Integrative multi-omics profiling reveals distinct evolutionary and immunogenic features of brain oligometastasis in lung adenocarcinoma Authors: Rongxin Liao, Qian Yu, Yusheng Huang, Hengqiu He, Mengmeng Song, Xuan Gao, Lu Shen, Yihao Tao, Lin Li, Gang Xiao, Rui Kong, Cancan Wang, Shunping Huang, Xiaoyue Zhang, Zaicheng Xu, Rongrong Zhou, Zhenzhou Yang, Yuan Peng Published: 2026 ABSTRACT AbstractBackgroundEmerging evidence has demonstrated that lung cancer patients with brain oligometastases (oligo-BMs) could benefit from local radical therapies. Despite rapid advancements in the treatment of oligometastatic disease, the molecular determinants governing the oligometastatic phenotype in the central nervous system remain elusive.MethodsWe performed 1021-panel sequencing, transcriptome profiling, DNA methylation mapping, and multiplex immunohistochemistry on 20 paired primary lung adenocarcinoma and oligo-BMs specimens to delineate their evolutionary dynamics and microenvironmental characteristics.ResultsA pronounced intertumor heterogeneity was observed between primary tumors (PTs) and oligo-BMs, while intratumoral heterogeneity was conserved across lesions. Subclonal analysis and phylogenetic reconstruction demonstrated that oligo-BMs exhibited predominant polyclonal seeding patterns and parallel progression trajectories. Moreover, oligo-BMs displayed a more immunosuppressed microenvironment compared to PTs, characterized by attenuated immunogenic cell death signatures, downregulation of immune-activated pathways, and diminished infiltration of activated immune cells (including B cells, NK cells, and Th1 cells). The methylation levels at functional genomic regions were highly concordant between PTs and oligo-BMs. Notably, we identified NLGN1 as a potential regulator of oligo-BM, where the hypermethylation at its genebody regions was mechanistically associated with transcriptional upregulation. Clinical validation revealed that NLGN1 was specifically overexpressed in BMs compared to PTs and extracranial metastases, correlating with advanced pathological stages and poor prognosis. In vivo studies further confirmed NLGN1 promotes BM in mice.ConclusionsOur findings provide novel insights into the evolutionary trajectory, immune landscape and methylation pattern of oligo-BMs, potentially paving the way for the development of innovative therapeutic strategies for lung cancer patients with oligo-BMs.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13073-026-01664-4.Keywords: LUAD, Brain oligometastasis, Evolutionary dynamics, Microenvironmental characteristics, NLGN1 Background Brain metastasis (BM) in advanced non-small cell lung cancer (NSCLC) is a prevalent and life-threatening complication, accounting for a substantial proportion of cancer-related mortality and posing formidable challenges in clinical management and treatment strategies [1, 2]. Approximately 10–20% of NSCLC patients present with BM at initial diagnosis, and up to 40% will subsequently develop BM during disease progression. With advancements in tumor biology, oligometastatic disease has emerged as a clinically distinct state of metastasis characterized by a restricted number of lesions and constrained dissemination potential. In contrast, polymetastatic tumors represent a more disseminated metastatic pattern with a higher systemic tumor burden. Specifically, brain oligometastases (oligo-BMs) refer to an indolent course in which metastases are confined to the brain with limited intracranial burden, characterized by tumor dormancy, reduced metastatic potential, and a more favorable prognosis. Consequently, management strategies for oligo-BMs focus on achieving complete tumor eradication rather than palliative approaches, aiming to further improve clinical outcomes for this patient population [3–5]. However, the limited understanding of the molecular mechanisms underlying the oligometastatic state has hindered the development of reliable methods for accurately distinguishing true oligometastases from early stages of widespread metastasis. As a result, some patients with undetected systemic progression may be misclassified and subjected to unnecessarily aggressive local therapies. The pathogenesis of BM involves a multistep cascade comprising the dissemination of tumor cells from primary lesions, disruption of the blood-brain barrier, adaptation to a new environment, and finally colonization in distant organs. This process is orchestrated by both intrinsic tumor cells properties, including genetic heterogeneity of primary tumors (PTs), transcriptional regulation, and epigenetic modifications in metastatic cells, as well as extrinsic interactions with the intracranial immune niche. Given the identification of major events in the metastatic cascade, numerous omics studies have made efforts to elucidate the molecular basis of BM. Several candidate driver genes, such as MYC, YAP1, and MMP13 amplifications, have been identified as contributors to BM formation in lung adenocarcinoma (LUAD). The genetic alterations driving the development of oligo-BMs are poorly understood, and it remains unclear whether BM tumor cells acquire neurotropic properties during metastatic progression. Brain is considered an immunologically privileged organ, creating a “hostile” environment that impedes the seeding and colonization of metastatic tumor cells compared with other organs. Several publications have revealed a highly immunosuppressive microenvironment in BMs compared to PTs, with reduced B and T-cell abundance and increased neutrophil infiltration [6, 7]. However, the immune profile of oligo-BMs has not been extensively studied. To date, limited research has investigated the role of DNA methylation events in lung cancer tumourigenesis and progression, and even fewer studies have examined methylation changes in BM. Moreover, these studies primarily concentrated on individual genes or small genes panels, and NGS-based methylation analyses of primary and paired BM samples from NSCLC patients have been scarce. Although current omics profiling of patient specimens has enhanced our understanding of the systems basis underlying brain metastasis [6, 7], most existing studies have predominantly focused on either primary lesions or metastasized brain tumors using single-level omics approaches. Comparative multiomics analysis of matched primary tumors and metastases remains rare due to the difficulty of obtaining brain metastatic tissues, hindering a comprehensive depiction of molecular changes linked to the development of metastasis. Furthermore, previous studies have primarily centered on conventional brain metastasis, with limited exploration into the mechanisms responsible for the formation of oligo-BM. Consequently, there is an urgent need for comprehensive molecular analyses of PTs and their paired brain oligometastatic lesions. Herein, we conducted an integrative multi-omics analysis of 20 pairs of LUAD and their corresponding brain oligometastatic tumors, incorporating multi-omics profiling, including 1021-panel sequencing, transcriptome profiling, DNA methylation mapping and multiplex immunohistochemistry (mIHC) (Fig. 1). This comprehensive approach enables systematic characterization of evolutionary trajectory from PTs to oligo-BMs, dynamic remodeling of the tumor immune microenvironment (TIME), and epigenetic reprogramming events associated with oligometastatic progression. Our findings provide novel insights into the molecular mechanisms driving oligometastasis formation and progression, with potential implications for refining therapeutic strategies in this clinically distinct patient population. Workflow for the collection, sequencing, multi-omic analysis, and validation of primary LUAD and paired oligo-BM samples Methods Study population and sample collection We enrolled 20 patients with brain oligometastatic LUAD who underwent surgical resection of both the PTs and the corresponding oligo-BMs at the Second Affiliated Hospital of Chongqing Medical University and Xiangya Hospital of Central South University between 2010 and 2021. The major inclusion criteria for this study were: histologically confirmed LUAD, stage IV disease according to the 8th edition of the American Joint Committee on Cancer staging system, with only one metastatic lesion confined to the brain at the time of initial diagnosis of brain metastasis, no progression at 3 or more months after brain metastasis. Regional lymph node involvement was considered part of the PT and not categorized as metastatic disease. In addition, between 2021 and 2024, a total of 71 brain oligometastatic lesions from LUAD patients who received first-line immunotherapy were collected at the Second Affiliated Hospital of Chongqing Medical University for multiplex immunohistochemistry staining of PD-L1 and CD8A. At the same institution, 7 liver metastases, 9 bone metastases, and 16 lymph node metastases from LUAD patients were collected for immunohistochemistry (IHC) of NLGN1 to assess its organ-specific expression. Clinical and histopathological information was retrieved from electronic medical records. This study received approval from the Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University, and all patients provided written informed consent before participation. RNA/DNA dual extraction from formalin-fixed and paraffin-embedded (FFPE) tissue FFPE tissue sections were subjected to deparaffinization using mineral oil. Specifically, 800µL of mineral oil (Fisher Scientific, #AC415080010) and 180µL of Buffer PKD were combined with the sections. Proteinase K was added to facilitate tissue digestion, followed by incubation at 56 °C for 15 min. To achieve phase separation, the mixture was centrifuged, and the aqueous phase was then chilled for 3 min to precipitate RNA. After centrifuging at 20,000 g for 15 min, the RNA-containing supernatant was carefully collected for further extraction, while the DNA remained in the pellet. Nucleic acids were isolated using the AllPrep DNA/RNA Mini Kit (QIAGEN, #80204) according to the manufacturer’s guidelines. RNA was eluted in nuclease-free water, while DNA was eluted in 0.5X Buffer ATE. Target DNA sequencing For tumor DNA, 400 to 800 ng of DNA was fragmented to achieve a peak size of 200 to 250 bp using a Covaris S2 ultrasonicator (Covaris, Inc). Indexed next-generation sequencing (NGS) libraries were then constructed with the NEBNext Ultra DNA Library Prep Kit for Illumina. To enrich for specific regions, a custom-designed panel (Integrated DNA Technologies, Inc.) that spans approximately 1.5 Mbp and targets 1,021 cancer-associated genes was utilized for hybridization. Sequencing of the indexed libraries was performed in a 100-bp paired-end configuration on a DNBSEQ-T7RS sequencer (MGI Tech), generating 5 Gb of data for FFPE. Raw data processing Sequenced reads were aligned to the reference human genome (GRCh37) using BWA (version 0.6.2) with default settings after adaptor trimming and the removal of low-quality reads [8]. For tumor DNA, duplicate reads were identified and