Document resource
Introduction Non-small cell lung cancer (NSCLC) accounts for ~85% of lung cancers and remains the deadliest cancer globally. 1 Surgical resection offers a curative option for approximately 30% of patients with stage I–IIIA disease.2 Postoperative surveillance relies on clinical examinations and imaging, but molecular residual disease (MRD)—a driver of early relapse—often escapes detection due to limited sensitivity.3 4 Neoadjuvant (NA) and adjuvant (AC) chemotherapy (CT) improve survival, likely by targeting MRD,5–9 but concerns about toxicity and modest long-term benefit highlight the need for biomarkers to better guide treatment and avoid over- or undertreatment.10 ctDNA has emerged as a promising non-invasive biomarker for MRD detection, monitoring NA response and identifying early recurrence, with growing validation across cancers.11–16In advanced stages, identifying actionable genomic alterations (AGAs) has enabled targeted therapies (TT) that improve outcomes, but accurate response assessment remains critical for guiding treatment.17 This review explores the evolving role of liquid biopsy (LB) in early-stage NSCLC for MRD detection and risk stratification and predicting response to systemic therapies in advanced disease (figure 1A,B).Figure 1Emerging landscape for ctDNA application in (A) early stage, and (B) advanced stage non-small cell lung cancer. ctDNA, circulating tumour DNA; MPR, major pathological response; MRD, molecular residual disease; NGS, next generation sequencing; PD-L1, programmed death-ligand 1; RECIST, Response Evaluation Criteria in Solid Tumours; TPS, tumour proportion score; WES, whole exome sequencing; WGS, whole genome sequencing.Analytical methods and challenges in ctDNA detection Plasma cell-free DNA (cfDNA) consists of short DNA fragments released into circulation via apoptosis or necrosis, with a half-life of 1–2 hours. 18 While most cfDNA originates from normal cells, tumours can release ctDNA, which is typically more abundant in advanced-stage NSCLC.19 20The variant allele frequency (VAF)—the proportion of mutant to total reads at a given locus—is the primary metric used to quantify ctDNA. Detection methods include PCR and next-generation sequencing (NGS).21 Digital droplet polymerase chain reaction (ddPCR) enables high-sensitivity mutation detection at VAFs as low as 0.001% under optimal conditions (figure 2).21 NGS offers broader mutation profiling, including single nucleotide variants (SNVs), indels, copy number variations and rearrangements. Amplicon-based NGS is fast and cost-effective but prone to allele dropout. Hybrid capture-based platforms, including Cancer Analysis of Plasma Proteins Sequencing (CAPP-Seq), provide higher sensitivity by capturing larger fragments and tolerating mismatches.21 Whole exome sequencing (WES) and whole genome sequencing (WGS) expand detection to coding and non-coding regions but require high-quality input and complex data analysis (table 1).21 Advances like SAGA, NeXT Personal and PhasED-Seq enhance WGS sensitivity for MRD detection by targeting rare variants and phasing multiple mutations on individual ctDNA molecules (figure 2).22 23 Finally, methylation profiling detects cancer-specific (cs) epigenetic alterations in ctDNA, offering a sensitive approach for early detection, particularly in low-shedding tumours like early-stage lung cancer (table 1).24Figure 2Evolution of circulating tumour DNA detection platforms for MRD analysis. CAPP-Seq, Cancer Analysis of Plasma Proteins Sequencing; ddPCR, digital droplet PCR; MRD, molecular residual disease; VAF, variant allele frequency.Table 1Comparison of ctDNA detection platforms for MRD analysisTestSensitivityTissue requirementTurnaround timeProsConsClinical applicabilityPCR1%–5% VAFNo4–24 hoursFast and low-costDetects mutations with high VAF (big tumours)Depends on primers designProvides relative quantificationLimited genomic coverageUseful when speed matters more than ultra-high sensitivityDigital droplet PCR0.01%–0.1% VAFNo1–3 daysHigh reproducibility because each droplet acts as an independent reactionProvides absolute quantification (copies/mL)Not ideal for multiple mutations or broad profilingBetter quantification. Useful for MRD or early recurrenceAmplicon-based NGS0.1%–1% VAFOptional for design1–7 daysGood for multiplexed hotspot testsLower cost than broad NGS and relatively fast.PCR amplification biases and limited to amplicon regionsHotspot panels, fast genotyping of multiple known hotspotsHybrid capture NGS—CAPP-Seq0.1%–0.01%—CAPP-Seq: 0.01%–0.001%Helpful but not mandatory3–10 days (3–14 days for CAPP-Seq)Biotinylated probes hybridise to target regions translating in broad genomic coverage.Clinically validated commercial optionsLess sensitive than the most optimised MRD-focused assaysComprehensive genotyping for advanced cancer, guiding targeted therapy, trial matching, longitudinal monitoring (with some sensitivity limits)PhasED-SeqExtremely high—0.0001% VAFUsually required to identify phased mutations7–14 daysTarget rare variants by phasing multiple mutations on individual ctDNA moleculesRequires multiple patient-specific phased sites or panel designSophisticated bioinformaticsNot yet universally availableBest-in-class sensitivity for MRD; very low false positive; ideal for postsurgical MRD detection and earliest recurrence signalsWES>0.5% VAFcan improve with ultra-deep WESRequire sufficient tumour tissue (issues with CR or small tissue samples)3–4 weeksCapture of exonic regions (all coding genes)Broad discovery: can detect many coding mutations without prior knowledgeRequires