Guiding the Application of Immunotherapy in Nonsmall Cell Lung Cancer: The Role of Biomarkers

ABSTRACT

Although immunotherapy has completely transformed the treatment landscape of nonsmall cell lung cancer (NSCLC), its wide application is still limited by the heterogeneity of patient responses, primary and acquired resistance, as well as the management challenges of immune‐related adverse events (irAEs). To achieve precise individualized immunotherapy, this review systematically summarizes the multidimensional biomarker profiles that can predict the efficacy, resistance, and safety of immunotherapy. Beyond the single PD‐L1 expression, the efficacy prediction system has expanded to integrate multimodal information such as dynamic tumor microenvironment, tumor genomic characteristics, systemic inflammation/immune status, and radiomics. At the same time, studies have revealed that immune‐suppressive microenvironments and specific genetic variations are key mechanisms mediating treatment resistance. Moreover, clinical characteristics, blood markers, and imaging findings provide a basis for predicting the risk of irAEs. Looking to the future, the core to overcoming current bottlenecks lies in constructing dynamic and integrated prediction models. By integrating multiomics data, longitudinal liquid biopsies, and artificial intelligence algorithms, an intelligent decision‐making system is expected to be developed to achieve real‐time monitoring of treatment response, early identification of resistance mechanisms, and proactive management of toxicity risks, ultimately optimizing the clinical practice of NSCLC immunotherapy and advancing it towards higher‐order precision medicine.

The content is divided into three main parts: (1) Predictive biomarkers: this includes six detailed subcategories. (2) Prediction of immune resistance: Lists relevant cell and gene markers. (3) Prediction of immune‐related adverse reactions.

1. Introduction

Lung cancer represents a significant global public health challenge, as documented by the World Health Organization (WHO), due to its high mortality rate. Approximately 15% of lung cancer cases are classified as small cell lung cancer (SCLC), whereas the remaining 85% are categorized as nonsmall cell lung cancer (NSCLC) [1]. NSCLC encompasses various histological subtypes, including squamous cell carcinoma, adenocarcinoma, and sarcomatoid tumors, among others [2]. Therapeutic approaches for NSCLC comprise surgical intervention, chemotherapy, radiotherapy, targeted therapy, and immunotherapy [3]. Recent advancements in targeted and immunotherapeutic modalities have fostered the development of more precise treatment strategies [4]. Within the context of immunotherapy for NSCLC, the identification and screening of biomarkers to select patients most likely to benefit from such treatments may enhance therapeutic efficacy and mitigate drug resistance [5, 6]. The construction of a multidimensional and dynamic model based on biomarker detection could facilitate the stratification of patients into distinct immune response subgroups, optimize treatment selection, and enable real‐time monitoring of biomarkers during therapy to promptly address emerging drug resistance [7]. This approach supports the realization of individualized and precision immunotherapy. Furthermore, integrating individual biomarker data with artificial intelligence and machine learning algorithms to develop comprehensive predictive models holds promise for more accurately forecasting immunotherapy outcomes.

Based on the KEYNOTE‐010 [8] and CheckMate 057 [9]/CheckMate 017 [10] studies, pembrolizumab and nivolumab have both demonstrated therapeutic effects on NSCLC that has previously undergone chemotherapy, ushering in the era of immunotherapy. The KEYNOTE‐024 study compared pembrolizumab with chemotherapy as the first‐line treatment for advanced NSCLC with high PD‐L1 expression (TPS ≥ 50%), suggesting precise first‐line immunotherapy for patients with high PD‐L1 expression [11]. For metastatic NSCLC without EGFR or ALK mutations (KEYNOTE‐189 (nonsquamous NSCLC) and KEYNOTE‐407 (squamous NSCLC)) [12, 13], the efficacy of pembrolizumab combined with chemotherapy was superior to chemotherapy alone. For Stage IV or recurrent NSCLC with high tumor mutational burden (TMB) (≥ 10 mutations per megabase), the efficacy of nivolumab plus ipilimumab in combination with chemotherapy was significantly longer than chemotherapy alone, and the proportion of patients with pathological complete response was higher than that of simple chemotherapy [14]. With the progress of research, attention is gradually shifting to operable patients. IMpower010 showed that in resected II–IIIA stage NSCLC, adjuvant chemotherapy after atezolizumab was superior to best supportive care, and the benefit was significant in the subgroup of tumors expressing PD‐L1 [15]. In the CheckMate 816 study, neoadjuvant nivolumab plus chemotherapy intervention for resectable IB–IIIA stage NSCLC significantly prolonged event‐free survival, and the proportion of patients with pathological complete response was higher than that of simple chemotherapy [16]. Ivonescimab (a dual‐specificity antibody targeting PD‐1 and vascular endothelial growth factor) has good efficacy in patients with advanced or metastatic NSCLC [17]. In addition, a trial of pembrolizumab combined with Datopotamab deruxtecan for NSCLC is currently underway [18]. FDA‐approved agents such as PD‐1 inhibitors (Nivolumab, Pembrolizumab, Cemiplimab), PD‐L1 inhibitors (Atezolizumab, Durvalumab, Avelumab), CTLA‐4 inhibitors (Ipilimumab, Tremelimumab), and the lymphocyte activation gene 3 (LAG‐3) inhibitor Relatlimab have been developed to target these pathways. Ongoing research is also exploring novel immune checkpoints, including T cell immunoglobulin and mucin domain 3 (TIM‐3), V‐domain Ig suppressor of T cell activation (VISTA), B7‐H3, B and T lymphocyte attenuator (BTLA), and T cell immunoglobulin and ITIM domain (TIGIT) [19]. The background is shown in Figure 1. However, for the stratification of the advantaged population and the inevitable treatment resistance, the application of biomarkers can achieve more precise patient stratification and overcome treatment resistance.