eliminated with the MarkDuplicates function in Picard (version 2.25.5; Broad Institute). This study employed the OncoTOP software [9], which is based on a comprehensive genomic profiling assay targeting 1,021 genes frequently mutated in cancer, covering an approximate genomic region of approximately 1.6 megabases (Mb). RNA next generation sequencing (RNA-seq) The NEBNext rRNA Depletion Kit (Human/Mouse/Rat) (NEB #Z1955E) was selected to specifically remove targeted ribosomal RNA. For RNA samples with a percentage of RNA fragments > 200 nucleotides (DV200) ≤ 50%, the fragmentation step was omitted, and the samples were directly prepared for library construction. Following rRNA depletion and, where applicable, fragmentation, cDNA synthesis and library preparation for NGS were carried out using the NEBNext® Ultra™ II Directional RNA Library Prep Kit (NEB#E7760L). The resulting libraries were quantified using a Qubit 3.0 (life Invitrogen, USA) and their quality assessed via the LabChip GX Touch system (PerkinElmer, USA). Ribosomal RNA sequences were filtered out by aligning the processed reads to the NCBI rRNA database using Bowtie2 (version 2.2.8) [10] after trimming adaptor sequences and removing low-quality reads with Fastp (version 0.20.0) [11]. The rRNA-free clean reads were then aligned to the human reference genome (hg19) using STAR (version 2.6.1) [12]. Gene expression was quantified using StringTie (version 2.0.6) [13], and expression levels were normalized to transcripts per million. DNA methylation mapping The quality of the samples was evaluated using the following DNA inclusion criteria: a total amount of 30 ng as measured by a Qubit 3.0 fluorometer (Life Technologies) and a fragment length distribution with a dominant peak around 170 bp, as described earlier. For each sample, 50 ng of DNA was utilized for library preparation. The samples underwent two enzymatic treatments, followed by the KAPA Hyper Library construction protocol (KAPA Biosystems, Wilmington, MA). The Gm-seq process was executed according to previously established protocols in collaboration with Gene-Plus company. We employed the Hieff NGS® Ultima Pro DNA Library Prep Kit for Illumina (Yeason, cat. no. 12201ES96) for end repair, A-tailing, and adaptor ligation. After magnetic bead purification, 5-methylcytosine and 5-hydroxymethylcytosine were oxidized to 5-formylcytosine (5fC) or 5-carboxylcytosine (5caC) using TET enzymes, followed by reduction of 5fC or 5caC to dihydrouracil via treatment with pyridine borane. Finally, PCR amplification was performed, and barcodes were added to generate the sequencing library. After this series of steps, methylated cytosines were identified as thymine using the MGI T7 sequencing platform. Somatic mutation calling Single-nucleotide variants (SNVs) and insertions and deletions (InDels) were identified using the MuTect2 [14] software (v3.4.46, refer to https://software.broadinstitute.org/gatk/documentation/tooldocs/3.80/org_broadinstitute_gatk_tools_walkers_cancer_m2_MuTect2.php) and realDcaller2 (in-house pipeline, version 1.1.3, Geneplus-Beijing), both of which were integrated into the Oncotop platform [9]. The following criteria were applied to filter out candidate mutations step by step: (1) Allele frequency below 1% was excluded; (2) Variants were flagged as potential cross-contamination if found in more than 0.1% of samples in single nucleotide polymorphism (SNP) databases, including dbSNP (https://www.ncbi.nlm.nih.gov/projects/SNP/) [15], 1000 Genomes Project (https://www.1000genomes.org/) [16], ESP6500 (https://evs.gs.washington.edu/) [17], and ExAC (http://exac.broadinstitute.org/) [18]; (3) Variants satisfying criterion 2 with a SIFT score greater than 0.05 or a PolyPhen2 score lower than 0.85 were filtered out unless they were predicted to be harmful mutations capable of causing disease. The final list of candidate variants underwent manual review using the Integrative Genomics Viewer [19], and only the mutations that passed this verification were considered validated somatic variants. Somatic copy number variations (CNVs) calling Regions with CNVs in tumor samples were identified using CNVkit (version 0.9.6) [20], and the final CNV profiles were generated after annotation and filtering of the raw calls. Chromosome arm-level copy number variations were further determined using an in-house developed CNVkitLOH algorithm. Mutational signatures We utilized the sigminer R package [21] to identify mutational signatures based on the COSMIC Signature V2 database (https://cancer.sanger.ac.uk/signatures/). Initially, somatic mutations were loaded using the read_maf function, and components for each sample were tallied, resulting in a sample-by-component matrix through the sig_tally function. Subsequently, the sig_auto_extract function was employed to automatically determine the optimal number of signatures and extract them using Bayesian non-negative matrix factorization. After the signatures were extracted, we applied the get_sig_similarity function to compare the identified signatures with those in COSMIC (version 2) to determine their etiologies. Additionally, the sig_fit function was used to calculate both absolute and relative exposure levels of each identified signature across samples. Phylogenetic analysis We employed PyClone [22] to analyze intraclonal structures within tumors, focusing on the evolutionary transition from PTs to oligo-BMs, considering segmental copy-number alterations and contamination by normal cells. The Citup tool [23] was utilized to infer clonal phylogeny and clone prevalences based on the cellular prevalences generated by PyClone. The “cellular prevalence” output from PyClone aligns with the concept of cancer cellular fraction (CCF). The visualization of clonal evolution was used by Timescape [24]. DNA methylation data preprocessing and methylation sites calling GMseq employs TAPS technology to convert mC to T, a process that affects only about 5% of the C bases in the entire genome, thereby minimizing the impact on commonly used alignment tools. The filtered clean reads were aligned to the reference genome using Sentieon software, after which duplicates were sorted and marked to generate a BAM file of the alignment results. Following alignment, methylation sites were detected using asTair [25]. The distribution of methylated C bases across the genome exists in three forms: CG, CHG, and CHH (where H represents A, T, or C). These methylation types vary in their establishment (e.g., symmetric versus asymmetric sites), maintenance mechanisms (e.g., different methyltransferases), and underlying genetic principles, each playing distinct roles in genomic regulation. Typically, DNA methylation data in these three sequence contexts are analyzed separately in the literature. asTair is a specialized tool designed for analyzing data from TAPS and other “mC to T” conversion technologies that do not rely on bisulfite treatment. After identifying all methylation sites, the average methylation level, sequencing depth, conversion rate, and coverage of CG, CHG, and CHH contexts were calculated for the samples. Methylation clustering and CpGs basis analysis To identify subtypes based on the cancer methylome, we performed clustering and visualization on the 1,000 most variable CpG sites using the ComplexHeatmap R package [26] with default parameters. Fisher’s exact test was employed to calculate the differential distribution of CpG types between the two groups. Differential expression and pathway enrichment analysis Differential gene expression analysis was conducted using the DESeq2 R package [27]. Genes with a p-value less than 0.05 and an absolute log2(fold change) greater than 1 were considered statistically significant. Functional enrichment analysis and visualization of the differentially expressed genes (DEGs) were carried out using the fgsea R package. Pathways with an adjusted p-value below 0.05 and an absolute normalized enrichment score greater than 1 were deemed significant. Immune infiltration classification We applied the single-sample gene set enrichment analysis (ssGSEA) method from the gene set variation analysis (GSVA) R package [28] to quantify the relative infiltration levels of 28 different immune cell types within the tumor microenvironment. Marker genes for each immune cell type were sourced from a previous study [29]. The ssGSEA analysis was conducted with the following parameter settings: method=’ssgsea’, kcdf=’Gaussian’, and abs.ranking=TRUE. The enrichment score generated by ssGSEA represented the relative infiltration of each immune cell type. These scores were then normalized to a unity distribution, where zero represents the minimal and one represents the maximal score for each immune cell type. We utilized the ConsensusClusterPlus R package [30] with the k-means clustering algorithm and Euclidean distance measurement to identify the optimal clustering solution and assess the stability and consistency of the immune infiltration clustering results. The final clustering outcome was visualized using the ComplexHeatmap R package [31]. IHC staining and quantification analysis The tumor sections were stained with NLGN1 antibodies (1:600, S97A-31, Novus, USA), and the immunoreactivity was quantified using an immunoreactivity score (IRS) system. All stained slides were independently evaluated by two experienced pathologists under predefined scoring criteria. Representative tumor areas were selected for assessment, and staining intensity as well as the proportion of positive tumor cells were recorded. The IRS was calculated by multiplying the intensity of the staining (graded from 0 for absent staining to 3 for strong staining) with the percentage of positive cells, which was scored as follows: 0 for < 5%, 1 for 5%-25%, 2 for 25%-50%, 3 for 51%-75%, and 4 for > 75%. The pathologists were blinded to all clinical information and patient outcomes during scoring, and discrepant scores were resolved by joint re-review to reach consensus. mIHC To better examine the tumor immune microenvironment and accurately visualize different immune cell populations, mIHC was conducted using specific antibodies. These included CD4 for helper T cells, CD8A for effector T cells, Foxp3 for regulatory T cells (Tregs), CD56 for natural killer (NK) cells, CD20 for B cells, and cytokeratin (CK) for tumor cells. Primary antibodies were applied sequentially to tissue slides, including anti-CD4 (1:100, #48274, Cell Signaling Technology, USA), anti-CD8A (1:250, #38553, Cell Signaling Technology, USA), anti-Foxp3 (1:100, #98377, Cell Signaling Technology, USA), anti-CD56 (1:200, #99746, Cell Signaling Technology, USA), anti-CD20 (1:250, ab78237, Abcam, USA), and anti-CK (1:100, ab7753, Abcam, USA). After antibody incubation, tissue sections were treated with HRP-conjugated polymer goat anti-mouse and rabbit secondary antibodies (10013001050, Panovue, China). Nuclei were counterstained with DAPI. Tumor microenvironment immune types (TMIT) were classified based on RNA-seq results in our internal