Analytical validation of Oncomine™ Lung cfDNA assay
Sensitivity testing was initially performed starting from 30 ng cfDNA of each Multiplex I cfDNA Reference Standard at 5, 1, 0.1 and 0% mutation frequencies and analyzing specifically 4 different EGFR hotspots, that is E746_A750del, L858R, T790M and V769_D770ins (Table 1). The tag-based NGS detected all EGFR mutations down to the 0.1% allele frequency with high concordance between the measured allele frequencies with those expected for each reference cfDNAs (Table 1). Since the mutant DNA copies for L858R resulted underestimated compared to mutant copies found for the other three variants (Table 1), we checked reproducibility of L858R variant call in critical samples (i.e. those with input cfDNA < 30 ng) and tested the assay using 20 ng of the reference standard at 0.1% mutated allele frequency. A positive call for the L858R mutation was achieved with the minimum allele molecular coverage, i.e. two molecular tags. In contrast, E746_A750del call was missed, although it can still be visualized on IGV and identified by a single molecular tag. No false positive call were observed with the test cfDNA containing wild type EGFR gene (0% mutated allele frequency) only, even when the VCF was visualized on IGV software.
Comparison of sensitizing and T790M EGFR mutations detected by Real Time PCR and tag-based NGS
The plasma cfDNA from a cohort of 42 patients progressing while under TKI was tested by Real Time PCR for EGFR mutations. The same samples were subsequently re-tested using Oncomine™ Lung cfDNA Assay and the results of the two technologies compared.
Using Real Time PCR, 26/42 (61.9%) cfDNA samples displayed the initial sensitizing EGFR mutation seen at the diagnosis in the primary tumor and 9/42 (21.4%) also showed a concurrent T790M mutation, whereas the remaining cases (16/42, 38.1%) resulted negative for both mutations (Fig. 1a, Additional file 4: Table S4). Eight out of nine T790M-positive plasma samples by Real Time PCR were also exon 19 deletion positive, while only one case had a L858R co-occurring sensitizing mutation.
According to the above data, the samples with concurrent sensitizing EGFR mutations and the T790M substitution were 9/26 (34.6%), a finding consistent with the current reports for Real Time PCR assays (Thress et al., 2015a; Mayo-de-las-Casas et al. 2017).
When tag-based NGS was employed on the same cfDNA samples, the cases positive for the original sensitizing EGFR mutation were 36/42 (85.7%), and those harboring the T790M resistance mutation were 18/42 (42.85%) (Fig. 1a, Additional file 4: Table S4). All of the latter cases also had the original sensitizing mutation; therefore, of the 36 cases with a sensitizing mutation, 50% (18/36) also had the T790M mutation, in line with data reported for the detection of resistance mutation in post-TKI tissues (Yu et al. 2013; Hata et al. 2013; Sequist et al. 2011; Oxnard et al. 2011). Among the 18 T790M-positive cases, 11 cases harbored the exon 19 deletion, 5 cases had the L858R mutation and two an unusual EGFR mutations, i.e. one had the rare A763_Y764insFQEA exon 20 insertion and one the G719C exon 18 substitution (Additional file 4: Table S4).
Most of tag-based NGS-positive and Real Time PCR-negative cases for sensitizing EGFR mutations (6/10 cases) had the L858R mutation, leading to the conclusion that the coincidence rate between the two methods was of 94% for exon 19 deletions and 57 and 40% for the L858R and uncommon EGFR mutations, respectively (Fig. 1b).
No difference was observed in each patient regarding the original EGFR sensitizing mutation between the primary tumor tissue and the plasma samples at progression with either the Real Time PCR or the tag-based NGS test, indicating that the specificity of both methodologies was 100% (Additional file 4: Table S4).
Characterization of EGFR allelic fraction detected by tag-based NGS
Subsequently, we investigated the proportion of mutant EGFR alleles (expressed as variant allele frequency, VAF) present in each patient. First of all, no relationship was found between the whole cfDNA (range 2.8–277 pg/μl of plasma) and the mutational load assessed on EGFR gene by tag-based NGS (Additional file 5: Figure S1). Second, when sensitizing EGFR mutations were considered (36 cases), the VAF median percentage was 1.705, with a 0.06–31.3 range (Fig. 2a). In detail, the VAF median value of the 10 cases that were Real Time PCR-negative/tag-based NGS-positive for sensitizing EGFR mutations, was significantly lower compared to that of the 26 cases that were EGFR positive with both methods (0.135 vs 3.68, respectively; p = 0.0002 Mann-Whitney test, Fig. 2b). These data indicate that tag-based NGS detects EGFR mutations present at low-frequency in cfDNA. Third, the majority of these low frequency EGFR mutations (6/10, 60%), were observed among cases harboring the L858R-type mutation (median VAF 0.105, range: 0.06–0.75). Conversely, the median VAF of cases found L858R positive with both Real Time PCR and tag-based NGS was definitely higher, i.e. 5.49 (Fig. 2c, p = 0.0002, Mann-Whitney test).
Fourth, the T790M allele frequencies of the 18 positive cases were shifted towards lower values (median VAF 0.57, range 0.07–14.47) compared to those of EGFR mutations (Fig. 2a and d, Del Re et al. 2017). Again, there was a statistically significant difference between the median VAF value of the 9 cases found T790M-positive by tag-based NGS only and that of the 9 cases that resulted positive by both technologies (0.24 and 1.74 respectively).