FIGURE 1.

Among lung cancer, approximately 15% are classified as SCLC, while the remaining 85% are categorized as NSCLC. Immune checkpoints PD‐1/PD‐L1 and CTLA‐4 mediate interactions between tumor cells and immune cells, thereby influencing immune regulatory responses. CTLA‐4, cytotoxic T‐lymphocyte‐associated protein 4; NSCLC, nonsmall cell lung cancer; PD‐1, programmed cell death protein 1; PD‐L1, programmed cell death ligand 1; SCLC, small cell lung cancer.

With the advancements in targeted therapy and immunotherapy, the treatment of NSCLC has entered the era of precision medicine [4]. Immune checkpoint inhibitors, such as PD‐1/PD‐L1 inhibitors, have expanded the therapeutic benefits from advanced second‐line treatment to first‐line and even adjuvant and neoadjuvant treatments for patients who are eligible for surgery, significantly changing clinical practice. However, immunotherapy faces two core bottlenecks: one is the need for more precise biomarkers to select the populations with superior benefits. The other is the inevitable occurrence of primary or acquired resistance. Therefore, constructing multidimensional and dynamic biomarker models and integrating artificial intelligence for prediction has become a key direction for achieving individualized precise immunotherapy, optimizing efficacy and overcoming resistance.

2. Predictive Biomarkers for Immunotherapy in NSCLC

The efficacy prediction system has now been expanded to integrate multimodal information such as dynamic tumor microenvironment (TME), tumor genomic characteristics, systemic inflammation/immune status, and radiomics. The predictors used in the clinical trial formulation mode shown in Table 1 and the predictive factors tested using clinical samples shown in Table 2.

TABLE 1.

The predictors used in the clinical trial formulation mode.

Participants	Clinical stage	Histology	Treatment stage	Intervention	Results	References
258 NSCLC	Stage (I–II; III–IV)	Adenocarcinoma; squamous		Nivolumab; pembrolizumab; atezolizumab	Baseline T cell density and exhaustion markers stratify outcomes in PD‐L1 positive NSCLC on ICI	[20]
30 NSCLC	Stage (IIIA; IIIB)	Adenocarcinoma; squamous cell	Neoadjuvant	Sintilimab combined with nab‐paclitaxel and carboplatin	Pretreatment CD8 overexpression predicts DFS	[21]
184 NSCLC	Stage (III; IV)	Adenocarcinoma; large‐cell carcinoma; squamous cell carcinoma; adenosquamous carcinoma; NSCLC–NOS	Line (1; 2; ≥ 3)	Chemotherapy and PD‐1 axis blockade; dual PD‐1 and CTLA4 blockade; other PD‐1 axis‐based combinations; single‐agent PD‐1 axis blockade	CD44 is a novel biomarker for PD‐1 axis sensitivity	[22]
57 NSCLC	Stage (I; II; III; IV)			Pembrolizumab; nivolumab; atezolizumab	CD39+CD8+ T cell ratio predicts response to PD‐1/PD‐L1 blockade	[23]
19 NSCLC	Stage (IIIB/C; IV)	LUAD; LUSC	Line (1; 2)	PD‐1 inhibitor monotherapy	CD57+CD8+ T cells predict immunotherapy success in NSCLC	[24]
101 NSCLC	Stage (II; III; IV)	Adenocarcinoma; squamous cell carcinoma; others	Line (1)	Pembrolizumab monotherapy or pembrolizumab in combination with chemotherapy	Pretreatment circulating TCR predicts outcomes in advanced NSCLC	[25]
38 NSCLC	Stage (III; IV)	Adenocarcinoma; squamous; unclassified	Line (1; ≥ 2)	Anti‐PD‐1; anti‐PD‐L1	TMR score predicts immunotherapy response in advanced NSCLC	[26]
252 NSCLC		SqNSCLC		Nivolumab or nivolumab plus ipilimumab	Higher immunogene scores or mIF immune infiltration predict better ICI response and survival	[27]
265 NSCLC				Nivolumab‐based; pembrolizumab‐based; others	Immunoscore‐IC predicts ICI efficacy in NSCLC	[28]
35 NSCLC	Stage (III; IV)	Squamous cell carcinoma; adenocarcinoma		ICB alone or in combination with chemotherapy (durvalumab; nivolumab; pembrolizumab)	MAIT cell reactivation contributes to tumor regression on ICI therapy	[29]
30 NSCLC	Stage (IIIB–IIIC; IV)	Adenocarcinoma; squamous cell carcinoma; underdifferentiation		Pembrolizumab; sintilimab; atezolizumab; tislelizumab; toripalimab	CD8+MAIT cells predict PD‐1 therapy response in NSCLC	[30]
61 NSCLC		Adenocarcinoma; squamous cell carcinoma; adenosquamous carcinoma; large‐cell neuroendocrine carcinoma		ICI and platinum‐doublet; ICI monotherapy; ICI (nivolumab; pembrolizumab; atezolizumab)	The ERBB2 mutation is a predictor for the efficacy of ICI intervention	[31]
43 NSCLC	Stage (III; IV)	Adenocarcinoma; squamous cell carcinoma; other	Line (1; 2; 3; 4)	Atezolizumab; durvalumab; nivolumab; pembrolizumab	CXCL9, CXCL10, IL‐15, CASP8, and ADA are associated with PFS	[32]
33 NSCLC	Stage (IIIB; IV)	Non‐SCC; SCC	Line (1; 2; 3)	Pembrolizumab; nivolumab	Baseline EV TGF‐β high expression predicts ICI nonresponse and shorter PFS/OS	[33]
88 NSCLC	Advanced NSCLC	Adenocarcinoma; squamous cell carcinoma	Line (1; 2; 3)	Nivolumab; pembrolizumab	EV‐miR‐625‐5p predicts ICI response and survival in NSCLC, especially in PD‐L1 ≥ 50%	[34]
78 NSCLC	Stage (IIA; IIB; IIIA; IIIB)	Adenocarcinoma; squamous cell carcinoma	Neoadjuvant	PD1 inhibitors	The exLR features accurately predict neoadjuvant ICI response in NSCLC	[35]

