Correlation between preoperative T-cell receptor profile characteristics and postoperative pathological stage/treatment response in non-small cell lung cancer: a prospective cohort study

Abstract

Background

Non-small cell lung cancer (NSCLC) is the leading cause of cancer mortality worldwide, with a substantial proportion of patients diagnosed at advanced stages and recurring after surgery. Traditional prognostic tools like TNM staging do not fully account for the host immune response. T-cell receptor (TCR) repertoire profiling, which reflects T-cell diversity and clonality, shows promise as a biomarker for predicting disease progression and therapy response. This study investigates the correlation between preoperative peripheral blood TCRβ repertoire characteristics and postoperative pathological stage, survival outcomes, and response to adjuvant therapies in NSCLC.

Materials and methods

A prospective cohort of 120 treatment-naïve NSCLC patients scheduled for curative surgical resection was enrolled. Peripheral blood mononuclear cells were collected preoperatively to perform high-throughput sequencing of the TCRβ complementarity-determining region 3 (CDR3). TCR repertoire diversity, clonality, CDR3 length distribution, and clonotype sharing were quantified. Clinical data including postoperative pathological staging and treatment responses were collected. Statistical analyses included correlation tests, survival analyses (Kaplan-Meier, Cox regression), hierarchical clustering, and ROC curve analyses to evaluate prognostic performance.

Results

As expected, pathological stage demonstrated strong prognostic value for both DFS and OS (log-rank P < 0.001). TCR diversity was significantly lower in advanced-stage NSCLC (III-IV) compared to early stages (I-II), while clonality positively correlated with higher pathological stage. High TCR clonality (≥ 0.35) independently predicted worse progression-free survival (DFS; adjusted HR = 1.7, P = 0.04) and overall survival (OS; adjusted HR = 1.9, P = 0.02). Time-dependent ROC analyses showed TCR clonality outperformed pathological stage and PD-L1 status in 2-year OS prediction (AUC = 0.72). CDR3 length distribution skewed toward shorter lengths in advanced stages, correlating with higher clonality and poorer survival. Hierarchical clustering identified distinct TCR repertoire patterns segregating early and advanced stages, with three public clonotypes enriched in stage III-IV patients. Among treatment subgroups, high TCR diversity predicted better DFS in patients receiving immunotherapy (HR = 0.29, P = 0.02) but not in chemotherapy or targeted therapy groups.

Conclusion

Preoperative peripheral blood TCRβ repertoire profiling provides valuable prognostic and predictive biomarkers in NSCLC. Reduced TCR diversity and increased clonality correlate with advanced pathological stage and poorer survival, while preserved diversity identifies patients likely to benefit from immunotherapy. These findings support the potential of preoperative TCR repertoire analysis as a baseline for post-operative monitoring and a component for refining risk stratification models, which could guide adjuvant therapy intensity and selection.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-16046-7.

Introduction

Lung cancer persists as the foremost cause of global cancer mortality [1]. The majority of cases (80–85%) are classified as non-small cell lung cancer (NSCLC), which is frequently diagnosed at an advanced stage, contributing to its poor prognosis [2]. While surgical resection is the cornerstone of curative-intent treatment for early-stage disease, only a minority of patients present with resectable tumors. Furthermore, a significant proportion (30–55%) of those undergoing surgery experience disease recurrence [3–5]. To address these limitations, the neoadjuvant paradigm has gained prominence. Administering systemic therapy prior to surgery offers the dual benefit of targeting micrometastases and downstaging tumors to convert unresectable cases into resectable ones [6, 7]. The drawbacks of this strategy include the risk of interim disease progression (which can disqualify patients from surgery) and potential adverse effects on surgical outcomes.

