Utility of cell-free DNA from bronchial washing fluid in diagnosis and genomic determination for radiology-suspected pulmonary nodules

Abstract

Background

Bronchial washing fluid (BWF) is a less-invasive specimen. Due to the limited sensitivity of BWF cellular component diagnosis, the aim of this study was to explore the potential role of BWF supernatant as a source of liquid biopsy of lung cancer.

Methods

This prospective study enrolled 76 suspected and 5 progressed lung cancer patients. Transbronchial biopsy tissues, BWF supernatant (BWF_Sup) and BWF precipitant (BWF_Pre) were tested by a targeted panel of 1021 genes.

Results

BWF_Sup cell-free DNA (cfDNA) was superior to tissue biopsy and BWF_Pre in determining mutational allele frequency, tumour mutational burden, and chromosomal instability. Moreover, BWF_Sup and BWF_Pre achieved comparable efficacy to tissue samples in differentiating malignant and benign patients, but only BWF_Sup persisted differentiated performance after excluding 55 malignancies pathologically diagnosed by bronchoscopic biopsy. Among 67 malignant patients, 82.1% and 71.6% of tumour-derived mutations (TDMs) were detected in BWF_Sup and BWF_Pre, respectively, and the detectability of TDMs in BWF_Sup was independent of the cytological examination of BWF. BWF_Sup outperformed BWF_Pre in providing more subclonal information and thus might yield advantage in tracking drug-resistant markers.

Conclusions

BWF_Sup cfDNA is a reliable medium for lung cancer diagnosis and genomic profiles and may provide important information for subsequent therapeutic regimens.

Introduction

Genotype-directed cancer therapy has revolutionised the management of non-small cell lung cancer (NSCLC), especially for advanced-stage diseases [1, 2], and achieved prominent improvement in patient survival and living quality compared with conventional systemic treatment [3, 4]. Tumour tissue samples play a fundamental role during molecular diagnose. Unfortunately, clinicians often have difficulty obtaining adequate tumour tissue samples to meet the increasing demands for identifying the growing number of molecular biomarkers. Up to 30–50% of NSCLC patient biopsies lack adequate tumour cell amount for standard biomarker analyses [5]. In addition, molecular testing results of tissues may be delayed due to the time required for the necessary processing and diagnostic steps prior to genomic analysis. The aforementioned reasons highlight the need for an alternative and rapid analytical source for cancer genotyping.

Circulating cell-free DNA (cfDNA) consists of free-floating nucleic acid fragments shed into the blood and other body fluid that primarily originate from cellular breakdown or active release [6]. Circulating tumour DNA (ctDNA) refers to the somatic DNA of tumour cells that is shed or released into the circulatory system after apoptosis. It is cfDNA derived from tumour cells and belongs to a type of cfDNA [7]. Studies have demonstrated that ctDNA can serve as a noninvasive molecular proxy for determining the tumour biology of solid malignancies [8]. However, blood-based ctDNA detection is usually challenging due to the extremely low detection rate, especially in early-stage cancers [9]. Averagely, the sensitivity is about 66% using PCR-based assays and 79% based on the next-generation sequencing (NGS) platform [10–12]. To expand the substituted application, other clinically accessible liquid specimens, such as urine, cerebrospinal fluid and pleural effusion, have been used to assess ctDNA determination, revealing great concordance with tumour tissue genetic information [13–16].

Bronchial washing fluid (BWF) is collected during the bronchoscopic procedure, and may contain cfDNA [17, 18]. It has been demonstrated that the ctDNA of BWF can be used to detect EGFR mutations with high sensitivity and consistency compared to tumour tissue [19]. Compared with tissue samples and blood-based liquid biopsy, the utility of cfDNA detection in BWF has several obvious advantages. First, due to direct contact with tumour lesions, it is expected that more ctDNA will be shed than in peripheral blood. Second, the collection of BWF is less invasive than tissue biopsies. Third, ctDNA can reflect a thorough molecular profile of tumour biology, whereas tissue biopsies may be hampered by paraffin section sampling bias [20, 21].

Although small case series have shown advantages of BWF ctDNA [18, 19], a comprehensive comparison with tumour biopsy has not been prospectively evaluated in larger-scale patient cohorts. Therefore, this study enrolled 83 subjects with suspected or definite lung cancer, from which matched BWF and tissue biopsy samples were obtained. Specially, supernatant and cellular precipitant were separated from BWF, and all specimens were performed with panel-based NGS of deep depth.

Materials and methods

Patients

Between January 2018 and August 2018, patients with radiology-suspected pulmonary nodules or progressed NSCLC after tyrosine kinase inhibitor (TKI) treatment were prospectively enrolled in this study. All patients signed written informed consent before any study-related procedures were performed. This study was registered and approved by the Zhongshan Hospital ethics committee (No. B2018-027).

Sample collection and processing

Transbronchial biopsies were performed under the guidance of endobronchial ultrasound with guide sheath (EBUS-GS). Forceps biopsy samples were submitted for pathologic diagnosis after formalin-fixation and paraffin-embedding (FFPE) procedures. Extra FFPE slides (ten slides, 5-μm thickness) were collected and stored at room temperature for genomic analysis, regardless of the pathologic diagnosis results. After forceps biopsies, the guide sheath remained and was fixed in the tumour lesion, then the fixed sheath was utilised to guide normal saline (20–40 ml) directly to the lesion where biopsies were taken. However, the traditional BWF collection method utilised the direct instillation of saline through a bronchoscope wedged into the segment. The recovery fluid had a minimum volume of 6 ml. After centrifugation at 600 × g for 10 min, the BWF supernatant and cellular precipitant were separately preserved for subsequent molecular or cytologic analysis.

