Efficiency and safety of sintilimab-based induction therapy in potentially resectable stage III non-small cell lung cancer (LungMate-009): an open-label, phase 2 trial Authors: Tao Ge, Hongyi Zhang, Xinsheng Zhu, Meixin Teng, Yifei Zhou, Suyu Wang, Dongliang Bian, Lele Zhang, Haiyang Hu, Qiji Guo, Jincheng Su, Shiqi Hu, Huansha Yu, Yirui Zhou, Jie Huang, Jie Yang, Yuming Zhu, Yi Bao, Minwei Bao, Xuefei Hu, Wenxin He, Peng Zhang Published: 2025 ABSTRACT SummaryBackgroundApproximately 30% of patients with non-small-cell lung cancer (NSCLC) have initially resectable disease. This trial aimed to evaluate the safety and feasibility of administering sintilimab-based induction therapy for potentially resectable stage III NSCLC to offer a more advantageous therapeutic strategy.MethodsThis investigator-initiated, open-label, phase 2 trial (NCT04728724) aimed to explore the efficacy and safety of sintilimab-based induction therapy in patients with potentially resectable stage III NSCLC on the basis of tumour PD-L1 expression. Eligible patients with PD-L1 expression in at least 50% of tumour cells received four cycles of sintilimab (200 mg every 3 weeks) as monotherapy, and patients with PD-L1 expression in less than 50% of tumour cells or of unknown status received four cycles of sintilimab combined with carboplatin-based chemotherapy before surgical resection. Evaluation of efficacy was conducted by a multidisciplinary team every two cycles. Surgical resection was performed within 21–28 days of the last dose. The primary endpoint was the major pathological response rate. Additionally, bulk RNA sequencing of baseline and surgical samples was conducted to investigate the tumour microenvironment and identify potential biomarkers.Findings100 patients were enrolled between November 16, 2022, and September 11, 2023, and 97 patients were included in analyses. 75 (77%) of 97 patients completed four cycles of neoadjuvant treatment (range 1–6 cycles). An overall response was observed in 80 patients (82%; 95% CI 75–90), and all had a partial response. Surgery was performed in 58 patients (60%), and all patients had R0 resection. 38 (66%, 95% CI 53–78) of 58 patients had a major pathological response. As of September 24, 2025, the median follow-up duration was 31.0 months (IQR 27.5–32.6), and median progression-free survival and overall survival were not reached. 24-month progression-free survival was 78% (95% CI 71–87), and 24-month overall survival was 88% (81–94). Grade 3 or higher treatment-related adverse events occurred in 40 (41%) of 97 patients, and the most common grade 3–4 treatment-related adverse events were neutropenia (n = 35 [36%]), leukopenia (n = 18 [19%]). Immune-related adverse events occurred in 51 (53%) of 97 patients receiving study treatment. Differential gene expression analysis in the combination of baseline and surgical tumour samples revealed the upregulation of immune-related genes in responders to neoadjuvant immune checkpoint blockade, and the upregulation of metabolism-related genes in non-responders.InterpretationSintilimab-based induction treatment strategy could be a feasible option for potentially resectable NSCLC patients with stage III disease. Notably, we found that increased plasma cell signatures were predictive of response, while elevated AKR1C family gene expression substantially correlated with resistance to immune checkpoint blockade.FundingThis work was supported by the 10.13039/501100001809National Natural Science Foundation of China (No. 82430053); 10.13039/501100003399Shanghai Science and Technology Committee, China (Grant No. 24SF1904500); Innovation Program of 10.13039/501100003395Shanghai Municipal Education Commission, China (Grant No. 2023ZKZD33, Grant No. AI for Science); Tongji University Medicine-X Interdisciplinary Research Initiative, China (Grant No. 2025-0554-ZD-03); and Foundation of Shanghai Pulmonary Hospital, China (Grant No. LYRC202402, FKLY20004). Research in contextEvidence before this studyWe systematically searched PubMed for English-language clinical trials published between November 1980 and November 2022, using the terms: “induction” OR “neoadjuvant” AND “stage III” AND “non-small-cell lung cancer”. At the outset of our study, several clinical trials had shown that immune checkpoint blockade, either as monotherapy or combined with chemotherapy, conferred substantial pathological and survival benefits in resectable non-small-cell lung cancer (NSCLC) regardless of PD-L1 expression, including resectable stage III NSCLC patients. Nevertheless, owing to the limited enrolment of stage III patients and focus on resectable disease, the efficacy and safety of sintilimab-based induction therapy in potentially resectable stage III NSCLC stratified by PD-L1 expression requires further investigation.Added value of this studyOur clinical trial aimed to evaluate the safety and feasibility of sintilimab-based induction therapy in patients with potentially resectable stage III NSCLC on the basis of tumour PD-L1 expression. Our results showed that this approach enabled a subset of initially unresectable patients to reach R0 resection with a high major pathological response rate and provided favorable survival outcomes. Additionally, our study found that enhanced plasma cell signatures predicted improved response to immune checkpoint blockade, whereas elevated expression of AKR1C family genes was associated with resistance.Implications of all the available evidenceFindings from our study and existing evidence indicate that immunotherapy-based induction treatment is a feasible strategy for potentially resectable stage III NSCLC patients, which may provide an opportunity for receiving surgery and more effective treatment options for such patients. However, the small cohort of patients with PD-L1 expression of at least 50% precludes definitive conclusions regarding immunotherapy monotherapy. Thus, large-scale, multicentre trials are warranted to further validate these findings and identify predictive biomarkers for optimal candidates. Research in context Evidence before this study We systematically searched PubMed for English-language clinical trials published between November 1980 and November 2022, using the terms: “induction” OR “neoadjuvant” AND “stage III” AND “non-small-cell lung cancer”. At the outset of our study, several clinical trials had shown that immune checkpoint blockade, either as monotherapy or combined with chemotherapy, conferred substantial pathological and survival benefits in resectable non-small-cell lung cancer (NSCLC) regardless of PD-L1 expression, including resectable stage III NSCLC patients. Nevertheless, owing to the limited enrolment of stage III patients and focus on resectable disease, the efficacy and safety of sintilimab-based induction therapy in potentially resectable stage III NSCLC stratified by PD-L1 expression requires further investigation. Added value of this study Our clinical trial aimed to evaluate the safety and feasibility of sintilimab-based induction therapy in patients with potentially resectable stage III NSCLC on the basis of tumour PD-L1 expression. Our results showed that this approach enabled a subset of initially unresectable patients to reach R0 resection with a high major pathological response rate and provided favorable survival outcomes. Additionally, our study found that enhanced plasma cell signatures predicted improved response to immune checkpoint blockade, whereas elevated expression of AKR1C family genes was associated with resistance. Implications of all the available evidence Findings from our study and existing evidence indicate that immunotherapy-based induction treatment is a feasible strategy for potentially resectable stage III NSCLC patients, which may provide an opportunity for receiving surgery and more effective treatment options for such patients. However, the small cohort of patients with PD-L1 expression of at least 50% precludes definitive conclusions regarding immunotherapy monotherapy. Thus, large-scale, multicentre trials are warranted to further validate these findings and identify predictive biomarkers for optimal candidates. Introduction Lung cancer is still the leading cause of cancer-related death worldwide.1 Non-small-cell lung cancer (NSCLC) accounts for 80%–85% of all lung cancers, but only 30% of NSCLC patients are initially diagnosed with resectable disease.2,3 Multimodality therapies have been considered the standard treatment in recent decades, with surgical resection being the optimal