Abstract 1781P
Background
High-throughput sequencing and ctDNA are driving the precision medicine paradigm shift. However, the analysis and the interpretation of the resulting data is an open challenge. By an integrated and multidisciplinary work among wet and dry lab experts, we have explored the feasibility of an automated BAM data solution with built-in variant calling.
Methods
A cohort of 49 women with metastatic breast cancer was prospectively enrolled within the CRO-2018-56 clinical trial and characterized for ESR1 and PIK3CA genes by NGS of ctDNA. The resulting BAM files were analyzed using FreeBayes, GATK, Miseq Reporter, LoFreq, Mutect2, SAMtools and SNVer variant callers. Their concordance was evaluated by Cohen’s kappa (k) with respect to manual annotation, both before and after filtering for clinical significance, based on the OncoKB database.
Results
Of the overall 49 analyzed samples, 7 were manually annotated as ESR1 mutated and 11 as PIK3CA mutated. Detected ESR1 variants were Y537S (3/7), D538G (2/7), H377R and Y537N (1/7, each). For PIK3CA, the H1047R (8/11), E545K (2/11) and H1047L (1/11) were observed. Number of variants called for ESR1 and PIK3CA, respectively, were: 5 and 30 in FreeBayes; 1 and 5 in GATK; 6 and 22 in Miseq Reporter; 8 and 52 in LoFreq; 5 and 11 in Mutect2; 1 and 2 in SAMtools; 4 and 21 in SNVer. Overall, FreeBayes and Miseq Reporter were the most concordant for ESR1 (k=0.8091 and 0.9105, respectively; P<0.0001). A significantly lower concordance was observed for PIK3CA, mainly due to pseudogene interference calling the E545A variant. Mutect2 and GATK were the most concordant (k=0.5684 and 0.5591, respectively; P<0.0001). After filtering for clinically significant variants, Miseq Reporter, FreeBayes and SNVer were the most concordant for ESR1 (k=1.0000, 0.8955 and 0.7742, respectively; P < 0.0001). Filtering did not change the concordance displayed by Mutect2 and GATK for PIK3CA.
Conclusions
NGS-based precision medicine workflow generates complex results that need ad hoc data solutions for their analysis. A multidisciplinary approach is essential for correct interpretation of results and successful transfer to the clinic.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
CRO Aviano National Cancer Institute.
Funding
CRO Aviano National Cancer Institute.
Disclosure
All authors have declared no conflicts of interest.