Assessing in house comprehensive genomic profiling by liquid biopsy for NSCLC patients

Patients' cohort

This is a prospective observational study conducted between May 2022 and March 2024, enrolling 36 consecutive male and female patients over 18 years of age with locally advanced/metastatic NSCLC who provided informed consent, which were molecularly characterized on bioptic samples prior to target therapy initiation. At disease progression, plasma samples were collected from all patients included in the study for ctDNA analysis with both TruSight Oncology 500 ctDNA version 1 (TSO500, Illumina Inc, San Diego, CA, USA) and Guardant360 CDx (G360, Guardant Health, Palo Alto, CA, USA). Whenever feasible, patients also underwent a tissue core biopsy of one of the metastatic deposits. Therapeutic decisions were undertaken based on the results obtained with the G360 LB results. The study was approved by the Internal Audit Committee and Ethics Committee (INT 283-21).

cfDNA extraction, next-generation sequencing and data analysis

Plasma was obtained from whole blood collections (2 × 10 mL Streck tubes per assay, total volume = 20 mL). G360 is a comprehensive LB test detecting SNVs, indels, CNAs, and fusions across 74 genes in ctDNA, serving as a companion diagnostic (CDx) for multiple targeted therapies (EGFR, ERBB2, KRAS mutations) in NSCLC. ¹⁴ Blood samples allocated for G360 were promptly shipped to Guardant Health facilities, and molecular reports delivered within 7-10 business days. The TSO500 is a LB NGS test for in-house use, designed to detect point mutations (SNV), indels and CNVs across 523 cancer-related genes, as well as fusions and splicing variants in 55 genes, and to characterize key immuno-oncology gene signatures, including TMB and MSI. After double centrifugation, cell-free DNA (cfDNA) was isolated using the QIAamp Circulating Nucleic Acid Kit (cat. 55114; Qiagen, Hilden, Germany). cfDNA was quantified using the Qubit Fluorometer 3.0 (Thermofisher, Waltham, MA, USA) and the dsDNA HS Assay (quantification range: 10 pg/µL–100 ng/µL; Thermofisher). The cfDNA fragment size was determined with the Cell-free DNA ScreenTape Analysis on TapeStation (Agilent Technologies, Inc., Santa Clara, CA). Briefly, at least 30 ng of cfDNA samples were used to generate libraries with TSO500 ctDNA KIT (Illumina). Libraries were quality-controlled with Qubit and TapeStation. Sequencing was carried out on the Illumina NovaSeq 6000, (800 million reads per sample). Raw data were analyzed using the Dragen TruSight Oncology 500 ctDNA v2.1.1 on an Illumina DRAGEN server v4. All variants that passed manufacturer’s quality check (“PASS” tag in VCF files) were used for further analyses. Polymorphisms were filtered by using publicly available databases (e.g. dbSNP, GnomAD, 1000Genomes). Variants were examined using OpenCravat ²¹ and clinically annotated using ClinVar ²² and cBio portal database ²³ ; variants classified as benign or likely benign were excluded from further analyses. The Integrative Genomic Viewer IGV tool ²⁴ was used for final check. Clinical actionability of pathogenic and likely pathogenic (P/LP) variants was defined according to the ESMO Scale for Clinical Actionability of molecular Targets (ESCAT). ²⁵ The final reported data were discussed by the institutional Molecular Tumor Board. ¹⁹ The FDA-approved G360 assay was used as a reference to evaluate the analytical performance of TSO500. For comparative analysis we considered the 74 genes shared between TSO500 and G360 tests. For variants detected by G360 but not confirmed by TSO500, the relevant genomic regions were verified to be included in the TSO500 target design and to be adequately covered in terms of sequencing depth, and Binary Alignment Map (BAM) files were manually reviewed using the Integrative Genomics Viewer (IGV) tool. Standard packages for R software v 4.1.2 was used to explore the data and for descriptive statistics. Complexheatmap package in R was used to draw oncoplots of gene variants.