treatment for stage II-III NSCLC patients.4,5 Preoperative chemotherapy can provide prolonged overall survival (OS) in patients with resectable NSCLC.6 However, the discrepancy in 5-year recurrence-free survival (RFS) and OS between preoperative chemotherapy and surgery alone is minimal, with only a 5 percentage point difference among patients with resectable NSCLC.6 PD-1/PD-L1 immune checkpoint inhibitor-based regimens have been approved as standard treatments for advanced or metastatic NSCLC and resected NSCLC without targetable molecular drivers, revolutionising treatment for these patients.7, 8, 9, 10, 11 Multiple large phase 3 clinical trials have confirmed that neoadjuvant chemoimmunotherapy significantly improved prognosis and pathological response compared to neoadjuvant chemotherapy alone in patients with resectable NSCLC.12, 13, 14, 15 Immunotherapy-based induction therapy boosts anti-tumour immunity before tumour and lymph node interventions,16 leading to better tumour/nodal downstaging, increased complete resection rates, and improved micrometastatic disease surveillance.13, 14, 15, 16, 17, 18 For initially unresectable stage III NSCLC, the standard treatment strategy is definitive chemoradiotherapy followed by consolidation immunotherapy, based on the phase 3 PACIFIC trial and GEMSTONE-301 trial, which showed median progression-free survival (PFS) of 16.8 and 14.3 months, respectively.19,20 From these results, the optimal treatment remains controversial and warrants further exploration in patients with potentially resectable stage III NSCLC who might convert to resectable disease after immunotherapy-based induction. Surgical intervention could enhance locoregional control and potentially improve patient outcomes, especially in individuals that do not have a pathological complete response to initial treatment, thus broadening the scope for achieving a cure. To date, there is limited evidence supporting the feasibility and effectiveness of neoadjuvant chemoimmunotherapy followed by surgery in improving survival among potentially resectable stage III NSCLC patients.21,22 Additionally, the predictive biomarkers of response and resistance in this specific patient subset remain unclear. Sintilimab, a domestically developed fully humanised IgG4 monoclonal antibody targeting PD-1, received approvals from the National Medical Products Administration as a first-line therapy in advanced or recurrent NSCLC in combination with chemotherapy.23 We initiated this trial to further evaluate the efficacy and safety of sintilimab-based induction therapy, and to identify predictive biomarkers of response in patients with potentially resectable stage III NSCLC. Methods Study design and participants LungMate-009 is an investigator-initiated, open-label, phase 2 trial of neoadjuvant sintilimab monotherapy or combined with platinum-based chemotherapy conducted at Shanghai Pulmonary Hospital. Eligible patients were aged 18–75 years and had histologically or cytologically confirmed, treatment-naive stage III NSCLC, based on the eighth edition of the American Joint Committee on Cancer staging system.24 All enrolled patients were deemed to have the possibility of converting to be resectable after neoadjuvant therapy by a multidisciplinary clinical team.25 An Eastern Cooperative Oncology Group performance score of 0 or 1 was required for every eligible patient.26 Patients were required to have adequate organ function, sufficient pulmonary function to tolerate subsequent surgery, and measurable lesions according to the Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. The main exclusion criteria included the presence of EGFR mutations or ALK translocations; histological evidence of small cell lung cancer; receiving any related anti-tumour treatment including surgery, local radiotherapy, cytotoxic drugs, targeted drugs, experimental therapies and so on; history of clinically significant malignancies within the past five years; comorbidity with unstable or ongoing systemic disease; use of systemic corticosteroid or other immunosuppressive therapy; previous or current interstitial lung disease; HIV positivity or acute or chronic hepatitis; active or previous autoimmune disease. Additional details are listed in the trial protocol (Supplementary File: Trial Protocol). The clinical stage of all patients was determined by routine clinical practice, including contrast-enhanced chest CT, brain MRI, bone scans, and abdominal CT scans at baseline. Fluorodeoxyglucose PET-CT scans were conducted if necessary. Percutaneous needle biopsy or endobronchial ultrasound biopsy was performed to determine the pathological diagnosis of the tumour. The status of mediastinal lymph nodes was evaluated through radiological imaging and/or pathological evaluation of endobronchial ultrasound biopsy. The lymph node with a diameter greater than 1.5 cm on CT or reported on PET-CT was considered radiologically positive. High suspicion of N1/2 lymph node involvement was determined by invasive staging approaches, including endobronchial ultrasound-guided transbronchial needle aspiration, endoscopic ultrasound-guided biopsy, or PET-CT. N3 lymph node metastasis was determined through pathological biopsy or PET-CT, and further pathological assessments were performed using endobronchial ultrasound- or supraclavicular lymph node biopsy. Ethics This study was performed in accordance with the guidelines of the Declaration of Helsinki and Good Clinical Practice. The trial protocol was reviewed and approved by the ethics committee of Shanghai Pulmonary Hospital (L21-214). All enrolled patients provided written informed consent before treatment. Procedures Patients with potentially resectable NSCLC were evaluated by a multidisciplinary team comprising medical oncologists, radiation oncologists, and thoracic surgeons. The related definitions of initially unresectable stage III NSCLC were provided in previously published clinical trials at our center (LungMate 003, NCT04379739),27 and were updated in our study. Potentially resectable NSCLC was characterised by the following characteristics and evaluated as having the probability of radical resection after sintilimab-based induction therapy by multidisciplinary assessment: (1) the tumour invading adjuvant unresectable tissue and organs, such as large vessels, the trachea, or primary bronchus, with the possibility of radical resection after tumour downgrading by induction therapy and the comprehensive preoperative evaluation; (2) patients diagnosed metastatic lymph nodes, with unresectable hilar lymph nodes, single-station N2 disease, multi-station N2 disease, or bulky N2 disease and patients could undergo mediastinal lymph node dissection after tumour downstaging through induction therapy based on preoperative evaluation; (3) pathological confirmed N3 disease, with the possibility of radical resection after lymph node downstaging by induction therapy; (4) R0 resection couldn't be achieved with pneumonectomy after preoperative evaluation at baseline, with the possibility of radical resection after induction treatment. Enrolled patients may present with multiple clinical characteristics. Patients with tumour PD-L1 expression level of less than 50% or unknown received 200 mg of sintilimab every 3 weeks in combination with platinum-based chemotherapy for four cycles, and those with tumour PD-L1 expression of 50% or more received four cycles of sintilimab monotherapy every 3 weeks before surgical resection intravenously. The platinum-based chemotherapy regimens were determined by the investigators according to institutional practice. These regimens included gemcitabine (1000 mg/m2) or paclitaxel (200 mg/m2) with carboplatin (5 mg/mL/min for area under the curve) for squamous or other histological subtype, and pemetrexed (500 mg/m2) with carboplatin (5 mg/mL/min for area under the curve) for adenocarcinoma. Following the completion of sintilimab-based induction treatment, surgical treatment was performed 21–28 days after the last dosage, provided it was deemed feasible by multidisciplinary evaluation. Patients also could undergo surgery early if they could not tolerate all 4 cycles of induction treatment or refused follow-up treatment. Otherwise, additional cycles were considered depending on the tumour response, patient's tolerance and strong willingness for surgical treatment, and the outcome of multidisciplinary evaluation. If the patient remained inoperable after induction therapy, therapeutic strategies were reformulated