TABLE 2.

The predictive factors tested using clinical samples.

Samples	Factor	Result	References
Samples of peripheral blood and tissue	T‐cell receptor	Pretreatment tissue TCR evenness predicts CPR to neoadjuvant chemoimmunotherapy	[36]
Tumor tissue and paired peripheral blood specimens	TIR index	TIR index evaluates tumor‐specific TCR and anti‐PD‐L1 therapy response	[37]
Surgical specimens	irHPC	irHPC score predicts pathological response to neoadjuvant immunotherapy	[38]
NSCLC tissues	MAIT cells	Circulating MAIT subsets predict anti‐PD‐1 efficacy	[39]
NSCLC tissues	TILs	TIL density predicts favorable postimmunotherapy prognosis	[40]
Tissue specimens	ALKBH5	ALKBH5 status predicts anti‐PD‐L1 clinical response	[41]
FFPE samples	E3 ligase Trim35	Trim35 predicts immunotherapy outcome	[42]
Blood samples	CRP	Early CRP flare predicts ICI response	[43]
ctDNA samples	ctDNA	ctDNA parameters predict long‐term benefit from chemoimmunotherapy	[44]
ctDNA samples	ctDNA‐defined clonality	ctDNA‐defined clonality predicts immunotherapy benefit in LA‐NSCLC	[45]
Blood samples	Metabolite butyrate of the gut microbiota	Gut microbiota‐derived butyrate is a promising biomarker for antitumor immunity	[46]
Fecal samples	Gut microbiome	Gut macrogenomic features predict ICI efficacy and PD‐L1 expression	[47]

2.1. TME and Immune Contexture

2.1.1. PD‐L1 Expression and Dynamic Assessment

The PD‐1/PD‐L1 pathway plays a crucial role in tumor immune escape [48]. The PD‐L1 dose (CN) shows high stability in various tumor regions, and low CN and absence may identify patients unlikely to respond to ICIs treatment, and can be used as an auxiliary biomarker [49]. Additionally, in metastatic cases, biopsy of the metastatic site may improve the detection of high PD‐L1 expression, reflecting the heterogeneity of expression, and recommending performing a biopsy on the targeted area [50]. The functional assessment of the PD‐1/PD‐L1 interaction has been used as a predictive biomarker for patient stratification, significantly improving the response rate of immunotherapy. Based on the PD‐1/PD‐L1 interaction, it is possible to predict patients with a lower PD‐L1 score who will benefit from immunotherapy [51]. The co‐localization score is used to measure the binding of PD‐1 expression to PD‐L1 expression, and it is related to the OS after immunotherapy [52]. The soluble PD‐1 protein (sPD‐1) may block the binding of PD‐L1 to PD‐1. The combined use of sPD‐L1 and sCTLA‐4 has been proven to effectively distinguish the response to PD‐1/PD‐L1 blockade therapy, supporting application in clinical patient stratification [53].

