Picoliter-Droplet Digital Polymerase Chain Reaction-Based Analysis of Cell-Free Plasma DNA to Assess EGFR Mutations in Lung Adenocarcinoma That Confer Resistance to Tyrosine-Kinase Inhibitors

Lung adenocarcinoma patients who received epidermal growth factor receptor (EGFR)-tyrosine-kinase inhibitor (TKI) therapy were subjected to picoliter-droplet digital polymerase chain reaction (ddPCR)-cell-free plasma DNA (cfDNA) analysis to determine the fraction of cfDNA with TKI-sensitive and -resistant mutations, as well as their concordance with mutation status in rebiopsied tumor tissues. Picoliter-ddPCR examination of cfDNA, supported by next-generation sequencing analysis, enables noninvasive assessment of EGFR mutations that confer resistance to TKIs.

Abstract

Purpose.

The objective of this study was to evaluate the utility of analyzing cell-free plasma DNA (cfDNA) by picoliter-droplet digital polymerase chain reaction (ddPCR) to detect EGFR mutations that confer resistance to tyrosine-kinase inhibitors (TKIs) used for treatment of lung adenocarcinoma (LADC).

Experimental design.

Thirty-five LADC patients who received epidermal growth factor receptor (EGFR)-TKI therapy, including ten who received tumor rebiopsy after development of resistance, were subjected to picoliter-ddPCR-cfDNA analysis to determine the fraction of cfDNA with TKI-sensitive (L858R and inflame exon 19 deletions) and -resistant (i.e., T790M) mutations, as well as their concordance with mutation status in rebiopsied tumor tissues.

Results.

cfDNA samples from 15 (94%) of 16 patients who acquired resistance were positive for TKI-sensitive mutations. Also, 7 (44%) were positive for the T790M mutation, with fractions of T790M (+) cfDNA ranging from 7.4% to 97%. T790M positivity in cfDNA was consistent in eight of ten patients for whom rebiopsied tumor tissues were analyzed, whereas the remaining cases were negative in cfDNA and positive in rebiopsied tumors. Prior to EGFR-TKI therapy, cfDNAs from 9 (38%) and 0 of 24 patients were positive for TKI-sensitive and T790M mutations, respectively. Next-generation sequencing of cfDNA from one patient who exhibited innate resistance to TKI despite a high fraction of TKI-sensitive mutations and the absence of the T790M mutation in his cfDNA revealed the presence of the L747P mutation, a known driver of TKI resistance.

Conclusion.

Picoliter-ddPCR examination of cfDNA, supported by next-generation sequencing analysis, enables noninvasive assessment of EGFR mutations that confer resistance to TKIs.

Implications for Practice:

Noninvasive monitoring of the predominance of tumors harboring the secondary T790M mutation in the activating mutation in EGFR gene is necessary for precise and effective treatment of lung adenocarcinoma. Because cells harboring the T790M mutation are resistant to epidermal growth factor receptor-tyrosine-kinase inhibitors (TKIs), the predominance of tumor cells harboring the T790M mutations influences the choice of whether to use conventional or next-generation TKIs. Digital polymerase chain reaction-based examination of cfDNA is a promising method; however, its feasibility, including its consistency with examination of rebiopsied tumor tissue, has not been fully proven. Here, picoliter-droplet digital polymerase chain reaction technology is presented as a candidate method for testing cfDNA and assessing the predominance of T790M-mutant tumors.

Introduction

EGFR (epidermal growth factor receptor) is a driver gene of non-small cell lung cancer (NSCLC), particularly lung adenocarcinoma (LADC). Activating somatic mutations in this gene define a subset of cases that respond to specific EGFR-tyrosine-kinase inhibitors (TKIs) such as gefitinib and erlotinib [1, 2]. The most frequent mutations in EGFR occur in the exons encoding the kinase domain of EGFR, including various types of in-frame deletions in exon 19 (19del) and a point mutation in exon 21 leading to the substitution of leucine for arginine at position 858 (L858R). Tumors harboring these TKI-sensitive mutations nearly always acquire resistance to TKIs within 2 years [3, 4]. The most common mechanism of resistance, accounting for 60% of cases, is the occurrence of the secondary mutation T790M (replacing a gatekeeper amino acid) in the EGFR allele harboring the TKI-sensitive mutation [5]. To overcome resistance to conventional EGFR-TKIs, a new generation of drugs (including AZD9291, CO-1686, and HM61713) that suppress the kinase activity of EGFR proteins harboring secondary T790M substitutions is currently being developed [6–9]. Phase I clinical trials demonstrate that progressed NSCLC patients who are diagnosed with T790M-positive tumors by genetic testing of rebiopsied tumor tissues respond to these new drugs [10]. However, because the new drugs bind their targets irreversibly, they are associated with severe side effects that are not observed during conventional EGFR-TKI therapy. In addition, other mutations in EGFR also confer resistance [11]. Therefore, to achieve precise and effective treatment of EGFR mutation-positive NSCLC patients, it is necessary to monitor the predominance of EGFR mutations that confer TKI resistance during therapy; the choice between conventional and next-generation EGFR-TKIs must be made based on the identities of the EGFR mutations conferring TKI resistance [6, 7].

