Admixture Mapping Identifies an Amerindian Ancestry Locus Associated with Albuminuria in Hispanics in the United States

Abstract

Increased urine albumin excretion is highly prevalent in Hispanics/Latinos. Previous studies have found an association between urine albumin excretion and Amerindian ancestry in Hispanic/Latino populations. Admixture between racial/ethnic groups creates long-range linkage disequilibrium between variants with different allelic frequencies in the founding populations and it can be used to localize genes. Hispanic/Latino genomes are an admixture of European, African, and Amerindian ancestries. We leveraged this admixture to identify associations between urine albumin excretion (urine albumin-to-creatinine ratio [UACR]) and genomic regions harboring variants with highly differentiated allele frequencies among the ancestral populations. Admixture mapping analysis of 12,212 Hispanic Community Health Study/Study of Latinos participants, using a linear mixed model, identified three novel genome-wide significant signals on chromosomes 2, 11, and 16. The admixture mapping signal identified on chromosome 2, spanning q11.2–14.1 and not previously reported for UACR, is driven by a difference between Amerindian ancestry and the other two ancestries (P<5.7 × 10⁻⁵). Within this locus, two common variants located at the proapoptotic BCL2L11 gene associated with UACR: rs116907128 (allele frequency =0.14; P=1.5 × 10⁻⁷) and rs586283 (C allele frequency =0.35; P=4.2 × 10⁻⁷). In a secondary analysis, rs116907128 accounted for most of the admixture mapping signal observed in the region. The rs116907128 variant is common among full-heritage Pima Indians (A allele frequency =0.54) but is monomorphic in the 1000 Genomes European and African populations. In a replication analysis using a sample of full-heritage Pima Indians, rs116907128 significantly associated with UACR (P=0.01; n=1568). Our findings provide evidence for the presence of Amerindian-specific variants influencing the variation of urine albumin excretion in Hispanics/Latinos.

Increased urine albumin excretion is associated with a higher lifetime risk of ESRD and with increased cardiovascular disease risk.^1,2 Both increased urine albumin excretion and ESRD differ by racial/ethnic groups in the United States, with the lowest and highest risks noted in European and Amerindian populations, respectively. Using sex-specific cut-points, increased urine albumin excretion prevalence in the United States is 10.3% in whites, 13.6% in blacks, 9.9% in Mexican Americans,³ >20% in American Indians,⁴ and 12%–14% in Hispanics/Latinos of the Hispanic Community Health Study/ Study of Latinos (HCHS/SOL).⁵ Hispanics/Latinos also have an approximately two-fold higher risk of ESRD than whites.⁶ However, Hispanics/Latinos are a heterogeneous group who show diversity in ancestry background including Amerindian, European, and West African.⁷ The percentages of African and Amerindian ancestry have previously been associated with increased urine albumin excretion prevalence in Hispanic/Latino populations.^8,9

Despite the strong evidence for a role of ancestry in CKD susceptibility, few studies of kidney traits have leveraged the known genetic admixture in Hispanic/Latino populations to discover potential chromosomal regions that may harbor variants that confer risk for CKD traits such as urine albumin excretion. Among the genome-wide significant loci that have been identified and consistently replicated for urine albumin excretion, CUBN (chromosome 10) genetic variants are associated with urine albumin excretion in individuals of European ancestry and Hispanics/Latinos,¹⁰ and the HBB variant related to sickle cell trait (chromosome 11) is African-specific and associated with urine albumin excretion in Hispanics/Latinos with African admixture.¹¹ An additional African-specific gene associated with increased urine albumin excretion and CKD is APOL1.^12,13 Our recent work in the HCHS/SOL has confirmed a high proportion of Amerindian ancestry among Mainland Hispanics (Mexican, Central, and South American), who also had a low proportion of African ancestry.¹⁴ However, Mainland Hispanics have similar mean urine albumin excretion and frequency of increased urine albumin excretion compared with individuals of Caribbean background (Cuban, Dominican, Puerto-Rican), despite the absence of African-specific risk variants (APOL1 or HBB). This evidence suggests the presence of Amerindian ancestry variants in Hispanics/Latinos influencing urine albumin excretion.

