Neoadjuvant camrelizumab plus chemotherapy in resectable stage IIIA–IIIB non-small cell lung cancer: integrated multidimensional biomarker analysis from a phase II study Authors: Guoxin Cai, Pingping Song, Zhendan Wang, Jiarui Zhao, Kaiyue Wang, Min Li, Xingpeng Wang, Jingyu Zhu, Jing Xu, Baiyang Huang, Xiao Han, Haiyan Zhou, Chao Xie, Shuai Fu, Hui Jia, Feifei Teng, Suzhen Wang, Hongfu Sun, Wanhu Li, Jinming Yu, Xue Meng Published: 2026 ABSTRACT AbstractBackgroundThis study evaluated biomarker profiles associated with therapeutic response and prognosis in resectable non-small cell lung cancer (NSCLC) receiving neoadjuvant camrelizumab plus chemotherapy.MethodsIn this single-arm phase II trial, patients with resectable stage IIIA–IIIB NSCLC received three cycles of camrelizumab plus chemotherapy, followed by surgery. Primary endpoints were pathological complete response (pCR). Secondary endpoints included major pathological response (MPR), disease-free survival (DFS), and safety. Exploratory analyses assessed immune subsets in the tumor microenvironment (TME) and peripheral blood, cytokine profiles, circulating tumor DNA (ctDNA) dynamics, and spatial immune–tumor interactions.ResultsThirty patients were enrolled (22 squamous, 8 adenocarcinoma); 27 underwent radical resection. The pCR and MPR rates were 33.3% and 50.0%, respectively. At a median follow-up of 27.2 months, the 2-year DFS rate was 77.4%. Grade≥3 treatment-related adverse events occurred in 13.3%. Achieving pCR was associated with higher baseline levels of CD8+ T cells and naïve-like T cells in both blood and TME. Elevated baseline CD8+ naïve-like T cells in both peripheral blood and tumor nests and their closer proximity to tumor cells correlated with improved DFS. Higher circulating baseline chemokine (C-C motif) ligand 3 (CCL3) levels were associated with both pCR and prolonged DFS. Postneoadjuvant treatment ctDNA positivity predicted shorter DFS, while longitudinal ctDNA monitoring identified relapse as early as 17.6 months before radiographic progression.ConclusionsBaseline CD8+ naïve-like T-cell abundance and spatial proximity to tumor cells, circulating CCL3 levels, and ctDNA dynamics may serve as potential prognostic biomarkers in resectable NSCLC receiving neoadjuvant camrelizumab plus chemotherapy.Trial registration numberNCT06241807. WHAT IS ALREADY KNOWN ON THIS TOPIC Neoadjuvant chemoimmunotherapy improves pathological response and survival in resectable non-small cell lung cancer (NSCLC), but clinical benefit remains heterogeneous and robust predictive biomarkers for patient selection and treatment monitoring are lacking. Neoadjuvant chemoimmunotherapy improves pathological response and survival in resectable non-small cell lung cancer (NSCLC), but clinical benefit remains heterogeneous and robust predictive biomarkers for patient selection and treatment monitoring are lacking. WHAT THIS STUDY ADDS In this prospective phase II study, neoadjuvant camrelizumab plus chemotherapy demonstrated favorable efficacy and manageable safety in patients with resectable stage IIIA–IIIB NSCLC. Higher baseline levels of CD8+ naïve-like T cells in peripheral blood and tumor nests, elevated circulating chemokine (C-C motif) ligand 3 (CCL3), and closer immune–tumor spatial proximity were associated with pathological complete response and prolonged disease-free survival. Longitudinal circulating tumor DNA (ctDNA) monitoring identified minimal residual disease and predicted relapse up to 17.6 months before radiographic progression. In this prospective phase II study, neoadjuvant camrelizumab plus chemotherapy demonstrated favorable efficacy and manageable safety in patients with resectable stage IIIA–IIIB NSCLC. Higher baseline levels of CD8+ naïve-like T cells in peripheral blood and tumor nests, elevated circulating chemokine (C-C motif) ligand 3 (CCL3), and closer immune–tumor spatial proximity were associated with pathological complete response and prolonged disease-free survival. Longitudinal circulating tumor DNA (ctDNA) monitoring identified minimal residual disease and predicted relapse up to 17.6 months before radiographic progression. HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY These findings support an integrated biomarker-driven framework combining immune profiling, spatial immune architecture, and ctDNA dynamics to optimize patient selection and postoperative surveillance in neoadjuvant chemoimmunotherapy for NSCLC. These findings support an integrated biomarker-driven framework combining immune profiling, spatial immune architecture, and ctDNA dynamics to optimize patient selection and postoperative surveillance in neoadjuvant chemoimmunotherapy for NSCLC. Introduction Non-small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancer cases and remains the leading cause of cancer-related mortality worldwide.1 Approximately 20%–25% of patients are diagnosed at a resectable stage.2 Neoadjuvant therapy (NAT) has been shown to improve survival by enhancing R0 resection rates and eliminating micrometastatic disease.3 However, traditional neoadjuvant chemotherapy yields only modest survival benefits, improving 5-year survival by approximately 5% compared with surgery alone, with pathological complete response (pCR) rates of just 5%–10% and limited long-term disease control.4 The emergence of immune checkpoint inhibitors has significantly altered the perioperative treatment paradigm for resectable NSCLC. Recent pivotal trials—including CheckMate 816,5 AEGEAN,6 KEYNOTE-671,7 and CheckMate 77T8—have demonstrated the addition of programmed cell death protein 1 (PD-1) or programmed death-ligand 1 (PD-L1) inhibitors to neoadjuvant chemotherapy significantly improves pCR rates and event-free survival (EFS). Consequently, neoadjuvant chemoimmunotherapy (NCIT) is now endorsed as a first-line treatment approach in the National Comprehensive Cancer Network (NCCN) Clinical Practice Guidelines.9 Camrelizumab is a humanized monoclonal antibody targeting PD-1. The TD-FOREKNOW phase II trial, the first randomized controlled study evaluating neoadjuvant camrelizumab plus chemotherapy for resectable NSCLC, demonstrated significantly higher pCR rates and improved EFS compared with