To achieve the goal of personalized and precision medicine, CGP has been recognized as a key tool that can potentially transform cancer risk prediction, detection, diagnosis, treatment, and monitoring. CGP can augment clinical decision making in the form of either In vitro diagnostic (IVD) or companion diagnostics (CDx), and most importantly, the continuous price reduction across small- to large-sized CGP panels has improved patient access [22]. Consequently, therapy assignment for newly-diagnosed breast cancers and monitoring of patients who are in treatment are expected to benefit from CGP. In this study we compare the performance of medium- to large-sized gene panels in a TNBC patient cohort.
A total of 108 breast cancers previously assayed with the OCP v3 with sufficient nucleic acid left were re-sequenced with TSO500, a large-sized CGP panel. Nowadays there are plenty of commercial and in-house targeted sequencing panels for cancer research as well as for clinical application. In the project AACR GENIE which integrated genomic data from panels varying in size from approximately 50 to 500 genes, a core of 44 genes had been recognized across various platforms [23]. A large-scale basket study, the NCI-MATCH, adapted the Oncomine AmpliSeq panel (ThermoFisher Scientific) targeting 143 genes with more than four thousand variants annotated [24–26]. On the other hand, the MSI-IMPACT trial developed a customized and hybridization capture-based 468-gene panel, which was Clinical Laboratory Improvement Amendments (CLIA) compatible and was advocated for the potential of identifying variants with clinical significance [27]. Commercial products such as the FoundationOne CDx, ActOnco and TruSight Tumor 170 followed [28–30].
Tumor tissues from the VGH-TAYLOR study were sequenced with the OCP v3 assay, which was designed as a research use only (RUO) assay and detected thousands of variants across 161 cancer-relevant genes [31]. Types of mutations detected such as frameshift, missense, synonymous, SNV, Indel, and CNV observed in individual breast cancer patients were reported [11]. Contrary to the OCP v3, TSO500 assay interrogates 523 cancer-related genes as well as multi-gene biomarkers such as microsatellite instability (MSI), tumor mutational burden (TMB) and homologous recombination deficiency (HRD) [32]. Both platforms performed tumor-only sequencing, which meant that reflex germline testing might be needed if the differentiation of somatic from germline origin was needed, such as the case of PARP inhibition [33–34]. On the other hand, the Todai OncoPanel reimbursed in Japan did provide paired tumor-normal test [35–37].
The four HER2-positive breast cancers provided an opportunity to evaluate the correlation between clinical HER2-status and NGS-based ERBB2 CNV. None of these cases were reported as HER2-amplified by OCP while two out of three HER2-amplified cases reported by TSO500 coincided with the four clinically HER2-positive breast cancers (Figs. 1, bracket). More HER2 gained breast cancers were identified by TSO500. As NTRK translocation was not listed as the targeted alteration of OCP, we did not compare the detectability of structure variants (CNV and fusion) further.
Table 1 details the genes and variants reported from either OCP v3 or TSO500 when genomic, cDNA and protein coordinates were used for variant sorting. This subset represented the most actionable variants revealed by medium- and large-sized CGP. AKT1 is an intra-cellular kinase and is predominantly altered in breast and endometrium cancer. The AKT1 E17K is the most common alteration across various tumor types [38–39]. The oral competitive kinase domain inhibitor capivasertib (AZD5363) has been shown to inhibit AKT1 E17K-mutated breast cancer. The CAPItello-291 phase III trial demonstrated that capivasertib and fulvestrant combination therapy resulted in significantly longer progression-free survival than treatment with fulvestrant alone among HR-positive/HER2-negative advanced breast cancer patients whose disease had progressed during or after previous aromatase inhibitor therapy with or without a cyclin-dependent kinase (CDK) 4/6 inhibitor [40]. The three AKT1 E17K-mutant cases were identified by both panels (Table 1).
There were two BRCA1 (S405* and R1203*) and one BRCA2 (S521*) truncating mutations; reflex germline testing should be proceeded if PARP inhibitior was considered for these patients [41]. The BRCA1 c.5470-1G > A SNP has been recognized as pathogenic by both the Breast Cancer Information Core (BIC) and Consortium of Investigators of Modifiers of BRCA1/2 (CIMBA). The BRCA2 S521* truncating mutation impairs nuclear localization of BRCA2, which is essential for normal BRCA2 function [42]. Despite common truncating and frameshift mutations (5 for BRCA1 and 3 for BRCA2) called by both panels, only TSO500 did reveal more suspicious variants (8 variants among 9 cases for BRCA1 and 13 variants among 21 subjects) while OCP was flawed by spurious mutations (BRCA1: S1180fs, T1376fs and BRCA2: S538fs, T598fs, S973fs, E33*, T912fs, S3041fs, S3147fs).
PALB2 K353fs is a truncating mutation in a tumor suppressor gene, and therefore is likely oncogenic. The phase II TBCRC 048 study also ascertained the predictive power of germline PALB2 mutation as high response rate (82%) was observed when olaparib, a PARP inhibitor, was used [43]. TSO500 identified 3 additional variants in 5 cases (Table 1) while OCP reported one false positive case with up to 4 variants (D1125fs, N368fs, S357fs and N342fs). In current study, no reflex germline testing was performed and the germline/somatic origin of such mutations remained unknown.