2.1.2. Immune Cell and Immune Cell Subsets

T cells initiate, infiltrate, recognize, and kill tumor cells. Baseline T cell density and the expression of T cell exhaustion markers can stratify the treatment outcomes of patients and can be used to determine the best treatment plan [20]. In patients with operable IIIA/IIIB stage NSCLC undergoing neoadjuvant immunotherapy combined with chemotherapy, the pretreatment CD8⁺ T cell level has been identified as a predictor of disease‐free survival (DFS). The neoadjuvant chemotherapy immunotherapy for patients with potentially resectable Stage IIIA/IIIB NSCLC has expanded the possibility of achieving R0 resection in initially unresectable cases [21]. CD44 can serve as a sensitivity marker; it is expected to complement the existing biomarkers and achieve the optimal stratification of patients [22].

The ratio of CD39⁺ CD8⁺ T cells has been determined as a predictive indicator; the predictive ability is independent of other confounding factors [23]. Additionally, CD57⁺ CD8⁺ T cells are associated with successful immunotherapy outcomes; measurements based on blood and tissues may become promising biomarkers [24].

2.1.3. T‐Cell Receptor (TCR)

The TCR is the decisive molecule for the identity and function of T cells. The parameters of TCR are associated with the complete pathological response (CPR) of patients receiving neoadjuvant chemotherapy immunotherapy. The parameter prediction effect based on TCR combination analysis is superior to that of PD‐L1 TPS and TMB [36]. Patients treated with PD‐1 or PD‐L1 inhibitors can be stratified based on the TCR immunotherapy response index (TIR index). Weighting of tumor‐specific TCR combinations can serve as a clinically feasible classification method [37]. The pretreatment circulating TCR level has been identified as a predictive indicator for pembrolizumab or combined chemotherapy [25]. Integrating TMB and TCR into a comprehensive TMB–TCR (TMR) score is associated with the response to immunotherapy and a higher TMR score is associated with a better PFS [26].

2.1.4. Immune Gene Signatures and Composite Scores

An increase in the immune score derived from the expression profile of immune genes or multiple immunofluorescence (mIF) for immune cell infiltration is associated with a favorable response to ICI and a higher survival rate. There is potential to monitor the therapeutic effect of ICI treatment on patients with SqNSCLC [27]. The immuno‐score‐IC may predict efficacy; utilizing spatial quantitative analysis of CD8 and PD‐L1 markers can predict therapeutic efficacy [28]. The immunohistochemical phenotype assessment standard (irHPC) scoring system can predict the pathological response to neoadjuvant immunotherapy; it may also be possible to make further predictions regarding the long‐term prognosis [38].

MAIT cells are an essential component of the immune system. Circulating MAIT cells have been shown to predict the efficacy of anti‐PD‐1 therapy [39], and the activation of MAIT cells facilitates tumor regression in patients receiving ICB. There is evidence of association between different MAIT subgroups and the response rate to ICIs [29]. CD8⁺ MAIT cells can also predict the efficacy of anti‐PD‐1 therapy and may be potential biomarkers that can serve as predictors of the efficacy of immunotherapy [30].

The success of anti‐PD‐1 therapy depends on the preexisting TIL in the tumor that are in an adaptive immune resistance state and the treatment outcome is related to this, the additional cost is extremely low, and it may also enhance the effectiveness of precise treatment [40, 54]. The immunophenotype score that assesses the degree of tumor‐infiltrating immune cells (TIIC) in the TME can predict the efficacy of immunotherapy; it may have potential clinical applications [55].

The stability and functionality assessment of PD‐L1 expression as well as sPD‐1/sPD‐L1 can be used for patient stratification and efficacy prediction. The density, proportion, and status of various immune cells and their subsets are closely related to the treatment response and survival outcome. The characteristic parameters of TCR and the TMR score integrating it with TMB also have predictive value. In addition, comprehensive scoring systems based on immune gene characteristics, immune cell infiltration, or TME immune phenotype can effectively evaluate the efficacy of immunotherapy. These markers jointly reveal the heterogeneity of the immune response and provide an important basis for individualized treatment.

2.2. Tumor Genomic Characteristics

2.2.1. TMB

High MTV before treatment is an adverse prognostic factor for tumor treatment. Baseline metabolic tumor burden measured by MTV has been confirmed as a prognostic biomarker for treatment response and may be an independent predictor of ICI treatment response [56]. Higher TMB cutoff values can identify individuals who may benefit from long‐term benefits. Specifically, TMB ≥ 20 mutations/megabase can identify patients who may benefit from ICIs even if their PD‐L1 level is low [57].

2.2.2. Driver Gene Mutations and co‐Mutation Patterns

BRAF mutations may serve as a good prognostic indicator for advanced and locally advanced patients receiving immunotherapy, demonstrating a high degree of sensitivity [58]. Additionally, patients without ERBB2 mutations are more likely to benefit from ICI treatment [31]. Genomic mutation features involving ZFHX3, NTRK3, EPHA7, MGA, STK11, EPHA5, and TP53 have been identified as predictive markers; they may serve as a potential biomarker for guiding clinical decisions regarding ICI treatment [59, 60]. Models based on ERBB4 and TP53 mutation status have also been developed to assess the efficacy of ICI treatment; they are particularly applicable to patients with TP53 mutations [61]. The variant allele frequency (VAF) model covering 15 genes indicates that the VAF mutation spectrum can serve as a marker. The VAF model can distinguish the ICI response in NSCLC [62]. Based on mutation‐based predictive models, the potential benefits of ICI therapy can be evaluated; it is conducive to the optimization of NSCLC treatment prediction algorithms [63].

