New DNA Methylation Chip Passes Its Biggest Test Yet, With Surprises for Cancer Research

DNA methylation is one of biology’s most powerful storytellers. Layered onto cytosine bases across the genome, these chemical tags help decide which genes are switched on or off, encode a cell’s identity, record its history of division, and shift dramatically in cancer and aging. For more than a decade, the workhorse technology for reading these marks in large human studies has been Illumina’s Infinium Methylation BeadChip family, a microarray platform prized for its cost-effectiveness, quantitative accuracy, and the deep ecosystem of community software built around it. Now a team led by researchers at the Children’s Hospital of Philadelphia, the Van Andel Institute, and the University of Southern California has published the most comprehensive independent evaluation to date of the platform’s newest generation, the Infinium MethylationEPIC v2 BeadChip, and their verdict is largely enthusiastic, with a few important caveats.

The study, published in Epigenetics Communications, systematically compared EPICv2 against its predecessors, the HumanMethylation450 array (HM450) and the first-generation EPIC array (EPICv1). The new chip carries 937,690 probes, up from 866,552 on EPICv1 and 486,427 on HM450. More than 99 percent of these probes target CpG cytosine methylation, the canonical epigenetic mark, while smaller fractions query non-CpG methylation, common single-nucleotide polymorphisms, and quality controls. Crucially, the team found that EPICv2 retains 83 percent of EPICv1 probes and 81 percent of HM450 probes, including 24,463 CpG probes from HM450 that had been dropped from EPICv1 and are now reintroduced. This backward compatibility matters enormously: over 100,000 samples profiled on HM450 alone have been deposited in public repositories, and any new chip must speak the same analytical language as that mountain of legacy data.

The technical heart of the Infinium platform lies in its two chemistries. Infinium-I probes use two bead types per target, one for the methylated cytosine and one for the unmethylated version, while Infinium-II probes use a single bead type and discriminate the two states through a color-discriminating single-base extension. All probes hybridize to bisulfite-converted genomic DNA, in which methylation status is preserved as a sequence difference. The evaluation found that the shared probes between EPICv1 and EPICv2 largely kept their original designs, with only 70 probes switching from Infinium-I to Infinium-II chemistry and 12 switching the other way. The researchers caution, however, that when integrating data across platforms, subtle methylation differences measured by these redesigned probes should be interpreted carefully.

One of the most consequential improvements concerns probe quality and population applicability. Of the probes deleted in the transition from EPICv1 to EPICv2, 72.9 percent had known problems with cross-reactivity or direct interference from sequence polymorphisms, whereas only 0.1 percent of retained probes suffered from such issues. The new array also maps more cleanly onto the GRCh38 human reference genome and is less susceptible to direct influence from ancestry-specific genetic variation, although probes targeting African ancestry populations remain the most affected, consistent with the higher genetic diversity of that population. These design decisions should make EPICv2 more reliable across diverse human cohorts, a long-standing concern in epigenome-wide association studies where poorly mapping probes can generate spurious findings.

Reproducibility testing reinforced the platform’s credentials. The team profiled technical replicates of several human cell lines, including the B-cell line GM12878, the prostate cancer line LNCaP, the lymphoblast line K562, and the colorectal carcinoma line HCT116, and found methylation measurements between replicates were highly correlated, with Spearman correlation coefficients significantly exceeding those between different cell lines. Interestingly, the newly added EPICv2 probes showed lower inter-cell line correlation, indicating that the expanded probe set discriminates cell identities better than the old one. EPICv2 also adopts a new probe naming convention, borrowed from the recent mouse methylation array, in which replicate probes share a prefix identifying the 122-mer template but differ in suffixes encoding strand, chemistry, and replicate index. Of 5,483 replicate probes covering 5,621 loci, most showed correlations close to one, validating the alternative designs without measurable signal loss.

Accuracy was benchmarked against whole genome bisulfite sequencing, the gold standard. On the same cell lines, Spearman correlations between EPICv2 beta values and sequencing-derived methylation fractions reached 0.854, 0.874, and 0.866 for GM12878, LNCaP, and K562 respectively. In a titration experiment using DNA with known methylation fractions, 89.8 percent of probes correlated above 0.99 with the expected level, and 98 percent above 0.9. The roughly 0.2 percent of CpG probes that performed poorly were enriched for sequence polymorphisms, poor mapping, and repetitive regions such as satellite DNA and retrotransposable elements, and the authors provide masking recommendations through their public annotation resources. The team also demonstrated the chip’s biological sensitivity by profiling HCT116 derivatives lacking the methylation writers DNMT1, DNMT3A, DNMT3B, or the histone methyltransferase SETD2, capturing dramatic global methylation loss in double-knockout lines and revealing that surviving methylation in these cells concentrates at imprinted regions, RNA polymerase III binding sites, and transposable elements.