There is great potential for genetic studies of CKD in Hispanics/Latinos to provide new insight into population-specific variants that confer risk for CKD traits but have not been uncovered through genome-wide association studies (GWAS). Admixture mapping leverages the known genomic heterogeneity of admixed individuals for improved genetic discovery, by identifying loci that contain genetic variants with allelic frequencies that are highly differentiated among ancestral populations and also significantly associated with a trait. It can use local ancestry at genomic regions to capture both common and rare variants. Prior research leveraging admixture has successfully identified the APOL1 alleles as a strong risk factor for clinically attributed hypertensive-associated CKD, FSGS, and HIV nephropathy in blacks.^12,15

The HCHS/SOL study reported here used admixture mapping analysis in a large population sample of Hispanics/Latinos to identify loci that may harbor variants that increase urine albumin excretion. We identified a new locus at chromosome 2 that harbors an Amerindian-specific variant associated with urine albumin excretion and these findings were replicated in a cohort of Pima Indians.

Results

We analyzed a sample of 12,212 Hispanic/Latino individuals from the HCHS/SOL, with mean age of 46 years (SD=14 years), and with 41% men, 20% of patients with diabetes, 28% hypertensive, and 43% with obesity. The UACR was 6.54 mg/dl (interquartile range, 4.49–12.24). Three percent of individuals had an eGFR<60 ml/min per 1.73 m² and 14.1% had increased UACR (≥17 mg/g in men and ≥25 mg/g in women).

Local ancestry was estimated as European, African, or Amerindian at each of 419,645 genotyped SNPs across the autosomes.¹⁶ Admixture mapping consisted of a joint test of the effects of all three ancestry types on log-transformed UACR at each of the 419,645 SNP positions (using a linear mixed model with multiple fixed and random effects, as described in Concise Methods). This analysis identified three regions with P value(s) <5.7 × 10⁻⁵, indicating genome-wide significance (Figure 1A). The most significant peak spans 2q11.2–q14.1 with a minimum P value achieved (P=5.6 × 10⁻⁷) at 2q13. The second spans 11q13.2–q13.4 with a peak at 11q13.3 (P=3.4 × 10⁻⁵). The third peak is located in the 16p13.3 gene region (P=3.6 × 10⁻⁵). Secondary admixture mapping analyses that tested the effect of each local ancestry type separately revealed that the 2q13 locus was significantly associated with local Amerindian ancestry, whereas the 11q13 locus was associated with both local Amerindian and local European ancestries, and the 16p13 locus was associated with local African ancestry (Figure 2). The estimates for admixture mapping are shown in the Supplemental Table 1.

Figure 1.

Manhattan plot of admixture mapping of UACR in the HCHS/SOL population, n=12,212. (A) Three genome-wide loci reached the significance threshold (P<5.7 × 10⁻⁵, dashed lines). (B) Admixture mapping conditional on rs116907128. (C) Admixture mapping conditional on rs586283. In each plot, the x axis is chromosome position and the y axis is the −log10(P value) for the joint test of variation among the three local ancestry types.

Figure 2.

Manhattan plot of admixture mapping of UACR for each local ancestry type separately in the HCHS/SOL population, n=12,212: (A) Amerindian versus non-Amerindian; (B) European versus non-European; (C) African versus non-African. x axis is chromosome position and y axis is the −log10(P value) for admixture mapping analyses. The dashed line is the threshold for genome-wide significance (P<5.7 × 10⁻⁵).

Single-variant association analyses performed using imputed genotypes within a 10 Mb window around 2q11.2–q14.1 yielded two SNPs significantly associated with UACR: rs116907128 (allele frequency for A allele =0.14; P=1.5 × 10⁻⁷) and rs586283 (allele frequency for C allele =0.35; P=4.2 × 10⁻⁷) (Figure 3A, Supplemental Material, Table 1). In analysis excluding diabetic individuals, the association of rs116907128 with UACR was unchanged (β=0.10; SEM=0.02; P=1.8 × 10⁻⁷), but the association of rs586283 with UACR was attenuated (β=0.05; SEM=0.01; P=2.1 × 10⁻⁴). Linkage disequilibrium between these two SNPs is low (r²=0.11), suggesting these are independent associations. There was no compelling evidence for associations within the regions containing the chromosomes 11 and 16 admixture signals (Figure 3, B and C).