based on research protocols, the effectiveness of neoadjuvant therapy, and multidisciplinary assessment. For patients who were still ineligible for surgery or declined surgical intervention, radiotherapy was recommended, followed by consolidation immunotherapy. In cases where patients exhibited a response to induction therapy and declined radiotherapy, continued systemic treatment was advised based on drug toxicity. Eligibility for surgical resection was determined according to the standard procedures of research institutions and guidelines from the International Association for the Study of Lung Cancer.28 The status of lymph nodes was reevaluated through pathological diagnosis after induction treatment. Notably, for every patient with N2 or N3 lymph node metastases at baseline, evaluation for complete surgical resection was necessary after lymph node downstaging was confirmed by pathological diagnosis. Adjuvant treatment was formulated according to the NCCN guidelines and multidisciplinary evaluation, once they had recovered from surgical treatment. Additionally, all patients were recommended to receive full-body imaging surveillance every 3–6 months. Contrast-enhanced CT scans were performed to assess radiological tumour response every two treatment cycles before surgery. All changes in target tumour size were evaluated by specialists at the Department of Radiology, Shanghai Pulmonary Hospital, according to RECIST version 1.1. Patients underwent weekly laboratory blood tests to monitor related complete blood cell counts and biochemical parameters. Abnormal laboratory results and adverse events were graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE) version 5.0. Investigators analysed and concluded whether adverse events were related to the current treatment according to the study protocol and standard regulatory requirements. If treatment-related adverse events (TRAEs) occurred, the related treatment could be delayed, discontinued, and resumed only if certain criteria were met. Sintilimab treatment was permitted for interruption, delay, or discontinuity depending on patient tolerability. However, dose reductions were only allowed for chemotherapy drugs including gemcitabine, paclitaxel, pemetrexed, and carboplatin per the adjusted level of dosage specified in the trial protocol (Supplementary File: Trial Protocol). The next treatment cycle could be delayed by up to 3 weeks if hematological recovery was not achieved on day 21 (±3 days). During treatment, consent could be withdrawn if specific criteria were met, such as patient withdrawal of consent, intolerable toxicity, non-compliance, intercurrent conditions, or other investigator-determined reasons impacting the patient's safety. The assessment of PD-L1 expression levels was routinely recommended and determined using the 22C3 pharmDx assay (Agilent Technologies, CA, USA). The objective pathological response was assessed by specialists at the Department of Pathology, Shanghai Pulmonary Hospital. The percentage of residual viable tumours in resected primary tumours and lymph nodes was measured by pathologists and subsequently confirmed by two masked pathologists with 100% concordance. Major pathological response (MPR) was defined as the presence of 10% or fewer viable tumour cells in the resected primary tumour or viable metastatic carcinoma in lymph nodes. The definition of non-MPR was the presence of over 10% viable tumour cells in the resected primary tumour, and pathologic complete response was defined as the absence of viable tumour cells in both the resected tumour specimen and all sampled regional lymph nodes.28, 29, 30 Total RNA from fresh frozen tissues was extracted with TRIzol. Sequencing libraries were generated using a NEBNext Ultra RNA Library Prep Kit for Illumina, and index barcodes were added to attribute sequences to each sample. The libraries were pooled, and paired-end sequencing (2 × 150 bp reads) was performed using an Illumina NovaSeq 6000. Subsequently, raw fastq files were trimmed via fastp (v0.23.4),31 and aligned to the human reference genome (hg38) by STAR (v2.7.11b) with default settings.32 After obtaining the BAM files, read counts were summarised by featureCounts (v2.0.6),33 and TPMs (Trans Per Million) were calculated. Batch effects were adjusted using the “ComBat_seq” function in the Sva package. Genomic DNA was extracted from tumours using the QIAamp Fast DNA Tissue Kit (QIAGEN) following the manufacturer's protocol. Total DNA was quantified using the Qubit 2.0 Fluorometer (Life Technologies) and NanoDrop 2000 (Thermo Fisher Scientific), and its integrity was assessed with TapeStation (Agilent Technologies). For library construction, the DNA was fragmented to 180–280 bp using a Covaris ultrasonicator. WES libraries were prepared and captured using the Agilent SureSelect Human All Exon V6 kit, then sequenced on the Illumina NovaSeq 6000 platform with 150 bp paired-end reads. Raw fastq files were then trimmed via fastp (v0.23.4) and aligned to hg38 by BWA (v0.7.18).34 The Genome Analysis Toolkit (GATK, v4.5.0.0) was used to mark duplicates and base quality recalibration. Somatic SNVs and indels were then detected by GATK Mutect2, annotated by VEP (v111.0), and converted to MAF format by vcf2maf (v1.6.21),35 all with default parameters. Outcomes The primary endpoint was the MPR rate. Secondary endpoints included the proportion of patients achieving complete resection, radiological response (assessed per RECIST version 1.1), the proportion of patients experiencing serious adverse events (SAEs; assessed per NCI-CTCAE version 5.0), PFS, and OS. PFS was defined as the time from diagnosis until objective tumour progression or death. OS was defined as the period from diagnosis to death from any cause. Surgical outcomes included morbidity, mortality, and any complications occurring within the first 30 days post-operation. The toxicity profile was monitored for 3 months following the final dose of neoadjuvant treatment according to NCI-CTCAE (version 5.0) guidelines. Postoperative follow-up was conducted at three-month intervals during the first two years and at six-month intervals thereafter. The prespecified exploratory endpoint was the identification of predictive biomarkers and mechanisms of response and resistance to neoadjuvant chemoimmunotherapy in these patients. It may not be possible to assess the proportion of patients with disease progression or mortality at five years due to the current insufficient follow-up duration. Statistics Analyses related to the drug's safety, radiological response, PFS, and OS at 12 months and 24 months were conducted in the treated population, which included all patients who received at least one cycle of neoadjuvant treatment. The evaluation of the R0 resection rate, pathological response, and surgical outcomes was performed in patients who underwent surgical treatment. For the estimation of sample size, a one-sample proportion test based on an exponential distribution was used to calculate the sample size using the primary endpoint of this research (the MPR rate). Considering the MPR of neoadjuvant immune checkpoint blockade monotherapy in resectable lung cancer was 45% in the CheckMate-159 trial36 and 43% for neoadjuvant immune checkpoint blockade plus chemotherapy (LungMate-001) in stage IIIA NSCLC,37 compared to 20% for neoadjuvant chemotherapy,38 we expect the MPR of neoadjuvant sintilimab-based treatment to be 45% in this study, considering the potential for overestimation due to the limited sample size in these trials. The calculation was conducted with a significance level (α) of 0.05 (two-tailed) and a power of 95% (1-β). The one-sample rate test calculation indicates that with P0 = 20% and P1 = 45%, the required sample size is 42 for patients with R0 resection. Based on the data from our hospital's LungMate-001 trial (60%),37 we anticipate achieving R0 resection in 45% of patients, considering the advanced stage and higher burden of disease. Therefore, 93 patients need to be recruited. Considering a 5% loss to follow-up or dropout, at least 98 patients need to be recruited. Continuous data were analysed using the median and IQR. Categorical data were presented as frequency and percentage. The Clopper-Pearson method was applied to calculate the exact two-sided 95% confidence intervals (CIs). The Kaplan–Meier method was used to calculate PFS and OS along with corresponding 95% CIs. The median follow-up time and corresponding IQR were calculated using the reverse Kaplan–Meier method. The Mann–Whitney U test was used for comparisons between two independent samples, while the Kruskal–Wallis test was employed for analyses involving multiple independent samples. Detailed statistical tests were described in the corresponding figure legends. A two-sided p value of <0.05 was deemed significant. All statistical analyses were performed using R Studio version 4.1.2 and IBM SPSS Statistics v20.0. This study is registered with ClinicalTrials.gov, NCT04728724. Role of funding source The sponsor and funders had no role in study design, data collection, data analysis, interpretation, or manuscript writing. Results Between November 16, 2022, and September 11, 2023, 100 patients were assessed for eligibility at Shanghai Pulmonary Hospital, and 97 patients were included in the analyses (Fig. 1). Baseline characteristics of the patients are shown in Table 1. The median age was 66.0 years (IQR 60.0–71.0). 