Genome-wide association for hemoglobin levels was performed in 6,316 Europeans and 9,685 Indian Asians participating in the London Life Sciences Population (LOLIPOP) study and the North Finland Birth Cohort of 1966 (NFBC1966). All LOLIPOP participants were resident in London, UK. Within this study, participants of European ancestry self-reported as white and born in Europe on a questionnaire; those of Indian Asian ancestry reported having all four grandparents born on the Indian subcontinent. All participants in the NFBC were of European ancestry. Clinical characteristics of participants, and the genotyping platforms used, are summarized in the Supplementary Methods and Supplementary Table 1. Imputation was used to infer missing genotypes, using the HapMap CEU sample as reference for Europeans and pooled founder haplotypes from all three HapMap populations as reference for Indian Asians (HapMap build35, dbSNP build 125)⁵. Imputed SNPs with minor allele frequency (MAF) <0.01 or low quality score (r² < 0.30) were removed.

Single SNP marker tests were performed for association with hemoglobin level using linear regression under an additive genetic mode and adjusting for age and sex. Population substructure was characterized using principal components analyses⁶ and was included as covariates in the regression models. Results for NFBC1966 Europeans, LOLIPOP Europeans and LOLIPOP Indian Asians were analyzed separately; results were then combined between studies using z scores weighted to the square root of sample size. Quantile-quantile plots showed good adherence to null expectations (Supplementary Fig. 1). The genome-wide association study had 80% power to identify SNPs associated with ~0.5% of population variation in hemoglobin in either ethnic group, or ~0.3% in combined analysis, at P < 5 × 10⁻⁸.

We found four SNPs among Europeans and three SNPs among Indian Asians that showed association with hemoglobin at a genome-wide significance threshold of P < 5 × 10⁻⁸ (ref. ⁷, Table 1 and Supplementary Fig. 2). All seven SNPs identified are located in the TMPRSS6 locus on chromosome 22. The SNPs identified among Europeans replicated among Indian Asians, and those identified in Indian Asians replicated among Europeans (all P < 0.001). In combined analysis of European and Indian Asian data, rs855791 (G→A) in TMPRSS6 showed the strongest association with hemoglobin level. The association of rs855791 with hemoglobin level was replicated in a further sample of 5,187 Europeans (P = 4.3 × 10⁻⁷) and 6,721 Indian Asians (P = 1.4 × 10⁻¹¹) from the LOLIPOP study (Supplementary Tables 2 and 3); effect sizes were similar among Europeans and Indian Asians, with no evidence for heterogeneity (P > 0.1). The proportion of population variance in hemoglobin explained by rs855791 was 0.25% in Europeans and 0.31% among Indian Asians. In addition, rs855791 was strongly associated with erythrocyte mean cell volume (MCV), mean cell hemoglobin (MCH) and mean cell hemoglobin concentration (MCHC), key indices of hemoglobin synthesis (Supplementary Table 3). The A allele of rs855791 associated with lower hemoglobin levels was more frequent in Indian Asians than in Europeans in the replication sample (0.52% versus 0.43%, P = 3.5 × 10⁻³³). The 19% of individuals of European ancestry and the 27% of Indian Asian ancestry with AA genotype at rs855791 had hemoglobin concentrations on average 0.2 g/dl lower than did persons with GG genotype.

The linkage disequilibrium (LD) structure of the TMPRSS6 locus was similar among Europeans and Indian Asians (Supplementary Fig. 3). The NFBC1966 and LOLIPOP data indicate that SNPs rs855791 and rs4820268 are in high LD (r²: Europeans 0.83; Indian Asians 0.65) but that rs228918 is in low LD with these two SNPs (r² < 0.1 in Europeans and Indian Asians). In a stepwise analysis conditioned on rs855791, SNP rs228918 was independently associated with hemoglobin levels in Europeans and Indian Asians (P < 0.05).