One patient harbored two mutations in ERBB2 (L811V and L839R) and another with L725S mutation was recognized by both platforms, none of which were among the oncogenic mutations approved by FDA for the use of neratinib, a pan-HER kinase inhibitor that binds irreversibly to the ATP-kinase domain of HER2 inhibiting the downstream phosphorylation of AKT and MAPK [44]. TSO500 identified three additional ERBB2 mutations (P1140A, V743_M744insHV and Y742_A745dup). It deserved notice that TSO500 detected more complex mutations, i.e. Y742_A745dup and V743_M744insHV, both classified as VUS, than OCP.
A couple of PIK3CA hotspot mutations have been identified from the SOLAR-1 trial for the usage of alpelisib, a selective PI3K-alpha inhibitor, namely C420R, E542K, E545A, E545D, E545G, E545K, Q546E, Q546R, H1047L, H1047R and H1047Y [45]. Our previous study also observed double mutations of PIK3CA among Taiwanese population [14]. Only four cases with E726K, G1049R, N345I and Q546R mutations were outside the hotspot region, while two cases with D350N or D725N were co-mutant with other hotspots, which were only reported by TSO500 (Table 1). Consequently, both platforms identified the same number of PIK3CA-altered breast cancers.
On the other hand, selective PI3K-beta inhibitor, GSK2636771 and AZD8186, are ATP-competitors and have shown preclinical anti-tumor response and durability for PTEN loss (deficiency) solid tumors including TNBC [46]. The PTEN c.1026 + 1G > A SNP has been reported as clinically pathogenic at least twice (Invitae and ClinGen PTEN Variant Curation Expert Panel) with germline origin [47–48]. The remaining 5 consensual variants were truncating mutations; TSO500 identified two additional cases with PTEN mutations (G127_G129del and V290Sfs*8) and OCP called one inappropriate variant due to malalignment (c.386G > A).
A plethora of TP53 alterations were identified by both OCP and TSO500. Despite not being currently druggable, TP53 may act as an independent prognostic factor for early and advanced-stage cancer with a wide range of mutational frequency [49]. It has been argued that all TP53 mutations from tumor-only sequencing are somatic, and rarely representative of the germline Li-Fraumeni syndrome [41]. TSO500 identified more variants than OCP, while the latter reported three cases with V73fs in homopolymer region and one case with mal-aligned V73fs mutation, both of which were fictitious.
Figure 2 shows 272 variants identified among actionable genes and TP53 from at least one panel. Interestingly, the concordance rate was low at slightly more than one-third (34.6%). With extensive bioinformatics analyses and manual curation, three-fifths of concordance (58.9% of 202 variants) was achieved after exclusion of unfiltered polymorphisms, low VAF variants and those beyond the targeted scope of OCP. An example of such analysis is given in Supplementary Fig. 1. Table 1 also highlights missed opportunities by OCP as an additional 62 variants were revealed by TSO500. Fundamental discrepancy in sequencing technology may provide some explanation. The amplicon-based OCP used PCR amplification of target regions while the hybrid capture-based TSO500 used hybridization with biotinylated probes to capture target regions. Inherited methodology indicates that PCR amplification can introduce biases and artifacts, potentially affecting the accuracy of the results. Among inconsistent variants, some were identical after manual inspections, further indicating the necessity of analytical abilities and domain knowledge in genomic nomenclature [50].
Despite being one of the pioneers directly comparing two commercialized targeted panels, there were some limitations of the study. First, sequencing was not conducted concurrently, and the 1–2 years of lag of TSO500 analysis following OCP might introduce some bias from nucleic acid degradation, despite all samples being stored under temperature-controlled conditions. Second, variants called by each panel were considered as they were based on the standard or formal algorithm of each platform (Ion Reporter/Oncomine Knowledgebase Reporter for OCP and PierianDx for TSO500), which limited the comparability between distinct platforms. For unbiased comparison, the same aligner, caller and annotator should be applied. However, BED files were unavailable from manufacturers to confirm the jointly interrogated regions. On the other hand, all commercial CGP solutions are under regulation as either laboratory developed tests (LDTs) or in vitro diagnostics (IVDs), and practically these pre-set algorithms should not be modified arbitrarily to enhance reproducibility. Despite this, we were able to conduct an exhaustive bioinformatic analysis from BAM files to dissect the conflicting results from the same samples. Third, NTRK is not within the targeted fusion genes of OCP and therefore a comparison with TSO500 is not possible. Moreover, NTRK sequence variants, amplifications or fusions are not actionable for breast cancer. As larotrectinib and entrectinib are approved in many countries, the importance of tumor-agnostic marker NTRK fusion cannot be overemphasized, and its detection ability should be incorporated into CGP for breast cancer [51].
In conclusion, this study compared the yield of actionable mutations between two CGP commercial assays that vary in size and found that three-fifths could be detected by both platforms. TSO500, the larger panel, detected more variants than OCP even from the same set of ESCAT-defined actionable genes and TP53. On the other hand, a proportion of inconsistent variants could be manually curated and were identical with aliasing coordinates or starting positions. Finally, there were variants detectable only by TSO500, indicating potential and fundamental differences in sequencing technology, bioinformatic algorithms and variant filtering. The value of a large-sized panel in clinical usage is ascertained. Given the experience from this study, the updated Oncomine Comprehensive Assay Plus with more than 500 genes profiled are anticipated in future studies.