Cadonilimab plus chemotherapy as first-line treatment in PD-L1-negative advanced non-small cell lung cancer: a phase II clinical trial

Abstract

Although immunotherapy is approved for patients with high PD-L1 expression, optimal therapeutic strategies for PD-L1-negative populations remain undefined. This study (ChiCTR2300071681) assessed the efficacy and safety of cadonilimab (PD-1/CTLA-4 bispecific antibody) plus chemotherapy in patients with PD-L1-negative advanced non-small-cell lung cancer (NSCLC). The primary endpoint, 12-month progression-free survival (PFS) rate, is 42.1% (95% CI, 29.6%–60.0%), which has reached the prespecified threshold. Secondary endpoints include a median overall survival of not reached, a median PFS of 9.7 months, an objective response rate of 66.0%, a disease control rate of 100.0%, and a median duration of response of 9.5 months. Grade ≥3 treatment-related adverse events occur in 26 (52.0%) patients. cfDNA methylation-based molecular response predicts the actual clinical response approximately 5 cycles earlier than conventional radiographic evaluation. Baseline differentially methylated fragments scores show a significant correlation with PFS, with low-risk patients demonstrating a longer median PFS compared to high-risk patients (11.4 months versus 6.9 months). Overall, first-line cadonilimab plus chemotherapy shows an encouraging efficacy with a manageable safety profile for challenging-to-treat PD-L1-negative advanced NSCLC, warranting further evaluation in controlled studies.

Therapeutic options for PD-L1–negative non–small-cell lung cancer (NSCLC) remain limited despite advances in immunotherapy. This study reports the efficacy and safety outcomes of a phase II trial evaluating cadonilimab plus chemotherapy in patients with PD-L1– negative advanced NSCLC.

Introduction

Lung cancer remains the leading cause of cancer-related deaths globally¹. Immune checkpoint inhibitors (ICIs) have significantly improved outcomes in non-small-cell lung cancer (NSCLC) with high programmed cell death-ligand 1 (PD-L1) expression; however, patients with PD-L1-negative advanced or metastatic NSCLC, which account for up to 48% of cases^2,3, consistently derive limited benefit from current ICI-based therapies^4–6. Despite the large proportion of such patients, this population has not been the primary focus of most prospective clinical trials. Existing evidence is largely derived from subgroup analyses. In the KEYNOTE-189 and KEYNOTE-407 randomized studies, pembrolizumab combined with chemotherapy demonstrated an objective response rate (ORR) of 50.6%, a median progression-free survival (PFS) of 6.5 months, and a median overall survival (OS) of 18.3 months in PD-L1-negative patients⁷. There is an urgent need for more effective and tailored therapies in this population.

Dual immune checkpoint blockade targeting programmed cell death 1 (PD-1) and cytotoxic T-lymphocyte antigen 4 (CTLA-4) represents a promising strategy for low-immunogenic tumors. PD-1 blockade reinvigorates exhausted T cells in the tumor microenvironment (TME), while CTLA-4 inhibition enhances T-cell priming in lymphoid tissues^8–12. The pooled analyses from CheckMate 227 and CheckMate 9LA demonstrated superior efficacy of nivolumab plus ipilimumab compared to chemotherapy alone among PD-L1-negative NSCLC patients (ORR: 29% versus 22%; median PFS: 5.4 months versus 4.9 months)^13,14. The meta-analyses further suggested that nivolumab plus ipilimumab may have an OS advantage over pembrolizumab combined with chemotherapy in PD-L1-negative NSCLC patients¹⁵. Despite encouraging efficacy, dual checkpoint blockade is linked to an increased incidence of toxicity. In the phase III JCOG2007 trial, treatment-related deaths were reported in 7% of patients who received platinum-based chemotherapy (two cycles) plus nivolumab-ipilimumab¹⁶. In this context, bispecific antibodies represent a rational approach to cancer treatment, providing enhanced immunotherapy efficacy while reducing treatment-related toxic effects.

Cadonilimab (AK104), a bispecific antibody targeting PD-1 and CTLA-4, is capable of reducing Fc-mediated toxicity while enhancing dual checkpoint inhibition¹⁷. It has been approved for marketing in China in June 2022 for the treatment of relapsed/metastatic cervical cancer patients who progressed on or after platinum-based chemotherapy. Furthermore, it has demonstrated efficacy in advanced gastric or gastroesophageal junction adenocarcinoma, regardless of PD-L1 status¹⁸. Early-phase data in advanced NSCLC revealed that cadonilimab, particularly at a dose of 10 mg/kg every three weeks (Q3W), showed encouraging efficacy and a manageable safety profile when in combination with antiangiogenic agents¹⁹ supporting this dosing regimen for future clinical trials.