Circulating plasma cell-free DNA (cfDNA), which is released into plasma from tumor tissues or circulating tumor cells (CTCs), represents a promising source of material for noninvasive liquid biopsy that could provide genetic information about CTCs and residual tumor cells [12–14]. cfDNA is particularly attractive for use in the lung cancer clinic due to the occasional difficulty of obtaining tumor tissues with high cellularity [15, 16]. Indeed, EGFR mutations present in tumor cells can be detected in the cfDNA of NSCLC patients using digital polymerase chain reaction (PCR) [17–20] and next-generation sequencing (NGS) [21, 22]. In particular, TKI-sensitive and T790M mutations in the cfDNA of NSCLC patients have been successfully detected using a digital PCR-based method called BEAMing (beads, emulsion, amplification, and magnetics) [15, 21, 23, 24]. Thus, cfDNA represents a promising source of material for noninvasive monitoring of tumor burden. However, several issues need to be resolved before these methods can be applied in the lung cancer clinic, including the concordance of T790M mutation status between cfDNA and rebiopsied lung cancer tissues, as well as their compatibility with EGFR mutation tests currently performed for biopsied tumor tissue. Regarding the latter point, routine EGFR mutation tests, such as the Scorpion/ARMS assay, provide information about the presence/absence and location of driver mutations, but not the exact sequence of the mutants, in each tumor [15, 21]. Therefore, cfDNA examination during EGFR-TKI therapy should use this information.

In this study, we established a picoliter-droplet digital PCR (ddPCR) system to quantify TKI-sensitive and -resistant EGFR mutations in the cfDNA of patients who were treated with conventional EGFR-TKIs. Our picoliter-ddPCR system has several advantages. First, it can identify the major types of EGFR exon 19 in-frame deletions in a single assay using a common probe, enabling us to detect >95% of known mutations [25] without prior information about the exact mutant DNA sequence; this cannot be provided by routine tests. Second, picoliter-ddPCR is performed in millions of droplets, including hundreds to thousands of droplets containing a single molecule of template DNA, preventing inaccuracy because of the inclusion of two or more EGFR DNA molecules in a droplet; therefore, this assay yields accurate estimates of the fraction of mutant DNA. Third, the assay is simple and rapid, consisting only of the PCR and detection procedures (supplemental online Fig. 1), making it feasible for routine use in the lung cancer clinic.

We examined cfDNA samples from 35 LADC patients who received EGFR-TKI therapy: 19 provided cfDNA before EGFR-TKI therapy, and 16 provided cfDNA after developing resistance to EGFR-TKIs,. We examined the samples using picoliter-ddPCR to determine the fraction of cfDNAs with EGFR-T790M mutation to assess the predominance of T790M-positive tumor cells and the concordance of T790M mutation status between cfDNA and rebiopsied tumor tissues.

Materials and Methods

Picoliter-ddPCR

The assay for detecting representative exon 19 in-frame deletions used two TaqMan probes (Fig. 1A). The FAM-labeled wild-type probe was designed to hybridize to a region in EGFR exon 19, in which in-frame deletions occur; therefore, this probe does not hybridize with deletion alleles. The VIC-labeled reference probe was designed to hybridize to a region in EGFR exon 19, in which in-frame deletions do not occur; therefore, this probe hybridizes with both wild-type and deletion alleles. This assay was designed to detect 34 representative exon 19 deletions covering >95% of deletion mutations in LADC [25] (supplemental online Table 1). The assays to detect the L858R and T790M mutations also used two TaqMan probes, one specific for the mutant allele and the other specific for the wild-type allele. For L858R, the mutant-specific probe detected 2753T→G, the predominant form of this mutation. For the T790M mutation, the probe detected 2369C→T [26].