Perhaps the most striking practical finding is how little DNA the platform actually needs. Although Illumina recommends 250 nanograms of input, the researchers obtained informative methylomes from as little as one nanogram of DNA and from as few as 500 flow-sorted cells. Probe success rates declined with lower input but stayed above 50 percent even at one nanogram, and low-input data remained highly correlated with high-input profiles. The failures that did occur were not random: quiescent and heterochromatic regions lost detection first, while CpG-dense bivalent promoters and enhancers resisted failure. This low-input capability opens the door to liquid biopsy applications, where circulating cell-free DNA from plasma is chronically scarce, and to archival or minimally invasive clinical samples.

The evaluation also confirmed that EPICv2 preserves the biomarkers that have made methylation arrays central to translational research. The team checked coverage of nine epigenetic clocks, seven cell-type deconvolution panels, and CpGs from 26 categories of human trait associations drawn from more than a thousand EWAS studies. Most clocks and EWAS hits were retained at higher-than-random rates, with gene-expression-associated CpGs the best preserved; telomere clocks and fertility-related CpGs were the notable exceptions. Using a recent whole-genome bisulfite sequencing atlas of normal human cell types, the researchers calculated that only 15.6 percent of pairwise cell-type contrasts lack any coverage, and 43 percent are covered by more than 100 probes, meaning the chip can robustly resolve the cellular composition of complex tissues.

The boldest new feature is a category of 824 probes, designated with the prefix “nv,” that target recurrent somatic mutations found in cancer rather than methylation at all. Most use Infinium-I chemistry, with multiple probes per site covering different alternative alleles; 59 genes are targeted, with TP53 the most heavily represented at 113 probes, concentrated in its DNA-binding and tetramerization domains. In a proof of concept, the nv probes correctly detected the KRAS G13D mutation in HCT116 cells. The authors note that these probes are more prone to detection failure than methylation probes, partly because internal CpGs within the probe sequence create uncertain hybridization contexts, and they flag this as an area for future improvement. Separately, the team showed that total signal intensities on the chip can detect copy number alterations, recovering the chromosome 9 deletion and chromosome 22 amplification tied to K562’s BCR-ABL1 fusion, and deletions on chromosomes 2 and 13 in LNCaP cells.

The overall picture is of a mature platform evolving intelligently rather than radically. EPICv2 shifts its probe content toward the regulatory genome, with new probes enriched in enhancers and depleted in quiescent chromatin, while pruning the problematic probes that plagued earlier generations. It cannot fully escape the physics of its chemistry: residual background signal tempers beta values toward intermediate levels, so completely unmethylated and fully methylated states are not captured perfectly, and roughly two percent of probes still show suboptimal titration correlation for reasons that are not always identifiable. But with its expanded enhancer coverage, improved cross-ancestry performance, one-nanogram sensitivity, preserved clocks and deconvolution panels, and the novel ability to interrogate somatic mutations and methylation on the same chip, EPICv2 gives researchers a genuinely multi-omics instrument. For a field racing to turn epigenetic signatures into diagnostics for cancer, aging, and disease, that combination may prove hard to beat.

Subject of Research: Evaluation of the Infinium MethylationEPIC v2 BeadChip for DNA methylation profiling

Article Title: Comprehensive evaluation of the Infinium human MethylationEPIC v2 BeadChip

Article References: Kaur, D., Lee, S. M., Goldberg, D., Spix, N. J., Hinoue, T., Li, H.-T., Dwaraka, V. B., Smith, R., Shen, H., Liang, G., Renke, N., Laird, P. W., & Zhou, W. (2023). Comprehensive evaluation of the Infinium human MethylationEPIC v2 BeadChip. Epigenetics Communications, 3(1), Article 6. https://doi.org/10.1186/s43682-023-00021-5

Image Credits: AI Generated

DOI: 10.1186/s43682-023-00021-5

Keywords: DNA methylation, epigenetics, MethylationEPIC v2, Infinium BeadChip, epigenetic clocks, cell type deconvolution, cancer genomics, somatic mutations, EWAS, bisulfite sequencing, liquid biopsy, genomics