chemotherapy alone, with a median follow-up of 14.1 months.10 Based on these findings, the Chinese Society of Clinical Oncology (CSCO) 2024 guidelines recommend neoadjuvant camrelizumab combined with chemotherapy for resectable NSCLC (level II recommendation). Despite these advances, clinical benefit remains heterogeneous among patients. Thus, it is crucial to identify robust biomarkers associated with treatment response to refine patient selection and guide therapeutic decision-making. In this context, we performed a single-arm phase II study to characterize biomarkers linked to therapeutic response and clinical outcomes in resectable stage IIIA–IIIB NSCLC treated with neoadjuvant camrelizumab combined with chemotherapy. Materials and methods Study design and participants A prospective, open-label, single-arm phase II clinical trial was conducted at Shandong Cancer Hospital and Institute between December 2022 and September 2023. Eligible patients were aged 18–75 years and had histologically confirmed, resectable stage IIIA or IIIB (T3–4N2) NSCLC based on the eighth edition of the American Joint Committee on Cancer (AJCC) staging system.11 Key inclusion criteria included an Eastern Cooperative Oncology Group (ECOG) performance status score of 0 or 1, at least one measurable lesion according to immune-based Response Evaluation Criteria in Solid Tumors (iRECIST) V.1.1,12 no prior systemic anticancer treatment, and absence of known activating epidermal growth factor receptor mutations or anaplastic lymphoma kinase rearrangements. Exclusion criteria included a prior history of other malignancies, documented immunodeficiency, use of immunosuppressive agents within 14 days prior to enrollment, and either a current or previous diagnosis of pulmonary fibrosis or interstitial lung disease. A comprehensive list of inclusion and exclusion criteria is available in the study protocol (online supplemental material). Procedures Patients received three cycles of neoadjuvant camrelizumab combined with platinum-based chemotherapy. Camrelizumab was administered intravenously at a fixed dose of 200 mg on day 1 of each 3-week cycle. Chemotherapy regimens were also administered every 3 weeks for three cycles. For patients with squamous cell carcinoma, regimens included nab-paclitaxel (260 mg/m² on day 1) or paclitaxel (175 mg/m² on day 1) combined with either cisplatin (75 mg/m² on day 1) or carboplatin (area under the curve = 5 mg/mL·min on day 1). For non-squamous cases, pemetrexed (500 mg/m² on day 1) was combined with cisplatin or carboplatin at the same respective doses. Surgery was scheduled 4–6 weeks following completion of NAT. Supportive care, including antiemetics and growth factors, was administered as needed according to institutional guidelines. Whole-body 18F-fluorodeoxyglucose positron emission tomography/CT (18F-FDG PET/CT) was performed within 1 week before initiation and after completion of treatment. Radiological imaging follow-up was conducted every 3 months for the first 2 years, followed by every 6 months thereafter. Adverse events (AEs) were assessed systematically at each study visit and graded based on the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE), V.5.0. Study endpoints The primary endpoint was the pCR rate, defined as the absence of viable tumor cells in both the resected primary tumor and all sampled regional lymph nodes. Secondary endpoints included the major pathological response (MPR) rate, defined as ≤10% residual viable tumor cells in the resected specimens, as well as disease-free survival (DFS), EFS, and safety. DFS was measured from the date of surgery to the first documented disease recurrence or death from any cause. EFS was defined as the interval from enrollment to disease progression, recurrence, or death from any cause. Exploratory analyses included profiling of immune subsets in the tumor microenvironment (TME) and peripheral blood, circulating cytokine/chemokine profiles, circulating tumor DNA (ctDNA) dynamics, and spatial immune–tumor interactions. Sample size and design The sample size was determined based on the primary endpoint of pCR, assuming a 30% rate for the investigational regimen (p₁) and 10% under the null hypothesis (p₀; historical neoadjuvant chemotherapy). Simon’s two-stage design was applied with a one-sided α=0.05 and 80% power. In stage 1, 11 evaluable patients were enrolled; if ≤1 patient achieved pCR, the study would be terminated for futility. Otherwise, accrual continued to 27 evaluable patients, with the regimen deemed effective if ≥6 achieved pCR. Allowing for a 10% dropout rate, the maximum planned enrollment was 30 patients. Statistical analysis Efficacy and safety analyses were conducted in the intention-to-treat (ITT) population. Categorical variables were reported as frequencies and percentages, while continuous variables were reported as medians with IQRs. Median follow-up time, IQRs, and time-to-event endpoints were estimated and analyzed using the Kaplan–Meier method. Exact two-sided 95% CIs were calculated using the Clopper–Pearson method. Correlations between 18F-FDG metabolic parameters and ctDNA levels were assessed using Spearman’s rank correlation coefficient. Comparisons of categorical biomarker variables between relapsed and disease-free patients, as well as between squamous and non-squamous histological subtypes, were performed using Fisher’s exact test. Posthoc analyses stratified patients based on pathological response, PD-L1 expression, and adjuvant therapy status. Statistical significance was defined as a two-sided p<0.05. All statistical analyses and graphical outputs were generated using R (V.3.6.1 and 4.3.0; Vienna, Austria) and GraphPad Prism (V.8; GraphPad Software). Detailed methods for ctDNA detection, flow cytometry, cytokine assays, multiplex immunofluorescence staining, and sample size/design are described in the online supplemental methods. Results Patient characteristics Between December 26, 2022 and September 20, 2023, a total of 30 patients with NSCLC from Shandong Cancer Hospital and Institute were enrolled (table 1, figure 1a). The median age was 63 years (IQR, 60–68 years). The majority of patients were male (28/30, 93.3%), and most had squamous