Iron deficiency is a global health concern and usually attributed to chronic blood loss or inadequate dietary iron intake [1]. In rare cases iron deficiency can be referred to insufficient duodenal iron absorption which cannot be sufficiently treated by oral iron supplementation. This type of anaemia is termed iron-refractory iron-deficiency anaemia (IRIDA) and is an autosomal-recessive disorder, characterized by (i) congenital, hypochromic, microcytic anaemia; (ii) very low mean corpuscular volume (MCV); (iii) low serum iron and low transferrin saturation; (iv) normal ferritin or ferritin levels in the lower limits of normal; (v) no response to oral iron treatment; and (vi) inappropriately high levels of hepcidin [2], [3].

At the luminal side of the gut dietary iron is transferred into the enterocyte by the divalent metal transporter 1 (DMT1, SLC11A2) [4]. SLC11A2 (solute carrier family 11 (proton-coupled divalent metal ion transporters), member 2) also transfers iron from the endosomes to the cytosol following the uptake of iron via a transferrin receptor complex, is also of importance for the shuttling of iron in several tissues such as the liver or kidney and importantly for the transfer of iron taken up via transferrin receptor mediated endocytosis from the endosome into the cytoplasm [4]–[6]. Dysregulation of SLC11A2 can lead to disturbances of iron homeostasis. For example, activation of SLC11A2 in the brain was associated with toxic iron accumulation, autophagy and cell death in mouse models of Parkinson disease [7] whereas pharmacological modulation of SLC11A2 activity can reverse hepatic iron overload in mouse models of hemochromatosis [4]. Importantly, a loss of function mutation was responsible for the development of severe microcytic anaemia in mk/mk mice [6]. Subsequently, several rare mutations in SLC11A2 were identified and were linked to the development of microcytic anaemia in a total of 4 patients [8]–[11]. Interestingly, such patients present with low serum ferritin levels but normal or increased transferrin saturation along with low hepcidin concentrations [8]–[12].

TMPRSS6 (transmembrane serine protease 6) mutations were described to cause IRIDA [13]–[16]. TMPRSS6 encodes for matriptase-2, a type II transmembrane serine protease mainly produced by the liver. TMPRSS6 belongs to a large group of type two serine proteases which modulate a variety of cellular processes including the selective cleavage of specific substrates which is fulfilled by a conserved catalytic motif. Type two serine proteases act as membrane bound proteases, however, soluble forms – as for TMPRSS6 – have also been described. Accordingly, TMPRSS6 expression is increased in early embryogenesis and mislocalization of this protease has been associated with high grade prostate cancer [17].

Recently, TMPRSS6 has been identified as a modifier of iron homeostasis because it regulates the expression of the systemic iron regulatory hormone hepcidin [18] and inhibits hepcidin activation by cleaving membrane hemojuvelin [19]. Hepcidin controls iron absorption by binding to the only known cellular iron export protein ferroportin thereby leading to ferroportin degradation and blockage of iron egress from the enterocyte into the circulation [18], [20]. In addition, hepcidin blocks the transfer of iron from macrophages into the circulation, which is the major iron source for erythropoiesis following erythrophagocytosis and re-utilization of the metal from senescent erythrocytes [20]–[22]. Thus, under physiologic conditions high levels of hepcidin as observed with iron overload reduce iron absorption from the diet. In iron deficiency, however, low iron levels inhibit hepcidin formation and thus enables iron to be transferred from the gut to the blood [18]. Part of the iron-mediated control of hepcidin can be referred to the action of TMPRSS6 and thus functional mutations in this gene are associated with insufficient iron absorption on the basis of increased hepcidin levels [13]–[16]. In addition, in genome-wide association studies (GWAS) common variants in TMPRSS6 were associated with alterations of serum iron status, erythrocyte volume [23]–[25], or hemoglobin levels [26], [27].

We identified a family with asymptomatic non-consanguineous parents with three of four children presenting with severe anaemia. After excluding all known causes responsible for iron deficiency anaemia we searched for mutations in SLC11A2 and TMPRSS6 that could explain the severe anaemia in these children.

