Introduction

Atopic dermatitis (AD) is a chronic inflammatory skin disease that is characterized by intensely itchy skin lesions. AD is one of the most common chronic diseases in childhood affecting 10%–20% of children in industrialized societies [1], with a steady increase over the past decades [2,3]. Along with asthma and hay fever, AD is commonly associated with an abnormal immune response and the formation of allergy antibodies (IgE) against innocuous environmental allergens.

AD is often the first clinical manifestation of allergic disease. The onset of disease is typically observed during the first two years of life [4]. For the majority of affected children, AD heralds a lifetime of allergic disease. A susceptible child commonly passes a characteristic sequence of transient or persistent disease stages that is known as the “atopic march” which begins with AD and food allergy in the young infant and continues with the development of allergic airways disease later in childhood and adulthood [5]. The close familial and intra-individual association of the allergic disorder strongly suggests shared genetic etiology.

A strong genetic component in allergic disorders has been recognized almost a century ago. Cooke and van der Veer first reported that the relatives of patients are at significantly increased risk of developing allergic disease [6]. Even today, a positive family history for allergic disorders is the single strongest predictor for the development of AD [7]. Additional evidence for the importance of genetic factors in atopic disease comes from twin studies. The concordance rate for AD among monozygotic twins of about 80% far exceeds the concordance rate of 20% observed among dizygotic twins [8,9]. These data clearly indicate that the genetic contribution to the expression of AD is substantial. Furthermore, studies on the vertical transmission of AD and atopic disease show that children are more likely to inherit these disorders if the mother is affected (parent-of-origin effect) [10]. The predominance of maternal inheritance may be due to environmental factors such as uterine milieu or breast feeding, but they may also arise due to genetic mechanisms such as parent-specific gene expression (genomic imprinting) [11].

AD and atopic disorders are regarded as multifactorial conditions, the onset and severity of which are influenced by both genetic and environmental factors. The data are consistent with an immune etiology shared by all allergic diseases and a congenital target organ defect, the penetrance of which is modified by multiple environmental factors during early childhood. The identification of genes underlying AD and allergic disorders has the capacity to define primary physiologic mechanisms, thereby clarifying disease pathogenesis, identifying pathways and targets for therapeutic intervention.

Although several genome-wide linkage screens for AD have been conducted [12–14], there was no substantial overlap between the identified regions of highest linkage, and the underlying genes remained elusive [15]. We have previously mapped a major susceptibility locus for AD on Chromosome 3q21 [16]. Here we report the identification and characterization of a novel epidermal collagen gene as the underlying disease gene.

Results

To narrow the candidate region spanning 12.75 centiMorgans (cM) (13.5 Mb) on Chromosome 3q21 (Figure 1A), 96 additional microsatellite markers at an average distance of 140 kb were genotyped in 199 affected sibling families with AD from the original linkage scan. Linkage analysis yielded a 1-lod support interval of 5.4 Mb between the markers M3CS075 and M3CS233 (Figure 1B). Subsequently, we conducted an association scan of the 5.4-Mb region using 212 single nucleotide polymorphisms (SNPs) at an average distance of 25.5 kb (Table S1). The SNPs were selected primarily to cover known and predicted genes. Because we had observed a strong maternal effect in the linkage study [16], we chose a family-based association analysis that allowed us to search for risk alleles preferentially transmitted from the mother [17]. Two adjacent SNPs, rs5852593 (p = 0.0079) and rs1497309 (p = 0.016), located 36 kb apart, were associated with AD (Table S1 and Figure 1B). To define the critical region, we typed 16 additional SNPs. Eight of these markers showed association with AD and a maternal transmission pattern to affected children, which was consistent with our previous linkage results (Table 1). The strongest association with a single marker was observed for rs4688761 (p_all = 0.0016, p_maternal = 0.0006). We selected eight markers spanning 96 kb that were associated with AD and carried nonredundant information based on the linkage disequilibrium (LD) in the region, and we performed haplotype analysis which confirmed the results (transmissions (T)_all:non-transmissions (NT)_all = 222:168, p_all = 0.0076, T_mat:NT_mat = 105:68, p_maternal = 0.0070). In addition, we assessed the significance of the difference in maternal-versus-paternal haplotype over transmissions empirically using the permutation procedure for parent-of-origin transmission disequilibrium test (TDT) implemented in PLINK (P_POO_{maternal emp} = 0.014) [18].

