Genetic association of fetal-hemoglobin levels in individuals with sickle cell disease in Tanzania maps to conserved regulatory elements within the MYB core enhancer

Background Common genetic variants residing near upstream regulatory elements for MYB, the gene encoding transcription factor cMYB, promote the persistence of fetal hemoglobin (HbF) into adulthood. While they have no consequences in healthy individuals, high HbF levels have major clinical benefits in patients with sickle cell disease (SCD) or β thalassemia. Here, we present our detailed investigation of HBS1L-MYB intergenic polymorphism block 2 (HMIP-2), the central component of the complex quantitative-trait locus upstream of MYB, in 1,022 individuals with SCD in Tanzania. Methods We have looked at 1022 individuals with HbSS or HbS/β0 in Tanzania. In order to achieve a detailed analysis of HMIP-2, we performed targeted genotyping for a total of 10 SNPs and extracted additional 528 SNPs information from a genome wide scan involving the same population. Using MACH, we utilized the existing YRI data from 1000 genomes to impute 54 SNPs situated within HIMP-2. Results Seven HbF-increasing, low-frequency variants (β > 0.3, p < 10−5, f ≤ 0.05) were located in two partially-independent sub-loci, HMIP-2A and HMIP-2B. The spectrum of haplotypes carrying such alleles was diverse when compared to European and West African reference populations: we detected one such haplotype at sub-locus HMIP-2A, two at HMIP-2B, and a fourth including high-HbF alleles at both sub-loci (‘Eurasian’ haplotype clade). In the region of HMIP-2A a putative functional variant (a 3-bp indel) has been described previously, but no such candidate causative variant exists at HMIP-2B. Extending our dataset through imputation with 1000 Genomes, whole-genome-sequence data, we have mapped peak association at HMIP-2B to an 11-kb region around rs9494145 and rs9483788, flanked by two conserved regulatory elements for MYB. Conclusions Studies in populations from the African continent provide distinct opportunities for mapping disease-modifying genetic loci, especially for conditions that are highly prevalent there, such as SCD. Population-genetic characteristics of our cohort, such as ethnic diversity and the predominance of shorter, African-type haplotypes, can add to the power of such studies.


Background
Sickle cell disease (SCD) is a hemoglobin disorder caused by the Glu6Val mutation in the β chain of adult hemoglobin. The resulting hemoglobin variant, HbS, is prone to polymerization, disrupting red blood cell shape, function and life span. SCD is prevalent in Sub-Saharan Africa, where it is a significant contributor to childhood mortality [1]. In Tanzania, 8,000-11,000 affected children are born annually [2]. The most common and severe forms of the disease are due to homozygosity for the mutation (HbSS) or compound heterozygosity with β 0 thalassemia (HbS/β 0 thalassemia). Where newborn screening and prophylactic penicillin are available, childhood mortality due to SCD is significantly reduced, but patients nevertheless remain at risk for chronic complications and premature death. The disease is milder in those patients that carry significant amounts of fetal hemoglobin (HbF) in their circulating red blood cells [3]. Similar to healthy populations, HbF persistence in patients with SCD is partially genetically controlled, and three HbF quantitative-trait loci (QTLs) -HBG2 [4,5], BCL11A [6,7] and HBS1L-MYB [8] -have been identified. Knowledge of the genetic factors underlying HbF persistence is helping to interpret the clinical variability of SCD and has led to the identification of novel molecular targets for the therapeutic reactivation of HbF.
HBS1L-MYB is unique among the HbF modifier loci because it has marked pleiotropic effects, i.e., in healthy individuals it affects general hematological parameters [9] as well as HbF. It has been postulated that changes in HbF levels caused by this locus are secondary to altered kinetics of erythropoiesis [10]. The locus consists of several linkage disequilibrium (LD) blocks of common variants, which affect erythroid traits independently [8]. The most effective of these, termed HMIP-2 (HBS1L-MYB intergenic polymorphism, block 2) has been shown to influence disease severity in patients with SCD [11] and β thalassemia [8,12]. HMIP-2 variants reside within the core enhancer for MYB [13], a key hematopoietic regulator gene [14]. It is divided further into sub-loci HMIP-2A and -2B, which provide independent HbF association in African populations, including SCD patients [11,[15][16][17][18]. A 3-bp deletion (rs66650371) at HMIP-2A is suspected to directly cause HbF variability [19], but is independent of the trait association seen at HMIP-2B. Therefore, causative variants acting at HMIP-2B are still to be discovered.
To better define the HbF association signal at HMIP-2B, and to identify candidate variants for trait causation, we dissected HMIP-2 and its effect on HbF persistence in a large SCD patient cohort from Tanzania. The Tanzanian population is well-suited to genetic fine-mapping studies, with a marked ethnic diversity [20,21] and the increased mapping resolution that is characteristic for African chromosomes [22,23].

