Clinical Implementation of Matched Tumor/Germline Sequencing Improves Accuracy of Tumor Genomic Profiling and Therapeutic Recommendations

Abstract

Genomic profiling of cancers informs diagnostic and prognostic classification and aids in selection of targeted therapeutics. Targeted, next-generation sequencing of cancer-specific genes is clinically feasible and enables comprehensive somatic reporting; without a matched germline specimen, germline alterations can confound analyses of the somatic profile and generate uncertainty in interpretation. This work reports the validation and implementation of optional matched tumor/germline sequencing in a precision cancer medicine program. DNA from 63 patient samples was analyzed using OncoPanel, a hybrid capture-based sequencing assay of 461 genes. Three analytical pipelines were implemented: tumor only, matched tumor/germline, and germline only. For matched tumor/germline, germline alterations in 19 genes with actionable/therapeutic implications were rescued. Retrospective analysis of the first 1600 matched cases was done to determine the potential clinical utility of this approach. Limit of detection for point mutations/insertions and deletions was 3% allele fraction; reproducibility was >98%. Matched tumor/germline concordance across 938 somatic calls was 100%. The average tumor mutational burden (TMB) was approximately 4 mutations/Mb lower than tumor-only sequencing. TMB-high patients were accurately reclassified as TMB-low in 14% of cases. Twenty-five percent of validation cases (14% after launch) had a pathogenic or likely pathogenic germline variant conferring cancer susceptibility; 14% of validation cases (7% after launch) harbored a germline variant of therapeutic significance. Matched tumor/germline sequencing is more accurate than tumor-only sequencing, while still encompassing all genomic findings that inform targeted therapy selection.

Figure 1

Process overview of the Profile program. Patient consents to a protocol to enable biobanking and sequencing; part of the consent deals with return of germline results. The OncoPanel test is ordered by a physician. Tumor and blood specimens are obtained from the patient (or outside institution) and sent for pathology review. DNA is extracted, followed by library preparation, hybrid capture, and sequencing. Data are analyzed and interpreted in tumor-only, matched tumor-germline, or germline-only format, and corresponding reports are generated. Results are returned to the patient and physician for clinical decision-making, and enter into an electronic data warehouse (EDW) and linked to pathology and clinical data. Positive germline results are returned to a genetic counselor and the patient; genetic counselor will initiate follow-up, including clinical confirmation of research findings. Transmission of consent, order, laboratory processes, and results are all electronic.

Figure 2

Validation analyses of matched tumor and germline pairs. Matched tumor and germline validation samples were processed through the individual germline-only (G-only) and tumor-only (T-only) pipelines, as well as the paired tumor/germline (T/G) pipeline. The G-only pipeline used the HaplotypeCaller, and the T-only and T/G pipelines used Mutect2 for single-nucleotide variant (SNV)/small insertion/deletion (Indel)/multinucleotide variant (MNV) detection. Tumor and germline copy number analyses were performed by RobustCNV and presented for simultaneous manual review. Structural variants (SVs) in somatic and germline samples were identified with SvABA and Manta tools. CNV, copy number variant; gnomAD, Genome Aggregation Database; GVPTI, germline variants with potential therapeutic implications; PON, panel of normals.

Figure 3

Sensitivity of single-nucleotide variant (SNV)/insertion/deletion (Indel)/multinucleotide variant (MNV) detection by the tumor/germline pipeline at varying variant allele frequencies (VAFs). The vertical dashed line represents a VAF of 3%. When the expected VAF was ≥0.03, >95% of the variants were detected.

Figure 4

Copy number variant detection relative to tumor purity. Green colored segments were called gain or loss by the pipeline. The colored arrowheads indicate the loci of relevant genes, with the gene name indicated close by and color coded correspondingly. A:RB1 2D was detected in 70% tumor sample TL5, was 1D in the 35% replicates, and the signal was lost by 14%. B: TL4 at 30% tumor harbored gains at different levels in adjacent baited loci: RAD21 at 9N and EXT, MYC, and RECQL4 each at 7N. The signals were maintained at 15% and diminished by 6%; RAD21 was called as a gain by the pipeline at the lowest replicates, but the signal may not be sufficient for sign out. C: TL2 had amplification of CDK12 at 9N adjacent to a high amplification of ERBB2 at an estimated 72N; the CDK12 signal was no longer detected at 6% tumor purity, but ERBB2 gain was detected at an average log₂ ratio of 1. Chr, chromosome.

Figure 5

Tumor mutational burden (TMB) correlation between tumor-only (T-only) and tumor/germline (T/G) data. A: The entire comparator validation cohort. B: Cases falling below a TMB of 20 (boxed area in A); the clinical cutoff of TMB = 10 mutations/Mb is denoted by the dashed red line. C: The difference in TMB (T-only – T/G matched) was calculated for 69 patients with self-reported ancestry and both sequencing tests. Non-White samples have a difference in TMB that is statistically higher than White samples (Wilcoxon rank-sum test with continuity correction W = 384.5, P = 0.04198).

Figure 6

Simultaneous manual review of tumor and germline copy number variant plots. The colored arrowheads indicate the loci of relevant genes, with the gene name indicated close by. A: A germline gain in RAD52 was confirmed, whereas the NAB2 gain on the same chromosome was in the tumor only. B: Germline single copy loss of RAD52 was confirmed in a different patient on review of both plots.

Figure 7

A: Oncoprint of germline variants with potential therapeutic implications (GVPTIs) identified in first 1628 patients tested with matched tumor/germline sequencing. A total of 129 variants were identified in 18 genes across 121 patients; 7 patients had two co-occurring variants. B: GVPTIs by Oncotree code.