NEW YORK (GenomeWeb) – Multiple tumor types are recurrently methylated at a specific CpG island, according to researchers from the National Institutes of Health. In addition, the biomarker could potentially be analyzed noninvasively to detect late stage and potentially intermediate and early stage cancers.
In a study published today in the Journal of Molecular Diagnostics, the researchers described a specific site, ZNF154, that is methylated across five different tumor types — lung, stomach, colon, breast, and gynecologic — and can be identified through a targeted bisulfite sequencing strategy. In addition, the researchers performed a computational simulation of circulating tumor DNA, demonstrating that they could detect the methylation biomarker on tumor DNA in a solution composed of 99 percent normal DNA.
The work builds on a previous study in which the team discovered that the epigenetic marker was found across a wide variety of samples in the Cancer Genome Atlas. In the more recent study, the researchers sought to test the reproducibility of the marker as well as whether it could be identified from blood.
The researchers looked at 184 tumor samples across the five different cancer types as well as genomic DNA extracted from 34 normal tissue samples. First, they validated the idea that the targeted bisulfite sequencing protocol would be robust even on small amounts of DNA, since bisulfite conversion can fragment and damage DNA. After verifying that the methylation signal was "robust" with "minimal variation" they moved on to testing the 184 tumor samples.
The DNA from the samples was bisulfite converted and the researchers designed a 302-base PCR product covering the ZNF154 CpG island, including the transcriptional start site, which they had previously found to be hypermethylated in the TCGA data. They then sequenced the amplicons on the Illumina MiSeq instrument.
The researchers looked first at DNA from 42 endometrial tumors and eight normal endometrial samples, finding that tumor samples had a 66 percent increase in median methylation levels, on average. In addition, they found increased levels of methylation in both late and early stage tumors, although more methylation in the later stage cancers compared to earlier stages.
Next, they compared sequencing data with methylation array data from the same samples finding good correlation, although it was more variable among samples with lower levels of methylation.
They next looked at data from lung, colon, breast, and stomach tumor samples. For each tumor type, mean methylation was higher than in the normal samples. For normal samples, mean methylation was 20 percent, 8 percent, 19 percent, and 4 percent for lung, stomach, colon, and breast tissue, respectively. Methylation levels increased by as much as 44 percent and 45 percent in the lung and colon samples, while methylation in stomach and breast tumor samples increased by 20 percent and 31 percent, respectively.
"The large magnitude of this hypermethylation bodes well for a strong discriminant in each tissue type," the authors wrote.
Nonetheless, they noted that there were still a few tumor samples that had lower levels of methylation than the average methylation in the normal samples, including one sample each from lung, stomach, and colon, and three breast samples.
Within each tumor type, the researchers noted a couple of additional interesting findings. For instance, within the lung cancer samples, small cell carcinomas and squamous cell carcinomas had average methylation levels around 15 percent higher than adenocarcinomas and bronchioalveolar carcinomas.
The group next wanted to assess whether ZNF154 methylation could be used as a pan-cancer marker in blood-based screening. The researchers randomly matched one of the 34 normal samples to each of the 184 tumor samples and created mixtures of various dilutions ranging from 10 percent normal to 99 percent normal.
They were able to reliably identify tumor DNA in mixtures containing up to 90 percent normal tissue. Using the area under a curve measurement (AUC), which plots the false positive rate and true positive rate, the team achieved an AUC of .89. When the percentage of normal data was increased to 99 percent, AUC dropped to .74.
For individual tumor types, endometrial and colon tumors were the easiest to classify with AUCs of .95 even in dilutions of 90 percent normal. Breast tumors were the hardest to classify, and the researchers never achieved an AUC greater than .90.
"Overall, our in silico dilution estimates suggest that hypermethylation of the ZNF154 region should be robustly detected as long as the ctDNA fraction is at or above single-digit percentage levels," the authors wrote, adding that such "sensitivity should be adequate to detect advanced cancer and some intermediate and early tumors, depending on the tumor type."