2.2.3. Other Somatic Mutations

The ALKBH5 (alkB homolog protein 5) status can predict the therapeutic effect and is expected to become a predictive biomarker [41]. Overexpression of TROP2 indicates a higher survival rate. The strategy based on the TROP2 biomarker may be applicable to select patients who are more likely to benefit from the combination of immunotherapy and anti‐TROP2 drugs [64]. Trim35 (an E3 ligase) is a predictive marker and a promising biomarker for predicting the outcome of immunotherapy for NSCLC [42]. Binding protein markers (CXCL9, CXCL10, IL‐15, CASP8, and ADA) are related to PFS and may predict the response to anti‐PD‐(L)1 therapy [32]. Using the genetic characteristics of FCER2, CD200R1, RHOV, TNNT2, WT1, AHSG, and KRTAP5‐8 genes to construct a model can guide the immunotherapy. It has excellent predictive capabilities and is of great significance for guiding immunotherapy [65].

A high TMB can identify potential benefit populations, especially when PD‐L1 expression is low. Specific driver gene mutations and co‐mutation patterns have predictive value. In addition, models constructed based on the VAF of multiple gene variations or specific gene sets and the expression status of other somatic mutations can evaluate the potential efficacy of immune checkpoint inhibitor therapy, providing a basis for precise treatment decisions.

2.3. Host Systemic and Peripheral Blood Biomarkers

2.3.1. Circulating Immune Cells and Soluble Factors

The baseline NLR and dNLR at the second cycle (C2) are related to the treatment prognosis. The persistently high dNLR is associated with the failure of early ICI treatment [66]. The early dynamic changes of CRP can predict the response of patients to ICIs within the first month of treatment. It is expected to become a biomarker that is easy to determine, cost‐effective, and noninvasive [43]. These easily accessible hematological indicators provide practical tools for dynamic monitoring of the efficacy and prognosis of immunotherapy.

2.4. Dynamic and Liquid Biopsy‐Based Biomarkers

2.4.1. Circulating Tumor DNA and Cell‐Free DNA

The efficacy of chemotherapy and immunotherapy can be evaluated based on the lowest ctDNA level in the early stage, which is a major predictor of both PFS and OS [44]. Clonal (defined by ctDNA) predicts the benefits of immunotherapy. It is applicable in the prognostic stratification of LA—NSCLC and the prediction of the benefits of immunotherapy [45]. In the longitudinal changes of ctDNA, the complete clearance of extracellular free tumor burden (cfTL) is significantly correlated with the PFS and OS. Molecular reactions help to explain heterogeneous clinical responses and are particularly applicable to patients with stable diseases [67].

2.4.2. Circulating Tumor Cells and Extracellular Vesicles

The dynamic changes of extracellular vesicles PD‐L1 (EV PD‐L1) in plasma samples could predict the rate of durable response. It can be used as a predictive model to identify patients with advanced lung cancer who can benefit from ICI [68]. Conversely, the high expression of EV TGF‐β on the surface of tumor cells was associated with no response and shorter PFS and OS. It is more accurate than TGF‐β in the cycle and the PD‐L1 in the tissue [33]. EV‐miR‐625‐5p has been identified as an independent biomarker. In particular, for patients with PD‐L1 ≥ 50% [34].

2.4.3. RNA Biomarkers

RNA sequencing (RNA‐seq) has the potential to help optimize the immunotherapy regimens. It may improve the options for immunotherapy in patients with NSCLC [69]. Exosomal long‐chain RNA in plasma has been identified as a predictive indicator of a good prognosis for patients undergoing neoadjuvant immunotherapy [35]. The Writer‐Score system predicts the prognosis of neoadjuvant PD‐1 inhibitor therapy by quantifying RNA modification patterns [70].

The early lowest level of ctDNA, the clonal status, and the dynamic changes during the treatment process are significantly correlated with the PFS and OS. The dynamic changes of plasma EV PD‐L1 can predict the rate of durable response, while high expression of EV TGF‐β is associated with no response and shorter survival. EV‐miR‐625‐5p can serve as an independent biomarker. Additionally, RNA‐seq helps optimize treatment plans. Long‐chain RNA in plasma exosomes and a scoring system based on RNA modification patterns also provide new approaches for predicting the efficacy and prognosis of immunotherapy.