Next, we investigated whether the observed association accounted for the linkage in the region. We used the marker that had shown the strongest association, rs4688761, to identify 73 of the 199 families from the original linkage scan in whom the disease-associated allele had been transmitted to affected offspring. Nonparametric linkage analysis in the 73 associated families yielded significant evidence for linkage (Z_all = 4.18 versus 4.31 in the complete cohort), demonstrating that the majority of the linkage signal was attributable to these families. The significance of this finding was assessed by performing nonparametric linkage analysis in 10,000 random selections of 73 families. An empirical significance level was calculated as the proportion of replicates for which the maximum Z_all score was equal or greater than that obtained in the actual analysis. The probability of obtaining a Z_all score of ≥4.18 by chance in a random selection of 73 families was estimated by 10,000 simulations to be 0.027.

To confirm the association result, we used a large independent replication set consisting of 292 complete nuclear families including 481 children with AD. We genotyped the selected eight SNPs covering an interval of 96 kb that were associated with AD in the discovery dataset. We confirmed the association with AD across all markers with the strongest association with AD observed for marker A36637742 (p = 0.00038), which also showed a significant overtransmission of the maternal allele (p = 0.0013) (Table 2). The association remained significant after correction for multiple testing. For each marker, it was the more common allele that was overtransmitted. Haplotypes were constructed over the region, which confirmed the association with AD and showed that this phenotype is associated with the overtransmission of the most common haplotype (haplotype frequency 65%, Table 3), and of the maternal allele (p_maternal = 0.000025 for AD, P_POO_{maternal emp} = 0.033). Next, we compared the AD status among the parents: in the discovery cohort, significantly more mothers (n = 63) than fathers (n = 19) suffered from AD (odds ratio [OR] 4.39, 95% confidence interval [CI] 2.51 – 7.68, p = 5.46 × 10⁻⁸). Similarly, in the replication cohort, mothers (n = 83) were significantly more frequently affected with AD than fathers (n = 55) (OR 1.71, 95% CI 1.16 – 2.52, p = 8.3 × 10⁻³). To assess whether the parent-of-origin effect that was observed originated from the discrepancy in AD prevalence among the parents, we compared haplotype transmissions from affected and unaffected mothers. In the discovery cohort, the excess in maternal transmissions was not predominantly attributable to affected mothers (T:UT = 24:21, p = 0.14) compared to unaffected ones (T:UT = 81:47, p = 0.0059). Similarly, affected mothers (T:UT = 25:17, p = 0.14) of the replication cohort did not contribute a larger transmission excess than the unaffected ones (T:UT = 70:29, p = 0.00013). We conclude that the observed maternal overtransmission pattern in both cohorts was not due to the higher AD prevalence among mothers.

A database search within the associated 96-kb interval revealed a single predicted gene, FLJ35880, extending 11.6 kb into the associated region. No other expressed sequence tag was detected. To identify any additional transcripts, we used putative exons predicted with The National Center for Biotechnology Information (NCBI) Modelmaker within and bordering the critical region to perform rapid amplification of cDNA ends (RACE) from human skin mRNA. We thus identified a single transcript of 9226 bp that consisted of 42 exons (Figure 1D) and included all eight exons of FLJ35880. The corresponding gene spanned 139 kb of genomic sequence and completely encompassed the associated region. Pairwise LD measures (D′) of the genotyped markers indicated that the gene was contained within one 170-kb region of increased LD (Figure 2), whereas the neighboring genes, LOC440978 and LOC131873, were located in separate blocks. The LD structure was consistent with the data for the European population in the HapMap database (Figure S1) [19]. The size of the transcript was confirmed by Northern blot hybridization with an FLJ35880-specific probe detecting a single transcript of 9.6 kb in human skin mRNA (unpublished data), which was in good agreement with the RACE experiments.

The open reading frame yielded a protein of 2614 amino acids with an estimated molecular weight of 289.9 kDa. The predicted protein contained a collagenous domain in the central part and was therefore classified as a new member of the collagen superfamily, collagen XXIX. A BLAST search revealed the human collagen VI alpha 3 chain as its closest neighbor (32% identity). Homology with collagen VI alpha 3 was further strengthened by a similar domain architecture consisting of six N-terminal and three C-terminal von Willebrand factor–type A domains (vWAs) flanking a short collagen triple helix (Figure 1E), and an 18–amino acid secretion signal [20]. Sequence analysis of all 42 exons in 46 unrelated children with AD and 2 unaffected individuals revealed 13 common and six rare sequence variations (frequency <2%) predicted to cause nonsynonymous amino acid substitutions (Table S2). Four variants were located within the triple helix repeat of a collagen domain, however none of them changed the first glycine residue in the repeating sequence patterns (Gly-X-Y), which is critical for triple helix formation [21]. In addition, we performed in silico comparisons of the six variants located within a vWA with the crystal structure of the human vWF A3 domain. One variant, V669G, affects an amino acid within a highly conserved stretch of eight amino acids. The mutation of an adjacent amino acid within this conserved region of a vWA in the homologous gene COLVIA1 has been reported to cause the monogenic muscle disorder Bethlem myopathy [22]. This variant, however was rare (allele frequency 1,3%). One additional variant (E455K) is located in a region predicted to be important for integrin–collagen interaction, and another one (M56T) is located in a helix near a rather conserved region that may also affect protein–protein interaction. All coding SNPs were genotyped in the discovery cohort. Four of them showed a positive association with AD that did, however, not account for the observed haplotype association (Table S2).