Although clinical factors like tumor-node-metastasis (TNM) staging are cornerstone tools for prognosticating post-surgical outcomes and guiding adjuvant therapy in lung cancer [8, 9], their underlying assumption that progression is solely a tumor cell-autonomous process is limiting [10]. This view fails to incorporate the critical role of the host immune response, a well-established predictor of tumor progression and survival across various cancers [11–13]. Consequently, integrating immune-related markers with traditional clinical factors may provide a more robust prognostic framework for postoperative outcomes [14]. The recognition of neoantigens by T-cells is a critical mechanism in the anti-tumor immune response [15]. This process is initiated when TCRs bind to antigens presented by the major histocompatibility complex (MHC), triggering adaptive immunity [16]. Structurally, TCRs are heterodimers composed of alpha (α) and beta (β) chains. Their remarkable diversity is generated through the somatic recombination of variable (V) and joining (J) gene segments, with the TCRβ chain incorporating an additional diversity (D) segment. Antigen specificity is primarily determined by three complementarity-determining regions (CDRs), with the hypervariable CDR3 region being particularly crucial for antigen binding and contributing substantially to the overall diversity of the TCR repertoire [17]. The frequency and diversity of TCRβ clones are strongly linked to the strength and breadth of the cellular immune response against tumor antigens [18]. Advances in high-throughput sequencing have enabled detailed profiling of the TCR repertoire, leading to growing interest in its use as a biomarker for cancer diagnosis, prognosis, and predicting responses to immune checkpoint inhibition (ICI) [19–22]. Nonetheless, the predictive utility of TCR repertoire metrics for immunotherapy outcomes is not yet definitive and remains a subject of debate [19–22]. Consequently, further investigation is essential to validate the clinical applicability of cancer-specific TCR-based biomarkers for forecasting efficacy in ICI treatment. Research into predicting responses to immune-based therapies via TCR repertoire analysis has traditionally relied on sequencing tumor-infiltrating lymphocytes (TILs). This approach is limited by its dependence on invasive tissue biopsies that must contain sufficient tumor material, a requirement that is often difficult to fulfill [23–25]. Emerging evidence, however, indicates that the TCR repertoire found in peripheral blood can mirror that of the tumor microenvironment. This suggests that less-invasive blood-based biomarkers could serve as predictive tools for response to PD-1 inhibitor therapy in some cancers [20]. Reuben et al. [24] demonstrated that intratumoral TCR heterogeneity is a significant predictor of postoperative relapse and disease-free survival in localized lung adenocarcinoma, their methodology required multi-region TCR sequencing (2–5 regions per tumor) to accurately assess this heterogeneity [24]. This requirement for extensive sampling presents a substantial barrier to the translation of this biomarker into routine clinical practice.

To address this, the present prospective study aimed to evaluate the correlation between preoperative peripheral blood TCRβ repertoire features including diversity, clonality, CDR3 length distribution, and clonotype sharing and postoperative pathological staging, survival endpoints (DFS and overall survival (OS)), and response to adjuvant therapies in a cohort of 120 NSCLC patients undergoing curative resection. By integrating high-throughput TCR sequencing with clinicopathological data, we sought to establish the prognostic value of preoperative TCR profiling and explore its potential as a baseline biomarker. This could, in future applications, inform postoperative monitoring strategies and help optimize adjuvant therapy decisions.

Materials and methods

Study design and cohort

This prospective cohort study was conducted at Second Hospital of Shanxi and Beijing Cancer Hospital between 2022 and 2025. We enrolled 120 treatment-naïve patients with histologically confirmed non-small cell lung cancer (NSCLC) scheduled for surgical resection with curative intent. Eligible participants were adults (≥ 18 years) with adenocarcinoma, squamous cell carcinoma, or large cell carcinoma subtypes, and no prior systemic anticancer therapy. Molecular profiling for key oncogenic drivers (e.g., EGFR, ALK) was performed as part of routine clinical practice for patients with non-squamous histology, consistent with standard guidelines. Exclusion criteria comprised small cell or mixed histology, distant metastases (AJCC M1b/c) at diagnosis, active autoimmune disease, immunosuppressive therapy, or inability to complete follow-up. The target sample size of 120 patients was determined through power analysis using G*Power (v3.1), assuming a 30% prevalence of advanced-stage (III–IV) disease, a hazard ratio of 2.0 for low versus high TCR diversity, 80% power, a two-sided α of 0.05, and a 15% anticipated attrition rate. This study was conducted in accordance with the ethical principles of the Declaration of Helsinki and received Ethical approval from the Institutional Review Board of Second Hospital of Shanxi Medical University (Protocol No. 2025YX314), and written informed consent was secured from all participants prior to enrollment.

