Homozygosity for a Haplotype in the HBG2-OR51B4 Region is Exclusive to Arab-Indian Haplotype Sickle Cell Anemia

Introduction

Homozygosity for rs334 (GAG-GTG, glu6val) or sickle cell anemia is common in Saudi Arabia. In the Eastern Province the sickle hemoglobin (HbS) gene is usually on the autochthonous Arab Indian (AI) β-globin gene (HBB) haplotype [1]. The higher fetal hemoglobin (HbF) level in the AI haplotype is often associated with milder disease compared with other HbS-associated haplotypes, especially in children [2, 3].

The Xmn1 G-A (C-T) polymorphism (rs7482144) is present in both the AI haplotype and the Senegal haplotype. Rs10128556, which is in linkage disequilibrium (LD) with rs7482144, could be the functional SNP of the Senegal haplotype [4]. The minor alleles of these SNPs also characterize the AI haplotype where adults had ~20% HbF compared with ~10% HbF in the Senegal haplotype. This suggests that elements linked to rs10128556 and rs7482144 in the AI but not the Senegal haplotype might affect HbF gene expression.

We compared the genome-wide distribution of SNPs in Saudi patients with the AI haplotype and Benin haplotype. Variants distinguishing these populations were limited to chromosome 11p15.5. Annotation of these SNPs narrowed the selection to 3 that we analyzed jointly using haplotype analysis. Homozygosity for a T/A/T haplotype of rs16912979 (in HS-4 of the LCR), rs7482144 and rs10128556 was exclusive to the AI haplotype. The AI haplotype might include a functionally important sub-haplotype that accounts for high HbF.

Methods

Patients

We recruited 245 sickle cell disease patients from the Eastern and Southwestern Provinces of Saudi Arabia for a genome-wide association study (GWAS). Forty-two AI (cohort 1) and 71 Benin cases were compared. Sixty-two additional 62 AI Saudi cases (cohort 2) and 41 Indian AI haplotype cases were studied by GWAS or targeted SNP analysis [5, 6]. Fourteen Saudi AI haplotype homozygotes had whole genome sequencing (WGS) including 7 AI homozygotes with high and 7 with low HbF, 3 Indian AI homozygotes, 3 African American Benin haplotype homozygotes selected because of unusually high HbF [7] and 1 African American Senegal haplotype homozygote. GWAS in 339 African American HbS homozygotes of diverse haplotypes examined the presence of SNPs found in the Saudi cohorts [8]. (Table I)

Table I.

Patient cohorts

	Saudi AI Cohort 1 (n=42)	Saudi AI Cohort 2 (n=62)	Indian AI (n=41)	AA SEN.SEN (n=8)	AA SEN.het (n=75)	Saudi BEN.BEN (n=71)	AA BEN.BEN (n=232)	AA CAR.CAR (n=23)
HbF (%)	17.6±5.0	18.8±7.5	23.0±4.8	8.4±4.5	8.5±4.7	10.8±4.6	6.8±5.7	5.5±3.7
Age (yrs.)	26.4±11.1	23.9±9.6	14.6±4.7	32.1±12.3	19.4±13.9	18.6±11.0	22.0±16.3	20.0±16.8

	Saudi AI		Indian AI (n=3)	AA BEN.BEN (n=3)	AA SEN.SEN (n=1)
	↑HbF (n=7)	↓HbF (n=7)
HbF (%)	23.5±2.6	8.2±1.3	26.0±4.5	19.8±0.4	16
Age (yrs.)	25.9±6.8	34.1±10.3	22.7±5.5	26.0±3.6	5.9

A. Saudi AI (cohort 1) and Saudi Benin haplotype patients were selected by their region of origin in the Eastern and Southwestern Provinces of Saudi Arabia. Saudi AI (cohort 2) was an independent patient group. Indian AI patients were from Raipur, Chattisgarh, India. Other cohorts include African Americans (AA) homozygous for Benin, Bantu (CAR) and Senegal haplotypes and Senegal haplotype compound heterozgotes with Benin and CAR haplotypes [18]. AI-AI homozygotes; SEN.SEN-Senegal homozygotes; SEN.het-Senegal compount heterozygotes; BEN.BEN-Benin homozygotes; CAR.CAR-CAR homozygotes.Values are mean ± SD. B. In addition to the 14 Saudi AI haplotype patients, African American samples chosen for whole genome sequencing included 3 Benin and 1 Senegal haplotype homozygotes. The Benin cases were selected for WGS because of their unusually high HbF, as reported previously [7].