Figure 1.

Picoliter-ddPCR for EGFR-cfDNA. (A): Assay design. Left: An assay to detect 19DEL. The assay was designed to detect the loss of wild-type sequence and could therefore detect multiple in-frame deletions in exon 19. Wild-type DNA generated signals from both the wild-type probe (FAM, green) and reference probe (VIC, red), whereas deletion alleles generated a signal only from the 19DEL reference probe. Right: Assays to detect the L858R and T790M mutations. Mutant DNA generated signals from the mutation-specific (FAM, green) probe, whereas wild-type DNA generated a signal from the wild-type probe (VIC, red). (B): Assessment of the predominance of T790M-positive tumor cells among all tumor cells. cfDNA was subjected to T790M and 19DEL or L858R mutation assays. The fraction of T790M-cfDNA in 19DEL/L858R-mutant (i.e., tumor-derived) cfDNA reflects the predominance of T790M-positive (i.e., TKI-resistant) tumor cells among all tumor cells. The result from case 1 after acquisition of resistance (one of the triplicated assays in supplemental online Table 2) is shown as a representative. FAM and VIC intensities are shown in arbitrary units.

Abbreviations: 19DEL, exon 19 deletion; cfDNA, cell-free plasma DNA; Mut, mutation.

Digital PCR was performed using the RainDrop Digital PCR System (RainDance Technologies, Billerica, MA, http://raindancetech.com), in which PCR takes place in millions of droplets with volumes of ∼5 pL [27–29] (supplemental online Fig. 1). PCR solutions (40 µL) were prepared by mixing 20 µL of QuantStudio 3D Master Mix (Life Technologies, , Rockville, MD, http://www.lifetech.com), 4 µL of 10× Droplet Stabilizer (RainDance Technologies), 2 µL of TaqMan SNP genotyping assay (Life Technologies), and DNA. The mixture was subjected to emulsification, followed by PCR with the following parameters: 95°C × 10 minutes (1 cycle); 45 cycles of 95°C × 15 s and 60°C × 1 minute; 98°C × 10 minutes; and a 10°C hold. The endpoint fluorescence signal (i.e., the fluorescence intensities of VIC [red] and FAM [green]) of each individual droplet was measured and visualized as clusters in a two-dimensional histogram. Spectral compensation was applied to each sample to eliminate contamination of fluorescence signals between the VIC and FAM fluorophores. Compensation factors, as well as the respective thresholds to define droplets positive for exon 19 deletions and the L858R and T790M mutations, were set based on data from positive-control cell lines.

Analysis of Genomic DNA From Lung Cancer Cell Lines

Genomic DNA extracted from cell lines was used to assess the accuracy and reproducibility of picoliter-ddPCR. The II-18 lung cancer and ACC-MESO-1 mesothelioma cell lines were obtained from the RIKEN BioResource Center (Tsukuba, Japan, http://en.brc.riken.jp/index.shtml). The NCI-H1975 lung cancer cell line was obtained from the American Type Culture Collection (Manassas, VA, http://www.atcc.org). The PC-9 lung cancer cell line was obtained from the late Dr. Yoshihiro Hayata of Tokyo Medical University, who established this cell line. In the II-18, H1975, and PC-9 cell lines, the alterations of cancer-related genes were consistent with those in previous reports and in the COSMIC database, confirming cell line authenticity [30–32]. We confirmed a lack of EGFR mutations in ACC-MESO-1 by real-time PCR. Genomic DNA was extracted using the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany, http://www.qiagen.com). After DNA was sheared to a size range of 2–4 kb using a Covaris S2 System (Covaris, Woburn, MA, http://covarisinc.com), DNA fragments were purified using the MinElute PCR purification kit (Qiagen). Quantitative and qualitative analyses of the purified products were performed on a 2100 Bioanalyzer (Agilent Technologies, Boeblingen, Germany, http://www.agilent.com). Genomic DNA from those cell lines, as well as H1975 DNA serially diluted from 50% to 1.56% with ACC-MESO-1 DNA, was subjected to picoliter-ddPCR in triplicate. The amounts of DNA used in picoliter-ddPCR are described in the legend of supplemental online Figure 2. The linear correlation coefficient values (R²) were calculated using the IBM SPSS statistics software (version 20.0; IBM, Armonk, NY, http://www-01.ibm.com/software/analytics/spss). The picoliter-ddPCR experiments were designed and performed following the essential requirements in the MIQE guidelines for ddPCR [33].