cohort, using the median expression values of PD-L1 and CD8A. Then, TMIT were classified according to PD-L1 (1:300, #13684, Cell Signaling Technology, USA) expression status and the density of tumor-infiltrating CD8+ T cells (high vs. low) based on mIHC staining. All tissue sections were digitized with the Polaris System (PerkinElmer, Waltham, Massachusetts, USA), and the generated multispectral images were subsequently analyzed using InForm software, version 2.4.8 (PerkinElmer, Waltham, Massachusetts, USA). Differential methylation analysis Differentially methylated CpGs (DMCs) were detected using the metilene software [32]. CpGs with a p-value < 0.05 and a methylation percentage difference exceeding 10% were deemed significant. Pairwise comparisons were carried out between oligo-BMs and PTs, EGFR + and EGFR- BMs, EGFR + and EGFR- PTs, as well as TP53 + and TP53- tumors. Motif enrichment analysis Differential DNA methylation can affect the binding of transcription factors (TFs). To investigate the TF motifs that may be overrepresented due to methylation differences between PTs and oligo-BMs, we used the “findMotifsGenome.pl” tool from HOMER [33] to analyze DMCs within ± 50 base pairs of gene promoter regions. DMC regions with either hypo- or hyper-methylation were examined separately to assess the impact of different methylation patterns. Transcription factors were considered significantly enriched if they had a q value ≤ 0.05. Transcription factors activity profiles We utilized the Reconstruction of Transcriptional Networks (RTN) package [34] to assess the activity of transcription factors based on DMC motifs. Initially, transcriptional regulatory networks (regulons) for the specified TFs were reconstructed. Next, mutual information analysis and Spearman rank-order correlation were used to identify potential associations between regulators and their target genes from the transcriptome expression profile. Associations with a p-value > 0.05 were excluded through permutation analysis, and unstable associations were removed using a bootstrapping strategy involving 100 resamplings, retaining associations with a consensus bootstrap value greater than 95%. Subsequently, the weakest interactions within any triplet comprising two TFs and a common target gene were eliminated, leaving the most significant TF-target pairs with a Regulon size greater than 5. Finally, the activity of each regulon was estimated using two-sided GSEA with the “tri.get” method. External database analysis Gene Expression Profiling Interactive Analysis (GEPIA) [35] was used to explore the correlation between gene expression and pathological stage in TCGA-LUAD. The Kaplan Meier plotter [36] is a comprehensive database containing EGA (https://ega-archive.org/), GEO (https://www.ncbi.nlm.nih.gov/geo/) and TCGA (https://www.cancer.gov/ccg/research/genome-sequencing/tcga), which was utilized to assess the prognostic value of genes enriched in cell adhesion molecules among patients with LUAD. Gene expressions of nonbrain metastases of LUAD were from Pleasance and colleagues [37]. Western blot Parental LLC cells, vector control LLC-BrM3 cells (LLC-BrM3 shCTRL), and NLGN1-knockdown LLC-BrM3 cells (LLC-BrM3 shNLGN1) were harvested and washed with ice-cold PBS. Total protein was extracted using RIPA lysis buffer supplemented with protease inhibitors. Cell lysates were centrifuged at 4 °C, and the supernatants were collected for protein quantification using a BCA protein assay kit. Equal amounts of protein were separated by SDS-PAGE and transferred onto PVDF membranes. After blocking, the membranes were incubated overnight at 4 °C with the indicated primary antibodies, followed by incubation with HRP-conjugated secondary antibodies at room temperature. Protein bands were visualized using a chemiluminescence detection system. Band intensities were quantified using ImageJ software and normalized to β-actin. In vivo brain metastasis assay Six-week-old C57BL/6 mice were obtained from Jiangsu Jicui Yaokang Biotechnology and maintained in the Animal Experimental Center of the Second Affiliated Hospital of Chongqing Medical University under specific pathogen-free conditions (22 ± 1 °C, 12 h light/dark cycle), with free access to food and water. All animal procedures were approved by the Animal Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University and were conducted in accordance with the institutional guidelines for animal care and use. The knockdown efficiency of NLGN1 in LLC-BrM3 shNLGN1 cells was confirmed by western blot before in vivo experiments. Ten mice were randomly allocated into two groups. For the experimental brain metastasis model, luciferase-tagged LLC-BrM3 shNLGN1 cells or LLC-BrM3 shCTRL cells (1 × 106 cells per mouse) were injected into the internal carotid artery. After 3 weeks, tumor growth in the brain was monitored using an in vivo imaging system. Then, the mice were euthanized by CO2 inhalation, and brain tissues were collected for pathological examination. Outcomes and statistical analysis Overall survival (OS) was defined as the period from the beginning of treatment to death, or, for surviving patients, the date of their last follow-up. Progression-free survival (PFS) refers to the time from the initiation of treatment until disease progression or death, whichever occurs first. Intracranial progression-free survival (iPFS) denotes the time from the start of treatment to either intracranial disease progression or death from any cause, whichever comes first. Kaplan-Meier survival curves were generated using the ‘survival’ R package (https://github.com/therneau/survival). The log-rank test was employed to compare the survival distributions between different groups of patients. For comparing continuous variables between two groups, the Wilcoxon rank-sum test was used. Fisher’s exact test was applied to compare the categorical variables between two groups. A p-value of less than 0.05 was considered statistically significant. Results Patient characteristics The clinicopathological characteristics of all participants were summarized in Additional file 1: Table S1. We included 20 patients with LUAD and matched oligo-BMs who underwent surgery intervention. The histological subtype of all patients was adenocarcinoma, with a median age of 59 years old (range 31–72). The cohort consisted of more men (55.0%) than women (45.0%), and 65% of the patients were smokers. Seven patients were diagnosed with brain metastases concurrently with their initial LUAD diagnosis and were categorized as synchronous brain metastasis. These patients underwent simultaneous resection of both the primary tumor and brain metastases. Correspondingly, the remaining thirteen patients who developed brain metastases during or after treatment of the primary tumor were categorized as metachronous brain metastasis. Patients with synchronous metastases all diagnosed with stage IV disease, whereas those with metachronous metastases initially presented with stage I to III disease. The median follow-up duration for our cohort was 719 days (range 299–1859), with 90% of patients experiencing mortality within this period. Comparison of mutational features between paired primary LUAD and oligo-BMs To investigate the discrepancy and concordance of the mutational landscape between paired PTs and oligo-BMs, we first compared the genomic features in these samples. C > T transitions were the most common base substitutions in both cohorts, whereas the transition/transversion ratio in PTs was higher than that in the oligo-BMs group (Additional file 2: Fig. S1A and S1B). The mutational landscape of 20 paired brain oligometastatic LUAD is shown in Fig. 2A. TP53 and EGFR emerged as the most frequently mutated driver genes, displaying a high degree of concordance between PTs and oligo-BMs (Fig. 2A), which underscores their pivotal role in tumorigenesis. Further analysis of the mutational gene frequency in both cohorts revealed no alterations specifically associated with oligo-BMs (Data not shown). To assess intertumor heterogeneity, genetic alterations were categorized as either shared (found in both primary and metastatic lesions) or private (unique to primary or metastatic lesions). Fig. 2B and C demonstrated that an average of 34.9% (range, 0%–76.0%) of SNVs and an average of 16.3% (range, 0%–50.0%) of CNVs were identified as shared. Notably, the ratio of shared SNVs did not significantly correlate with that of CNVs (R = 0.062, p = 0.81, Fig. 2D), indicating that SNVs and CNVs occur independently as genomic events within these tumors. Mutation landscape of paired PTs and oligo-BMs of LUAD. A Integrated genomic and clinical data from 20 primary tumors and their corresponding brain metastatic lesions. Samples are arranged in descending order based on the proportion of samples harboring high-frequency mutated genes and the mutation frequency of these genes. B and C Proportions of SNV and CNV that are either shared (present in both primary and metastatic lesions) or private (present exclusively in either primary or metastatic lesions) for each patient. D Correlation between SNV concordance and CNV concordance, with the concordance ratio defined as the number of shared events between primary tumors and paired metastases divided by the total number of events in the same primary tumor and metastasis pairs. E Enrichment of TCGA-defined oncogenic signaling pathways in primary tumors and brain metastases. F Comparison of MSI status between PTs and oligo-BMs. G Comparison of TMB between PTs and oligo-BMs. H Correlation analysis of TMB in PTs and oligo-BMs. I, J, and K Comparison of CNV all, CNV gain and CNV loss burden between the PTs and oligo-BMs groups. SNV, single nucleotide variation; CNV, copy number variation; PTs, primary tumors; oligo-BMs, brain oligometastases; MSI, microsatellite instability; TMB, tumor mutational burden Next, we conducted enrichment analysis of oncogenic signaling pathways in TCGA to elucidate the biological significance of mutations occurring in PTs and oligo-BMs [38]. Notably, IFNγ signaling pathways associated with the regulation of PD-L1 expression was enriched exclusively in BM compared to PT among the altered pathways (Fig. 2E). Additionally, we compared the microsatellite instability (MSI) status, tumor mutation burden (TMB), and CNV burden between PTs and oligo-BMs. No significant differences were observed in MSI status or TMB between the two groups (p = 1, Fig. 2F; p = 0.66, Fig. 2G). However, TMB showed a strong positive correlation between PTs and oligo-BMs (p = 4.9e-05, R = 0.78, Fig. 2H), which also reflects the genetic consistency between the metastatic and primary lesions. Moreover, oligo-BMs exhibited a slightly higher burden of CNV gains compared to PTs (p = 0.046, Fig. 2I), while the overall CNV burden and CNV loss burden were similar between the two groups (p = 0.098, Fig. 2J; p = 0.42, Fig. 2K). Subclonal architecture, phylogenetic relationships and mutational signatures in matched primary LUAD and oligo-BMs Monoclonal and polyclonal seeding are