A volume of 10 ml peripheral blood was collected on the same day of the bronchoscopy procedure and was processed within 2 h. After centrifugation at 1600 × g for 10 min, the plasma was separated and removed, and peripheral blood lymphocytes (PBLs) were collected for the extraction of germline genomic DNA [22, 23]. At least 2 ml of BWF supernatant (BWF_Sup), BWF precipitant (BWF_Pre) and PBLs were immediately transported to the laboratory for DNA extraction and genomic profiling, regardless of the pathologic diagnosis.

DNA extraction

PBL DNA and BWF_Pre DNA were extracted using the DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany), whereas cfDNA was isolated from 1.5–3.0 mL BWF_Sup using the QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany). Genomic DNA was extracted from FFPE samples using the Maxwell^® RSC DNA FFPE Kit (Promega, Madison, WI, USA). A Qubit fluorometer and the Qubit dsDNA HS (High Sensitivity) Assay kit (Invitrogen, Carlsbad, CA, USA) were used to determine the DNA concentration, and the fragment size distribution of cfDNA was assessed using an Agilent 2100 BioAnalyzer and the DNA HS kit (Agilent Technologies, Santa Clara, CA, USA).

Target capture and next-generation sequencing

Before library construction, 1 µg each of genomic DNA extracted from PBL, BWF_Pre and FFPE samples was sheared to 200–250-bp fragments, using a Covaris S2 ultrasonicator (Covaris, Woburn, MA, USA). A volume of 80 ng DNA from BWF_Sup was used for library construction. Indexed Illumina NGS libraries were constructed from PBL DNA, BWF_Pre DNA, FFPE DNA, and BWF_Sup DNA using the KAPA Library Preparation Kit (Kapa Biosystems, Wilmington, MA, USA). Captured probes were constructed for the entire regions of the most common driver genes across 12 solid tumour types to enhance detection sensitivity [24]. Next, genomic regions related to the effects of chemotherapy, targeted drugs, and immunotherapy per available clinical and preclinical research were added. Finally, frequently mutated regions recorded in the Catalogue of Somatic Mutations in Cancer (http://cancer.sanger.ac.uk/cosmic) and TCGA (https://cancergenome.nih.gov/) were included. In total, the probes covered 1.6 Mbp of the genome. All genes and coordinates of selected regions were provided in Supplementary Table S1. Capture hybridisation was carried out according to the manufacturer’s protocol. After hybrid selection, the captured DNA fragments were amplified and pooled to generate several multiplex libraries. DNA sequencing was performed with Illumina 2 × 75 bp paired-end reads, using an Illumina HiSeq 3000 instrument (Illumina, San Diego, CA, USA). The TruSeq PE Cluster Generation Kit V3 and the TruSeq SBS Kit V3 (Illumina, San Diego, CA, USA) were employed, according to the manufacturer’s recommendations. The preset amount of data was 1, 2, 2, and 2 Gb for PBL DNA, FFPE DNA, BWF_Pre DNA, and BWF_Sup DNA, respectively.

Sequencing data analysis

After removing terminal adaptor sequences and low-quality data, the remaining reads were mapped to the reference human genome (hg19), using the Burrows–Wheel Aligner (0.7.12-r1039) (19451168), with default parameters. MuTect2 (3.4–46-gbc02625) [25] was used to call somatic small insertions and deletions (Indels) and single nucleotide variants (SNVs). Indels and SNVs, at an allele frequency of ≥1%, were profiled in FFPE and BWF_Pre samples. CONTRA (v2.0.8) was performed to identify somatic copy-number variations (CNVs) [26]. To detect gene fusions and structure variations (SV), baits were designed to capture selected exons and introns for the ALK, ROS1, RET and NTRK1 oncogenes, according to previously reported SVs. An in-house algorithm was employed to utilise split-reads and discordant read-pairs to identify SVs. All final candidate variants were manually confirmed with the integrative genomics viewer browser. The average depth was 1420 ± 402.4× for FFPE samples, 1140 ± 341.4× for BWF_Pre samples, and 1200 ± 1397.3× for BWF_Sup samples. Mistakes introduced by PCR or sequencing procedures were corrected according to clustered reads. Local realignment and base quality recalibration were performed using the Gene Analysis Toolkit (https://www.broadinstitute.org/gatk/).

Determination of tumour heterogeneity and clonal population structure

Each tumour’s mutant-allele tumour heterogeneity (MATH) value was calculated from the median absolute deviation (MAD) and the median of its mutant AFs: MATH = 148.26 × MAD/median [27]. The key implication of MATH value was to reflect the fluctuation range of AFs in the same sample. PyClone [28] was used to infer clonal population structures of specific cancer patients via the joint analysis of FFPE, BWF_Sup and BWF_Pre samples. Generally speaking, PyClone is a Bayesian clustering method for grouping sets of deeply sequenced somatic mutations into putative clonal clusters while estimating their cellular prevalence and accounting for allelic imbalances introduced by segmental copy-number changes and normal-cell contamination.

