INTRODUCTION
Lung cancer is one of the most common and deadly cancers worldwide, with non-small cell lung cancer (NSCLC) accounting for approximately 85% of all cases, posing a serious threat to human health[1]. Over the past decade, the treatment of NSCLC has entered an era of precision medicine centered on molecularly targeted drugs and immune checkpoint inhibitors. Targeted therapies for driver genes such as epidermal growth factor receptor (EGFR) and anaplastic lymphoma kinase (ALK), as well as programmed cell death protein 1/programmed death-ligand 1 (PD-1/PD-L1) inhibitors, have significantly improved survival outcomes in specific patient populations, marking a profound shift in the treatment paradigm from traditional cytotoxic chemotherapy to molecularly guided, individualized strategies[2,3]. However, acquired resistance has become a common outcome of nearly all targeted and immunotherapies, constituting a core bottleneck limiting long-term patient survival. This dilemma stems from the high complexity of tumor biology: Spatiotemporal heterogeneity within tumor cells and across metastatic sites, adaptive molecular evolution under therapeutic pressure, and the dynamic remodeling of the multi-cellular immune microenvironment collectively form a multidimensional network of resistance. Consequently, systematically deciphering and effectively intervening in these dynamically evolving resistance mechanisms has become a critical scientific challenge requiring urgent breakthroughs in the field of NSCLC research. This article aims to review novel therapeutic modalities, represented by antibody-drug conjugates (ADCs), bispecific antibodies, and targeted protein degradation technologies, and to explore their mechanistic advantages in overcoming refractory targets and resistance; simultaneously outlining the prospects for personalized dynamic diagnosis and treatment based on the integration of multi-omics data with advanced computational models, liquid biopsy-driven adaptive therapy, and “patient digital twins”. This review provides an integrated theoretical framework for understanding the NSCLC drug resistance network and offers forward-looking scientific insights for the development of next-generation therapeutic strategies.
This is a narrative review rather than a systematic review or structured review. As such, this work does not attempt to provide a comprehensive or exhaustive synthesis of all available evidence, nor does it claim to have followed prespecified protocols, systematic screening, or risk-of-bias assessment. A targeted literature search was conducted in PubMed from 2018 to 2026 using the following keywords: Non-small cell lung cancer, drug resistance, targeted therapy, immunotherapy, antibody-drug conjugates, liquid biopsy, and spatial heterogeneity. Inclusion criteria were clinical studies on NSCLC, phase III clinical trials, large-scale real-world studies, and high-quality basic research. Exclusion criteria were non-lung cancers, studies not addressing drug resistance mechanisms, duplicate publications, and low-quality literature. Key studies were selected based on relevance to the thematic focus of each section, novelty of findings, methodological rigor, and clinical impact, with priority given to peer-reviewed publications in high-impact journals and registrational clinical trials. No formal systematic evaluation, meta-analysis, or risk-of-bias assessment was performed.
MOLECULAR MECHANISMS AND SPATIAL HETEROGENEITY: FROM TARGET MUTATIONS TO SYNERGISTIC INHIBITION STRATEGIES
Molecular heterogeneity in tumors represents a core challenge for precision therapy. Distinct molecular characteristics may exist within different regions of the same tumor, or even between individual cells, directly influencing treatment response and ultimate clinical outcomes.
Direct and bypass mechanisms of driver gene mutations
The success of targeted therapy has ushered in a new era in cancer treatment, but the emergence of acquired resistance is almost inevitable. It is important to note that resistance mechanisms differ substantially across molecular subtypes and therapeutic contexts. The following sections distinguish resistance patterns according to specific driver alterations - EGFR mutations, ALK rearrangements, and KRAS mutations - because the biological basis, clinical trajectory, and therapeutic implications of resistance to EGFR-tyrosine kinase inhibitors (TKIs), ALK inhibitors, and KRAS-targeted agents are fundamentally distinct. Furthermore, resistance to immunotherapy, chemotherapy, and ADCs involves separate biological processes that are discussed in their respective sections below. While resistance patterns vary significantly across different driver gene subtypes (EGFR, ALK, KRAS, etc.), they can generally be categorized into two types: Direct resistance caused by further mutations in the target itself, and resistance resulting from the activation of alternative signaling pathways[4]. Direct resistance is caused by secondary or subsequent mutations in the driver gene itself, leading to a reduction or complete loss of the drug’s ability to bind to the target. Taking EGFR-mutated NSCLC as an example, approximately 50% to 60% of patients develop the T790M mutation after receiving first- or second-generation EGFR-TKI therapy. This target mutation reduces the drug’s affinity for the mutated protein, weakens enzyme inhibitory activity, reactivates downstream proliferation signals, and allows tumor cells to regain their growth advantage, resulting in clinical resistance[5]; Although the third-generation EGFR TKI - osimertinib effectively inhibits the T790M mutation, new mutation sites such as C797S and L792F/H can still be detected subsequently, further reducing the drug’s binding affinity. In ALK-rearranged NSCLC, the sequential use of ALK-TKIs is similarly accompanied by dynamic changes in the spectrum of resistance mutations. Following resistance to first-generation TKIs crizotinib, mutations such as L1196M are commonly observed. However, after treatment with second-generation TKIs (such as alectinib), the G1202R mutation has become the most common and challenging form of resistance, exhibiting high resistance to both first- and second-generation TKIs[6].
Bypass activation resistance primarily achieves functional compensation by activating other signaling pathways, thereby maintaining cell proliferation and survival. In EGFR-TKIs resistance in EGFR-mutant NSCLC one of the most common bypass activation mechanisms is MET gene amplification, which activates downstream pathways such as phosphoinositide 3-kinase/protein kinase B, allowing cancer cells to proliferate even when EGFR is effectively inhibited[7]. In patients with ALK-rearranged NSCLC following treatment with second-generation ALK-TKIs, MET amplification also accounts for a significant proportion, with an incidence rate of approximately one-third following treatment with second-generation TKIs. In addition, numerous clinical and translational studies have validated human epidermal growth factor receptor 2 (HER2) amplification, KRAS mutations, B-Raf proto-oncogene, serine/threonine kinase (BRAF) mutations, phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) mutations, and phosphatase and tensin homolog deleted on chromosome 10 (PTEN) loss as common bypass activation mechanisms[8]. Among these, HER2 amplification and MET amplification have well-established clinical evidence in EGFR TKI and ALK TKI resistance, whereas KRAS, B-Raf proto-oncogene, serine/threonine kinase (BRAF), phosphatidylinositol-4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) alterations, and phosphatase and tensin homolog deleted on chromosome 10 (PTEN) loss are supported by robust preclinical and early phase clinical data. These mechanisms may occur individually or concurrently in different patients, further increasing the complexity of drug resistance. These two mechanisms often coexist, forming multi-drug-resistant clones that drive disease progression[9]. It is worth noting that even as research into resistance mechanisms deepens, clinical practice has not become simpler. In actual patients, direct resistance and bypass activation are often not mutually exclusive; rather, they frequently coexist or occur sequentially, jointly driving disease progression. This implies that post-resistance treatment decisions can no longer rely on a linear approach of “identifying a single mechanism - matching a single drug”, but instead require a reinterpretation of the nature of treatment failure from the perspective of coexisting mechanisms and clonal competition.