Figure 3.

Single-variant and admixture association analyses within the regions on chromosomes 2, 11, and 16 showing admixture mapping signals in the HCHS/SOL population, n=12,212. For each plot, the x axis is chromosome position and the y axis is the −log10(P value) for admixture mapping analyses (blue lines) and single SNP association with UACR (black dots). The red dashed line is the threshold for the genome-wide admixture mapping (P<5.7 × 10⁻⁵). The gray dashed line is the threshold for association if the study was a GWAS.

Table 1.

Association findings for SNP rs116907128, which lies within the 2q11.2–q14.1 locus identified by admixture mapping

Study	Alleles	A allele frequency	UACRa		eGFR P Value	Type 2 Diabetes		Diabetic Nephropathyb
			β (SEM)	P Value		OR	P Value	OR (95% CI)	P Value
HCHS/SOL study	A/C	0.14	0.11 (0.02)	1.5 × 10−7	0.97	0.99	0.79	NA
Full-heritage Pima Indiansc	A/C	0.54	0.32 (0.13)	9.6 × 10−3	0.85	1.07	0.33	1.22 (1.01 to 1.48)	0.04

OR, odds ratio; 95% CI, 95% confidence interval; NA, not available.

^aTest for heterogeneity estimated from HCHS/SOL Hispanics/Latinos and Pima Indians not significant (Q=2.55 on 1 degrees of freedom; P=0.11).

^bDiabetic nephropathy defined as case (UACR≥300 mg/g or ESRD; n=398) versus control (UACR<300 mg/g; n=1251) for a total of n=1649.

^cUACR (n=1568); eGFR (n=1552); type 2 diabetes (total n=3699; 1724 cases; and 1975 controls).

To determine if the two identified SNPs on chromosome 2 account for the admixture mapping findings, we repeated the admixture mapping analyses using these SNPs as covariates. Adjusting for rs116907128 markedly reduced the admixture mapping signal (P for admixture =0.002) at the chromosome 2 locus (Figure 1B), whereas conditioning on rs586283 slightly attenuated the association findings (P for admixture =1.4 × 10⁻⁴) (Figure 1C, Supplemental Table 1). Analyses including both SNPs showed similar results to conditional analysis on rs116907128 (data not shown). Neither SNP was associated with eGFR in Hispanics/Latinos (rs116907128, P=0.97; and rs586283, P=0.15) or diabetes (P=0.81 and 0.46, respectively). SNP rs116907128 is monomorphic in 1000 Genome Project European and African samples, but it is a common variant among full-heritage Pima Indians (A allele frequency =0.54). In a replication analysis conducted in full-heritage Pima Indians, rs116907128 associated with UACR as a continuous trait (n=1568; β=0.32; P=0.01; P for heterogeneity Hispanics/Latinos and Pima Indians =0.11) and nominally with diabetic nephropathy in a case (UACR≥300 mg/g or ESRD) versus control (UACR<300 mg/g) analysis (n=1649; P=0.04; odds ratio=1.22; 95% confidence interval, 1.01 to 1.48). There was a nonsignificant increase in prevalence of diabetic nephropathy by duration of diabetes among Pima Indians with AA/AC genotypes compared with CC genotypes but the genotype-duration of diabetes interaction was not significant (P=0.12; Supplemental Figure 1). The rs116907128 SNP was not associated with eGFR (n=1552; P=0.85) or type 2 diabetes (n=3699; P=0.33) in full-heritage Pima Indians.