larger sequencing inputHigh cost to reach high depthGenerally, more cell-free DNA needed for high sensitivity.Research, discovery of novel alterations, building patient-specific panels for downstream MRD assaysWhole genome sequencingVery lowRequire sufficient tumour tissue (issues with CR or small tissue samples)3–4 weeksSequence entire genome (coding+non-coding)Can detect single nucleotide variants, structural variants, copy number alterations, fragmentomics and methylation signals if integrated.Large input and budget required for ultra-deepResearch, pan-omic approaches (fragmentomics and methylation), exploratory studies, copy number profiling; not routine for sensitive MRD detectionMethylation profilingVery high (similar to or better than hybrid capture depending on the panel)No3–10 daysHigher signal density than point mutationsGood for tissue-of-originMutation-agnosticBioinformatics intensivePanels differ widelyExcellent for MRD detection, early cancer detection and tissue-of-origin inferenceCAPP-Seq, Cancer Analysis of Plasma Proteins Sequencing; CR, complete responses; ctDNA, circulating-tumour DNA ; MRD, molecular residual disease; NGS, next generation sequencing; VAF, variant allele frequency; WES, whole exome sequencing.LB in early-stage NSCLC Correlation of LB and clinicopathological features Detecting ctDNA in early-stage NSCLC remains challenging, particularly for diagnosing and monitoring MRD, where accuracy and timing are critical for treatment decisions. Two main strategies are used: tumour-informed and tumour-agnostic approaches. 25 Tumour-informed methods involve WES or WGS of the primary tumour to design personalised panels for ctDNA tracking. While offering high sensitivity, they require sufficient tumour tissue—often lacking after major pathological response (MPR) or pathological complete response (pCR)—and are further limited by yield of small biopsies. These assays also have longer turnaround times (3–4 weeks) and may exclude patients without detectable mutations, leading to potential bias. Tumour-agnostic approaches use predefined NGS panels to detect common mutations without prior tumour sequencing. They are faster (7–10 days), tissue-independent and more broadly applicable but may miss low-frequency mutations or produce false positives from clonal haematopoiesis (CH) if not corrected.25 Nonetheless, to mitigate CH-related artefacts, modern ctDNA assays perform parallel sequencing of matched leukocyte DNA to identify and computationally remove haematopoietic-derived variants, thereby increasing specificity.Both strategies are being evaluated in trials of NA and adjuvant therapy (AT), but no standard has been established. A meta-analysis of 16 studies (1251 patients) confirmed the clinical value of ctDNA in predicting relapse, with high specificity across landmark and surveillance strategies.25 Notably, tumour-agnostic MRD assays—especially WGS-based—showed the highest pooled sensitivity, though amplicon- and hybrid-capture methods performed similarly.26Among the earliest studies that described the correlation of ctDNA and clinicopathological features, the TRACERx study (Abbosh et al) assessed the prognostic value of preoperative ctDNA in early-stage NSCLC using a tumour-informed anchored-multiplex PCR assay (ECLIPSE) targeting 200 mutations, with 95% sensitivity at 0.008% VAF. Among 197 patients, 88 had lung adenocarcinoma (LUAD). In this subgroup, 2-year overall survival (OS) was 90% in ctDNA-negative (−) patients (n=52), compared with 63% (n=25) with low ctDNA and 24% (n=11) with high ctDNA levels. No significant OS correlation was seen in non-LUAD histologies (p=0.3), suggesting histology-dependent prognostic value.19 In addition, an ancillary analysis from the TRACERx study also linked ctDNA detectability to tumour necrosis, proliferation, lymphovascular invasion and size—mean VAF reached 0.1% at 10 cm³ tumour volume.27Consistent with this finding, the ctDNA Lung-DETECT study (n=151), using a tumour-informed RaDaR panel, identified male sex, smoking, squamous histology, nodal involvement, tumour size, pleural and lymphovascular invasion as predictors of ctDNA positivity.28 In a secondary analysis by Coyne et al, ctDNA was detected in 57% (8/14) of patients with occult nodal disease versus 5% (6/118) without (p<0.01).29 Similarly, in patients harbouring AGAs, a recent updated analysis of MRD in the NeoADAURA study (n=358)—which compared osimertinib plus CT (n=121), osimertinib monotherapy (n=117) and CT alone (n=120) in patients with completely resectable stage II–III EGFR mutant NSCLC—found that MRD positivity was associated with stage II (vs III, p<0.0001), larger tumours (p=0.02) and baseline nodal involvement (p=0.0001).30 31 These findings suggest that ctDNA detection may help guide invasive mediastinal staging in clinically node-negative NSCLC.Tumour-agnostic approaches have also shown promising results. Hong et al analysed 895 patients with stage I–II EGFR/ALK wild type LUAD using a methylation-based assay (MBA). Among 414 with stage I disease, preoperative ctDNA was detected in 13%, correlating with worse 2-year recurrence-free survival (RFS, 69% vs 91%, p<0.001) and high-risk features like grade 3 histology, upstaging and elevated programmed death-ligand 1 (PD-L1) (55% vs 28%, p<0.001).32 Likewise, Driussi et al evaluated ctDNA in n=260 patients with stage I NSCLC using a MBA. Detection correlated with pathological stage, tumour grade and volume across LUAD and squamous cell carcinomas. These findings suggest that the presence of ctDNA in LUAD may be associated with aggressive