Single-cell and spatial studies (15-19) have shown that the cell TME is highly heterogeneous in NSCLC. Therefore, tumor-associated macrophages (TAMs) should not be interpreted solely within a rigid M1/M2 binary model. Historically, although M1-like macrophages have been associated with pro-inflammatory and antitumor activity and M2-like macrophages have been generally linked to tissue repair, angiogenesis, immune suppression and tumor progression, accumulating evidence (16-20) has indicated the presence of a continuum of macrophage activation states rather than two distinct subtypes. Within the TME, hypoxia, tumor-derived cytokines and stromal-derived signals, activate TAM programs that produce vascular endothelial growth factor (VEGF), matrix metalloproteinases, IL-10 and several chemokines that recruit regulatory T cells (Tregs), while suppressing effector T cell activity (15-20). Collectively, these factors can determine macrophage phenotype.

Consistently, myeloid-derived suppressor cells (MDSCs) warrant subtype-specific interpretation. Monocytic MDSCs predominantly suppress immunity through the expression of arginase 1 (ARG1), inducible nitric oxide synthase (iNOS), IL-10 and TGF-β, whereas polymorphonuclear MDSCs are more closely associated with reactive oxygen species (ROS) production and oxidative stress. These mechanisms deplete L-arginine, impair T cell receptor signaling, inhibit natural killer and CD8⁺ T cell function and promote the expansion of Tregs (21,22). This suppresses antitumor immunity through several mechanisms, including the secretion of IL-10 and TGF-β, CD25 expression-mediated IL-2 consumption and CTLA-4-mediated competition for CD80/CD86 on antigen-presenting cells (23-25). Cancer-associated fibroblasts and ECM remodeling promote immune evasion by increasing stromal stiffness and secreting TGF-β, thus attenuating lymphocyte infiltration and promoting the development of immune-excluded tumors (26-30).

Within the TME, lymphocyte activation gene 3 (LAG-3) serves as an inhibitor of T cell activation and proliferation while enhancing the immunosuppressive activity of Tregs through high-affinity binding to major histocompatibility complex (MHC) class II molecules (31). T cell immunoglobulin and mucin-domain-containing protein 3 (TIM-3) is another inhibitory immune checkpoint receptor that restrains T-cell activation and effector function and is predominantly expressed on T helper (Th)1 and type 1 cytotoxic (Tc1) cells. After binding ligands such as galectin-9, TIM-3 suppresses Th1/Tc1-mediated responses and promotes T cell exhaustion (32). T cell immunoreceptor with Ig and ITIM domains (TIGIT), which is predominantly expressed on the surface of CD8⁺ T cells, mediates immunosuppression in the TME by binding to receptors on antigen-presenting cells, including CD155(33). TIGIT is also present on Tregs, where it enhances suppressive capacity and inhibits the activation, proliferation and effector activity of immune cells (34).

The TME is typically characterized by hypoxia, which activates hypoxia-inducible factor-1α, induces VEGF expression and directly or indirectly upregulates PD-L1(35). Hypoxia also serves a key role in promoting lactate accumulation, nutrient competition and acidic stress through distinct but interconnected mechanisms. Lactic acid preferentially supports the metabolism of Tregs (36). By contrast, extracellular acidosis and glucose deprivation impair CD8⁺ effector T cell function by inhibiting cytokine production, proliferation and cytotoxic activity (37). In parallel, hypoxia, impaired perfusion, high interstitial pressure and metabolic restriction attenuate immune infiltration and activity, eventually resulting in a non-inflamed or dysfunctional immune phenotype.

The TME represents a dynamic and highly heterogeneous ecosystem in which cell factors, molecular pathways and physicochemical characteristics can collectively determine the efficacy of immunotherapy. Consequently, targeting key regulatory mechanisms within the TME and reshaping the immune milieu through focused intervention represent promising strategies for improving clinical responses to immunotherapy in NSCLC.

Tumor-intrinsic resistance is primarily characterized by the loss or impairment of antigen presentation. Mutations or deletions in β2 microglobulin disrupt the assembly and transport of MHC class I, while human leukocyte antigen class I downregulation decreases recognition by CD8⁺ T cells (38,39). Alterations in serine/threonine kinase 11 (STK11), also known as liver kinase B1 (LKB1) and KEAP1 define a clinically important immune-resistant state. Loss of STK11 suppresses stimulator of interferon genes-dependent type I interferon signaling, thereby attenuating innate immune activation and dendritic cell priming. These alterations are commonly associated with low PD-L1 expression and decreased CD8⁺ T cell infiltration. Concurrent loss of KEAP1 results in activation of nuclear factor erythroid 2-related factor 2-dependent antioxidant and metabolic programs, promoting immune evasion and resistance to ICIs (40,41). EGFR- and ALK-driven NSCLC is commonly characterized by lower tumor mutational burden (TMB), decreased neoantigenicity and a less inflamed immune contexture (the composition, density, functional state and spatial distribution of immune cells within the tumor) (6). In EGFR-mutant NSCLC, resistance to EGFR-TKIs promotes immune escape through increased PD-L1 expression (7), whereas Ras homolog family member B (RHOB)-dependent AKT signaling represents a distinct tumor-intrinsic mechanism of EGFR-TKI resistance (42).

