Efficacy of different platforms in detecting EGFR mutations using cerebrospinal fluid cell‐free DNA from non‐small‐cell lung cancer patients with leptomeningeal metastases

Abstract Background Cell‐free tumor DNA (ctDNA) obtained through liquid biopsy is useful for the molecular analysis of advanced non‐small‐cell lung cancer (NSCLC). Few studies have directly compared analysis platforms in terms of their diagnostic performance in analyzing ctDNA obtained from the cerebrospinal fluid (CSF) of patients with leptomeningeal metastasis (LM). Methods We prospectively analyzed patients with epidermal growth factor receptor (EGFR)‐mutant NSCLC who were subjected to CSF analysis for suspected LM. To detect EGFR mutations, CSF ctDNA was analyzed using the cobas EGFR Mutation Test and droplet digital polymerase chain reaction (ddPCR). CSF samples from osimertinib‐refractory patients with LM were also subjected to next‐generation sequencing (NGS). Results Significantly higher rates of valid results (95.1% vs. 78%, respectively, p = 0.04) and EGFR common mutation detection (94.3% vs. 77.1%, respectively, p = 0.047) were obtained through ddPCR than through the cobas EGFR Mutation Test. The sensitivities of ddPCR and cobas were 94.3% and 75.6%, respectively. The concordance rate for EGFR mutation detection through ddPCR and the cobas EGFR Mutation Test was 75.6% and that for EGFR mutation detection in CSF and plasma ctDNA was 28.1%. In osimertinib‐resistant CSF samples, all original EGFR mutations were detected through NGS. MET amplification and CCDC6‐RET fusion were demonstrated in one patient each (9.1%). Conclusions The cobas EGFR Mutation Test, ddPCR, and NGS appear to be feasible methods for analyzing CSF ctDNA in patients with NSCLC and LM. In addition, NGS may provide comprehensive information regarding the mechanisms underlying osimertinib resistance.


INTRODUCTION
In recent years, the treatment paradigm for advanced nonsmall-cell lung cancer (NSCLC) has markedly shifted toward precision medicine because of the discovery of driver oncogenes and targeted therapies. 1,2 Epidermal growth factor receptor (EGFR) mutations are the most common oncogenic driver mutations observed in Asian patients with NSCLC. 3 EGFR tyrosine kinase inhibitors (TKIs) are superior to cytotoxic chemotherapy in terms of efficacy and have become the standard treatment option for patients with EGFR-mutant NSCLC. [4][5][6] The identification of driver oncogenes from tumor specimens has become a routine task in clinical practice. However, obtaining a sufficient amount of tumor tissue for molecular profiling may be difficult. Liquid (plasma) biopsy is a practical alternative in the case of absent or insufficient tissue samples and can be performed concurrently or sequentially with tumor genotyping in patients with advanced NSCLC. 7,8 Patients with EGFR-mutant NSCLC have a higher risk of central nervous system (CNS) involvement during the treatment course than do patients with EGFR wild-type NSCLC. 9 Among the neurological complications of NSCLC, leptomeningeal metastasis (LM) is the most devastating form of CNS metastasis, with a median overall survival (OS) <1 year. 10 The incidence of LM in patients with EGFR-mutant NSCLC was reported to be 9.4%. 11 However, obtaining tumor specimens from intracranial lesions is challenging, particularly in patients with LM, because biopsy of the leptomeninges is not clinically feasible. Therefore, the use of cerebrospinal (CSF) for liquid biopsy to access circulating tumor cells or cell-free tumor DNA (ctDNA) is a new approach for genotyping intracranial lesions in patients with NSCLC and LM. 12 Studies have indicated the clinical usefulness of detecting sensitive and resistant EGFR mutations in CSF ctDNA. 13,14 The rates of mutation detection in patients with EGFRmutant NSCLC vary across sequencing methods. Droplet digital polymerase chain reaction (ddPCR), which combines microfluidic technology with quantitative PCR, is an ultrasensitive assay for measuring absolute mutation alleles that has been used to detect low-abundance nucleic acids in clinical practice. 15 For the detection of plasma EGFR mutations, the sensitivity of ddPCR is higher than that of the amplification refractory mutation system (ARMS) PCR. 16,17 However, studies directly comparing the mutation detection rate of different sequencing platforms using CSF ctDNA are scarce. Therefore, in this study, we compared multiple PCR methods in terms of their diagnostic performance in analyzing CSF ctDNA and explored their clinical relevance.