2.5. The Gut Microbiome as an Ecosystem Modulator

The metabolite of the gut microbiota, butyric acid, can promote the efficacy of anti‐PD‐1 immunotherapy. It can significantly enhance the predictive effect of immunotherapy for patients [46]. The proportions of Bacteroidetes, Bacteroidaceae, and Proteobacteria are very high, while the presence of γ‐Proteobacteria is associated with adverse reactions to immunotherapy [71]. Short PFS is associated with increased abundance of the Bacteroidetes (F) and Actinobacteria families, while an increase in the abundance of the Eukaryotic Archaea family is associated with low PD‐L1 expression. The F/bacteroides (F/B) ratio is significantly increased in patients with short PFS. The PD‐L1 phenotype is best predicted through metabolic pathways, while the classification features are more suitable for predicting the outcome [47]. The composition and diversity of the gut microbiota are closely related to the efficacy of immunotherapy for NSCLC.

2.6. Radiomics and Imaging Biomarkers

CT imaging has been proven to be able to predict the treatment outcome, as well as the phenomena of pseudo‐progression (PP) and hyperprogressive disease (HPD). The radiological features within and around tumors based on noninvasive CT can be used to detect atypical tumor responses before immunotherapy [72]. The radiological features targeting CD274 derived from CT scans can be used to assess the suitability of anti‐PD‐1 or PD‐L1 treatment [73]. Moreover, ^[68Ga] GA‐NOTA‐RW102 immunopET imaging, can be used to screen patients who respond to immunotherapy [74].

A subregional radiomics model (SRRM) was established using 11 subregional imaging features for stratifying the risk of PFS. The predictive performance of SRRM is superior, outperforming traditional radiology, PD‐L1 expression, and TMB scoring [75]. Additionally, the delta‐radiomics model constructed from short‐term follow‐up CT radiomics data demonstrated effectiveness in predicting PFS and OS, with better predictive results for lung adenocarcinoma (LUAD) than lung squamous cell carcinoma (LUSC) [76]. A comprehensive model combining radiomics and tumor‐infiltrating PD‐L1 can stratify patients. This multiparameter model demonstrates high sensitivity and specificity in identifying ICI nonresponders and is superior to tissue PD‐L1. Moreover, it is directly related to tumor changes [77]. Simultaneously, baseline models and longitudinal models were implemented, and the results indicated that the combination of multidimensional and longitudinal data enhances the detection of sustained immune therapy benefits in advanced patients. This model can be used as a prognostic biomarker and a decision support tool [78]. The metastasis scoring system (METscore) based on baseline imaging can evaluate the therapeutic efficacy based on the baseline imaging assessment of advanced patients. It may be possible to noninvasively predict the efficacy of PD‐L1 blockade in patients with advanced NSCLC through baseline radiological assessment [79].

Imaging biomarkers provide an important noninvasive assessment method for predicting the efficacy and prognosis of immunotherapy in NSCLC. Based on imaging features from CT, MRI, and PET/CT, they can effectively predict treatment responses. Moreover, by combining imaging features targeting PD‐L1, immune PET imaging, and clinical pathological factors, a fusion model further improves the prediction accuracy, facilitating the selection of patients benefiting from immunotherapy and enabling individualized efficacy assessment.

2.7. Integrative Models and Artificial Intelligence

Automatic analysis of radiological scans using deep learning can provide Supporting Information to guide immunotherapy strategies. It can provide independent biomarkers that are distinct from existing clinical and pathological markers, thereby facilitating precise immunotherapy for patients with NSCLC [80]. ML and deep learning technologies have shown significant potential in predicting the efficacy of immunotherapy for NSCLC.

3. Predicting Immune Resistance

The presence of CD68⁺ cells, CD33⁺ cells, FOXP3⁺ cells, and the nonnuclear expression of FOXP3⁺ cells are associated with resistance to ICIs [81]. The expression of CD66b is related to resistance to ICI treatment [82]. The level of autocrine factors (ATX) is negatively correlated with the total CD8⁺ T cell count and the effector CD8⁺ T cell count, and the ATX/LPA axis promotes resistance to PD‐1 treatment [83]. PTEN deficiency is associated with immune treatment resistance in LUSC [84]. The enrichment of tumor‐associated macrophages (TAMs) is related to immune treatment resistance [85]. Acquired inactivating mutations in STK11, B2M, APC, MTOR, KEAP1, and JAK1/2 are associated with resistance to ICI [86]. RPL13A and GNL3 are proposed as markers of resistance to PD‐1 blockers [87]. Activating signal co‐integrator 3 (ASCC3) has been confirmed to be able to enhance resistance to PD‐1 treatment [88]. Analysis of the methylation of plasma free DNA can reveal genes related to anti‐PD‐1 resistance [89]. Table 3 lists the predictors of immune resistance in clinical trials.

TABLE 3.

lists the predictors of immune resistance in clinical trials.