cell carcinoma (22/30, 73.3%), while eight patients (26.7%) had adenocarcinoma. Clinical staging at enrollment indicated stage IIIA disease in 19 patients (63.3%) and stage IIIB in 11 patients (36.7%). All 30 patients completed NAT, of whom 27 (90.0%) subsequently underwent surgical resection, with all achieving R0 status. The median interval from the final NAT cycle to surgery was 4.9 weeks (IQR, 4.6–5.4 weeks). Surgical delays occurred in 3 of 27 patients (11.1%) due to anemia (n=2) and thrombocytopenia (n=1). Among those who underwent surgery, 26 patients (96.3%) received lobectomy, and one patient (3.7%) underwent pneumonectomy, which occurred in a patient with stage IIIA disease. The median operative time and blood loss were 120 min (IQR, 96–150) and 100 mL (IQR, 50–120), respectively, in the overall population; 128 min (IQR, 101–159) and 100 mL (IQR, 50–120) in stage IIIA patients; and 115 min (IQR, 92–120) and 100 mL (IQR, 80–120) in stage IIIB (N2) patients. Three patients did not undergo surgery: two declined the procedure, and one was deemed ineligible due to clinical reassessment. Clinicopathological characteristics, outcomes, and survival in patients with resectable stage IIIA–IIIB NSCLC treated with neoadjuvant camrelizumab plus chemotherapy. (a) Clinicopathological characteristics and outcomes. (b) Disease-free survival (DFS). (c) Event-free survival. (d) DFS in patients with versus without pathological complete response (pCR). (e) DFS in patients with versus without major pathological response (MPR). (f) DFS in patients with PD-L1 expression≥1% compared with those with<1%. (g) DFS in patients with versus without adjuvant immunotherapy (IT). iCR, immune complete response; iPD, immune progressive disease; iPR, immune partial response; iRECIST, immune response evaluation criteria in solid tumors; iSD, immune stable disease; NSCLC, non-small cell lung cancer; PD-L1, programmed death-ligand 1. Table 1Baseline characteristicsCharacteristicTotal (n=30)Age, median (IQR), years63 (60–68)Sex, n (%) Male28 (93.3) Female2 (6.7)Histologic characteristics, n (%) Squamous cell carcinoma22 (73.3) Adenocarcinoma8 (26.7)Chemotherapy regimen, n (%) Paclitaxel or Nab-paclitaxel+Platinum22 (73.3) Pemetrexed+Platinum8 (26.7)ECOG performance status, n (%) 021 (70.0) 19 (30.0)Smoking, n (%) Nonsmoker6 (20.0) Former or current smoker24 (80.0)T stage, n (%) T1–219 (63.3) T3–411 (36.7)N stage, n (%) N02 (6.7) N16 (20.0) N222 (73.3)Clinical stage, n (%) IIIA19 (63.3) IIIB11 (36.7)Resection extent*, n (%) Pneumonectomy1 (3.7) Lobectomy26 (96.3)PD-L1 expression level, n (%) <1%9 (30.0) 1%–49%10 (33.3) ≥50%7 (23.3) Unknown†4 (13.3)*Denominator based on patients with surgery (n=27).†PD-L1 status was unavailable in four patients due to insufficient tumor tissue or unavailability of biopsy specimens from outside institutions.ECOG, Eastern Cooperative Oncology Group; PD-L1, programmed death-ligand 1. Baseline characteristics Denominator based on patients with surgery (n=27). PD-L1 status was unavailable in four patients due to insufficient tumor tissue or unavailability of biopsy specimens from outside institutions. ECOG, Eastern Cooperative Oncology Group; PD-L1, programmed death-ligand 1. Efficacy The tumor downstaging rate was 76.7% (23/30). In the ITT population, the rates of MPR and pCR were 50.0% (15/30) and 33.3% (10/30), respectively (figure 1a). Among patients with squamous cell carcinoma, MPR and pCR rates were 63.6% (14/22) and 40.9% (9/22), respectively; for adenocarcinoma, both MPR and pCR rates were 12.5% (1/8). When stratified by stage, the pCR and MPR rates were 38.9% and 44.4% in patients with stage IIIA disease, and 33.3% and 66.7% in those with stage IIIB disease, respectively. As of the data cut-off (July 31, 2025), the median follow-up duration was 27.2 months (IQR, 24.6–29.8). In the ITT population, six patients (20.0%) experienced relapse; one patient (3.3%) died of pneumonia 2 weeks after surgery, and one patient died due to disease progression. Median DFS and overall survival (OS) had not yet been reached. The 2 year DFS rate in the surgery cohort and the 2 year EFS rate in the ITT population were 77.4% (95% CI 56.4% to 89.2%) and 76.0% (95% CI 56.0% to 87.8%), respectively (figure 1b,c). When stratified by stage, the 2 year DFS rate was 72.2% (95% CI 50.6% to 87.4%) in patients with stage IIIA disease and 87.5% (95% CI 38.7% to 98.1%) in those with stage IIIB disease. Subgroup analyses showed that patients achieving MPR had significantly longer DFS (p=0.026) and EFS (p=0.026) compared with non-MPR patients, whereas pCR status was not significantly associated with DFS (p=0.241) or EFS (p=0.241) (figure 1d,e, online supplemental figure 1a,b). There was a trend toward improved DFS (p=0.095) and EFS (p=0.077) in patients with PD-L1–negative tumors compared with those with PD-L1≥1% (figure 1f, online supplemental figure 1c). Among patients who underwent surgery, 12 received adjuvant camrelizumab and 15 did not. This exploratory, non-randomized analysis did not reveal statistically significant differences in DFS (p=0.505) or EFS (p=0.374) between the two subgroups and should be interpreted with caution due to potential selection bias (figure 1g, online supplemental figure 1d). Safety Among the ITT population, 29 of 30 patients (96.7%) experienced treatment-related adverse events (TRAEs) during NAT (table 2), most of which were grades 1–2. The most frequent events were decreased white blood cell count (17/30, 56.7%) and alopecia (14/30, 46.7%). Four patients (13.3%) developed grade≥3 TRAEs, including leukopenia in 3 (10.0%) and neutropenia in 1 (3.3%). Immune-related adverse events occurred in 15 patients (50.0%), all of which were grades 1–2, including reactive cutaneous capillary endothelial proliferation (RCCEP) (13/30, 43.3%), hypothyroidism (2/30, 6.7%), and hyperthyroidism (1/30, 3.3%) (table 2). No dose reductions, treatment delays, or deaths were reported during the NAT period. Among the 27 patients who underwent surgery, 13 (48.1%) experienced surgery-related adverse events (SRAEs), mostly grades 1–2 (table 2). The most common SRAEs were anemia (7/27, 25.9%) and intraoperative bleeding (6/27, 22.2%). When stratified by stage, SRAEs