considered the two principal hypothesized modes of cancer metastasis. In monoclonal seeding, a single clone that prevails within the primary tumor seeds a distant metastasis, giving rise to a lesion that is founded by one ancestral clone and is relatively less diverse, although subsequent evolution, potentially driven by ongoing chromosomal instability, may still generate additional diversification. Conversely, polyclonal seeding involves multiple distinct clones from the primary tumor establishing metastases either through cooperation or independently, ultimately leading to a metastatic tumor with greater clonal diversity (Fig. 3A) [39–41]. To delineate the seeding patterns during brain oligometastasis, we inferred clonal lineages and visualized the data using TimeScape. We listed the representative case of each pattern in Fig. 3B. The clonal lineages of all cases were listed in Additional file 2: Fig. S2. The results revealed that only three patients (P2, P4, and P5) were monoclonal seeding, while the majority of patients (85%, 17/20) were found to have polyclonal seeding, containing multiple identical subclones in both PTs and paired oligo-BMs (Fig. 3C). Moreover, we found that all subclones in the primary lesions originated from a common ancestor prior to primary tumors and subsequently co-migrated to the brain, where they underwent either clonal expansion or contraction. These findings suggested that genome evolution initiated from a single cell or cellular cluster in the lung tissue and gave rise to diverse clonal subpopulations in PTs. TimeScape revealed the predominance of polyclonal seeding mode, wherein multiple distinct subclones from the primary site metastasized to the brain, ultimately establishing oligo-BMs. Subclonal architecture and phylogenetic relationship of paired PTs and oligo-BMs. A Schematic of monoclonal and polyclonal metastasis. B Clonal lineages of the major tumor subpopulations plotted with TimeScape with inferred common ancestors indicated in pink, where representative patients are shown: monoclonal (P4); polyclonal (P6). C Pie chart summarizing the proportions of monoclonal and polyclonal metastasis in our cohort. D Timing of metastasis divided into parallel and linear progression model based on the genetic divergence between paired PTs and oligo-BMs. E Distinct progression model based on the CCF of SNVs between paired PTs and oligo-BMs for representative patients with parallel progression model (P3) and linear progression model (P6). For each patient, the left scatter plot shows CCF values for SNVs. Each dot represents one mutation. Clonal mutations in PT-private are labeled in red, BM-private are labeled in green, and shared are labeled in blue. Phylogenetic trees on the right, depicting the proportion of ubiquitous mutations (trunk) and unique mutations (branch) to illustrate the evolutionary relationship between PTs and oligo-BMs within individual cases. F Pie chart summarizing the proportions of parallel and linear progression model in our cohort. G Violin plot showing the number of BM-private mutations with parallel or linear progression model from our group. H Statistical analysis of the proportions of trunk and branch mutations, with trunk and branch lengths representing the number of SNVs. I Percentage of clonal and subclonal mutations for top genes in PTs and oligo-BMs, respevtively. CCF, cancer cellular fraction; PTs, primary tumors; oligo-BMs, brain oligometastases Metastatic disease has traditionally been considered as a late-stage event in cancer progression, signifying that tumor cells have entered a phase of uncontrollable sustained proliferation and widespread dissemination to multiple organs [42]. However, emerging evidence suggests that the dissemination of metastasis can also occur at early stages, depending on the specific type and biology of the cancer [43, 44]. The timing of brain oligometastatic dissemination during the course of LUAD is still unclear. To explore the temporal dynamics of metastatic dissemination (parallel versus linear progression model [45]), we developed an approach centered on the genomic heterogeneity between paired PTs and oligo-BMs (Fig. 3D). These progression models are distinguished by two key parameters: the relative timing of metastasis emergence in the primary tumor and the genetic divergence between primaries and corresponding metastases [45]. The parallel progression model was characterized by a range of 0 < R* ≤ 0.5, while the linear progression model was defined as 0.5 < R* <1.0. For instance, R* was 0.18 in P3, indicating early onset of brain oligometastasis, whereas R* was 0.64 in P6, indicating late brain oligometastasis (Fig. 3E). The CCF plots and phylogenetic trees of all patients were listed in Additional file 2: Fig. S3. Our results indicated that most oligo-BMs (75%, 15/20) followed the parallel progression model and tended to emerge at a relatively early stage in the molecular evolution process (Fig. 3F). Notably, parallel-progression oligo-BMs harbored significantly more BM-private mutations, consistent with an early dissemination model where metastatic seeding cells acquire additional mutations during prolonged colonization in the brain niche (p = 0.0017, Fig. 3G). Furthermore, the proportion of mutations in metastatic branches was higher than in primary branches, approaching borderline significance (p = 0.072, Fig. 3H). This observation suggested that brain oligometastatic lesions might accumulate more mutations under evolutionary selection pressure during their development. Next, we investigated the association between clonal patterns (monoclonal vs. polyclonal) and seeding mode (parallel vs. linear) with clinical parameters, including BM tumor size, driver gene status, systemic treatment, metastatic timing (synchronous vs. metachronous), and survival outcomes. However, no significant differences were observed in any of the analyzed parameters (all p > 0.05, Additional file 2: Fig. S4A and S4B). Additionally, we examined the clonal distribution of high-frequency mutations. Among the Top10 genes, EGFR, TP53 and RBM10 exhibited a mixture of clonal and subclonal mutations in both primary and metastatic lesions, while other frequently mutated genes, such as NF1 and LRP1B, mainly presented as subclonal mutations with no clonal mutations observed (Fig. 3I). Intratumoral heterogeneity (ITH), which refers to the diversity of individual cells within a neoplasm, is associated with tumor evolution, progression, and metastasis. ITH can be quantified using two methods: clone number estimated by PyClone and the ITH score. Our findings revealed a striking concordance between PTs and oligo-BMs for both clone number (p = 0.13, Additional file 2: Fig. S5A) and ITH score (p = 0.35, Additional file 2: Fig. S5B), indicating that brain oligometastases exhibited intratumoral heterogeneity comparable to primary tumors. Therefore, we did not observe a significant difference in ITH metrics between PTs and oligo-BMs. Each organ possesses a unique microenvironment, and tumor cells may encounter different environmental mutagens in various organs, potentially resulting in distinct mutational processes that eventually affect tumor invasion, proliferation, and metastasis [46]. However, there is currently a lack of comprehensive research on which mutational signatures play a crucial role in the progression of LUAD to oligo-BM. To address this, we reconstructed the mutation profiles of all paired samples using a predetermined set of 30 COSMIC signatures. Our analysis revealed that in PTs, approximately 45.6% of the mutational signatures were associated with signature 4 (smoking), 14.4% with signature 11 (alkylating agent exposure), 23.6% with signature 23 (unknown etiology), 10.0% with signature 7 (ultraviolet light exposure), and 6.4% with signature 9 (DNA polymerase η). In contrast, in oligo-BMs, 54.0% of the mutational signatures were attributed to signature 4, 33.1% to signature 23 (unknown etiology), and 12.9% to signature 11 (alkylating agent exposure) (Additional file 2: Fig. S6A and Additional file 3: Table S2). Additionally, we observed that the smoking-related signature 4 was the most prevalent mutational signature in both PTs and oligo-BMs, likely due to the high proportion (13/20, 65%) of smokers in the study population. To further delineate the dynamic changes in mutational signatures during the formation of oligo-BM, we analyzed the mutational processes related to initial clonal expansion prior to the occurrence of metastasis, as well as subsequent subclonal diversification. Specifically, mutational signature analyses were performed separately for shared mutations, which represent early genomic events, and for private mutations in PTs or oligo-BMs, which represent later subclonal mutations. Through this analysis, signature 4 was also identified as the predominant mutational signature in shared mutations, aligning with previous studies that alterations related to exposure to tobacco carcinogens often occur early in the clonal evolution [47]. In comparison to shared mutations, primary-only mutations were enriched for mutagenic processes, including signature 7, signature 9, and signature 12 (unknown), whereas metastasis-only mutations were enriched for signatures 3 and 7 (Additional file 2: Fig. S6B and Additional file 4: Table S3), indicating that distinct mutational processes may be active at different molecular times during the evolutionary trajectory of oligo-BM. Particularly, signature 3 (Homologous recombination deficiency, HRD) was exclusively observed in the BM-only group, as opposed to the PT-only and shared groups, suggesting that HRD may represent a late event evolving independently from PT. In summary, the smoking-related processes may drive mutagenesis during early clonal expansion of tumor evolution, other mechanisms such as HRD, DNA polymerase η, and an ultraviolet light-like mutational pattern, likely play significant roles in subclonal diversification with PTs or oligo-BMs. Comparison of TIME between primary LUAD and oligo-BMs The intracranial immune microenvironment plays a crucial role in modulating tumor progression and brain metastasis through interactions with tumor cells. Shaped by the unique intracranial immune microenvironment, the immune landscape of brain metastatic tumor cells gradually diverges from that of the primary tumor, leading to the development of distinct immune heterogeneity. However, the differences in the immune microenvironment between PTs and oligo-BMs still remain largely unknown. To characterize the immunological features of the tumor microenvironment in these paired samples, we first conducted Estimate algorithm to assess overall immune cell infiltration. The result showed that the immune score of oligo-BMs was significantly lower than that of PTs (Fig. 4B), indicating a general decrease in immune cell infiltration in metastases. To further identify the specific immune cell subsets contributing to this reduction, we utilized ssGSEA to evaluate the relative abundance of immune infiltration