The anatomical extent of disease and surgical feasibility usually define the management of resectable non-small cell lung cancer (NSCLC), yet postoperative recurrence is often driven by biological events that precede surgery. Despite advances in neoadjuvant, adjuvant, and perioperative systemic therapies, postoperative relapse remains a major cause of treatment failure following curative-intent therapy (1–5). A tumor that has been completely resected macroscopically may harbour genetically diverse subclones with the capacity for invasion, dissemination, immune escape, and survival under therapeutic pressure. This distinction is important because postoperative recurrence is not simply a failure of local control. In many patients, it may represent the delayed clinical manifestation of tumor populations that have disseminated and evolved before resection.

Evolutionary studies have provided a biological explanation for the disconnect between anatomical resectability and postoperative risk of recurrence. The TRACERx program has shown that intratumoral heterogeneity, subclonal architecture, copy number instability, and whole-genome doubling are associated with patterns of tumor evolution, progression, and clinical outcomes in lung cancer (6,7). Further analyses of metastatic progression have suggested that metastases may arise from pre-existing or evolving tumor clones subjected to selective pressures, rather than through a simple linear process (8). In parallel, ultrasensitive preoperative circulating tumor DNA (ctDNA) analysis of early-stage lung adenocarcinoma has shown that molecularly detectable disease before surgery can identify patients at increased risk of adverse outcomes, providing evidence that clinically occult tumor burden may persist despite apparently localized disease (9).

Genomic instability provides a conceptual framework that links these observations. Rather than representing a single molecular abnormality, genomic instability encompasses diverse processes and genomic features, including chromosomal instability (CIN), copy number alterations (CNAs), structural variants, aneuploidy, DNA repair defects, replication stress, and genome-wide mutational patterns (10). These processes can increase the genetic diversity available for selection, promote the survival of clinically important subclones, and in some contexts, create candidate therapeutic vulnerabilities. Therefore, genomic instability may contribute to tumor initiation and progression, metastatic dissemination, and persistence of residual disease.

The present review evaluated the genomic instability as a biological framework for interpreting recurrence after NSCLC resection. Rather than treating it as a single, interchangeable construct, each instability process was assessed across three translational dimensions: i) Its principal biological consequence; ii) its potential clinical role as a prognostic marker, treatment-predictive biomarker, or therapeutic vulnerability; and iii) the maturity, directness, and clinical relevance of the supporting evidence in resectable NSCLC. This approach distinguished features with relatively direct evolutionary or prognostic relevance from those supported mainly by mechanistic studies or evidence derived from advanced disease. Notably, the relationship between genomic instability and molecular residual disease (MRD) was treated as a testable evolutionary hypothesis rather than an established biological or causal relationship. Key unresolved controversies were also highlighted and strategies for integrating tumor profiling with longitudinal MRD assessment were outlined. The aim was to define the boundaries of current evidence, identify research priorities, and clarify where genomic instability may complement, but should not replace, established clinical, pathological, or molecular decision-making. The conceptual relationships between genomic instability, clonal evolution, occult residual disease, and postoperative recurrence are summarized in Fig. 1.

Genomic instability in NSCLC is not a single-molecule phenomenon. It arises from several related but biologically distinct processes and genomic alterations. For example, CIN results from ongoing errors in chromosome segregation and maintenance, leading to aneuploidy, chromosome-arm gains or losses, and highly complex karyotypes in some tumors (11). CNAs represent another important component of genomic complexity. These changes may affect oncogenes, tumor suppressor genes, or immune-related loci, and their distribution may vary substantially across different regions of the same tumor. Notably, a recent lung cancer study linked clone-level copy-number diversity to survival, suggesting that copy-number complexity is not simply a descriptive feature but may reflect clinically relevant evolutionary behavior (12).