Gene expression analysis in human tissues revealed a tissue-specific expression pattern of COL29A1. The highest expression was observed in the skin, but also in the lung, small intestine, colon, and testis (Figure 3A). Overall, COL29A1 expression is moderately low compared to more abundant epidermal transcripts such as keratin 10 (unpublished data).

To specify the expression sites in the layers of the skin and to assess the role of collagen XXIX in AD, we performed in situ hybridization using a COL29A1-specific cRNA-probe on skin biopsies of five patients with AD and five healthy patients (controls). In normal skin, COL29A1 was exclusively expressed in the epidermis with the strongest staining in the suprabasal viable layers. In contrast, the skin of patients with AD revealed a striking absence of COL29A1 mRNA staining in the most differentiated upper spinous and granular layers (Figure 3B and Figure S2). No significant difference in the expression of COL29A1 was observed comparing patients with AD to controls (1.28- ± 0.53-fold down-regulation in AD patients, p = 0.41) using quantitative Taqman reverse-transcriptase (RT)-PCR. These results indicate that while differences in mRNA quantity were not detected, AD patients show a distinct abnormal cellular distribution pattern of COL29A1 expression in the differentiated outer epidermis.

We generated a polyclonal antibody to visualize the collagen XXIX protein in the skin of five patients with AD and five normal controls, including four and three of each group, respectively, in whom in situ analysis was performed. Consistent with the in situ findings, we observed collagen XXIX staining in the differentiated suprabasal layers of the epidermis in normal human skin and a remarkable absence of staining in the most differentiated upper spinous and granular layers (Figure 4 and Figure S3).

Discussion

In a whole-genome linkage scan for AD, we previously identified a susceptibility locus on human Chromosome 3q21. The candidacy of this chromosomal region was further supported by Bradley et al., who mapped a locus for AD severity in close proximity (3q14) in a Swedish population [13]. By positional cloning, we have now identified the disease-causing gene, COL29A1, which encodes a novel epidermal collagen.

The disease gene was located in a two-staged investigation consisting of systematic linkage and association scanning of the region and subsequent confirmation of the association in a large, independent, replication dataset. In the first stage, we genotyped additional microsatellite markers in the candidate region, which narrowed the initial 12.75-cM linkage interval to 5.4 Mb. The association analysis with an average marker distance of 25.5 kb, finally, revealed an association that was confined to a haplotype block of 170 kb, which included a single gene, COL29A1. Pairwise LD measures indicated LD across the entire gene and defined two subblocks of increased LD. The strongest association was observed within the 96-kb subblock encoding the collagenous domain and the C terminus of collagen XXIX. A rapid decay of LD at the borders of the COL29A1 haplotype block and lack of association of the SNPs located within the neighboring genes clearly limited the association to COL29A1. In addition, we demonstrated in the discovery dataset that the families that contributed to the association of SNP rs4688761 with AD also accounted for most of the linkage signal. This finding corroborated that variants in COL29A1 explained the previously reported linkage of AD to 3q21 [16]. Finally, we confirmed association in a large independent family cohort, making COL29A1 the first AD susceptibility gene, to our knowledge, that is identified by positional cloning.

Consistent with the linkage analysis, we found a maternal transmission pattern to affected offspring in both family cohorts. Although the sexes were equally represented among the affected children of both cohorts, we observed a marked maternal preponderance in AD status among the parents in both cohorts. This finding clearly supports the notion of a maternal effect. It may, however, also raise the question whether the maternal overtransmission pattern observed for the COL29A1 haplotype was due to the different prevalence of AD in mothers and fathers. This is unlikely to be the case, because the analytical tools used in this study, nonparametric linkage and TDT, do not take into account the parental phenotype. Furthermore, we showed that the observed maternal effect was not predominantly attributable to transmissions from affected rather than unaffected mothers. Although parent-of-origin effects have not previously been reported for genes from Chromosome 3q21, they have been observed at other loci influencing allergic disease [23]. Tissue-specific imprinting of genes encoding extracellular matrix (ECM) proteins has been reported in the mouse [24], and their disruption has been shown to impair skin structure and function [25]. Interestingly, COL29A1 is expressed in human placenta, an organ of embryonic origin. Apart from classic genomic imprinting mechanisms, maternal effects may be due to an interaction of the child's genotype with the maternal environment during prenatal life.