Feasibility Assessment of Picoliter-ddPCR

The feasibility of the picoliter-ddPCR assay described above was examined using DNA from three EGFR-mutant lung cancer cell lines: PC-9, harboring an exon 19 deletion; II-18, harboring the L858R mutation; and NCI-H1975, harboring the T790M and L858R mutations [34, 35]. Detection of EGFR-mutated DNA was reproducible, and background mutations were not detected (supplemental online Fig. 2A). The quantitative nature of the assay was validated using serially diluted DNA (supplemental online Fig. 2B; R² = 0.96), and the limit of detection was defined as more than 0.75% mutant alleles. The calculated fraction of T790M alleles in all TKI-sensitive alleles was reproducible when as little as 2 ng of DNA was used in picoliter-ddPCR (supplemental online Fig. 2C); under these conditions, hundreds of droplets positive for EGFR wild-type alleles were detected. Thus, we concluded that this method was suitable for examining the proportion of alleles with TKI-sensitive mutations that also harbored the T790M mutation.

Analysis of cfDNA From NSCLC Patients

Peripheral blood samples were collected after obtaining written informed consent from 35 LADC patients receiving EGFR-TKI therapy. From 16 of the 35 patients, blood samples were collected after confirmation of resistance. Of these 16 patients, 5 also provided blood samples before the administration of EGFR-TKI; thus, a total of 21 blood samples were collected from the 16 patients who acquired resistance (16 + 5 = 21). From the remaining 19 patients, blood samples were collected only before the administration of EGFR-TKI. Thus, a total of 40 (21 + 19) peripheral blood samples were collected from 35 LADC patients. In addition, peripheral blood samples were also collected from two patients with LADC harboring the EML4-ALK fusion; these samples were used as EGFR mutation-free controls. From each patient, blood samples were collected in two 5-mL EDTA-containing Vacutainers and spun to separate plasma within 30 minutes of collection. Plasma samples were kept frozen at −80°C until DNA extraction. We also analyzed plasma samples obtained from patients 1, 8, 10, and 11 before the initiation of EGFR-TKI therapy. cfDNA was extracted from 2 mL of plasma using the QIAamp circulating nucleic acid kit (Qiagen) and quantitated using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Waltham, MA, http://www.thermofisher.com; Life Technologies) and a Qubit fluorometer (Invitrogen, Carlsbad, CA, http://www.invitrogen.com). cfDNA was used in picoliter-ddPCR without being sheared. Reaction mixtures containing 4 ng of cfDNA, as determined using a Qubit fluorometer (and corresponding to the amount of cfDNA obtained from 1 mL of plasma in the majority of cases), were subjected to picoliter-ddPCR in triplicate (patients 1–3, 13, and 17) or in single reactions. This study was approved by the institutional review board of the National Cancer Center. The study was registered in the University Hospital Medical Information Network Clinical Trial Registry (UMIN 000017581).

Threshold Setting for Judgment of Positivity

To set the threshold for calling EGFR mutations in patient cfDNA, cfDNA from two patients with LADC harboring the EML4-ALK fusion were subjected to picoliter-ddPCR. These samples were considered negative for EGFR mutations because the ALK fusion is mutually exclusive with EGFR mutations in lung cancer [36]. Only a few droplets among millions were positive for EGFR mutations (mean = 2.5, SE = 1.4) (supplemental online Fig. 3). It remained possible that a few EGFR mutations were present, even in ALK fusion-positive LADC patients; however, to avoid false-positive results, the threshold for a positive call was tentatively set to 10 droplets, based on the following equation: mean + 5 × SE = 9.5. Using this threshold, the rate of false-positive droplet detection was predicted to be less than 0.0002%.

EGFR Mutation Analysis in Biopsied Tumor Tissue

Formalin-fixed, paraffin-embedded tumor tissues from patients were submitted for high-resolution melting analysis or Scorpion-ARMS-based diagnosis of EGFR mutations during the course of standard clinical practice between December 2003 and June 2015. EGFR-TKI-resistant tumor tissues were obtained by rebiopsy of pericardium (patient 1), primary disease (patients 2 and 9), liver metastasis (patients 3, 5, and 7), pleural effusion (patients 8 and 10), or lymph nodes (patients 4 and 6) and were subjected to Scorpion-ARMS-based clinical examination for T790M mutations.