occurred in 44.4% (8/18) of patients with stage IIIA disease and 55.6% (5/9) of those with stage IIIB (N2) disease. Table 2Adverse eventsEventAny gradeGrade≥3Immune-mediated adverse events, n (%) Any15 (50.0)0 Reactive cutaneous capillary endothelial proliferation13 (43.3)0 Hypothyroidism2 (6.7)0 Hyperthyroidism1 (3.3)0Treatment-related adverse events, n (%) Any29 (96.7)4 (13.3) White blood cell count decreased17 (56.7)3 (10.0) Alopecia14 (46.7)0 Reactive cutaneous capillary endothelial proliferation13 (43.3)0 Neutrophil count decreased11 (33.3)1 (3.3) Fatigue7 (23.3)0 Nausea5 (16.7)0 Anemia4 (13.3)0 Thrombocytopenia3 (10.0)0 Vomiting3 (10.0)0 Constipation2 (6.7)0 Neurotoxicity2 (6.7)0 Hypothyroidism2 (6.7)0 Increased aminotransferases2 (6.7)0 Fever1 (3.3)0 Diarrhea1 (3.3)0 Hyperthyroidism1 (3.3)0Surgery-related adverse events, n (%)* Any13 (48.1)1 (3.7) Anemia7 (25.9)0 Intraoperative bleeding6 (22.2)0 Pulmonary infection2 (7.4)1 (3.7) Postoperative oozing of blood2 (7.4)0 Postoperative pneumothorax1 (3.7)0 Pleural effusion1 (3.7)0 Perioperative death1 (3.7)1 (3.7)*Denominator based on patients with surgery (n=27). Adverse events Denominator based on patients with surgery (n=27). One patient died postoperatively. The patient had no pre-existing pulmonary disease prior to treatment. On postoperative day 6, the patient developed fever and productive cough with purulent sputum. Sputum culture was positive for Pseudomonas aeruginosa. Chest CT demonstrated multiple irregular, large patchy areas of increased density in both lungs, with partial consolidation and air bronchograms. Based on the clinical presentation and imaging findings, the event was considered consistent with severe postoperative pneumonia rather than immune-related pneumonitis. Despite anti-infective treatment, the patient’s condition deteriorated and death occurred 2 weeks after surgery. Analysis of ctDNA The mutation landscape of baseline plasma ctDNA was delineated (online supplemental figure 2a). Overall, TP53 was the most frequently mutated gene (70.4%), followed by CSMD3 (33.3%) and ZFHX4 (25.9%). The predominant mutation type was missense mutation, along with other alterations including frameshift variants, splice-site mutations, in-frame deletions, gene fusions, stop-gained/loss variants, and synonymous variants. Stratified analyses based on pathological response (online supplemental tables 1 and 2) revealed that mutations in TNR and PXDNL were more common in the patients without MPR compared with those with MPR. Similarly, RYR2 and PXDNL mutations appeared enriched in the non-pCR subgroup relative to the pCR group. However, none of these differences reached statistical significance (all p>0.10). Collectively, ctDNA profiles in this cohort were characterized by high heterogeneity driven by TP53 alterations, with certain mutation distributions showing trends consistent with pathological response but lacking statistical significance. Evaluation of ctDNA dynamics revealed a significant decline in ctDNA positivity rate following NAT, decreasing from 88.9% at baseline to 22.2% post-treatment (p<0.001) (figure 2a, online supplemental figure 2b). When stratified by pathological response, no significant differences in baseline ctDNA levels were found between pCR and non-pCR groups or between MPR and non-MPR groups (p=0.382 and p=0.451, respectively). However, post-NAT assessments showed a decreasing trend in ctDNA levels in the pCR group versus the non-pCR group (p=0.264), as well as in the MPR group versus the non-MPR group (p=0.170) (figure 2b,c). The Sankey diagram illustrated a shift in population composition according to ctDNA positivity before and after NAT (figure 2e). Overall, the proportion of ctDNA-negative patients increased after NAT, and conversion from positive to negative was more frequent in patients achieving pCR or MPR than in those without. Patient circulating tumor DNA (ctDNA) status. (a–c) ctDNA concentration before and after neoadjuvant therapy (NAT) in (a) all patients, (b) patients with versus without pathological complete response (pCR), and (c) patients with versus without major pathological response (MPR). (d) Disease-free survival (DFS) stratified by ctDNA status assessed post-NAT and within 1 month after surgery. (e) Sankey plot illustrating transitions in ctDNA status and correlation with pathological response. (f) Swimmer plots illustrating longitudinal ctDNA monitoring. Prognostic analyses showed that patients who remained ctDNA-negative after NAT exhibited a non-significant trend toward improved DFS compared with those who were ctDNA-positive (p=0.075) (online supplemental figure 2c). Moreover, patients who remained ctDNA-negative both after NAT and within 1 month postoperatively demonstrated significantly superior DFS compared with those positive at either timepoint (p=0.031) (figure 2d). Longitudinal ctDNA surveillance, depicted in a swimmer plot (figure 2f), showed that 77.3% (17/22) of evaluable patients maintained ctDNA negativity during postoperative follow-up and remained disease free. In contrast, three patients with postoperative ctDNA conversion (negative to positive) or persistent ctDNA positivity subsequently experienced recurrence, accounting for 60% (3/5) of relapsed patients under long-term minimal residual disease (MRD) monitoring. Notably, ctDNA positivity preceded radiographic recurrence by a median of 17.6 months (range, 7.4–20.3 months). To assess whether metabolic tumor activity correlates with ctDNA burden, we analyzed ctDNA concentration in relation to 18F-FDG uptake parameters before and after NAT. At baseline, ctDNA concentration showed no significant correlation with maximum standardized uptake value (SUVmax), SUVmean, total lesion glycolysis (TLG), or metabolic tumor volume (r=0.114, 0.097, 0.317, and 0.319; all p>0.1) (online supplemental figure 3a–d). Likewise, baseline FDG parameters did not differ significantly between ctDNA-positive and ctDNA-negative subgroups (all p>0.7) (online supplemental figure 3e–h). After NAT, ctDNA-positive patients exhibited higher metabolic activity, reflected by elevated SUVmax, SUVmean, and TLG values (p<0.001, p<0.001, p=0.070, respectively) (online supplemental figure 3). These