Statistical analysis

Nonsynonymous SNVs and indels with ≥3% allele frequencies were included in the calculation of tumour mutational burden (TMB). The percentage of positive agreement was defined as the number of positively consistent cases out of the total number of evaluations. IBM SPSS software 23 (IBM, Armonk, NY, USA) and GraphPad Prism 7 (GraphPad Software, La Jolla, CA, USA) were used for statistical analyses. Continuous variables are expressed as the mean (±standard deviation [29]) and were compared using Mann–Whitney U test (unpaired) and Wilcoxon signed-rank test (paired). Categorical variables were compared using the χ² or Fisher’s exact test. Spearman’s coefficient was used to evaluate the correlation between two variables. The receiver operating characteristic (ROC) curve was used to identify the performance of mutational burden of different specimens in the diagnosis of malignancies. All tests were two-sided, and P < 0.05 was considered to be statistically significant.

Results

Patient characteristics and study flow

A total of 81 patients (76 with suspected peripheral lung cancer and 5 with progressed NSCLC) were prospectively enrolled, whose clinical characteristics are listed in Table 1 and detailed in Supplementary Table S2. The workflow of the study is presented in Fig. 1. A total of 55 patients with lung cancer were pathologically diagnosed by bronchoscopic biopsy. Among the 26 patients initially diagnosed with non-malignant lesion, 12 patients were subsequently diagnosed with lung cancer (5 by surgical resection, 3 by CT-guided lung biopsy, 1 by pleural biopsy under video-assisted thoracic surgery, 1 by a second bronchoscopy and 2 by clinical diagnoses) and 14 were finally diagnosed with benign diseases (pulmonary infection, granulomatous lesion, or tuberculosis). Totally, 67 patients were diagnosed with lung cancer (62 were TKI-naive and 5 progressed after EGFR-TKI exposure) and the diagnostic yield of bronchoscopic biopsy was 68% (55/81) for malignant lesions. Malignant or atypical cells were observed in the BWF precipitant samples of 39 (58.2%) lung cancer patients.

Table 1.

Patient characteristics.

Variables	N = 81
Age, years
Average (range)	62.5 (35–84)
Gender, n (%)
Male	54 (66.7)
Female	27 (33.3)
Smoking history, n (%)
Current	22 (27.2)
Former	7 (8.6)
Never	52 (64.2)
Lesion size, n (%)
≤30 mm	22 (27.2)
>30 mm	59 (72.8)
Biopsy location showed by EBUS, n (%)
Central	64 (79.0)
Peripheral	17 (21.0)
Atypical cells in BWF_Pre, n (%)
Yes	39 (48.1)
No	42 (51.9)
Histologic type, n (%)
Adenocarcinoma	54 (66.7)
Squamous cell carcinoma	8 (9.9)
Adenosquamous cell carcinoma	1 (1.2)
Large cell lung endocrine carcinoma	3 (3.7)
Small cell lung carcinoma	1 (1.2)
Non-malignancy	14 (17.3)
Clinical stage, n (%)
I	8 (9.9)
II	11 (13.6)
III	15 (18.5)
IV	33 (40.7)
Not applicable	14 (17.3)
TKI treatment, n (%)
Yes	5 (6.2)
No	76 (93.8)

EBUS endobronchial ultrasound, BWF_Pre bronchial washing fluid precipitant, TKI tyrosine kinase inhibitor.

Fig. 1. Workflow of samples collection and detection.

Transbronchial biopsies were performed under the guidance of EBUS-GS. After formalin-fixation and paraffin-embedding procedures, tissue samples were divided for subsequent pathologic examination and genomic analysis. Clinical tumour purity of tissue samples was shown in Supplementary Table S5. After forceps biopsies, the fixed sheath was utilised to guide normal saline (20–40 ml) directly to the lesion where biopsies were taken. The recovery fluid was centrifuged to gather separated supernatant and precipitate. A 1021-gene panel was used to identify somatic mutations, taking PBLs as germline controls. The quality control information of the sequenced samples was shown in Supplementary Table S7.

BWF_Sup cfDNA yielded advantages over plasma cfDNA and BWF_Pre DNA in genomic profiling

To appraise the genetic information load of BWF_Sup cfDNA, 323 NSCLC patients (79 Stage I/II, 144 Stage III/IV) were enrolled, and plasma cfDNA was extracted and examined. Another plasma cfDNA concentration of 87 healthy individuals came from the Gene-plus database [17]. The results showed that the median concentrations of cfDNA extracted from BWF_Sup in non-malignant patients were superior to that of plasma cfDNA from healthy individuals (P = 0.0012, Supplementary Fig. S1A). As for cancer patients, the median concentrations of BWF_Sup cfDNA in early-stage and advanced-stage patients were 80.4 ng/ml and 288.7 ng/ml, respectively, both of which were significantly higher than the respective means of 6.7 ng/ml (P = 0.0023) and 28.3 ng/ml (P < 0.0001) from NSCLC plasma (Supplementary Fig. S1A). The length distribution of BWF_Sup cfDNA fragment is mainly concentrated at about 167 bp, which was similar to 166 bp in plasma samples (Supplementary Fig. S1B).