Genome-Capture Deep Sequencing Using a Next-Generation Sequencer

Nucleotide sequences of EGFR were examined by targeted genome capture and massively parallel sequencing using a MiSeq sequencer and a 90-gene targeted panel, the National Cancer Center (NCC) oncopanel (catalog no. 931196; Agilent). One microgram of cfDNA was subjected to enrichment using the NCC oncopanel probes. The mean depth of sequencing was 1,027.

Results

Picoliter-ddPCR to Detect EGFR Mutations in cfDNA

Picoliter-ddPCR assays to detect exon 19 deletions and L858R and T790M mutations were designed as shown in Figure 1A. In this analysis, each cfDNA sample was subjected to two picoliter-ddPCR assays: one for the T790M mutation and the other for either a TKI-sensitive exon 19 deletion or a L858R mutation, chosen based on a routine clinical test of tumor samples before EGFR-TKI therapy. The assay for exon 19 deletions was designed to detect loss of the wild-type signal and could therefore detect representative in-frame deletion mutations in exon 19. The L858R and T790M assays were designed to detect wild-type and mutant alleles with each probe. Based on data from these two picoliter-ddPCR assays, the proportion of T790M cfDNA among all tumor cfDNA, as represented by TKI-sensitive mutations, was deduced as shown in Figure 1B.

Study Cohort Consisting of 35 LADC Patients Who Received EGFR-TKI Therapy

We prepared a cohort of 35 patients with advanced LADCs (Table 1). Their tumor tissues obtained before EGFR-TKI therapy were diagnosed as positive for TKI-sensitive EGFR mutations; all patients received molecularly targeted therapy with EGFR-TKIs (Table 1). In 16 cases (patients 1–16: TKI-resistant cohort), cfDNA was obtained after the confirmation of resistance to EGFR-TKIs (Figure 2). From five cases (patients 1, 8, 10, 11, and 12), cfDNA was also available from before EGFR-TKI therapy. In the other 19 cases (patients 17–35: pre-TKI cohort), cfDNA was obtained only before initiation of EGFR-TKI therapy (supplemental online Fig. 4).

Table 1.

Study subjects for picoliter-droplet digital polymerase chain reaction analysis of cfDNA

Figure 2.

Therapeutic process of LADC patients in the TKI-resistant cohort. Content and duration of therapy, as well as the timing of EGFR mutation tests, are illustrated for 16 patients whose cfDNA was obtained after acquisition of resistance to EGFR-TKI therapy. From patients 1, 8, 10, 11, and 12, cfDNA samples were also obtained before EGFR-TKI therapy. All patients were diagnosed as having tumors harboring TKI-sensitive mutations, based on analysis of biopsied tissue samples. In patients 1–10, rebiopsied tumor tissues obtained after acquisition of resistance to EGFR-TKIs were also tested for T790M mutations. Positivity for TKI-sensitive (TKI-mut) and T790M mutations in cfDNA and rebiopsied tumors is shown on the right.

Abbreviations: BSC, best supportive care; ddPCR, droplet digital polymerase chain reaction; EGFR, epidermal growth factor receptor; LADC, lung adenocarcinoma; RT, radiotherapy; TKI, tyrosine-kinase inhibitor.

cfDNA Analysis for TKI-Sensitive and T790M Mutations

We first examined cfDNA obtained after acquisition of resistance from 16 cases of the TKI-resistant cohort. cfDNA samples from 15 (94%) and 7 (44%) patients were positive for TKI-sensitive and T790M mutations, respectively (Figs. 2, 3; detailed data in Table 1 and supplemental online Table 2). In 10 of these 16 cases, rebiopsied tumor tissues were subjected to a T790M test after the tumor became resistant to EGFR-TKI (patients 1–10 in Fig. 2); seven (patients 1–7) were positive for the T790M mutation, and the remaining three (patients 8–10) were negative. Positivity and negativity for the T790M mutation was concordant between cfDNA and rebiopsied tumors in 8 patients (concordance rate, 0.8), whereas in the remaining 2 cases, only the rebiopsied tumors, but not the cfDNAs, were positive. If we assume that the test results of rebiopsied tumors were correct, our cfDNA assay had a sensitivity of 71% (5 detected in cfDNAs and 7 detected in rebiopsied tumors), with no false positives (supplemental online Fig. 5).

