NSCLC presents two biologically distinct therapeutic challenges. Oncogene-addicted NSCLC (OA-NSCLC), comprising roughly 50% of nonsquamous cases, harbors targetable driver mutations such as EGFR, ALK, ROS1, KRAS G12C, and MET exon 14. Non-oncogene-addicted NSCLC (NOA-NSCLC) lacks these drivers and relies primarily on immunotherapy, yet a substantial proportion of patients derive limited benefit. Multiomic approaches offer the greatest potential to unlock new treatment strategies for this harder-to-treat NOA-NSCLC group.
Genomics alone cannot capture cancer's full complexity. Clinical oncology has been transformed by DNA-level discoveries, yet tumor biology is governed by layers of regulation beyond the genome: transcription, post-translational modification, metabolism, epigenetics, and the microenvironment. Single-omic analyses miss the epistatic interactions between these layers that ultimately determine drug sensitivity, immune evasion, and metastatic potential.
Non-genomic omics remain largely preclinical in NSCLC. While genomic sequencing now routinely informs treatment decisions, the clinical translation of transcriptomics, proteomics, epigenomics, metabolomics, microbiomics, and spatial omics is still nascent. This review synthesizes the current state and future trajectory of each domain and their integration in NSCLC, with emphasis on actionable findings.
Spatial omics adds a critical topographic dimension. Bulk omic methods average signals across heterogeneous cell populations, obscuring clinically relevant spatial organization. Spatial transcriptomics and proteomics platforms now resolve gene and protein expression within tissue architecture, revealing how the positions of immune, stromal, and tumor cells relative to one another determine immunotherapy response and prognosis.
RNA sequencing reveals actionable fusion transcripts and molecular subtypes. Transcriptomic profiling by RNA-seq enables detection of gene fusions that are missed by DNA sequencing when breakpoints fall in intronic regions. Clinical trials such as the WINTHER study demonstrated that RNA-based tumor profiling could guide therapeutic decisions beyond what genomic testing alone provided, particularly for identifying fusions driving aberrant signaling.
Molecular subtyping stratifies prognosis within histological groups. Transcriptomic clustering of lung adenocarcinomas and squamous cell carcinomas reveals gene expression subtypes with distinct survival outcomes and treatment sensitivities. These molecular subtypes can refine the TNM staging framework by identifying patients with aggressive biology even within early-stage disease categories.
RNA therapeutics represent an emerging translational opportunity. Small interfering RNA (siRNA) and mRNA-based therapeutic strategies exploit transcriptomic discoveries to silence oncogenes or restore tumor suppressors. Nanoparticle delivery systems have improved the stability and tumor targeting of RNA-based agents, opening a new class of lung cancer therapeutics grounded directly in transcriptomic insights.
Liquid biopsy transcriptomics enables noninvasive tumor monitoring. Cell-free RNA in plasma and exosomal RNA from tumor cells can be profiled to track treatment response and early resistance emergence. This noninvasive approach is particularly valuable in NSCLC, where serial tissue biopsies are often impractical, and complements cell-free DNA approaches already entering clinical practice.
The proteome reveals biology invisible to genomic and transcriptomic analysis. Mass spectrometry-based proteomics, reverse phase protein array (RPPA), and Olink proximity extension assays quantify protein abundance and post-translational modifications including phosphorylation, ubiquitination, and acetylation. The APOLLO network study demonstrated that proteomic and phosphoproteomic profiling of lung adenocarcinomas identified biologically distinct subtypes not captured by genomic or transcriptomic data alone.
PD-L1 protein and mRNA levels are poorly correlated. A critical finding from proteomic studies is that PD-L1 protein expression measured by immunohistochemistry does not reliably reflect PD-L1 mRNA levels. This discordance, driven by post-translational regulation, helps explain why mRNA-based biomarker assays for immunotherapy response have underperformed compared to protein-level measurements, and underscores the necessity of proteomic data for immune checkpoint biomarker development.
DNA methylation alterations are among the earliest epigenetic events in NSCLC. Promoter hypermethylation of tumor suppressor genes including CDKN2A silences critical cell cycle regulators, while hypomethylation at loci such as CD147 promotes tumor invasiveness. These methylation events represent both diagnostic biomarkers detectable in cell-free DNA and potential therapeutic targets for demethylating agents combined with immunotherapy.
Histone modifications and non-coding RNAs shape the NSCLC epigenome. KDM2A and HDAC6 are among the chromatin-modifying enzymes altered in NSCLC, driving gene expression reprogramming that supports tumor growth. MicroRNAs including miR-29c and miR-26A1 function as tumor suppressors frequently lost in NSCLC, while lncRNAs such as MEG3 regulate oncogenic pathways. These epigenetic regulators are increasingly targeted by small molecule inhibitors under clinical investigation.
The Warburg effect and its metabolic consequences shape the NSCLC microenvironment. NSCLC cells preferentially use aerobic glycolysis, generating lactate that acidifies the tumor microenvironment and suppresses immune cell function. Plasma lactate dehydrogenase levels have been correlated with outcomes in patients receiving immune checkpoint inhibitors, suggesting that metabolic profiling could serve as a dynamic biomarker of immunotherapy benefit.
Lipidome and bile acid profiles predict immunotherapy response. Metabolomic profiling has identified lipidomic signatures and altered bile acid pathways that associate with response to PD-1/PD-L1 inhibitors in NSCLC. Changes in citrate cycle intermediates have also been linked to immune checkpoint inhibitor outcomes, reflecting the bidirectional relationship between cancer metabolism and immune cell activity within the tumor microenvironment.