87 (90%) of 97 patients were male, and 71 (73%) had a history of smoking. 58 patients (60%) had an Eastern Cooperative Oncology Group performance status of 0. 53 patients (55%) had stage IIIA NSCLC at baseline, and 51 (53%) had squamous cell carcinoma. Additionally, 56 patients (58%) had N2 disease. Five patients (5%) received only one treatment cycle due to the refusal of subsequent treatment (Supplementary Table S1), and six patients (6%) patients received more than four cycles of neoadjuvant treatment (Supplementary Table S2). 75 patients (77%) completed four cycles of neoadjuvant treatment (range 1–6 cycles). Two patients with PD-L1 ≥50% received sinitilimab monotherapy (Fig. 2).Fig. 1The flow chart of the enrolled patients with potentially resectable stage III NSCLC. PD, Progressive disease; SD, Stable disease; PR, Partial response; MPR, major pathologic response; pCR, pathologic complete response.Table 1Baseline characteristics.CharacteristicParticipants, no. (%)Age, median (IQR), y66.0 (60.0–71.0)Sex, no. (%) Male87 (90) Female10 (10)ECOG, no. (%) 058 (60) 139 (40)Smoking status, no. (%) Never26 (27) Ever71 (73)Clinical T stage, no. (%) T110 (10) T214 (14) T312 (12) T461 (63)Nodal stage, no. (%) N013 (13) N120 (21) N256 (58) Single-station26 (27) Multi -station30 (31) N38 (8)Clinical stage, no. (%) IIIA53 (55) IIIB38 (39) IIIC6 (6)PD-L1 expression, no. (%) <1%33 (34) 1–49%18 (19) ≥50%2 (2) Unknown44 (45)Pathology, no. (%) Squamous cell carcinoma51 (53) Adenocarcinoma21 (22) Adenosquamous carcinoma1 (1) NSCLC-NOS2 (2) NSCLC22 (23)Treatment cycle, no. (%) 15 (5) 25 (5) 36 (6) 475 (77) >46 (6)Treatment, no. (%) Sintilimab monotherapy2 (2) Sintilimab plus chemotherapy95 (98)Percentages may not total 100 owing to rounding. IQR, interquartile range; ECOG, Eastern Cooperative Oncology Group; PD-L1, programmed cell death ligand-1; NSCLC-NOS, non-small-cell lung cancer-not otherwise specified.Fig. 2Swimmer plot of clinical characteristics and follow-up in the treated population who received at least one cycle of treatment. PD, Progressive disease; SD, Stable disease; PR, Partial response; MPR, major pathologic response; pCR, pathologic complete response; LUAS, lung adenosquamous cell carcinoma; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma; NSCLC, non-small-cell lung cancer; PD-L1, programmed cell death ligand-1. The flow chart of the enrolled patients with potentially resectable stage III NSCLC. PD, Progressive disease; SD, Stable disease; PR, Partial response; MPR, major pathologic response; pCR, pathologic complete response. Table 1Baseline characteristics.CharacteristicParticipants, no. (%)Age, median (IQR), y66.0 (60.0–71.0)Sex, no. (%) Male87 (90) Female10 (10)ECOG, no. (%) 058 (60) 139 (40)Smoking status, no. (%) Never26 (27) Ever71 (73)Clinical T stage, no. (%) T110 (10) T214 (14) T312 (12) T461 (63)Nodal stage, no. (%) N013 (13) N120 (21) N256 (58) Single-station26 (27) Multi -station30 (31) N38 (8)Clinical stage, no. (%) IIIA53 (55) IIIB38 (39) IIIC6 (6)PD-L1 expression, no. (%) <1%33 (34) 1–49%18 (19) ≥50%2 (2) Unknown44 (45)Pathology, no. (%) Squamous cell carcinoma51 (53) Adenocarcinoma21 (22) Adenosquamous carcinoma1 (1) NSCLC-NOS2 (2) NSCLC22 (23)Treatment cycle, no. (%) 15 (5) 25 (5) 36 (6) 475 (77) >46 (6)Treatment, no. (%) Sintilimab monotherapy2 (2) Sintilimab plus chemotherapy95 (98)Percentages may not total 100 owing to rounding. IQR, interquartile range; ECOG, Eastern Cooperative Oncology Group; PD-L1, programmed cell death ligand-1; NSCLC-NOS, non-small-cell lung cancer-not otherwise specified. Baseline characteristics. Percentages may not total 100 owing to rounding. IQR, interquartile range; ECOG, Eastern Cooperative Oncology Group; PD-L1, programmed cell death ligand-1; NSCLC-NOS, non-small-cell lung cancer-not otherwise specified. Swimmer plot of clinical characteristics and follow-up in the treated population who received at least one cycle of treatment. PD, Progressive disease; SD, Stable disease; PR, Partial response; MPR, major pathologic response; pCR, pathologic complete response; LUAS, lung adenosquamous cell carcinoma; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma; NSCLC, non-small-cell lung cancer; PD-L1, programmed cell death ligand-1. Of the 39 patients who didn't have surgery, five (13%) received chemoimmunotherapy as maintenance therapy, and ten (26%) received radiotherapy alone. Detailed treatment information is shown in Fig. 1. The primary reasons that patients were not candidates for surgery included the patient's decision (23 [59%] of 39), incomplete resectability (12 [31%]), and financial reasons and travel restrictions (4 [10%]). Among the 39 patients without surgical treatment, 14 (36%) received radiotherapy, and 25 (64%) did not receive any local treatment. Postoperative complications were observed in 21 (36%) of 58 patients who underwent surgery. The most common complications were air leakage (15 [26%] of 58 patients) and pleural effusion (10 [17%]). According to the trial protocol, two patients received only one cycle due to severe TRAEs and withdrew consent. All patients had at least one TRAE during neoadjuvant treatment, and 40 (41%) of 97 patients had adverse events of grade 3 or worse. The most common treatment-related grade 3 or worse adverse events were neutropenia (35 [36%] of 97 patients), leukopenia (18 [19%]), asthenia or fatigue (7 [7%]), alanine aminotransferase increased (6 [6%]), paraesthesia (4 [4%]), and aspartate aminotransferase increased (4 [4%]; Table 2). 15 (15%) of 97 patients had dose reduction due to grade 4 neutropenia and grade 3 or 4 thrombocytopenia. TRAEs leading to treatment delay occurred in one patient (1%), and treatment discontinuation occurred in two (2%). None of the adverse events reported during neoadjuvant treatment led to surgery delay. Four patients (4%) had severe TRAEs, including severe thrombocytopenia and liver and kidney injury. No adverse events led to death during neoadjuvant treatment. Immune-related adverse events of any grade occurred in 51 (53%) of 97 patients, including 21 (22%) who had hyperglycemia and 40 (41%) who had rash. No grade 3 or worse immune-related adverse events were observed.Table 2Treatment-related adverse events occurring in ≥10% of patients.Any treatment-related adverse eventGrade 1–2, no. (%)Grade 3, no. (%)Grade 4, no. (%)Any TRAE57 (59)23 (24)17 (18)Asthenia or Fatigue70 (72)7 (7)0Neutropenia38 (39)22 (23)13 (13)Anaemia70 (72)1 (1)0Alopecia70 (72)00Leucopenia51 (53)17 (18)1 (1)Decreased appetite or Anorexia54 (56)00Arthralgia53 (55)2 (2)0Myalgia52 (54)00Cough44 (45)00Lost weight39 (40)1 (1)0Constipation39 (40)1 (1)0Alanine aminotransferase increased34 (35)3 (3)3 (3)Skin disorders (Rash)38 (39)00Paraesthesia34 (35)4 (4)0Thrombocytopenia31 (32)4 (4)1 (1)Back pain31 (32)00Aspartate aminotransferase increased25 (26)2 (2)2 (2)Pulmonary infection22 (23)00Hyperglycaemia21 (22)00Vomiting18 (19)00Fever17 (18)00Gamma-glutamyl transferase increased14 (14)3 (3)0Dyspnoea16 (16)00Diarrhoea12 (12)00Increased creatinine10 (10)1 (1)0Haemoptysis10 (10)00Dacryorrhea7 (7)1 (1)0Hypertension6 (6)1 (1)0TRAE, treatment-related adverse event. Table 2Treatment-related adverse events occurring in ≥10% of patients.Any treatment-related adverse eventGrade 1–2, no. (%)Grade 3, no. (%)Grade 4, no. (%)Any TRAE57 (59)23 (24)17 (18)Asthenia or Fatigue70 (72)7 (7)0Neutropenia38 (39)22 (23)13 (13)Anaemia70 (72)1 (1)0Alopecia70 (72)00Leucopenia51 (53)17 (18)1 (1)Decreased appetite or Anorexia54 (56)00Arthralgia53 (55)2 (2)0Myalgia52 (54)00Cough44 (45)00Lost weight39 (40)1 (1)0Constipation39 (40)1 (1)0Alanine aminotransferase increased34 (35)3 (3)3 (3)Skin disorders (Rash)38 (39)00Paraesthesia34 (35)4 (4)0Thrombocytopenia31 (32)4 (4)1 (1)Back pain31 (32)00Aspartate aminotransferase increased25 (26)2 (2)2 (2)Pulmonary infection22 (23)00Hyperglycaemia21 (22)00Vomiting18 (19)00Fever17 (18)00Gamma-glutamyl transferase increased14 (14)3 (3)0Dyspnoea16 (16)00Diarrhoea12 (12)00Increased creatinine10 (10)1 (1)0Haemoptysis10 (10)00Dacryorrhea7 (7)1 (1)0Hypertension6 (6)1 (1)0TRAE, treatment-related adverse event. Treatment-related adverse events occurring in ≥10% of patients. TRAE, treatment-related adverse event. According to RECIST 1.1 criteria, 80 (82%; 95% CI 75–90) of the 97 patients had an overall response, and all had a partial response (80 of 80, 100%). 