After sequencing procedure and mutation calling, a total of 483, 519 and 379 SNVs/Indels/SVs were detected in 67 FFPE (82.7%), 69 BWF_Sup (85.2%), and 62 BWF_Pre samples (76.5%), respectively (Supplementary Tables S3 and S4 and Fig. 2a). The overall distribution of mutational allele frequencies (AFs) in different sample types was displayed in Fig. 2b. BWF_Sup showed modestly higher AFs than FFPE samples (median 9.8% vs. 7.3%, P = 0.046), and both were significantly higher than that of BWF_Pre samples (P < 0.001 and P < 0.001, respectively) (Fig. 2b). Specifically, the percentage of mutations with >10% AFs in BWF_Sup was superior to that in FFPE and BWF_Pre samples (Supplementary Fig. S2), suggesting the enrichment of genomic information in BWF_Sup.

Fig. 2. Genomic characteristics of different sample types.

a Overview of mutation detection in FFPE, BWF_Sup and BWF_Pre. The final diagnosis of each patient is illustrated in the left column. b Distribution of mutational AFs in different samples types. Each spot indicates a specific mutation from each sample. The horizontal bars represent the range of AFs, and the vertical bars represent the mean of AFs for each sample. Statistics is performed via two-sided Mann–Whitney U test. c, d Comparison of TMB and CNV fraction in the whole capture region among different sample types. Statistics is performed via two-sided Wilcoxon signed-rank test. For the calculation of TMB, only nonsynonymous SNVs and indels with ≥3% allele frequencies were included.

TMB had been validated as a predictive biomarker for the efficacy of immune checkpoint inhibitors in NSCLC [30] and was evaluated in different sample types. Overall, the TMB of BWF_Sup was comparable to that of FFPE, and both were superior to BWF_Pre according to matched-pair analysis (P < 0.001 for FFPE, P < 0.001 for BWF_Sup, Fig. 2c). Based on Spearman correlation coefficient, the TMB of FFPE and BWF_Sup exhibited a fairly stronger linear dependence (r = 0.701, P < 0.001) compared with FFPE and BWF_Pre (r = 0.634, P < 0.001) or BWF_Sup and BWF_Pre (r = 0.594, P < 0.001) (Supplementary Fig. S3A).

Chromosomal instability was another common cancer-related event and often challenging to be detected in liquid biopsy attributed to the low-tumour DNA content. Having demonstrated the abundant DNA content in BWF_Sup, we next assessed the fraction of CNV fragments in the total captured genomic regions for each sample. CNV fractions in both FFPE and BWF_Sup were significantly higher than that in BWF_Pre (P < 0.001, P < 0.001, Fig. 2d). Unexpectedly, BWF_Sup displayed a prominent advantage over FFPE in the comparison of CNV fraction, which might be due to the interference of normal lung tissue in FFPE (P = 0.002, Fig. 2d). The illustrations of chromosomal instability in matched FFPE, BWF_Sup and BWF_Pre from ZS013, ZS022, ZS055 and ZS078 were shown in Supplementary Fig. S4, and the instable regions were quite similar in different sample types from the same patients. However, further analysis suggested that CNV fractions in any two sample types were weakly correlated, indicating that the constituents of DNA fragments in each sample type were not exactly the same (Supplementary Fig. S3B).

Taken together, the aforementioned results manifested that BWF_Sup was enriched with tissue-derived cfDNA and outperformed BWF_Pre in comprehensive genomic profiling and FFPE in specific aspects, such as mutation AFs and chromosomal instability.

Genotyping results of BWF as a complementary diagnostic tool to tissue biopsy

Among 67 patients who were finally diagnosed with malignant diseases, somatic mutations (including SNVs, Indels and SVs) were detected in 94.0% (63/67) of FFPE, 97.0% (65/67) of BWF_Sup, and 88.1% (59/67) of BWF_Pre samples, which were all significantly higher than those from 14 benign patients (FFPE, 35.7%, 5/14; BWF_Sup, 35.7%, 5/14; BWF_Pre, 21.4%, 3/14; P < 0.001 for FFPE, BWF_Sup and BWF_Pre, Fig. 3a). According to ROC analysis, the AUC of differential diagnosis was 0.874 for TMB in BWF_Sup, and a slightly inferior 0.733 for TMB in BWF_Pre (Fig. 3a). Another two non-specific parameters, maximal somatic AFs (MSAFs) and CNV fractions, exhibited similar tendency: both were significantly higher in malignant lesions than non-malignant lesions despite of the sample types (except for MSAF in BWF_Pre), and all of the AUC demonstrated the high accuracy for the prediction of malignant lesions in BWF_Sup and BWF_Pre samples, comparing to FFPE samples (Fig. 3b, c).

Fig. 3. Genomic discrepancies of different sample types between malignant and non-malignant patients.

a The mutation-positive rate differs between malignant and non-malignant patients. Two-sided Mann–Whitney U test is used to estimate discrepancies. ROC analysis displayed in the right panel indicates the sensitivity and specificity of TMB in identifying malignancies from benign nodules. b, c Comparison of MSAF and CNV fraction of each sample between malignant and non-malignant patients. Two-sided χ² or Fisher’s exact test is used to estimate discrepancies. ROC analysis displayed in the right panels indicates the sensitivity and specificity of continuous MSAF and CNV fraction in identifying malignancies from benign nodules. d Heatmap illustrates the mutational spectrum of different samples types. Patients are divided on the basis of final pathological diagnosis. Asterisks represent the driver genes for lung cancer according to literature records. LUAD lung adenocarcinoma, LUSC lung squamous carcinoma, LUAD − LUSC lung adenosquamous carcinoma, LCLC large cell lung endocrine carcinoma, SCLC small cell lung cancer.

