A Novel Segmentation-Based Purity-Informed Approach Accurately Detects Total Copy Number and Its Subclonal Fraction in Myeloma Targeted Panel Sequencing Data

Chromosomal copy number aberrations (CNAs) are common in multiple myeloma (MM) and play a major role in patient prognosis and treatment outcome. Detection of CNAs is possible using targeted region sequencing (TRS), however, accurate detection and quantification of the total copy number and the subclonal fraction of each targeted genomic region is challenging, especially in low-purity samples. Using the Myeloma Genome Project (MGP) Panel [PMID:35522533], we have developed a segmentation-based CNA calling approach for TRS to (a) identify copy number breakpoints, (b) estimate total copy number and its subclonal fraction (defined as significant deviation from expected clonal state), based on tumor coverage ratio (LogR) normalized by its paired germline sample and (c) adjust for the purity of the tumor sample as estimated from the TRS panel single nucleotide variant (SNV) data.

We initially used a set of 27 longitudinal samples from 14 MM patients with both TRS and whole-genome sequencing (WGS) data to optimize the method for accuracy. Results showed a high concordance (96.3%) in total copy number in the targeted regions between WGS (by using Battenberg) and TRS. Cohen’s Kappa between the two was 0.93 (95%CI 0.84-1), showing strong agreement between TRS and gold-standard WGS calls. In addition to standard calling of gain and deletion events even in tumors with purity as low as 17%, the method could also differentiate different total copy number states (e.g. distinguishing between Gain1q and Amp1q). We could detect subclonal events with CCF > 0.3 in tumors with a minimum purity of 50%. The ability to detect CCF depended on sample sequencing evenness as quantified by mean absolute difference of LogR across the genomic regions of the panel (MAD(LogR)<0.3). Our method also allowed the detection of biallelic (e.g., Del17p-TP53) and combinatorial (e.g., Gain1q-Del1p) events in myeloma.

To validate this method, the same analysis was then undertaken on two independent, recently generated TRS datasets (Oxford dataset, N=48 and the UK RADAR clinical trial, N=80), albeit with single time-point samples. Similar frequencies of high-risk CNAs were identified in the validation datasets (Del1p32-p12, Gain1q21, Amp1q21 and Del17p at 23%, 23%, 4% and 15% in the Oxford dataset and 19%, 28%, 6% and 20% in the RADAR dataset respectively) which also match well with previous reports, corroborating the utility of the developed CNA calling method. Altogether, these results demonstrate that this computational method accurately detects CNAs with prognostic value. Also, with the increasing use of subclone size to define high-risk myeloma in new myeloma high-risk guidelines, our method may enable better risk stratification from TRS data at diagnosis. This method has the potential to be applied to other hematological cancer targeted panels to call accurate CNA in the regions targeted by NGS panels.