Gut microbiota composition influences systemic immunotherapy efficacy. The gut microbiome modulates systemic immune tone through bile acid metabolism, short-chain fatty acid production, and direct immune cell programming. In NSCLC patients, baseline gut microbiome composition has been associated with differential response to PD-1 inhibitors, and broad-spectrum antibiotic use that disrupts the microbiome has been linked to inferior immunotherapy outcomes.
Intratumoral bacteria are an emerging component of the NSCLC microenvironment. Recent studies have detected bacteria from the phyla Proteobacteria, Firmicutes, and Actinobacteria within NSCLC tumor tissue. Intratumoral Escherichia species have been associated with improved survival in patients receiving immune checkpoint inhibitors, suggesting that the intratumoral microbiome participates in shaping local immune responses in ways that can be therapeutically exploited.
Spatial transcriptomics platforms enable in situ gene expression mapping. Technologies including GeoMx Digital Spatial Profiler and 10x Genomics Visium measure transcriptomic profiles while preserving tissue spatial context. These platforms can resolve gene expression within tumor nests, invasive margins, stroma, and tertiary lymphoid structures simultaneously, revealing how physical location within the tumor ecosystem shapes cellular identity and function.
Spatial positioning of CD8+ T cells and macrophages predicts immunotherapy outcomes. Spatial proteomic approaches including multiplexed immunofluorescence and imaging mass cytometry have demonstrated that the proximity of CD8+ cytotoxic T cells to tumor cells, rather than their total abundance, is a stronger predictor of response to PD-1/PD-L1 inhibitors. Tumor-associated macrophage spatial distribution and polarization state within defined tissue niches similarly influence prognosis independently of bulk immune cell counts.
Cancer-associated fibroblasts create immunosuppressive spatial barriers. Spatial omics analyses have identified distinct cancer-associated fibroblast subpopulations that physically exclude cytotoxic lymphocytes from tumor nests by forming stromal barriers. These fibroblast-rich exclusion zones predict resistance to immunotherapy and represent potential targets for stromal reprogramming strategies designed to restore immune cell infiltration.
Tertiary lymphoid structures are spatially defined immune hubs. Tertiary lymphoid structures (TLS), which are ectopic lymphoid organs formed within tumors, have been associated with improved outcomes in NSCLC patients receiving immunotherapy. Spatial omics has clarified that the density and maturation state of TLS, as well as the spatial interactions between TLS B cells, T follicular helper cells, and dendritic cells, are the biologically active features rather than TLS presence alone.
Early versus late integration represent fundamentally different analytical philosophies. Early integration concatenates data from multiple omic layers before modeling, preserving cross-modal interactions but amplifying the curse of dimensionality. Late integration trains separate models on each omic dataset and combines their outputs, which is more computationally tractable but may miss inter-omic synergies. Selection between these strategies depends on sample size, data modality, and the specific biological question being addressed.
Autoencoders compress multiomics data into biologically meaningful representations. Deep learning autoencoders trained on combined TCGA genomic, transcriptomic, methylation, and proteomic data from NSCLC samples produced integrated representations that outperformed any single omic layer for tumor subtype classification. These latent representations capture non-linear cross-modal interactions that linear dimensionality reduction methods such as PCA cannot detect.
Network and graph-based methods model inter-omic relationships explicitly. Graph neural networks and similarity network fusion approaches model the relationships between omic layers as edges in a biological network, enabling propagation of signals across data modalities. These methods are particularly suited to capturing the cascading effects of upstream genomic events on downstream proteomic and metabolomic phenotypes.
The high-p, small-n problem remains the central challenge for multiomics in NSCLC. NSCLC multiomics studies face a fundamental statistical challenge: the number of measured features across omic layers far exceeds the number of available patient samples with complete multi-layer data and long-term clinical follow-up. Addressing this limitation requires harmonized data collection across multicenter trials, standardized biobanking protocols, and regulatory frameworks that incentivize comprehensive molecular profiling alongside clinical endpoints.
Only genomics has achieved broad clinical impact in NSCLC to date. Despite the scientific richness of transcriptomic, proteomic, epigenomic, metabolomic, and spatial omic discoveries, none of these domains has yet been translated into routine clinical decision-making for NSCLC. Genomic sequencing for targetable alterations remains the sole omic approach with established clinical utility, reflecting the translational gap that multiomic research must bridge.
NOA-NSCLC represents the highest-priority target for multiomic clinical translation. Patients with non-oncogene-addicted NSCLC lack actionable genomic targets and have heterogeneous responses to immunotherapy. Multiomic profiling offers the greatest potential to stratify this population into biologically coherent subgroups with distinct therapeutic vulnerabilities, making NOA-NSCLC the natural proving ground for integrative omic approaches.
Prospective multicenter trials with embedded multiomic endpoints are urgently needed. Retrospective analyses of archived samples have generated hypotheses but cannot establish clinical utility. Prospective trials that collect matched tissue, blood, and microbiome samples across treatment time points with prespecified multiomic endpoints are essential for validating whether any individual omic signature or integrated multiomics panel can guide treatment selection.
Standardization, data sharing, and computational infrastructure must advance in parallel. The multiomic field is fragmented by platform heterogeneity, batch effects, and siloed datasets that prevent cross-study validation. Establishing consensus standards for sample collection, processing, and bioinformatic analysis, combined with federated data-sharing frameworks that protect patient privacy, will be critical prerequisites for accelerating clinical translation of NSCLC multiomics.