In this multicenter, single-arm phase II study (LungCadX), we aimed to evaluate the efficacy and safety of first-line cadonilimab (10 mg/kg Q3W) combined with platinum-based chemotherapy in patients with advanced PD-L1-negative NSCLC. In parallel, circulating cell-free DNA (cfDNA) methylation profiling as a noninvasive biomarker for early response assessment and long-term prognosis was investigated to establish a precision immunotherapy strategy for this underserved population.

Results

Patients and treatment

Between July 2023 and August 2024, 50 of 55 screened patients from three centers in China were enrolled, with 3 patients not meeting the inclusion criteria and 2 declining to participate. All enrolled patients received cadonilimab plus chemotherapy and were included in the safety analysis. Of these, 3 patients were removed from the FAS due to discontinued follow-up after one cycle and lack of efficacy assessment (Fig. 1). At the cut-off date (November 11, 2025), 8 of 47 patients were still receiving treatment, while 39 discontinued treatment due to disease progression (PD; n = 25), deaths (n = 7), adverse events (AEs; n = 3), and alternative therapy (n = 4).

Fig. 1. Study flow chart.

A total of 55 patients provided informed consent. 5 patients were excluded before treatmentinitiation because they did not meet the eligibility criteria or withdrew informed consent. 50 patients received atleast one dose of study treatment and comprised the safety set. 3 patients voluntarily withdrew after receiving onetreatment and were excluded from the effi cacy analysis. The full analysis set included 47 patients. At the data cutoff, 39 patients had discontinued treatment because of progressive disease, death, adverse events, or alternativetherapy, whereas 8 patients remained on treatment.

In the FAS cohort, the median age was 63.4 (range: 35–78; Table 1) years, and the majority were male (43 [91.5%]). Most had stage IV disease (38 [80.9%]), and 23 (48.9%) patients had squamous cell carcinoma. Eighteen patients (38.3%) had metastatic disease, including 4 (8.5%) with brain metastases, 5 (10.6%) with liver metastases, and 9 (19.1%) with bone metastases.

Table 1.

Baseline characteristics of patients

Characteristic	Patients (N = 47)
Age (years), median (range)	63.4 (35–78)
≥65	23 (48.9)
<65	24 (51.1)
Gender, n (%)
Male	43 (91.5)
Female	4 (8.5)
Smoking status, n (%)
Current/former	39 (83.0)
Never	8 (17.0)
Clinical stage, n (%)
IIIB/IIIC	9 (19.1)
IV	38 (80.9)
Pathological subtypes, n (%)
Squamous cell carcinoma	23 (48.9)
Non-squamous cell carcinoma	24 (51.1)
Brain metastasis, n (%)
Yes	4 (8.5)
No	43 (91.5)
Liver metastasis, n (%)
Yes	5 (10.6)
No	42 (89.4)
Bone metastasis, n (%)
Yes	9 (19.1)
No	38 (80.9)

Efficacy

As of the data cut-off date (November 11, 2025), after the median follow-up time was 12.9 (95% interquartile range [IQR]: 9.0–16.6) months, the median PFS was 9.7 (95% CI: 7.3–13.8) months, with the PFS rate at 12 months of 42.1% (95% CI: 29.6%–60.0%) (Fig. 2a). Subgroup analysis showed that patients with squamous cell carcinoma had a median PFS of 12.9 (95% CI: 7.1–not estimable [NE]) months, whereas those with non-squamous cell carcinoma had a median PFS of 8.9 (7.2–17.6) months; the corresponding 12-month PFS rates were 54.2% (95% CI: 36.8%–79.8%) and 30.0% (15.5%–58.0%), respectively (Fig. 2b).

Fig. 2. Kaplan–Meier survival curves.

a PFS in all PD-L1 negative NSCLC patients treated with cadonilimab plus chemotherapy. b PFS in squamous cell carcinoma and non-squamous cell carcinoma. Source data are provided as a Source Data file. PFS: progression-free survival; PD-L1: Programmed cell death 1 ligand 1; NSCLC: non-small cell lung cancer; CI: confidence interval.