histological features, including necrosis, inflammation, spread through air spaces and lymphovascular invasion.33Prognostic significance of preoperative ctDNA in NSCLC Several studies have reported the increasing application of preoperative ctDNA as a biomarker of tumour burden for guiding potential postoperative treatment strategies. A meta-analysis including prospective and retrospective studies by Li et al showed that preoperative ctDNA was associated with worse RFS (HR 3.0, p<0.05) and OS (HR 2.77, p<0.05), particularly in patients with LUAD (HR=3.46) and patients with stage I–II NSCLC. Moreover, patients with positive (+) preoperative ctDNA had better survival outcomes from postoperative AC when compared with baseline ctDNA− patients.To investigate the potential of deeper ctDNA sequencing to assess prognosis in preoperative NSCLC, Black et al analysed 171 TRACERx patients using the tumour-informed NeXT Personal assay with enhanced sensitivity.34 In stage I LUAD, ctDNA was detected in 57% (16/28), a marked improvement over prior studies (13%–14% in Abbosh et al). For stage II LUAD, detection reached 79% (23/29) versus 44% previously reported. ctDNA shedding correlated significantly with smoking history (p=0.021) and high-grade LUAD subtypes, particularly solid and cribriform patterns (p<0.005). Stratifying patients by ctDNA burden (negative, low and high) showed strong prognostic value, with 5-year OS rates of 100%, 61.4% and 48.8%, respectively. HRs for OS were 11.08 (low) and 19.33 (high), with similar trends for RFS.34Preoperative ctDNA is also a prognostic biomarker in early-stage NSCLC with AGAs. Jung et al, prospectively studied 278 patients with resected stages I–IIIA EGFR-mutant NSCLC using ddPCR for longitudinal ctDNA monitoring.35 Preoperative ctDNA was detected in 24% (67/278), with 76% clearing ctDNA by 4 weeks postsurgery, showing no difference between exon 19 and L858R mutations.35 Patients were grouped by baseline ctDNA and postoperative MRD status, with 3-year disease-free survival (DFS) rates of 84% (ctDNA−), 78% (ctDNA+MRD−) and 50% (ctDNA+MRD+) (p=0.02). MRD detection occurred earlier in exon 19 deletions than L858R (19 vs 33 months).35 These results support the potential utility of preoperative and longitudinal ctDNA monitoring using simple techniques for risk stratification in resected EGFR-mutant NSCLC.Dynamics of ctDNA during NA treatment Perioperative treatments may limit subclonal expansion in early-stage NSCLC by removing emerging metastatic clones before they spread. Using phylogenetic tracing, Abbosh et al reported that in 43% (18/42) of patients with evaluable postoperative plasma samples, a primary-tumour subclone later dominated all metastatic sites.27 36In the 5-year outcomes of the phase II NADIM trial (n=46) in resectable stage IIIA/IIIB NSCLC, higher pretreatment mutant allele frequency (sumMAF ≥1%) was associated with shorter progression-free survival (PFS) (48.6% vs 83.8%, HR 4.26, p=0.008) and OS (56.3% vs 86.2%, HR 6.88, p=0.007), using a tumour-agnostic amplicon-based assay (Oncomine).37 Similarly, ctDNA clearance (<0.1% MAF or undetectable) after NA treatment correlated with improved 5-year PFS (85.2% vs 60.6%) and OS (92.3% vs 59.2%).37 38Consistently, the phase III CheckMate 816 study (n=358) in patients with resectable NSCLC, NA nivolumab plus CT significantly improved event-free survival (EFS) and pCR and, notably, showed a positive 5-year OS benefit compared with CT alone.39 40 ctDNA analysis was performed in 89 patients, revealing that the percentage of patients with ctDNA clearance was higher with CT-immunotherapy (IT) compared with CT alone (56% vs 35%, respectively).40 Furthermore, EFS was longer in patients with ctDNA clearance than those without in both arms (HR 0.60 for CT-IT and HR 0.63 for CT alone), and the percentage of patients with pCR was higher among those with ctDNA clearance.40 In the same way, in the phase III AEGEAN study (n=802) of perioperative durvalumab in stage II–III NSCLC, early ctDNA clearance (before cycle 2) was linked to higher pCR rates (50% vs 15.1% with durvalumab; 14.3% vs 3.1% with placebo) and MPR (66.7% vs 35.8%; 38.1% vs 12.25%, respectively).41 42 Additionally, mean VAF (mVAF) decreases were greater in patients with pCR/MPR (p≤0.003), supporting ctDNA clearance as a biomarker for NA-IT efficacy in early-stage NSCLC.42In patients with AGAs receiving NA treatment, a phase II multi-institutional study (NCT03433469) of NA osimertinib monotherapy (n=27) in resectable stage I–IIIA EGFR mutant NSCLC reported a MPR rate of 15%.30 MPR was not predictive of EFS, but notably, in the multivariate analysis, patients with detectable ctDNA prior to cycle 1, day 1 (C1D1) of NA osimertinib had significantly shorter median EFS compared with those with undetectable ctDNA (8.5 vs 45.6 months; p=0.03)43Following this idea, the NeoADAURA study (n=358) that compared osimertinib plus CT (n=121), osimertinib monotherapy (n=117) and CT alone (n=120) in resectable stage II–III EGFR-mutant NSCLC44 showed that osimertinib-containing regimens had higher MPR rates (26% with osimertinib+CT, 25% with osimertinib alone) versus CT alone (2%, p<0.0001).44 Furthermore, Blakely et al presented the MRD analysis from the NeoADAURA study (n=189) using Personalis’ NeXT Personal assay.31 Compared with plasma-based PCR (Cobas), the tumour-informed MRD NeXT Personal panel showed a 71% detection versus 30% with plasma PCR for baseline ctDNA. The EFS was significantly higher in patients with MRD not detected at baseline versus MRD detected (EFS at 18 months=96% vs 82%, HR 0.24, 