Orthogonal validation of T790M by ddPCR
26/42 patients (10 T790M-negative and 16 T790M-positive cases by tag-based sequencing) also were tested for the T790M mutation by the ddPCR assay. In three cases, classified as T790M-positive by tag-based NGS, the ddPCR test was unsuccessful due to low cfDNA quantity available. These were excluded from the comparative analyses. Thirteen of the remaining 23 cases, that were classified as T790M-positive by the tag-based NGS, were confirmed to be positive by ddPCR with a very similar VAF, likewise the 10 cases classified as negative by the tag-based NGS also were confirmed to be negative by ddPCR (Fig. 3).
Comparative analyses of EGFR mutations in plasma and post-TKI tissues
Post-TKI specimens from 15 patients were tested for EGFR mutations by Real Time PCR and results compared with those of cfDNA testing by tag-based NGS (Table 2).
All the post-TKI tissue specimens were positive for the original sensitizing EGFR mutation and 5 of them displayed the T790M resistance mutation. When the results obtained on tissue specimens were compared with those of the corresponding plasma samples, 4/15 cases resulted discordant (26.7%). The original sensitizing EGFR mutation in patient 4 and the T790M resistance mutation in patient 14 were not detected in plasma by tag-based NGS, although they were both present in post-TKI tissue specimens (Table 2). Patient 14 also resulted T790M-negative by ddPCR on cfDNA. In contrast, the remaining discordant cases were found T790M-positive in cfDNA and not in tissues (patients 25 and 39). In both the two post TKI tissues, absence of T790M mutation was confirmed by ddPCR (cut-off > 0.5%; Additional file 3: Table S3). Collectively, the concordance between tissue and plasma was of 93.3% for sensitizing EGFR mutations and 80% for the T790M mutation. Considering mutations found in tumor tissues as reference values, the tag-based NGS appeared to have 80% of sensitivity and specificity for the T790M detection in plasma, in line with other reports (Mayo-de-las-Casas et al. 2017).
Clinical characteristics of the patients with T790M-positive and T790M-negative cfDNA
Clinical data were available for 40/42 patients within the cohort (Additional file 1: Table S1); among these, 18 patients resulted positive for the T790M mutation with tag-based NGS (Additional file 4: Table S4). No significant correlation was observed between T790M status (positive vs. negative) and gender, ECOG (Eastern Cooperative Oncology Group) performance status, smoking habit, age, line of treatment in which EGFR TKI was administered, response to EGFR TKI, or sites of disease progression (extra-thoracic or intra-thoracic) during TKI treatment. However, we observed that positivity for cfDNA T790M mutation was more frequent in patients with the exon 19 mutation than in those with the exon 21 mutation (11/18 vs 5/18, Fisher p-value: 0.046). In addition, all T790M mutation-positive patients had received gefitinib, whereas T790M was not found in patients treated with erlotinib or afatinib (Chi Squared p-value: 0.013); however, this result might be influenced by the substantial disproportion of the administered TKI which favored gefitinib. Among the 18 T790M-positive patients, 17 received treatment with osimertinib (80 mg/day) and were considered evaluable for clinical outcomes (Table 3).
All but one of these were positive for the T790M mutation by tag-based NGS on plasma, while in one the mutation was observed in tissue specimen. Eight of these 17 patients also were T790M mutation positive by Real Time PCR on plasma.
Fourteen patients were evaluable for objective response assessment by RECIST (Response Evaluation Criteria in Solid Tumors) 1.1 as their CT-scans were available at our Institution (Table 3), while all the 17 patients were evaluable for progression-free survival (PFS) and overall survival (OS). All the patients but one achieved at least disease control as best response. Among the 16 patients who were evaluable for RECIST best objective response, nine achieved partial response (PR, 64.28%), six stable disease (SD, 37.5%) and one patient experienced progressive disease (PD, 6.25%). The waterfall plot for objective response of 14/16 patients is reported in (Additional file 6: Figure S2). Most patients with exon 19 deletions (7/9 cases) achieved objective response with osimertinib compared to those patients with other sensitizing mutations (1/4 cases). The median PFS of the osimertinib-treated patients was 8.8 months and the median OS was 16.7 months. There were no differences between patients with exon 19 deletions and those with other mutations in terms of PFS (8.8 vs. 8.6 months; Log Rank p-value: 0.550) or OS (18.0 vs. 16.7 months; Log Rank p-value: 0.513). When we compared the clinical outcome of patients receiving osimertinib according to the results of NGS and Real Time PCR on plasma, we observed the following results. Among the 17 patients who were T790M-positive at tag-based NGS on plasma, eight were positive also at Real Time PCR on plasma, while nine were negative. With regards to RECIST response, T790M Real Time PCR-positive patients achieved the following outcomes: six PR, one SD, one PD; Real Time PCR-negative patients were divided as if follows: three PR, five SD. With regards to survival, among the T790M tag-based NGS-positive patients receiving osimertinib, no significant difference was observed between Real Time PCR-positive and negative patients, both in terms of PFS (12.2 vs. 8.6 months; Log Rank p-value: 0.177) and OS (19.2 vs. 11.6 months; Log Rank p-value: 0.143); similarly to response, these data were based on a small patient population and limited follow up.