For the secondary endpoints, median OS was not reached (NR; 95% CI: NR–NR) overall and in both squamous (95% CI: NR–NR) and non-squamous cell carcinoma (95% CI: 9.7–NR) subgroups. The ORR was 66.0% (95% CI: 50.7–79.1%), with a higher ORR observed in squamous cell carcinoma (87.0%; 95% CI: 66.4–97.2%) compared to non-squamous cell carcinoma (45.8%; 95% CI: 25.6–67.2%; Supplementary Table 1 and Supplementary Fig. 1). Most patients showed disease control and a reduction in tumor size (Fig. 3a and b). The DCR was 100% (95% CI: 92.5–100%) (Fig. 3c), and the median DoR was 9.5 (95% CI: 6.0–NE) months.

Fig. 3. Anti-tumor activity and treatment duration.

a Waterfall plot of best percentage change from baseline of sum of target lesion diameters. b Spider plot of percentage change from baseline of sum of target lesion diameters. c Swimmer plot of treatment duration and best overall response. Source data are provided as a Source Data file. PR: partial response; SD: stable disease.

Safety

TRAEs of any grade occurred in 46 (92.0%) patients (Table 2), with neutrophil count decreased (28/50, 56.0%), anemia (28/50, 56.0%), white blood cell decreased (26/50, 52.0%), alanine aminotransferase increased (16/50, 32.0%), and platelet count decreased (16/50, 32.0%) being the most common. Grade ≥ 3 TRAEs were observed in 26 (52.0%) patients, most commonly neutrophil count decreased (18/50, 36.0%), pneumonitis (4/50, 8.0%), platelet count decreased (4/50, 8.0%), and serum amylase increased (3/50, 6.0%). The rate of treatment discontinuation due to TRAEs was 10.0% (5/50). No deaths attributed to TRAEs were reported. Immune-related adverse events (irAEs) of any grade occurred in 18 patients (36.0%). Grade ≥3 irAEs were observed in 6 (12.0%) patients, including alanine aminotransferase increased (n = 2), infusion-related reaction (n = 2), immune-mediated pneumonitis (n = 1), and blood creatinine increased (n = 1). All AEs are presented in Table 2.

Table 2.

Summary of adverse events (AEs) and frequent treatment-related adverse events (incidences ≥10%)

	Overall (N = 50)
AEs, n (%)	47 (94.0)
TRAEs, n (%)	46 (92.0)
≥Grade3 AEs, n (%)	27 (54.0)
≥Grade 3 TRAEs, n (%)	26 (52.0)
Serious AEs, n (%)	10 (20.0)
Serious TRAEs, n (%)	10 (20.0)
TRAEs leading to interruption of any drug, n (%)	15 (30.0)
TRAEs leading to discontinuation of any drug, n (%)	5 (10.0)
AEs leading to death, n (%)	0
TRAEs leading to death, n (%)	0
irAEs, n (%)	18 (36.0)
≥Grade 3 irAEs, n (%)	6 (12.0)
irAEs leading to interruption of any drug, n (%)	8 (16.0)
irAEs leading to discontinuation of any drug, n (%)	4 (8.0)
irAEs leading to death, n (%)	0

	Overall (N = 50)
Frequent TRAEs (incidences ≥10%), n (%)	All Grade	>= Grade 3
Neutrophil count decreased	28 (56.0)	18 (36.0)
Anaemia	28 (56.0)	0
White blood cell count decreased	26 (52.0)	2 (4.0)
Alanine aminotransferase increased	16 (32.0)	2 (4.0)
Platelet count decreased	16 (32.0)	4 (8.0)
Aspartate aminotransferase increased	14 (28.0)	2 (4.0)
Blood bilirubin increased	9 (18.0)	0
Infusion-related reaction	7 (14.0)	2 (4.0)
Serum amylase increased	7 (14.0)	3 (6.0)
Blood creatinine increased	6 (12.0)	1 (2.0)
Gamma-glutamyltransferase increased	6 (12.0)	1 (2.0)
Pneumonitis	5 (10.0)	4 (8.0)

AE adverse event, TEAE treat-emergent adverse events, TRAE treatment-related adverse events; irAE immune-related adverse events.

TRAE treatment-related adverse events.

cfDNA methylation analysis

cfDNA methylation profiling was performed in 47 patients, identifying 27,333 differentially methylated fragment regions (DMFRs) primarily in intergenic, intronic, and promoter regions, with over half within CpG islands (Fig. 4a). Changes in differentially methylated fragment (DMF) scores from baseline to C3D1 reflected distinct methylation patterns between responders and non-responders, with 25 patients showing decreased scores (Fig. 4b). We further evaluated the correlation between DMF score-based molecular response and imaging-based clinical response. Clinical responses increased over time (PR in 17, 24, and 22 patients at C3D1, C5D1, and C8D1, respectively), with a significant association emerging at C8D1 (P = 0.039) (Fig. 4c), driven by patients who showed early molecular but delayed clinical responses. At C3D1, responders (per C8D1 imaging) had significantly lower DMF scores than at baseline (P = 0.006), while non-responders showed no significant change (P = 0.134) (Fig. 4d), indicating that molecular response preceded radiographic response by approximately five cycles (∼105 days).