Study design and patients
We prospectively analyzed patients with EGFR-mutant NSCLC who were subjected to CSF analysis for suspected LM between August 2020 and November 2021. Data regarding the patients' clinical characteristics were collected, including age, sex, Eastern Cooperative Oncology Group (ECOG) performance status, primary or metastatic tumor's EGFR mutation status, history of systemic treatment before and after CSF sampling, brain magnetic resonance imaging (MRI) findings, CSF analysis reports, and survival status. The European Association of Neuro-Oncology (EANO)-European Society for Medical Oncology (ESMO) guideline was used to diagnose LM in this study. 18 In brief, positive CSF cytology indicated "confirmed LM," classical neurological image findings (i.e., MRI), typical clinical signs indicated "probable LM," and only typical neurological symptoms indicated "possible LM." This study was approved by the Institutional Review Board of Taipei Veterans General Hospital, Taiwan (approval number: 2020-07-008C) and conducted in accordance with the ethical principles of Declaration of Helsinki.

Sample preparation and ctDNA isolation
Lumbar puncture was performed to diagnose LM and CSF was collected. CSF collection was also collected through a ventriculoperitoneal shunt. Then 5-8 mL of CSF was reserved for cytological assessments, and the residual sample was sent for cell count and biochemistry analyses. The CSF sample was centrifuged at 400g for 5 min at 4 C, and the pellet was collected and sent for cytological assessments, which were performed by experienced cytologists. The supernatant was recentrifuged at 2000g for 10 min at 4 C, and the secondary supernatant was stored at À80 C until ctDNA extraction. CSF ctDNA was extracted using the cobas DNA Sample Preparation Kit (Roche Molecular Systems) according to the manufacturer's instructions.
To obtain plasma ctDNA, venipuncture was performed to collect peripheral blood samples (10 mL), which were kept in K2 EDTA tubes at the time of CSF sampling and processed within 1 h, when feasible. The tubes were centrifuged twice at 2000g for 10 min at 4 C, and ctDNA was extracted from 2 mL of plasma sample using the QIAamp Circulating Nucleic Acid Kit (Qiagen) according to the manufacturer's instructions.

EGFR mutation analysis using the cobas EGFR Mutation Test
The ctDNA extracted from CSF and plasma samples was analyzed using the cobas EGFR Mutation Test (version 2; Roche Diagnostics), which is a mutant allele-specific realtime PCR. This test was performed to detect 42 types of EGFR mutations in exons 18-21. The results were interpreted according to the manufacturer's instructions. "Mutation detected" or "no mutation" indicated the detection of at least one mutation or no mutation in the targeted EGFR region, respectively. "Invalid" signified that the extracted DNA was of insufficient quantity or quality to repeat the amplification and detection steps in each sample.
EGFR mutation analysis using ddPCR Each PCR sample was partitioned into 20 000 droplets and analyzed using a QX200 Droplet Digital PCR system (Bio-Rad Laboratories). Following amplification, each droplet was scored as positive or negative after the detection of the fluorescence signal emitted by the target sequence. Poisson statistical analysis was performed for the absolute quantification of the target sequence. To analyze EGFR mutations, in addition to a reference assay, TaqMan EGFR T790M mutation assay, EGFR Exon 19 deletions assay, EGFR L858R mutation assay, and EGFR C797S mutation assay (Invitrogen Life Technologies) were performed. In brief, 20 μL of a ddPCR reaction mix containing 2X Master Mix (Bio-Rad Laboratories), 20X primer, and TaqMan Probe mix (Applied Biosystems Life Technologies) in addition to the DNA template was prepared and subjected to emulsification using the QX200 droplet generator. The emulsified samples were then transferred into 96-well plates for PCR, and the PCR products were loaded into a droplet reader (QX200 Droplet Digital System; Bio Rad Laboratories) for analysis of mutations through QuantaSoft (version 1.7.4.0917; Bio-Rad Laboratories) The result was interpreted as positive if the mutant copy number was ≥3 in the ddPCR assays. Fractional abundance (FA), which indicates the abundance of mutant DNA alleles in a wild-type background, was also evaluated. 19

Next-generation sequencing
In osimertinib-resistant patients with LM, the extracted ctDNA from CSF was sent for Oncomine Lung cfDNA Assay if residual samples were available. The MagMAX Cell-Free DNA Isolation Kit was used for ctDNA isolation. We applied the Ion Chef System and Ion S5 XL sequencing system for template preparation and next-generation sequencing (NGS), respectively. Eleven genes (ALK, BRAF, EGFR, ERBB2, KRAS, MAP2K1, MET, NRAS, PIK3CA, ROS1, and TP53) with ≥150 hotspots were analyzed. Ion Reporter was used for postsequencing analysis.