Whole-genome doubling increases the complexity. Once a tumor cell duplicates its entire chromosomal complement, it may tolerate and accumulate additional chromosomal alterations. TRACERx-based analyses have shown that FAT atypical cadherin 1 (FAT1) loss, Hippo pathway dysregulation, CIN, and whole-genome doubling are associated with NSCLC. Similarly, studies across cancer types have shown that genome-doubled tumors can develop complicated chromosomal histories that may be difficult to reconstruct from a single sampled region (13,14). Structural variants contribute in different ways; they can generate actionable fusions, disrupt tumor suppressor loci, or alter regulatory elements. Genomic changes do not occur in isolation either; during NSCLC evolution, they may interact with epigenetic remodeling, underscoring the close relationship between genome structure and transcriptional state (15). Extrachromosomal DNA (ecDNA) represents a dynamic, non-chromosomal form of amplified DNA that can undergo rapid structural and copy-number evolution. Experimental research suggests that ecDNA-positive cancers may develop dependencies related to transcription-replication conflicts and could be vulnerable to checkpoint kinase 1 (CHK1) inhibition (16).

Instability can also arise from defective DNA repair pathways and persistent replication stress. Deficiencies in homologous recombination, mismatch repair, nucleotide excision repair, base excision repair, or non-homologous end joining can increase the accumulation of mutations, replication-associated lesions, and structural genomic abnormalities (17). When one repair pathway is impaired, tumor cells may become more dependent on alternative repair pathways or cell-cycle checkpoints for survival. This provides a biological rationale for DNA-damaging therapies and DNA damage response (DDR)-targeted strategies in selected settings, although the clinical relevance is likely to vary according to specific molecular defect and cellular context (18). Mutational signatures provide a complementary view of genomic instability and tumor evolution. Rather than identifying a single alteration, they capture patterns of mutations that reflect the processes shaping the tumor genome over time, including carcinogen exposure, enzymatic editing, and repair deficiency. In lung cancer, apolipoprotein B mRNA editing enzyme catalytic polypeptide-like activity has been linked to tumor evolution and treatment resistance, and whole-genome studies have identified distinct mutagenic processes in smoking-associated and never-smoker lung cancers (19–21).

These differences may have distinct clinical implications. Tumors with extensive CIN may have marked clonal diversity and greater capacity for ongoing genomic evolution; however, CIN itself does not currently define a clinically actionable repair defect. Conversely, a tumor with a specific homologous recombination defect may provide a rationale for a therapeutic hypothesis, even when broad aneuploidy is not prominent. Collapsing these different processes into a single label such as ‘high genomic instability’ would therefore risk obscuring the biological and therapeutic differences that may eventually make them useful.

Measurement is equally important. Targeted DNA or RNA sequencing can identify established driver alterations and selected copy-number events but is less suited to comprehensively characterizing global chromosomal complexity, whole-genome doubling, complex structural variation, or mutational signatures (22). Broader genomic profiling is often required to assess these features, while characterization of clonal diversity may additionally require multi-region sampling or phylogeny-informed analysis (23). Repair deficiency presents a particular challenge because alterations in homologous recombination repair genes and genomic-scarring or mutational-signature measures show incomplete concordance in NSCLC and should not be treated as interchangeable readouts (24). This situation is even less mature for ecDNA, which remains a research-stage structural feature for which standardized perioperative assays are not yet available (25). Surgical specimens obtained from patients with resectable NSCLC offer important opportunities for exploratory profiling. However, clinical application requires pre-specified analytical methods, reproducible thresholds, and prospective evidence that these measurements add useful information beyond standard pathology, driver status, and longitudinal MRD assessment. The potential measurement strategies and their current translational relevance for major genomic instability processes in resectable NSCLC are summarized in Table I.

Genomic instability can lead to carcinogenesis and tumor progression by continuously generating genetic diversity within the primary tumor. Resectable NSCLC should therefore not be viewed as a single uniform lesion, but as a spatially structured population of genetically related subclones. Single-cell and spatial transcriptomic studies have shown that malignant cells, immune populations, and stromal components in NSCLC form organized spatial communities rather than random cellular mixtures (26,27). This complexity is clinically important because the subclone that ultimately drives recurrence may not be the dominant clone represented in a single resected specimen.

However, sampling limitations further complicate this interpretation. A focal biopsy or a single tissue block may capture only a fraction of the genomic diversity within a lung tumor (28). Longitudinal evolutionary analyses have shown that multiple subclones from a primary tumor may disseminate and contribute to subsequent metastatic disease (29). Therefore, a minor subclone may become clinically important if it acquires or possesses selective advantages in invasion, survival, immune escape, or adaptation to distant microenvironments.

