Glioblastoma is monitored mostly by MRI, which can miss early changes in tumor burden. Cell-free DNA (cfDNA) in blood is a promising alternative, but we know little about what governs its release. In our new paper in Neuro-Oncology, led by Sharvari Mankame, we dissect cfDNA release in patient-derived glioblastoma models under tightly controlled conditions.
We cultured two patient-derived lines from the Mayo Clinic GBM PDX National Resource, GBM43 and GBM12, and sampled the conditioned media every 24 hours for five days to measure cell counts, cfDNA yield, fragment size and variants by targeted sequencing. All variants in each line’s genomic DNA were detected in its cfDNA. In GBM43 their allele frequencies rose over the five days, while in GBM12 they stayed stable.
We expected cfDNA to come mostly from dying cells. Instead, cfDNA yield correlated more strongly with live cell counts than with dead cell counts (GBM43: R = 0.88 versus 0.47). Release in these models is therefore not driven by apoptosis alone.
To model the tumor microenvironment, we co-cultured GBM43 with normal human astrocytes (NHAs). Variants unique to each cell type let us attribute cfDNA to its source: GBM43 variants were absent from astrocyte cfDNA, and astrocyte variants were absent from tumor cfDNA.
Over five days, GBM43 outgrew the astrocytes and astrocyte death increased. In co-culture cfDNA, NHA-specific allele frequencies rose 2.4% per day while GBM43-specific allele frequencies fell 3.6% per day. DNA from dying astrocytes diluted the tumor signal, much as DNA from non-malignant cells dilutes tumor DNA in patient plasma.
Fragment sizes told the same story. Astrocyte cfDNA showed a mononucleosomal peak (~185 bp) typical of apoptosis, GBM43 cfDNA lacked nucleosomal peaks, and co-cultures produced a multimodal pattern (174, 330 and 486 bp) seen in neither monoculture.
Temozolomide (TMZ) changed the picture. In treated GBM43 cultures, cfDNA yield rose sharply and correlated with dead cell counts (R = 0.85), no longer with live cell counts (R = −0.04). Treated cfDNA also showed nucleosomal laddering and a narrower pool of detectable variants.
Finally, we turned to mice carrying orthotopic GBM43 tumors, treated with TMZ or vehicle. Plasma cfDNA was elevated in tumor-bearing mice relative to PBS-injected controls. After separating human from mouse reads, we found that tumor-derived fragments were shorter than host fragments, and that their copy number profile recapitulated the chromosome 7 and 9 gains of the parental cells.
Together: cfDNA composition depends on tumor proliferation, the microenvironment and treatment. Interpreting a liquid biopsy in glioblastoma will require accounting for that context, and our next step is to test these principles in patient samples.
Getting the patient-derived cultures to grow reliably took two years, and each experiment was run in triplicate. Huge credit to Sharvari for leading this work, to Hersh Nanda, Maria Kyriakidou and Mimi Mbegbu from our lab, to Nanyun Tang and Michael Berens at TGen, and to Angad Beniwal, Matthew Dufault and Nhan Tran, who ran the mouse studies at Mayo Clinic Arizona. This research was supported by a 2022 American Brain Tumor Association Discovery Award, the Ivy Foundation, the Lane Spyrow GBM Fellowship and Students Supporting Brain Tumor Research.
Read more: the paper in Neuro-Oncology · the preprint on bioRxiv