Clinical studyPatients

Adult patients with histologically or cytologically confirmed NSCLC eligible for anatomic resection were enrolled. Clinical stages IB, II and selected stage IIIA (T3 N1, T4 with satellite nodule in the same lung N0/N1, selected T1a–T2b N2 cases considered suitable for primary surgical approach by the multidisciplinary tumor board) according to the Union International Contre le Cancer (UICC) eighth edition were eligible. Additional inclusion criteria (see Supplementary information for full study protocol) are women and men ≥18 years of age, Eastern Cooperative Oncology Group performance score ≤1, exclusion of extensive mediastinal lymph node metastases (multilevel N2, N3), exclusion of distant metastases, measurable target tumor before immunotherapy using standard imaging techniques, sufficient pulmonary function to undergo curative lung cancer surgery (percentage of predicted forced expiratory volume at 1 s (ppFEV1) > 30%, percentage of predicted diffusion capacity of the lung for carbon monoxide (ppDLCO) > 30%, percentage of predicted maximal oxygen consumption (ppVO2max) ≥ 10 ml min−1 kg−1 (if cardiopulmonary exercise testing was mandated per local guidelines)), adequate hematological, hepatic and renal function parameters, sufficient cardiac left ventricular function defined as left ventricular ejection fraction ≥50% documented either by echocardiography or multigated acquisition (MUGA) scan, ability and willingness to provide written informed consent and to comply with the study protocol and with the planned surgical procedures. Gender was determined based on self-report.

Exclusion criteria (see Supplementary information for full study protocol) are: active or history of autoimmune disease or immune deficiency; subjects with a condition requiring systemic treatment with either corticosteroids (>10 mg daily prednisone equivalents) or other immunosuppressive medications within 14 days of study drug administration; subjects who have undergone organ transplant or allogeneic stem cell transplantation; ppFEV1 < 30%, ppDLCO < 30%, ppVO2max < 10 ml min−1 kg−1 (if cardiopulmonary exercise testing was mandated per local guidelines); uncontrolled or significant cardiovascular disease (including myocardial infarction, stroke or transient ischemic attack, uncontrolled angina, clinically significant arrhythmias, QTc prolongation >480 ms, pulmonary hypertension); history of other clinically significant cardiovascular disease (including cardiomyopathy, congestive heart failure, pericarditis, pericardial effusion, coronary artery stent occlusion, deep venous thrombosis); cardiopulmonary disease-related requirement for daily supplemental oxygen; subjects with a history of myocarditis, regardless of etiology; elevated troponin T or I; active neurological disease; active malignancy or previous malignancy within the past 3 years; known history of positive test for human immunodeficiency virus (HIV-1 and HIV-2) or known acquired immunodeficiency syndrome; any positive test result for hepatitis B virus or hepatitis C virus (HCV) indicating presence of virus, for example hepatitis B surface antigen (Australia antigen) positive or hepatitis C antibody (anti-HCV) positive (except if HCV RNA negative); any other disease, metabolic dysfunction, physical examination finding or clinical laboratory finding that contraindicates the use of an investigational drug, may affect the interpretation of the results or may render the patient at high risk from treatment complications; receipt of live attenuated vaccine within 30 days before the first dose of study medication; peripheral neuropathy National Cancer Institute Common Terminology Criteria for Adverse Events grade ≥2; history of gastric perforation or fistulae in past 6 months; serious or nonhealing wound, ulcer or bone fracture within 28 days before enrollment; major surgery within 28 days before enrollment except staging mediastinoscopy, diagnostic video-assisted thoracoscopic surgery (VATS) or implantation of a venous port-system; any other concurrent preoperative antineoplastic treatment including irradiation, pregnant or breastfeeding women; insufficient cardiac left ventricular function defined as left ventricular ejection fraction <50% by echocardiography or MUGA scan; confirmed history of encephalitis, meningitis or uncontrolled seizures in the year before informed consent; subjects with a history of severe toxicity or life-threatening toxicity (grade 3 or 4) related to previous immune therapy (for example anti-CTLA-4 or anti-PD-1/PD-L1 treatment or any other antibody or drug specifically targeting T cell co-stimulation or immune checkpoint pathways) except those that are unlikely to reoccur with standard countermeasures (for example, hormone replacement after endocrinopathy); history of severe or life-threatening (grade 3 or 4) infusion-related reactions to previous immuno therapy; previous treatment with LAG-3 targeting agent; participation in another interventional clinical study within the past 3 months before inclusion or simultaneous participation in other clinical studies; previous treatment with nivolumab or relatlimab; previous immunotherapy for lung cancer; criteria that in the opinion of the investigator preclude participation for scientific reasons, for reasons of compliance or for reasons of the subject’s safety; or any contraindications against nivolumab or relatlimab.