results indicate that persistent ctDNA positivity after NAT may reflect residual metabolic activity and viable tumor burden, underscoring its potential as a complementary biomarker to metabolic imaging for treatment response assessment. The predictive performance of MRD for pathological response or recurrence was assessed using sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) based on ctDNA status after NAT or at 1 month postoperatively. Using ctDNA positivity, ctDNA predicted non-pCR with a sensitivity of 35.3%, specificity of 80.0%, PPV of 75.0%, and NPV of 42.1%, and predicted non-MPR with a sensitivity of 38.5%, specificity of 78.6%, PPV of 62.5%, and NPV of 57.9%. In contrast, ctDNA showed stronger performance for relapse, with a sensitivity of 83.3%, specificity of 85.7%, PPV of 62.5%, and NPV of 94.7%. Analysis of circulating immune cell subsets and cytokines We next investigated the association between circulating immune cell subsets, cytokine levels, and pathological response as well as prognosis. Prior to NAT, patients who subsequently achieved pCR demonstrated a non-significant trend toward a higher proportion of peripheral CD8+ T cells compared with non-pCR patients (p=0.063), a difference that reached statistical significance after NAT (p=0.044) (figure 3a). Additionally, baseline levels of CD8+ naïve T cells were significantly higher in the pCR group compared with the non-pCR group (p=0.050), though this difference was no longer observed post-NAT (p=0.471) (figure 3b). At baseline, there were no significant differences between the two groups in B cells, CD3+ T cells, CD4+ central memory T cells (Tcm), CD4+ naïve T cells, CD4+ effector memory T cells (Tem), Regulatory T cells (Tregs), PD-1+ CD8+ T cells, M1/M2-like monocytes, myeloid and plasmacytoid dendritic cells (mDCs and pDCs), or myeloid-derived suppressor cells (MDSCs) subsets (polymorphonuclear MDSCs (PMN-MDSCs), early MDSC (e-MDSCs), and monocytic MDSC (M-MDSCs)), except for a significantly higher proportion of CD4+ T cells in non-pCR patients (p=0.038) (online supplemental figure 4). In addition, CD8+ Tcm and CD8+ Tem tended to be increased in non-pCR patients (p=0.092 and 0.072; online supplemental figure 4). After NAT, non-pCR patients continued to exhibit a higher proportion of CD4+ T cells (p=0.032), along with significantly higher levels of M1-like monocytes (p=0.040) and M-MDSCs (p=0.049). Other immune cell subsets showed no significant differences between groups (all p>0.05) (online supplemental figure 4). In terms of prognostic implications, baseline levels of total CD8+ or CD4+ T cells were not significantly associated with DFS (p=0.368 and p=0.486, respectively), whereas patients with elevated baseline CD8+ naïve T cells showed a non-significant trend toward prolonged DFS (p=0.061) (figure 3c,d, online supplemental figure 5a). Circulating immune cell subsets and cytokines profiles in relation to pathological response and survival. (a–b) Proportion of circulating (a) CD8+ T cells and (b) CD8+ naïve T cells before and after neoadjuvant therapy (NAT), stratified by pathological complete response (pCR). (c–d) Disease-free survival (DFS) stratified by baseline levels of circulating (c) CD8+ T cells or (d) CD8+ naïve T cells. (e–g) Baseline circulating levels of (e) IL-2, (f) CXCL9, and (g) CCL3 in patients with and without major pathological response (MPR). (h–j) Post-NAT levels of (h) IL-2, (i) CXCL9, and (j) CCL3 in patients with and without MPR. (k–m) DFS stratified by the baseline levels of circulating (k) IL-2, (l) CXCL9, or (m) CCL3. IL-2, interleukin-2; CCL3, C-C motif chemokine ligand 3; CXCL9, C-X-C motif chemokine ligand 9. Regarding circulating cytokines and chemokines, patients who achieved MPR exhibited significantly higher baseline plasma levels of interleukin-2 (IL-2), chemokine (C-X-C motif) ligand 9 (CXCL9), and chemokine (C-C motif) ligand 3 (CCL3) compared with non-MPR patients (p=0.040, 0.001, and 0.001, respectively) (figure 3e–g). After NAT, only CCL3 levels remained significantly associated with MPR (p=0.030), while IL-2 and CXCL9 were no longer significant (p=0.128 and 0.604, respectively) (figure 3h–j). Most other cytokines and chemokines showed no significant differences between groups either before or after NAT (online supplemental figure 6). When patients were stratified by pCR status, CCL3 levels remained consistently higher in those achieving pCR both before and after NAT, whereas IL-2 and CXCL9 levels showed no significant differences (online supplemental figure 7). Prognostic analyses further demonstrated that high baseline levels of CXCL9 and CCL3 were significantly associated with longer DFS (both p=0.003), while IL-2 levels were not correlated with DFS (p=0.414) (figure 3k–m). Multiplex immunofluorescence analysis of TME We further characterized immune cell subsets within the primary TME to assess their associations with pathological response and prognosis. Among lymphocyte subsets, baseline CD8+ T-cell infiltration was significantly higher in patients achieving pCR than in those without pCR (TME: p=0.025; stroma: p=0.006; tumor nests: p=0.027) (figure 4a). Following NAT, no significant differences in CD8+ T-cell levels were observed between groups across TME compartments (TME: p=0.218; stroma: p=0.446; tumor nests: p=0.110) (figure 4a). At baseline, the level of CD8+ naïve-like T cells tended to be higher in the pCR group, particularly in the TME and stroma (TME: p=0.096; stroma: p=0.058; tumor nests: p=0.221) (figure 4b). Following NAT, CD8+ naïve-like T cells were significantly enriched in the tumor nests of the pCR group (TME: p=0.187; stroma: p=0.361; tumor nests: p=0.036) (figure 4b). In contrast, at baseline, CD4+ T cells (p=0.024) and CD4+ naïve-like T cells (p=0.045) were more abundant in the stroma of non-pCR patients, whereas no differences were observed in tumor nests. CD8+ Tcm (TME: p=0.057; stroma: p=0.038; tumor nests: p=0.101) tended to be enriched in the pCR group, while CD4+ Tem (TME: p=0.032; stroma: p=0.016; tumor nests: p=0.437) and CD8+ Tem (TME: p=0.013; stroma: p=0.001; tumor nests: p=0.045) were enriched in the non-pCR group (online supplemental figure 8). After NAT, CD8+ Tem remained significantly