The paradigm of synergistic inhibition of “undruggable” targets
For a long time, many key oncogenes were considered “undruggable” because their protein structures lacked distinct drug-binding pockets; the most well-known of these is KRAS, one of the most commonly mutated genes in human tumors[10]. Sotorasib and Adagrasib, covalent inhibitors targeting KRAS G12C, received clinical approval starting in 2021 based on robust data from registrational trials. By specifically binding to the cysteine at the KRAS G12C mutation site and locking the protein in an inactivated state, they established the druggability of this target for the first time in clinical settings[11]. However, the objective response rate for monotherapy is only 35%-45%, with a median progression-free survival of 6-8 months, and acquired resistance may develop following treatment. Studies have shown that combined inhibition can delay resistance and enhance efficacy: Src homology 2 domain-containing protein tyrosine phosphatase 2 (SHP2) serves as a key node linking multiple receptor tyrosine kinases (RTKs) to the rat sarcoma viral oncogene homolog/mitogen-activated protein kinase signaling pathway[12]. Mechanistic studies indicate that upon inhibition of KRAS G12C, the SHP2-mediated RTK→Son of Sevenless homolog 1 (SOS1) feedback pathway rapidly reloads GTP, restoring downstream signaling. This provides a preclinical rationale for combining KRAS G12C inhibitors with src homology 2 domain-containing protein tyrosine phosphatase 2 (SHP2) inhibitors (JAB-3312, RLY-1971) to disrupt this adaptive resistance loop[13]. Early-phase clinical studies demonstrate that first-line treatment of NSCLC patients with colestrazib (a KRAS G12C inhibitor) plus SHP2 inhibitors (such as sitneprotafib or JAB-3312) achieves an objective response rate of 64.7%-71% and median progression-free survival of 12.2 months, representing encouraging but still preliminary clinical activity[14]. Additionally, combination with EGFR inhibitors (such as cetuximab) is a reasonable strategy aimed at blocking RTK activation that may occur following KRAS inhibition[15]; studies indicate that KRAS G12C inhibitors can enhance intratumoral CD8+ T-cell infiltration and remodel the immune microenvironment, providing a theoretical basis for combination with immune checkpoint inhibitors[16]. Recent studies have confirmed that redox imbalance plays a key role in treatment resistance in NSCLC, with nicotinamide adenine dinucleotide phosphate oxidase-mediated reactive oxygen species (ROS) production and the nuclear factor erythroid 2-related factor 2 (NRF2)-driven antioxidant response pathway serving as central regulatory mechanisms[17]. Excessive ROS accumulation promotes tumor cell proliferation, invasion, and DNA damage repair, thereby reducing sensitivity to targeted therapies and immunotherapy; conversely, excessive activation of the NRF2 pathway initiates the expression of downstream antioxidant genes, enhancing tumor cell tolerance to drug-induced oxidative stress and ultimately mediating drug resistance phenotypes. This signaling axis not only contributes to targeted drug resistance but also reshapes the tumor immune microenvironment, inhibiting immune cell infiltration and activation, thereby mediating dual resistance to both targeted and immunotherapy. A growing body of evidence suggests that interventions targeting the nicotinamide adenine dinucleotide phosphate oxidase-ROS-NRF2 axis hold promise as a new strategy to reverse resistance in NSCLC.
It is important to contextualize NRF2 activation within the broader adaptive redox-protective network that includes the thioredoxin (Trx) system. While NRF2 drives the transcriptional upregulation of glutathione (GSH)-dependent defenses - such as glutamate-cysteine ligase and GSH S-transferases - the Trx system operates as a parallel and complementary antioxidant axis[18]. The Trx system, comprising Trx, Trx reductase, and peroxiredoxins, directly reduces oxidized proteins and scavenges hydrogen peroxide, thereby maintaining intracellular redox homeostasis independently of GSH[19]. Under drug-induced oxidative stress, concurrent upregulation of both the NRF2-GSH pathway and the Trx system provides tumor cells with redundant protective mechanisms, enabling them to evade ferroptosis - a form of iron-dependent lipid peroxidation cell death that is increasingly recognized as a vulnerability in resistant NSCLC cells[20]. This dual-system antioxidant defense explains why monotherapy targeting either NRF2 or the Trx pathway alone may be insufficient to restore drug sensitivity. Integrating NRF2 inhibition with Trx system blockade represents a rational combinatorial strategy to exhaust cellular antioxidant capacity, resensitize resistant tumors to targeted agents and immunotherapy, and potentially induce ferroptotic cell death[21]. This broader redox-adaptive framework should guide the design of future therapeutic strategies aimed at overcoming redox-mediated drug resistance in NSCLC.
The impact and challenges of spatial heterogeneity
Spatial heterogeneity refers to differences in genomic, transcriptomic, and proteomic profiles among distinct regions within the same tumor; it constitutes a key biological basis for the failure of targeted therapy and the development of acquired resistance. Because mutation burden, copy number alterations, and signaling pathway activation states are distributed non-randomly, a single-site biopsy can only represent a localized molecular profile, resulting in significant sampling bias that may overlook clinically significant resistant clones. Following targeted therapy, drug-resistant clones may emerge in certain tumor regions, while other regions remain sensitive to the drug. If treatment is guided solely by the genotype from the initial biopsy, this may lead to continued growth in resistant regions and disease progression. Therefore, it is necessary to construct a spatial mutation map through multi-region sampling, identify functionally heterogeneous regions using radiomics, and utilize liquid biopsy - which can dynamically track the overall molecular status - to comprehensively monitor the spatiotemporal evolution of drug-resistant clones and provide a reliable basis for precision therapy[22]. From a clinical perspective, spatial heterogeneity has several practical implications for patient management. First, in biopsy interpretation, a negative result for a known resistance mutation from a single site does not exclude its presence elsewhere in the tumor; clinicians should consider repeat biopsy from a different lesion or concurrent liquid biopsy when radiographic progression is discordant with molecular testing results. Second, for treatment selection, the presence of spatially distinct resistant subclones may necessitate combination regimens that can target multiple coexisting mechanisms rather than relying on a single agent matched to one alteration. Third, at disease progression, re-biopsy should ideally sample the progressing lesion rather than the original site, as the dominant clone may have shifted under therapeutic pressure. These practical considerations underscore the need for integrating multi-lesion assessment into routine clinical decision-making.
Although spatial heterogeneity is a key factor underlying drug resistance, the invasive nature, limited accessibility, and high cost of multi-site biopsies make it difficult to implement them widely in routine clinical practice; results obtained from single site biopsy may not be generalizable across different subtypes or lesions, and still result in significant sampling bias. While multi-site biopsies can improve the completeness of molecular profiles, their invasiveness, reproducibility, and sample accessibility limit routine application; liquid biopsy and radiomics can compensate for local sampling bias, yet they still cannot fully replace histological and molecular evidence at the tissue level. Therefore, the core of future research lies not in acquiring more data, but in balancing information completeness with clinical feasibility.
HETEROGENEITY OF THE TUMOR IMMUNE MICROENVIRONMENT: FROM COLD-HOT CLASSIFICATION TO PRECISION INTERVENTION
The heterogeneity of the tumor immune microenvironment is a key determinant of treatment response in cancer[23]. This heterogeneity manifests at multiple levels: From spatial distribution within the tumor, such as the immunosuppressive core and the relatively immunologically active periphery[24], to dynamic changes driven by disease progression and therapeutic interventions, and to significant variations among individual patients. Based on the degree of immune cell infiltration and functional status, NSCLC has traditionally been classified into “cold tumors” and “hot tumors”. This binary classification provides a convenient descriptive framework but oversimplifies the spatiotemporal complexity and functional plasticity of the tumor immune microenvironment. To avoid inconsistencies and provide a more precise alternative, we utilize a three state immune stratification model (hot, immune excluded, cold) supported by single cell and spatial transcriptomic data, which better reflects T cell infiltration distribution, stromal barriers, functional states, and dynamic transitions under treatment. This refined framework serves as the basis for subsequent discussion instead of the over-simplified hot/cold dichotomy. Within this refined immune stratification framework, this heterogeneity directly determines the efficacy of treatment regimens such as immunotherapy[25]. For example, “hot tumors” are typically more sensitive to immune checkpoint inhibitors, whereas “cold tumors” require modification of the microenvironment through combination therapies such as radiotherapy, anti-angiogenic drugs, or immune stimulators to induce a “hot” transformation[26]. Therefore, the key to future precision interventions lies in utilizing multi-omics technologies to decipher individualized, dynamic microenvironmental characteristics and, based on this, formulate comprehensive combination therapy strategies to achieve an effective transition from “cold” to “hot” and improve treatment outcomes[27].