Study design and treatment

NEOpredict-Lung (NCT04205552) is an open-label, randomized phase 2 trial (see Supplementary information for the version of the study protocol pertinent to this report). This manuscript reports results from arms A and B of the study, which treated patients with two doses of nivolumab (240 mg every 14 days per intravenous infusion, arm A) or nivolumab and relatlimab (240 and 80 mg, respectively, every 14 days per intravenous infusion, arm B). The dose and schedule of nivolumab and nivolumab plus relatlimab were selected to align with the biweekly administration of nivolumab in other studies of preoperative ICI combinations in NSCLC patients, such as NEOSTAR10 and CheckMate 816 (ref. 13). It is supported by findings of the ongoing study RELATIVITY-020 (ref. 39), which explores multiple doses and schedules of relatlimab-based combinations.

The study was not designed to formally compare both treatment arms. No gender analysis was performed because of the limited cohort sizes and the nature of the study.

Patients were randomly assigned (1:1) via an interactive web response system provided by Alcedis GmbH (https://www.alcedis.de/en); there was no stratification or blinding. Patients were treated for a maximum of two cycles (14 days each), which was followed by standard of care surgery and, if clinically indicated, postoperative medical therapy and/or radiotherapy. Surgery and postoperative treatments were not part of the clinical study intervention. All patients are followed up to 12 months within the study protocol. Subsequent follow-up is provided within standard of care.

Endpoints

The primary study endpoint is the number of patients proceeding to curatively intended surgery of NSCLC within 43 days of the initiation of study therapy.

Secondary endpoints include: the objective response rate (RECIST 1.1) before surgery; the pathological response rate (complete pathological responses defined as the absence of viable tumor cells on routine H&E staining of resected tumors and lymph nodes, and rate of MPRs defined as 10% or less viable tumor cells on routine H&E staining of resected tumors); the R0 resection rate; the DFS rate at 12 months per RECIST 1.1; the OS rate at 12 months; the safety and tolerability of preoperative immunotherapy; and morbidity and mortality within 90 days of surgery.

Exploratory endpoints are assessed in tumor and lymph node samples, blood cells, plasma and serum.

All primary and secondary endpoints were assessed in the intention-to-treat population and in the full analysis set.

Clinical data are captured in the clinical database using a proprietary electronic case report system provided by Alcedis GmbH (https://www.alcedis.de/en).

Assessments

Radiographic and nuclear imaging assessments at baseline were conducted within standard of care at the study sites. Specifically, all 60 patients underwent whole-body imaging by FDG-PET/CT. For exclusion of brain metastases, 41 patients underwent contrast-enhanced brain magnetic resonance imaging (MRI) scanning, 18 patients underwent contrast-enhanced brain CT scanning (because of contraindications or intolerance of MRI imaging or unavailability of an MRI slot within the protocol-defined screening period). In one patient with stage IB NSCLC no brain imaging was performed as per Dutch guidelines. All patients underwent CT or PET/CT imaging immediately before surgery. Radiographic response was evaluated at the study sites following RECIST 1.1. Exploratory analyses were conducted on nuclear imaging data acquired before surgery.