95% CI 0.07 to 0.80, respectively). In addition, baseline MRD clearance (defined as a 10-fold decrease in ctDNA or MRD not detected in presurgical samples after baseline MRD detected) was enriched with osimertinib-containing regimens (MRD clearance 83% with osimertinib+CT, 84% with osimertinib monotherapy, and 58% with CT alone) and in patients with MPR (MPR 24% in patients with MRD clearance vs 6% in MRD non-clearance, p=0.03).31 These results provide preliminary evidence that MRD could complement MPR and that NA osimertinib may be beneficial in EGFR-mutant stage II–III NSCLC.Postoperative surveillance of ctDNA and AT strategies Several studies have evaluated the prognostic value of postoperative and longitudinal ctDNA-based MRD detection. Guo et al conducted a meta-analysis of 11 studies (1104 patients with early-stage NSCLC) examining ctDNA detection at 1 month and >3 months postsurgery and its association with survival.45 Detection rates were higher in stage III than in stages I–II at both time points (1 month: 46.1% vs 9.9%; long term: 82.4% vs 28.3%).45 ctDNA positivity at either time point was associated with shorter RFS (HR 4.43 and 7.99) and worse OS (HR 5.07 and 7.49), independent of ethnicity.45 Additionally, ctDNA+ patients benefited from AC (HR 0.3), while ctDNA− patients showed no significant RFS improvement (HR 1.18).45Regarding the optimal timing for postsurgery ctDNA detection, the LUCID study (n=88) examined optimal timing for postsurgery ctDNA detection in NSCLC (stage IA–IIIB) using a 48-variant tumour-informed assay. ctDNA was detected in 25% (12/48) of patients 1–3 days after surgery (mVAF 0.0026%) but was not linked to recurrence.46 In contrast, ctDNA detected >2 weeks postsurgery was strongly associated with shorter RFS and OS (HR 14.8 and 5.48, p<0.05).46 Additionally, patients who were ctDNA+ before but negative after treatment had excellent outcomes, with only 2 of 16 relapsing, suggesting AT could help de-escalate in some selected cases.46 These findings suggest that ctDNA testing beyond 2 weeks postsurgery offers greater prognostic value than immediate postoperative detection.47Aligning with de-escalating strategies for AC based on relapse risk, Zhang et al conducted an observational study (n=261) in patients with resected stage I–III NSCLC (table 2).48 The study showed that patients with ctDNA− MRD over 18 months (using a tumour-informed ctDNA panel) may represent a ‘cured’ population, unaffected by clinical stage.48 Furthermore, the detection of MRD preceded radiographic progression in 87.2% of patients, with a median lead time of 3.4 months.48 Consistently, Qiu et al conducted a prospective study (n=116) demonstrating that ctDNA positivity, both after surgery and following AC, was significantly associated with worse RFS (HR 3.95, p<0.05 and HR 3.22, p=0.009, respectively).49 In contrast, ctDNA− patients with stage II–III NSCLC exhibited a low risk of relapse, regardless of whether AC was administered.49 Moreover, longitudinal ctDNA monitoring enabled earlier detection of recurrence compared with imaging, with a median lead time of 88 days.49 The use of joint modelling combining serial ctDNA measurements with time-to-recurrence data provided a superior predictive performance at 12 and 15 months postsurgery compared with traditional Cox models (HR 8.55, p<0.05).49 Collectively, as supported by other studies (table 2), ctDNA dynamics after surgery and AT represent an evolving biomarker for predicting recurrence risk and guiding personalised treatment decisions in resected NSCLC.26 49 50Table 2Clinical studies testing postoperative ctDNA in patients with early-stage NSCLC (including ≥100 patients)Study (n)Disease stagePostsurgical ctDNA collection timePanel usedPostoperative ctDNA detection rate (%)Effect of postoperative ctDNA status on prognosisAbbosh et al,19 (TRACERx), n=197IA–IIIA120 daysTumour-informed: anchored-multiplex PCR tracking (200 mut)25%ctDNA landmark+ patients exhibited a HR of 5.3, p<0.05 for OS, and a HR of 6.8 for freedom from recurrence, p<0.05, compared with ctDNA landmark - patientsChen et al,83 (PROPHET), n=181IA–III3–7 days and 1 month postsurgeryTumour-informed: personalised panel from whole exome sequencing, 50 patient-specific variants16% at 3–7 days after surgery; 12% at 1 month after surgeryHR 5.31 and 16.40 at 3–7 days and 1 month after surgery (both p<0.001)Fu et al,84n=177I–III1 month (±7 days) after surgeryTumour-informed: NGS panel of 338 cancer-related genes24.7% (36/146) landmarkPatients with ctDNA+ at landmark or longitudinal time points had markedly reduced DFS (HR 5.07 and 7.14, respectively, both p<0.01)Xia et al,85(LUNGCA-1), n=330I–IIIBefore surgery, 3 days and 1 month after surgeryTumour-informed: customised NGS- 769-gene panel6.4% (19/296) at 3 days after surgery; 5.9% (19/324) at 1 months after surgeryThe presence of MRD+ (ctDNA positivity at postoperative 3 days and/or 1 month) was a strong predictor for disease relapse (HR 11.1; p<0.001).MRD+ patients who received AT had improved RFS over those not receiving AT (HR 0.3; p=0.008)Zhang et al,48n=261I–III1 month after surgery and every 3–6 months.Primarily tumour-informed: customised NGS 1021 gene panel8.5% (21/245) LandmarkDFS (landmark) of MRD− versus MRD+ NR versus 12.1 months. HR 0.08, p<0.001DFS (longitudinal) of MRD− versus MRD+ NR versus 15.9 months. HR 0.02, p<0.001Li et al,86n=123Resectable I–IIIABefore, 1 month after surgery, and every 3–6 months up to 3 years.Tumour agnostic: NGS panel of 425 cancer-related genes10.3% (12/116)Longitudinal ctDNA positive 31.1% (37/119)Postsurgical ctDNA