Fine-tuning of the immune microenvironment and precision immunotherapy strategies
The immune phenotype of the tumor microenvironment can be preliminarily classified into three states - hot, immune-excluded, and cold - based on the density of CD8+ T-cell infiltration. Among these, hot tumors with high infiltration and high PD-L1 expression can achieve objective response rates of 30%-45% with PD-1/PD-L1 inhibitors[28]; whereas immune-excluded tumors, despite T-cell accumulation in the tumor periphery[29], are unable to effectively infiltrate the tumor core due to physical barriers such as collagen, resulting in limited efficacy of immunotherapy[30]; as for “cold” tumors, which are sparse in immune cells and lack immunogenicity, their response rates to immune checkpoint inhibitors are typically below 5%. This classification framework provides a preliminary explanation for the phenomenon of varying responses to immune checkpoint inhibitors among different tumors[31].
With the application of high-resolution technologies such as single-cell sequencing and spatial transcriptomics, current research has moved beyond the limitations of classification based solely on the degree of infiltration, shifting toward a systematic analysis of the functional states and three-dimensional spatial configurations of key cellular components within the microenvironment[23,32]. In-depth studies have shown that macrophages associated with NSCLC do not exist in a simple binary opposition of antitumor or pro-tumor states within the microenvironment, but rather constitute a functionally continuous cell population[33]. Studies have shown that macrophages associated with NSCLC typically exhibit a continuum of activation biased toward M2 polarization[34,35]. By highly expressing factors such as interleukin-10, transforming growth factor-β, and C-C chemokine ligand 22, they directly suppress CD8+ T cell function and promote the recruitment of regulatory T cells, thereby maintaining an immunosuppressive state[36]. In solid tumors such as NSCLC, lung cancer-associated fibroblasts secrete chemokines such as transforming growth factor-β and C-X-C motif chemokine ligand 12 and deposit large amounts of type I collagen and fibronectin, jointly constructing a dual barrier of chemical chemotaxis and physical obstruction that significantly restricts T-cell infiltration and activity. In contrast, tertiary lymphoid structures - highly organized aggregates of immune cells that form ectopically within the tumor microenvironment - exhibit structural integrity and functional maturity that are significantly associated with favorable clinical outcomes and improved responses to immunotherapy[37], with their maturity depending on the presence of CD21-positive follicular dendritic cells, PNAd-positive high-endothelial microvessels, and intact B-cell follicles. Mature tertiary lymphoid structures can effectively enhance local antigen presentation efficiency, a finding that establishes them as a biomarker with independent predictive value[38,39]. Based on these in-depth cellular and molecular insights, current immunotherapy strategies are evolving from single-agent immune checkpoint blockade toward combined intervention models targeting multiple cellular components and signaling pathways[40]. By implementing combination interventions targeting key nodes such as colony-stimulating factor 1 receptor or phosphoinositide 3-kinase gamma signaling in tumor-associated macrophages, transforming growth factor-β or C-X-C motif chemokine receptor 4 pathways in NSCLC-associated fibroblasts, and FMS-like tyrosine kinase 3 ligand/Lymphotoxin-β receptor - which promotes the formation of tertiary lymphoid structures - it is expected that the immune-exclusionary or “cold tumor” microenvironment can be transformed into a “hot tumor” phenotype characterized by high immune cell infiltration[41,42]. It should be emphasized that immune microenvironment classification is not a fixed, static label. Whether due to adaptive changes in the tumor cells themselves or treatment-mediated remodeling of cellular components, the same patient may exhibit different immunological characteristics at different time points. Therefore, while the immune subtype derived from a single biopsy has reference value, it is limited by temporal heterogeneity and has limited guidance for long-term efficacy prediction and treatment stratification. This suggests that precision immunotherapy should focus more on dynamic assessment rather than one-time typing.
Interactions between targeted therapy and the immune microenvironment
Targeted therapy and immunotherapy are not isolated; there is a complex interaction between the two. Targeted drugs, represented by TKIs, exhibit a distinct duality and temporal pattern in their remodeling of the immune microenvironment in NSCLC, characterized by “initial activation followed by suppression”. During the initial or sensitive phase of treatment, TKIs exert an immune-activating effect, manifested by upregulating major histocompatibility complex molecule expression on tumor cells, promoting CD8+ T-cell infiltration, and reducing regulatory T cells. This transforms the immunosuppressive “cold” microenvironment into an immune-activated “hot” state, a process that can be viewed as “preheating” the tumor immune microenvironment. However, as treatment progresses and resistance develops, tumor cells adaptively upregulate the expression of immune checkpoint molecules such as PD-L1, while immune-suppressive components in the microenvironment - such as myeloid-derived suppressor cells - increase, causing the microenvironment to revert to an immunosuppressive state[43,44].
The mechanisms underlying these dynamic changes provide a scientific basis for combination therapy involving targeted and immunotherapy, but clinical practice has shown that the direct combination of EGFR-TKIs with immune checkpoint inhibitors is often associated with a higher risk of toxicity. Therefore, the current research focus has shifted toward exploring safer and more effective multimodal combination regimens, such as triple-combination strategies involving “targeted therapy combined with chemotherapy and immunotherapy” or “targeted therapy combined with anti-angiogenic agents and immunotherapy”. The groundbreaking results of the IMPower150 study provide reliable evidence for this shift. The study confirmed that the triple-combination regimen of atezolizumab, bevacizumab, and chemotherapy significantly prolonged overall survival in first-line treatment of non-squamous NSCLC compared to conventional regimens, with sustained benefits observed across different PD-L1 expression subgroups. This marks a new phase in cancer treatment characterized by the synergistic regulation of the tumor microenvironment through multiple pathways[45].
This also suggests that synergies in molecular and immunological mechanisms do not necessarily translate into effective combinations at the clinical level. In theory, the transient window of immune activation induced by targeted therapy provides a foundation for combination immunotherapy; however, in clinical practice, issues such as cumulative toxicity, unclear optimal timing, and patient selection criteria may all undermine the overall benefit of this strategy. Therefore, the focus of future optimization of combination regimens may not lie in simply increasing the number of drugs, but rather in determining the most appropriate timing of intervention, treatment sequence, and target patient populations.
THE TEMPORAL DYNAMICS OF DRUG RESISTANCE: CLONAL EVOLUTION AND THERAPEUTIC RE-NAVIGATION
With the widespread application of molecularly targeted therapies and immunotherapy in NSCLC, treatment resistance has been identified as one of the core factors limiting long-term efficacy. A large body of basic and clinical research consistently indicates that resistance is not triggered instantaneously by a single molecular event, but rather is a dynamic process that develops gradually as tumors undergo multi-stage, multi-level changes under sustained therapeutic selection pressure. This process is biologically underpinned by clonal evolution and exhibits high variability and individual differences over time[46]. A systematic understanding of the temporal dynamics of drug resistance is a key prerequisite for achieving precision therapy and optimizing clinical decision-making[47,48].
Clonal evolution and sequential drug resistance
The theory of clonal evolution provides a unified biological framework for understanding acquired drug resistance. At diagnosis, NSCLC typically consists of multiple cell subpopulations with distinct genetic backgrounds, which differ in terms of genetic mutations, copy number alterations, and epigenetic states[40,49]. In the absence of therapeutic intervention, a relatively stable competitive equilibrium is maintained among these cell subpopulations; however, following the introduction of targeted therapy or immunotherapy, the selective pressure exerted by the drugs significantly alters this equilibrium, allowing cell subpopulations with drug resistance potential to gain a relative survival advantage and gradually expand.