We used Haploreg¹⁷ (Epigenome Roadmap Project and the Encyclopedia Of DNA Elements [ENCODE] Project) to query the evidence for regulatory function of SNPs in the chromosome 2 region with significant admixture mapping signal. The rs116907128 position, located 5′ upstream of BCL2L11, overlaps enriched regulatory annotations for DNase I hypersensitivity sites in multiple tissues including fetal kidney and immune cells. This variant also overlaps transcription factor binding sites, including histone H3K4me1, H3K4me3, H3K27ac, and H3K9ac in several cell lines.

Discussion

This study used admixture mapping analyses to identify novel loci that may harbor variants affecting UACR and conferring risk for UACR. The main findings of our study are the identification of three novel genomic regions on chromosomes 2, 11, and 16 for which UACR is associated with local ancestry in a large and diverse population sample of Hispanics/Latinos. This approach identified an Amerindian-specific locus on chromosome 2, driven by a variant located 5′ of BCL2L11, common in Amerindians but rare in European and African populations. Amerindian genetic ancestry has been significantly associated with urine albumin excretion among Hispanics/Latinos,^8,9 and Amerindian variants have been previously shown to associate with type 2 diabetes susceptibility in Hispanics/Latinos.^18,19 The admixture mapping approach is on the basis of local ancestry—i.e., the number of alleles at a specific variant position that derive from each member of a given set of ancestral populations (in this case Amerindian, West African, and European). This approach also accounts for global ancestry (across all autosomes) as a fixed covariate in the linear mixed-model regression. In contrast, although GWAS typically adjust for global ancestry, they do not directly account for local ancestry in detecting genotype-phenotype associations. Our findings provide important information on the underlying genetic architecture of ancestry-specific genetic risk for UACR. Here, we have identified three specific loci that may contribute to associations between global ancestry and UACR or increased UACR. These results support further investigation of Amerindian ancestry to better understand Hispanic CKD risk.

To date, the only other validated genome-wide significant loci for UACR are CUBN and HBB (which is related to sickle cell trait^10,11). Two additional UACR loci located at 2q21 (HS6ST1) and 11q14 (near RAB38/CTSC) were recently described in a GWAS of diabetic individuals (n=5509–5825) of European ancestry.²⁰ These regions do not overlap our identified loci. We previously have shown significant genome-wide associations of CUBN and HBB with UACR in the HCHS/SOL study.²¹ The most significant CUBN variant has similar allele frequency in West African (AFR=0.17), European (EUR=0.12), and Hispanic (0.14) populations using 1000 Genomes data. Therefore, this variant will not be expected to be identified in admixture mapping.

Our main admixture mapping finding is in a region containing BCL2L11, which encodes a proapoptotic protein that belongs to the BCL-2 protein family. Several lines of evidence suggest a role of this gene and its protein in kidney development and kidney disease states. MicroRNA (miRNA)–mediated regulation of Bcl1211 expression in mice plays an important role in nephron progenitor survival during kidney development.²² miRNAs are involved in post-transcriptional repression of target mRNAs. Loss of miRNAs in nephron progenitors increases apoptosis and elevates the expression of protein Bim (also known as BCL2L11) leading to a decrease in nephron number.²² Bim null mice manifest systemic autoimmune disease and immune complex GN.²³ In an experimental study of diabetic nephropathy, increased advanced glycation end products promoted Bcl2l11 expression, leading to podocyte apoptosis.²⁴ Establishing the functional role of this gene in CKD will require targeted in vivo studies using knockout and/or transgenic mouse models, in addition to longitudinal population studies. The variant we identified, rs116907128, located at 5′ of BCL2L11, overlaps enriched regulatory annotations for DNase I hypersensitivity sites and histone marks in fetal and adult kidney, and in several other cell lines including immune cells. This evidence suggests a regulatory function in these cells and tissues. Among full-heritage Pima Indians, this variant is associated with increased UACR (n=1568; P=0.01) and diabetic nephropathy (n=1649; P=0.04), but not with eGFR or type 2 diabetes. In HCHS/SOL Hispanics/Latinos, there is also no association with either type 2 diabetes or eGFR. Interestingly, the “A” allele for rs116907128 is more common in Pima Indians than admixed Hispanics/Latinos, and the association with UACR was stronger in full Pima Indians.