HBB haplotypes

The AI haplotype was ascertained directly by genotyping rs7482144, rs3834466 and rs549964658; other haplotypes were determined as described [6, 9, 10].

GWAS

Genotyping used Illumina SNP arrays [8].

Imputation

To capture all SNPs in the HBB gene cluster region, SHAPEIT and IMPUTE2 were used to impute non-genotyped SNPs using reference files from the1000 Genome phase I database [11]. Only variants with a quality score threshold of ≥0.90, a minor allele frequency (MAF) difference >0.85 between the Saudi AI cohort1 and Saudi Benin haplotype cohorts, and SNPs annotated with RegulomeDB that were examined only in blood cells were included in our final analysis (Fig. S1)

Whole genome sequencing

Paired-end 100 bp reads were sequenced on the Illumina HiSeq 2000 platform to a depth of 40X per individual. (supplemental material) [12, 13].

Amplification refractory mutation system (ARMS)

ARMS was used to genotype rs16912979 in 41 AI haplotype Indian HbS homozygotes (supplementary material).

Epigenetics

Transcription factor binding data sets from the RegulomeDB of the ENCODE Consortium and Roadmap Epigenomics Project were searched for enhancer marks and transcription factor binding in blood cells [14].

Statistical analysis

Analysis was done in PLINK using logistic regression with additive coding of the SNPs [15]. Genomic control (GC) approach corrected the genomic inflation caused by sub-population stratification. QQ plots and genomic control estimates were generated using R (Fig. S2). Regional association plots were generated by the LocusZoom [16]. Haplotype analysis was conducted using the haplostats package in the R software as described [17].(supplemental material)

Results

Directly genotyped SNPs

Two hundred-twenty three variants in chromosome 11p15.5 from positions 3.5 to 6.5 mb distinguished Saudi AI (cohort 1) and Benin cases (p-values 9.6E-07-2.7E-45). (Fig. 1A, Table SI). Thirteen SNPs were present in all Saudi AI haplotype but rare in the Saudi Benin haplotype (allele frequency < 0.05) (Table SII, Fig. S3). These results were replicated in the 62 AI haplotype cohort 2 cases and in 14 Saudi and 3 Indian cases genotyped by WGS. The 13 SNPs were not present in the 3 high HbF African American Benin and 1 Senegal haplotype samples and were rare or absent in 93 Senegal and 606 Benin haplotype chromosomes. A regional LD plot for rs16912979 is shown in Fig. S4. MAF of SNPs in MYB, BCL11A and KLF1 were similar in AI and Benin haplotype cohorts.

Figure 1A.

Manhattan plot from the GWAS comparing Saudi Eastern AI vs. Saudi Southwestern Benin haplotype patients. P-values (−log10 P) of 599,131 SNPs after correction by Genomic Control is plotted against its physical chromosomal position. Odd chromosomes are in blue and even chromosomes in orange. Genome-wide significant variants separating these populations are clustered in chromosome 11p15.5.

Imputed SNPs

Ten SNPs were between OR51B4 upstream of HBG2 and OR52A5 downstream of HBB. Rs113622911 at coordinate 5269855 and rs59495893 at coordinate 5274126 were about 20 kb upstream of HBG1 in an area with H3K27Ac marks and POLR2A binding but were upstream of the canonical promoters of this gene. (Fig. 1B)

Figure 1B.

Epigenetic marks of AI haplotype-specific and other SNPs. Shown in the 1st track are the chromosomal locations of the genes of the HBB cluster including the upstream (OR51B4) and downstream (OR52A5) olfactory receptors genes. A blue box shows the position of hypersensitive site 4 (HS-4) of the LCR. The 2nd track in light blue shows marks for histone H3 acetyl lysine 27 (HeK27Ac) throughout this region. Track 3 (DNase) shows DNase1 hypersensitive sites and track 4 (TFBS) shows binding sites for the transcription factors that include GATA1, GATA2, POLR2A and JUND. Below this are the locations of the SNPs detected by GWAS (black arrowheads (Table SII). SNPs shown in red arrowheads form a triallelic haplotype whose minor alleles are present only in the AI haplotype. Blue arrowheads show the positions of imputed SNPs that differ between Saudi AI and Saudi Benin haplotype patients (see text).