positivity was associated with shorter RFS (HR 3.04, p=0.01). Patients who remained ctDNA+ during longitudinal monitoring had significantly shorter RFS (HR 3.46, p<0.001) and OS (HR 9.99, p=0.01) compared with those who were longitudinally ctDNA−Wang et al,87n=128I–IIIBefore surgery, 7 days after surgery and every 3 months.Tumour agnostic: NGS panel 425 cancer-associated genes32.3%ctDNA positivity 7 days after surgery was associated with a high risk of recurrence (HR 3.90, p<0.001). Patients with longitudinal ctDNA positivity had an even higher risk of recurrence (HR 7.59, p<0.001)Zhang et al,50n=108I–III(n=73)After lung surgery and ATTumour-informed: personalised panel multiplex PCR-based (Illumina HiSeq 4000)28.8% (21 patients) postsurgical17.8% (13) after ATThe ctDNA+ patients had significantly worse RFS (HR 8.84, p<0.001) compared with ctDNA−Wang et al,47n=270 (four cohorts)I3 days to 1 months after surgeryTumour informed: (1) Amplification of 127 genes; (2) NGS 769 cancer-related genes; (3) RaDaR assays; (4) NGS −139 cancer-related genes3.3% (9/270)The risk of recurrence was significantly lower in the ctDNA− group compared with the ctDNA+ group (HR 0.11, p<0.0001). Among ctDNA+ patients, there was no significant difference in RFS between those who received AT and those who did not (p=0.58).Qiu et al,49n=116Resected NSCLC I–IV1 month after surgery and before the startof ATTumour-informed: NGS panel with 139 cancer-related genes21.2%Within both the AT and non-AT groups, ctDNA+ patients exhibited a higher risk of recurrence compared with ctDNA− patients (p<0.05). Among ctDNA+ patients, those who received AT showed significantly improved RFS compared with those who did not (p<0.05).Zhou et al,53(IMpower010), n=1005IB–IIIAPrechemotherapy and pre-atezolizumab/placeboTumour-informed: Natera Signatera RUO assay20% (118/600)The 5-year DFS HRs were 0.70 for ctDNA− patients and 0.67 for ctDNA+ patients in stages II–IIIAJung et al,35n=278IA–IIIA EGFRmut4 weeks after surgeryddPCR EGFR6% for Group C.Group A: baseline ctDNA− (n=211); Group B: baseline ctDNA+ but MRD (n=51); Group C: baseline ctDNA+ and MRD+ (n=16). The 3-year DFS rates differed significantly among the three groups: 84% for group A, 78% for group B and 50% for group C, with group C showing an HR of 3.28 (p=0.02).AT, adjuvant therapy; ctDNA, circulating tumour DNA; DFS, disease-free survival; EGFR, epidermal growth factor receptor; MRD, molecular residual disease; mut, mutations; NGS, next generation sequencing; NR, not reported; NSCLC, non-small cell lung cancer; OS, overall survival; RFS, recurrence free survival.Considering escalation strategies in stage I disease, Wang et al (n=270) assessed the role of ctDNA detection in estimating the prognosis and guiding AC for resectable stage I NSCLC (table 2).47 Only nine patients (3.3%) had ctDNA+, and the risk of recurrence was significantly lower in the ctDNA− group compared with the ctDNA+ group (HR 0.11, p<0.0001). In addition, in the ctDNA+ group, there were no significant differences in RFS between patients who received AC and patients who did not receive AC (p=0.58), and in the ctDNA− group, those who received AC had a worse RFS in comparison with those who did not receive AC (HR 2.36, p=0.029).47 This study suggests that postoperative ctDNA may serve as a prognostic marker for recurrence but has limited ability to guide AT in resected stage I NSCLC. However, prospective studies with larger cohorts are needed to validate this approach, given the small number of ctDNA+ patients.Incorporating emerging technologies with lower limits of detection into postoperative ctDNA detection, Isbell et al analysed longitudinal ctDNA in early-stage NSCLC using data from the TRACERx study, using a tumour-informed SNV-based assay (CAPP-Seq) and PhasED-Seq on 269 samples from 46 patients. Twelve cases were MRD+ by PhasED-Seq, all of whom recurred (100%). In contrast, six cases were MRD+ by the CAPP-Seq, of whom five (83%) had tumour recurrence. Accordingly, PhasED-Seq had a higher clinical sensitivity than CAPP-Seq (12/18 (67%) vs 5/18 (28%), p=0.022). Notably, MRD+ patients identified by PhasED-Seq who received AT demonstrated significantly improved outcomes compared with those who did not (HR 8.2, p=0.00035). This therapeutic benefit was not observed when MRD was assessed using the SNV-based assay CAPP-Seq. Furthermore, with PhasED-Seq, 80% (4/5) of MRD+ patients who underwent AT achieved MRD clearance, whereas none (0/3) of the patients without AT cleared their MRD.51 Overall, this study shows that ultrasensitive techniques for MRD detection could accurately identify patients who derive meaningful benefit from AT.With regard to adjuvant IT, the phase III IMpower010 study (n=1005) evaluated adjuvant atezolizumab versus placebo in patients with resected stage IB–IIIA NSCLC following CT.52 A subset of 600 patients (stage II–IIIA, n=534) was evaluable for ctDNA analysis using Natera’s tumour-informed Signatera assay (table 2).53 Postsurgical ctDNA positivity was observed in 21% (112/534) of patients and increased with disease stage (IB: 9%, II: 14% and IIIA: 29%).53 In the 5-year DFS update, ctDNA+ patients with stage II–IIIA disease had a worse prognosis (5-year DFS: atezolizumab—median not reached (NR) vs 52.6 months for placebo; HR 0.60), compared with ctDNA− patients (NR vs 60.6 months; HR 0.74).54 Nevertheless, atezolizumab improved DFS over placebo in both ctDNA+ and ctDNA− subgroups, with a greater benefit observed in patients with PD-L1 expression ≥1% (HR 0.54 for ctDNA+ and HR 0.57 for ctDNA−) compared with those with