14 (14%) of 97 patients had stable disease, and one (1%) had progressive disease during neoadjuvant therapy and received subsequent radiotherapy. 50 (52%) of 97 patients had downstaging of lymph nodal status (Table 3). 58 patients (60%) received surgical resection successfully, and all 58 had complete tumour resection without associated mortality. One patient with tumour PD-L1 expression of at least 50% who received sintilimab monotherapy underwent surgery successfully. Lobectomy was the most frequent surgical procedure (49 [84%] of 58), and two patients received segmentectomy due to insufficient pulmonary function. 36 (62%) of 58 patients received video-assisted thoracoscopic surgery. Details on surgical procedures are shown in Table 3. 38 (66%; 95% CI 53–78) of 58 patients who received surgery had MPR; 23 (40%; 95% CI 27–53) had pathological complete response (Fig. 3). Of 14 patients with stable disease, three (21%) reached MPR, whereas 35 (44%) of 80 patients with a partial response reached MPR. The distribution of MPR stratified by clinicopathological characteristics is presented in Table 4. The median PFS and OS were not reached in either the surgical or non-surgical cohort, and a significant difference was observed between patients with surgical treatment and those without surgery regarding PFS and OS (PFS hazard ratio [HR] 0.22, 95% CI 0.08–0.57; p = 0.0001; Supplementary Figure S1A; OS HR 0.15, 95% CI 0.03–0.72; p = 0.0002; Supplementary Figure S1B).Table 3Evaluation of efficacy of neoadjuvant treatment.No. (%)Radiological response Partial response (PR)80 (82) Stable disease (SD)14 (14) Progressive disease (PD)1 (1) Not evaluable2 (2)Overall response rate, % (95% CI)82 (75–90)Pathological response (N = 58) Non-MPR20 (34) MPR38 (66) pCR23 (40)Residual tumor cell, median (IQR), %3.5 (0–45.0)Downstaging of lymph nodal status (N = 50) N3–N27 (14) N3–N11 (2) N2–N117 (34) N2–N013 (26) N1–N012 (24)Type of surgery (N = 58) Video-assisted thoracoscopic surgery (VATS)36 (62) Robot-assisted thoracoscopic surgery (RATS)7 (12) Thoracotomy15 (26)Margins Positive0 (0) Negative58 (100)Surgical resection (N = 58) Lobectomy49 (84) Pneumonectomy2 (3) Sleeve Lobectomy5 (9) Segmentectomy2 (3)Median duration of surgery (IQR), min152.5 (120.0–180.0)Median blood loss (IQR), mL50.0 (50.0–100.0)Median duration of hospital stay postoperatively (IQR), days6.0 (4.0–7.0)Median chest drainage time (IQR), days9.0 (5.0–16.5)Percentages may not total 100 owing to rounding. MPR, major pathologic response; pCR, pathologic complete response; IQR, interquartile range; CI, Confidence Interval; min, minutes.Fig. 3The baseline clinical characteristics and pathological response of patients who received surgery (n = 58). SD, Stable disease; PR, Partial response; MPR, major pathologic response; pCR, pathologic complete response; LUAS, lung adenosquamous cell carcinoma; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma; NSCLC, non-small-cell lung cancer; PD-L1, programmed cell death ligand-1.Table 4Distribution of the MPR rate in patients with different clinicopathological characteristics.MPR, no (%)Patients38 (66)Age, no. (%) ≥65 y21 (55) <65 y17 (45)Sex, no. (%) Male35 (92) Female3 (8)Smoking history, no. (%) Ever31 (82) Never7 (18)PD-L1 expression, no. (%) <1%11 (29) 1–49%6 (16) ≥50%1 (3) Unknown20 (53)Pathology, no. (%) Squamous cell carcinoma21 (55) Adenocarcinoma6 (16) Adenosquamous carcinoma1 (3) NSCLC10 (26)Clinical T stage, no. (%) T16 (16) T24 (11) T34 (11) T424 (63)Nodal stage, no. (%) N04 (11) N17 (18) N223 (61) N34 (11)Clinical stage, no. (%) IIIA19 (50) IIIB16 (42) IIIC3 (8)Percentages may not total 100 owing to rounding. MPR, major pathologic response; PD-L1, programmed cell death ligand 1; NSCLC, non-small-cell lung cancer. Table 3Evaluation of efficacy of neoadjuvant treatment.No. (%)Radiological response Partial response (PR)80 (82) Stable disease (SD)14 (14) Progressive disease (PD)1 (1) Not evaluable2 (2)Overall response rate, % (95% CI)82 (75–90)Pathological response (N = 58) Non-MPR20 (34) MPR38 (66) pCR23 (40)Residual tumor cell, median (IQR), %3.5 (0–45.0)Downstaging of lymph nodal status (N = 50) N3–N27 (14) N3–N11 (2) N2–N117 (34) N2–N013 (26) N1–N012 (24)Type of surgery (N = 58) Video-assisted thoracoscopic surgery (VATS)36 (62) Robot-assisted thoracoscopic surgery (RATS)7 (12) Thoracotomy15 (26)Margins Positive0 (0) Negative58 (100)Surgical resection (N = 58) Lobectomy49 (84) Pneumonectomy2 (3) Sleeve Lobectomy5 (9) Segmentectomy2 (3)Median duration of surgery (IQR), min152.5 (120.0–180.0)Median blood loss (IQR), mL50.0 (50.0–100.0)Median duration of hospital stay postoperatively (IQR), days6.0 (4.0–7.0)Median chest drainage time (IQR), days9.0 (5.0–16.5)Percentages may not total 100 owing to rounding. MPR, major pathologic response; pCR, pathologic complete response; IQR, interquartile range; CI, Confidence Interval; min, minutes. Evaluation of efficacy of neoadjuvant treatment. Percentages may not total 100 owing to rounding. MPR, major pathologic response; pCR, pathologic complete response; IQR, interquartile range; CI, Confidence Interval; min, minutes. The baseline clinical characteristics and pathological response of patients who received surgery (n = 58). SD, Stable disease; PR, Partial response; MPR, major pathologic response; pCR, pathologic complete response; LUAS, lung adenosquamous cell carcinoma; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma; NSCLC, non-small-cell lung cancer; PD-L1, programmed cell death ligand-1. Table 4Distribution of the MPR rate in patients with different clinicopathological characteristics.MPR, no (%)Patients38 (66)Age, no. (%) ≥65 y21 (55) <65 y17 (45)Sex, no. (%) Male35 (92) Female3 (8)Smoking history, no. (%) Ever31 (82) Never7 (18)PD-L1 expression, no. (%) <1%11 (29) 1–49%6 (16) ≥50%1 (3) Unknown20 (53)Pathology, no. (%) Squamous cell carcinoma21 (55) Adenocarcinoma6 (16) Adenosquamous carcinoma1 (3) NSCLC10 (26)Clinical T stage, no. (%) T16 (16) T24 (11) T34 (11) T424 (63)Nodal stage, no. (%) N04 (11) N17 (18) N223 (61) N34 (11)Clinical stage, no. (%) IIIA19 (50) IIIB16 (42) IIIC3 (8)Percentages may not total 100 owing to rounding. MPR, major pathologic response; PD-L1, programmed cell death ligand 1; NSCLC, non-small-cell lung cancer. Distribution of the MPR rate in patients with different clinicopathological characteristics. Percentages may not total 100 owing to rounding. MPR, major pathologic response; PD-L1, programmed cell death ligand 1; NSCLC, non-small-cell lung cancer. The data cutoff date for this interim analysis was September 24, 2025, with a median follow-up of 31.0 months (IQR 27.5–32.6). 24 (25%) of 97 patients had disease progression, and 16 (16%) died. The median duration of PFS and OS was not reached in these populations (Fig. 4A and B). In the treated population, 12-month PFS was 85% (95% CI 78–92), and 24-month PFS was 78% (95% CI 71–87). 12-month OS was 93% (95% CI 88–98), and 24-month OS was 88% (95% CI 81–94).Fig. 4Kaplan–Meier curves of progression-free survival (A) and overall survival (B) in the treated population who received at least one cycle of neoadjuvant treatment. CI, Confidence Interval. Kaplan–Meier curves of progression-free survival (A) and overall survival (B) in the treated population who received at least one cycle of neoadjuvant treatment. CI, Confidence Interval. 