The tumor microenvironment helps determine which subclones persist and expand. Hypoxia, inflammation, stromal pressure, nutrient limitation, vascular invasion, and immune surveillance can act as selective forces during tumor evolution before surgery. Spatial Tumor Atlas studies have shown that malignant cell evolution is closely linked to the surrounding microenvironment, and the spatial organization of immune cells in lung cancer is associated with immune evasion and clinical outcomes (30,31). In this setting, genomic instability should not be viewed as a direct or sufficient cause of recurrence. Rather, it expands the pool of variants, from which microenvironment and environmental and therapeutic selection may favor subclones with greater metastatic potential, immune-evasive capacity, and treatment resistance. This provides a biological link between genomic instability and metastatic progression. Tumor cells capable of leaving the primary lesion before surgery must survive in circulation, adapt to distant tissue environments, and either establish immediate growth or enter a state of disseminated tumor cell dormancy. Dormant disseminated tumor cells may remain clinically invisible for prolonged periods while evading immune surveillance, and their later outgrowth can be supported by stromal, immune, and tissue-niche interactions (32–34). A recent functional study on NSCLC showed that immune-evasive and non-immune-evasive subclones can coexist within the same tumor, suggesting that immune escape may emerge during subclonal evolution (35). The clinical relevance of genomic instability therefore lies not in a direct deterministic effect on recurrence, but in its potential to generate diversity that enables adaptation, dissemination, immune evasion, and persistence.

Postoperative recurrence is the clinical endpoint of a biological process that may begin well before recurrence becomes clinically detectable. Although recurrence is typically recognized by imaging, pathology, or clinical symptoms, the underlying events may include preoperative clonal diversification, occult dissemination, persistence of residual tumor cells, and subsequent metastatic outgrowth. Multiomics profiling of recurrent stage I NSCLC has shown that tumors that subsequently relapse may differ in their genomic, epigenomic, transcriptomic, and microenvironmental features (36). This biological heterogeneity may help explain why patients with similar anatomical stages and apparently complete resection can experience markedly different postoperative outcomes.

Genomic instability may contribute to this process on several levels. First, it increases the diversity from which invasive or stress-tolerant subclones emerge within the primary tumor. Second, it may support metastatic competence by generating alterations that favor cellular motility, vascular invasion, immune escape, or adaptation to distant microenvironments. Third, it may facilitate the persistence of residual tumor cells under postoperative and therapeutic stress. Pulmonary venous circulating tumor cell research has shown that tumor cells may enter the circulation before resection, and that such early dissemination is associated with subsequent relapse and metastatic disease (37). Tumor dormancy research further suggests that disseminated residual tumor cells can persist for prolonged periods before undergoing clinical outgrowth (38).

This provides an important context for understanding the role of MRD. Presurgical ctDNA detection and postoperative ctDNA-defined MRD both have demonstrated prognostic value in early-stage NSCLC (39,40). Imaging-based features, including radiological tumor volume and positron emission tomography/computed tomography-derived parameters, may add another layer of relapse risk assessment when interpreted in conjunction with ctDNA (41,42). In resected epidermal growth factor receptor-mutated NSCLC, exploratory analyses from the ADAURA trial further suggested that MRD signals can remain informative, even in patients receiving effective adjuvant epidermal growth factor receptor (EGFR)-targeted therapy (43).

However, MRD should not be interpreted as a complete or infallible measure of residual disease. The recurrence risk remains heterogeneous, and MRD assays require further analytical and clinical standardization before they can routinely guide treatment decisions (44,45). A positive postoperative ctDNA result may indicate that one or more tumor populations have survived resection and remain detectable in the circulation. Conversely, a negative result is reassuring but does not exclude residual disease, particularly in tumors with low shedding, limited representation of the primary tumor in the sampled tissue, or marked spatial heterogeneity.

Complete resection remains essential, but it cannot reverse evolutionary events that have already occurred. Resected tumors with extensive clonal diversity may harbor subclones that have disseminated, entered dormancy, or acquired immune-evasive properties before surgery. Another tumor at the same stage may not have undergone the same extent of evolutionary diversification. What later appears as postoperative recurrence may, therefore, be the delayed manifestation of carcinogenic and metastatic processes that were established before surgery in some patients.