In the present paper, we report the clinical case of three siblings with severe iron deficiency anaemia being refractory to oral iron substitution. Given the wealth of publications describing mutations in two genes (SLC11A2 and TMPRSS6) being associated with iron deficiency anaemia [8]–[10], [13], [15], [16], [36], [37], we initially hypothesized that mutations in these two candidate genes were the reason for this anaemia. Sequencing of the exons and exon–intron boundaries of both SLC11A2 and TMPRSS6 revealed seven known SNPs to occur in the family, but no novel mutations. No allele clearly segregated with the IRIDA trait in this family.

Four SNPs were found within SLC11A2: rs6580779, rs161044, rs150909, and rs149411. All family members except for the first son were heterozygotes for these SNPs (Table 2). Given the length of 8678 bp of non-redundant sequence information, we found one SNP in 2170 bp. Within the entire gene region of SLC11A2 of ∼50 kb, ∼730 SNPs were found in dbSNP (build 135).

Although mutations in TMPRSS6 are described to be linked to severe iron deficiency anaemia in patients with relatively increased hepcidin levels, we decided to sequence the exons and exon–intron boundaries of the TMPRSS6 gene. We found three SNPs: rs11704654, rs4820268, and rs855791 (Table 2). Given the length of 7075 bp of non-redundant sequence information, we found one SNP in 2358 bp. Within the entire gene region of TMPRSS6 of ∼44 kb, ∼784 SNPs were described in dbSNP (build 135). The SNP rs855791 results in a nonsynonymous (V736A) change in the serine protease domain of TMPRSS6 and was recently described as the top-hit of genome-wide association studies (GWAS) to be associated with alterations of serum iron, transferrin saturation, erythrocyte mean cell volume, blood hemoglobin levels, and glycated hemoglobin [23], [26], [38]. Similarly, rs4820268, located in exon 13, was associated with lower serum iron concentrations, lower hemoglobin levels, smaller red cells, and more variability in red cell size [24]. This variant was also discussed to have a lowering effect on hepcidin levels in urine [25].

We genotyped the seven SNPs in the population-based SAPHIR study (n = 1,726) and replicated them in NBS (n = 1,832) and found that the TMPRSS6-SNPs were very frequent in these populations, but that the three SNPs in SLC11A2 occurred at minor allele frequencies of 6.1% (Table 3). In addition, the SNPs rs6580779, rs161044, and rs150909 were highly correlated with each other, although they were dispersed over the entire gene region of ∼40 kb (Figure S3). Due to this correlation structure, the number of effective loci was calculated to be 5 out of the 7 SNPs. Therefore, the significance level for the single genotype analysis was set to 0.01.

When performing genetic association analysis of these seven SNPs with either iron or ferritin levels in the SAPHIR population, we found that only the two top-hit SNPs of recent GWAS on iron levels [23], [24], [26] (rs4820268 and rs855791) exhibited a slight, though significant effect on iron and ferritin (Table 4). These associations remain significant even after correction for multiple testing. The remaining five SNPs showed no influence on the analyzed outcome variables. Since rs4820268 and rs855791 were highly correlated (D' = 0.95, r² = 0.83), they rather reflect the same information. In order to test for a possible interaction of the only two significant SNPs rs4820268 and rs855791 on both iron and ferritin levels, we introduced the interaction term rs4820268*rs855791 to the linear regression model, but it was neither significant on iron (p = 0.469) nor on ferritin levels (p = 0.686). Therefore, we could not detect an epistatic effect of these two SNPs on iron-related traits.

The haplotypes of all family members within TMPRSS6 and SLC11A2 are summarized in Table S5 together with the haplotype frequencies in SAPHIR. Interestingly, the most common TMPRSS6-haplotypes in SAPHIR was not observed in the family members. When combining the haplotypes to haplotype pairs as observed in family members, the TMPRSS6 haplotype-pair of son 2 showed significantly reduced iron values as well as ferritin levels (Table 5). The corresponding individuals from the SAPHIR population (n = 6) showed markedly reduced iron as well as ferritin values than the average SAPHIR population (Table S6).