12-month PFS for patients with surgery was 95% (95% CI 89–100), and 24-month PFS was 90% (95% CI 82–98), and both the 12-month and 24-month OS were 97% (95% CI 92–100). Of the 38 patients who had a MPR, including 23 patients with pathologic complete response, 12-month and 24-month PFS was 95% (95% CI 88–100) and 92% (95% CI 84–100), and both the 12-month and 24-month OS were 97% (95% CI 92–100). Among the 23 patients with pathologic complete response, only one patient had disease progression at the time of analysis. The median PFS and OS were not reached in patients with MPR or non-MPR, and no significant differences in PFS and OS were observed among patients with different pathological responses (PFS HR 0.52, 95% CI 0.10–2.55; p = 0.19; OS HR 0.53, 95% CI 0.03–8.53; p = 0.24; Supplementary Figure S1C and D). To explore predictive biomarkers for immunotherapy in NSCLC, we prospectively collected 48 tumour samples from patients who underwent surgical resection. All these tumour samples, including 9 baseline and 39 surgical samples, underwent bulk RNA sequencing. Additionally, whole-exome sequencing was performed on eight baseline samples for genomic characterisation (Fig. 5A). Our subsequent analysis primarily focused on transcriptomic variations in tumour samples, as gene mutation frequency showed no significant differences between MPR and non-MPR patients (Supplementary Figure S2A).Fig. 5Plasma cell signature predicts response to sintilimab. (A) Overview of the SPH cohort displaying patient data, including treatment response, PD-L1 IHC levels, vital and relapse status, as well as RNA-seq and WES sample types. (B) Differential gene expression between MPR (n = 31) and non-MPR groups (n = 17) in tumour samples. (C) The clinicopathologic and immunophenotypic characteristics of tumour samples (n = 48) were outlined across four predefined groups. Clinical attributes, including pathology, clinical TNM stage, smoking history, and PD-L1 IHC expression, were presented in the top rows. Immune and stromal scores from the ESTIMATE algorithm are depicted below. The following heatmap depicted the relative abundance of immune cells calculated by xCell, expression of immune effector molecules, and immune checkpoint molecules. Data were analysed by the Kruskal–Wallis test. (D) The proportion of immune phenotype in tumour samples (n = 48) across the four clusters was illustrated by a bar chart. (E) Uniform manifold approximation and projection (UMAP) plot of all cells colored by major cell types according to canonical markers. The outer circle indicated the proportion of Scissor+ or Scissor− cells in each cell cluster. (F) The waterfall plot demonstrated scaled plasma cell scores in baseline patients (n = 9). The box plot below compared the BCR Shannon scores between the MPR and non-MPR groups, which were analysed by a one-sided Mann–Whitney U test. (G) Multivariate Cox regression analysis of overall survival using multiple signatures in OAK (n = 344) and POPLAR cohort (n = 95). ∗, p ≤ 0.05; ∗∗, p ≤ 0.01; ∗∗∗, p ≤ 0.005. ∗∗∗∗, p ≤ 0.001. pCR, pathologic complete response; MPR, major pathologic response; PD-L1, programmed cell death ligand-1; NSCLC-NOS, non-small-cell lung cancer-not otherwise specified; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma; CI, Confidence Interval; LN, lymph node; TMB, Tumour Mutational Burden, BCR, B Cell Receptor; DDIR, DNA damage immune response; TIDE, Tumour Immune Dysfunction and Exclusion. Plasma cell signature predicts response to sintilimab. (A) Overview of the SPH cohort displaying patient data, including treatment response, PD-L1 IHC levels, vital and relapse status, as well as RNA-seq and WES sample types. (B) Differential gene expression between MPR (n = 31) and non-MPR groups (n = 17) in tumour samples. (C) The clinicopathologic and immunophenotypic characteristics of tumour samples (n = 48) were outlined across four predefined groups. Clinical attributes, including pathology, clinical TNM stage, smoking history, and PD-L1 IHC expression, were presented in the top rows. Immune and stromal scores from the ESTIMATE algorithm are depicted below. The following heatmap depicted the relative abundance of immune cells calculated by xCell, expression of immune effector molecules, and immune checkpoint molecules. Data were analysed by the Kruskal–Wallis test. (D) The proportion of immune phenotype in tumour samples (n = 48) across the four clusters was illustrated by a bar chart. (E) Uniform manifold approximation and projection (UMAP) plot of all cells colored by major cell types according to canonical markers. The outer circle indicated the proportion of Scissor+ or Scissor− cells in each cell cluster. (F) The waterfall plot demonstrated scaled plasma cell scores in baseline patients (n = 9). The box plot below compared the BCR Shannon scores between the MPR and non-MPR groups, which were analysed by a one-sided Mann–Whitney U test. (G) Multivariate Cox regression analysis of overall survival using multiple signatures in OAK (n = 344) and POPLAR cohort (n = 95). ∗, p ≤ 0.05; ∗∗, p ≤ 0.01; ∗∗∗, p ≤ 0.005. ∗∗∗∗, p ≤ 0.001. pCR, pathologic complete response; MPR, major pathologic response; PD-L1, programmed cell death ligand-1; NSCLC-NOS, non-small-cell lung cancer-not otherwise specified; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma; CI, Confidence Interval; LN, lymph node; TMB, Tumour Mutational Burden, BCR, B Cell Receptor; DDIR, DNA damage immune response; TIDE, Tumour Immune Dysfunction and Exclusion. To assess transcriptomic differences across patients grouped by MPR status and sampling time, we performed principal component analysis and observed clear segregation between these groups (Supplementary Figure S2B, C, and D). Differential gene expression analysis was conducted on all tumour samples to further illustrate the transcriptomic features (Fig. 5B; negative-binomial exact test, FDR <0.05). The significant upregulation of immune-related genes, such as CCL18 and HLA-G, was observed in the MPR group, underscoring the outstanding role of the immune microenvironment in responders to immune checkpoint blockade. Additionally, the MPR group was also characterised by the upregulation of SCGB1A1, which was reported to enhance the efficacy of immune checkpoint blockade in NSCLC by promoting the infiltration of cytotoxic cells.39 Meanwhile, the upregulation of steroid hormone-related genes, such as AKR1C1, AKR1C2, and AKR1C3, was found in the non-MPR group, implicating the relation between metabolic dysfunction and inferior response to immune checkpoint blockade. Moreover, differential gene expression analysis in lymph node samples validated the upregulation of AKR1C3 in non-responders (Supplementary Figure S2E). Given the enrichment of immune-related genes in MPR patients, we subsequently investigated the immunological phenotype among four sample groups: baseline samples from MPR patients (MPR & Baseline), surgical samples from MPR patients (MPR & Surgery), baseline samples from non-MPR patients (non-MPR & Baseline), and surgical samples from non-MPR patients (non-MPR & Surgery). We first performed the xCell and MCP-counter algorithm to quantify the immune cell infiltration and found that the MPR & Surgery group was featured by increased immune cell infiltration, particularly CD8+T cells and macrophages (Fig. 5C and Supplementary Figure S3A). This group also possessed higher expression of effector molecules, such as GZMA and GZMK, as well as immune checkpoint molecules, such as PDCD1 and CTLA4 (Fig. 5C). Consistently, the MPR & Surgery group exhibited the highest proportion of the immune-enriched, non-fibrotic phenotype, highlighting the inflamed immune contexture in responders to immune checkpoint blockade (Fig. 5D).40 Prompted by the role of immune cells in immunotherapy efficacy, we aimed to identify the key cell cluster shaping tumour response to immune checkpoint blockade.41,42 Based on the Scissor algorithm, we integrated bulk RNA-seq data from the aforementioned 9 baseline samples with single-cell RNA-seq data from baseline samples derived from our previous studies.43 Owing to the high dimensionality of single-cell RNA-seq data, we found that plasma cells possessed the highest proportion of Scissor+ cells, which were associated with better response to immune checkpoint blockade (Fig. 5E). Consistently, gene set enrichment analysis also underscored the enrichment of immunoglobulin complex pathway and B cell-mediated immunity pathway in MPR & Baseline group (Supplementary Figure S3B). Subsequent correlation analysis revealed that plasma cells were significantly associated with a broad spectrum of anti-tumour cell subsets, indicating the potential value of plasma cells in predicting the response to immunotherapy (Supplementary Figure S3C). To better quantify plasma cell abundance, we identified a plasma cell signature comprising 13 transcript markers according to the Scissor data (Supplementary Methods). We then tested this signature in 9 baseline tumour samples, and found that all MPR patients exhibited elevated plasma cell scores (Fig. 5F), Moreover, plasma cell signature showed improved predictive power for response to immune checkpoint blockade in the baseline setting (area under the curve 0.70, 95% CI 0.28–1.00) compared to the plasma cell signature raised by Patil and colleagues (Supplementary Figure S3D).44 Furthermore, the BCR Shannon score, indicative of antigen-specific plasma cell response, was significantly higher in the MPR group at baseline (Fig. 5F; one-sided Mann–Whitney U-test p = 0.032). Although two non-MPR patients had exceptionally high plasma cell scores, the fairly low tumour mutational burden they harbored could bridge the gap between plasma cell abundance and immunotherapy efficacy (Supplementary Figure S2A). We further validated the plasma cell signature in two independent NSCLC immunotherapy cohorts including OAK and POPLAR cohorts,45,46 where elevated plasma cell scores were significantly associated with improved OS (Supplementary Figure S3E; OAK Log-rank p = 0.00020, POPLAR Log-rank p = 0.0027) and PFS (Supplementary Figure S3F; OAK Log-rank p = 0.0024, POPLAR Log-rank p = 0.023). Additionally, multivariate analysis confirmed that the plasma cell signature was independently associated with favorable clinical outcomes (Fig. 5G) and exhibited the highest predictive performance for treatment response in both OAK and POPLAR cohorts (Supplementary Figure S3G and H), suggesting its potential as an independent and robust predictor of response to immune checkpoint blockade compared with other proposed biomarkers. Inspired by the metabolic genes uncovered in the non-MPR group, we then measured the pathway variance and noted the significant diversity of metabolic pathways among all samples (Fig. 6A), deciphering the role of metabolic heterogeneity in resistance to immune checkpoint blockade. To further assess metabolic pathway activity, we conducted single-sample gene set enrichment analysis and found significant enrichment of lipid-associated metabolic pathways in non-MPR patients, particularly in their baseline samples (Fig. 6B). Furthermore, Gene Ontology enrichment analyses demonstrated the upregulation of steroid hormone-related pathways in the non-MPR & Baseline group, which corresponded to the elevated expression of AKR1C family genes observed in differential gene expression analysis (Figs. 5B and 6C). Notably, AKR1C family genes were significantly upregulated in non-MPR patients exclusively at baseline rather than after treatment, implicating their potential as biomarkers for identifying patients refractory to immune checkpoint blockade (Fig. 6D). Moreover, survival analyses in OAK and POPLAR cohorts substantiated the association between high expression of AKR1C family genes and poor response to immune checkpoint blockade (Fig. 6E).Fig. 6AKR1C family genes associate with resistance to immune checkpoint blockade. (A) Evaluation of variance between metabolism and immune pathways in SPH cohort (n = 48). p values were computed using the Mann–Whitney U test. (B) Clinicopathological information and metabolic pathway scores in SPH cohort (n = 48). Clinical attributes such as pathology, clinical TNM stage, smoking history, and PD-L1 IHC expression were displayed in the top rows. The following heatmap exhibited the metabolic pathway scores across four groups. (C) GO enrichment analysis of differentially expressed genes underscoring distinct biological pathways of MPR and non-MPR groups in pre-treatment and post-treatment settings. (D) In the SPH cohort (n = 48), AKR1C family gene scores were compared between MPR and non-MPR groups across all samples, baseline samples, and surgical samples. p values were computed by using the Mann–Whitney U test. (E) Kaplan–Meier curves for overall survival of two groups stratified by AKR1C family gene scores in the OAK cohort (n = 344) and POPLAR cohort (n = 95). MPR, major pathologic response; PD-L1, programmed cell death ligand-1; NSCLC-NOS, non-small-cell lung cancer-not otherwise specified; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma. AKR1C family genes associate with resistance to immune checkpoint blockade. (A) Evaluation of variance between metabolism and immune pathways in SPH cohort (n = 48). p values were computed using the Mann–Whitney U test. (B) Clinicopathological information and metabolic pathway scores in SPH cohort (n = 48). Clinical attributes such as pathology, clinical TNM stage, smoking history, and PD-L1 IHC expression were displayed in the top rows. The following heatmap exhibited the metabolic pathway scores across four groups. (C) GO enrichment analysis of differentially expressed genes underscoring distinct biological pathways of MPR and non-MPR groups in pre-treatment and post-treatment settings. (D) In the SPH cohort (n = 48), AKR1C family gene scores were compared between MPR and non-MPR groups across all samples, baseline samples, and surgical samples. p values were computed by using the Mann–Whitney U test. (E) Kaplan–Meier curves for overall survival of two groups stratified by AKR1C family gene scores in the OAK cohort (n = 344) and POPLAR cohort (n = 95). MPR, major pathologic response; PD-L1, programmed cell death ligand-1; NSCLC-NOS, non-small-cell lung cancer-not otherwise specified; LUAD, lung adenocarcinoma; LUSC, Lung squamous cell carcinoma. Discussion Locally advanced stage III disease accounts for approximately 20%–30% of patients with newly diagnosed NSCLC.47,48 Stage III NSCLC encompasses a highly heterogeneous group of cancers characterised by considerable variation in tumour size, invasion, and nodal involvement,49 most of which are deemed as unresectable diseases.50 Guidelines for distinguishing resectability from unresectability in stage III NSCLC are currently not uniformly defined, leading to significant variability in treatment strategies among different centers based on the experience of thoracic surgeons and diagnostic imaging.51,52 For patients with unresectable stage III NSCLC, the current standard recommendation is concurrent chemoradiotherapy followed by consolidation therapy with durvalumab for those without progressive disease, with a median PFS of 16.8 months and the 12-month PFS of 55.9%.19,53 The significant pathological and survival benefits associated with neoadjuvant immunotherapy plus chemotherapy stimulated the use of this treatment regimen for this population, with the expectation of converting their tumour to a complete resection status. In our study, surgery was successfully performed in 58 patients (60%), and all achieved R0 resection with MPR and pathologic complete response rates of 66% and 40%, respectively. Subsequently, the ORR in these patients reached 82%. To our knowledge, our study is the largest clinical trial evaluating the combination of PD-1 inhibitor monotherapy or combined platinum-based doublet chemotherapy in the neoadjuvant setting for potentially resectable stage III NSCLC to date, and provides a preliminary conclusion regarding the application of this treatment approach in these patients. Previous phase 3 clinical trials regarding neoadjuvant chemoimmunotherapy followed by surgery in patients with resectable NSCLC have revealed a remarkable improvement in pathological response and survival outcomes, revolutionising the treatment of resectable NSCLC.14,17,18,54 However, the role of surgical resection in potentially resectable stage III NSCLC patients remains controversial, even when the patients have achieved tumour downstaging after induction therapy. Moghanaki et al. didn't recommend that patients with unresectable stage III NSCLC undergo re-evaluation for surgery due to elevated risks of perioperative complications and mortality, alongside the ambiguity surrounding the potential benefits of surgical intervention.55 However, Dickhoff et al. suggested that the resectability of unresectable stage III NSCLC can be reevaluated through a more effective systemic treatment strategy in patients without distant metastases, potentially leading to further improvements in survival outcomes and locoregional control.56 Deng et al. retrospectively analysed the surgical and pathological outcomes of 51 patients with initially unresectable stage cIIIB NSCLC and found that 31 patients (61%) achieved the conversion of resectability successfully and underwent subsequent radical surgery after chemoimmunotherapy, with the MPR rate of 67.7%. The follow-up results