In a combined analysis of genome-wide data from Europeans and Indian Asians, ten SNPs in the TMPRSS6 locus, and a further six SNPs in the HFE locus on chromosome 6, were associated with hemoglobin at genome-wide significance (Supplementary Table 4 and Supplementary Fig. 2). At the HFE locus, rs198846 (G→A) was most strongly associated with hemoglobin levels. SNP rs198846 is located near the mutation in HFE that results in a C282Y substitution in the HFE protein (rs1800562), a variant that causes hereditary hemochromatosis and influences hemoglobin levels⁸. Based on the HapMap CEU population, rs198846 is in weak LD (r² = 0.006) with rs1800562; in regression analysis, the relationship of rs198846 with hemoglobin was independent of rs1800562 (P < 0.001).

The association of rs198846, near HFE, with hemoglobin level was confirmed in the replication sample (Supplementary Table 3); the proportion of population variance in hemoglobin level explained by rs198846 was 0.32% among Europeans and 0.03% among Indian Asians. SNP rs198846 was also associated with MCV, MCH and MCHC (Supplementary Table 3). The major allele (G) of SNP rs198846 was associated with lower hemoglobin and was more frequent in Indian Asians than in Europeans (92% versus 84%, P = 6.9 × 10⁻⁸¹). The effects of rs855791 and rs198846 on hemoglobin levels were independent and additive. Hemoglobin levels were ~0.4g/dl lower among the 23% of Indian Asians and the 13% of Europeans homozygous for allele A of rs855791 and allele G of rs198846, compared with persons with ≤1 copy of these alleles.

SNPs in the HBB and HBA1 loci, reported to be associated with hemoglobin in a founder population⁴, were not related to hemoglobin, MCV, MCH or red blood cell count in Europeans or Indian Asians (all P > 0.05 after Bonferroni correction).

We report that the association of rs855791 in TMPRSS6 with hemoglobin levels in Europeans and Indian Asians. SNP rs855791 is nonsynonymous and causes a valine-to-alanine amino acid change at position 736 of TMPRSS6, a type II plasma membrane serine protease expressed mainly in liver (Fig. 1)⁹. TMPRSS6 has a key role in iron homeostasis^10,11, inhibiting hepatic hepcidin production. Hepcidin is a direct inhibitor of ferroportin, a membrane iron transport protein present on enterocytes and macrophages, and thereby inhibits intestinal iron absorption and the release of iron from cellular stores^10,12. Rare TMPRSS6 mutations result in unregulated hepcidin synthesis, reduced iron absorption, and iron-deficiency anemia refractory to oral iron therapy³. Recent studies show that rs855791 and rs4820268 in TMPRSS6 are associated with reduced iron and transferring saturation¹³, consistent with the hypothesis that rs855791 influences hepcidin-regulated iron homeostasis.

Animal and in vitro studies have shown that deletion of the serine protease domain of TMPRSS6 eliminates inhibition of hepcidin levels³. Comparison with other serine proteases, and molecular modeling using PHYRE (Supplementary Methods)¹⁴, showed that the amino acid altered by rs855791 is located close to both the catalytic and the specificity site of the serine protease (Fig. 1), suggesting that rs855791 may be a causal variant, acting through altered protease activity or substrate binding.

The genome-wide association and replication data from Europeans and Indian Asians additionally demonstrate association of genetic variants in the HFE locus with hemoglobin. This association is also likely to be mediated through iron metabolism. HFE is a key component of the signaling pathway through which iron-loaded transferring stimulates hepcidin synthesis¹⁵, and genetic variants in HFE are well known to be associated with abnormal iron status and with hemoglobin levels^8,13.

We report here the association of common genetic variants in TMPRSS6 with hemoglobin levels among individuals of both European and Indian Asian ancestry. This association may be mediated through alteration of protease function and hepcidin-mediated control of iron homeostasis. Our findings could provide new insight into the genetic factors influencing anemia and related blood disorders.

supplementary