Study subjects, sample collection and phenotyping
Only patients with Hb SS or HbS/β 0 thalassemia genotype were included in this study. Enrollment of patients, diagnosis and confirmation of sickle phenotypes as well as the quantification of hemoglobin subtypes were performed as previously described [24]. Informed consent was obtained for each patient and ethical approval given by the Muhimbili University Research and Publications Committee (MU/RP/AEC/VOLX1/33). During followup clinics, a 2-ml blood sample was collected from nontransfused SCD patients (confirmed Hb SS genotype) who are not on hydroxyurea treatment. This study includes 1,022 individuals with HbF measured (by HPLC, Variant I, Biorad, Hercules, CA, USA) at the age of 5 years or older. The median age of the SCD population is 11 years; males and females are represented equally. HbF values vary significantly, with a median of 5.4% (of total hemoglobin).
GWAS data was processed with the PLINK software package (http://pngu.mgh.havard.edu/purcell/plink/). Test for genetic association with ln[%HbF], including conditional analysis, was performed with STATA v12 (Stata Corp, College Station, TX) using multiple linear regression with age and sex as covariates. Haplotype relative effects were estimated using multifactor ANOVA in R (http://www.r-projects.org/), correcting for pair-wise comparison using Tukey's method, including age and sex as covariates.

African-specific trait association at HMIP-2B
In contrast to HMIP-2A, a good candidate variant for causing the HbF association at HMIP-2B has so far not been identified. We therefore studied HbF association at this sub-locus in more detail. We included additional variants from 1000 Genomes project sequence data [28] through imputation with YRI (Yoruba, West African) haplotypes, which are well-matched with our population regarding the allele frequencies of regional markers [18]. The YRI dataset harbors 51 polymorphic variants within the HMIP-2B region (chr6:135,426,573-135,435,501), and another 17 such variants in an additional 5-kb fragment we decided to include on the 3' side of the locus, adjacent to MYB. 15 patients carrying European-type high-HbF haplotypes (tagged by rs1376090 [18]) were excluded from imputation and analysis to make full use of the increased mapping resolution of the shorter African haplotypes, with the goal to localize the QTL more precisely. Association analysis subsequent to imputation identified an additional SNP with strong HbF association, rs9483788 ( Figure 2). The area of significant association (p < 0.01) thus extends between chr6:135,427,000 and 135,438,000, with the peak between MYB enhancer elements −71 and −63 (Figure 2).

HMIP-2 haplotypes
To dissect the haplotype architecture underlying the trait association pattern at HMIP-2, we phase-aligned genotypes for the seven strongly-associated markers. To relate our data to findings in other populations, we also included rs9376090, which tags European and Asian high-HbF alleles, and rs4895441, which is part of the HMIP-2B sub-locus in other SCD patient populations.
The most prevalent haplotype carried the reference (low-HbF) allele in all positions, representing the global ancestral situation, universally associated with low HbF levels [18]. The other haplotypes contained high-HbF associated alleles in at least one position (Figure 3). A'Eurasian-type' haplotype [18] (high-HbF alleles at all seven positions, including the ancestry-informative marker rs9376090) was detected at low frequency, but with high HbF levels (9.3%, median, Figure 4).
The remaining haplotypes belonged to two minor clades more common in African populations, one with HbFpromoting alleles at HMIP-2A only (termed clade ' A -b'), and one with HbF-promoting alleles at HMIP-2B only (clade 'a -B'). In Tanzanian patients, ' A -b' was represented by a single haplotype, whereas the 'a -B' clade was more diverse (haplotypes 'a -B 1 ', 'a -B 2 ', 'a -B 3 '). ' A -b', 'a -B 2 ' and 'a -B 3 ' haplotypes all significantly boost HbF levels, compared to the ancestral haplotype 'a -b' ( Table 2), while 'a -B 1 ' does not. 'a -B 1 ' contains only a single nonancestral allele, rs4895441-G, which is not strongly HbF-associated in Tanzanian patients. 'a -B 3 ' , which contains two high-HbF associated alleles, rs9494145-C and rs9483788-C, significantly increases HbF levels over the ancestral haplotype. A stronger effect is detected when all four HbF-boosting alleles, rs9389269-C, rs9402686-A, rs9494145-C and rs9483788-C, are present (in 'a -B 2 ', Table 2).