PD-L1 <1% (HR 0.88 and HR 0.95, respectively).53 54Building on this concept, the phase III NADIM-ADJUVANT study (n=210) assessed the role of adjuvant CT versus CT plus nivolumab followed by nivolumab maintenance for 6 months (every 4 weeks) in patients with completely resected stage IB–IIIA NSCLC.55 At the interim analysis with 57% maturity, the cancer-specific DFS (csDFS) was 77.9% for the treatment arm versus 62.5% with CT alone (HR 0.54, p=0.02), and there was a meaningful reduction in the relapse rates of 20.4% in the CT-IT arm when compared with 38.8% with CT alone. Regarding the MRD status, patients with MRD− had higher median DFS with CT-IT versus CT alone (NR vs NR, HR at 3 years=0.31, p=0.04). In addition, in the treatment arm, MRD negativity was also associated with improved csDFS (HR 5.7, 95% CI 1.0 to 31.2, p=0.04).55Within the subset of patients harbouring AGA, MRD has been investigated to assess recurrence risk and guide AT. In the phase III ADAURA study (n=682), osimertinib significantly improved DFS in resected stage IB–IIIA EGFR-mutant NSCLC.56 57 A post hoc analysis of 220 patients (112 osimertinib and 108 placebo) using a tumour-informed customised assay, RaDaR, showed that MRD detection preceded radiographic recurrence by a median of 4.7 months.58 Among MRD− patients, 36-month DFS was 86% with osimertinib versus 36% with placebo (HR 0.23). MRD positivity at any time occurred in 25% (n=28) of osimertinib-treated patients versus 69% with placebo.58 Most MRD events in the osimertinib arm occurred after treatment completion (68%, 19/28), with 58% (11/19) arising within 12 months.58 At 24 months post-treatment, 66% of patients in the osimertinib group remained MRD event-free.58 These findings suggest that MRD monitoring may help identify patients who could benefit from extended osimertinib therapy, although further clinical validation is warranted.ctDNA detection after radical chemoradiotherapy The standard treatment for locally advanced unresectable NSCLC is concurrent chemoradiotherapy (CRT), followed by consolidation with durvalumab (PACIFIC study) 59 or osimertinib in patients harbouring EGFR exon 19 deletions or L858R mutations (LAURA study).60 In a retrospective study (n=65), Moding et al investigated whether ctDNA could identify patients who may benefit from IT and monitor treatment response after CRT, using CAPP-Seq NGS.61 Plasma samples were collected before treatment and within 4 months post-CRT, prior to initiating IT. Thirteen patients (20%) were excluded due to the absence of detectable tumour variants or unavailable tissue.61 Patients with undetectable ctDNA after CRT had excellent outcomes, regardless of IT use, with a 2-year freedom from progression (FFP) rate of 100% compared with 0% in ctDNA+ patients (p=0.0006). ctDNA detection preceded radiographic progression by a mean of 4.1 months.61 Additionally, patients who showed a reduction in ctDNA during IT experienced significantly improved outcomes, with a 1-year FFP of 100% (p=0.003).61 Similarly, in a prospective study by Pan et al (n=139), 27.3% of patients (38/139) had early undetectable ctDNA during or after radiotherapy and demonstrated better survival outcomes, with a 2-year PFS rate of 88.4%, regardless of whether they received consolidation IT.62 These findings highlight ctDNA as a valuable biomarker for identifying patients who may benefit from consolidation IT and monitoring treatment response to CRT.Limitations and future directions of ctDNA in early-stage NSCLC MRD has high evidence of clinical validity in anticipating future relapse in many cancer types. 15 However, molecular relapse detection cannot be recommended in routine clinical practice, as currently there is no evidence for clinical utility in directing treatment.16 63 Key challenges for ctDNA implementation in early-stage lung cancer include (1) the risk of false-negative results due to low or absent ctDNA shedding, particularly in small-volume or indolent tumours; (2) the lack of assay standardisation across platforms with varying sensitivity, specificity and turnaround times (eg, ddPCR, NGS and PhasED-Seq) and (3) the limited prospective data demonstrating that ctDNA-guided therapeutic decisions improve survival outcomes. Although the prognostic value of postoperative ctDNA is well established, evidence supporting ctDNA-directed interventions remains largely exploratory, and current consensus statements from the European Society for Medical Oncology and the International Association for the Study of Lung Cancer underscore the need for randomised trials to define its clinical utility.16 Cautious interpretation of published findings is therefore warranted.64Advanced stage NSCLC Assessment of ctDNA in molecular response In advanced NSCLC, IT and TT for driver mutations have improved PFS and quality of life. 65 However, resistance inevitably develops, making timely detection crucial to guide therapy. Tissue biopsies are invasive, often impractical and may miss tumour heterogeneity.63 In contrast, ctDNA offers a dynamic, comprehensive view of tumour evolution, reflecting the genomic profile of multiple lesions and subclones.66As mentioned, the most common metric for mutations in ctDNA is the VAF, especially when comparing baseline VAF to early on-treatment levels to assess response and predict outcomes.67 Molecular response (MR) can be measured via (1) ctDNA clearance, (2) delta VAF (dVAF) and (3) VAF ratio.67ctDNA clearance is binary—detectable or not after therapy—but in some cases misses meaningful ctDNA reductions without full clearance.67 In a study by Zhang et al (n=978), the prognostic and predictive value of ctDNA was evaluated in patients with advanced cancer treated with immune checkpoint inhibitors (ICIs).68 The study found that higher (≥median) pretreatment VAF was associated with poorer OS (HR 0.58, p<0.001) but not with objective response rate (ORR), suggesting a primarily prognostic role for pretreatment VAF in patient outcomes.68 Similarly, Jee et al analysed OS in an advanced NSCLC cohort of patients (n=1127) receiving ctDNA-guided therapy. The authors found that patients with detectable ctDNA and max VAF ≥median had significantly shorter OS compared with ctDNA− patients (n=358), with a modified OS (mOS) of 10 versus 32 months, respectively (HR 2.4).20 This reinforces the prognostic significance of ctDNA burden in advanced disease beyond simple binary clearance metrics.The dVAF measures the change in ctDNA levels after treatment initiation, typically calculated by subtracting the mean baseline VAF of all mutations from the mean on-treatment VAF.67 In the abovementioned study by Zhang et al, reductions in on-treatment VAF and lower on-treatment VAF values were independently associated with longer PFS and OS, as well as higher ORR (p<0.0001), suggesting that ctDNA dynamics could be predictive of benefit from ICI therapy.68Ultimately, the VAF ratio method compares on-treatment ctDNA levels to baseline to assess response, considering both the change and the remaining ctDNA. Using the mean VAF instead of the maximum VAF helps reduce bias from sampling variability, gene amplification or CH. For example, if the VAF is 10% at baseline and 5% during treatment, the ratio would be 0.5.67ctDNA in MR to TTs EGFR mutations are present in 15%–20% of advanced NSCLC and around 60% of patients who never smoked.69 Currently, the first-line treatment for patients with EGFR exon 19 or 21-positive NSCLC includes monotherapy or a combination of third-generation EGFR inhibitors with CT or EGFR/mesenchymal epithelial transition bispecific antibodies.70–72 In an exploratory analysis of the FLAURA study, which compared osimertinib to standard of care (gefitinib or erlotinib), patients who achieved ctDNA clearance at 3 and 6 weeks after starting EGFR-tyrosine kinase inhibitor (TKI) therapy had numerically longer PFS than those without ctDNA clearance (median PFS: 13.5 vs 9.5 months at 3 weeks and 13.5 vs 8.3 months at 6 weeks; table 3).73Table 3Effect of ctDNA status post-treatment in patients with advanced NSCLC and AGAAGAStudy (n)Panel usedEffect of ctDNA status post-treatmentEGFRFLAURA (n=556) 1L osimertinib versus SOC gefitinib/erlotinibddPCRPatients who achieved ctDNA clearance at 3 and 6 weeks had numerically longer PFS than those without ctDNA clearance (mPFS: 13.5 m vs 9.5 m at 3 w, and 13.5 m vs 8.3 m at 6 w)EGFRFLAURA-2 (n=557) 1L osimertinib±chemotherapyddPCROf patients randomised to osimertinib alone, patients with non-detected ctDNA had longer mPFS versus those with detectable plasma ctDNA negative (mPFS 13.9 m vs 30.3 m)EGFRMARIPOSA (n=1074) 1L amivantamab+lazertinib versus osimertinibddPCRIn both arms, 85% of patients cleared ctDNA at C3D1. Amivantamab+lazertinib significantly improved mPFS versus osimertinib in patients without ctDNA clearance at C3D1 (16.5 m vs 9.1 m; HR 0.49, p=0.01)ALKCROWN (n=296), 1L lorlatinib versus crizotinib74-gene ctDNA NGSPatients treated with lorlatinib who had a MR at 4 w had longer PFS compared with non-responders (12 m PFS 89% vs 56% HR=0.37,0.85). Patients treated with crizotinib who had a MR had similar PFS as those without MR.ALKALEX (n=303), 1L alectinib versus crizotinibFoundationACTmPFS was longer with alectinib than crizotinib in both ≤median and >median ctDNA biomarker-evaluable population (not estimable vs 14.8 m, HR 0.38, p<0.0001)KRASPhase II KRYSTAL-1 study (n=60) adagrasib in pretreated KRAS G12Cmut NSCLC (cohort A)ddPCRPatients with complete ctDNA clearance at cycle 2 had higher objective response rate (60.6% vs 33.3%) compared with incomplete ctDNA clearance. Patients with complete ctDNA clearance at cycle 4 correlated with longer overall survival (14.7 vs 5.4 m, HR-0.1, p<0.001) and improved PFS (9.8 m vs 4.3 m, HR 0.3, p=0.01)KRASPhase II CodeBreak 100 (n=124) and phase III CodeBreak 200 (n=345). Sotorasib versus docetaxel in pretreated NSCLC KRAS G12CmutNGSctDNA clearance occurred sooner with sotorasib versus docetaxel (43% vs 14% at cycle 2). In patients receiving sotorasib, ctDNA clearance at cycle 1/day 8 was associated with improved PFS (7.26 m vs 4.01 m, p=0.006).AGA, actionable genomic alteration; ctDNA, circulating tumour DNA; ddPCR, digital droplet PCR; m, months; mPFS, modified PFS; MR, molecular response; MR, molecular response; NGS, next generation sequencing; NSCLC, non-small cell lung cancer; PFS, progression free survival; w, weeks.Analogously, ALK rearrangements occur in 3%–5% of advanced NSCLC. Newer-generation ALK inhibitors provide superior PFS and intracranial activity in patients with brain metastasis over crizotinib, which establishes them as the preferred first-line therapy.74 LB enables detection, resistance mutations and dynamic monitoring of ALK-TKI therapy. Kwon et al (n=92) evaluated longitudinal ctDNA dynamics in a cohort of patients with ALK-positive NSCLC using Guardant360.75 The study found that absence of ctDNA at baseline was significantly associated with longer median PFS (36.1 vs 11.4 months, p=0.004) and OS (NR vs 29.3 months, p=0.02). Additionally, patients who achieved ctDNA clearance at 2 months (n=29) had significantly longer PFS (25.4 vs 11.6 months, p=0.001) and OS (NR vs 26.1 months, p=0.03) compared