Statistical analysis
Patient characteristics were summarized using descriptive statistics. Categorical variables are presented as numeric and percentage values, whereas continuous variables are presented as median and range values. Chi-square and Fisher's exact test were used to analyze the correlations between detectable mutations in CSF ctDNA and the patients' demographic characteristics. The sensitivity, specificity, positive predictive value, negative predictive value (NPV), and accuracy of each platform was measured according to the relevant formulas. The Kaplan-Meier method was used to plot survival curves, and the log-rank test was used to perform between-group comparisons. p < 0.05 was considered statistically significant. OS was calculated as the interval between the date of CSF sampling and that of death or most recent follow-up. All analyses were performed using SPSS for Windows (version 22.0; IBM Corporation).

Patient characteristics
This study enrolled 40 patients; 30 patients and seven patients were diagnosed to have confirmed LM or probable LM, respectively. The median age of patients with LM was 64.5 years (range 37.3-99.4 years). Most were women (67.6%), never smoked (75.7%) and had an ECOG performance score of 0-1 (51.4%). Exon 21 L858R mutation (54.1%) was found to be the most common EGFR mutation in our study, followed by exon 19 deletion (32.4%). Fourteen patients (37.8%) received whole-brain radiotherapy (WBRT) before CSF analysis. Of all the patients, >90% had received at least one targeted therapy before CSF analysis; osimertinib was the most commonly used EGFR-TKI (32.4%). Regarding the brain MRI images, the most common finding was parenchymal metastasis (81.1%), followed by leptomeningeal enhancement (78.4%) and ventricular dilation (64.9%). The median duration from initial NSCLC diagnosis to CSF sampling was 30.9 months (range 0-103.2 months). The most common method used for CSF collection was lumbar puncture (78.4%); CSF sampling was performed more than once for three patients. The patient characteristics are summarized in Table 1.

Diagnostic performance of the cobas EGFR Mutation Test and ddPCR
Supporting Information Table S1 summarizes the analysis results of the 41 CSF samples obtained from patients with confirmed or probable LM (n = 37). The median white blood cell (WBC) count in the CSF was 3/cumm, and the median level of protein was 69.6 mg/dL. A significantly higher rate of valid results was obtained through ddPCR than through the cobas EGFR Mutation Test (95.1% vs. 78%, respectively, p = 0.04; Supporting Information Table S2).
The detection rate of EGFR mutation in the CSF ctDNA analyzed using the cobas EGFR Mutation Test was 75.6% (31/41). This value was 81.1% (30/37) in patients with abnormal cytology results. However, using ddPCR, the mutation detection rates were 82.9% (34/41) and 86.5% (32/37) in the overall and cytological abnormal samples, respectively ( Figure 1). For patients carrying common EGFR mutations (exon 19 deletion and exon 21 L585R mutation) with LM, the rate of mutation detection was significantly higher with ddPCR than with the cobas EGFR Mutation Test (94.3% vs. 77.1%, respectively, p = 0.047). The diagnostic performance of each platform is presented in Table 2.
Overall, the concordance rate for EGFR mutations detected using the cobas EGFR Mutation Test and ddPCR was 75.6% (31/41; Table 3), rising to 82.9% (29/35) if patients with uncommon mutations were excluded (Supporting Information Table S3). Six patients returned invalid results in cobas but were revealed to have detectable EGFR mutations in ddPCR testing. The clinical characteristics of these six patients are listed in Supporting Information Table S4. Five patients (13.5%, 5/37) had EGFR T790M mutation, which were detected through both the cobas EGFR Mutation Test and ddPCR. The mutation profiles of all enrolled patients are illustrated in Supporting Information Figure S1.
The median FA of EGFR mutations detected through ddPCR was 53.7% (range 0.15-94%); this result did not correlate with the WBC count or protein level in the CSF. Similarly, no correlation was observed between the presence of adenocarcinoma cells in the CSF and high FA (p = 0.253).