With regards to the specific detected mutations, only 16 SNVs and Indels (5 in FFPE, 13 in BWF_Sup, 5 in BWF_Pre) with low AFs were detected in 14 patients diagnosed with non-malignant diseases, and none of them were commonly identified driver mutations for lung cancer (Fig. 3d). As a comparison, the most recurrent mutant genes in 67 malignant patients included TP53 (47.8% in FFPE, 46.3% in BWF_Sup, 40.3% in BWF_Pre), EGFR (35.8% in FFPE, 35.8% in BWF_Sup, 32.8% in BWF_Pre), KRAS (16.4% in FFPE, 16.4% in BWF_Sup, 16.4% in BWF_Pre), and FAT1 (11.9% in FFPE, 13.4% in BWF_Sup, 11.9% in BWF_Pre) (Fig. 3d), all of which were identified driver genes of lung cancer [29]. Particularly for four patients with endocrine carcinoma (three large cell, one small cell), the detectability of specific markers, TP53 or RB1 mutations, were 100% (4/4) for FFPE and BWF_Sup and 50% (2/4) for BWF_Pre, implying the advantage of BWF_Sup over BWF_Pre in the diagnosis of rare subtypes (Supplementary Table S3). Via the integration of non-specific parameters and specific driver events, we could well differentiate malignant lesions from non-malignant lesions using both FFPE and BWF specimens without any prior knowledge.

Excluding 55 malignant patients pathologically diagnosed by bronchoscopic biopsy, we further evaluated the diagnostic utility of BWF among 26 patients initially diagnosed with non-malignant lesions. Unlike in the total cohort, the mutation-positive rate of FFPE showed no significant difference between malignant and non-malignant patients (P = 0.238, Supplementary Fig. S5A). Besides, the MSAF and CNV fraction of FFPE and BWF_Pre were similar between malignant and non-malignant patients (Supplementary Fig. S5B, C). Only BWF_Sup persisted differentiated genomic performance (Supplementary Fig. S5), and the detection of driver genes in BWF_Sup samples showed a positive predictive value of 100% (11/11) and a negative predictive value of 93.3% (14/15) for subsequent lung cancer diagnoses among the suspicious cohort without a determined malignant diagnosis, indicating that mutation detection of BWF_Sup was independent of initial clinical diagnosis. The poor detection of FFPE and BWF_Pre might be attributed to the spatial sampling bias of FFPE sections of bronchoscopy biopsy tissue, which led to the different tumour cell fraction in both specimens, whereas the flow characteristic of BWF_Sup might overcome the tissue obstacle in vivo to some extent and provide more lesion-specific information.

Detection of tumour-derived mutations in BWF

In 67 malignant patients, a total of 468 mutations (SNVs, Indels, SVs) were detected in 63 FFPE samples, benchmarking against which 384 (82.1%) and 335 (71.6%) were traced in BWF_Sup and BWF_Pre samples, respectively, and 68.6% (321/468) of mutations were shared by all three sample types. In addition, 115 mutations were exclusively detected in BWF_Sup samples while 39 mutations were exclusively detected in BWF_Pre samples compared to FFPE samples (Fig. 4a). With respect to tumour-derived mutations (TDMs) in BWF_Sup (Fig. 4b) and BWF_Pre (Fig. 4c), the AFs in both samples were positively correlated to those in FFPE samples (BWF_Sup, r = 0.491, P < 0.001; BWF_Pre, r = 0.515, P < 0.001). The positive correlation persisted when only including those driver mutations, although the coefficient between BWF_Sup and FFPE was weaker than total TDMs (BWF_Sup, r = 0.289, P < 0.001, Fig. 4b; BWF_Pre, r = 0.521, P < 0.001, Fig. 4c). To elucidate the functional significance of BWF_Sup and BWF_Pre exclusive mutations, pathway enrichment was performed in terms of Kyoto Encyclopedia of Genes and Genomes (KEGG) categories. As the result, microRNAs in cancer, pathways in cancer and focal adhesion were jointly enriched in both mutation cohorts. Meanwhile, other cancer-related pathways, such as Transcriptional misregulation in cancer, TGF-beta signalling pathway, and Homologous recombination, were also marked in BWF_Sup exclusive mutations (Supplementary Fig. S6). The median tumour purity of BWF_Pre, BWF_Sup and FFPE samples are 0.037, 0.110 and 0.101, respectively (Supplementary Table S6). These results suggested that both BWF_Sup and BWF_Pre exclusive mutations were likewise associated with tumorigenesis and development, and their mismatch in FFPE might be due to spatial heterogeneity.

Fig. 4. Detection of TDMs in BWF_Sup and BWF_Pre samples.

a Venn diagram indicates the mutational overlap of different sample types. b, c The AFs of TDMs in BWF_Sup and BWF_Pre are linearly correlated with that in FFPE according to Spearman correlation analysis. The red and blue spots represent driver and non-driver TDMs, respectively. d The fractions of TDMs detected in each BWF_Sup or BWF_Pre. Each row indicates a specific malignant patient with a different stage manifested by colour gradation. Statistics is performed via two-sided Wilcoxon signed-rank test. e The fractions of main driver TDMs detected in BWF_Sup and BWF_Pre. The difference between both samples is estimated using two-sided χ² or Fisher’s exact test.