Baseline assessments included the collection of tumor tissue samples for centrally performed exploratory analyses. Diagnostic tumor tissue was obtained by endobronchial ultrasound-guided biopsy (31 patients), CT-guided transthoracic biopsy (17 patients) or by other approaches including bronchoscopy-guided forceps biopsy and miniprobe/navigation-guided biopsy (13 patients). For mediastinal staging, 47 patients underwent systematic endobronchial ultrasound including sampling of suspicious lymph nodes, and 11 patients had staging mediastinoscopy.

Histology and biomarker studies were conducted within standard of care at the study sites. PD-L1 expression by tumor cells was assessed locally using the primary antibody clone 22C3 (DAKO/Agilent M3653) following validated protocols with continuous external quality assurance (QUIP, UK NEQAS, NordiQC).

Additional tumor tissue samples were collected during surgery, and blood samples were collected at protocol-defined time points.

Statistical analyses

Based on published results of a study with preoperative nivolumab9 each study arm included up to 30 evaluable patients with the expectation that at least 26 of 30 patients treated in each study arm will undergo curatively intended surgery within 6 weeks of initiation of study treatment. At maximum 4 of 30 patients may experience a delay of curatively intended surgery beyond day 43 (with study treatment being administered on day 1), either because of toxicities or disease progression, to declare the study arm feasible. Continuous monitoring of prespecified stopping boundaries was applied to facilitate early termination of nonfeasible study arms to reduce patient risks. Details can be reviewed in the clinical study protocol (Supplementary information).

All secondary parameters were evaluated in an explorative or descriptive manner, providing means, medians, ranges, standard deviations and/or confidence intervals.

Trial oversight

The protocol and amendments were approved by the responsible ethics committees and competent regulatory authorities at each participating study site and country. In the legislature of the study sponsor and study site Essen the Ethics Committee of the Medical Faculty of the University Duisburg-Essen, Essen, Germany, granted primary approval on 10 September 2019 (19-8828-AF). The competent regulatory authority in the legislature of the study sponsor and study site Essen, the Paul-Ehrlich-Institut (Federal Institute for Vaccines and Biomedicines), Langen, Germany, granted primary approval on 27 November 2019 (EudraCT-Nr. 2109-007278-29, Vorlage-Nr. 3834/01). For study site Hasselt, approval was granted by the Ethics Committee OLV Ziekenhuis VZW, Aalst, Belgium (EudraCT-Nr. 2109-007278-29 Pilot 262-SM001, Reference 202/082), and the Federal Agency for Medicines and Health Products, Brussels, Belgium (EudraCT-Nr. 2109-007278-29 Pilot 262, 1240640 M). For study site Amsterdam, approval was granted by the METC—The Netherlands Cancer Institute, Antoni van Leeuwenhoek (NKI-AVL), Amsterdam, The Netherlands (NL72532.031.20), and by the Centrale Commissee Mensgebonden Onderzoek, The Hague, The Netherlands (Decree NL72532.031.21 CA).

The study was conducted according to the principles of the Declaration of Helsinki and the International Conference on Harmonization Good Clinical Practice guidelines.

All patients provided written informed consent before enrollment. The study is sponsored by the University Hospital Essen and was designed by employees of the sponsor, who were also study investigators.

A data safety monitoring committee, which is independent of the sponsor and the study investigators, reviewed all safety and efficacy data, including radiographic and pathological response data.

The clinical data were collected by the investigators, analyzed by statisticians employed by a contract research organization commissioned by the sponsor, and interpreted by the authors. Authors had full access to the data and are responsible for all content and editorial decisions.

Metabolic hybrid imaging

As per national and international practice guidelines, patients received FDG-PET/CT (or PET) at initial staging. Patients treated at study site Essen underwent a second FDG-PET/CT scan before surgery to confirm curative resectability. Images were acquired at a median of 4 days (range 1–29 days) before surgery. Imaging data were collected on three different PET/CT scanner types (Biograph Vision 600 (Siemens Healthineers), Biograph mCT (Siemens Healthineers), Vereos (Philips Healthcare)). It was ascertained that each individual patient underwent both scans on the same scanner type. Data acquisition started 67 ± 18 min (PET/CT 1) and 72 ± 12 min (PET/CT 2) after injection of 305 ± 58 MBq FDG (PET/CT 1) and 280 ± 58 MBq FDG (PET/CT 2), respectively. Patient handling and data processing were performed as detailed elsewhere40. After attenuation correction metabolic response rates were estimated according to PERCIST 1.0 (ref. 28).