cells in the tumor microenvironment (Fig. 4A). The oligo-BMs displayed lower infiltration scores of activated B cell (p = 0.02), natural killer (NK) cell (p = 0.043), immature B cell (p = 0.046), and type 1 T helper cell (p = 0.014) compared to PTs (Fig. 4C). We subsequently investigated whether the differences in immune infiltration observed between PTs and oligo-BMs were also reflected in the tumor immune cell density. For this purpose, we analyzed the densities of CD4+, Foxp3+, CD8+, CD56+, and CD20+ cells in paired tumor tissues utilizing mIHC. Representative images for each marker are presented in Fig. 4D. Indeed, we observed no significant changes in the density of CD4+, Foxp3+, CD8+ T cells, or CD20+ B cells between oligo-BMs and PTs. However, the infiltration density of CD56+ NK cells was significantly lower in oligo-BMs (Fig. 4E). Tumor infiltrating immune cell subtypes between paired PTs and oligo-BMs. A Relative fractions of tumor infiltrating immune cells estimated using the ssGSEA algorithm, based on RNA sequencing data from PTs and oligo-BMs. B Overall immune cell infiltration levels assessed by ESTIMATE algorithm. C Comparison of immune cell infiltration between PTs and paired oligo-BMs based on RNA-seq data. D Representative mIHC images of immune related markers including CD4+ (yellow), Foxp3+ (gray), CD8A+ (red), CD56+ (green), CD20+ (purple), and CK (tumor cell, cyan). E Mean cell density comparison of CD4+, Foxp3+, CD8A+, CD56+, CD20 + cells between PTs and oligo-BMs The TIME can be categorized into four TMIT types according to the median expression of CD8A and PD-L1, providing valuable insights for optimizing precision immunotherapy, as previously reported [48]. Tumors with TMIT I (CD8A-high/PD-L1-high) can derive better benefit from anti-PD-L1/PD-1 therapies [49]. In our study, we found that oligo-BMs contained a higher proportion of TMIT I than PTs (40.0% vs. 20.0%, Fig. 5A, Additional file 5: Table S4). To further validate the correlation between TMIT classification and the efficacy of immunotherapy in oligo-BMs of LUAD patients, we collected 71 brain oligometastatic tumor samples from LUAD patients without EGFR/ALK driver mutations. These patients were treated with first-line immunotherapy and classified into different TMIT types according to CD8 and PD-L1 expression from mIHC results: TMIT I (30 samples, 42.3%), TMIT II (14 samples, 19.7%), TMIT III (19 samples, 26.8%), and TMIT IV (8 samples, 11.2%) (Fig. 5B). We found that TMIT subtypes were significantly associated with OS and iPFS in LUAD patients with oligo-BMs (Fig. 5C, OS, p = 0.046; iPFS, p = 0.027). The TMIT I subgroup demonstrated the most favorable survival outcomes, while the TMIT III subgroup exhibited the worst survival benefit. Comparison of TMIT subtypes and immune-related signatures between paired PTs and oligo-BMs. A Comparison of the proportion of TMIT between PTs and oligo-BMs. B Representative mIHC images of four TMIT in brain oligometastatic lesions of LUAD patients receiving first-line immunotherapy. C Kaplan-Meier survival analysis of OS and iPFS for brain oligometastatic LUAD patients stratified by TMIT. TMIT I, n = 30; TMIT II, n = 14; TMIT III, n = 19; TMIT IV, n = 8; D Comparison of PD-L1 mRNA expression between PTs and paired oligo-BMs. E Correlation of PD-L1 mRNA expression between PTs and paired oligo-BMs. F, G and H The correlation between PD-L1 expression and CD4/CD8A/IFNG expression in oligo-BMs, respectively. I Comparison of immune-related gene signatures between paired PTs and oligo-BMs. J Volcano plot showing differentially expressed genes between PTs and oligo-BMs, with multiple collagen family genes downregulated in oligo-BMs. K Comparison of collagen enrichment scores between PTs and oligo-BMs. TMIT: tumor microenvironment immune types; TIME, tumor immune microenvironment; ssGSEA, single-sample gene set enrichment analysis; PTs, primary tumors; oligo-BMs, brain oligometastases Given the pivotal role of immune checkpoint molecules in modulating the immune response and their potential as therapeutic targets in cancer treatment, we next examined the differential expression of immune checkpoint molecules between PTs and oligo-BMs and found a significant upregulation of PD-L1 in oligo-BMs (Fig. 5D and Additional file 2: Fig. S7). Additionally, the expression levels of PD-L1 in BMs were poorly correlated with those in PTs (R = 0.31, p = 0.18, Fig. 5E). Currently, immunotherapy with PD-L1 inhibitors has revolutionized the treatment landscape for lung cancer. Two primary mechanisms have been proposed to drive PD-L1 expression in tumor cells, including the secretion of interferon-γ (IFN-γ) by tumor infiltrating lymphocytes (TILs) and the activation of aberrant oncogenic signaling pathways [49–52]. Given the differential expression patterns of PD-L1 between paired samples, we inquired whether the regulatory mechanisms governing PD-L1 expression might also differ. Thus, we further investigated the correlation between TILs markers (CD4 and CD8A) and PD-L1 expression. In oligo-BMs, both CD4 (R = 0.63, p = 0.0029) and CD8A (R = 0.66, p = 0.0032) expression levels showed a positive correlation with PD-L1 expression (Fig. 5F and G). Notably, IFN-γ expression was also strongly and positively correlated with PD-L1 expression in oligo-BMs (Fig. 5H). However, these correlations were not observed in primary lesions (Additional file 2: Fig. S8). Therefore, PD-L1 induction in oligo-BMs may be influenced by the secretion of IFN-γ produced by TILs, while it is primarily driven by oncogenic pathway activation or other mechanisms in PTs. Increasing evidence suggests that various forms of cell death could reshape the TIME and affect antitumor immune responses [53, 54]. We implemented GSVA to assess the relative activity of common cell death pathways in PTs and oligo-BMs. A notable increase in scores for parthanatos and ferroptosis was observed in oligo-BMs compared to PTs, whereas the score for immunogenic cell death (ICD) was significantly decreased (Additional file 2: Fig. S9). Activated ICD could stimunate the immune system through the release of damage-associated molecular patterns from dying tumor cells, leading to the recruitment of antigen presenting cells and enhancing tumor immunotherapy. Overall, the propensity for pathways of cell death varied between primary and metastatic lesions, with the lower ICD scores observed in oligo-BMs compared to paired PTs, indicating a less active or “cold” immune microenvironment. In addition, some recently reported immune-related gene signatures have been shown to be associated with immunotherapy response and tumor immunogenicity [29, 55, 56]. Our findings revealed that the expression scores of chemokines and MHC-I signatures were significantly higher in oligo-BMs compared with PTs (p < 0.05, Fig. 5I). Next, we conducted differential gene expression analysis and identified 1,090 DEGs, including 835 genes upregulated in oligo-BMs and 255 in PTs (Fig. 5J). Notably, we observed many genes in the collagen family exhibited significantly higher expression levels in PTs (Fig. 5J). Interestingly, the enrichment score of collagen and several pathways related to collagen synthesis were also significantly upregulated in PTs (p = 0.001, Fig. 5K, Additional file 2: Fig. S10A). GSEA analysis also demonstrated that multiple immune associated pathways were significantly enriched in PTs than their oligo-BMs (Additional file 2: Fig. S11), while several pathways linked to metabolism, synaptic transmission, and neurotransmitter release were enriched in oligo-BMs (Additional file 2: Fig. S12A and S12B, and Additional file 2: Fig. S10B). In conclusion, we performed a comprehensive analysis of the TIME between primary lesions and paired brain oligometastasis. Our findings indicated that oligo-BMs exhibited lower immune and ICD scores, suppression of immune-activated pathway and decreased infiltration of activated immune cells (B cells, NK cells and Th1 cells), suggesting a more immunosuppressed microenvironment compared to PT. Moreover, we observed a discrepancy in PD-L1 expression between PTs and oligo-BMs, with distinct regulatory mechanisms, implying potential inconsistency in response to anti-PD-L1/PD-1 therapy. DNA methylation profiles between primary LUAD and oligo-BMs DNA methylation refers to an epigenetic modification where a methyl group is attached to the cytosine base in DNA, typically occurring at CpG dinucleotides. Apart from genomic alterations and transcriptional regulation, aberrant methylation also commonly occurs in cancer and contributes to tumor proliferation and metastasis [57, 58]. However, the alterations in DNA methylation during brain oligometastasis have not been extensively studied. We first analyzed the methylation patterns at regulatory genomic regions, which are typically correlated with transcript expression levels. The ratio of 5-methylcytosine to cytosine remained consistently lower across all functional genomic regions in oligo-BMs than PTs (Fig. 6B). In addition, no significant differences were observed in the methylation level of mC, mCpG, mCHG and mCHH between primary lesions and metastases (Additional file 2: Fig. S13). Considering that transcriptionally relevant DNA methylation changes in cancer frequently occur at CpG sites [59], we further compared the mCpG levels across different functional genomic regions and found them to be highly concordant between PTs and oligo-BMs (Additional file 2: Fig. S14). To gain a comprehensive view of methylome, we subsequently performed unsupervised hierarchical clustering of primary and metastatic samples separately or together using the 1,000 most variably methylated CpG sites (Fig. 6A). Notably, oligo-BMs clustered together with their corresponding PTs rather than with metastases from other patients, indicating substantial inter-individual heterogeneity and relative homogeneity among lesions within the same patient in methylation patterns. In addition, TP53 mutant samples tended be enriched together in the clustering of all paired tumors. This separation based on genetic background was also evident when metastases were clustered, but not in primary lesions. Methylome analysis of paired PTs and oligo-BMs. A Unsupervised hierarchical consensus clustering of all primary and metastatic samples (left), primary (middle), and metastatic samples (right), respectively. B Ratio of 5-methylcytosine to cytosine across functional genomic regions in PTs and oligo-BMs. C Proportion of differentially methylated CpG sites, with hyper-methylated sites indicated in red and hypo-methylated sites in blue. D Comparison of mean methylation levels of CpG sites between EGFR wild-type and EGFR mutant tumors. E Enrichment analysis for differentially methylated CpG sites between various groups: primary tumors versus brain metastases, EGFR-mutant versus EGFR wild-type metastases, EGFR-mutant versus EGFR wild-type primary tumors, and TP53-mutant versus TP53 wild-type tumors. Values represent the log-OR from Fisher exact tests. F Enrichment of