Figure 3.

Positivity for EGFR mutations in cfDNA. Percentages of cfDNA samples positive for TKI-sensitive and T790M mutations are shown. In total, 16 samples were obtained after the acquisition of resistance to EGFR-TKI therapy (patients 1–16), and 24 samples were obtained before EGFR-TKI therapy (patients 1, 8, 10, 11, 12, and 17–35).

Abbreviations: EGFR, epidermal growth factor receptor; TKI, tyrosine-kinase inhibitor.

Among the 16 patients in the TKI-resistant cohort, cfDNA obtained before EGFR-TKI therapy was available from 5 (patients 1, 8, 10, 11, and 12; Fig. 2). In total, 3 of them were positive for a TKI-sensitive mutation, whereas all were negative for the T790M mutation. Consistent with this, cfDNAs obtained from 19 patients before EGFR-TKI therapy (patients 17–35) were all negative for T790M mutations, whereas 6 cases (32%) were positive for TKI-sensitive mutations (supplemental online Fig. 4). Of the 24 cfDNA samples obtained before EGFR-TKI therapy, 9 (36%) were positive for TKI-sensitive mutations, whereas all were negative for the T790M mutation (Fig. 3), further supporting the idea that the T790M mutation is enriched after acquisition of EGFR-TKI therapy.

Deduction of Tumor Predominance by Plasma cfDNA Profiling

In each cases, the predominance of TKI-resistant tumors was deduced from profiles of EGFR mutations in cfDNA. Seven cases (patients 1–5, 14, and 15 in Fig. 2) acquired EGFR-TKI resistance, associated with the occurrence of the T790M mutation in cfDNAs. In these cases, the proportion of T790M mutant tumor alleles ranged from 7.4% to 97% (Table 1). Thus, the proportion of tumor cells harboring T790M mutations differed among the cases following acquisition of resistance.

cfDNAs obtained both before and after acquisition of EGFR-TKI resistance were analyzed in five of the cases (patients 1, 8, 10, 11, and 12). The fraction of cfDNAs with TKI-sensitive mutations was similar or greater than the fraction before therapy in all of these cases (shown by circle size in Fig. 4A–4C). Therefore, tumor progression during EGFR-TKI therapy often (but not always) results in an increase in the proportion of cfDNA from tumor cells, whereas the T790M mutation appears after acquisition of resistance in a subset of cases (shown by black areas in Fig. 4A, 4B, 4D).

Figure 4.

EGFR-cfDNA profile of representative patients who acquired resistance to EGFR-TKI. (A): Circle graph showing fractions of tumor cell-derived cfDNA, as well as fractions of T790M-mutant cfDNA. (B): Profile of four cases who provided cfDNA both before EGFR-TKI therapy and after acquisition of EGFR-TKI resistance. (−) indicates no positivity for the T790M mutation. (C): Results of NGS analysis of cfDNA of case 12 after expressing resistance to EGFR-TKI therapy, shown in the Integrated Genomics Viewer. L747P mutant alleles were detected in 69% (2,922 of 4,232) of sequencing reads, whereas no T790M reads were observed (0 of 4,274). (D): Profile of three representative cases who provided cfDNA only after resistance acquisition to EGFR-TKI therapy. A question mark indicates no results.

Abbreviation: cfDNA, cell-free plasma DNA.

Patient 12 had the highest fraction (69.3%) of TKI-sensitive mutant DNAs among the TKI-resistant cohort but did not have the T790M mutation (Table 1). Notably, this case had a high fraction (60.6%) of TKI-sensitive mutant DNAs before EGFR-TKI therapy and maintained it throughout the therapy (Fig. 4C). This patient did not respond to gefitinib and was the only case in this cohort with progressed disease (Table 1). To determine the mechanism underlying this innate EGFR-TKI resistance, the patient’s cfDNA was further analyzed by targeted deep sequencing. The results revealed the L747P (c.2239_2240TT→CC) mutation at a high allele frequency (69%), but no in-frame deletion mutation in EGFR exon 19 (Fig, 4C). L747P is a known driver mutation in EGFR that is misdiagnosed as an exon 19 deletion mutation by routine EGFR mutation tests [11].