more abundant in tumors of non-pCR patients (p=0.018), whereas no significant differences were found in CD4+ T cells, CD4+ naïve-like T cells, or Tcm between groups. PD-1+ CD8+ T cells showed no significant differences between groups either before or after NAT (online supplemental figure 8). Regarding myeloid subsets, baseline M2-like monocytes were significantly more abundant in tumor nests of non-pCR patients (p=0.023). However, no significant differences were detected for M1-like monocytes, mDCs, pDCs, or MDSC subsets (PMN-MDSCs and M-MDSCs) across compartments either before or after NAT (online supplemental figures 9 and 10). Intratumoral immune cell composition, spatial distribution, and associations with response and survival. Dynamic changes in immune cell subsets within the tumor microenvironment (TME) pre-neoadjuvant and post-neoadjuvant therapy (NAT). (a–b) Proportion of CD8+ T cells (a) and CD8+ naïve-like T cells (b) within the TME before and after NAT in patients with and without pathological complete response (pCR). (c) Schematic of the spatial interaction analysis, defining the effective score as the number of immune–tumor cell pairs within a defined radius, normalized to total immune cells. (d) Effective scores of immune subsets in tumor nests pre-NAT, measured at 10, 20, and 30 µm intervals; statistical comparisons based on 20 µm. (e–f) Disease-free survival (DFS) stratified by the proportion of CD8+ T cells in the (e) TME and the (f) tumor nests before NAT. (g–h) DFS stratified by the proportion of CD8+ naïve-like T cells in the (g) TME and the (h) tumor nests before NAT. (i–j) DFS stratified by the distribution of effective scores (0–20µm) of (i) CD8+ T cells and (j) CD8+ naïve-like T cells prior to NAT. *ns, not significant; *p<0.05. MDSC, myeloid-derived suppressor cells; M-MDSC, monocytic MDSC. To assess spatial relationships between immune and tumor cells, we calculated an effective score (representing the number of tumor cells within a specified radius per immune cell), defined as the number of immune–tumor cell pairs divided by the total immune cell count, stratified by distance thresholds (0–10, 0–20, and 0–30 µm; figure 4c). In baseline tumor nests, patients who achieved pCR exhibited higher effective scores across nearly all immune cell subsets compared with non-pCR patients. When analyzed at a 20 µm interaction radius, CD8+ T cells, CD8+ Tem, and CD8+ Tcm reached statistical significance (all p<0.05) (figure 4d), suggesting that closer proximity of cytotoxic and memory T cells to tumor cells at baseline may contribute to improved pathological response. By contrast, no significant differences in effective scores were observed between groups after NAT (online supplemental figure 10). In terms of prognosis, stratification by baseline CD8+ T-cell infiltration showed that patients with higher CD8+ T-cell infiltration in tumor nests tended to have longer DFS (p=0.068), whereas no difference was observed in the overall TME (p=0.289) (figure 4e,f). Stratification based on CD8+ naïve-like T-cell infiltration revealed that patients with high baseline levels in the TME or tumor nests experienced no disease progression during follow-up, although the associations with DFS did not reach statistical significance (TME: p=0.119; tumor nests: p=0.233) (figure 4g,h). Notably, high baseline infiltration of PD-1+ CD8+ T cells in both the TME and tumor nests was significantly associated with longer DFS (both p=0.005) (online supplemental figure 5b,c), despite showing no significant association with pathological response. In contrast, baseline levels of CD4+ Tem, CD8+ Tem, and M2-like monocytes were not significantly associated with DFS in any compartment (online supplemental figure 5d–i). Furthermore, when patients were stratified by baseline effective scores (0–20 µm), no significant correlations were observed between DFS and either CD8+ T cells or CD8+ naïve-like T cells (p=0.313 and p=0.233, respectively) (figure 4i,j), although none of the patients with higher CD8+ naïve-like T-effective scores experienced progression during follow-up, likely owing to the limited sample size. Biomarker associations with disease recurrence A comparative biomarker analysis was performed between patients who experienced disease relapse and those who remained disease free. ctDNA positivity detected either after NAT or at 1 month postoperatively was significantly associated with disease relapse (p=0.004). Compared with relapsed patients, disease-free patients exhibited significantly higher baseline circulating CCL3 levels (p=0.016) and a higher abundance of CD8+PD-1+ T cells within the tumor nests of the TME (p=0.014). Baseline circulating CD8+ naïve T-cell levels were also higher in disease-free patients, with a non-significant association with relapse status (p=0.077). Notably, although the spatial proximity between CD8+ naïve-like T cells and tumor cells did not differ significantly between relapsed and non-relapsed patients (p=0.539), all patients with a high CD8+ naïve-like T-cell effective score remained disease free during follow-up. Discussion In this single-arm phase II study, we employed a multidimensional biomarker profiling approach to investigate predictors of therapeutic response and prognosis in patients with resectable stage IIIA–IIIB NSCLC treated with neoadjuvant camrelizumab combined with chemotherapy. While most previous studies on NCIT have typically examined a single aspect, our analysis integrated intratumoral and peripheral immune profiles, spatial immune–tumor interactions, cytokine signatures, and ctDNA dynamics, correlating these features with both pathological response and prognosis. This integrative strategy offers a more comprehensive view of the immune response mechanisms and suggests a thoughtful shift from single-point indicators toward an interconnected biomarker network in NCIT. Overall, this approach may facilitate the identification of patients most likely to benefit from immunotherapy (IT) and contribute to the development of more personalized treatment strategies. Our findings align favorably with previous neoadjuvant IT trials. In the CheckMate 816 study,5 nivolumab plus chemotherapy achieved a pCR rate of 