known TFBM due to differential methylation using HOMER. The Y-axis represents the logarithm of the adjusted p-value, with positive values indicating TFBM in hypermethylated regions and negative values indicating TFBM in hypomethylated regions. The X-axis shows the rank of motifs based on their significance. G Heatmap displaying regulon activity scores for transcription factors estimated by RTN. Regulons with significant differential activity between PTs and oligo-BMs were marked in red with p-values listed on the left. TFBM, transcription factor binding motifs; PTs, primary tumors; BM, brain oligometastasis; RTN, reconstruct transcriptional networks To further investigate the methylation differences between PTs and oligo-BMs, we used metilene software to identify DMCs in PTs against oligo-BMs. DMCs were significantly enriched in introns and intergenic regions (Additional file 2: Fig. S15A), with a predominance of hypermethylation over hypomethylation in various gene regions (Fig. 6C and Additional file 2: Fig. S15B), suggesting an uneven distribution of methylation levels across the genome. Given that EGFR mutations play a pivotal role in the pathogenesis of brain metastasis and are closely linked to alterations in gene methylation patterns [60], we investigated whether this specific mutational background was closely associated with distinct methylation profiles in oligo-BMs. Compared to EGFR wildtype tumors, the methylation level of CpG was significantly lower in EGFR-mutant tumors (p = 0.039, Fig. 6D). Furthermore, we observed notable enrichment of hypermethylation (OR = 1.2, p < 0.05) and depletion of hypomethylation (OR = 0.4, p < 0.05) in CpG islands of EGFR-mutant metastases (i.e., metastases with EGFR mutations vs. metastases without EGFR mutations) (Fig. 6E). This pattern of differentially methylated CpG islands was also evident when comparing EGFR-mutant PTs with wildtype PTs, where similar trends of hypermethylation enrichment (OR = 1.2, p < 0.05) and hypomethylation depletion (OR = 0.3, p < 0.05) were observed. However, a contrasting trend emerged when comparing all metastases with primary tumors. As shown in Fig. 6E, metastases exhibited a depletion of hypermethylation (OR = 0.5, p < 0.05) and an enrichment of hypomethylation (OR = 1.5, p < 0.05) in CpG islands. Collectively, these findings showed that oligo-BMs had distinct distribution of CpG islands compared to PTs. However, EGFR mutations appeared to reverse this distribution pattern, highlighting their unique role in transcriptional regulation. Methylation changes in regulatory regions of genome can influence gene expression by interfering with TFs binding to specific sequences, known as motifs [61]. To identify overrepresented TF motifs resulting from differential methylation between PTs and oligo-BMs, the HOMER tool was utilized on DMCs (± 50bps) overlapping with gene promoter regions. We identified 60 potential TFs in hypermethylated regions and 37 in hypomethylated regions, respectively (Fig. 6F, Additional file 6: Table S5 and Additional file 7: Table S6). Subsequently, we reconstructed the transcriptional regulatory networks (regulons) for TFs predicted by DMC motifs and compared their activity using the RTN package. This approach enabled us to model the impact of upstream TFs on the expression changes of downstream target genes. As shown in Fig. 6G, several regulons exhibited significant differential activity between PTs and oligo-BMs. Among them, FOXF1 and GATA6 had significantly lower activity in oligo-BMs and were also enriched in hypermethylated regions according to HOMER, which have been reported to be associated with progression and metastasis of LUAD [62–64]. Hypermethylation typically inhibits the binding of TF, resulting in a diminished capacity of TFs to regulate their target genes. Consequently, this leads to a reduction in the activity of the gene sets they govern. These results suggested that DNA methylation changes in motifs of FOXF1 and GATA6 may regulate target gene expression by altering TF binding ability, thereby facilitating the progression of brain oligometastasis. Integrative analysis of transcriptome and methylome identified NLGN1 as a prognostic biomarker associated with olig-BMs Given that DMCs at promoters or gene bodies could significantly affect gene expression profiles, we conducted an integrative analysis of transcriptome and methylome by taking the intersection of differentially methylated genes (DMGs) and DEGs. Genes with at least one DMC in their promoter or gene body regions were considered DMGs. These genes were then categorized into four groups based on their expression and methylation patterns: hypermethylated-upregulated (hyper-up), hypomethylated-upregulated (hypo-up), hypermethylated-downregulated (hyper-down), and hypomethylated-downregulated (hypo-down) genes (Fig. 7A). Among the 696 upregulated genes, we identified 333 hypomethylated CpG sites in the promoter regions and 28,488 hypermethylated CpG sites in the gene body regions. Conversely, of the 171 downregulated genes, there were 146 hypermethylated CpG sites in the promoter regions and 2,577 hypomethylated CpG sites in the gene body regions (Additional file 2: Fig. S16). To explore the biological pathways influenced by these methylation driver genes and identify which might potentially facilitate or drive the development of oligo-BMs, we performed KEGG pathway enrichment analysis separately for upregulated and downregulated genes (Fig. 7B and C). The neuroactive ligand-receptor interaction, calcium signaling pathway, cell adhesion molecules, and long-term potentiation were significantly enriched in oligo-BMs, while pathways in cancer, focal adhesion, and ECM receptor interaction were found to be significantly enriched in PTs. Particularly, we focused on genes related to cell adhesion molecules that are enriched in oligo-BMs, as they play a critical role in tumor invasion, cancer migration and metastasis [65]. To screen key target genes involved in brain metastasis, we first explored the relationship between gene sets within this pathway and pathological stages using the GEPIA database. Our analysis showed that expressions of CDH4, CNTNAP2 and NLGN1 increased with advancing pathological stage in LUAD (Fig. 7D and Additional file 2: Fig. S17A and E). Moreover, the analysis of Kaplan-Meier plotter database found that higher expression of CDH4, CNTNAP2, and NLGN1 was also significantly associated with worse overall survival (OS) and shorter time to first progression (FP) (Fig. 7C and Additional file 2: Fig. S17B, C, F and G). To further pinpoint genes specifically associated with brain metastasis, we examined the mRNA levels of CDH4, CNTNAP2 and NLGN1 in an external dataset of extracranial LUAD metastases from a previous study. Notably, the expression of NLGN1 was markedly elevated in BMs compared to extracranial metastases, while CDH4 and CNTNAP2 expression levels showed no significant difference (Fig. 7E and Additional file 2: Fig. S17D and H). To assess the organ-specific expression of NLGN1, IHC was further performed to evaluate its protein expression in oligo-BMs, liver, bone, and lymph node metastases, as well as paired PTs. The results showed that the IRS of NLGN1 was significantly higher in BMs than in extracranial metastases (Fig. 7F). Moreover, oligo-BMs also displayed a markedly elevated IRS for NLGN1 compared to their paired PTs (Fig. 7G). Additionally, survival analysis in our internal cohort also validated that higher NLGN1 expression correlated with poor PFS and OS (Fig. 7H), and the prognostic significance was retained after adjusting for age, gender, smoking, and stage in a multivariable cox regression model (Additional file 2: Fig. S18). Intriguingly, NLGN1, characterized by its brain-specific elevated expression, emerged as a gene frequently subjected to methylation, with all associated CpGs exhibiting hypermethylation in its gene body regions (Additional file 2: Fig. S19 and Additional file 8: Table S7). This methylation regulatory mechanism aligns with the trend that hypermethylation in gene body regions was often associated with increased transcription. Integrative analysis of gene expression and DNA methylation. A Scatter plot displaying the intersection of differentially methylated genes and differentially expressed genes in the promoter or genebody regions, with each point representing a CpG-gene pair. B KEGG pathway enrichment analysis was conducted for the genes that were hyper-up and hypo-up in oligo-BMs, as well as for the hyper-down and hypo-down genes in PTs. C Correlation of NLGN1 expression with overall survival (left) and first progression time (right) (Kaplan-Meier plotter database). D Relative expression of NLGN1 mRNA in stage I, stage II, stage III, stage IV of LUAD (GEPIA database). E Comparison of NLGN1 expression in BM samples from our cohort and extracranial metastases of LUAD from Pleasance and colleagues. F Representative immunohistochemical staining images and immunoreactive scores of NLGN1 in liver (n = 7), bone (n = 9), lymph node (n = 16) and brain metastases (n = 91) of LUAD. G Representative immunohistochemical staining images and immunoreactive scores of NLGN1 in 20 paired PTs and oligo-BMs. H The survival analysis of overall survival and progression free survival in our cohort based on NLGN1 expression. I Western blot was used to detect NLGN1 expression. Representative Western blot images (left) and quantitative analysis of NLGN1 protein levels (right). J The mice received an internal carotid artery injection of the vector control (LLC-BrM3 shCTRL) and the NLGN1-knockdown LLC cell line (LLC-BrM3 shNLGN1). K Quantification of bioluminescence intensity in mice 3 weeks after injection. L Brain metastasis-free survival curves for mice injected with LLC-BrM3 shCTRL and LLC-BrM3 shNLGN1, respectively Lastly, we sought to validate the functional role of NLGN1 expression in promoting BMs in vivo. We first established a NLGN1-knockdown LLC cell line (LLC-BrM3 shNLGN1) and the corresponding vector control (LLC-BrM3 shCTRL) through lentiviral transduction. After NLGN1 knockdown, the expression of NLGN1 protein was reduced (Fig. 7I). Next, we examined the in vivo tumorigenesis in mouse syngeneic models by internal carotid artery injection. After NLGN1 knockdown, the growth rate of BMs decreased significantly, as evidenced by bioluminescence imaging (Fig. 7J and K). Additionally, Kaplan-Meier survival analysis demonstrated prolonged brain metastasis-free survival in the LLC-shNLGN1 cohort (Fig. 7L). These in vivo findings substantiate NLGN1 as a critical molecular driver of cerebral metastasis. Taken together, our findings suggested that NLGN1 expression was elevated at both mRNA and protein levels in BMs, and its upregulation was significantly linked to BMs and poor survival outcomes, independent of other clinical parameters. Hypermethylation-induced upregulation of NLGN1 in gene body regions might promote brain metastasis because these changes seemed to be specific to BMs when compared to extracranial