Finally, when we combined the TMPRSS6 and SLC11A2 genotypes to a joint profile, we saw that the one SAPHIR individual exhibiting the profile of the anaemic son had extremely low iron and ferritin levels lying below the 5^th percentile of the iron and ferritin level distribution in SAPHIR. This individual exhibited also the same combination of TMPRSS6 and SLC11A2 haplotype-pairs as son 2, which was also associated with lower iron and ferritin levels (Table 5). Although this analysis is only based on one person, the laboratory parameters are in line with the observations made in the family.

However, this observation could not be confirmed in the Nijmegen Biomedical Study [39]. Although the NBS population was of similar size as the SAPHIR study, the SLC11A2/TMPRSS6-profile of the anaemic son was not observed (Table S7).

Taken together, neither the single SNP effects nor the haplotypes could explain the observed phenomenon that both the family and unrelated individuals carrying the family's joint TMPRSS6/SLC11A2-profiles in SAPHIR showed reduced iron and ferritin levels.

The children reported in the present study showed all features of iron deficiency anaemia, especially the inability to respond to oral iron administration. Interestingly, neither the father nor the mother was anaemic, but both were borderline iron-deficient, which is also reflected by low or undetectable serum hepcidin levels. This observation is in line with previous reports that the degree of anaemia is more evident in infancy than during adult life [16], [45].

In accordance, the results of our comparative observation with the SAPHIR study indicated that adult individuals from the SAPHIR population with identical SNP combinations and haplotypes as seen in the children of the family under investigation in this study had low iron levels but not that severe iron deficiency as seen in the children. These discrepancies might thus possibly be attributed to the fact that iron parameters of very young children were compared with adults and that the phenotype could only be seen in case of high iron demand as it occurs in early childhood and adolescence where high amounts of the metal are needed for haemoglobin synthesis and growth. This also fits to observations made in TMPRSS6 mutated subjects which have been shown to give a more severe phenotype in young children and which becomes milder with aging [16], [36].

The results of the haplotype analyses indicated that the combined SLC11A2/TMPRSS6 haplotype of SON 2 was significantly associated with lowered iron and highly significantly associated with lowered ferritin values in SAPHIR. Unfortunately, we could not replicate this finding, as the SLC11A2/TMPRSS6 haplotype of SON 2 was not present in the Nijmegen Biomedical Study. However, it would be interesting to validate this finding in further populations.

Although GWAS have been very successful at identifying risk alleles for complex genetic traits, there still remains the fact that the identified risk alleles for a particular trait, together, don't fully explain the trait heritability. The results of our study fit well into this observation. Various explanations for this missing-heritability problem have been proposed, but a recent study by Song et al. [46] supports the idea that perhaps this heritability is not missing at all, or at least not completely. Typically, the risk information from multiple alleles is combined by using a multiplicative model, thereby assuming that the individual loci behave independently. Song et al., however, calculated recurrence risks to close relatives under different models and found that the results yielded higher estimated recurrence risks than were predicted under a multiplicative model, particularly when the risk alleles were rare [46]. This indicates that gene-gene interactions could account for more heritability than is often assumed. The empirical data from our family study confirm the results of Song and colleagues [46], thus implying that gene-gene interactions might account for an important section of the missing heritability of complex diseases. In addition to gene-gene interactions one has also to consider interaction of the gene products at the protein level, which is a form of epistatic interaction, as has recently been shown for DMT1 and hepcidin [47].

With this in mind, we hypothesized that the combined impairment of SLC11A2 and TMPRSS6 activities by the different polymorphisms effected both transmembrane iron-flux and hepcidin regulation which in summary would lead to the iron-deficient phenotype in situations with an increased iron demand (growth phase of children).

Although our data are suggestive for a gene-gene interaction, based on the methodology used, one cannot fully exclude another gene defect that might contribute to the dramatic anaemia in the children which might be identified by whole exome sequencing.