Development of a highly sensitive ctDNA assay to map the evolution of metastatic breast cancer.

e13048 Background: Treatment of metastatic (met) breast cancer (mBC) remains challenging due to tumor evolution, where tumors may acquire additional somatic alterations conferring resistance top treatment. We piloted a novel ctDNA assay, ProvSeq-Liquid (PS-L) in the longitudinal profiling of ctDNA in mBC. PS-L is a next generation sequencing assay that detects somatic cfDNA variants in 523 genes. We determined the reproducibility of PS-L and applied the PS-L to a cohort of cfDNA from newly diagnosed mBC patients compared to paired primary and met tissue samples. Methods: 26 adult patients with a new diagnosis of de novo or first recurrence HER2 negative mBC were enrolled. Patient cfDNA will be sampled approximately every 12 weeks for 3 years. PS-L and ProvSeq-Solid (PS-S) are lab-developed tests utilizing Illumina TruSight Oncology 500 High Throughput Liquid or Solid assay chemistries, respectively. Comprehensive genomic profiling (CGP) of ctDNA samples extracted from plasma was performed using the PS-L at a ≥0.5% variant allele frequency (VAF) cutoff and CGP of met/primary tumor was performed using PS-S. Libraries were sequenced on an Illumina NovaSeq 6000 and analyzed using Illumina DRAGEN Bio-IT. Variants were filtered based on germline whole-exome sequencing testing, an internal database of curated variants, and evaluated as clinically signficiant (CS) with assistance from the Onco-KB precision oncology knowledge base. Results: Average median coverage of target genes was 1920X in PS-L (sd = 764X, n = 105) and 1048X in PS-S samples (sd = 350X, n = 54). PS-L controls with 37 known small nucleotide variants and indels between 0.3% - 0.9% (mean = 0.5%, n = 4) were recovered at an average rate of 97%. Median time between most recent tissue sampling and earliest ctDNA was 36 days. Across solid tumor and ctDNA results, 539 unique variants were identified. Of those, 117 variants were annotated as CS and 11 were biomarkers predictive of response to an FDA-approved drug in mBC. The most common variants found in either solid tumor or ctDNA that were predictive of response to an FDA-approved drug were PIK3CA E545 (n = 6 patients) and PIK3CA H1047R (n = 3 patients). For patients with both primary and met tissue samples tested (n = 16), the mean concordance of CS variants between the sample types was 87% (range = 14-100%). An average of 2.8 CS variants (range = 0-6) were found in primary and/or met samples (n = 26 patients). Of these CS variants, an average of 38% (range = 0-100%) were concordant with the first available ctDNA sample time point and an average of 45% (range = 0 - 100%) were concordant with at least one ctDNA sample timepoint. Conclusions: We observed high reproducibility of PS-L, with potential to validate at low VAFs. There was variable overlap in CS variants between solid tissue and ctDNA, with some variants exclusive to ctDNA depending on time between solid tumor at ctDNA collections. Enrollment and longitudinal PS-L testing are ongoing.