Following surgery, ctDNA-defined MRD may provide one of the earliest indications that tumor-derived molecular material remains detectable after resection of the primary lesion. A positive result is consistent with the presence of residual tumor-derived DNA and may reflect clones that disseminated before surgery and persist during the early postoperative period. However, ctDNA-defined MRD is an assay-dependent molecular signal and is not a direct measure of viable residual tumor cells, residual tumor burden, or genomic instability. Therefore, it is best understood as an imperfect but potentially informative biomarker of molecular residual disease and early recurrence risk, rather than a direct measure of residual clonal burden or evolution (46,47).

The relationship between genomic instability and MRD should be regarded as a biologically plausible, testable hypothesis rather than an established causal pathway. Experimental and evolutionary studies have suggested that genomic instability expands clonal diversity and increases the probability that invasive, stress-tolerant, or immune-evasive populations emerge and persist (48–50). However, direct clinical evidence linking specific instability features, such as CIN, copy number complexity, whole-genome doubling, or repair deficiency, to the presence or longitudinal dynamics of postoperative MRD in resectable NSCLC remains limited. Studies characterizing genomic instability patterns and their clinical associations (51) and those linking postoperative MRD with subsequent relapse in resectable NSCLC (52) currently represent related but largely distinct evidence streams.

An MRD result reflects both the biological presence of residual disease and its detectability in the sampled blood compartment. Detectability is influenced by the residual tumor burden, biological shedding, access to the circulation, blood collection timing, and assay sensitivity (53,54). Spatial and temporal intratumoral heterogeneity introduce additional uncertainty, particularly in tissue-informed assays. A tissue block selected for sequencing may represent the dominant primary tumor clone while failing to capture a minor clone that subsequently persists or disseminates. Tumor-derived sequence variants can be used as tracking markers in tumor-informed or targeted sequencing assays (55–58), whereas fragmentomic (59–61) and methylation-based approaches (62) may capture broader signals; however, none eliminate the effects of low shedding, limited vascular access, or very small residual burden. Consequently, a negative result does not establish the biological absence of residual disease (63). Conversely, an association between a genomic instability feature and MRD positivity could partly reflect differences in tumor burden, proliferation, necrosis, or ctDNA shedding rather than a direct effect of genomic instability on the persistence of residual tumor cells.

Therefore, three levels of evidence should be distinguished. First, mechanistic evidence may explain how an unstable process facilitates genomic diversification, dissemination, and survival under therapeutic pressure. Second, prognostic evidence may indicate that a genomic instability feature or postoperative MRD status is associated with recurrence or survival. Third, predictive evidence must demonstrate that a biomarker identifies differential treatment benefit, ideally through a biomarker-by-treatment interaction. Mechanistic plausibility and prognostic associations do not establish predictive utility. At present, no genomic instability feature has demonstrated treatment-predictive value in resectable NSCLC, and no validated perioperative model has established that genomic instability profiling provides prognostic or predictive information beyond standard clinicopathological assessment and longitudinal MRD.

Tissue profiling and serial blood monitoring can provide complementary but distinct information. Analysis of resected tumors can characterize their evolutionary architecture, including chromosomal complexity, repair defects, and other instability-related features. Serial ctDNA testing can determine whether tumor-derived molecular signals persist, clear, or re-emerge after surgery or perioperative treatment (64). Persistent MRD after the resection of a genomically complex tumor may represent a different biological context from a transient low-level signal arising from a tumor with limited pathological risk; however, whether such combined profiling provides clinically actionable information remains unproven and should not yet be considered a validated basis for treatment intensification or surveillance planning.

Prospective studies should pair prespecified, analytically validated tissue-based measures of genomic instability with standardized postoperative and longitudinal blood sampling. Such studies should account for residual tumor burden, DNA shedding, vascular accessibility, spatial and temporal heterogeneity, sampling time, and assay sensitivity while using clinically relevant endpoints such as recurrence-free survival and MRD clearance or conversion. They should also assess whether integrating instability profiling with MRD improves risk stratification beyond established clinicopathological factors and driver status, and whether any instability phenotype identifies differential benefit from a matched intervention through a biomarker-by-treatment interaction. Until these questions are resolved, the genomic instability-MRD relationship should remain a research framework rather than a basis for routine perioperative treatment decisions.