Based on individual level, the fraction of detected TDMs was significantly higher in BWF_Sup (median, 100%, 0–100%) than in BWF_Pre (median, 85.7%, 0–100%) according to Wilcoxon matched-pairs signed-rank test (P = 0.001, Fig. 4d). At least one TDM was detected in 95.2% (60/63) of BWF_Sup and 82.5% (52/63) of BWF_Pre samples (χ² test P = 0.023). Total TDM concordance was underlined in over half of BWF_Sup samples (34/63) and 44.4% (28/63) of BWF_Pre samples. In view of pathological stages, only Stage IV patients exhibited a differentiated fraction of detected TDMs between paired BWF_Sup and BWF_Pre (P = 0.028, Fig. 4d). However, the relatively small cases might interfere the robustness of analysis within Stage I–III patients.

High concordances were observed among the TDMs in identified driver genes (29625050), including TP53, EGFR, KRAS, FAT1, etc., and the concordant fractions was similar in BWF_Sup and BWF_Pre except for TP53 (P = 0.040, Fig. 4e). EGFR-mutated lung cancers account for a significant subgroup of NSCLC patients whose prognosis could be dramatically improved by EGFR-TKIs. A total of 27 EGFR mutations with known clinical relevance were detected in 24 FFPE samples (13 with L858R, 9 with e19del, 1 with L62R, 1 with R108K, 1 with S768I, 1 with R671H and 1 with T790M, Supplementary Table S3), and BWF_Sup and BWF_Pre respectively harboured 23 EGFR TDMs (85.2%, Fig. 5e). Notably, BWF_Sup samples identified three additional EGFR mutations that were not found in paired FFPE samples, including 1 with e19del, 1 with L858R and 1 with resistant L718Q mutation that was detected in a progressive tumour after treatment with the third-generation EGFR-TKI Osimertinib (ZS029, Supplementary Table S3). The additionally detected EGFR mutations in BWF_Sup (e19del and L858R) were verified via the Competitive Blocker Q-PCR (Supplementary Fig. S7). Other FFPE-naive mutations that may be associated with drug sensitivity (BRAF V600E, ZS064) to targeted therapies were also detected in BWF_Sup samples with relatively higher AFs than BWF_Pre (Supplementary Table S3). As another marker associate with TKI primary resistance, the point mutations in 12/13 exon of KRAS showed 91.7% (11/12) and 100% (12/12) concordance with FFPE samples in BWF_Sup and BWF_Pre samples, respectively (Fig. 5e). Several types of gene fusion were also identified as driver events in lung cancer, such as ALK, ROS1 and RET fusions, which were also explored in the detection of BWF samples. A total of six gene fusions were detected in FFPE samples, including three CD74/LRIG3-ROS1 fusions, two EML4-ALK fusions, and one KIF5B-RET fusion, of which five (83.3%) and four (66.7%) cases were detected in BWF_Sup and BWF_Pre samples, respectively (Supplementary Table S4 and Fig. 5e). Based on aforementioned results, the driver mutations of tumour tissues were well traced in BWF_Sup and BWF_Pre samples, and the combination of BWF with histologic samples might improve the overall assessment of targeted therapy.

Fig. 5. TDM detection of cytologically negative BWF.

a Driver TDM spectra of BWF_Sup and BWF_Pre from 28 BWF atypical cell (−) malignant patients. The AFs are indicated by colour gradation. b The sensitivities of detecting TDMs in BWF_Sup and BWF_Pre from BWF atypical cell (+) or BWF atypical cell (−) patients. Mann–Whitney U test (unpaired) or Wilcoxon signed-rank test (paired) is used to compare the concordant rate in each patient.

During the microscopic examination of BWF, atypical cell was absent in 28 patients subsequently diagnosed with malignancies. Within those patients, a total of 199 mutations were detected in FFPE samples, of which 164 (82.4%) and 124 (62.3%) were traced in BWF_Sup and BWF_Pre samples, respectively (P < 0.001, Table 2). From the perspective of driver mutations, BWF_Sup still showed a higher consistency with FFPE compare to BWF_Pre (87.5%, 49/56 versus 66.1%, 37/56, P = 0.007, Table 2, Fig. 5a). Furthermore, the AFs of detected driver mutations were superior in BWF_Sup compared to BWF_Pre (P < 0.001, Fig. 5a). Based on an individual level, the fractions of TDMs in BWF_Pre without atypical cell were significantly lower than that in BWF_Pre with atypical cell (P < 0.001) and BWF_Sup without atypical cell (P < 0.001) (Fig. 5b). However, the detectability of TDMs between BWF_Sup and BWF_Pre from patients with atypical cell were similarly based on either mutation level or individual level (Table 2 and Fig. 5b). In summary, these results suggested that cytologically negative BWF still contained sufficient ctDNA for mutation profiling, meanwhile the performance of supernatant and precipitate seemed equivalent in cytologically positive BWF. The TDMs detection sensitivity of BWF_Sup was independent of cytological examination in BWF.

Table 2.

Detectability of TDMs in BWF_Sup and BWF_Pre in patients with or without BWF atypical cells.