Fig. 4. Evaluation of cfDNA methylation dynamics and association with treatment response and prognosis.

a Distribution of DMFRs identified from blood samples collected at C1D1 and C3D1. DMFRs are reported across genomic regions (intergenic, intronic, promoter) and CpG islands. b Distribution of changes in DMF scores between C1D1 and C3D1 among patients, compared with clinical response and patients’ characteristics. c Longitudinal analysis of PR rates by imaging across treatment time points. The correlation between molecular response (DMF score) and imaging-based response at C3D1, C5D1, and C8D1 was shown. d Box plot of DMF scores. DMF scores were calculated from cfDNA samples of 41 patients with available clinical response data (imaging evaluated at C8D1). The cfDNA samples originated from distinct biological replicates; each point represents an independent biological measurement. The x‑axis shows the two timepoints (C1D1 and C3D1), and the y‑axis shows the log2‑transformed DMF scores. Group-wise comparisons were performed between C1D1 and C3D1 timepoints by two-sided Wilcoxon rank-sum test. No adjustments for multiple comparisons were applied. In clinical responders, samples size N (C1D1) = 23 and N (C3D1) = 23, Wilcoxon test statistic W = 389, the 95% CI = [0.52, 6.32], effect size = 0.40, P-value = 0.006. In clinical non-responders, samples size N (C1D1) = 18 and N (C3D1) = 18, Wilcoxon test statistic W = 210, the 95% CI = [−0.24, 1.91], effect size = 0.25, P-value = 0.134. e Kaplan–Meier analysis of PFS based on baseline molecular risk, defined by DMF score at C1D1. f Forest plot of multivariable Cox regression. HRs and 95% CI are shown for each clinical and molecular factor. The solid square indicates the HR point estimate, and the horizontal line represents the 95% CI. The vertical dashed line at HR = 1 denotes no effect. Only the Molecular Risk Group was significantly associated with survival (HR = 0.419, 95% CI: 0.182–0.963, P-value = 0.0405). For the global model: log-rank P-value = 0.24466, C-index = 0.72, AIC = 209.9 and events = 32.

Associations between molecular response and clinical features (pathology and clinical stage) were also assessed; however, no significant correlations were observed (Supplementary Fig. 2). This analysis revealed that molecular response at C3D1 was not predictive of better PFS (Supplementary Fig. 3). However, baseline molecular risk, defined by a C1D1 DMF score cutoff of 10 (median DMF score), was prognostic for PFS: median PFS was 6.9 months in the high-risk group vs. 11.4 months in the low-risk group (P = 0.032) (Fig. 4e). Multivariate Cox analysis confirmed low molecular risk as an independent protective factor (Hazard ratios [HRs] = 0.419; 95% CI: 0.182–0.963; P = 0.041), outperforming conventional clinical variables (Fig. 4f).

Discussion

Patients with PD-L1-negative NSCLC respond poorly to conventional immunotherapy, highlighting an urgent need for effective therapeutic strategies. The addition of anti-PD-1 antibodies to CTLA-4 inhibitors may enhance immune activation in poorly immunogenic tumors, offering a rational approach that has generated growing interest in dual immune checkpoint blockade strategies, especially in PD-L1-negative NSCLC.

This clinical study evaluated a PD-1/CTLA-4 bispecific antibody plus chemotherapy in patients with PD-L1-negative advanced NSCLC in the first-line setting. Within the current therapeutic landscape, where treatment for this population is primarily defined by PD-1/PD-L1 inhibitors combined with chemotherapy, cadonilimab combined with chemotherapy exhibited encouraging efficacy, achieving a 12-month PFS rate of 42.1%, meeting the primary endpoint. ORR was 66.0%, and median PFS was 9.7 months. Although comparison between trials should be viewed cautiously, these results were numerically higher than historical data reported from a pooled analysis of KEYNOTE-189 and KEYNOTE-407 studies in patients with previously untreated PD-L1-negative advanced NSCLC, where pembrolizumab plus chemotherapy reported a 12-month PFS rate of 30.1%, an ORR of 50.6%, and a median PFS of 6.5 months⁷. Notably, even when compared with nivolumab plus ipilimumab with 2 cycles of chemotherapy that demonstrated survival benefit in CheckMate 9LA (ORR: 31.1%; median PFS: 5.8 months; 12-month PFS rate: 31%), cadonilimab combined with four cycles of chemotherapy remains encouraging with numerically comparable or better outcomes, thereby warranting longer follow-up and larger-scale randomized studies to evaluate its potential to confer durable survival¹². Additionally, an early clinical study supported the strategy of combining PD-1/CTLA-4 bispecific antibody with chemotherapy. MEDI5752 (a PD-1/CTLA-4 bispecific antibody) plus chemotherapy demonstrated promising efficacy in NSCLC. Among PD-L1-negative patients, the ORR was 48% with the 750 mg Q3W (n = 35) and 55.6% with the 1500 mg Q3W (n = 9), with a PFS of 13.4 months in the higher-dose group²⁰. Collectively, these findings provide preliminary evidence that cadonilimab in combination with chemotherapy could serve as a potential first-line approach for PD-L1-negative advanced NSCLC, meriting further clinical investigation.