In EGFR-mutant NSCLC, the sequential relationship between clonal evolution and drug resistance has been systematically elucidated[50]. Following treatment with first- and second-generation EGFR-TKIs, the detection rate of the T790M mutation significantly increased in patients with disease progression; this mutation confers a selective advantage to mutant clones in the therapeutic environment by reducing the binding affinity of the drug to the target protein[51]. With the clinical application of the third-generation EGFR-TKI osimertinib, T790M-associated clones have been effectively suppressed[52], but new resistant subclones have gradually emerged, including EGFR C797S mutations, MET amplification, HER2 amplification, and activation of downstream signaling pathways[53,54].
It is worth emphasizing that recent single-cell DNA sequencing studies have revealed the complexity of resistant clones at a higher resolution[55]. In patients resistant to osimertinib, multiple resistance mechanisms are not necessarily distributed across different tumor regions but can coexist within the same cell or the same cell subpopulation; for example, the EGFR C797S mutation and MET amplification can occur simultaneously. This finding directly demonstrates that acquired resistance is not a simple linear evolutionary process, but rather the result of the gradual accumulation of multiple molecular events at the subclonal level, providing a biological basis for explaining the limited efficacy of single-target inhibitors in post-line therapy[56,57].
The treatment course of ALK-rearranged NSCLC also provides a clear example of the sequence of resistance development[58]. Following treatment with the first-generation ALK-TKI crizotinib, mutations such as L1196M gradually accumulate[59,60]; under second-generation ALK-TKI treatment, solvent-front mutations such as G1202R become the primary form of resistance[61]. A consistent correlation exists between the use of different generations of ALK-TKIs and the profile of resistance mutations, indicating that prior treatment history plays a decisive role in shaping the structure of resistant clones.
In addition to genetic alterations, clonal evolution can also manifest through significant phenotypic changes[62]. Some EGFR-mutated lung adenocarcinomas undergo histological transformation into small cell lung cancer following long-term TKI therapy; this process is typically accompanied by loss of function in RB1 and TP53 and exhibits a loss of dependence on the original EGFR signaling pathway[63]. Such transformed drug-resistant clones differ significantly from the original adenocarcinoma in both biological behavior and treatment response, highlighting the clinical significance of the diversity of drug resistance mechanisms. Large-scale prospective studies have further validated the association between clonal evolution and clinical outcomes at the population level. The TRACERx study, through multi-region sampling and longitudinal sequencing, found that evolutionary events such as increased mutation burden in cell subpopulations and whole-genome duplication were significantly associated with the risk of recurrence and poor survival outcomes, thereby establishing the central role of clonal heterogeneity in treatment failure[64].
For clinical readers, it is essential to distinguish between ADCs with established regulatory approval or guideline-supported indications and those still in investigational stages. Trastuzumab deruxtecan is approved for HER2-mutant NSCLC based on phase 3 data. Datopotamab deruxtecan has completed phase 3 evaluation and is approved in multiple jurisdictions. Telisotuzumab vedotin has received accelerated approval for c-MET-overexpressing NSCLC. These agents represent the current standard-of-care options for their respective biomarker-defined populations. In contrast, ADCs targeting B7-H3 (MGC018, DS-7300), AXL, CEACAM5, NaPi2b, tissue factor, PD-L1 (HLX43), and bispecific targets (PTK7/EGFR) remain in early-phase clinical trials or preclinical development; their efficacy and safety profiles in NSCLC are not yet established, and they should be considered investigational therapies available only through clinical trials.
Liquid biopsy and dynamic monitoring
Given that drug resistance is viewed as a dynamic evolutionary process, achieving real-time monitoring of clonal changes has become a critical issue in clinical practice. Traditional tissue biopsies, due to their invasive nature, limited reproducibility, and susceptibility to spatial heterogeneity, struggle to meet the demands of longitudinal follow-up. Liquid biopsy, particularly detection technologies based on circulating tumor DNA (ctDNA), offers a viable solution to these challenges[65,66].
ctDNA originates from the apoptosis or necrosis of tumor cells in primary and metastatic lesions and can, to a certain extent, reflect the molecular characteristics of systemic tumors[67]. During EGFR-TKI therapy, multiple studies have shown that resistance mutations such as T790M or C797S can be detected via ctDNA prior to radiographic progression[68,69]. Similar results have been observed in patients treated with ALK-TKIs, where newly emerging resistance mutations or MET amplification can be identified in advance via liquid biopsy[70].
In addition to qualitative detection, quantitative changes in ctDNA also have significant clinical implications[71]. A dynamic decline or clearance of variant allele frequency is closely associated with treatment response; multiple prospective studies have shown that ctDNA clearance is significantly associated with longer progression-free survival and overall survival[72]. However, this still needs to be verified with a larger sample size. Conversely, persistent ctDNA positivity or a recurrence of elevated levels often indicates the presence of residual disease or drug-resistant clones, providing an early warning signal of disease progression[73]. Although liquid biopsy still has limitations regarding detection sensitivity, the risk of false negatives, and the determination of tumor origin, its value in dynamically monitoring the evolution of drug resistance has been widely recognized and is gradually being integrated into clinical decision-making processes.
Clinical practice of treatment re-navigation
Given that drug resistance is recognized as a dynamic evolutionary process, treatment strategies must shift from static regimen selection to dynamic adjustment. Treatment re-navigation emphasizes optimizing treatment regimens based on the current dominant resistance mechanisms by reassessing molecular characteristics at different stages of disease progression.
In patients with EGFR-TKI resistance accompanied by MET amplification, the treatment strategy of combining EGFR inhibitors with MET inhibitors has demonstrated superior efficacy compared to monotherapy in multiple clinical studies, including higher objective response rates and longer disease control duration[74]. For patients with coexisting multiple resistance mechanisms, single-target inhibition often fails to achieve sustained benefit, and combination or multi-target therapy is considered a more rational choice[75,76]; for patients who have undergone histological transformation or lost their driver pathways, continuing the original targeted therapy typically lacks a biological rationale, and treatment should be promptly adjusted to a chemotherapy-based regimen. This practice further highlights the necessity of dynamic monitoring in guiding clinical decision-making. Overall, viewing drug resistance as a time-dependent dynamic process helps redefine treatment goals for NSCLC. By integrating clonal evolution theory, liquid biopsy technology, and dynamic treatment strategies, treatment re-navigation provides a practical pathway to achieving personalized, continuously optimized precision therapy[77].
NOVEL THERAPEUTIC STRATEGIES: BREAKTHROUGHS BEYOND TRADITIONAL MODELS
In the field of NSCLC treatment, although targeted therapies - represented by kinase inhibitors - and immune checkpoint inhibitors have achieved significant success, the emergence of acquired resistance remains a fundamental barrier to long-term patient benefit. Consequently, the scientific and clinical communities are actively exploring and developing a series of novel therapeutic strategies that transcend the mechanisms of action of traditional inhibitors. By introducing entirely new drug action paradigms and biological intervention logic, these innovative strategies aim to overcome the resistance limitations of existing therapies and provide new strategies to clinical dilemmas where effective treatment options are lacking. Against this backdrop, ADCs, bispecific antibodies, and targeted protein degradation technologies are emerging as promising approaches to improve resistance and explore novel biological mechanisms[78].
