Stable ethnic variations in DNA methylation patterns of human skin.

TO THE EDITOR Epigenetic mechanisms regulate the interpretation of genetic information and mediate phenotypic plasticity (Bird, 2007; Feinberg, 2007). Although it is now well established that DNA methylation patterns differ between human cell types and tissues (Mohn and Schubeler, 2009), little is known about epigenetic variation. Primary human epidermis samples represent an excellent model system for epigenetic analyses because they are characterized by a high degree of cellular homogeneity, with keratinocytes being the main cell type (Kanitakis, 2002). In addition, it has been shown that functional characteristics of human skin can vary between ethnicities (Berardesca and Maibach, 2003), which rendered the analysis of ethnic DNA methylation patterns particularly interesting. We have shown previously that human epidermis shows very little interindividual methylation differences, which permits the identification of statistically significant methylation changes with comparably small sample numbers (Gronniger et al., 2010). To identify ethnic methylation differences, we therefore obtained epidermis samples from 30 healthy volunteers (10 Africans, 10 Asians and 10 Caucasians, see Supplementary Table S1 online). All samples were suction blister roofs (Sudel et al., 2003) from the inner forearms of adult females, distributed in similar age groups (Supplementary Table S1 online). Tissue samples were obtained according to the recommendations of the Declaration of Helsinki Principles and the guideline of the International Conference on Harmonisation Good Clinical Practice, as applicable to a nondrug study. All volunteers were long-time residents of Hamburg, Germany, and provided written, informed consent. DNA from all 30 samples was analyzed by Illumina HumanMethylation27 BeadChip arrays (Illumina, San Diego, CA; Bibikova et al., 2009) to determine the methylation status of 27,578 CpG dinucleotides. This analysis generated 30 million data points, with beta values for individual markers ranging from 0 (unmethylated) to 1 (completely methylated). Raw data were normalized and statistically corrected as described previously (Gronniger et al., 2010), before they were used for subsequent data analysis. Unsupervised hierarchical clustering of the 30 methylation profiles revealed an overall close proximity of samples from specific ethnic groups (Figure 1). This result suggested that different ethnic groups might be defined by specific methylation differences. As a first step toward the identification of ethnic methylation differences, we established median methylation values for all markers and for each ethnic group. These methylation values were subsequently used to define ethnic methylation profiles. The results showed a high degree of similarity between individual methylation patterns, which is consistent with earlier observations (Gronniger et al., 2010). In addition, we also identified markers that were substantially (|Δβ|>0.15, with a Benjamini–Hochberg adjusted P-value <0.01) hyper- or hypomethylated when two ethnic groups were compared (Figure 2a). Numbers of differentially methylated markers were relatively low and ranged from 21 to 100, which corresponds to 0.1–0.4% of the probes analyzed (Figure 2a). To validate and further analyze the array-predicted methylation differences, we used deep bisulfite sequencing of three selected genes that were among the most differentially methylated genes from the array analysis. Fragments were PCR amplified from sample pools and sequenced using 454 technology, which routinely generated sequencing coverages exceeding 1,000 × . The resulting 454 sequencing profiles essentially confirmed the Infinium-based results. The 5′ gene body of the VWCE (von Willebrand factor C and EGF domains) gene, which was predicted to be hypermethylated in Africans, also showed substantially increased methylation in the 454 profile obtained from African samples (Figure 2b, 61% methylation in Africans vs. 37% and 36% in Asians and Caucasians, respectively). Similarly, the intragenic CpG island of the CPXM2 (carboxypeptidase X member 2) gene, which was predicted to be hypermethylated in Africans and in Caucasians, also showed notably higher methylation levels in the bisulfite sequencing results from these ethnic groups (Figure 2b, 8% methylation in Asians vs. 32% and 38% in Caucasians and Africans, respectively). Last, the promoter and 5′-gene region of the PM20D1 peptidase gene, which was predicted to be hypermethylated in Caucasians, showed substantially increased methylation in the 454 sequences obtained from the Caucasian sample pool (Figure 2b, 39% methylation in Caucasians vs. 9% and 6% methylation in Asians and Africans, respectively). These results further illustrate and substantiate the ethnic differences in methylation patterns. Altogether, our data analysis identified a nonredundant set of 205 markers that were differentially methylated across all three ethnicities. Markers showing ethnic methylation differences were preferentially found at sites outside of CpG islands (Figure 2c). Interestingly, six of these markers were previously identified to be hypomethylated in sun-exposed skin (Gronniger et al., 2010). This overlap is highly significant (P<0.01, Fisher's exact test) and suggests a functional relevance of the observed ethnic methylation differences. The functional relevance was further investigated by pathway analysis, which showed an enrichment in functional categories associated with dermatological diseases and conditions (Supplementary Figure S1 online), consistent with known ethnic variations in skin function (Berardesca and Maibach, 2003). Finally, we also used principal component analysis to identify specific ethnic methylation profiles. Principal components PC1 and PC2 accounted for 32% and 10% of the total variation, respectively, which was sufficient to accurately separate methylation profiles from Caucasians, Asians, and Africans (Figure 2d). It was notable that the African samples formed two distinct subclusters (Figure 2d), but additional experiments will be required to investigate this aspect further. In conclusion, principal component analysis confirmed the presence of ethnic methylation differences and suggested that a comparably small set of methylation markers can be used to accurately predict the ethnic origin of human skin samples. The degree of epigenetic heterogeneity among human individuals is a topic of considerable scientific debate. Although it is widely assumed that environmental signals can modulate epigenetic marks to mediate phenotypic changes (Jaenisch and Bird, 2003), the human epigenome appears to be relatively stable in differentiated cells (Eckhardt et al., 2006). We have analyzed skin samples from African, Asian, and Caucasian volunteers to identify ethnic methylation differences. Our data suggest that ethnic methylation differences affect only a small fraction of the genome, but appear in many individuals from a specific ethnicity. This stability appears remarkable and distinguishes ethnic methylation differences from the stochastic epigenetic variation described in previous studies (Feinberg and Irizarry, 2010; Feinberg et al., 2010). The stability of ethnic methylation differences can possibly be explained by the involvement of chromatin factors and/or genetic variations. Further work will be required to elucidate the mechanisms that establish and maintain ethnic methylation differences. MW, EG, FS, and HW are employees of Beiersdorf AG. FL received consultation fees and a commercial research grant from Beiersdorf AG. We thank Jan Batzer for his help in obtaining skin samples and Günter Raddatz, Cassandra Falckenhayn, Roger Fischer, and Oliver Heil for their valuable support in the methylation analysis. SUPPLEMENTARY MATERIAL Supplementary material is linked to the online version of the paper