Subgroup analysis in our study showed that outcomes in patients with squamous cell carcinoma, with an ORR of 87.0%, a median PFS of 12.9 months, and a 12-month PFS rate of 54.2%, were superior to those reported in the KEYNOTE-407 trial (ORR: 67.4%; median PFS: 6.3 months) and the CheckMate 227 and 9LA trials (ORR: 42.0%; median PFS: 5.3 months)^4,14. This observation holds particular clinical significance for this NSCLC subgroup, given that patients with squamous cell carcinoma generally have a poor prognosis and derive limited long-term benefits from chemotherapy or PD-(L)1–based regimens⁷. Our data support that the first-line PD-1/CTLA-4 bispecific antibody treatment regimen may be a suitable option in PD-L1-negative squamous cell carcinoma patients.

All AEs were generally manageable with standard supportive care, indicating that cadonilimab plus chemotherapy may have a manageable safety profile. The most common TRAE was neutrophil count decreased. Grade 3–4 TRAEs occurred in 52.0% of patients, which was slightly higher than that reported for cadonilimab plus anlotinib in the AK104-208 trial (49.3%) and for nivolumab plus ipilimumab (34%)^19,21, but lower than that observed with pembrolizumab plus chemotherapy in the KEYNOTE-189 and KEYNOTE-407 trials (59.1%)⁷, and markedly lower than the 80% reported for MEDI5752 (PD-1/CTLA-4 bispecific antibody) 1500 mg Q3W combined with chemotherapy²⁰. Together with the predominance of chemotherapy-related myelosuppressive toxicities in the safety profile, these data supported the interpretation that the observed toxicities were largely driven by the addition of cytotoxic chemotherapy. Notably, single-molecule design of cadonilimab may offer a more controllable safety profile than other dual immunotherapy combinations, thereby providing a wider therapeutic window when used in combination with chemotherapy. This was reflected by the ability to maintain a numerically comparable incidence of grade 3–4 TRAEs (52.0% vs. 47%) despite a longer duration of chemotherapy with cadonilimab (4 cycles) compared with nivolumab plus ipilimumab (2 cycles)¹². In terms of irAEs, our study demonstrated considerably lower incidence for both any grades (36.0% versus 30.9%) and grade 3 or 4 events (12.0% versus 12.6%) compared to the rates reported in the pembrolizumab arms of KEYNOTE-189 and KEYNOTE-407 study⁷. No TRAE-related deaths were reported. These findings suggest that cadonilimab plus chemotherapy can be safely applied in patients with PD-L1-negative NSCLC.

Of interest, our study is the first to reveal that cfDNA methylation profiling can enable early, non-invasive prediction of treatment response dynamics and long-term prognosis in this patient population. In this study, we performed an enzyme-transfer-based whole-genome sequencing methylation assay on blood samples collected at C1D1 and C3D1 to identify NSCLC-specific cfDNA methylation signatures. This technique has been successfully applied to monitor treatment response in gastroesophageal cancer and to assess PD in NSCLC with leptomeningeal metastases using blood and CSF samples^22,23, and the detection platform has been employed across multiple indications to assess tumor burden dynamics during treatment. Our analysis revealed a significant correlation between methylation results at C3D1 and radiographic results at C8D1, suggesting that molecular assessments could potentially predict treatment response approximately 105 days earlier than conventional imaging. Meanwhile, baseline molecular risk levels were significantly associated with PFS. cfDNA methylation has been shown to reflect tumor burden and stratify risk via tumor-specific patterns^24–26, and it has been utilized for longitudinal monitoring of advanced NSCLC in patients undergoing targeted therapies²⁷. Overall, blood-based methylation assays represent a promising tool for early prediction of treatment response and long-term clinical outcomes.

