek WD, Sorlie PD, et al. and glucose regulation? Hum Mol Genet 2017; glycation end products and diabetic complica- Sample design and cohort selection in the 26:1966–1978 tions. Korean J Physiol Pharmacol 2014;18:1–14