Epigenetic marks

Thirteen SNPs annotated by their putative functional role using the RegulomeDB database (Table SII). Rs16912979, rs4910743 and rs4601817 were within the HBB gene cluster LCR; rs16912979 was located in DNase1 HS-4. Rs6912979 lies with a region of H3K27Ac marks and strong binding signals for POLR2A, GATA1, GATA2 and JUND (Fig. 1B); rs4601817 has weak binding signals for JUND and POLR2A; rs4910743 is in a region with H3K27Ac marks.

A unique AI haplotype

The epigenetic marks associated with rs16912979 and its presence in LCR HS-4 suggested this SNP as a component of a putative functional AI haplotype that included rs7482144 and rs10128556. Homozygosity for the minor allele of these 3 SNPs was limited to individuals with this haplotype. The T allele of rs16912979 was present in all 46 Bantu (CAR) haplotype chromosomes; however, these chromosomes had the T/T/C haplotype. (Table SIII)

Discussion

The Xmn1 G-A polymorphism or rs7482144, a marker of the Senegal HBB haplotype, was associated with increased HBG2 expression [18]. Both rs7482144 and rs10128556 were also present in AI haplotype sickle cell anemia. These two SNPs, along with rs16912979, constituted a unique haplotype that included or was in LD with functional elements that might contribute to high HbF in AI haplotype. Homozygosity for this T/A/T haplotype distinguished the AI from all other HBB haplotypes. This suggested that maximum cis-acting modulation of HbF requires elements of both the AI and Senegal haplotype. Other variants exclusive to the AI haplotype within this region could also be the functional elements.

Transcription factor binding data and enhancer marks for some of the AI haplotype-associated SNPs suggested the presence of functional variants in this region and a haplotype effect on HBG expression [14]. We focused on rs16912979 because of its location in HS-4 of the LCR and strong binding signals for GATA1 and GATA2 and POLR2A in K562 cells. Recruitment of RNA polymerase II (Pol II) to the LCR, which is dependent on GATA1, is important for transcriptional activation of the downstream globin genes [19]. Relative concentrations of the GATA transcription factors play an important role in in the regulation of HBG and HBE expression [20]. A binding signal for the large subunit of Pol II (POLR2A) is also present [21, 22]. Interrogation of all SNPs present at high frequency in the AI haplotype found many in regions with H3K27Ac marks in erythroid cells suggesting the open chromatin characteristic of an active enhancer [23].

The T/A/T sub-haplotype might mark a functional domain for the cis-acting regulation of HBG2 expression [24]. HS-4, is required for high-level globin gene expression in definitive erythroid cells and contains a strictly conserved GATA1 binding site [25]. The distal LCR physically contacts the proximal globin gene promoters via chromatin looping that is developmentally and stage-specifically regulated [26]. DNA sequence motifs that are most conserved include GATA sequences in HS-2, HS-3 and HS-4, KLF1 binding sites in HS-2 and HS-3, and an E-box motif in HS-2 [27]. GATA1 is required for chromatin loop formation between hypersensitive sites and gene promoters [28]. LCR looping to globin gene promoters is facilitated by the LDB1/LMO2/GATA-1/TAL-1 erythroid specific protein complex (Lbd1 complex) [29-31]. TAL1 is a transcription factor that binds to regulatory regions of many erythroid genes as part of a complex with GATA1, LMO2 and Ldb1; TAL1 binds at HS-4, HS-2 and HS-1 [29]. TAL1 overexpression increases its occupancy at HS-4 and HS-2 enhancing the expression of HBG. GATA1 and NF-E2 are both required for chromatin loop formation between the LCR and the active γ-globin genes in K562 cells [28]. RNA sequencing revealed a transcript from this region in human CD34⁺ cells and non-coding RNAs might play a role in modifying histones across the LCR and looping with HBG2 promoter [32]. In adult erythroblasts that express HBB, forced chromosome looping using the self-association domain of the Ldb1 complex established strong interactions of the HBG2 promoter with HS-2, −3 and −4, diminished interactions with the adult globin genes, and increased HBG expression to about 85% of total globin [33].

Variants cis to HBB that are not exclusive to the AI haplotype, in the context of homozygosity for the T/A/T haplotype might also account for further modulation of HbF. The T/A/T haplotype or a more extended haplotype of SNPs in LD, might be required for optimally functional looping of the LCR to the HBG2 promoter and its robust transcription (Fig. S5)

These genetic association studies provide a rationale for functional studies of HBG2 expression in wild-type and T/A/T haplotype erythroblasts and mechanistic studies like chromatin conformation capture experiments, to evaluate the role of chromatin looping as a mediator of the T/A/T haplotype effects on HbF.