The resolution of admixture mapping depends on the lengths of local ancestry tracts (i.e., the contiguous positions on a given chromosome having the same local ancestry). In Hispanics/Latinos, local ancestry tracts are relatively large because the admixture among European, African, and Amerindian ancestors was relatively recent so recombination has not yet broken them down into small segments. Therefore, the regions with genome-wide significance in admixture mapping are relatively large. Here, we used single-variant association testing to further localize the genetic effect(s) within these broader regions. There is an SNP within the admixture mapping signal for UACR on chromosome 2 (rs116907128), which has a P value for association with UACR of 1.5 × 10⁻⁷. This P value is significant after multiple-testing adjustment within that region, but it does not meet the genome-wide significance level of 5 × 10⁻⁸. Therefore, it was not identified as in our previous GWAS for UACR in HCHS/SOL.²¹ This example illustrates how admixture mapping can be complementary to genome-wide association testing of single variants as well as localizing genetic factors contributing to global ancestry associations with diseases and their component traits.

In the regions detected on chromosome 11 and 16, we found no SNPs that passed the region-specific multiple-testing adjustment. This could be due to low power in association testing (e.g., due to relatively low combined allele frequencies of the tag SNPs), or due to lack of good proxies of the causal SNP(s). Further studies will require sequencing the chromosome 11 and 16 regions to identify variants (common and rare) accounting for these admixture signals in Hispanics/Latinos. In addition, fine-mapping of loci associated with Amerindian local ancestry are likely to be more successful when performed in American Indian populations.

Our study is limited by the available genomic markers (imputed and genotyped) for fine-mapping of regions. Although we found strong associations of rs116907128 (and rs586283) at the chromosome 2 locus, which accounted for most of the local ancestry association in the admixture mapping, there was still a fairly small admixture P value after adjusting for the SNPs, suggesting that there must be other variants in the region also contributing to the signal, or that the SNPs are only in partial LD with one or more causal variants. Admixed mapping approaches capture both rare and common variants, but association analyses are underpowered to detect low-frequency variants. Further studies using sequencing are needed to identify additional low-frequency and rare variants in the region that could account for the admixture mapping findings. The replication of our findings in Pima Indians supports the role of the locus on UACR variation. Future studies will examine longitudinal changes in kidney function and UACR to better characterize the relevance of BCL2L11 to increased UACR and CKD, which may provide important clues to the clinical effect of this gene in CKD. HCHS/SOL is currently examining participants in a second visit, from which data could be used for these follow-up studies.

In summary, we identified three novel loci for UACR using admixture mapping in Hispanics/Latinos, including a region on chromosome 2 within which the signal is driven by an Amerindian-specific variant near the BCL2L11 gene, which has a UACR association that replicated in Pima Indians. Our study provides important information on the presence of Amerindian-specific genetic variants associated with UACR in Hispanics/Latinos and American Indians.

Concise Methods

Study Population

HCHS/SOL is a community-based cohort study of 16,415 self-identified Hispanic/Latino adults aged 18–74 years from randomly selected households near four United States field centers (Chicago, Bronx, Miami, and San Diego) with baseline examination (2008–2011) and yearly telephone follow-up assessment for at least 3 years as previously described.²⁵ Participants self-identified as having Hispanic/Latino background, with the largest groups being Central American (n=1730), Cuban (n=2348), Dominican (n=1460), Mexican (n=6471), Puerto-Rican (n=2728), and South American (n=1068). The clinical examination included physical measures, behavioral/lifestyle factors, and sociodemographic assessments,²⁵ in addition to collection of fasting blood and spot urine samples. The study was approved by the Institutional Review Board at each participating institution, and all participants provided written informed consent. The final sample used for analyses was 12,212 individuals.