Figure 2
African-specific association with HbF at HMIP-2B in Tanzanian SCD patients. Association analysis was performed with the same individuals as in Figure 1, but 15 patients with Eurasian-type high-HbF haplotypes ('A -B', carrying the ancestry-informative allele rs1376090-C) were excluded. Data for 54 markers imputed from 1000 Genomes YRI (Yoruba, Ibadan, Nigeria) sequence were added to the analysis. While the exclusion of selected individuals resulted in a weaker overall association signal, the potential to map African-specific variants at higher resolution was considered more important at this stage. Association scores are shown; unconditioned (black dots) and conditioned on the first three principal components derived from genome-wide SNP data (open circles) [25]. Shown also is the location of the conserved MYB upstream regulatory elements −71 and −63 [13].
MYB enhancer region [13]. Some of these variants have been also associated with white blood counts, mean cell volume and mean cell hemoglobin in our population [36]. Our interest was focused on sub-locus HMIP-2B, where a causative variant has not yet been identified. After excluding patients with longer, 'Eurasian'-type [18] high-HbF haplotypes and including imputed variants from the YRI (Yoruba, 1000 Genomes sequence data) population, we determined the most likely map location of HMIP-2B as an 11-kb segment including the enhancer core element −71 and the interval between elements −71 and −63 (Figure 2), where peak association (rs9494145, rs9483788) was detected. The two HbF-boosting haplotypes underlying this association peak, 'a -B 2 ' and 'a -B 3 ' , share rs9494145-C and rs9483788-C (Figure 3). 'a -B 2 ', which contains all four HbF-boosting alleles (rs9389269-C, rs9402686-A, rs9494145-C and rs9483788-C), has the stronger effect of the two. This means that none of the four strongly trait-associated SNPs detected at HMIP-2B in Tanzanian patients appears to fulfill the conditions for being the singular causative variant, i.e. both, being necessary to show a significant effect and sufficient to produce the maximum genetic effect originating from this sub-locus. Thus, additional variants, not present in the 1000 Genomes dataset, might contribute to trait variability.
Long, 'Eurasian-type' (with high-HbF associated alleles across all of HMIP-2 [18]), high-HbF haplotypes were present in the patient cohort at a low frequency. These haplotypes are tagged by the ancestry-informative allele rs9376090-C (Figure 3). 24% of individuals with such haplotypes reported Arabic parental ethnicity, compared to 2% in the general cohort. The high HbF levels we observed in such patients (a median of 9.3% in ' A-B'/'a-b' heterozygotes, Figure 4) are likely due to the presence of Figure 3 HMIP-2 haplotypes detected in Tanzanian SCD patients. Nine critical variants at HMIP-2 were used to investigate haplotypes present at the locus (alignment by Phase v. 2.1). Haplotypes were assigned to the principal clades described previously [18], dependent on whether they contain HbF-increasing alleles (shaded in gray) at HMIP-2A (capital 'A') or HMIP-2B (capital 'B'). a-b: ancestral haplotype present in all human populations, composed entirely of low-HbF associated alleles; A-b: HbF increasing alleles at HMIP-2A, but lacking the European/Asian-specific allele rs9376090-C; a-B: a-B 1 : one HbF-increasing allele at HMIP-2B, rs4895441-G, a-B 2 : HbF-increasing alleles across HMIP-2B; a-B 3 : two HbF-increasing alleles at HMIP-2B, rs9494145-C and rs9483788-C, A-B: Eurasian haplotype, HbF-increasing alleles across all of HMIP-2; Rare haplotypes (frequency < 0.5%) are not shown. *imputed Figure 4 HbF levels according to HMIP-2 haplotype combination. %HbF of total hemoglobin was measured by HPLC in patients carrying one copy of the ancestral haplotype, with diverse haplotypes on the sister chromosome. Boxes show median and inter-quartile range. Relative effects of these haplotypes are shown in Table 2. "Difference" represents the difference in effect size between the two haplotypes on log-transformed HbF levels; negative difference values indicates that the effect size of the second haplotype is bigger than that of the first. 'lower' and 'upper' represent boundaries for family-wise 95% confidence intervals. 'p adjusted' is the P-value adjusted for age, sex as well as multiple testing. Three haplotypes (shown in bold) have significant HbF-boosting effects at the 0.05 level.
the 3-bp deleted allele at HMIP-2A and possibly another functional allele at HMIP-2B. Population stratification might also contribute to higher levels of HbF: Arab/Indian sickle mutation haplotypes on chromosome 11 are known to result in milder disease and high HbF levels [37]. We also observed a residual association after conditioning on HMIP-2A and HMIP-2B. We suspect that these are part of a group of linked SNPs that overlaps the physical location of HMIP-1, a HbF QTL detected upstream of HMIP-2 ( Figure 1A) in the European population [8]. However we didn't feel we have the power to investigate this further with the present dataset.

Conclusions
We have localized HMIP-2B, a QTL for fetal-hemoglobin persistence, to an 11-kb region within the core enhancer for MYB. So far, we have not identified a likely functional variant within or at this locus. Further studies will involve extended sequence analysis in groups of patients carrying a-B 2 and a-B 3 haplotypes.