NEW YORK – A new tool for profiling the composition and activity of immune cells in the blood without the need for isolating intact cells could provide a powerful tool for uncovering biomarkers and revealing unknown aspects of immune function in cancer and other diseases.
A group led by Dartmouth University researchers published a study in Nature Communications last week describing the creation of a set of libraries for analyzing methylation array data, and showed that it can yield an accurate picture of the immune cell composition in a sample without the need for isolating actual intact cells via flow cytometry.
According to the authors, the approach can distinguish 12 different cell types in peripheral blood, including closely related cell types such as T-cell subtypes and various granulocytes. In addition, the library includes multiple cell-derived ratios and proportions, such as the neutrophil-to-lymphocyte ratio or naïve-to-memory ratios, totaling more than 56 total immune profile variables.
"With flow cytometry you need a fresh sample or cryopreserved sample, and you are mostly concentrating on cell surface markers, which is a limited set of proteins," said Brock Christensen, co-director of the Cancer Population Sciences Research Program at Dartmouth-Hitchcock's Norris Cotton Cancer Center. "Whereas here we can measure nearly a billion discrete sites in the epigenome to get a super high-resolution picture of … the profile for all of these specific cell types."
"Once you know that information, and you use some robust machine learning approaches to discern the optimal set of CPG sites for the library, then you can take that information and apply it to a whole blood sample," he explained.
According to Christensen and his co-authors, this opens the possibility of studying things like cell counts and characteristics, which would otherwise have required fresh blood samples, using the wealth of archived DNA now collected in cancer and other disease biobanks.
RNA-based approaches are already being used in this vein, most notably, T-cell repertoire sequencing.
But Christensen said that DNA methylation offers a much better venue. "One of the nice things about methylation is that the methylation mark on a given site is seen as either present or absent. So the accuracy of DNA methylation-based methods are always going to outperform an RNA-based deconvolution method," he said.
"It just will never be as good because you can have widely variable numbers of cell-specific transcripts within a given cell type." This means the deconvolution operation, "the math underlying separating all of the components … it just won't work as well."
The DNA methylation array methodology behind the new approach is something investigators "stumbled on" several years ago in studies comparing methylation in blood from cancer patients and controls.
"We were seeing these huge [differences] between cases and controls, but it was really just that the sick people have a different immune profile, so we started to think, how do we deal with that? Can we adjust for the differences in immune cell proportions so that we can look for the sort of independent effects of methylation?"
According to Christensen, the landmark paper debuting an approach to use immune cell contributions to clean up methylation array data was published by Oregon State University researcher Andy Houseman in 2012. This has since become widely used in epigenetic epidemiology, with Christensen's co-author and CPS colleague Lucas Salas following up in 2018 with an updated reference library optimized for the current standard of Illumina array platforms.
Initially, Christensen and colleagues were focused on using these methods to help adjust methylation analyses for the relative contribution of immune cells. "It was only a few years ago that that we started to realize that … instead of just using it to adjust in models where we're interested in methylation, why don't we actually ask questions about the immune system?" Christensen said. "Now that's become our focus, particularly in an era of … rapidly emerging immunomodulatory therapies," he added.
In the study, Christensen, Salas, and their co-authors described the creation of their library for translating methylation data into a readout of immune cell populations, which required initial isolation and quantification of individual cell types from a cohort of volunteers.
Using machine learning to compare this data to methylation array data for the same individuals, the group was able to create and validate a reverse-predictor that then infers cell populations from methylation patterns.
Important for future applications of the approach, Salas said, is the fact that he and his team were able to collect a diverse population to inform their immune cell deconvolution library. "We have males, females, different ages, and we had a wide range of information so that we could extrapolate independently of ancestral background [and other] genetic differences," he said.
The group's success in isolating cell types was also crucial, he said. "It sounds like it's something that may be simple … but that was the most challenging part for developing this library," he noted. "Several other groups have tried, and they were not successful … so it's a real success for us to have something that is really standardized."
The study's authors validated the resulting flow cytometry samples with known cell counts, as well as an independent set of artificial mixtures from the Gene Expression Omnibus.
The team also completed some early tests of the approach in a variety of contexts, using publicly available datasets from multiple sclerosis, rheumatoid arthritis, breast cancer patients, and COVID-19 infection.
In the breast cancer cohort, the researchers compared methylation-predicted immune-cell proportions in patients before and after receiving chemotherapy or a combination of chemo/radiation. In 74 patients who received radiation therapy only, there was a significant relative increase of the neutrophil proportion and a mirrored decrease of several lymphoid lineages after treatment. In 70 patients who got both treatments, there was a significant increase in eosinophil, monocyte, and regulatory T-cell proportions, and a decrease of memory B cells after treatment.
Christensen is now leading an evaluation of the approach to predict response to cancer immunotherapy. He is a co-PI on a study at Dartmouth and Dana-Farber in patients with head and neck cancers, and said the researchers also have an IRB under review which would allow them to recruit patients across all FDA-approved immunotherapy treatments in order to track their response and look for predictive immune signatures.
Beyond cancer, he said the team also sees immediate applications in the monoclonal antibody drug space for immune disorders like rheumatoid arthritis.
"Every time you turn on the TV you see a new ad for some monoclonal antibody therapy, and they go through this whole list of potential side effects," Christensen said. "It would be great if we could do a better job of predicting who is going to respond well and who might not, based on a more detailed immune profiling tool."
The method could also help in the early-phase clinical trial setting, as a companion diagnostic tool to select a subset of patients who are most likely to respond, thereby improving the odds of a successful readout.
Finally, the authors wrote that the approach could potentially aid current clinical assessments in which flow cytometry is standard, in cases where samples can't be processed immediately. Because the method allows assessment of more cell types than can be accommodated by existing tests, it could also allow new, more comprehensive clinical assays.
For example, Christensen said, the currently used complete cell blood count, a common clinical assessment, is limited to five general immune cell types. But the Dartmouth team showed its DNA-based method could extend this to twelve immune cell types.
In that light, CBC is "not really complete at all," he said, "So we believe there's some room in the future to take what we're doing and give patients and clinicians a much more detailed immune profile that also doesn't require cells."