Sequencing of the exons and promoter region revealed 19 nonsynonymous coding SNPs, six of which were located within a vWA. In silico analysis of these variants revealed only one rare mutation altering a highly conserved amino acid, but all of them may affect protein–protein interaction. None of the nonsynonymous coding SNPs explained the observed association on its own. It has been demonstrated in other complex diseases that multiple independent variants may occur in a single disease gene [26]. It is therefore possible that several variants or combinations thereof which are associated with the most common haplotype of COL29A1 are involved in the disease pathogenesis. The functional influence of the associated variants remains to be determined.

Involvement of COL29A1 in AD is further supported by its tissue- and cell-specific expression pattern. Like COL29A1, a growing number of collagens recently identified show a restricted expression pattern. These are not mainly found in fibrous connective tissue, but rather in the ECM of more specialized tissue structures pointing to a distinctive function of these proteins [27]. Highest COL29A1 expression was observed in the skin, but also in other epithelial tissues such as the lung, small intestine, and colon, which are the main manifestation sites of allergic disorders, including asthma and food allergies. This gene expression pattern might indicate a role of collagen XXIX in a wider spectrum of allergic diseases and suggests a molecular link between AD, respiratory airways disease, and food allergies, which are epidemiologically closely associated [28,29].

In human skin, collagen XXIX was detected throughout the viable layers of the epidermis with an increase toward the differentiated cells of the granular layer. Comparative expression analysis of COL29A1 by in situ hybridization and immunohistochemistry in skin biopsies of patients with AD and healthy controls revealed a distinct lack of COL29A1 mRNA and protein in the outer viable layers of the epidermis. The process of epidermal stratification is tightly regulated by an increasing gradient of extracellular Ca²⁺ concentration and a specific special and temporal expression pattern of transcriptional regulators [30]. Our findings indicate that the specific cellular milieu acquired during terminal epidermal differentiation affect the regulation or degradation of COL29A1 mRNA in the outer epidermis. However, our findings do not allow us to distinguish between these two processes.

Lack of collagen XXIX in the outer epidermis of AD patients indicates that a defective ECM may give rise to the disease, proposing a new pathomechanism for AD. Collagens are the most abundant ECM proteins in vertebrates and play a crucial role in maintaining tissue integrity. Their importance for tissue function has been highlighted by the wide spectrum of human diseases caused by mutations in collagen genes [31]. Although a large number of collagens in the connective tissue–rich dermis have been characterized, little is known about collagens in the ECM of the epidermis [32,33]. Collagen XXIX belongs to the vWA containing collagens. They form filaments with globular domains containing the vWA motifs, which are involved in protein–ligand interactions for the organization of tissue architecture and cell adhesion [34]. It is therefore conceivable that collagen XXIX plays an important role in keratinocyte cohesion. Lack of collagen XXIX may facilitate antigen penetration through the skin, which may explain the association found between COL29A variants and allergic sensitization, a common feature in AD patients [35]. Recent findings indicate that structural and functional integrity of the epidermis is a key factor in the development of AD [36] and in the disease progression to allergic airways disease [37].

Furthermore, ECM collagens influence the migration of epidermal antigen-presenting Langerhans cells and T cells [38,39] and may thus play an important role in the initiation and maintenance of cutaneous immune responses. In addition, ECM collagens participate in immune regulation by binding to inhibitory immune receptors [40], rendering collagen XXIX an interesting novel susceptibility gene for AD. Impairment of the immune defense function of the skin is a clinical hallmark of AD. Patients with AD show a diminished resistance against microbes resulting in the colonization of nonlesional skin with Staphylococcus aureus in nearly 90% of patients and an increased susceptibility to bacterial and viral skin infections [41]. Our findings led to the identification of collagen XXIX as a novel component of the epidermal ECM and propose a new disease mechanism in the etiology of atopic dermatitis and allergies.

Materials and Methods

Supporting Information

Acknowledgments

We thank the patients with atopic dermatitis and their families for their participation; Monika Schwarz, Jenny Pech, Christina Flachmeier, and Sabine Enigk for technical assistance; and Dr. Sylke Schneider and Dr. Claudia Rasche for performing the skin biopsies.