Evaluation of the Infinium Methylation 450K technology

Aims: Studies of DNA methylomes hold enormous promise for biomedicine but are hampered by the technological challenges of analyzing many samples cost-effectively. Recently, a major extension of the previous Infinium HumanMethylation27 BeadChip (R) (Illumina, Inc. CA, USA), called Infinium HumanMethylation450 (Infinium Methylation 450K; Illumina, Inc. CA, USA) was developed. This upgraded technology is a hybrid of two different chemical assays, the Infinium I and Infinium II assays, allowing (for 12 samples in parallel) assessment of the methylation status of more than 480,000 cytosines distributed over the whole genome. In this article, we evaluate Infinium Methylation 450K on cell lines and tissue samples, highlighting some of its advantages but also some of its limitations. In particular, we compare the methylation values of the Infinium I and Infinium II assays. Materials & methods: We used Infinium Methylation 450K to profile: first, the well-characterized HCT116 wild-type and double-knockout cell lines and then, 16 breast tissue samples (including eight normal and eight primary tumor samples). Absolute methylation values (beta-values) were extracted with the GenomeStudio (TM) software and then subjected to detailed analysis. Results: While this technology appeared highly robust as previously shown, we noticed a divergence between the b-values retrieved from the type I and type II Infinium assays. Specifically, the b-values obtained from Infinium II probes were less accurate and reproducible than those obtained from Infinium I probes. This suggests that data from the type I and type II assays should be considered separately in any downstream bioinformatic analysis. To be able to deal with the Infinium I and Infinium II data together, we developed and tested a new correction technique, which we called `peak-based correction'. The idea was to rescale the Infinium II data on the basis of the Infinium I data. While this technique should be viewed as an approximation method, it significantly improves the quality of Infinium II data. Conclusion: Infinium 450K is a powerful technique in terms of reagent costs, time of labor, sample throughput and coverage. It holds great promise for the better understanding of the epigenetic component in health and disease. Yet, due to the nature of its design comprising two different chemical assays, analysis of the whole set of data is not as easy as initially anticipated. Correction strategies, such as the peak-based approach proposed here, are a step towards adequate output data analysis.