			BWF_Sup
			Total			Driver
			+	−	P value	+	−	P value
BWF_pre	All patients (n = 63)*	+	321	14	<0.001	110	2	0.036
		−	63	70		14	11
	Atypical cells (+) (n = 39)	+	199	12	0.347	74	1	1.000
		−	21	37		1	5
	Atypical cells (–) (n = 24)	+	122	2	<0.001	36	1	0.007
		−	42	33		13	6

^*Only include patients whose FFPE was successfully detected with somatic SNVs, Indels or SVs.

BWF as an auxiliary for the reconstruction of tumour subclonal structure

Having shown that BWF_Sup/BWF_Pre exclusive mutations might also be associated with tumorigenesis and development, we hypothesised the joint analysis of FFPE, BWF_Sup and BWF_Pre samples probably enhanced the recognition of intratumoral heterogeneity and clonal population structure. Herein, we used two independent strategies, MATH value and clonal cluster number identified by Pyclone algorithm, to evaluate the intratumoral heterogeneity. Mathematically, two parameters demonstrated modest linear correlation no matter in FFPE (r = 0.432, P < 0.001), BWF_Sup (r = 0.364, P = 0.002) or BWF_Pre (r = 0.534, P < 0.001) (Supplementary Fig. S8). In terms of either MATH value or clonal cluster number, FFPE and BWF_Sup indicated significantly higher heterogeneity than BWF_Pre (Fig. 6a, b). Whereafter, the analysis of clonal and subclonal mutations in each sample exhibited that the number of clonal mutations were similar in different sample types, while the number of subclonal mutations in BWF_Pre were significantly less than that of FFPE and BWF_Sup (Fig. 6c). It seemed that BWF_Sup dealt more effectively with subclone tracing than BWF_Pre. Within the entire clonal mutations, up to 76.6% (141/184) were shared by three sample types, and only 3.3% (6/184) were exclusively detected in BWF_Sup. As a comparison, the distribution of subclonal mutations was quite dispersive. BWF_Sup privately harboured 79.8% (327/410) of subclonal mutations, which was comparable to FFPE (72.0%, 295/410), and mutually exclusive subclonal mutations in both samples accounted for considerable fractions (Fig. 6d). Considering that genomic markers related with acquired resistance usually occurred in subclones, BWF might serve as a promising auxiliary approach to trace drug-resistant mechanism.

Fig. 6. Joint clonal analysis of FFPE, BWF_Sup and BWF_Pre.

a, b Comparison of MATH value and cluster number estimated by PyClone algorithm among different sample types. Statistics are performed via a two-sided Wilcoxon signed-rank test. c Number of clonal or subclonal mutations in each sample. A two-sided Wilcoxon signed-rank test is used to compare the mean of different sample types. d The distribution of estimated clonal and subclonal mutations in different sample types.

Discussion

Flexible bronchoscopy has greatly improved the clinical course of pulmonary medicine. As technology develops, the role of the flexible bronchoscope for both diagnostic and therapeutic indications is continually expanding [31]. However, cellular microscopy and tissue biopsy are frequently disturbed by spatial bias, which also impacts the genetic determination due to insufficient tumour cells. Bronchial washing is a simple and safe procedure during bronchoscopy. Our previous study had shown that BWF could detect EGFR mutations with high consistency to tissue biopsy [32–34]. In this study, we firstly compared genotyping results between BWF_Sup, BWF_Pre and tissue biopsies via NGS platform in a prospective, multi-centre clinical trial. We proved that ctDNA profiling using BWF_Sup samples represented an efficient and accurate liquid biopsy approach for cancer diagnosis to identify clinically/biologically relevant variants and assess multiple genetic characteristics in lung cancer patients, which was potential for compensating for the deficiency of insufficient tissue biopsy.

Since nearly 60–80% of lung adenocarcinomas harbour actionable genetic alterations and locate in the peripheral of the lung [35], we enrolled patients with peripheral lung lesions to enrich cases more worthy of genetic testing. Besides, since the diagnostic yield of bronchoscopic biopsy for peripheral lesions is generally lower than that for endobronchial visible lesions [36], quality controls were set up in order to minimise the bias of bronchoscopic operation on the comparative results, including EBUS confirmed biopsy location and sheath-guided bronchial washing. The guide sheath fixed in the radiology-suspected nodules after biopsy could work as an extended working channel, elevating sampling precision by diminishing the futile drains of normal saline into the uninvolved bronchus and collecting BWF from the same location where biopsies were taken. Additionally, we used a NGS platform to obtain somatic mutations, this sequencing platform has been extensively utilised and validated during R&D and subsequent clinical application [37, 38], and several articles have been published based on this NGS assay and 1% VAF cutoff [39–41], thus the sequencing method and VAF cutoff value in this study are reliable.