Genotypes and Imputation

HCHS/SOL participants were genotyped using a Custom Illumina array containing a total of 2,536,661 SNPs of which 2,427,090 are from a standard Illumina Omni2.5M array and the remaining 109,571 are custom SNPs. Quality control was completed using methods previously described.²⁶ Genotype imputation was performed using the 1000 Genomes Project phase 1 cosmopolitan reference panel. We estimated kinship coefficients and principal components using methods previously described.¹⁴

Phenotypes and Covariates

Albumin (milligrams per deciliter) and creatinine (grams per deciliter) were measured in urine specimens using an immunoturbidometric method and a creatinase enzymatic method, respectively. Log-transformed UACR, defined as the ratio of albumin-to-creatinine, was the main phenotype. For variants associated with UACR within regions, we also examined the association with eGFR, estimated from the Chronic Kidney Disease Epidemiology Collaboration equation on the basis of the measured serum creatinine, sex, age,²⁷ and diabetes, with cases defined as having fasting time >8 hours and fasting glucose levels ≥126 mg/dl; or fasting≤8 hours and fasting glucose ≥200 mg/dl; or postoral glucose tolerance test glucose ≥200 mg/dl; or hemoglobin A1C ≥6.5%; or if on current treatment with a hypoglycemic agent.

Statistical Analyses

The HCHS/SOL has approximately 2000 related individuals, and a complex sampling design. To account for survey design and for correlation structure due to relatedness, we fit variance components for kinship, household, and census block group using linear mixed models.¹⁴ All analyses were additionally adjusted for age, sex, center, and the five first principal components. Sampling weights were also included as a fixed effect to account for the sampling design.¹⁴

Admixture Mapping

We inferred local ancestry at a set of 236,456 SNPs in common between HCHS/SOL and reference-panel datasets (selected populations from the Human Genome Diversity Panel,²⁸ HapMap 3,²⁹ and 1000 Genome phase 1³⁰) used to detect and estimate European, West African, and Amerindian ancestry at genomic locations, as previously described.¹⁶ We ran an unsupervised analysis using ADMIXTURE software³¹ with K=3 populations, and retained individuals with >90% estimated ancestry from one of the inferred ancestral populations. We phased the HCHS/SOL data combined with the reference panels using Beagle 4.0³² and inferred local ancestry using RFMix.³³ Using these inferred local ancestries,¹⁶ we implemented a genome-wide admixture mapping approach for UACR utilizing a linear mixed model as described above with a joint test for the local ancestry count from all three ancestries (Amerindian, West African, European). As a secondary analysis, we performed admixture mapping, testing the local ancestry of each ancestral group separately. On the basis of previously reported simulation analyses in HCHS/SOL,³⁴ a P value of 5.7 × 10⁻⁵ controls the family-wise error rate of admixture mapping at level 0.05.

Association Mapping within a Region

We used linear mixed models as described above within each identified region from the admixture mapping to test the association of each variant with UACR. P values were adjusted for the number of tests performed.

Functional Annotation of Variants

We used Haploreg to examine the overlap of two noncoding SNPs of interest on chromosome 2 with chromatin state and protein binding annotations. These annotations were obtained from the Roadmap Epigenomics and ENCODE projects for specific tissues and cells.¹⁷

Replication

Replication of rs116907128 was assessed in a population-based sample of Southwest United States American Indians who are full Pima Indian heritage and had participated in a longitudinal study conducted by the National Institute of Diabetes and Digestive and Kidney Diseases.^35,36 This included n=1568 individuals for UACR; n=1552 for eGFR; n=3699 for diabetes case-control; and n=1649 for diabetic nephropathy case-control. Rs116907128 was previously genotyped using a Pima Indian custom Axiom array (Affymetrix). All genotypes analyzed on this array met quality control metrics (call rate ≥90%; discrepant rate ≤2 pairs among 100 blind duplicate pairs; and lack of deviation from Hardy–Weinberg equilibrium with a P>10⁻⁴). The SNP associations with UACR and diabetic nephropathy were performed using linear and logistic models, respectively, adjusted for age and sex. We tested the heterogeneity across Hispanics/Latinos and Pima Indian estimates using the Cochran Q test.

Disclosures

None.