24% and a 2 year EFS rate of 63.8% in the neoadjuvant setting. Similarly, the KEYNOTE-671 study7 reported a pCR rate of 18.1% and a 2 year EFS rate of 62.4% with perioperative pembrolizumab plus chemotherapy. In contrast, our trial achieved a higher pCR rate of 33.3% and a 2 year EFS of 76.0%, consistent with the results of the TD-FOREKNOW trial of camrelizumab plus chemotherapy in resectable NSCLC, which demonstrated a pCR rate of 32.6% and a 24 month EFS rate of 76.9%.10 It should be noted that NCIT has since become an accepted standard of care for resectable stage IIIA NSCLC. However, the present study was designed and approved prior to this paradigm shift, when neoadjuvant chemotherapy alone was the predominant reference regimen. Therefore, the efficacy of the camrelizumab-based regimen in this single-arm study should be interpreted in the context of historical chemotherapy controls, and direct comparisons with contemporary chemoimmunotherapy regimens should be made with caution. Patients with squamous cell carcinoma derived greater benefit than those with adenocarcinoma, as reflected by the marked differences in pCR and MPR rates, consistent with prior reports.13 14 Beyond the well-recognized higher immunogenicity of squamous tumors, we examined the association between histology and immune or molecular biomarkers. Baseline PD-L1 expression, CD8+ naïve-like T-cell abundance and spatial proximity to tumor cells, circulating CCL3 levels, and post-treatment ctDNA levels did not differ significantly between squamous cell carcinoma and adenocarcinoma, suggesting that histology-specific treatment responses cannot be fully explained by the biomarkers assessed in this study. These findings imply that additional biological factors may contribute to differential treatment sensitivity between histological subtypes, including differences in genomic landscape, intrinsic chemotherapy sensitivity, antigenicity, and TME composition. Collectively, these results suggest that histology-specific immune biology may contribute to differential responses to NCIT and highlight the need for integrated molecular and immune profiling in larger cohorts to clarify the mechanisms underlying differential responses between histological subtypes. Additionally, selected stage IIIB (N2) patients were included based on multidisciplinary assessment of technical resectability, consistent with recent phase III trials, including AEGEAN,6 KEYNOTE-671,7 CheckMate 77T,8 and IMpower130,15 which also enrolled selected stage IIIB (N2) patients. Although pathological response data for this subgroup remain limited, our exploratory findings suggest that carefully selected IIIB (N2) patients may derive meaningful benefit from neoadjuvant camrelizumab-based therapy, warranting validation in larger cohorts. Notably, a subset of patients who did not achieve MPR exhibited undetectable ctDNA after NAT. This phenomenon may be attributed to several interrelated biological and technical factors. Biologically, low ctDNA shedding may occur in tumors with localized residual disease, dense stromal architecture, or indolent growth kinetics, resulting in limited systemic release of tumor-derived DNA despite the presence of viable tumor cells. In addition, the timing of blood sampling relative to NAT and surgery may capture a transient phase of ctDNA clearance before residual tumor cells re-establish detectable shedding. From a technical perspective, although our tumor-naïve MRD assay covers a broad spectrum of cancer-related genes and incorporates multilayered error suppression strategies, residual tumor clones may harbor mutations outside the panel’s coverage or exist at burdens below the assay’s detection threshold, particularly in the neoadjuvant setting characterized by low tumor burden. Consistent with prior neoadjuvant IT studies,16 17 this discordance suggests that ctDNA monitoring and pathological assessment capture complementary aspects of treatment response, and that ctDNA should be interpreted as a systemic biomarker of residual disease rather than a surrogate for pathological response in the neoadjuvant setting. Our exploratory analyses of immune profiling provide mechanistic insight into the efficacy of neoadjuvant camrelizumab plus chemotherapy. Patients achieving pCR exhibited higher baseline CD8+ T-cell infiltration in both peripheral blood and tumor nests, indicating that therapeutic response relies on robust cytotoxic T-cell activity. Enrichment of CD8+ Tem cells in non-pCR tumors may reflect terminal differentiation or functional exhaustion, thereby limiting their proliferative and cytotoxic capacity. Conversely, higher baseline levels of CD8+ naïve-like T cells in both peripheral blood and tumor nests were also observed in patients achieving pCR and showed a trend toward longer DFS. An expanded naïve T-cell compartment may contribute to increased clonal diversity and differentiation capacity, facilitating effective priming following PD-1 blockade and chemotherapy-induced antigen release. This observation is consistent with recent single-cell and spatial transcriptomic studies in NSCLC treated with NCIT,18 which identified CD8+ naïve T cells as candidate predictive biomarkers. Notably, in the TME, these cells are unlikely to represent bona fide naïve T cells and are more appropriately interpreted as phenotypically naïve-like subsets, given the limitations of marker-based classification. Furthermore, consistent with previous studies,1921 we found that intratumoral CD8+ naïve-like T cells were relatively sparse, whereas CD8+ Tem cells predominated in the TME. This scarcity may partly explain why stratification by CD8+ naïve-like T-cell proportion was not significantly associated with DFS, even though all patients with relatively higher baseline levels remained recurrence-free during follow-up. Chemokine analyses further underscored the importance of immune cell trafficking. Elevated baseline levels of CXCL9 and CCL3 were significantly associated with improved pathological response and prolonged DFS, consistent with findings from a previous study of neoadjuvant IT in breast cancer.22 CXCL9, a CXCR3 ligand, facilitates recruitment of activated effector T cells to the TME, whereas CCL3 