Phenotyping of peripheral blood T cells

T cell phenotypes were determined by multiparametric flow cytometry. Briefly, cryopreserved peripheral blood mononuclear cells were thawed and rested overnight in RPMI medium supplemented with 10% FCS, 100 U ml−1 penicillin and 100 µg ml−1 streptomycin (PAA Laboratories) at 37 °C in a 5% CO2 atmosphere. Antibody staining of cell surface molecules (30 min, 4 °C) was followed by fixation and permeabilization for staining of intracellular markers (30 min, 4 °C). Stained samples were analyzed using a Gallios flow cytometer (Beckman Coulter) and Kaluza software (Beckman Coulter). Antibodies and gating strategy are depicted in Supplementary Fig. 2a.

Phenotyping of immune cell subsets in resected tumorsDissection of resected tumors

Tumor tissue was put in 1 ml of digestion medium (Dulbeccoʼs modified Eagleʼs medium/F12/HEPES solution supplemented with penicillin/streptomycin and 1% BSA and containing collagenase, hyaluronidase and DNAse I) and cut into small pieces. To facilitate dissociation the tissue was incubated for 40 min at 37 °C and pipetted every 10 min during the incubation period. The resulting cell suspension was transferred to a 50-ml centrifuge tube and centrifuged at 300g for 10 min at ambient temperature. The pellet was resuspended in trypsin/EDTA and incubated for 5 min at ambient temperature. After inactivation of the trypsin by Dulbeccoʼs modified Eagleʼs medium/F12/HEPES solution containing 10% FCS, the cell suspension was again triturated and filtered through a 40-µm cell strainer. After washing the filter with 50 ml of phosphate-buffered saline (PBS) the cells were centrifuged at 400g for 5 min at ambient temperature. Following one more washing step with PBS, cell number and viability was measured using the NucleoCounter NC-3000 and one to two million cells per vial were cryopreserved in FCS-containing 10% DMSO.

Flow cytometry

The cryopreserved tumor cell suspensions were analyzed in batches using two panels of antibodies. The staining method, antibodies and gating strategy for T lymphocyte subsets (Supplementary Fig. 2b) have been described previously41. Myeloid immune cells were detected using a separate antibody panel and gating strategy (Supplementary Fig. 2c). Flow cytometry was run on a CytoFLEX LX (Beckman Coulter) using the CytExpert v.2.3 software. Final data analysis was performed using FlowJo Software v.10 (Tree Star).

Gene expression analysesRNA isolation and quantification

For nucleic acid isolation, two to four sections each 10-µm thick (depending on sample size) from the respective formalin-fixed paraffin-embedded (FFPE) tissue sample were used. In total, RNA isolation could be performed on 46 resected tumors as well as 17 paired biopsies. Isolation procedures have been carried out semiautomatically on the Maxwell purification system (Maxwell RSC RNA FFPE Kit; Promega, cat. no. AS1440). All steps were performed following the respective protocol provided by the manufacturer. Total RNA was eluted in 50 µl RNase-free water and quantified using the RNA broad-range assay on a Qubit 2.0 fluorometer (Life Technology). One microliter of sample isolate volume was diluted for each quantification. RNA was stored at −80 °C until further use.

NanoString CodeSet design

Fluorescently barcoded RNA probes were synthesized and provided by NanoString. In total, gene expression was quantified using the dedicated PanCancer Immune Profiling panel as well as the PanCancer Pathway panel. Both panels consisted of the identical 40 reference and 770 individual target genes. The PanCancer Pathway panel comprises key players of the Notch, APC (Wnt), Hedgehog, transforming growth factor β, MAPK, STAT, PI3K and RAS signaling pathways as well as chromatin modification, transcriptional regulation, DNA damage control, cell cycle and apoptosis. The PanCancer Immune Profiling panel comprises targets associated with the various immunological processes and pathways of a host anti-cancer immune response. In total, both panels combined cover 1,398 different genes. For both panels, one sample served as a blank.