Notably, changes in molecular response from baseline to C3D1 were not predictive of better PFS in our analysis. Instead, a nonsignificant but counterintuitive trend was observed in which molecular non-responders had longer PFS than responders (median PFS 11.4 vs 9.0 months, p = 0.096). This pattern is likely driven by confounding due to baseline tumor burden: patients classified as molecular responders tended to have higher baseline tumor fraction, which is associated with poorer PFS. Collectively, these findings suggest that baseline tumor fraction may be a stronger determinant of PFS than short-interval early decreases in tumor fraction during treatment.

An additional consideration is that integrating baseline tumor WES could further refine the biological interpretation of cfDNA dynamics by defining the somatic mutation landscape and enabling assessment of concordance or complementarity between tissue and blood biomarkers. However, WES was not feasible in this advanced-stage NSCLC cohort because baseline specimens were typically small needle biopsies with insufficient qualified residual tissue after routine pathology and standard-of-care testing, and the pre-treatment plasma cfDNA remaining after methylation profiling was often below typical WES input requirements. Future prospective studies with planned biospecimen collection and nucleic-acid allocation could integrate WES with longitudinal methylation profiling to provide additional mechanistic insights. In this study, the cfDNA methylation analyses were exploratory and hypothesis-generating. Given the retrospective design, limited sample availability, and lack of independent or prospective validation, our cfDNA biomarker findings should be interpreted cautiously and may be influenced by cohort-specific factors. Validation in independent, prospectively enrolled cohorts will be essential to confirm both clinical utility and biological relevance.

There were several limitations to this study. First, the single-arm design and the relatively small sample size used in this exploratory study may limit the generalizability and robustness of our conclusions. Second, longer follow-up and analyses of mechanisms of resistance and subsequent therapy will be necessary to better characterize the durability and clinical relevance of the observed responses. Finally, these results require confirmation in future studies with a parallel control group.

In conclusion, in advanced PD-L1-negative NSCLC, the addition of cadonilimab to platinum-doublet chemotherapy as a first-line treatment demonstrated encouraging efficacy with manageable toxicity. These findings have informed the design of an ongoing multicenter phase III randomized trial for PD-L1-negative advanced NSCLC, including NCT05990127, which is evaluating cadonilimab or tislelizumab combined with chemotherapy, and NCT05690945, which is investigating another PD-1/CTLA4 bispecific antibody in combination with chemotherapy.

Methods

Ethics statement

The lead institutional review board and independent ethics committee of Shanghai Pulmonary Hospital (Approval No. L23-160-2), as well as the independent ethics committees of all participating centers, approved the study protocol and amendments, and the study was conducted in accordance with Good Clinical Practice Standards and the Declaration of Helsinki. Written informed consent was obtained from all patients before enrollment. While no direct financial incentives were provided for participation, participants received stipends for travel and nutritional support. Comprehensive trial-related injury insurance was maintained by the sponsor to provide compensation for any adverse events or injuries directly associated with the study procedures. Male and female patients were recruited based on self-reported sex information, with all study outcomes subsequently reported in a sex-disaggregated manner.

Study design and participants

This open-label, multicenter, single-arm, investigator-initiated, phase 2 study (LungCadX) was conducted at Shanghai Pulmonary Hospital, Shanghai Chest Hospital, and Huamei Hospital in China. Eligible patients were aged 18 to 80 years and had histologically or cytologically confirmed unresectable locally advanced (stage IIIB or IIIC) not amenable to definitive chemoradiotherapy, or metastatic, or recurrent (stage IV) NSCLC per the eighth edition of the Cancer Staging Manual of the American Joint Committee on Cancer and Union for International Cancer Control, without epidermal growth factor receptor (EGFR) or anaplastic lymphoma kinase (ALK) mutations. All participants were required to have an Eastern Cooperative Oncology Group (ECOG) performance status score of 0 or 1, a predicted life expectancy of at least 3 months, tumor PD-L1 tumor proportion score (TPS) < 1% (22C3), at least one measurable lesion per Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST v1.1), and adequate organ functions. Patients with previous ICIs (anti-PD-1/PD-L1 antibodies, anti-CTLA-4 antibodies), immune checkpoint agonists (CD137, Ox40), and immune cell therapy were ineligible. Other key exclusion criteria included: clinically uncontrollable pleural effusion or ascites; active hemoptysis, active diverticulitis, abdominal abscess, gastrointestinal obstruction, and peritoneal metastasis requiring clinical intervention; tumor compressing surrounding vital organs (such as the esophagus) with associated symptoms, compressing the superior vena cava, or invading major mediastinal vessels.

