To the editor

Approximately 30–55% of non-small-cell lung cancer (NSCLC) patients developed recurrence despite curative resection [1]. Circulating tumor DNA (ctDNA) is shed by tumor cells and may serve as an effective prognostic marker following multiple therapeutic modalities [2,3,4,5]. However, it remained not fully understood to what extent serial ctDNA monitoring could help identify the risk of recurrence in resectable NSCLC.

In this study, a total of 128 patients with resectable NSCLC were enrolled (Fig. 1A). Primary tumor and lymph node metastasis (LNM) samples were collected from curative surgeries as standard of care. Plasma samples were collected before surgery, 7 days after surgery, and every three months thereafter. Both tissue and plasma samples were sequenced using a comprehensive 425-gene panel (Fig. 1A, B). One patient was excluded during quality control (Additional files 1, 2: Table S1 and S2).

Fig. 1
figure 1

Study design and ctDNA detection. a Workflow of sample collection, sample exclusion, and data analysis. b Schematic diagram illustrating the timeline for sample collection and the number of plasma samples available for analyses at each time point. c Proportions of patients that showed different presurgical and postsurgical ctDNA status. The results of ctDNA detection of all postsurgical samples were included. d Proportions of patients positive for presurgical (upper panel) and postsurgical (lower panel) plasma samples, stratified by pathology histology, TNM stage, LNM status, and smoking history

Full size image

A total of 611 plasma and 593 tissue samples were included in the analyses (Fig. 1A, Additional file 4: Fig. S1). We reconstructed the clonal phylogeny of each patient from multi-region tissue sequencing to buttress the ctDNA detection (See Additional file 11: Methods). Near half (46.4%, 59/127) of the patients were ctDNA-negative throughout the investigation period. In 32.3% (41/127) of the patients, ctDNA was detected in at least one postsurgical plasma sample, most of whom (65.9%, 27/41) were ctDNA-positive in presurgical samples (Fig. 1C; Additional file 5: Fig. S2).

As shown in Fig. 1D, patients with lung squamous cell carcinoma (LUSC) were more frequently ctDNA-positive than those with lung adenocarcinoma (LUAD). The detection rate correlated with TNM stages and LNM status, and smokers were found with a higher ctDNA-positive rate than non-smokers in presurgical instead of postsurgical results.

Postsurgical ctDNA detection at as early as seven days after surgeries could indicate high risk of recurrence (HR = 3.90, P = 0.00011; Fig. 2A), independently of clinicopathological characteristics (multivariate-Cox: HR = 5.49, P = 0.002; Fig. 2B). ctDNA detection at following time points (3 months and 6 months) could also serve as prognostic markers (3 months—HR = 4.32, P < 0.0001; 6 months—HR = 6.19, P < 0.0001) and remained statistically significant after adjusted for clinicopathological characteristics (multivariate-Cox: 3 months—HR = 4.17, P < 0.001; 6 month—HR = 4.59, P < 0.003; Additional file 6: Fig. S3). Longitudinal ctDNA detection accurately identified high risk of disease recurrence (univariate Cox: HR = 7.59, P < 0.0001, Fig. 2B; multivariate-Cox: HR = 8.33, P < 0.001, Fig. 2C) and covered the most of relapsed cases (73.5%, 25/34). In these cases, ctDNA detection led radiographic relapse by a median of 145 days. The time intervals were similar between LUAD and LUSC (144 and 150 days, respectively) (Fig. 2D, E; Additional files 7, 8, 9: Figure S4-6). Other results were shown in Additional files 10 and 13.

Fig. 2
figure 2

Prognostic values of ctDNA mutation. a, b Analysis of recurrence-free survival of patients stratified by 7-day postsurgical (a) and longitudinal (b) ctDNA detection. Univariate Cox regression results were shown. c The results of multivariate-Cox regression for recurrence-free disease in patients stratified by longitudinal ctDNA detection. d Swimmer plot illustrating the ctDNA status, adjuvant therapy, and pathological events of cases with disease recurrence (n = 34). e Time of the earliest ctDNA detection and radiographic relapse, measured by days from the surgery. f Analysis of recurrence-free survival of patients stratified by the clonality of longitudinal ctDNA detection. The ctDNA-positive (Clone) group comprised patients with at least one clonal mutation detected in at least one postsurgical plasma sample. The ctDNA-positive (Subclone) group comprised patients with at least one subclonal mutation detected in at least one postsurgical plasma sample and no clonal mutation detected in any postsurgical samples. The ctDNA-negative group comprised patients with no mutation detected in any postsurgical plasma samples. g, h Clonal phylogenetic information of tissue and plasma samples of Patient 60 (g) and Patient 53 (h). Heatmaps denote mutation profiles of multi-regionally resected primary tumors, lymph node metastasis, and plasma samples with clonal annotation (leftmost column) representing mutation clusters. Phylo-groups comprise samples having identical clonal phylogeny. Colored nodes denote the detection of ctDNA mutations in respective clones, whereas gray nodes denote that no mutation in respective clones was detected

Full size image

We further found that clonal mutations in ctDNA were more prognostically informative than subclonal ones. During the longitudinal ctDNA surveillance, patients with clonal mutation exhibited a worse prognosis than ctDNA-negative ones (HR = 10.07, P < 0.0001), and no significantly differential survival was observed between those with only subclonal mutations (See Additional file 11: Methods) detected and the ctDNA-negative group (HR = 1.94, P = 0.305) (Fig. 2F). Nonetheless, tracking subclonal dynamics in ctDNA may inform the source of relapse. In Patient 60, at the six-month time point, three mutations from subclones 1 and 2 were detected in plasma. Subclones 1 and 2 were specific to three regions of primary tumor 2, suggesting that primary tumor 2 may be the active source of ctDNA. Later the sequencing of the relapse lesion confirmed primary tumor 2 as its clonal origin (Fig. 2G). In Patient 53, clonal EGFR 19Del and subclonal SMAD4 mutations were detected in plasma shortly before disease relapse. The subclonal SMAD4 mutation was absent from LNM, whereas LNM-specific STK11 mutation was undetectable in ctDNA, together suggesting that LNM may not be an active source of ctDNA or disease recurrence (Fig. 2H).

In summary, we found that ctDNA could serve as a promising biomarker for risk of recurrence in NSCLC patients who receive curative surgeries, and the results were further discussed in Additional files 3 and 12. As early as 7 days after the surgery, ctDNA detection identified patients at high risk. Longitudinal ctDNA surveillance could reliably predict recurrence, which opens a window of almost 145 days for optimal disease management. Furthermore, our results showed that tracking subclonal dynamics could inform the origin of tumor recurrence.