G360 and TSO molecular results

Among the 31 samples included in the comparative analysis, G360 identified 137 variants (mean per patient: 4.42, range 0-15). Ninety-five variants were annotated as P/LP (mean per patient: 3.06, range: 0-9), and 42 as variants of unknown significance (VUS; mean per patient: 1.35, range: 0-9). In particular, G360 identified 117 SNV/indels, five gene fusions and 15 CNVs. TSO500 ctDNA assay identified 723 variants (mean per patient: 23.32, range: 7-123). Two hundred and seventeen variants were P/LP (mean per patient: 7, range: 0-27), and 489 VUS (mean per patient: 15.77, range: 5-96), including 598 SNV/indels, five fusions and 103 CNV. No significant correlation was found between the DNA input and the number of variants obtained (Pearson’s r = -0.017, p = 0.92) (Online Supplementary Figure S1A). The mean value of TMB obtained by TSO 500 ctDNA assay was 8.9 mut/Mb (range: 0–73.4). No significant correlation was found between the DNA input and TMB value (Pearson’s r = 0.132, p = 0.47) (Online Supplementary Figure S1B). No significant association was found between ctDNA content and number of metastatic sites (Wilcoxon rank-sum test, p = 0.43) (Online Supplementary Figure S2).

G360 and TSO500 comparative analysis

Overall, TSO500 correctly identified 111 out of the 137 variants (81.02%) detected by G360. TSO500 found 102 variants (range per sample: 1-19, median: 3) which were not identified by G360, leading to an overall specificity of 0.97 (range: 0.74-1) (Figure 2A and 2B). TSO500 correctly identified 81 out of the 95 P/LP variants (85.26%) found by G360 (median sensitivity 0.89, range 0.33-1). Additionally, TSO500 detected 138 P/LP variants (range: 1-22, median: 3) not identified by G360 (specificity 0.99, range 0.88-1) (Figure 2C and 2D).

OPEN IN VIEWER

Figure 2.

Analytical performance of the TruSight Oncology 500 ctDNA v1 (TSO500) platform compared to the G360360 CDx test, in evaluating all variants (A,B) and pathogenic/likely pathogenic variants (C,D). The bar charts (A,C) display the proportion of correctly identified variants (sensitivity, blue) and correctly not identified (specificity, red) of TSO500 for each sample. Samples INT283/21_0025 and INT283/21_0035 resulted wild type by G360 testing, and therefore sensitivity calculation was not applicable. As well, sensitivity cannot be extracted in sample INT283/21_0046 where G360 did not identify pathogenic variants. Boxplots (B,D) show the sensitivity and specificity distributions across the whole dataset.

Comparative analysis through ESCAT tier classification

For the evaluation of the potential clinical impact, we compared alterations identified by TSO500 and G360 across ESCAT classes. TSO500 and G360 were consistent in detecting ESCAT I (20 and 21 variants by TSO500 and G360, respectively) and ESCAT II variants (three and one variants, respectively), while TSO500 identified a higher number of ESCAT III (17 and 12 variants) and ESCAT IV (30 and nine variants) variants (Figure 3). Compared with G360, TSO500 showed high sensitivity across ESCAT classes (95.2%, 100%, 58.3%, 88.8% for ESCAT I, II, III, IV class mutations, respectively).

OPEN IN VIEWER

Figure 3.

Boxplots showing the distribution of the absolute number of variants detected in each sample across ESCAT classes.

In particular, among standard of care actionable targets (ESCAT I), TSO500 did not detect an EGFR mutation (p.Leu858Arg) that was identified by G360; in this case, TSO500 run showed a low median coverage (927) (Figure 4).

OPEN IN VIEWER

Figure 4.

Oncoplot showing therapeutically actionable variants (ESCAT I) in EGFR, ALK, KRAS (p.G12C) and RET genes, detected by G360 (left side of each column) and TSO500 (right side of each column). Single nucleotide and indels variants are represented by short red rectangles, while gene fusions by long yellow rectangles.

Notably, all five gene fusions identified by G360 (four ALK and one RET fusions) were confirmed by TSO. TSO500 did not detect five ESCAT IIIA variants, including two PIK3CA, one BRCA1, one PTEN and one ATM variants. Except for the BRCA1 mutation (VAF: 8%), all these variants were characterized by a low variant allele frequency in G360 (range: 0.05%-0,38%). Of note, after a subsequent review of the BAM files using the IGV software, one of the PIK3CA variants and the PTEN variant were detected by TSO500, albeit at low frequency (PIK3CA, VAF: 0.07%) or affected by systematic noise (PTEN, VAF: 0.24). These variants had been excluded from the VCF files due to pre-specified quality filtering. On the other hand, no additional ESCAT I mutations missed by G360 were identified by TSO500 (Figure 4). A high sensitivity (14/15, 93,3%) was also found with regard to CNV for EGFR (six cases), BRAF (two), CCNE1 (two), MET (one), KRAS (one) and CCND1 (one). TSO500 identified 44 potential intrinsic or acquired resistance mutations in 12 out of 26 patients (46.2%), compared to 25 resistance mutations detected by G360 in 12 patients, including EGFR p.L718G, p.C797S and amplification, PIK3CA mutations, MET amplification, cell cycle genes alterations, FGF genes alterations and RAS genes alterations.^20,26-32 In particular, TSO500 detected all but three resistance mutations identified by G360, including the two aforementioned PIK3CA mutations in two patients treated with anti-EGFR therapy, and one ALK mutation (p.Gly1202Arg) in a patient with ALK fusion treated with lorlatinib.