Comparison of EGFR mutations in CSF and plasma ctDNA
Thirty-two patients received plasma ctDNA analysis at the time of CSF sampling, and most had no detectable EGFR mutation in plasma (65.6%, 21/32). Among the patients with detectable EGFR mutations in plasma ctDNA, 21.9% of cases (7/32) had L858R and 12.5% (4/32) had exon 19 deletion. We detected T790M mutation in only three cases (9.4%). The concordance rate for EGFR mutations detected in CSF ctDNA using ddPCR and plasma ctDNA was 28.1% (9/32; Table 4).

ctDNA profiling of Osimertinib-resistant CSF samples
The CSF samples of 11 patients who developed LM progression after osimertinib use were sent for cobas EGFR  Mutation Test, ddPCR, and NGS testing using the Oncomine Lung cfDNA Assay ( Figure 2). All patients had EGFR mutations detectable using the cobas EGFR Mutation Test and NGS. One patient (9%) exhibited EGFR L747S mutation. Following EGFR mutations, TP53 mutation was the second most common co-mutation detected through NGS (45.5%, 5/11). MET amplification and CCDC6-RET fusion, which were reported to be associated with osimertinib resistance, were each observed in one patient. The treatment history and serial genetic profiling results of patient VGH034 are presented in Supporting Information Figure S2. This patient also had systemic progression at the time of CSF sampling. CCDC6-RET fusion was also detected through plasma ctDNA analysis using FoundationOne LiquidCDx.

OS after LM
In this study, 32 (80%) patients received osimertinib after CSF sampling and 12 (30%) patients received WBRT after LM. The median OS was 7.5 months (95% confidence interval 3.54-11.46). The OS was numerically higher in patients receiving osimertinib than in those not receiving it (9.0 vs. 3.5 months, respectively, p = 0.253). In the patients receiving osimertinib after LM, the presence of a T790M mutation did not affect the OS after osimertinib use. Patients with a higher FA of EGFR mutation (>65%) had inferior OS after osimertinib use compared with those with lower FA, but the difference is not statistically significant (12 vs. 5.4 months, respectively, p = 0.076; Figure 3).

DISCUSSION
In patients with NSCLC and LM, molecular profiling is crucial for diagnosis and treatment decisions. Reports comparing the different sequencing platforms used for CSF ctDNA are limited. In this study, we demonstrated that CSF ctDNA analysis using the cobas EGFR Mutation Test, ddPCR, and NGS was feasible. Compared with the cobas EGFR Mutation Test, ddPCR had a significantly higher rate of valid results and a higher rate of EGFR mutation detection. In addition to EGFR sensitizing mutation, the genetic alteration of resistance could be detected through NGS following targeted therapy.
In this study, the rates of EGFR mutations detected in CSF ctDNA through the cobas EGFR Mutation Test were 75.6% and 81.1% in the overall patient samples and abnormal cytology samples, respectively. This finding is consistent with that in our previous report. 13 The rate of EGFR mutation detected in CSF ctDNA through ddPCR was 82.9%, which was similar to corresponding results reported in other studies conducted using the same technique. 20,21 Using primary tumor testing results as a reference, Xu et al. reported similar sensitivity for ddPCR and the ARMS in relation to CSF ctDNA analysis (95% and 89.2%, respectively). 22 The sensitivity of ddPCR in our study was 94.3%, which was higher than that of the cobas EGFR Mutation Test and cytology testing. Notably, the EGFR mutation detection rate in plasma in our study was only 28.1%. These findings further support the use of CSF ctDNA alongside cytological assessments for the diagnosis and treatment of patients with NSCLC and LM.
Acquired resistance is inevitable after the application of targeted therapy. The most common mechanism underlying resistance to first-and second-generation EGFR-TKIs is an acquired T790M mutation in EGFR, which has an incidence rate of 52.8-63%. 15,23 Despite the use of different platforms, the percentage of T790M mutation is lower in CSF ctDNA analysis. 13,14,20 Osimertinib, a third-generation EGFR-TKI T A B L E 3 Comparison of EGFR mutation status from CSF ctDNA using the cobas EGFR Mutation Test and ddPCR Test   ddPCR  E19D  E19D/T790M  L858R  L858R/T790M  Uncommon mutation a  Invalid  No mutation  Total   E19D  9  0  0  0  0  2  0  11   E19D/T790M  0  1  0  0  0  0  0  1   L858R  0  0  14  0  0  4  F I G U R E 2 Next-generation sequencing analysis of cerebrospinal fluid cell-free tumor DNA extracted from patients receiving osimertinib. ALK, anaplastic lymphoma kinase; ddPCR, droplet digital polymerase chain reaction; E19D, exon 19 deletion; EGFR, epidermal growth factor receptor; NA, not available that overcomes the resistance conferred by the T790M mutation, has a higher CNS penetration rate than do other EGFR-TKIs and has exhibited promising intracranial activity in prospective studies. 10,24 In addition, osimertinib has demonstrated efficacy in patients with EGFR-mutant LM. 25,26 In our study, 30% of the patients underwent CSF sampling after osimertinib treatment. The percentages of T790M mutation in the CSF ctDNA of all patients and those treated with osimertinib were 13.5% and 8.3%, respectively. The EGFR T790M mutation was detected in the patients' CSF samples by using ddPCR and the cobas EGFR Mutation Test. The presence of a T790M mutation did not affect the survival outcome after LM in patients receiving osimertinib. Because the number of patients harboring EGFR T790M mutation in their CSF was small in our cohort, further studies are required to compare the efficacy of different sequencing methods in detecting EGFR T790M mutation and to evaluate its prognostic significance. Circulating ctDNA is associated with tumor volume, and a high-variant allele frequency of ctDNA may lead to inferior survival in patients with metastatic malignancies. 27,28 However, few studies have focused on the association between the prognosis of patients with LM and the percentage of mutant DNA alleles in their CSF. In our study, patients with a higher FA of EGFR mutations were likely to have inferior outcomes after osimertinib use. However, the difference was nonsignificant. Further studies are warranted to validate the prognostic value of ctDNA level in patients' CSF.