promotes recruitment and activation of T cells and DCs. Notably, CCL3 remained associated with MPR after NAT, suggesting its role in sustaining immune response. Spatial interaction analyses provided additional insight. At baseline, pCR tumors exhibited higher effective scores across most immune subsets, underscoring spatial proximity—rather than cell density alone—as a key determinant of cytotoxic activity. Following NAT, these differences were no longer significant, suggesting partial immune remodeling. To the best of our knowledge, this is the first NCIT study in NSCLC to integrate peripheral blood and TME immune phenotyping with spatial immune–tumor interactions. Notably, CD4+ T cells and CD8+ Tcm showed discordant distributions between peripheral blood and the TME, with higher levels in the circulation of non-pCR patients but stromal enrichment in pCR tumors, suggesting that systemic immune expansion does not necessarily translate into effective intratumoral activity and may be limited by impaired recruitment or spatial positioning. In contrast, CD8+ Tem cells were consistently increased in non-pCR patients across both compartments, potentially indicating chronic antigen-driven but functionally ineffective immune activation. Together with the observation that higher baseline CD8+ naïve-like T-cell levels and closer immune–tumor spatial proximity were associated with pCR, these findings suggest that treatment response is determined not only by immune cell abundance but also by their differentiation state, spatial organization, and functional engagement. Collectively, our findings support an integrated, time-resolved predictive framework linking baseline immune readiness with dynamic molecular surveillance. High circulating CD8+ naïve T cells and elevated CCL3 levels, together with favorable spatial immune architecture, define an “immune-ready” TME conducive to chemotherapy-induced immune activation and subsequent PD-1 blockade, thereby increasing the likelihood of achieving pCR/MPR. In parallel, perioperative and postoperative ctDNA dynamics serve as a real-time indicator of MRD and emerging relapse risk, enabling longitudinal monitoring beyond initial treatment response. Together, this “baseline enrichment+dynamic surveillance” framework integrates immune profiling with ctDNA monitoring to predict NAT benefit and identify patients at high risk of relapse. Validation in larger cohorts incorporating multiomics is warranted to further confirm and refine this model. In this study, we observed an inverse association between tumor PD-L1 expression and response to NCIT, with higher PD-L1 levels associated with inferior outcomes. Although counterintuitive given the established predictive role of PD-L1 in advanced NSCLC, several factors may underlie this observation. First, elevated PD-L1 expression may reflect an adaptive immune resistance phenotype characterized by chronic immune activation, T-cell exhaustion, and an immunosuppressive TME, rather than effective antitumor immunity.23 24 In such tumors, PD-1 blockade alone may be insufficient to fully reinvigorate dysfunctional immune responses. In contrast, tumors with low PD-L1 expression are more likely to lack pre-existing antitumor T-cell responses and thus rely more heavily on chemotherapy-induced immunogenic cell death (ICD) to initiate antitumor immunity. ICD-associated release of tumor antigens and damage-associated molecular patterns promotes DC activation and T-cell priming in tumor-draining lymph nodes.25 26 Baseline immune readiness, reflected by higher circulating CD8+ naïve T-cell proportions and elevated CCL3 levels, may provide a favorable substrate for de novo T-cell priming, a key mechanism underlying neoadjuvant IT.27 28 Accordingly, tumors with lower PD-L1 expression and preserved immune plasticity may be more dependent on chemotherapy-mediated immune priming and derive greater benefit from the synergistic effects of chemotherapy and PD-1 blockade. Moreover, PD-L1 assessment is limited by spatial and temporal intratumoral heterogeneity despite assay standardization, potentially leading to tumor misclassification.29 30 The limited sample size may also contribute to variability in subgroup analyses. Collectively, these findings suggest that PD-L1 expression alone may not fully capture the immunobiology governing response to NCIT and that integrated immune profiling may improve patient stratification. This study has several limitations. First, the absence of a chemotherapy control arm, together with the lack of cross-validation or external validation, limits the robustness and generalizability of the biomarker findings, which should therefore be considered hypothesis-generating and warrant further validation in larger, independent cohorts. Second, in the absence of a chemotherapy-only control group, it is not possible to definitively distinguish predictive from prognostic biomarkers. Third, the follow-up duration, although sufficient for DFS analysis, remains insufficient for conclusive assessment of OS. Fourth, the contribution of adjuvant IT remains unresolved in this setting and warrants evaluation in prospective randomized trials. Fifth, the performance of the tumor-agnostic MRD assay in this study was suboptimal, which is likely due to its tumor-naïve design and limited sensitivity in detecting low-frequency variants. In addition, the identification of naïve-like CD8+ T cells within the TME in this study relied on phenotypic markers (eg, CD45RA and CCR7), which may not fully distinguish bona fide naïve T cells from other phenotypically similar or less differentiated T-cell subsets and therefore warrants cautious interpretation. In summary, baseline CD8+ naïve-like T-cell abundance and spatial proximity to tumor cells, circulating CCL3 levels, and ctDNA dynamics may serve as potential biomarkers of pathological response and prognosis in patients with resectable stage IIIA–IIIB NSCLC receiving neoadjuvant camrelizumab combined with chemotherapy. Given the limited sample size, larger randomized trials with extended follow-up are needed to validate these findings, refine patient selection, and optimize perioperative IT strategies in resectable NSCLC. Supplementary material