The immune consequences of genomic instability cannot easily be reduced to simple rules. An unstable genome may generate mutations, structural changes, and novel peptides that can serve as potential tumor antigens. However, a greater number of mutations does not automatically make a tumor more susceptible to immune recognition or control. What matters is whether the relevant neoantigens are expressed and clonally represented, whether T cells can recognize and reach the tumor, and whether the surrounding microenvironment permits an immune response (65,66). This explains why tumors with a high mutational burden may still evade immunity when antigen presentation is impaired, effector T cells are excluded, or immunosuppressive programs predominate. In NSCLC, acquired resistance to immunotherapy has been linked to defects in antigen processing, interferon signaling, and immune cell function (67). The genomic background also matters: Alterations in serine/threonine kinase 11 and Kelch-like ECH-associated protein 1 have been associated with distinct immune profiles and poorer outcomes after programmed cell death protein 1/programmed death-ligand 1 (PD-L1) blockade, particularly in KRAS-mutated lung adenocarcinoma (68,69).

CIN introduces an additional layer of complexity. Chromosome missegregation can generate micronuclei and activate the cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) pathway. In experimental settings, this response may induce type I interferon signaling and enhance antitumor immune responses (70). In other contexts, particularly when signaling becomes chronic or dysregulated, it may promote a pro-metastatic and immunosuppressive environment (71). CIN is therefore not simply ‘beneficial’ or ‘detrimental’ for antitumor immunity. Its effects are likely to depend on the magnitude and duration of cGAS-STING signaling, as well as on the local cellular and molecular context of the tumor.

This uncertainty is clinically relevant now that perioperative immunotherapy has become an established component of treatment for selected patients with resectable NSCLC. Phase III studies evaluating perioperative durvalumab and nivolumab-based strategies have demonstrated improvements in event-free survival, supporting the efficacy of perioperative chemoimmunotherapy in appropriately selected patients (72,73). Meta-analytic evidence similarly supports neoadjuvant chemoimmunotherapy across pathological response, surgical, and efficacy outcomes, although the magnitude and interpretation vary according to disease stage, PD-L1 expression, treatment regimen, and clinical setting (74). Genomic instability features may help explain some of this heterogeneity; however, they are not yet sufficiently validated to serve as independent biomarkers for perioperative immunotherapy selection. Their potential relevance should instead be interpreted alongside established clinical, molecular, and immune biomarkers (75). Postoperative ctDNA-defined MRD may provide a complementary, dynamic measure of residual disease risk within this broader assessment (76).

Genomic instability can accelerate tumor evolution. In some biological settings, it also creates dependencies on pathways that enable tumor cells to tolerate ongoing DNA damage, including residual repair capacity, cell-cycle checkpoints, and stress-response mechanisms. These dependencies provide a rationale for investigating DDR-directed strategies and agents targeting replication stress or checkpoint pathways. However, in resectable NSCLC, this remains a research hypothesis rather than an established treatment-selection strategy. Platinum-based chemotherapy is an established perioperative systemic treatment, and radiotherapy is used in selected locoregional settings. Neither approach is currently routinely selected on the basis of CIN, whole-genome doubling, homologous recombination deficiency, replication stress, or other genomic instability biomarkers (77,78).

Homologous recombination deficiency provides one potential example. In early-stage resectable NSCLC, a genomic scarring score has been explored as a candidate marker of poly(ADP-ribose) polymerase (PARP) inhibitor sensitivity, although the supporting therapeutic evidence remains largely preclinical, and PARP inhibitors are not part of standard perioperative treatment (79). Replication stress raises additional therapeutic possibilities. Tumors with oncogenic activation, tumor protein p53 dysfunction, or impaired checkpoint control may become increasingly reliant on ataxia telangiectasia and Rad3-related (ATR), CHK1, or WEE1 G2 checkpoint kinase (WEE1) to tolerate persistent DNA damage. ATR inhibitors have shown early clinical activity in biomarker-selected advanced solid tumors (80,81). WEE1 inhibition has preclinical support for KRAS/tumor protein p53-mutant NSCLC (82), whereas DNA-dependent protein kinase (DNA-PK) inhibition has entered early-phase clinical evaluation in advanced cancers (83). These findings provide a rationale for developing and prospectively validating predictive biomarkers, but they do not support treatment selection for resectable NSCLC.