Discussion

In this study, we evaluated the ability of picoliter-ddPCR-based analysis of plasma cfDNA to detect EGFR mutations conferring resistance to EGFR-TKIs. T790M mutations were detected in 7 (44%) of 16 cfDNA samples of patients with confirmed resistance to EGFR-TKI therapy but in none of 24 cfDNA samples obtained before EGFR-TKI therapy. Thus, our assay results also demonstrated that the T790M mutation is enriched in cfDNA after acquisition of resistance to EGFR-TKI therapy. In 10 patients whose tumor tissues were rebiopsied after confirmation of resistance, the concordance rate of T790M mutation status between cfDNA and rebiopsied tumor tissue was 0.8, and the sensitivity of the assay was 71%. This result was similar to that of a recent study using a conventional ddPCR method [37]. In our study, two cases (patients 6 and 7) exhibited negativity in cfDNA but positivity in tumors; the reason for this discordance is unclear. cfDNA is thought to reflect tumor predominance throughout the body [14, 16, 38]; however, intra- and/or intertumor heterogeneity might have caused low levels of T790M-mutated cfDNA in these two cases; because of the nature of cfDNA as a biomarker, a single rebiopsied tissue examined does not necessarily represent the status of tumors in the body. On the other hand, the negativity of those two cfDNA samples might also have been the result of low sensitivity resulting from the stringent threshold value set for this study. Taken together, the results of this study support the utility of picoliter-ddPCR for cfDNA-based monitoring of T790M mutation in tumors during EGFR-TKI therapy. However, the sensitivity of the assay could be improved by further adjusting the amounts of cfDNA subjected to ddPCR, as well as the threshold values.

The proportion of tumor cells that are positive for the T790M mutation informs the choice of whether to use next-generation TKIs that suppress the activity of EGFR-T790M mutants. Notably, the fractions of T790M-mutant cfDNA among all tumor cfDNA, indicating the predominance of T790M-positive tumor cells, differed among the seven cases that were positive for this mutation (Table 1). Hence, not only positivity but also the fractions of T790M cfDNAs might inform the therapeutic strategy. In patient 14, most tumor cells were deduced to have acquired the T790M mutation (Fig. 4D); therefore, this patient might benefit from next-generation EGFR-TKIs. On the other hand, in some patients, such as patients 3 and 15, T790M-mutant cfDNA constituted only a minor fraction of tumor cfDNA; therefore, the T790M mutation might have occurred in only a subset of tumor cells. Indeed, co-occurrence of more than one resistance mechanism (e.g., HER2 and MET amplification) within a single tumor has been observed [39]. Therefore, these patients might benefit from next-generation EGFR-TKIs; however, detailed monitoring of tumor shrinkage and the proportion of T790M alleles would be necessary to assess the therapeutic effects. Strategies for treating patients with EGFR-TKI resistance using next-generation EGFR-TKIs or other drugs are being actively discussed and tested in clinical trials [40]. Comonitoring of cfDNA for TKI-sensitive and T790M mutations will help to establish criteria for drug selection.

By combining our assay with NGS analysis of cfDNA, we deduced the molecular mechanism of resistance in a patient who did not respond to gefitinib, indicating that their tumor cells were innately resistant to this drug despite the absence of the T790M mutation. NGS analysis of cfDNA revealed the presence of a 2-bp indel (c.2239_2240delinsCC) in exon 19, causing the L747P amino acid substitution, but the absence of T790M and all other known TKI-sensitive EGFR mutations in cfDNA. Our picoliter-ddPCR analysis of cfDNA and the Scorpion-ARMs test of tumor tissues prior to gefitinib therapy misdiagnosed this case as being positive for exon 19 deletion mutation because of mispriming of the oligonucleotides used for PCR [11]. The EGFR-L747P mutation is a rare driver mutation conferring innate resistance of tumor cells to conventional EGFR-TKIs [41]. These findings suggest that NGS would be useful in detecting diverse mutations by complementing the ddPCR assay focused on TKI-sensitive and T790M EGFR mutations. Resistance to EGFR-TKIs is caused not only by EGFR-T790M mutations but also by other alterations in EGFR and other genes, for example, amplifications of MET and HER2 [39]. Thus, establishment of a comprehensive cfDNA analysis method that enables detection not only of TKI-sensitive and T790M EGFR mutations but also other genetic alterations that confer resistance is necessary for noninvasive diagnosis of EGFR-TKI resistance in the lung cancer clinic. Picoliter-ddPCR focusing on hot-spot mutations, complemented by NGS analysis, represents one way to perform such an analysis.