metastases. Discussion Brain metastases frequently occur in patients with advanced NSCLC, especially those with LUAD [66], which have historically been considered incurable and associated with poor survival outcomes. Growing evidence demonstrated that oligometastasis is increasingly recognized as a distinct pattern of metastasis with a more favorable natural history and prognosis compared to polymetastasis. This distinction suggests that oligometastatic cases may benefit from aggressive local therapies aimed at achieving clinically meaningful tumor control and, in some instances, even a cure. However, a significant proportion of patients in the early stage of polymetastasis receive excessive local treatments due to inadequate early detection, which can lead to rapid disease progression thereafter. Molecular aberrations, including genetic mutations, transcriptional regulation, and epigenetic modifications can profoundly impact tumor metastasis. Current research on brain oligometastatic disease is very limited, with existing studies mainly focusing on the preliminary exploration of genomic features and TCR repertoires, lacking in-depth analysis across multiple omics layers [67]. This gap has impeded an extensive understanding of the exact mechanism and metastatic process involved in brain oligometastasis, complicating the clinical assessment of true oligometastasis. Therefore, comprehensive molecular analyses of primary tumors and paired brain oligometastatic lesions are essential for gaining better insight into the evolution of cancer oligometastasis and elucidating the mechanisms driving brain oligometastasis. In our study, we first investigated the mutational landscape of PTs and matched oligo-BMs, and found a high concordance of TP53 and EGFR driver alterations. This finding aligns with previous studies that major driver-genes, such as TP53, EGFR, KRAS and ALK, exhibit over 80% consistency between PTs and paired BMs in NSCLC [68]. Similarly, pan-cancer analyses have also indicated that driver mutations were homogeneous in metastases originating from the same primary tumor [69]. To illustrate the evolutionary process of olig-BMs, two main progression models have been proposed: the parallel progression model and the linear progression model [45, 70]. The linear progression model suggests that the metastasis-competent clone spreads from the primary tumor at a later stage, exhibiting a limited degree of primary-metastasis genetic difference. In contrast, the parallel progression model posits that the metastatic clone emerges at an early stage of tumorigenesis, with both primary tumors and metastases evolving in parallel, leading to significant primary-metastasis genetic divergence [70]. In our study, phylogenetic tree analysis revealed that oligo-BMs were genetically more divergent from their PTs, indicating a predominance of the parallel progression model. Brastianos et al. analyzed whole-exome sequencing data from 35 conventional primary NSCLC and their paired brain metastases, finding that an average of 69% of SNVs and 59% of CNV events were shared [71]. This stands in stark contrast to our cohort, which showed an average of only 35% of SNVs and 16% of CNVs, respectively, suggesting a significantly higher intertumor heterogeneity between oligo-BMs and PTs. Overall, these findings imply that at the time of brain oligometastasis occurrence, tumors are still in the early stage of molecular evolution, with most somatic mutations yet to arise. Therefore, genomic sequencing of biopsy specimens from the primary tumor may not accurately reflect the alterations present in oligo-BMs. It is imperative to sequence metastatic samples to integrate genetic information from oligo-BMs and guide individualized treatment, particularly for patients lacking driver gene mutations. ITH has been identified as a mechanism of tumor evasion, progression, and metastasis [72]. Our data revealed that oligo-BMs exhibited comparable intratumoral heterogeneity to that of PTs in our study. Specifically, both PTs and oligo-BMs displayed similar levels of ITH, as assessed by two methods including clone number of PyClone and the ITH score. This finding was unexpected, as the existence of evolutionary bottlenecking is thought to reduce genetic diversity within tumors during metastasis, potentially leading to decreased heterogeneity in metastatic lesions [73]. Consistently, a recent pan-cancer analysis of unpaired primary and metastatic samples (MSK-MET study) also demonstrated that most cancers exhibit greater intratumoral homogeneity in metastases compared to their primaries [74]. We speculate that one possible explanation for this phenomenon is that ITH may have developed prior to oligometastatic spread and remained largely unchanged during tumor evolution in both primary and metastatic sites. The brain oligometastatic lesions are initially colonized by a subset of metastatic cells with brain tropism, which then maintain relatively stable genetic characteristics during growth. Therefore, ITH may not play a key role in the evolutionary process of brain oligometastasis. Oligometastasis is generally considered an indolent disease with limited metastatic capability. This raises an intriguing question about what factors suppress cancer metastasis and contribute to the formation of oligo-BMs in LUAD. Numerous evidence demonstrates that immune cell infiltration, as a key component of the TIME, plays a crucial role in controlling the spread of cancer cells and their subsequent growth in distant organs [75]. First, our study observed an increased infiltration of NK cells in PTs at both mRNA and protein levels. NK cells, as innate immune effectors, have a potent ability to rapidly eliminate malignantly transformed cells [76]. The role of NK cells in inducing tumor dormancy and controlling the metastatic dissemination from PTs has been well established [77, 78]. Previous studies have shown that NK cells can clear cancer cells and reduce the number of metastases without infiltrating established metastatic sites, as their surveillance occurred before tumor cells colonize distant organs [79]. Therefore, the elevated infiltration of NK cells in the primary lesion may serve as a potential mechanism contributing to the formation of oligometastasis. Second, collagen family genes, such as COL1A1 and COL3A1, as well as pathways associated with collagen synthesis, were found to be significantly upregulated in PTs. Collagen represents a principal component of extracellular matrix in the TIME and has profound effects on tumor expansion and distant metastasis during cancer progression [80]. Specifically, type I collagen has been documented to inhibit tumor growth and metastasis in pancreatic ductal adenocarcinoma [81]. Similarly, previous studies have demonstrated that type III collagen enrichment in the TIME can induce tumor dormancy and attenuate breast cancer aggressiveness [82]. In our cohort, the abundance of collagen in primary lesions may help limit metastatic capability of tumors and sustain the indolent state of oligometastatic disease. Third, we identified NLGN1 as a potential gene that drives brain-specific metastasis in lung cancer. This gene exhibits a high degree of consistency across transcriptional levels, methylation regulation patterns, and protein expression. Compared to primary lesions and extracranial metastases, NLGN1 was significantly upregulated in brain metastases. In our internal cohort, increased NLGN1 expression was associated with worse OS and PFS. Subsequent analyses of publicly accessible databases also confirmed that higher NLGN1 expression correlated with worse OS, FP, and more advanced pathological stages. Critically, functional validation in vivo demonstrated that NLGN1 knockdown reduced brain metastatic burden and prolonged brain metastasis-free survival. NLGN1, a postsynaptic protein initially identified in excitatory synapses, plays a key role in synaptic formation and function within the nervous system [83]. Dysfunction of NLGN1 has been linked to the development of cognitive disorders, such as Alzheimer’s disease and autism [84, 85]. Additionally, certain types of cancers have also shown the aberrant expression of NLGN1. In pancreatic and prostatic cancer, it has been demonstrated to promote tumor cell migration and invasion along nerves both in vivo and in vitro [86]. In colorectal cancer, NLGN1 has been implicated in driving aggressive cancer cell behavior and enhancing metastatic capacity via the WNT/APC/β-catenin pathway [87]. However, to our knowledge, no studies have yet explored the role of NLGN1 in promoting brain metastasis in LUAD. Our analysis of gene expressions and DNA methylation changes indicated that the hypermethylation of genebody regions may be a key factor driving the upregulation of NLGN1 expression. These changes appear to be specific to oligo-BMs and have not been observed in primary or extracranial metastatic sites. Taken together, the above results propose possible mechanisms underlying the limited metastatic capacity of oligometastasis and identify NLGN1 as both a novel biomarker and therapeutic target for brain metastasis. Further experimental studies are required to explore the mechanisms of these findings in the future. Cancer cells can evade immune surveillance through various mechanisms, including defects in antigen presentation and overexpression of inhibitory immune checkpoint molecules [88]. In our study, we observed a significant decrease in the MHC-I signature expression score in PTs, while the immune inhibitory receptor PD-L1 was significantly upregulated in oligo-BMs. The downregulation of MHC-I is closely associated with antigen presentation disorders, allowing tumors to escape detection by the immune system [89]. These findings suggest that PTs and oligo-BMs may utilize different mechanisms to escape immune detection. Specifically, PTs may block T cell recognition through defective antigen presentation, whereas oligo-BMs may increase the expression of inhibiting immune checkpoint molecules, resulting in diminished T cell recognition. Such differences reveal the dynamic nature of tumor immune escape, suggesting that tumors can adapt their immune escape strategies to various growth environments and immune selection pressures. PD-L1 is also a predictor of response to immune checkpoint inhibitor treatment. The elevated PD-L1 level in brain metastases presented in our research reflects the spatial heterogeneity of PD-L1 that has been reported in previous study [90], suggesting an inconsistent response that oligo-BMs may respond better to PD-1/PD-L1 inhibitors. This also highlights the importance of considering PD-L1 expression in the tumor microenvironment as a key factor for clinicians when making treatment decisions for patients with brain oligometastasis undergoing immunotherapy. Ock et al. classified a substantial number of samples from the TCGA pan-cancer database into four TMIT by assessing the median expression of PD-L1 and CD8A. Their findings suggest that TMIT I may respond more