This study was registered in the Chinese Clinical Trial Registry with the identifier ChiCTR2300071681 (https://www.chictr.org.cn/hvshowprojectEN.html?id=225437&v=1.0); the date of preregistration was May 22, 2023. All procedures and analyses reported in this manuscript were conducted in accordance with the preregistered protocol, and no deviations were made during the trial.

Study procedures

In this study, all patients received 4 cycles of cadonilimab (10 mg/kg Q3W) in combination with chemotherapy (pemetrexed plus carboplatin for non-squamous NSCLC and paclitaxel plus carboplatin for squamous NSCLC), followed by maintenance with cadonilimab plus pemetrexed. Tumor evaluation was performed at baseline, and week 6 (±7 days) and week 12 (±7 days) following enrollment, then every 9 weeks (±7 days) up to week 48, and every 12 weeks (±7 days) thereafter until radiographic PD, treatment discontinuation, initiation of other anti-tumor therapies, loss to follow-up, death, withdrawal of consent, or study termination, whichever occurred first. Baseline PD-L1 expression was determined using PD-L1 immunohistochemistry 22C3 pharmDx assay (Dako Omnis), and PD-L1 negative was defined as a PD-L1 TPS < 1%.

All adverse events (AEs) were recorded from the end of treatment to 30 days (90 days for serious AEs [SAEs]) after the last dose of the study drug or the initiation of subsequent anti-tumor treatment, whichever occurred first. AEs were assessed and graded by the investigator according to the National Cancer Institute Common Terminology Criteria for Adverse Events (NCI CTCAE) version 5.0.

Endpoints

The primary endpoint was the 12-month PFS rate based on RECIST v1.1, assessed by the investigator, defined as the proportion of patients who are still alive without PD. Secondary endpoints included OS (defined as the time from the first dose of cadonilimab plus chemotherapy to all-caused death), PFS (defined as the time from the first dose of cadonilimab plus chemotherapy to radiographic PD or death due to any cause, whichever occurred first), ORR (defined as the proportion of patients with complete response [CR] or partial response [PR]), duration of response (DoR, defined as the time from the first documented CR or PR to the first documented PD or death for any reason, whichever occurred first), DCR (defined as the proportion of patients with CR, PR, or stable disease [SD]), and safety.

Plasma cfDNA extraction and next-generation sequencing

For cfDNA analysis, peripheral blood samples were prospectively collected at two predefined time points: baseline before treatment initiation (C1D1) and an early on-treatment time point at Cycle 3 Day 1 (C3D1). These sampling time points were aligned with scheduled radiographic tumor assessments to evaluate whether early molecular changes could precede conventional imaging-based response assessment. cfDNA was extracted from plasma samples using the QIAamp circulating nucleic acid kit, and its quantity and quality were checked using Qubit fluorimeter and Bioanalyzer 2100 after purification. 5 to 20 ng of extracted cfDNA was prepared for library construction, including end-repair, dA-tailing, adapter ligation, and polymerase chain reaction amplification. Hybrid capture was performed on the amplified DNA libraries with sufficient yields. Briefly, the library was hybridized overnight with the panel probes and subsequently washed to remove unbound fragments. The purified libraries were quality-checked with Bioanalyzer 2100 and then sequenced on the Illumina paired-end sequencing platform (2 × 150 bp).

Bioinformatics and statistical analysis

cfDNA was analyzed using PredicineEPIC, a whole-genome methylation assay. Specific DMFs were identified using a proprietary in-house algorithm developed by Predicine²². DMFRs were defined as clusters of co-localized CpG sites covered by DMFs²⁸. To ensure reproducible quantification and enable future validation, fragment-level methylation status was assessed using wgbstools. A normalized DMF score was calculated for each sample by adjusting the total count of detected DMFs for the corresponding sequencing depth. Liquid biopsy response was determined using quantitative response criteria based on dynamic changes in these normalized DMF values (LB-RECIST)²⁹. PD was defined as an increase of more than 33%, PR as a decrease of more than 33%, and SD as changes within ±33% relative to baseline. For risk stratification analysis, patients were classified into Molecular Groups (High-Risk vs. Low-Risk) using a data-driven threshold. The mean normalized DMF score of the current cohort was established as the cutoff. Patients with scores above the mean were classified as High-Risk, while those below the mean were classified as Low-Risk. Statistical comparisons of numerical variables were performed using the Wilcoxon rank-sum test, while categorical variables were compared using the chi-squared test. Survival outcomes were analyzed using standard time-to-event methods. PFS was defined as the time from the baseline blood draw to the occurrence of the event of interest or last follow-up. Survival distributions were estimated using the Kaplan–Meier method and compared between groups using the log-rank test. HRs and corresponding 95% CIs were estimated using Cox proportional hazards regression models. For subgroup analyses, survival comparisons were performed using the same methods. All tests were two-sided, with a P-value < 0.05 considered statistically significant.