Cobas EGFR Mutation
The resistance mechanisms of osimertinib, which can be divided into EGFR-dependent and EGFR-independent, are heterogeneous and complex. 29,30 Eleven patients in our cohort underwent CSF ctDNA analysis through NGS. The detection of EGFR mutation was 100%, and no patient had detectable EGFR T790M mutation in CSF following osimertinib treatment. Using targeted sequencing, Zheng et al. observed that 21.7% of osimertinib-refractory patients maintained their T790M mutation. 31 EGFR C797X mutation is a major resistance mutation that emerges following osimertinib treatment that was also observed during CSF ctDNA analysis in a previous study. 32 However, no patient had detectable C797X mutation in our study. One patient exhibited MET amplification and another exhibited RET fusion following osimertinib treatment. MET copy number gains are a relatively common resistance mechanism in patients with NSCLC and LM. 14,33 Furthermore, oncogenic fusions were reported to be a bypass pathway of osimertinib resistance, with an incidence range from 1% to 10%. 30 CCDC6-RET fusion was detected in both the CSF and plasma ctDNA of one patient with progressive LM in our study. It is rare to observe CCDC6-RET fusion in CSF in the context of resistance to osimertinib. However, due to the fact that the patient also exhibited systemic progression, it is challenging to ascertain whether this oncogenic fusion is an intracranial resistance mechanism to osimertinib. Combined EGFR and RET inhibition was reported to be an effective treatment strategy for such patients 34 ; however, our patient developed rapid tumor progression and died before novel targeted treatment could be administered. Together, NGS of CSF ctDNA may provide more information for patients who have progressive LM after CNS-penetrant TKI therapy. More studies are required to elucidate the resistance mechanism of intracranial lesions following targeted therapy.
This study had some limitations. First, no other genetic alterations were investigated using the CSF and plasma samples, hence we could not construct the comprehensive mutation profiles of the patients with EGFR-mutant NSCLC and LM. NGS of CSF ctDNA is a feasible method for detecting resistance mutation after new-generation targeted therapy. 14,35 However, to the best of our knowledge, most countries lack a commercially available sequencing platform. Second, the mechanisms underlying osimertinib resistance could not be elucidated because of the small number of samples. Nevertheless, our study provided meaningful results and can be used to establish a reasonable strategy for EGFR mutation analysis of CSF in populations with a relatively high incidence of EGFR-mutant NSCLC.
In conclusion, our study demonstrated that the cobas EGFR Mutation Test, ddPCR, and NGS are all feasible methods for CSF ctDNA analysis in patients with NSCLC and LM. Our results revealed that the sensitivity of ddPCR is higher than that of the cobas EGFR Mutation Test. NGS may provide more comprehensive information on resistance mechanisms following osimertinib treatment.
F I G U R E 3 Overall survival in patients receiving osimertinib after leptomeningeal metastasis stratified by the presence of a T790M mutation (a) in the CSF and the FA of EGFR mutations (b). CI, confidence interval; CSF, cerebrospinal fluid; FA, fractional abundance