TKI-sensitive mutations were detected in most (15 of 16, 94% in Fig. 3) cfDNA samples obtained after confirming resistance. Notably, cfDNA obtained from patients who developed new extrapleural tumors upon disease progression following EGFR-TKI therapy exhibited high (>10%) fractions of cfDNA with TKI-sensitive mutations (p = .00070 by Fisher’s exact test; supplemental online Table 3). This finding confirms the utility of cfDNA in deducing the tumor burden in progressed cases, as suggested by previous studies [17, 22]. On the other hand, only a subset of cfDNA samples obtained before EGFR-TKI therapy were positive for TKI-sensitive mutation (9/24; 38% in Fig. 3), despite the extrapleural growth of the tumors (supplemental online Table 4). The reason for this difference is unknown, but one possible explanation is that epithelial-to-mesenchymal transition of tumor cells, which often occurs contemporaneously with acquisition of resistance to EGFR-TKIs, might increase the amount of plasma cfDNA by promoting the dissemination of tumor cells into plasma [42, 43].

This study has a few limitations. First, the sample size was small, particularly for cfDNA samples with corresponding rebiopsied tumor tissues. Hence, the utility, feasibility, and robustness of this picoliter-ddPCR assay should be further examined in a clinical trial that assesses subsequent response to treatments by conventional and third-generation TKIs by prospectively including a large set of samples. To this end, we have initiated a large-scale prospective study to validate the concordance of T790M predominance between cfDNA and rebiopsied tumor tissues. Second, this study focused on determining the predominance of the T790M mutation in cfDNA from advanced patients to monitor the burden of T790M-positive tumor cells during therapy. Because of the stringent criteria used here to judge positivity, the sensitivity of the assay was lower than that of other assays that focus on diagnosis of early-stage tumors [24, 44]. Increasing the amount of cfDNA used in the assay and/or setting more appropriate thresholds according to the amount of cfDNA should make our assay more sensitive.

See http://www.TheOncologist.com for supplemental material available online.

Author Contributions

Conception/Design: Yoshitaka Seki, Yutaka Fujiwara

Provision of study material or patients: Kuniko Sunami, Yasushi Goto, Hidehito Horinouchi, Shintaro Kanda, Hiroshi Nokihara, Shun-ichi Watanabe, Noboru Yamamoto

Collection and/or assembly of data: Yoshitaka Seki, Erina Takai

Data analysis and interpretation: Yoshitaka Seki, Takashi Kohno, Erina Takai, Hitoshi Ichikawa

Manuscript writing: Yoshitaka Seki, Yutaka Fujiwara, Takashi Kohno

Final approval of manuscript: Yoshitaka Seki, Yutaka Fujiwara, Takashi Kohno, Erina Takai, Kuniko Sunami, Yasushi Goto, Hidehito Horinouchi, Shintaro Kanda, Hiroshi Nokihara, Shun-ichi Watanabe, Hitoshi Ichikawa, Noboru Yamamoto, Kazuyoshi Kuwano, Yuichiro Ohe

Disclosures

Yutaka Fujiwara: AstraZeneca, Eli Lilly, Novartis Pharma (RF); Yasushi Goto: Eli Lilly, Boehringer Ingelheim (C/A, H), AstraZeneca (H); Hidehito Horinouchi: Taiho Pharmaceutical, MSD, Merck Serono (RF); Shintaro Kanda: AstraZeneca (RF); Hiroshi Nokihara: Merck, Pfizer, Taiho, Eisai, Chuga, Eli Lilly, Novartis, Glaxo, Yakult, Quintiles, Astellas, AstraZeneca, Ono (RF), Sanofi, Eli Lilly, Boehringer Ingelheim, AstraZeneca, Ono (H); Noboru Yamamoto: Chugai, Takeda, Quintiles, Bristol-Myers Squibb, Astellas, Taiho (RF), Eli Lilly (H). The other authors indicated no financial relationships.

(C/A) Consulting/advisory relationship; (RF) Research funding; (E) Employment; (ET) Expert testimony; (H) Honoraria received; (OI) Ownership interests; (IP) Intellectual property rights/inventor/patent holder; (SAB) Scientific advisory board