Statistical analysis

Sample size estimation was performed on the basis of the primary endpoint of the 12-month PFS rate. Referring to the PD-L1-negative NSCLC subgroup in the KEYNOTE-189 study (pembrolizumab plus chemotherapy), the historical control 12-month PFS rate was set at 26%³⁰. In this study, we hypothesized that cadonilimab plus chemotherapy would improve the 12-month PFS rate to 40.0%. With a one-sided α of 0.10, 80% power, a 12-month enrollment period, and a 12-month follow-up period, the final analysis will be triggered when 31 PFS events have accumulated. Accounting for a 10% rate of loss to follow-up, 48 treated patients were required. Furthermore, anticipating that approximately 10% of consented patients might not receive treatment due to ineligibility or withdrawal, a total of 54 subjects were planned for enrollment.

The efficacy analysis was performed in the full analysis set (FAS), including all enrolled patients who received at least one dose of cadonilimab plus chemotherapy and underwent at least one tumor response evaluation. The safety was analyzed in the safety analysis set (SS), which included all enrolled patients treated with at least one dose of cadonilimab plus chemotherapy.

Continuous variables were presented as means (standard deviations) or medians (ranges). The categorical data were described by counts and percentages. The 95% confidence intervals (CIs) of ORR and DCR were estimated using the Clopper-Pearson method. The median OS, PFS, DoR, and follow-up duration were calculated with the Kaplan-Meier method, and their 95% CIs were calculated using the Brookmeyer-Crowley method with a log-log transformation. Statistical analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC, USA) and R version 4.4.1. Survival analyses were performed using the Kaplan–Meier method and compared with the log-rank test. Statistical tests were two-tailed, with an α set at 0.1.

Author contributions

Conceptualization: C.S., C.R., T.C.; Methodology: L.W., J.C., H.T., H.D., S.J., T.C.; Formal analysis: C.S.; Investigation: C.S., L.W., K.H., J.Q., H.L., S.P., C.R., L.N., L.L., J.Z., Y.W., T.C.; Resources: C.S., L.W., Q.W., K.H., J.Q., H.L., S.P., C.R., L.N., L.L., J.Z., Y.W., Y.Z., T.C.; Data curation: C.S., L.W., K.H., J.Q., H.L., S.P., C.R., L.N., L.L., J.Z., Y.W., T.C.; Writing review & editing: C.S., T.C., L.W.; Supervision: C.S., C.R., T.C.; Project administration: C.S., C.R., T.C.; Funding acquisition: C.S., L.W.

Peer review

Peer review information

Nature Communications thanks Lin Wu, Alexander Haragan, and the other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was funded by the National Key R&D Program of China (grant numbers: 2023YFC2508601, 2023YFC2508604, and 2023YFC2508605, C.S.), the Shanghai Hospital Development Center Foundation (SHDC22025208, C.S.), the Shanghai Pulmonary Hospital 2024 Research-oriented Physician Talent Program (LYRC202401, C.S.), the Tongji University Medicine-X Interdisciplinary Research Initiative (2025-0554-ZD-08, C.S.), and the Shanghai Anticancer Association EYAS PROJECT (SACA-CY24A04, L.W.).

Data availability

The study protocol and all datasets generated during this study are included in the Article, Supplementary Information, and Source Data files. The raw sequence data reported in this article have been deposited in the Genome Sequence Archive at the National Genomics Data Center, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences, under accession code HRA012722. Access to raw sequencing data for non-commercial research requires approval from the study sponsor and relevant ethics committees, as well as the execution of a formal data use agreement. Due to patient privacy concerns, clinical data are not publicly available; however, de-identified individual patient data may be accessed from the corresponding author (susu_mail@126.com) for non-commercial academic purposes for up to three years. Such requests will typically be processed within three months. Source data are provided with this paper.

Competing interests

Haoran Tang, Hang Dong, and Shidong Jia disclosed that they were employees of Huidu Shanghai Medicine Ltd. Yichao Zang is an employee of Akeso Biopharma. No disclosures were reported by the other authors. The other authors declared no conflict of interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74241-3.