with those without clearance.75 Similarly, an exploratory analysis from the phase III CROWN study (n=296), which compared first-line lorlatinib versus crizotinib in ALK-positive advanced NSCLC, showed that a reduction in VAF at 4 weeks was associated with a Response Evaluation Criteria in Solid Tumours (RECIST) response in both treatment arms.76 Moreover, patients with a decrease in mean VAF at week 4 had a confirmed ORR of 75.8% with lorlatinib and 53.3% with crizotinib. In the lorlatinib arm, the HR for dVAF ≤0 versus >0 was 0.50 (95% CI 0.23 to 1.12; mPFS NR in either group) (table 3).76Regarding subsequent lines of treatment for ALK-positive NSCLC, Shaw et al (n=121) showed that early dynamics of ctDNA predict lorlatinib efficacy in pretreated patients. Patients with a reduction in VAF (dVAF <0) showed greater tumour response (26% vs 12%, p=0.04) and had significantly lower dVAF if they achieved complete response or partial response, compared with stable disease or progressive disease (PD). The mPFS was also longer in the dVAF <0 group (6.6 vs 2.6 months; HR 2.6), as was mOS (18.0 vs 8.6 months; HR 2.0), indicating better outcomes with early ctDNA reduction.The KRAS G12C mutation has recently emerged as a key AGA in NSCLC, with sotorasib as the first approved targeted inhibitor demonstrating clinical efficacy. Passiglia et al evaluated the correlation between ctDNA dynamics, measured using a customised NGS panel and treatment response in NSCLC patients harbouring KRAS G12 mutations (n=32) receiving sotorasib in a real-world setting.77 The study showed that patients who achieved clearance of ctDNA had a significant improvement in the ORR (80% vs 8%; p<0.001), mPFS (7.9 vs 2.8 months; p<0.001) and mOS (16.8 vs 6.4 months; p<0.001) compared with those who did not achieve ctDNA clearance.77 Moreover, an increase in mVAF anticipated radiologic PD in 70% of patients who were evaluable at the resistance time point.77 Similarly, Paweletz et al reported consistent results in patients receiving adagrasib in the KRYSTAL-1 study. Using ddPCR, 89.7% of patients (35/39) showed >90% reduction in KRAS G12C ctDNA, and 84.6% (33/39) achieved complete ctDNA clearance by cycle 2. Complete ctDNA clearance at cycle 2 was associated with a higher ORR (60.6% vs 33.3%), while clearance at cycle 4 correlated with longer OS (14.7 vs 5.4 months) and improved PFS (HR 0.3) (table 3).78RET rearrangements occur in around 3% of NSCLC, and RET inhibitors are the recommended first-line therapy. In the phase I/II LIBRETTO-001 study (n=247) evaluating selpercatinib in RET-driven cancers, the ctDNA analysis in 34 patients showed a median 96% reduction in RET allele frequency by day 15, with complete clearance in 44%. Clearance was more frequent in RET fusion (67%) versus RET-mutant (8%) patients.79 Along the same lines, MET alterations are primary oncogenic drivers that occur in 3%–4% of NSCLC. The phase II VISION study (n=152), which evaluated tepotinib in patients with confirmed MET exon 14 skipping mutations, 51 patients had available data for ctDNA. Among them, 34 (67%) showed an MR, and 24 (71%) achieved a RECIST response per independent review committee.80 Altogether, these findings correlate with ctDNA dynamics as a robust biomarker for assessing treatment response in advanced NSCLC harbouring genomic alterations.ctDNA in MR to IT The current predictive markers, such as PD-L1 or tumour mutation burden, are inconsistent when predicting therapeutic response. This has led to an urgent need to implement MR-driven approaches to interpret outcomes, guide treatment selection and future escalation strategies. The BR.36 phase II adaptive study (n=50) evaluated the MR to single-agent pembrolizumab in EGFR- and ALK-negative NSCLC using a tumour-agnostic NGS approach.81 The study demonstrated a ctDNA response sensitivity of 82% and specificity of 75% for RECIST, suggesting that MR is largely concordant with radiological responses. CH was confirmed in 17% of plasma samples, with 5% representing germline mutations and 78% tumour-derived alterations. The best overall RECIST response rate was 32%, significantly lower than the presumed 45% (p=0.04), with a median radiological response duration of 10.1 months. The median time to ctDNA response was 2.1 months.81 Patients with undetectable ctDNA at baseline (n=10) had longer PFS and OS compared with those with detectable ctDNA (n=35), with PFS of 8.3 versus 2.9 months and OS of 16.8 versus 10.9 months, respectively. Furthermore, patients with MR had longer PFS (5.03 vs 2.6 months) and OS (NR vs 7.23 months) compared with those with molecular PD.81 These findings, consistent with those of other investigators, support ctDNA-based MR as a valuable early marker of IT efficacy in NSCLC.68Regarding correlation with tissue biomarkers and ctDNA analysis with IT, Boscolo et al performed a prospective study (n=113) in patients with advanced NSCLC treated with single agent ICIs and correlated clinopathological features with ctDNA monitoring at baseline (T1), 3 weeks after (T2) and at the time of radiological evaluation (T3). Interestingly, among patients with high PD-L1 expression, elevated T2 max VAF and ctDNA increase between T2 and T1 correlated with worse PFS as well as worse OS.82Conclusions LB offers a promising tool for early risk stratification and minimising overtreatment in patients showing MR. It holds strong potential for guiding timely clinical decisions and accelerating therapeutic evaluation in trials. However, further validation is needed to confirm its predictive value before routine clinical use.