The broader experience with targeting genomic instability-related dependencies also underscores the need for caution. Therapeutic strategies directed at replication stress and DDR dependencies are biologically plausible (84), but most clinical evidence comes from advanced solid tumors or metastatic NSCLC rather than from patients treated with curative intent. In the HUDSON study, biomarker-directed treatment incorporating ATR inhibition and PD-L1 blockade showed clinical activity in advanced NSCLC. By contrast, phase III trials evaluating PARP inhibitor-immunotherapy maintenance approaches have not established a broadly applicable treatment strategy for metastatic disease (85–87). Therefore, it is premature to extend these approaches to routine perioperative care. Homologous recombination deficiency, replication stress phenotypes, and checkpoint dependence are better viewed as candidate biomarkers for prospective, biomarker-defined clinical trials (88). ecDNA represents a distinct candidate biomarker, and the bromodomain and extraterminal domain-containing protein 4-dependent organization of ecDNA hubs may represent a therapeutic vulnerability that requires prospective validation (89,90).

Notably, these candidate vulnerabilities are unlikely to be interchangeable. Repair-deficient tumors may be suitable for prospective evaluation of DDR-directed therapies, whereas tumors with prominent replication stress may be more relevant for studies on ATR, CHK1, or WEE1 dependence. ecDNA-positive cancers may represent another biological setting in which transcription-replication conflicts or CHK1-related dependencies warrant investigation. CIN is less directly actionable, but it may influence the tumor immune microenvironment through context-dependent inflammatory signaling. Its therapeutic implications remain uncertain and should not be used to guide treatment selection without prospective validation. The practical aim, then, is not to assign existing perioperative therapies based on genomic instability, but to identify well-defined instability phenotypes that can be prospectively and rigorously evaluated in future clinical trials.

The perioperative management of resectable NSCLC remains grounded in anatomical stage, pathological risk features, actionable driver alterations, and treatment strategies supported by prospective clinical trials. The ninth edition of the tumor-node-metastasis classification provides the current anatomical framework (91), whereas established molecular alterations identify patients eligible for specific targeted therapies (92–94). Postoperative ctDNA-defined MRD is emerging as a longitudinal indicator of residual disease risk, although its utility for treatment selection remains under investigation. Genomic instability features are less clinically mature and should be evaluated on a case-by-case basis, according to the specific clinical claim under consideration, rather than treated as a single, interchangeable biomarker category.

Therefore, a three-axis translational framework is proposed. The first axis defines the principal biological consequence of each feature: The generation of clonal diversity, tolerance to genomic disruption, defective DNA repair, dependence on replication-stress pathways, or dynamic adaptation mediated by ecDNA. The second axis defines its potential clinical role as a prognostic marker, treatment-predictive biomarker, or therapeutic vulnerability. The third axis considers evidence maturity, ranging from mechanistic and preclinical observations to retrospective clinical associations, prospective prognostic validation, and prospective treatment-specific predictive evidence. These levels are not interchangeable. A biologically plausible dependency is not necessarily a predictive biomarker, and an association with recurrence does not establish that treatment decisions based on the feature will improve clinical outcomes.

Among the current candidates, clonal diversity and copy-number complexity have the clearest near-term prognostic rationale. By contrast, CIN and whole-genome doubling have stronger evolutionary and retrospective clinical support than prospective clinical validation. Pan-cancer copy-number signature analyses have linked distinct CNA patterns to whole-genome doubling, aneuploidy, CIN-associated processes, and disease-specific survival (95). In lung adenocarcinoma, multi-region TRACERx validation of ORACLE showed that clonal expression patterns capture prognostic information associated with genetic evolutionary features, including CIN, and are associated with survival (96). However, ORACLE is a clonal expression-based biomarker, rather than a genomic instability phenotype. For genomic instability features, assay definitions, analytical thresholds, and incremental prognostic value beyond established clinicopathological factors remain insufficiently validated, and none have demonstrated treatment-predictive utility in resectable NSCLC. Accordingly, their appropriate near-term role is prospective validation within risk-stratification models, rather than routine treatment assignment.