The rise of ADCs
Over the past two decades, the treatment landscape for NSCLC has undergone significant changes. Targeted therapies and immunotherapies have become the mainstays of treatment for advanced NSCLC and are increasingly being applied to the treatment of early-stage NSCLC[79]. ADCs, as a highly promising therapeutic modality, have made rapid progress in recent years[80]. ADCs represent a targeted therapeutic strategy that combines the high specificity of monoclonal antibodies with the potent cytotoxicity of small-molecule drugs. By delivering the payload to tumor antigens, they enhance tumor-killing activity and reduce off-target toxicity[81]. Their precise structure consists of three core components: A monoclonal antibody capable of highly specific recognition and binding to antigens on the surface of tumor cells; one or more highly active cytotoxic payloads covalently linked via a chemical linker; and a linker that remains stable in the circulatory system but can be efficiently cleaved within target cells to release the payload. After binding to target cells, the ADC enters the cell via endocytosis. Within organelles such as lysosomes, the linker is cleaved by specific enzymes, precisely releasing the cytotoxic drug and thereby inducing tumor cell apoptosis. It is worth noting that the payloads released by many advanced ADC drugs possess high cell membrane permeability, enabling them to diffuse into neighboring tumor cells that do not express the target antigen and induce their apoptosis. This mechanism, known as the bystander effect, allows ADCs to effectively overcome tumor heterogeneity and achieve broader killing of tumor tissue[82].
In recent years, several ADC drugs have achieved breakthrough progress in the treatment of NSCLC and have been successfully translated into clinical practice. Trastuzumab-deruxtecan is an ADC that targets HER2. In the pivotal DESTINY-Lung02 clinical trial, trastuzumab-deruxtecan demonstrated unprecedented antitumor activity in patients with HER2-mutated advanced NSCLC who had previously received systemic therapy, with a confirmed objective response rate as high as 57.7%. These outstanding results led to regulatory approval, filling a long-standing treatment gap for this specific molecular subtype and establishing a new standard of care[83,84]. This agent has received regulatory approval for HER2-mutant NSCLC based on phase 3 data. Another highly anticipated ADC is datopotamab deruxtecan, which targets tropomyosin-related protein 2. In the TROPION-Lung01 study, datopotamab deruxtecan significantly improved progression-free survival in patients with advanced NSCLC who had previously received treatment, compared to the conventional chemotherapy drug docetaxel. Its clinical value is particularly notable in overcoming resistance to EGFR-TKIs, as it offers a novel mechanism of action entirely independent of the EGFR signaling pathway, providing a much-needed new treatment option for this resistant patient population[85]. Datopotamab deruxtecan has been evaluated in phase 3 trials and is approved in multiple jurisdictions. In addition, telisotuzumab vedotin, an ADC targeting the cell-surface-expressed c-MET protein, is able to directly target the MET signaling pathway, which is one of the primary bypass mechanisms of resistance to EGFR-TKIs[86]. Based on the positive results of the LUMINOSITY study, Teliso-V has received accelerated approval specifically for the treatment of patients with previously treated NSCLC who overexpress the c-MET protein, fully validating the clinical feasibility of the strategy to precisely target specific resistance signaling pathways using ADCs[87]. This agent is approved for patients with c-MET-overexpressing NSCLC who have received prior therapy.
Building on this, ADC development targeting other emerging targets is also actively advancing, further broadening the application prospects of ADCs in the field of NSCLC. B7-H3 is an immune checkpoint molecule overexpressed in NSCLC and associated with poor prognosis and immune evasion[88]. Two B7-H3-targeting ADCs, MGC018 (dual-drug payload) and DS-7300 (topoisomerase I inhibitor), have demonstrated early clinical activity[89,90]. AXL is an RTK associated with resistance to targeted therapy and immune evasion; ADCs targeting AXL are currently in the exploratory research phase[91].
Due to their limited efficacy, CEACAM5 and NaPi2b have not yet made significant progress in the field of NSCLC, while the development of receptor tyrosine kinase-like orphan receptor 2 has shifted toward head and neck cancer. ADCs targeting tissue factor are still undergoing active clinical development and are currently being evaluated in NSCLC. ADCs targeting PD-L1, such as HLX43, are currently in phase II clinical trials. Translational strategies for ADCs targeting bispecific targets such as PTK7/EGFR (e.g., BCG017) are also under investigation. Other emerging translational targets include folate receptor alpha (FRα; MORAb-202, PRO1184, STRO-002), transferrin receptor 1 (CD71; CX-2029), and ALCAM 77. Although ADCs have demonstrated significant efficacy in NSCLC, their clinical application remains subject to significant limitations. ADCs are associated with adverse events of various grades and increased risk, including treatment-related symptoms and toxicities affecting the gastrointestinal, nervous, ocular, and hepatic systems[92].
However, the emergence of drugs with new mechanisms of action does not simply mean that the problem of drug resistance has been resolved; rather, it likely indicates that the patterns of drug resistance themselves are shifting. ADCs, bispecific antibodies, and targeted protein degradation technologies have overcome the limitations of traditional inhibitors, but they also introduce new patient selection thresholds, toxicity profiles, and risks of secondary resistance. As clinically illustrated by Dato DXd, a trophoblast cell surface antigen 2 targeted ADC, even agents with demonstrated durable antitumor activity carry distinct adverse events of special interest, including interstitial lung disease, infusion related reactions, oral mucositis/stomatitis, and ocular surface events. These toxicities are particularly relevant in patients with NSCLC, who carry higher baseline risks for pulmonary complications and require proactive monitoring, standardized prophylaxis, and multidisciplinary management to ensure safety. Furthermore, long term durability remains incompletely characterized, and patient selection relies heavily on target expression status, which lacks universal standardization across clinical laboratories. Collectively, these factors highlight the need for balanced clinical decision making that integrates efficacy, toxicity risk, predictive biomarkers, and real world feasibility[85]. Beyond safety and standardization challenges, clinically meaningful ADC activity is also inherently limited by the development of acquired resistance. For ADCs, major resistance mechanisms include reduced or lost target antigen expression, antigen masking, impaired antibody-antigen binding, and tumor heterogeneity that limits uniform targeting. Additional pathways involve defective internalization and endocytic trafficking, lysosomal dysfunction and elevated pH, decreased payload processing, and enhanced drug efflux mediated by adenosine triphosphate-binding cassette transporters. Resistance can also arise from direct insensitivity to cytotoxic payloads, altered microtubule or topoisomerase I function, dysregulated apoptotic signaling, and remodeling of the tumor microenvironment that blunts therapeutic activity[93]. Therefore, the clinical value of these drugs lies not only in improved early response rates but also in achieving a stable balance between precise patient stratification, long-term benefits, and real-world applicability.
Bispecific antibodies and cell therapies
In the field of immunotherapy, bispecific antibodies and cell therapies are offering innovative solutions to overcome treatment resistance in NSCLC through their unique mechanisms. These two therapeutic classes achieve more precise and potent attacks on tumor cells by directly reprogramming the immune system or modifying it ex vivo, respectively.
