Cell-free DNA Fragmentation as a Non-Invasive Biomarker Across Different Oncology Settings

Introduction: Tumorigenesis is accompanied by genomic changes, and identifying these changes can improve cancer detection and enable personalized clinical management. Conventional genomic analyses have been difficult to use widely as they rely on expensive high-coverage sequencing and a priori knowledge of tumor alterations. To overcome this, we previously developed DELFI (DNA Evaluation of Fragments for Early Interception), a non-invasive liquid biopsy that uses cost-effective low-coverage whole-genome sequencing (WGS) to analyze millions of cfDNA fragments whose length and genome-wide distribution reflect cancer-related genetic and epigenetic changes that can be used for cancer detection. Here, we analyzed cfDNA from 355 individuals to evaluate fragmentation in two distinct clinical contexts where blood-based biomarkers are not standard of care, but their adoption could address clinical unmet needs. (a) Detection of ovarian cancer, where 80% of tumors are diagnosed at a late stage (90%, 70%, and < 30% five-year survival for stages I, II and III/IV respectively) and no screening strategy has yet shown a survival benefit. The protein biomarkers CA-125 (Cancer Antigen 125) and HE4 (Human Epididymis Protein 4) have shown promise, but their performance has been insufficient to justify population-wide screening. (b) Monitoring of an aggressive brain tumor, high grade glioma (HGG), following initial surgery and chemoradiation. Current response assessments require multiple MRIs months apart and fail to differentiate pseudo-progression from true progression, hindering treatment planning for a disease with median progression-free survival (PFS) around 7 months. Our results demonstrate the potential for cfDNA fragmentation profiles to serve as accessible blood-based biomarkers in diverse settings.

Materials and

Methods: For the ovarian cancer screening application, we performed whole genome sequencing (WGS) at 1-2x coverage on cfDNA extracted from the blood of 134 women with ovarian cancer and 204 women without cancer from academic institutions in Europe and the United States (Netherlands Cancer Institute trial NL58253.031.16, Danish Endoscopy III trial, Netherlands trial register ID NTR1829, Penn BioTrust Collection RRID SCR_022387). We computed the ratio of short (100-150 bp) to long (151-220 bp) cfDNA fragments in 473 non-overlapping 5Mb bins genome-wide, and quantified aneuploidy on each chromosomal arm as a z-score with respect to a panel of healthy individuals. We also quantified CA-125 and HE-4 levels in plasma or serum from the same individuals. We used the fragmentation features ensembled with protein levels to cross-validate a logistic regression model (DELFI-Pro) for ovarian cancer detection in the samples from patients in Europe (n=276) and externally validated the locked model on the samples from patients in the United States (n=62). For the brain tumor monitoring application, we extracted cfDNA and performed WGS on 58 liquid biopsies from 17 patients with HGG (with n=5, 16, 13, 8, 14 and 2 blood draws at pre-surgery, pre-chemoradiation treatment, 2, 4 and 6 weeks on-treatment, and 1-month post-treatment time points, respectively). We computed the same DELFI fragmentation features described above and correlated fragmentation profiles between timepoints for each patient. We estimated PFS as time between diagnosis and first indication of new treatment after initial treatment.

Results, Conclusions, and Discussions: In the cross-validated ovarian cancer screening cohort, DELFI-Pro detected ovarian cancer with high performance (AUC 0.96, 95% CI 0.93-0.99). At a score threshold corresponding to >99% specificity, required for widespread screening of a low prevalence disease, sensitivity was 72%, 69%, 87% and 100% for stage I-IV cancers, respectively. For the most common subtype, High-Grade Serous Ovarian Cancer (HGSOC), sensitivity was 90%. Particularly for early-stage disease, this was substantially higher than for CA-125 alone (sensitivity 34%, 62%, 63% and 100% for stage I-IV cancers). In the external validation cohort, DELFI-Pro maintained high performance with overall sensitivity of 73%, including 81% sensitivity for HGSOC at >99% specificity. These results illustrate the potential for assays of the cfDNA fragmentome to enhance currently available protein biomarkers, towards population-wide screening of ovarian cancer. In a completely different clinical setting, monitoring of patients with known HGG brain tumors, fragmentation profiles also demonstrated utility. Fragmentation profiles in these patients at the pre-treatment timepoint were more correlated to their on-treatment profiles than to pre-surgery profiles (median correlation 0.94 vs. 0.88). This observation may reflect a high pre-surgery tumor burden that falls after resection and adjuvant treatment and can be rapidly, non-invasively tracked in the circulation. After four weeks on-treatment, a higher correlation to the pre-treatment fragmentation profile was associated with longer PFS (p=0.01, median PFS 15.6 vs. 4.8 months for patients with correlations above and below the mean, respectively), suggesting that patients with fragmentomes resembling their lowest tumor burden profile survive longer before progression. This highlights the potential for cfDNA fragmentation as a biomarker of treatment response and disease progression in HGG and opens the door to supporting real time treatment modification. These two clinical scenarios are distinct - reliable detection of ovarian cancer, a devastating cancer type not currently screened for, or monitoring of HGG, an aggressive brain tumor which currently lacks an effective way to identify treatment response - but both feature a clinical unmet need that could benefit from an accessible, high-performance biomarker. Our results serve as a proof-of-concept for liquid biopsies of the cfDNA fragmentome in these complex, diverse clinical settings.

Acknowledgements (Optional):