Liquid biopsy-based genetic analysis relies on abundant amount of high-quality cancer-derived fraction of cfDNA. In our study, the median cfDNA content of BWF_Sup was 162.2 ng/ml, much higher than that of plasma samples, which was consistent with the study of Yang et al. [17]. The higher cfDNA levels in BWF_Sup samples may be due to the fact that BWF extraction is more invasive than blood extraction and BWF has more contact with tumours. Besides, there was a trend of higher cfDNA levels in blood NSCLC Stage III/IV than in Stage I/II and healthy controls, while there was no significant difference in BWF between III/IV Stage and non-malignant cases. Unfortunately, due to the lack of blood samples from non-malignant patients and BWF samples from NSCLC patients, we were unable to compare the cfDNA concentrations in blood of the non-malignant patients or in BWF from NSCLC patients, but it will be a meaningful exploration. Furthermore, BWF_Sup showed higher AFs than FFPE samples, demonstrating a high proportion of ctDNA [42, 43]. Notably, our study first explored the fragment size of BWF_Sup cfDNA. With a similar fragment size of cfDNA detected in BWF with that in plasma samples, we hypothesise that cfDNA may origin from apoptosis or necrosis tumour cells or active release of living tumour cells from tumour in situ [44]. BWF_Sup ctDNA may thus mainly reveal the genetic information from primary tumour lesions. However, the origin of cfDNA and its existence form needs further exploration.

For locally advanced or metastatic NSCLCs, determining the molecular genotyping in a precise manner can be crucial to improve the outcome [45]. We identified a consistency of 82.1% between BWF_Sup and FFPE samples in genetic variants detection, and an even higher consistency (90.5%) for library-putative driver mutations. Notably, for gene fusions, whose detection was hampered in plasma samples due to the low cfDNA amount and enzymatic degradation [46, 47], but BWF_Sup could also achieve 83.3% consistency with FFPE samples. A clinical trial compared the sensitivity of tissue biopsy and BWF_Sup samples to detect EGFR mutation in more than 100 lung cancer patients, and found that the sensitivity of BWF_Sup was 92.5%, while the total sensitivity of tissue biopsy was only 77.5% [48]. Furthermore, we observed a certain correlation between maximum allele frequency (MSAF) of individual BWF_sup samples and the purity of the corresponding FFPE samples (Spearman r = 0.379; P = 0.001) (Supplementary Fig. S9), in which MSAF could be used to estimate the amount of ctDNA [49]. Collectively, we believe that the application of BWF in genetic testing is a very good complement to tissue biopsy, especially in the case of insufficient biopsy specimens, and lesions unsuitable for biopsy. Except for targetable alterations, several immunotherapeutic biomarkers are commonly used to distinguish populations that benefited from emerging immunotherapies, such as TMB. Herein, BWF_Sup exhibited a strong correlation with FFPE in terms of TMB assessment, indicating that BWF_Sup might be a reliable substitute of tissue biopsy in immunotherapeutic biomarker prediction.

These results have profound diagnostic implications by highlighting the potential advantage of ctDNA over biopsy samples in the setting of false-negative results. In our study, the BWF_Sup was able to detect additional variants compared with FFPE and precipitant samples, providing more options for subsequent treatment strategies. Though the detection of driver genes has not been used as a criterion in differentiating malignant lesions from benign, the information could still be exploited to stratify patients who require more-aggressive diagnostic procedures. In addition, for patients with acquired resistance after TKI treatment, BWF_Sup may better capture the heterogeneity of acquired resistance [21, 50]. In one patient of our study, BWF_Sup was able to detect a drug-resistant mutation (EGFR-L718Q) when tissue biopsy failed to reveal malignancy in a progressive tumour after treatment with the third-generation EGFR-TKI Osimertinib. Recent studies have explored the feasibility of ctDNA from BWF supernatant for early-stage lung cancer [51]. Our study provides thorough insights into the potential of BWF_Sup ctDNA in elucidating the genetic characteristics of NSCLC patients.

Several limitations persisted. First, no blood-based liquid biopsy was collected, resulting an inability to compare cfDNA between paired blood and BWF_Sup samples. Second, this study was observational and the outcome of clinical intervention guided by BWF profiling was lacking. Third, the collection of BWF guided by a fixed sheath inevitably might lead to intra-tumour heterogeneity. However, through multiple perspective comparisons of genetic parameters, we evaluated the similarities and differences between BWF and tissue biopsy, indicating that BWF could be used as a supplement to molecular diagnosis of tissue specimens and exert its clinical application value. These findings could help to improve the diagnostic accuracy during the initial clinical course, thereby expanding the therapeutic window to achieve superior outcome. Additionally, the concordance of clinically relevant mutations between BWF_Sup and FFPE, as well as the extra detectability of actionable mutations in BWF_Sup, might enhance the likelihood to benefit from targeted therapies and trace drug-resistant clones.

Interpretation

CtDNA from BWF_Sup outperformed BWF_Pre in genomic profiling [18, 52], and could serve as a promising approach for the diagnosis of radiology-suspected pulmonary nodules without any prior knowledge, of which the performance was independent of the false-negative results derived from bronchoscopic biopsies. Genetic alterations identified in BWF_Sup were quite consistent with tissue biopsy, and could supply additional subclonal information to help achieve comprehensive treatment strategies.

Author contributions

Prof. Xin Zhang is the guarantor of the content of the manuscript, including the data and analysis. Xinyu Zhang and YX had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. ZY, Yencheng Chao, QH, CL, MY, X Zhu, LC, JB, Y Gong, Y Guan, MZ, JH, HZ, TR, QS, KW, YH, XX, DPC and XP contributed substantially to the study design, data analysis and interpretation, and the writing of the manuscript.

Funding

None.

Data availability

The datasets supporting the conclusions of this article are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

This study was approved by the Ethics Committee of the Zhongshan Hospital of Fudan University (No. B2018-027) and complied with all relevant ethical regulations. All participating patients signed written informed consent.

Consent to publish

Not applicable.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-022-01969-2.