Bispecific antibodies are a class of engineered antibody molecules whose core feature is the ability to simultaneously and specifically bind to two different antigen epitopes through a single protein structure. Among these, the most mature strategy involves constructing T-cell engagers - molecules that target the CD3 complex on the surface of T cells at one end and a tumor-associated antigen at the other. This design establishes a physical link between tumor cells and T cells, thereby bypassing the major histocompatibility complex-dependent antigen presentation required for traditional T-cell activation, directly activating cytotoxic T cells and inducing them to kill tumor cells[94]. Taking amivantamab, a bispecific antibody targeting EGFR and MET, as an example, its mechanism of action goes beyond simple signaling pathway blockade. As a clinically validated agent, amivantamab is approved for EGFR exon 20 insertion–mutant NSCLC and has demonstrated efficacy in phase 3 trials for patients with disease progression after osimertinib. By binding to the extracellular domains of EGFR and MET, this drug not only competitively inhibits ligand-mediated downstream signaling but also promotes the endocytosis and lysosomal degradation of both receptors, thereby reducing the burden of oncogenes at the protein level[95]. Furthermore, its intact antibody Fc domain can further recruit and activate innate immune cells, such as natural killer cells, through antibody-dependent cellular cytotoxicity, thereby generating a multi-layered antitumor immune response. Clinical data confirm the efficacy of this strategy: In a refractory patient population harboring EGFR exon 20 insertion mutations - a subgroup traditionally unresponsive to tyrosine kinase inhibitors - amivantamab monotherapy achieved a significant objective response rate. More importantly, the MARIPOSA Phase III clinical trial demonstrated that the combination of amivantamab with the third-generation EGFR TKI lazertinib significantly prolonged progression-free survival and overall survival compared to the current standard therapy, but this still needs to be validated with a larger sample size. Osimertinib monotherapy, in the first-line treatment of patients with EGFR-sensitive mutations, establishes the potential of bispecific antibody combination therapy as a new standard regimen[96]. However, clinical data from the phase III MARIPOSA-2 trial highlight key challenges that temper the broad application of amivantamab-based combinations. First, toxicity profiles are substantial and regimen-dependent: Amivantamab plus chemotherapy is associated with high rates of hematologic adverse events (grade ≥ 3 neutropenia, thrombocytopenia, anemia), infusion-related reactions (56%-58%), and rash, whereas the triplet combination of amivantamab plus lazertinib and chemotherapy further increases the risk of severe myelosuppression (grade ≥ 3 AEs up to 92%), venous thromboembolism (22%), and treatment discontinuations (34%). Prophylactic anticoagulation and supportive care are required to mitigate these risks. Second, long-term durability remains incompletely defined; although median progression-free survival is significantly prolonged, mature overall survival data are still lacking, and most patients eventually develop secondary resistance. Third, patient selection is currently limited to EGFR-mutant NSCLC with disease progression on osimertinib, and validated predictive biomarkers (e.g., ctDNA or tissue-based indicators) to stratify potential responders remain unavailable. Furthermore, the triplet regimen required protocol modification due to overlapping toxicities, underscoring the need for optimized sequencing and dosing strategies to balance efficacy and safety[96].
Acquired resistance also represents a major challenge for T-cell-redirecting bispecific antibodies, which limits their long-term efficacy despite promising initial responses. Resistance mechanisms are multifactorial and primarily include tumor-intrinsic factors, T-cell dysfunction, and an immunosuppressive tumor microenvironment. Common tumor-related mechanisms encompass high tumor burden, upregulation of T-cell inhibitory ligands, and target antigen loss or downregulation, which represents a major mode of antigen escape. Functional impairment of T cells also plays a critical role in both primary and acquired resistance; poor baseline T-cell activity is associated with primary refractoriness, while chronic T-cell stimulation induced by long-term bispecific antibody exposure leads to progressive T-cell exhaustion and further reduces antitumor activity. In addition, the immunosuppressive tumor microenvironment significantly dampens the effector function of bispecific antibodies. Rational strategies to overcome resistance include dual or trispecific antigen targeting to prevent antigen escape, combination interventions to reverse T-cell exhaustion by modulating inhibitory or costimulatory pathways, and cotreatment with immunomodulatory agents such as CD38-targeting antibodies to deplete immunosuppressive cell populations. A deeper understanding of these resistance mechanisms will guide rational combination strategies and further optimize the antitumor activity of bispecific antibodies[97].
Cell therapies, represented by chimeric antigen receptor (CAR) T-cell therapy, represent another paradigm shift. This therapy involves isolating autologous T cells from the patient, genetically engineering them in vitro using viral or non-viral vectors to stably express CARs capable of specifically recognizing tumor antigens, and then reinfusing them into the patient after in vitro expansion. The engineered CAR-T cells can autonomously recognize target antigens and, independent of their natural T-cell receptor signaling pathways, generate potent T-cell activation and proliferation signals, thereby exerting a sustained antitumor effect[98].
In the field of solid tumors such as non-small cell lung cancer, the clinical translation of CAR-T therapy faces a series of unique challenges, including physical and immunosuppressive barriers in the tumor microenvironment[99], the lack of ideal tumor-specific antigens to prevent severe “off-target” toxicity[100], and the insufficient persistence and homing ability of CAR-T cells within the body[101]. Current research efforts are focused on developing next-generation technologies, such as enhancing CAR-T cell infiltration by co-expressing cytokine or chemokine receptors[102], or designing safer, smarter CAR structures that incorporate co-stimulatory signal domains and “switch” control elements[103]. Despite these advances, CAR-T therapies remain in exploratory clinical development and are not yet approved for NSCLC.
A core challenge shared by bispecific antibodies and cell therapies concerns treatment safety and drug resistance. Both can trigger serious adverse reactions such as cytokine release syndrome and immune effector cell-associated neurotoxicity syndrome[104]. Furthermore, tumor cells can evade such highly specific immune attacks by downregulating or losing target antigens or by altering their immunophenotype, leading to secondary resistance. Future developments will focus on optimizing molecular and cellular designs, such as developing bispecific antibodies with lower-affinity CD3-binding domains to reduce systemic toxicity, or constructing dual-target CAR-T cells capable of simultaneously targeting multiple antigens to address tumor heterogeneity[105,106]. Furthermore, the rational combination of these immune redirection strategies with other therapeutic modalities, such as immune checkpoint inhibitors and tumor microenvironment modulators, represents a key area of exploration for overcoming immune suppression and achieving durable remission.
Targeted protein degradation technologies
Targeted protein degradation technologies, particularly proteolytic targeted chimeric technologies, represent a new paradigm in small-molecule drug development that transcends traditional “occupancy-driven” inhibition models[107]. This technology utilizes the cell’s intrinsic ubiquitin-proteasome system; by designing heterobifunctional molecules to bring the target protein into close proximity with an E3 ubiquitin ligase, it induces the ubiquitination of the target protein and subsequent proteasomal degradation. This “event-driven” mechanism of action eliminates the need for high-affinity binding to the target protein’s active site, thereby enabling the targeting of “undruggable” proteins lacking clear druggable pockets[108]. Its catalytic properties allow a single Proteolysis Targeting Chimera (PROTAC) molecule to mediate multiple rounds of degradation, offering high efficiency and long-lasting potential; this technology has rapidly emerged as a highly promising new strategy in the field of cancer therapy[109].
In the treatment of NSCLC, research on the degradation of key driver genes using PROTAC technology has achieved a series of substantial advances. Regarding the EGFR, studies have confirmed that PROTAC molecules can effectively degrade various EGFR mutant types, including those with exon 19 deletion, L858R single mutations, and L858R/T790M double mutations[110]. Some molecules have demonstrated inhibitory activity against the L858R/T790M/C797S triple mutation that emerges following resistance to osimertinib in preclinical models, offering a new approach to addressing resistance to third-generation tyrosine kinase inhibitors[111]. The research also focuses on the pharmacokinetic optimization of PROTAC molecules, with some candidate molecules already demonstrating good oral bioavailability and sustained in vivo degradation efficacy in mouse models. For example, studies have found that specific PROTAC molecules can maintain their degradation effect on mutant EGFR for up to 72 hours after discontinuation of administration[112]. Currently, PROTAC molecules capable of crossing the blood-brain barrier and targeting mutations have entered the preclinical development stage[113].
PROTAC technology also shows application potential in echinoderm microtubule-associated protein-like 4-anaplastic lymphoma kinase fusion-positive NSCLC. Research has successfully developed PROTAC molecules capable of effectively degrading the echinoderm microtubule-associated protein-like 4-anaplastic lymphoma kinase fusion protein, and their in vivo antitumor activity has been validated in mouse xenograft models[114]. To enhance therapeutic precision and minimize potential effects on normal tissues, novel regulatory strategies such as light-induced PROTACs have been designed, in which the degradation function is activated only upon specific exogenous light stimulation[115,116]. However, improving selectivity for ALK proteins and achieving effective degradation of common resistance mutations such as L1196M and G1202R remain important areas for future optimization[2].