Repair deficiency, replication stress, and ecDNA represent different translational categories. Their principal promise lies in identifying candidate therapeutic dependencies rather than established prognostic markers of recurrence. Repair-deficient tumors may warrant evaluation in DDR-directed trials, whereas tumors with prominent replication stress may be candidates for ATR-, CHK1-, or WEE1-directed strategies (97). ecDNA-positive cancers may represent another setting in which transcription-replication conflict or checkpoint dependencies can be therapeutically explored (98). However, alterations in repair genes, genomic scars, and functional measures of repair deficiency are not interchangeable, and most therapeutic evidence has been derived from preclinical studies, early phase trials, or advanced cancers. Therefore, these features should be regarded as priorities for biomarker development and prospective clinical testing, rather than validated predictive biomarkers for resectable NSCLC.

The immune effects of genomic instability are highly context-dependent. Mutational or structural complexity may increase the pool of potential tumor antigens; however, immune recognition also depends on antigen clonality and presentation, T-cell access and function, and the surrounding tumor microenvironment. CIN-associated inflammatory signaling may enhance immune recognition in some settings, while promoting immune suppression or metastatic behavior in others (99). Therefore, genomic instability features should not currently supersede established clinical, pathological, molecular, or immune factors in the selection of perioperative immunotherapy. Their immediate role is to generate testable hypotheses regarding heterogeneous treatment responses that can be evaluated prospectively.

Tissue profiling, longitudinal MRD assessment, and immune characterization may provide complementary but non-equivalent information. Tissue analysis describes the evolutionary architecture and candidate therapeutic dependencies of the resected tumors. Postoperative ctDNA-defined MRD provides a dynamic, assay-dependent signal of residual disease rather than a direct measurement of genomic instability. Immune biomarkers characterize the tumor-immune context in which residual clones may persist or respond to treatment. Among these layers, ctDNA-defined MRD currently has the most direct prognostic evidence for postoperative recurrence, although it is not itself a genomic instability phenotype and should not be used as a standalone basis for treatment escalation or de-escalation (100).

This translational prioritization must also be interpreted in the context of histological subtype, driver status, and treatment setting. Evidence derived from lung adenocarcinoma or molecularly selected cohorts should not be assumed to apply uniformly to all patients with resectable NSCLC. EGFR-mutated, anaplastic lymphoma kinase-positive, driver-negative non-squamous and squamous tumors may differ in their patterns of genomic instability, immune context, and therapeutic dependencies. Therefore, the proposed framework is neither a composite score nor a treatment algorithm. Its purpose is to identify which clinical claims are supported for each feature, where the evidence remains indirect or immature, and which biomarker-defined questions should be prioritized for prospective evaluations. This evidence-based prioritization is summarized in Table II, and Fig. 2 places the framework within the broader perioperative treatment pathway.

Clinical translation requires standardized, phenotype-specific, and analytically validated assays rather than a generic designation of ‘genomic instability’. Future studies should prospectively define the instability phenotype, assay, analytical thresholds, tissue and blood sampling strategy, and clinically relevant endpoints (101,102). The key questions are whether genomic instability features add prognostic information beyond stage, pathology, driver status, and longitudinal MRD, and whether they identify differential benefit from matched therapies. Prospective biomarker-defined trials will be required to establish clinical utility (103).

Genomic instability provides a useful evolutionary framework for understanding postoperative recurrence in resectable NSCLC, but its individual components have distinct and unequally mature clinical implications. Clonal diversity and copy-number complexity have the strongest near-term prognostic rationale, whereas CIN and whole-genome doubling remain primarily evolutionary and prognostic candidates. Repair defects, replication stress, and ecDNA represent candidate therapeutic vulnerabilities rather than validated predictive biomarkers. Postoperative ctDNA-defined MRD provides more direct prognostic information, but its relationship with specific genomic instability features remains unproven. At present, genomic instability profiling should support biomarker development and prospective clinical trials rather than routine perioperative treatment decisions.