Participants	Clinical stage	Histology	Treatment stage	Intervention	Results	References
25 NSCLC	Stage (IIB; III; IV)	Adenocarcinoma; squamous; others		ICI monotherapy; chemotherapy–immunotherapy	CD68+, CD33+, FOXP3+ (including nonnuclear) cells predict ICI resistance	[81]
56 NSCLC	Stage (I; II; III; IV)	Adenocarcinoma; adenosquamous; NOS; squamous; large‐cell carcinoma; NA	Line (1; 2; 3/4)	Atezolizumab; durvalumab/tremelimumab; ipilimumab/nivolumab; nivolumab; pembrolizumab	CD66b expression predicts ICI resistance in NSCLC	[82]
152 NSCLC		Adenocarcinoma; other	Line (≤ 1; > 1)	Anti‐PD1/PD‐L1 monotherapy; anti‐PD1/PD‐L1 in combination with another ICI	TAM enrichment in NSCLC TME predicts immunotherapy resistance	[85]
56 NSCLC	Stage (III; IV)	Adenocarcinoma; adenosquamous; NOS; squamous	Line (1; 2; 3/4)	Chemotherapy + anti‐PD‐1/anti‐PD‐L1 monotherapy; anti‐PD‐1 + anti‐CTLA‐4 monotherapy; other combinations; anti‐PD‐1/anti‐PD‐L1 monotherapy	RPL13A and GNL3 predict PD‐1 blockade resistance	[87]
179 NSCLC		Adenocarcinoma; SCC; others	Line (1; ≥ 2)	Anti‐PD‐1/PD‐L1	HLA‐I homozygosity predicts lower risk of severe irAEs in NSCLC patients on anti‐PD‐1/PD‐L1	[90]
556 NSCLC	Stage (IV)	Squamous; nonsquamous	Line (1; 2)	Nivolumab; pembrolizumab; other	CT imaging biomarkers differentiate immunotherapy‐related pneumonia in NSCLC	[91]

The presence or enrichment of specific immune cells in the TME is associated with drug resistance. The expression of CD66b and the absence of PTEN are also related to drug resistance. The autocrine factor ATX and the signaling axis formed with LPA promote PD‐1 treatment resistance by affecting the number of CD8⁺ T cells. Additionally, acquired inactivating mutations in multiple genes and abnormal expression of specific genes have been identified as markers of drug resistance. The methylation analysis of plasma free DNA can also reveal genes related to anti‐PD‐1 resistance. These findings are helpful in identifying potential resistant patients and exploring combined treatment strategies.

4. Predicting Immune‐Related Adverse Events (irAEs)

Haplotype homozygosity at certain human leukocyte antigen‐I (HLA‐I) loci can predict that patients are less likely to experience severe irAEs [90]. The difference in mean platelet volume (MPV) is associated with the incidence of irAEs after pembrolizumab treatment [92]. The relationship between immunotherapy and local immune‐mediated pneumonia (LIPI) is related to the occurrence of irAEs [93]. Baseline and dynamic plasma interleukin‐10 (IL‐10) levels are independently associated with the occurrence of irAEs [94]. CT imaging has been used to identify CIP [91]. Table 3 lists the predictors of immune‐related adverse reactions in clinical trials.

Existing studies have identified several potential predictive markers for irAEs, including homozygosity of HLA‐I, changes in MPV, association with LIPI, and baseline and dynamic levels of IL‐10. At the same time, techniques such as CT imaging have been applied to identify CIP. These markers provide important evidence for risk prediction of irAEs in clinical practice.

5. Discussion

Currently recommended biomarkers for clinical use include sPD‐L1 and sCTLA‐4 [53], early dynamic changes of CRP [43], and the lowest ctDNA level in the early stage [44]. Potential biomarkers for clinical application include baseline T cell density and the expression of T cell exhaustion markers [20], TROP2 [64], cfTL [67], CT imaging [72], RPL13A and GNL3 [87], and baseline and dynamic IL‐10 [94]. Biomarkers that may assist in clinical judgment include the TIR index [37], immune score [27], and RNA‐seq [69]. Additionally, MAIT [29, 39] and ALKBH5 [41] are also considered helpful for clinical assessment. Table 4 lists the recommendation level of biomarkers.

TABLE 4.

Lists the recommendation level of biomarkers.

Recommendation level	Participants	Sample type	Detection method	Biomarker	References
Recommended for clinical application	Advanced NSCLC	Blood samples	Highly sensitive chemiluminescence‐based assay	sPD‐L1 and sCTLA‐4	[53]
Recommended for clinical application	Advanced NSCLC	Blood samples	Measuring serum parameters	Early CRP kinetics	[43]
Recommended for clinical application	Stage IV or recurrent metastatic nonsquamous NSCLC	Blood samples	ctDNA sequencing‐330 kb assay targeting 311 genes	Lowest ctDNA level in the early stage and integrating radiographic	[44]
Can be used in clinical applications	Stage (I–II; III–IV) NSCLC	FFPE whole‐tissue sections of NSCLC samples	Multiplexed immunofluorescence staining	Baseline T cell density and T cell exhaustion marker expression	[20]
Can be used for clinical applications	NSCLC		CT scans	Radiomic features extracted from baseline CT scans	[72]
May be used in clinical applications	NSCLC who progressed following platinum‐based chemotherapy	Tumor tissue samples, blood samples	High‐throughput analysis of transcriptomic, multiplex immunofluorescence, proteomic profiling of plasma	TROP2	[64]
May be used for clinical applications	NSCLC	FFPE tumor specimens	High‐throughput transcriptome analysis	RPL13A and GNL3	[87]
May be used for clinical applications	Advanced NSCLC	Peripheral blood	Cytokine measurements	Baseline and dynamic IL‐10 plasma levels	[94]
May be used in clinical applications	Stage (I; II; III; IV) NSCLC	Peripheral blood, tumor, and adjacent tissue samples	Flow cytometry and single‐cell RNA sequencing	Circulating MAIT subset	[39]
Provided methods for clinical use	Advanced NSCLC	Blood samples	Plasma cell‐free DNA, WBC, and tumor NGS	cfTL	[67]