favorably to immune checkpoint blockade therapy, as these tumors exhibit evidence of preexisting intratumor T cells that are suppressed by PD-L1 engagement [48, 49]. The discordant response of immunotherapy was also supported by the higher proportion of TMIT I observed in oligo-BMs compared with PTs. In the cohort of this study, we observed a significant difference in PD-L1 expression between PTs and oligo-BMs, suggesting the potential existence of distinct molecular mechanisms that regulate PD-L1 expression in these two sites. Recent research showed that PD-L1 expression can be induced extrinsically by IFN-γ secreted from TILs or intrinsically in response to oncogenic signaling. In oligo-BMs, we identified a strongly positive correlation between PD-L1 levels and TILs markers (CD4 and CD8A) as well as IFN-γ, whereas no such correlation was observed in PTs. Therefore, PD-L1 appears to be upregulated by the IFN-γ secretion from TILs in oligo-BMs, implying that the efficacy of immunotherapy might be well predicted by measuring levels of TILs or IFN-γ in the tumor microenvironment. Conversely, PD-L1 expression in PT may be primarily dependent on the activation of oncogenic pathways, which suggests that a combined therapeutic strategy incorporating pathway inhibitors alongside immunotherapy could be more effective. Conclusions In summary, our study comprehensively elucidated the dynamic genomic, transcriptomic, and epigenomic features of LUAD and paired oligo-BM. These findings provided multi-omics evidence that depicts the evolutionary trajectory, immune landscape, and methylation pattern of these patients. A more detailed and precise comprehension of the biological mechanisms underlying the metastatic process may facilitate the development of innovative therapeutic strategies aimed at preventing and targeting brain oligometastases. However, several limitations warrant discussion. First, although we employed relatively more stringent inclusion criteria for brain oligometastasis, the absence of a diagnostic gold standard may still result in the recruitment of polymetastatic patients. Second, our study may be subject to potential biases arising from the unstratified classification of molecular subtypes, such as EGFR or ALK positive, owing to the relatively limited sample size which may constrain the power of our analyses. Third, due to the inherent limitations of targeted panel sequencing, whole-genome doubling status could not be assessed. Moreover, the restricted genomic coverage and single-region sampling may introduce potential biases in the evaluation of genetic heterogeneity between paired lesions. Fourth, a significant limitation of this study is the absence of validation for the molecular mechanisms underlying the key features identified in the sequencing dataset. Therefore, both in vivo and in vitro studies should be pursued in the future to provide stronger evidence. Supplementary Information Additional file 1: Table S1. Clinicopathological characteristics of patients. Additional file 2: Fig. S1. Mutation patterns in PTs and oligo-BMs. Fig. S2. Evolutionary trajectory of each patient. Fig. S3. CCF plots and phylogenetic trees of paired PTs and oligo-BMs. Fig. S4. The association of clonal patterns and seeding mode with clinical parameters. Fig. S5. Comparison of intratumoral heterogeneity between PTs and oligo-BMs. Fig. S6. Mutational signatures observed in PTs and matched oligo-BMs. Fig. S7. Comparison of immune checkpoint molecules expression between PTs and oligo-BMs. Fig. S8. Correlation of PD-L1 expression with CD4, CD8A, and IFNG expression in PTs. Fig. S9. GSVA score of cell death signaling between PTs and oligo-BMs. Fig. S10. GSEA analysis was carried out based on Reactome gene list (PTs vs. oligo-BMs). Fig. S11. GSEA analysis was performed based on KEGG gene list (PTs vs. oligo-BMs). Fig. S12. GSEA analysis was performed based on Gene Ontology gene list (PTs vs. oligo-BMs). Fig. S13. Methylation levels of mC, mCpG, mCHG and mCHH between PTs and oligo-BMs. Fig. S14. CpG methylation levels at different functional genomic regions between PTs and oligo-BMs. Fig. S15. Distribution patterns of DMCs across genomic regions in PTs and oligo-BMs. Fig. S16. Identification of methylation-related differentially expressed genes. Fig. S17. Clinical relevance of CDH4 and CNTNAP2 in LUAD. Fig. S18. Multivariate Cox proportional regression models for our cohort (n = 20). Fig. S19. High-frequency of hypermethylated or hypomethylated genes in our cohort. Additional file 3: Table S2. Mutational signatures in PTs and oligo-BMs. Additional file 4: Table S3. Mutational signatures for primary-only, metastasis-only and shared mutations. Additional file 5: Table S4. Distribution of TMIT in PTs and oligo-BMs. Additional file 6: Table S5. Enriched known transcription factor motifs in hypermethylated regions. Additional file 7: Table S6. Enriched known transcription factor motifs in hypomethylated regions. Additional file 8: Table S7. All CpGs of NLGN1 are hypermethylated in gene body regions. Acknowledgements Not applicable. Abbreviations Non-small cell lung cancer Lung adenocarcinoma Primary tumor Brain metastasis Brain oligometastasis Overall survival Progression free survival First progression Next-generation sequencing Single nucleotide variants Copy number variations Single nucleotide polymorphism Single-sample gene set enrichment analysis Gene set variation analysis Differentially methylated CpGs Transcription factor Reconstruction of Transcriptional Networks Gene Expression Profiling Interactive Analysis Intratumoral heterogeneity Interferon-γ Tumor infiltrating lymphocytes Tumor immune microenvironment Immunogenic cell death Tumor microenvironment immune types Differentially methylated genes Differentially expressed genes Tumor mutation burden Formalin-fixed and paraffin-embedded Kyoto Encyclopedia of Genes and Genomes Homologous recombination deficiency Multiplex immunohistochemistry Immunohistochemistry Authors’ contributions Rongxin Liao, Qian Yu and Yusheng Huang designed the study and wrote the manuscript. Mengmeng Song, Xuan Gao and Lu Shen performed the sequencing of all included samples. Rongxin Liao and Mengmeng Song performed bioinformatics and statistical analysis. Hengqiu He, Yihao Tao, Lin Li, Gang Xiao, Rui Kong, Cancan Wang and Rongrong Zhou collected the tissue samples and clinical data. Shunping Huang, Xiaoyue Zhang, Zaicheng Xu helped with the data analysis. Rongrong Zhou, Zhenzhou Yang and Yuan Peng provided critical comments, suggestions, and revised the manuscript. All authors read and approved the final version of the manuscript. Funding This work was supported by grants from the Natural Science Foundation of China (Grant No. 82573031), the China Postdoctoral Science Foundation (No. 2022M710555) and the Chongqing Medical University Program for Youth Innovation in Future Medicine (No. W0172) to Yuan Peng, the National Natural Science Foundation of China (No. 82273572) to Zhenzhou Yang and Public health key specialty construction project of Chongqing. Data availability Raw bulk RNA-seq, DNA methylation, and targeted panel sequencing data generated in this study have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA018046) [91] that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human. Code used for analysis is available on GitHub (https://github.com/smengmeng30-dev/luad-oligo-bm-multiomics/) [92]. Declarations Ethics approval and consent to participate The study involving human tissue samples was approved by the Ethics Committee of Chongqing Medical University (Reference No. 2020 − 171). Written informed consent was obtained from each participant prior to enrollment. All procedures were conducted in accordance with the principles of the Helsinki Declaration and under the supervision of the ethics committee. All animal experiments were approved by the Animal Ethics Committee of the Second Affiliated Hospital of Chongqing Medical University (Reference No. 2022 − 285) and were performed in accordance with the relevant institutional guidelines and regulations for animal care and use. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Rongxin Liao, Qian Yu, Yusheng Huang and Yuan Peng contributed equally to this article. Contributor Information Rongrong Zhou, Email: zhourr@csu.edu.cn. Zhenzhou Yang, Email: yangzz@cqmu.edu.cn. Yuan Peng, Email: pengyuan1127@cqmu.edu.cn. References Associated Data Supplementary Materials Additional file 1: Table S1. Clinicopathological characteristics of patients. Additional file 2: Fig. S1. Mutation patterns in PTs and oligo-BMs. Fig. S2. Evolutionary trajectory of each patient. Fig. S3. CCF plots and phylogenetic trees of paired PTs and oligo-BMs. Fig. S4. The association of clonal patterns and seeding mode with clinical parameters. Fig. S5. Comparison of intratumoral heterogeneity between PTs and oligo-BMs. Fig. S6. Mutational signatures observed in PTs and matched oligo-BMs. Fig. S7. Comparison of immune checkpoint molecules expression between PTs and oligo-BMs. Fig. S8. Correlation of PD-L1 expression with CD4, CD8A, and IFNG expression in PTs. Fig. S9. GSVA score of cell death signaling between PTs and oligo-BMs. Fig. S10. GSEA analysis was carried out based on Reactome gene list (PTs vs. oligo-BMs). Fig. S11. GSEA analysis was performed based on KEGG gene list (PTs vs. oligo-BMs). Fig. S12. GSEA analysis was performed based on Gene Ontology gene list (PTs vs. oligo-BMs). Fig. S13. Methylation levels of mC, mCpG, mCHG and mCHH between PTs and oligo-BMs. Fig. S14. CpG methylation levels at different functional genomic regions between PTs and oligo-BMs. Fig. S15. Distribution patterns of DMCs across genomic regions in PTs and oligo-BMs. Fig. S16. Identification of methylation-related differentially expressed genes. Fig. S17. Clinical relevance of CDH4 and CNTNAP2 in LUAD. Fig. S18. Multivariate Cox proportional regression models for our cohort (n = 20). Fig. S19. High-frequency of hypermethylated or hypomethylated genes in our cohort. Additional file 3: Table S2. Mutational signatures in PTs and oligo-BMs. Additional file 4: Table S3. Mutational signatures for primary-only, metastasis-only and shared mutations. Additional file 5: Table S4. Distribution of TMIT in PTs and oligo-BMs. Additional file 6: Table S5. Enriched known transcription factor motifs in hypermethylated regions. Additional file 7: Table S6. Enriched known transcription factor motifs in hypomethylated regions. Additional file 8: Table S7. All CpGs of NLGN1 are hypermethylated in gene body regions. Data Availability Statement Raw bulk RNA-seq, DNA methylation, and targeted panel sequencing data generated in this study have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2025) in National Genomics Data Center (Nucleic Acids Res 2025), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (GSA-Human: HRA018046) [91] that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human. Code used for analysis is available on GitHub (https://github.com/smengmeng30-dev/luad-oligo-bm-multiomics/) [92].