Regarding KRAS, a target long considered “undruggable”, PROTAC technology has also achieved significant breakthroughs. Research has successfully developed PROTAC molecules that can specifically degrade the KRAS G12C mutant protein, and their degradation capacity has been validated in multiple lung cancer cell lines carrying this mutation. This provides proof of concept and new tools for treating related lung cancers by directly eliminating the mutant KRAS protein[117,118].
In addition to PROTACs that primarily rely on the proteasome pathway, emerging technologies such as lysosome-targeting chimeric proteins have further expanded the scope of protein degradation. By directing cell surface or extracellular proteins to the lysosomal degradation pathway, LYTAC molecules have opened new avenues for degrading membrane proteins and secreted proteins - target categories traditionally difficult to reach for small molecules and PROTACs[119,120].
Currently, more than 40 PROTAC candidates have entered clinical trials worldwide. Phase II clinical trials of PROTACs targeting the androgen receptor and estrogen receptor have preliminarily demonstrated the safety and efficacy of this technology in humans[121,122]. Although PROTAC molecules generally face challenges such as high molecular weight and the need to optimize oral bioavailability, their unique advantages in targeting undruggable targets and overcoming drug resistance are driving rapid development in this field[123]. PROTAC research targeting key targets such as EGFR, ALK, and KRAS in NSCLC is steadily progressing from basic research to clinical applications and is expected to provide patients with potential new directions in the future[124]. It is critical to emphasize that, while PROTACs targeting the androgen receptor and estrogen receptor have reached phase II trials in other malignancies, all PROTAC candidates directed at EGFR, ALK, or KRAS in NSCLC remain in preclinical or very early-phase clinical stages. No PROTAC has yet received regulatory approval for NSCLC, and these agents should be regarded as investigational therapies with unproven clinical efficacy in this disease setting.
Targeted protein degradation technologies represented by PROTACs offer a distinct mechanism to eliminate pathogenic proteins and overcome limitations of traditional enzyme inhibitors. Nonetheless, the clinical application of PROTACs is severely hindered by suboptimal tumor distribution, low bioavailability, and the rapid emergence of acquired resistance. A well-characterized mechanism of PROTAC resistance involves overexpression of the drug efflux transporter ATP-binding cassette subfamily B member 1 (ABCB1/MDR1), which actively pumps PROTAC molecules out of tumor cells and reduces intracellular drug accumulation. Additional resistance mechanisms include mutations or downregulation of E3 ubiquitin ligase components, structural alterations in the target protein that abrogate binding, and dysfunction of the ubiquitin-proteasome pathway. To address MDR1-mediated PROTAC resistance, novel nanoplatforms have been developed using coassembled albumin-based nanoparticles that encapsulate alkylated PROTAC prodrugs together with mechanistic target of rapamycin kinase inhibitors such as rapamycin. This strategy restores intracellular PROTAC activity through esterase-mediated prodrug activation while suppressing MDR1 expression via rapamycin, thereby reversing efflux-related resistance. Such combinatorial nanodelivery systems have demonstrated potent antitumor activity in PROTAC-resistant tumor models, highlighting the feasibility of overcoming resistance through targeted delivery and simultaneous modulation of transporter function[125,126].
From a mechanistic perspective, ADCs, bispecific antibodies, and targeted protein degradation technologies are not merely parallel categories of new drugs, but rather represent three distinct pathways for addressing drug resistance. ADCs use antibody-mediated delivery to direct cytotoxic payloads to tumor cells and rely on the bystander effect to expand their killing range. They can, to some extent, overcome a tumor’s persistent reliance on a single driving pathway, making them more suitable for resistance states characterized by complex driving mechanisms or increased bypass activation. Bispecific antibodies can simultaneously target different sites or bridge immune cells and tumor cells, making them useful for addressing multi-pathway escape or imbalances in tumor-immune interactions. Targeted protein degradation technologies, by inducing the degradation of target proteins rather than merely inhibiting their activity, open up intervention possibilities for conformational resistance and “undruggable” targets that are difficult for traditional inhibitors to address. The value of these novel therapeutic strategies lies not only in changes to drug delivery formats but also in a shift in therapeutic philosophy - moving away from competitive inhibition centered on a single target toward multi-level regulation of the tumor ecosystem, protein homeostasis, and immune networks.
FUTURE DIRECTION: INTEGRATING MULTI-OMICS AND COMPUTATIONAL MODELS TO ACHIEVE PRECISION DIAGNOSIS AND TREATMENT
While the integration of multi-omics and artificial intelligence (AI) holds substantial promise for advancing precision NSCLC care, it is essential to acknowledge that these tools are not yet ready for routine clinical deployment. Before multi-omics and AI-driven models can guide standard NSCLC treatment decisions, they must undergo rigorous prospective validation, establish standardized data pipelines, demonstrate external reproducibility across diverse populations and institutions, achieve regulatory approval, provide clinically interpretable outputs, and integrate seamlessly into existing clinical workflows. The following sections outline the conceptual potential of these approaches while recognizing that most current evidence derives from retrospective studies and preclinical models. As research into tumor heterogeneity and the complexity of the microenvironment deepens, precision medicine is no longer confined to traditional genomic mutation screening but has shifted toward constructing high-resolution panoramic maps of the tumor ecosystem[127].
Recent studies have shown that by integrating multi-omics data - including genomic, transcriptomic, and proteomic data - and combining them with single-cell resolution and spatial positioning information, it is possible to analyze tumor development and resistance mechanisms with unprecedented depth[128]. Particularly in solid tumors such as NSCLC, multi-omics strategies have become a core driver for revealing tumor microenvironment remodeling, identifying rare cell subpopulations, and discovering novel therapeutic targets, providing a scientific basis for overcoming current bottlenecks in targeted and immunotherapy[129].
However, there are significant differences in the multi-omic characteristics among different driver subtypes and immune phenotypes, necessitating a stratified analysis. Breakthroughs in single-cell sequencing and spatial transcriptomics have provided unprecedented resolution for elucidating the multidimensional mechanisms underlying treatment resistance in NSCLC (Figure 1). By integrating multimodal data from single-cell RNA sequencing, spatial proteomics, and ctDNA liquid biopsy, researchers have been able to map the dynamic evolution of the tumor microenvironment at the single-cell level, revealing the spatiotemporal heterogeneity of resistant clones[23].
Figure 1 Multidimensional mechanisms of therapy resistance in non-small cell lung cancer and emerging strategies for overcoming resistance.
NSCLC: Non-small cell lung cancer; EGFR-TKI: Epidermal growth factor receptor tyrosine kinase inhibitor; pAKT: Phosphorylated protein kinase B; pERK: Phosphorylated extracellular signal-regulated kinase; MET: Mesenchymal-epithelial transition factor; TME: Tumor microenvironment; PD-L1: Programmed death-ligand 1; TGF-β: Transforming growth factor-beta; IL-10: Interleukin-10; TAM: Tumor-associated macrophage; ctDNA: Circulating tumor DNA; ADC: Antibody-drug conjugate; BsAb: Bispecific antibody; PROTAC: Proteolysis-targeting chimera; TPD: Targeted protein degradation. Created in BioRender. Available from: https://BioRender.com/76ufua3 (see Supplementary material).
Analysis of NSCLC samples before and after neoadjuvant immunochemotherapy revealed that treatment-induced tumor microenvironment remodeling involves the reconfiguration of epithelial-immune cell interaction networks, with the spatial distribution of proliferating tumor cells relative to M1/M2 macrophages effectively predicting immunotherapy response[130]. Meanwhile, a multi-omics integration study of EGFR-TKIs resistance models revealed that lysophosphatidic acid-mediated signaling and metabolic reprogramming are key nodes in third-generation inhibitor resistance, providing new targets for intervention strategies[131].