The marked expansion of immunosuppressive cell populations within the TME, including Tregs, MDSCs and TAMs, substantially contributes to immune evasion and tumor progression through several mechanisms. For example, Tregs display greater infiltration, abundance and suppressive capacity in tumor tissue than in peripheral blood and non-tumor tissue, thus impairing effector T cell function by secreting inhibitory cytokines such as IL-10 and TGF-β and by expressing CTLA-4, which competes for co-stimulatory signals for T cell activation (43). MDSCs accumulate in the peripheral blood and tumor tissue of patients with NSCLC and suppress T-cell proliferation and function through ARG1, iNOS, ROS and TGF-β (21). TAMs, including M2-like macrophages, promote angiogenesis and recruit Tregs by secreting VEGF, IL-10 and CCL17/22, while also driving T cell exhaustion through the production of tryptophan metabolites (17).

Systemic factors affect the response to ICIs but require disease-specific interpretation. In NSCLC, exposure to antibiotics and specific features of the gut microbiome are associated with ICI efficacy, while taxa such as Akkermansia muciniphila, Bifidobacterium spp. and Ruminococcus have been involved in promoting a favorable immune milieu in certain NSCLC cohorts (44-48). Associations involving Fusobacterium are stronger in colorectal cancer compared with NSCLC and should be considered extrapolative unless supported by NSCLC-specific datasets. Host nutrition, sarcopenia, hypoalbuminemia, elevated C-reactive protein levels and a high neutrophil-to-lymphocyte ratio reflect systemic inflammation and impaired immune competence (49).

Collectively, the TME-mediated resistance to ICIs in patients with NSCLC arises from the interplay between tumor genotype, antigen presentation, the local immunosuppressive cell ecosystem, stromal architecture, metabolic stress and the host systemic state. Fig. 1 summarizes the aforementioned integrated framework, including immune-desert and -excluded and exhausted inflamed phenotypes, and links each resistance domain to its corresponding therapeutic rationale.

Combination therapies should be evaluated according to the evidence level, patient selection and toxicity profile rather than biological plausibility alone. Dual immune checkpoint blockade with nivolumab + ipilimumab has shown notable efficacy in phase III clinical trials in selected patients with advanced NSCLC, including the CheckMate 227 trial. However, the interpretation of these results depends on PD-L1 expression, TMB-defined analyses, study endpoints and the comparator arm (50-52). By contrast, emerging immune checkpoint combinations require caution (53). Although the CITYSCAPE trial suggested clinical efficacy of tiragolumab + atezolizumab in patients with PD-L1-positive NSCLC, the subsequent phase III SKYSCRAPER-01 trial failed to meet the overall survival endpoint (54,55). Similarly, although LAG-3 and TIM-3 remain biologically promising, their clinical adoption in NSCLC requires confirmation through mature clinical trials (56-59).

Anti-angiogenic therapy has a strong mechanistic rationale because VEGF promotes abnormal vasculature, hypoxia, Treg/MDSC accumulation and impaired lymphocyte trafficking. The IMpower150 trial supported the use of atezolizumab + bevacizumab, carboplatin and paclitaxel in patients with metastatic non-squamous NSCLC and is clinically notable for demonstrating efficacy in subgroups such as patients with baseline liver metastases and those with EGFR-mutant disease after TKI failure (60-66). This regimen is therefore more clinically actionable compared with investigational combinations such as ICIs with anti-angiogenic TKIs, TGF-β-targeted agents or myeloid-cell-targeted agents, although its use is affected by regional approvals, histology, prior therapy and the risk of bleeding or vascular complications.

The combination of ICIs with chemotherapy or radiotherapy is currently considered a standard treatment approach in unresectable stage III NSCLC after concurrent chemoradiotherapy and metastatic driver-negative NSCLC treated with chemo-immunotherapy. Chemotherapy induces immunogenic cell death, decreases suppressive myeloid cell populations and increases antigen release, whereas radiotherapy predominantly serves as an in situ vaccine by releasing tumor antigens, increasing antigen presentation and promoting interferon-mediated immune activation (67-70). The PACIFIC regimen (durvalumab after concurrent chemoradiotherapy [CRT]) is an established treatment strategy following concurrent chemoradiotherapy for patients with unresectable stage III NSCLC and chemo-immunotherapy regimens such as KEYNOTE-189 represent the standard first-line option for patients with metastatic non-squamous NSCLC without actionable drivers (67-70). However, these clinical benefits should be balanced against treatment-related adverse events, including pneumonitis and bone marrow toxicity; baseline comorbidities should be assessed separately because they may increase treatment risk.

In EGFR- or ALK-driven NSCLC, ICI monotherapy typically shows limited efficacy, and combinations with TKIs can result in notable toxicity, including pneumonitis and hepatotoxicity (71,72). The KEYNOTE-789 trial did not demonstrate a clear clinical advantage of pembrolizumab + chemotherapy after EGFR-TKI failure (73). In the post-osimertinib setting, CHRYSALIS-2 evaluated amivantamab + lazertinib without the addition of an ICI. Therefore, this regimen should be interpreted as a targeted therapy regimen rather than an ICI-based combination therapy (74,75). In patients with oncogene-driven NSCLC, clinical decision-making should prioritize targeted therapy, chemotherapy-based treatment options and carefully selected anti-VEGF/ICI/chemotherapy combinations rather than the indiscriminate use of TKI-ICI combinations.

Microbiome- and metabolism-associated interventions remain promising but are largely investigational. Fecal microbiota transplantation, probiotics, dietary fiber manipulation, indoleamine 2,3-dioxygenase (IDO) inhibition, lactate metabolism targeting and related approaches have a strong mechanistic appeal but require prospective clinical validation, standardized assessment approaches and safety monitoring before their clinical implementation in patients with NSCLC (46-48,76-78).

The main therapeutic categories of combination treatment strategies, their biological rationale, level of clinical evidence and their key limitations are summarized in Table I, and an evidence-stratified overview is shown in Fig. 2.