Digital gene expression analysis by hybridization

Digital gene expression analysis was performed on the NanoString nCounter platform, utilizing the NanoString MAX/FLEX system. A minimum of 100 ng of total RNA sample input was hybridized to the probes for 21 h at 65 °C. Subsequent cartridge preparation was performed in a NanoString PrepStation using the high-sensitivity protocol. Finally, the cartridge was scanned on the DigitalAnalyzer (NanoString) at 555 fields-of-view.

Gene expression analysis

NanoString data was normalized and cleaned using NanoTube (v.1.6.0)42, entailing three steps. First, counts were scaled by comparing the geometric mean of positive control features between samples. Second, genes in which at least 50% of samples are <2 s.d. above the mean of negative controls were removed. Third, counts were scaled by comparing the geometric mean of housekeeping genes between samples. Afterwards, differential expression analysis was performed using the quasi-likelihood F-test approach of EdgeR (two-sided, v.3.40.0)43. First, genes differentially expressed between sample types (resected tumor versus biopsy) were determined, while correcting for additive batch effects induced by pathological response (MPR = 1/0) and tumor classification (LUAD, lung squamous cell carcinoma, large-cell neuroendocrine carcinoma, sarcomatoid). Second, genes differentially expressed between MPR and no MPR were determined separately within each sample type and study arm. Reproducibility was ensured by implementing above analysis as a Snakemake44 workflow.

Genome sequencingDNA isolation and quantification

For tumor samples, one to four FFPE sections (10-µm thick, number depending on sample size) were lysed for genomic DNA isolation. Isolation was performed semiautomatically on the Maxwell purification system (Maxwell RSC DNA FFPE Kit; Promega, cat. no. AS1450) as specified by the manufacturer. DNA was eluted in 50 µl of RNase-free water and quantified fluorescently for library preparation using a Qubit 2.0 fluorometer (Life Technology) with its appertaining DNA broad-range assay. Corresponding normal DNA was isolated from blood or peripheral blood mononuclear cells using routinely available QIAGEN technology. DNA was stored at −20 °C before use.

Sequencing and genomic variant calling

Whole-exome sequencing was performed using the Twist Human Core + RefSeq + Mitochondrial Panel (Twist Bioscience), and 2 × 100-bp fragment sizes were sequenced using a NovaSeq6000 (Illumina). Demultiplexing of sequenced reads was achieved using bcl2fastq (v.2.2). Further data analysis was performed using our open-source Snakemake workflow dna-seq-varlociraptor (v.3.24, https://github.com/snakemake-workflows/dna-seq-varlociraptor), entailing the following steps. Adapter trimming was performed using Cutadapt (v.4.1, https://doi.org/10.14806/ej.17.1.200). Quality was monitored using MultiQC (v.1.14)45 including FASTQC (v.0.11.9, https://www.bioinformatics.babraham.ac.uk/projects/fastqc/), Somalier (v.0.2.18)46 and samtools (v.1.14)47. Reads were mapped to GRCh38 using bwa-mem (v.0.7.17, https://doi.org/10.48550/arXiv.1303.3997) and deduplicated using Picard-Tools (v.2.26). Base qualities were recalibrated using GATK (v.4.2)48. Single nucleotide variants and small indels were detected using Freebayes (v.1.3.6, https://doi.org/10.48550/arXiv.1207.3907) and classified into events of interest (somatic in biopsy or resection, germline) using Varlociraptor (v.8.3)49. Variant calls were distinguished from noise by controlling the (Bayesian) local false discovery rate (FDR) using Varlociraptor. Variant annotation (with impact, previous knowledge) was performed using VEP (v.109.3)50. Extraction of variants of interest was performed using vembrane (v.1.0)51. Specifically, for Fig. 2a, variants were filtered to be nonsynonymous, having a REVEL score >0.5 if available (that is, being predicted as pathogenic), having a gnomad allele frequency <0.2, being not marked as benign or likely benign in ClinVar and impacting one of the TCGA LUAD 500 cancer genes. Missing whole-exome sequencing data was complemented with results from panel sequencing (TSO500) whenever available. To identify genes that had altered variant allele frequencies (VAFs) comparing the diagnostic biopsy and the resected tumor, genes defined by oncobk (https://www.oncokb.org/cancer-genes) were inspected. To adjust for the different tumor cell content between biopsies and resected tumors, probabilities were calculated that the variants were not present in the normal sample of the same patient and that the VAF had changed before surgery. Only variants that were not marked by ClinVar as benign or likely benign and had a REVEL score >0.7 are reported in Supplementary Fig. 3.