These studies demonstrate that the systematic integration of multi-omics data not only identifies rare resistant clones that are difficult to capture with traditional bulk sequencing but also elucidates the mechanisms underlying the interaction between intrinsic molecular alterations in tumor cells and external regulation by the microenvironment. By constructing high-resolution spatiotemporal maps of tumors, key cellular interaction axes driving resistance can be precisely localized, thereby expanding therapeutic targets from single oncogenes to complex tumor ecosystems and providing mechanism-driven strategies to overcome resistance[132].
In terms of computational model development, AI and machine learning (ML) algorithms are accelerating the extraction of clinically meaningful predictive features from massive multi-omics datasets. Deep learning frameworks such as single-cell State Transition Across-samples of RNA-seq data have enabled the automatic enrichment and enhancement of phenotypically associated features in single-cell sequencing data, effectively elucidating cell state transitions related to immune suppression and drug resistance[133,134]; meanwhile, support vector machine models built on spatial multi-omics data have demonstrated superior performance in predicting chemotherapy response, providing a new paradigm for the quantitative analysis of histopathological images[135].
More cutting-edge research focuses on the implementation of the “patient digital twin” concept - by integrating individualized genomic, transcriptomic, proteomic, and imaging data to construct computational models capable of dynamically simulating treatment responses, thereby enabling virtual testing and optimization of treatment regimens[136,137].
The development of such models will rely on standardized data collection protocols, the establishment of cross-center validation cohorts, and the introduction of causal inference algorithms to ensure predictive reliability and clinical interpretability. The cornerstone of personalized dynamic diagnosis and treatment lies in the shift from static histopathology to a real-time liquid biopsy monitoring paradigm. Traditional tissue biopsies struggle to capture the spatiotemporal heterogeneity of tumors and treatment-induced clonal evolution, whereas liquid biopsy technologies based on ctDNA, circulating tumor cells, and exosomes provide a non-invasive, continuous monitoring window[138].
Studies have shown that liquid biopsy can not only reveal the mechanisms of resistance to targeted drugs such as osimertinib by detecting genetic alterations like MET amplification or secondary EGFR mutations, but also predict disease progression by monitoring the dynamic levels of metabolites or non-coding RNA in exosomes[139,140]. High-sensitivity surface-enhanced Raman scattering immunoprobes have been developed to track the phenotypic evolution of circulating tumor cells over time, enabling clinicians to identify molecular-level resistance signals before radiological progression occurs and providing a critical time window for preemptive intervention[141,142].
Finally, the deep integration of AI with multi-omics computational models to transform massive amounts of heterogeneous data into actionable, dynamic treatment decisions is the only path to achieving precision medicine. Utilizing deep learning algorithms to extract phenotypic features from single-cell data, or constructing multi-omics-based prognostic risk stratification models, can accurately predict the probability of a patient’s response to immune checkpoint inhibitors or targeted therapies[143].
ML-driven integrative analyses have successfully identified C-X-C motif chemokine receptor 1-positive neutrophil infiltration as a feature associated with resistance to immunotherapy, as well as circulating CD137-positive Treg cell markers for predicting the efficacy of neoadjuvant anti-PD-1 therapy[144]. This computationally driven strategy supports the clinical implementation of adaptive treatment regimens: When monitoring models indicate a risk of resistance or changes in the microenvironment, treatment strategies can be promptly adjusted - for example, by sequentially using novel antibody-drug conjugates to precisely target heterogeneous tumor cells, or by combining cytotoxic T-lymphocyte-associated protein 4 inhibitors with metabolic inhibitors to reverse the immunosuppressive microenvironment[145,146]. This data-driven, mechanism-guided dynamic closed-loop diagnostic and therapeutic approach marks the evolution of cancer treatment from empirical regimen selection toward precise navigation throughout the entire disease course.
Despite the transformative conceptual promise of multi-omics integration and AI-driven computational models in refining precision NSCLC care, these approaches remain highly speculative and face substantial, unresolved limitations and validation challenges that restrict real-world clinical translation.
First, data-related constraints represent a fundamental bottleneck. AI and ML methodologies depend on large volumes of high-quality, standardized, and well-curated multi-omics data. Clinical oncology datasets are often heterogeneous, high-dimensional, incomplete, noisy, and subject to selection bias, all of which degrade model reliability and generalizability. Extensive preprocessing and manual curation are required but are rarely feasible in routine clinical settings. Second, model heterogeneity and inconsistent standardization hinder reproducibility. A wide variety of AI/ML frameworks exist without clear guidance for methodology selection, and many models are validated only in retrospective cohorts. Few studies adhere to established reporting guidelines such as Consolidated Standards of Reporting Trials - Artificial Intelligence, Standard Protocol Items: Recommendations for Interventional Trials - Artificial Intelligence, or Minimum Information about Clinical Artificial Intelligence Modeling, leading to inadequate transparency in training, validation, cohort selection, and analytical design. Regulatory certification for AI-based medical devices also lacks uniform standards, delaying responsible clinical deployment. Third, poor interpretability and limited explainability undermine clinical trust. Most AI models function as “black boxes”, and few studies incorporate explainable AI to clarify biological mechanisms or predictive rationales. This opacity complicates clinical acceptance and impedes mechanistic validation, especially for predictions derived from radiological or histological images. Fourth, generalizability and robustness remain unproven. Most models lack rigorous external validation across diverse populations, institutions, and technical platforms, leaving them vulnerable to algorithmic bias. Real-world performance often differs substantially from retrospective results, and few tools have been tested in prospective clinical trials. Finally, clinical implementation pathways remain undefined. Few studies include prospective, protocol-driven pipelines for embedding AI predictions into routine decision-making. Although prospective trials such as I3 LUNG (NCT05537922) and APOLLO 11 (NCT0555096) aim to address this gap, evidence remains preliminary.
Until these challenges are resolved through standardized data acquisition, rigorous prospective validation, regulatory alignment, and interpretable design, multi-omics and AI-driven strategies will remain limited to research contexts rather than routine clinical tools[147].
CONCLUSION
The field of NSCLC treatment is undergoing a fundamental shift from static classification to dynamic systems monitoring. This review systematically demonstrates that acquired drug resistance is an inevitable outcome of the co-evolution of the tumor genome, clonal populations, and the immune microenvironment under therapeutic pressure. The spatiotemporal heterogeneity and adaptability of this process determine that interventions targeting any single dimension alone are unlikely to achieve sustained efficacy.
At the mechanistic level, single-cell and spatial multi-omics technologies have revealed the highly heterogeneous nature of drug resistance, confirming that the coexistence of multiple clones within the same tumor and immune suppression in the microenvironment are structural causes of treatment failure. In terms of monitoring and early warning, liquid biopsy technologies can now detect drug resistance signals - such as ctDNA and exosomes - at an ultra-early stage, significantly advancing the window for clinical intervention. In terms of therapeutic interventions, ADCs, bispecific antibodies, and targeted protein degradation technologies represent the three major directions for novel mechanisms of action. Through precise delivery, immune redirection, and substrate clearance mechanisms, respectively, they effectively overcome the inherent limitations of traditional therapies and have substantially improved survival benefits in their respective target populations.
In summary, clinical treatment strategies for NSCLC are shifting from a reactive model focused on managing drug resistance to a new proactive management model centered on prospective molecular monitoring and early intervention. By deeply integrating dynamic monitoring technologies, novel mechanism therapies, and AI decision-making systems, clinical practice is expected to achieve proactive intervention in disease progression. This marks the official entry of cancer diagnosis and treatment into a new era grounded in systems biology and centered on adaptive monitoring, with the ultimate goal of transforming NSCLC into a chronic disease that can be managed long-term, thereby improving both patients’ quality of life and survival duration.