of this study revealed that the patients who underwent surgery presented significant improvements in survival benefits compared to those without surgical resection. Ricciuti et al. conducted an international, multicenter, retrospective study assessing the pathological and survival outcomes in borderline resectable and unresectable stage III NSCLC patients treated with neoadjuvant PD-1/PD-L1 inhibitors combined with chemotherapy and found that 84 (75.0%) patients received surgery successfully and exhibited significantly improved survival outcomes compared with those without surgery.57 In a previous LungMate-002 clinical trial, R0 resection was performed in 25 (68.4%) patients with potentially resectable NSCLC after neoadjuvant chemoimmunotherapy, and the patients receiving surgical treatment also showed prolonged survival benefit compared to non-surgical patients.58 In our study, 58 (60%) patients underwent surgery safely, and all patients achieved R0 resection, with a high MPR rate, which was similar to the results in the aforementioned studies. Only two patients underwent pneumonectomy in our study, and the incidence of postoperative complications was acceptable, with no patient deaths within 90 days of surgery. Additionally, our study also demonstrated acceptable survival outcomes among surgical patients, with a 24-month PFS rate of 90%, which was comparable to the outcomes of the NADIM trial, with the PFS rate of 87.9% at 24 months in the per-protocol population.59 Thus, these results indicated that neoadjuvant chemoimmunotherapy may be feasible for patients with potentially resectable stage III NSCLC, and surgical resection may become a viable treatment option following such therapy and through reassessment. The strategy of sintilimab-based induction therapy was well tolerated in this study, and the frequency of grade 3 or higher TRAEs was similar to that observed in the randomised pembrolizumab group of KEYNOTE-671, reporting 44.9% of the participants experienced TRAEs of grade 3 or higher.14 In our study, most of the patients had TRAEs of grade 1–2, and no grade 3 or higher immune-related adverse events were observed. The incidence of common AEs was similar to that of other studies with sintilimab-based therapy,37,60 and no new safety signals were detected. From these results, neoadjuvant sintilimab-based therapy appears to be safe and well-tolerated in potentially resectable stage III NSCLC. This study also aimed to identify predictors of response and resistance to immune checkpoint blockade. We identified plasma cell signatures as a key biomarker of response to immune checkpoint blockade. Previous studies have identified a strong association between increased intratumoural plasma cells and longer overall survival in NSCLC patients treated with PD-L1 blockade.44 Differing from previous research, our study first validated the predictive value of plasma cells in the neoadjuvant setting. Meanwhile, our plasma cell signature derived from the Scissor algorithm performed better than previously reported signatures in two independent cohorts. We also discovered a strong correlation of plasma cells with other immune cell subclusters, implicating potential interactions and cross-talk within the microenvironment. Therefore, despite the inflamed microenvironment in MPR patients, specific immune cell subclusters may serve as mechanistic triggers and pivotal regulators of clinical response. Further efforts are needed to exploit the mechanisms by which plasma cells contribute to anti-tumour immunity. Apart from the tumour immune phenotype, we found that metabolic heterogeneity, particularly variations in steroid hormone-related pathways, correlated with resistance to neoadjuvant immunotherapy. Notably, the AKR1C family genes, which play a key role in steroid hormone-related pathways, were found to be elevated in baseline samples of non-MPR patients. Previous studies have demonstrated that AKR1C family genes contribute to estrogen metabolism by catalysing the reduction of estrone to 17β-estradiol, thereby promoting drug resistance in various cancer types.61,62 This suggests the potential utility of AKR1C family genes as predictive biomarkers or therapeutic targets. Several AKR1C1 inhibitors have been identified, offering new opportunities for targeted interventions. Moreover, we also found the upregulation of other lipid-associated metabolic pathways in non-MPR patients, such as bile acid and bile salt biosynthesis, emphasising the critical role of metabolic mechanisms in resistance to immunotherapy. Several limitations in our study need to be acknowledged. First, as a phase 2 trial, it could not compare the efficacy of neoadjuvant immunotherapy followed by surgery with concurrent chemoradiotherapy followed by durvalumab, requiring a large-scale, multicentre, randomised controlled trial for further validation. Secondly, the single-center design limits the generalisability of our findings, which necessitates confirmation with multicenter data. Additionally, the short follow-up period may compromise the robustness of certain analyses, calling for an extended follow-up to ensure comprehensive evaluation. Third, the absence of detectable PD-L1 expression in a significant proportion of the enrolled patients may introduce a potential source of bias into our findings. This limitation stems primarily from the limited tissue yield from endobronchial ultrasound procedures, and these specimens are often processed for cytological diagnosis rather than histological examination at our institution. This approach facilitates accurate pathological assessment under conditions of limited tumour tissue availability63 and remains consistent with ethical standards in clinical management. The limited number of enrolled patients with PD-L1 expression ≥50% during consecutive enrollment precluded robust conclusions regarding the efficacy of immunotherapy monotherapy in this subgroup, underscoring the need for a comprehensive clinical trial. Fourth, the heterogeneity in treatment strategies following the induction phase may lead to undertreatment of certain patient subgroups and compromise the accurate evaluation of patient outcomes. Finally, our bioinformatics analysis is exploratory and conducted in a subset of resected patients. Due to limited sample size and validation solely in adjuvant settings, the predictive value of our findings regarding plasma cell infiltration and AKR1C gene expression may be overstated, requiring further validation in other neoadjuvant immune checkpoint blockade cohorts. In conclusion, sintilimab-based induction treatment is safe and effective for patients with potentially resectable stage III NSCLC, demonstrating a high MPR rate and manageable TRAEs. Furthermore, this treatment approach may provide an opportunity to convert potentially resectable into resectable disease status. Additionally, we identified the value of increased plasma cell signatures in predicting response to neoadjuvant immune checkpoint blockade and the association between AKR1C family gene expression and resistance to neoadjuvant immune checkpoint blockade. Contributors Peng Zhang, Wenxin He, Xuefei Hu, Minwei Bao and Yi Bao: conceptualisation, design, funding acquisition, data validation; Tao Ge and Hongyi Zhang: data analysis, formal analysis, writing (original manuscript) and data verification; Xinsheng Zhu, Meixin Teng, Yifei Zhou: follow-up, data access and data analysis; Suyu Wang, Dongliang Bian, Lele Zhang, Haiyang Hu, Qiji Guo, Jincheng Su, Shiqi Hu and Huansha Yu: data collection and data interpretation; Yirui Zhou, Jie Huang, Jie Yang and Yuming Zhu: data validation, investigation; all author: writing (review and editing). All authors have read and approved the final version of the manuscript. Data sharing statement Raw transcriptomic and genomic data generated in this study have been deposited in the Genome Sequence Archive for Human database under accession number HRA012938. Publicly available datasets used in our analyses are listed in the Supplementary Materials with corresponding accession numbers and data processing protocols. Codes used for analyses and generation of figures are available on GitHub (https://github.com/hongyizhang18/sintilimab). The related clinical data in our study are not publicly available due to involving patient privacy, but can be available from the corresponding author upon reasonable request. Declaration of interests The authors declare that there were no conflicts of interest or competing interests.