Heterogeneity adds value to our culture in numerous ways, but when it comes to analyzing large-scale case/control studies with blood, heterogeneity can be more frustrating than a Rubik’s Cube.
Since blood is an easy-to-get, relatively noninvasive specimen, it’s ideal for disease diagnosis and for monitoring the effectiveness of treatments. Scientists have reported that DNA methylation is different in many types of cancers versus controls in blood specimens. But blood is heterogeneous with many different types of cells in blood, each which can have a different DNA methylation signature from the next. As a result, it’s been kind of tough to say if observed differential DNA methylation is indicative of the whole sample, or just a consequence of the changing representation of cell types during different disease stages.
In a study published last summer, a clever team of researchers at institutions in five states found that shifts in leukocyte populations are the likely cause of the methylation signatures in the blood of cancer patients that scientists had been reporting. But how could they simultaneously quantify and account for the shifts that might be going on with all of the different types of leukocytes? Are the reported differences in DNA methylation due to real upticks or losses of methylation, or are those differences simply due to the presence of more B cells, for example, because these subjects are sick and the immune system is kicking in?
Sure, flow cytometry could help address the problem, but it requires fresh blood samples, fancy equipment, and tagging with antibodies—expensive and time-consuming.
Deconvolution with DNA Methylation
So, the team developed a method that allows them to monitor DNA methylation states in changing leukocyte populations without tagging and without being limited to fresh specimens. That opens up the possibility of doing the analysis in more laboratories, using frozen samples stored in biobanks.
The approach is similar to regression calibration. They knew that you can tell human white blood cell types apart by looking at differentially methylated regions (DMRs). They used these DMRs as markers, or surrogates, to identify the various immune cell subpopulations (B cells, granulocytes, NK cells, etc.) and monitor their changes. The team considers the “methylation signature to be a high-dimensional multivariate surrogate for the distribution of white blood cells.”
Then, researchers from Baltimore and Sweden used this approach in a recent paper linking DNA methylation and a genetic risk of rheumatoid arthritis. The method helped them correct for cellular heterogeneity in cases and controls. Without the adjustment, their data showed tons of methylation differences between subjects with rheumatoid arthritis and controls. But after using the filtering method, they were able to see just the methylation differences relevant to their study. So it seems that DNA methylation differences in blood probably could help diagnose disease and monitor treatment, but there are perhaps far fewer differences than previously reported.
But a little caution is still needed. For example, variations in the DNA methylation in the immune cells themselves could conceivably throw off the method. And it’s uncertain what effect other factors and exposures could have on DNA methylation—changes that wouldn’t be relevant to the disease that researchers are studying.