Up close on clonal hematopoiesis in cfDNA testing

The MSK data revealed that the number of WBC-matched variants in the cfDNA was strongly associated with age, he said. This was not the case with the biopsy-matched or biopsy-subthreshold mutations. “Some of the variants of unknown source also seemed to be a weak association with age, indicating that some of the VUSOs may also be coming from clonal hematopoiesis,” he said.

The mutations followed expectations. The majority of the WBC-matched mutations involved the canonical clonal hematopoiesis genes, such as DNMT3A, TET2, PPM1D, and TP53. “The list was long,” Dr. Razavi said, and notably many known cancer-driver mutations and even some pathogenic actionable mutations were found to originate from CH.

An overall strong association of CH with age was found, as was an increased number of mutations by age, though there was extensive variability in the rate of CH among healthy controls and cancer patients in each age category (Fig. 3). High levels of CH were seen in some of the younger patients, “something that hasn’t been reported to this extent before,” Dr. Razavi said.

Some of the younger patients in the cohort had high allele frequencies of the clonal hematopoiesis in their WBCs. Most of the mutations with the high allele frequencies, as expected, were canonical CH genes, indicating that those mutations are clones that expanded, Dr. Razavi said.

Consistent with previous reports, CH mutations both in WBCs and cfDNA were found to be strongly associated with prior treatment for cancer as some can result in resistance to treatment and may expand post-therapy (Hsu JI, et al. Cell Stem Cell. 2018;23[5]:700–713.e6). “We found a strong indication with truncating PPM1D mutations being much more prevalent in cancer patients who received chemotherapy or radiation,” compared with cancer patients who received no treatment, Dr. Razavi said. Most of these PPM1D mutations were focused in the C-terminus of the protein, and PPM1D was the only gene for which all the mutations clustered in one position and one region of the genome.

“This is very much consistent with more recent work showing PPM1D C-terminus mutations result in stabilization of the protein,” he said, “and that protein inhibits TP53 and prevents apoptosis and results in resistance to certain chemotherapies.”

Dr. Razavi said the group’s findings were consistent overall with previous studies in patients who do not have cancer. In a study of somatic mutations in the cfDNA of 259 healthy individuals, the authors identified clonal hematopoiesis in 60 percent of the subjects using two panels of 508 or 599 cancer-related genes with a depth of sequencing of 6200×. White cell sequencing was done to a depth of about 400×, lower than the sequencing depth of the MSK study. “Even in healthy individuals, most of the mutations involved the known canonical clonal hematopoiesis genes,” Dr. Razavi said (Liu J, et al. Ann Oncol. 2019;30[3]:464–470). The authors wrote, “Hematopoietic clone-derived mutations, including ‘driver mutations’ and ‘passenger mutations,’ are prevalent in the cfDNA of both healthy individuals and cancer patients and may be a potential source of false positives in the liquid biopsy.”

 

This prompted Dr. Razavi’s group to look at its data to determine the optimal WBC sequencing depth to find CH mutations in the cfDNA, he said. In looking at cfDNA variants and the variants that have been matched to the WBCs, the majority of mutations found with a variant allele frequency of less than one percent were WBC matched, “telling us this is a significant phenomenon,” he said. In looking at the mutations in the WBCs that were matched to cfDNA, “we see the same phenomenon, telling us the deep white cell sequencing is required to identify many of these CH cell-free DNA mutations found in the blood.” (Fig. 4).

In the MSK study, a small proportion of the variants of unknown source were detected in the control and cancer nonhypermutated groups. As expected, the vast majority (78 percent) of such variants were detected in the 10 hypermutated cancer samples, “telling us that most likely these are the mutations that are coming from the tumor,” Dr. Razavi said. In the control and nonhypermutated groups, the genes mutated in these variants are mainly clonal hematopoiesis mutations, “telling us there is still residual CH in the variant of unknown source.”

Dr. Razavi and colleagues found a strong correlation between gene size and the number of variants of unknown source in the hypermutated samples, indicating “that most of the mutations in the hypermutated cases are likely coming from subclonal mutations from different sites of the tumor.”

In looking at the distribution of mutations, they found the majority of variants of unknown source in patients with cancer mirror the mutations that are biopsy subthreshold or biopsy matched, indicating most are coming from tumor heterogeneity. “But some follow the patterns of the white-cell–matched mutations, again telling us some of the residual variants of unknown source are also coming from clonal hematopoiesis.”

Somatic clonal expansion and somatic mosaicisms are another source of mutation about which there was concern. “Recent work has shown you can find clonal expansion in the normal tissue,” Dr. Razavi said, with most mutations coming from skin, especially if exposed to the sun, or the lungs or esophagus (Yizhak K, et al. Science. 2019;364[6444]:eaaw0726). He and his group didn’t see this. “We expect this to be a smaller contributor to the source of cell-free DNA in cancer patients and maybe in the healthy individuals,” Dr. Razavi said. “Most of the mutations are coming from the CH.”

In the MSK study, CH variants accounted for the majority of the cfDNA variants with low variant allele frequencies, and that has implications in testing for measurable residual disease or early cancer detection, Dr. Razavi said in summing up. “When the levels of the tumor-driver mutations are very low or when we expand the panel and look at the wider part of the genome, if you don’t take into account these factors, many of the mutations that we find in the cell-free DNA may actually originate from the clonal hematopoiesis.”

CH sequencing at high depth can help. “We also think that addition of in silico CH filtering based on the type of variant, fragment size, and genomic content, combined with machine-learning algorithms that tell us where the site is, can also help with filtering out some of the CH variants,” Dr. Razavi said. “But in silico approaches are not by themselves enough to filter out CH, especially when we look at the large gene panel. And white cell sequencing at high depth probably at this point is the only solution.”

High-intensity cfDNA sequencing assays, with their high genomic footprint and high sensitivity, can be used, he said, “to assess the tumor mutational burden and assign mutational signatures and also provide the better, broader landscape of the genomic alterations that can be used to assess the tumor heterogeneity and also clonal evolution by time.”

In the view of Dr. Razavi and colleagues, fixed panels by themselves are likely to have limited utility in early-stage disease as a standalone test. And if they are used, he said, they need to be combined with other approaches to improve the sensitivity and specificity of these types of assays. 

Amy Carpenter Aquino is CAP TODAY senior editor.