Finally, we observed a high correlation in variant allele frequencies among gene alterations detected by both TSO and G360 (Sperman ρ = 0.935, CCC=0.993) (Online Supplementary Figure S3).

Comparison between LB and tissue NGS results at relapse

In nine out of 27 patients (33.3%), concurrent tissue-based CGP was performed at the same disease progression timepoint, allowing a direct comparison between tissue and LB NGS results.

Overall, in 7/9 cases TSO500 identified 16 potentially actionable (ESCAT I-IV) and resistance mutations which were not identified by concurrent tissue NGS, including two ESCAT IA variants (two KRAS G12C), five ESCAT IIIA variants (two BRCA1 deletions, one PIK3CA mutation, one KRAS mutation and one KMT2A mutation), one ESCAT IIIB (MET mutation), five ESCAT IV variants (one CHEK2 mutation, and PIK3CA, FGFR3, BRAF, and CDK6 amplifications), and three EGFR resistance alterations (one amplification and two mutations). Of note, in a patient with an ALK-EML4 positive NSCLC progressing after two lines of ALK inhibitors (Alectinib and Lorlatinib), both G360 and TSO500 detected a KRAS p.G12C and a PIK3CA p.E542K mutation, which were not identified by concurrent tissue NGS. Upon further progression to a third line of treatment with carboplatin-pemetrexed, this patient was subjected to additional molecular profiling in both tissue and LB samples. The KRAS variant was detected in subsequent analysis both in tissue and LB samples, while the PIK3CA variant was identified in the plasma sample but not in tissue. For the complete list of pathogenic mutations detected by TSO500, G360 and tissue NGS see Online Supplementary Table S1.

Comparison between LB at relapse and tissue NGS at diagnosis

Based on the availability of tissue NGS at diagnosis, we evaluated whether the resistance mechanisms identified by TSO500 LB were intrinsic or acquired. Comparable data for pre-treatment tumor tissue and LB at relapse were available for 5/12 patients (the remaining seven patients were characterized in tissue at diagnosis by targeted small NGS panels). In four of these cases, TSO500 LB identified mutations which were not detected in tissue at diagnosis, thus representing putative acquired mutations. These included acquired EGFR p.L718Q and BRAF p.V600E mutations in two patients with EGFR mutations treated with osimertinib, receiving platinum-based chemotherapy and savolitinib plus osimertinib/placebo within a clinical trial (NCT03778229), respectively; one MET amplification plus KRAS p.G12C mutation in an ALK positive patients previously treated with alectinib, receiving platinum based chemotherapy; and one BRAF p.V600E plus ALK missense mutation in two ALK-rearranged patients treated with lorlatinib. However, in patients harboring gene alterations with strong evidence of actionability (BRAF p.V600E and KRAS p.G12C) detected by LB, targeted therapy remained inaccessible due to the unavailability of tissue-based NGS confirmation for BRAF and limited off-label access or clinical trial availability for KRAS p.G12C. This highlights the critical need for broader drug access for alterations identified through LB.

In the remaining patient, TSO500 LB did not detect any acquired variant, confirming the ERBB2, CCNE1, and EGFR amplifications previously identified in the pre-treatment tissue biopsy by CGP.

At baseline, 11 patients had actionable gene fusions. At disease progression, LB detected gene fusions in four of these patients (five out of 12 plasma samples, 41.6%).

In the remaining seven patients, G360 testing also did not detect any gene fusion. Concurrent tissue biopsies at progression were available for three of these seven patients; in two cases, tissue analysis confirmed the persistence of the original RET fusion, while in the third case, both IHC and ISH confirmed the loss of the ALK fusion.

Among all tissue samples collected at progression, gene fusions were detected in seven cases. TSO500 assay successfully identified gene fusions in five of these cases, yielding a detection rate of 71.4%.

Comparative analysis between TMB values detected by tissue CGP (both prior target therapy and at disease progression) and LB TSO500 revealed poor correlation (Pearson’s r = 0.228, p = 0.498) (Online Supplementary Figure S4).