Among the biomarkers included in this article, immunophenotype score [55], TMB cutoff [57], ZFHX3, NTRK3, EPHA7, MGA, STK11, EPHA5, and TP53 genomic mutation features [59, 60], ERBB4 and TP53 mutation model [61], VAF model [62], mutation‐based predictive model [63], FCER2, CD200R1, RHOV, TNNT2, WT1, AHSG, and KRTAP5‐8 genes model [65], Writer‐Score system [70], and SRRM [75] were derived from database exploration. ALKBH5 [41], ^[68Ga] GA‐NoTA‐NB109 and PET imaging [74], ASCC3 [88], butyric acid [46] were derived from preclinical evidence. Summarize the biomarkers for immunotherapy in NSCLC shown in Figure 2. The remaining biomarkers were obtained from clinical trials and clinical sample tests, and have a high level of evidence, which can be used as a reference for clinicians when applying them. Furthermore, when faced with inconsistent biomarker results, clinicians should adhere to the principle that: time takes precedence over absolute values, integration is superior to singularity, and dynamics are better than staticity. It is suggested that in clinical practice, the latest test results should be the core, supplemented by multidimensional information, and dynamically monitored for verification. This will maximize the precision of decision‐making in immunotherapy.

FIGURE 2.

Summarize the biomarkers for immunotherapy in NSCLC. Source: Created in BioRender. Sun, Y. https://BioRender.com/0u4j8tc.

The inhibition and reprogramming of the immune system are one of the core mechanisms underlying tumor occurrence and progression. Immunotherapy aims to activate the anti‐tumor immune response and reverse the immune evasion of tumors. It is worth noting that tumor immunotherapy, represented by immune checkpoint inhibitors, has demonstrated remarkable clinical efficacy, enabling some previously treatment‐resistant malignant tumors to achieve sustained remission and promoting the development of tumor treatment towards a more precise direction. However, the absence of effective early diagnostic platforms and the typically delayed manifestation of symptoms during disease progression complicate the prognosis of NSCLC, thereby constraining therapeutic options and adversely impacting patient survival outcomes [95]. The effective management of patients with NSCLC necessitates the implementation of precise molecular testing to identify predictive biomarkers. The information obtained from such testing is essential for the development of personalized therapeutic approaches. The detection of specific biomarkers enhances the likelihood of administering more efficacious treatments while concurrently minimizing adverse effects linked to nontargeted therapies. This approach ultimately contributes to improved patient quality of life and more efficient allocation of healthcare resources [96].

However, although biomarker research has broad prospects, it still faces multiple challenges. Firstly, the spatial heterogeneity within tumors and metastatic foci, as well as the temporal heterogeneity induced by treatment pressure, severely limits the representativeness and timeliness of single‐point biopsy results. Secondly, most emerging biomarkers lack unified global detection standards and validation procedures, and their high analysis costs and complex technical requirements also hinder their widespread implementation in primary care institutions. Moreover, the majority of current biomarkers' evidence still remains at the stage of retrospective analysis or small‐scale prospective exploration, and urgently require rigorous validation through large‐scale, prospective, and interventional clinical trials to strictly verify their predictive efficacy and clinical applicability.

To systematically overcome these bottlenecks, future development should focus on the following key directions: First, build a dynamically integrated biomarker ecosystem. Second, promote multidimensional data fusion and artificial intelligence decision support. By deeply integrating multidimensional information such as genomics, transcriptomics, proteomics, metabolomics, microbiomics, imagingomics, and clinical data, and using machine learning and deep learning algorithms to uncover hidden complex patterns. Third, explore new biomarkers for overcoming drug resistance. A deeper understanding of the resistance mechanism will give rise to new biomarkers for predicting secondary resistance. At the same time, develop biomarkers that can accurately and early predict severe irAEs, improving patient safety and quality of life.

Author Contributions

Yuting Sun: conceptualization, methodology, writing – original draft. Guanghui Zhu: writing – original draft. Xinyi Ma: writing – original draft. Xuelei Chu: conceptualization, software, writing – original draft. Xue He: writing – original draft. Weina Li: methodology, writing – original draft. Xiaoyu Zhu: writing – original draft. Xinmiao Wang: conceptualization, software, writing – original draft. Jie Li: conceptualization, funding acquisition, writing – review and editing.

Funding

This work was supported by National Key Research and Development Program of China (2023YFC3503300, 2023YFC3503305) and The Science and Technology Innovation Projects for Graduate Students at China Academy of Chinese Medical Sciences (KC2025006).

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.