Inference of subclonal diversityTumor purity estimation

Previous estimates p1 and p2 of the tumor purity of samples from resected tumors were obtained by two independent pathologists evaluating sections stained with H&E. For the other samples, a posterior estimate of the tumor purity of each sample was obtained as follows. We plotted the somatic VAF distribution of the pretherapeutic biopsy and the resected tumor samples of each patient. For this, the maximum a posteriori allele frequency estimates provided by Varlociraptor without adjusting for purity were used (that is, no sample contamination assigned, see https://varlociraptor.github.io/docs/calling). The expectation is that without copy number variants any somatic variant may at most have a VAF equal to the tumor purity. Read sampling variance and copy number variation can generate peaks beyond the tumor purity. For resection samples, we proceeded as follows: Let v be the highest VAF of the distribution or a threshold for which higher VAFs could as well be explained by sampling or copy number variation. If v was consistent with the previous estimates (that is, within the interval [p1,p2]) and the previous estimates were agreeing to a sufficient degree (p2 − p1 ≤ 0.2) we reported v as the posterior purity. Otherwise, we considered the posterior purity as unknown (28 of 56 cases). For samples in which the resected tumor had a posterior purity, we compared the distribution of the pretherapeutic biopsy and the resected tumor, and inferred a posterior estimate by scaling the biopsy distribution to match the shape of the resection distribution. Such scaling was possible in all investigated cases.

Subclonal diversity

For patients with posterior purity estimates, subclonal diversity was visualized in the following way. During tumor evolution, each somatic mutation that does not lead to cell death can be seen as an event generating a new subclone. We made the simplifying assumption that each nonlethal somatic mutation during development of the tumor generates one new subclone. Thus, the number of somatic variants can be seen as a proxy for the number of subclones, and each somatic variant can be considered as a representative of the subclone that originates in it. Note that this neglects the fact that multiple somatic variants can occur during one cell division. However, under the assumption that all considered samples have a similar somatic mutation rate, the subclone counts obtained would still be proportional to the true number of subclones, and thereby comparable across patients.

Thus, for each patient, we obtained the sufficiently relevant subclones by considering variants with posterior probability ≥0.95 according to Varlociraptor for being somatic in the pretreatment biopsy or in the resected tumor, and purity adjusted VAF ≥ 0.1. To be able to be certain that a variant is detectable in both the pretreatment biopsy and the resected tumor, we further filtered them such that there would be an expectation that they would be represented by at least two reads if occurring at the same frequency in the respective other sample (pretreatment biopsy for resected tumor; resected tumor for pretreatment biopsy). Patients in whom both pretreatment biopsy and resected tumor had no such somatic variants/subclones after filtering were omitted because they would not allow any statement about subclonal gains and losses. Variants with VAF = 0.0 in the resected tumor but VAF ≥ 0.1 in the pretreatment biopsy were then counted as ‘lost subclones’ following study therapy. Variants with VAF = 0.0 in the pretreatment biopsy but VAF ≥ 0.1 in the resected tumor were counted as ‘gained subclones’ following study therapy. Note that because the pretreatment biopsy may not represent the entire primary tumor, a ‘gain’ is not distinguishable from enrichment of a variant that was spatially missed in the biopsy.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.