Sample collection and TCR sequencing

Preoperative peripheral blood samples (10 mL) were collected from all participants in EDTA tubes within 7 days prior to surgical resection. Peripheral blood mononuclear cells (PBMCs) were isolated within 2 h of collection using Ficoll-Paque PLUS density gradient centrifugation (GE Healthcare) according to manufacturer specifications. Genomic DNA was extracted from 2 × 10⁶ PBMCs using the QIAamp DNA Blood Mini Kit (Qiagen, Hilden, Germany), with DNA concentration and purity assessed by NanoDrop spectrophotometry (A260/A280 > 1.8).

T cell receptor β-chain (TCRβ) complementarity-determining region 3 (CDR3) sequencing was performed using the ImmunoSEQ hsTCRB assay (Adaptive Biotechnologies, Seattle, USA). Briefly, multiplex PCR amplification of rearranged TCRβ loci was conducted using V- and J-gene-specific primers, followed by high-throughput sequencing on the Illumina MiSeq platform (2 × 150 bp paired-end reads). Raw sequencing data were processed through the Adaptive Biotechnologies immuneACCESS platform (v3.0) for error correction, clonotype assembly, and quantification. Alternatively, for validation purposes, raw FASTQ files were reanalyzed using MiXCR (v3.0.13) with default parameters for alignment, error correction, and clonotype calling. PD-L1 expression was assessed on archival tumor tissue using the PD-L1 IHC 22C3 pharmDx assay (Agilent Technologies) following manufacturer’s protocols. Tumor proportion scores (TPS) were determined by qualified pathologists, with results categorized as negative (< 1%), low (1–49%), or high (≥ 50%) expression.

Quality control metrics included:

• Minimum library concentration: 10 ng/µL

• Minimum sequencing depth: 50,000 productive templates per sample

• Removal of clonotypes with < 2 reads or < 0.0001% frequency

TCR repertoire analysis

TCR repertoire characteristics were quantified using established immunogenetic metrics. Repertoire diversity was calculated via the Shannon entropy index H=−where S represents the total number of unique clonotypes and p_i​ denotes the frequency of the *i*-th clonotype. Higher H values indicate greater repertoire heterogeneity. Clonality was defined as 1-(Pielou’s evenness), ranging from 0 (polyclonal) to 1 (monoclonal). The top 10 expanded clonotypes were identified by ranking all clonotypes by their relative frequency within each sample.

CDR3 length distribution analysis

The distribution of CDR3 amino acid lengths was evaluated for each patient’s TCRβ repertoire. CDR3 lengths were determined from the assembled sequences obtained via the ImmunoSEQ assay or MiXCR analysis. Only productive CDR3 sequences were included, with lengths ranging typically from 5 to 25 amino acids. Histograms of CDR3 length frequencies were generated for visualization, stratified by pathological stage (I–II vs. III–IV). Statistical differences in length distributions between groups were assessed using the Kolmogorov-Smirnov test, with P < 0.05 considered significant. All analyses were performed in R (v4.2.0) using the ggplot2 package (v3.3.5) for plotting.

For comparative analysis, hierarchical clustering of TCRβ CDR3 amino acid sequences was performed using the pheatmap package (v1.0.12) in R (v4.2.0). A binary presence-absence matrix of CDR3β sequences across all samples was generated, and Euclidean distances were computed between samples. Ward’s minimum variance method was applied for dendrogram construction. Stage-specific clonotype enrichment was assessed using Fisher’s exact test with Bonferroni correction (P < 0.01 considered significant).

The relationship between TCR diversity and clonality was quantified using both Spearman rank correlation (for non-parametric assessment) and Pearson correlation (assuming linearity). Normality of distribution for each metric was assessed using Shapiro-Wilk tests. The proportion of shared variance was calculated as the square of the Pearson correlation coefficient (R²).

Clinical endpoints

Pathological staging was determined postoperatively according to the American Joint Committee on Cancer (AJCC) 8th edition TNM classification system by two independent pathologists blinded to TCR data. Discrepancies were resolved through consensus review.

Primary survival endpoints were defined as follows:

1. Disease-free survival (DFS): Time from surgical resection to first documented disease recurrence, new primary lung cancer, or death from any cause.

2. Overall survival (OS): Time from surgical resection to death from any cause.

Patients without events were censored at last follow-up (March 31, 2025). Treatment response for patients receiving neoadjuvant/adjuvant therapy was assessed via:

• Pathological complete response (pCR): Absence of viable tumor cells in resected specimens

• Radiological response: CT-based RECIST v1.1 criteria every 12 weeks

Pathological assessment, including pCR determination, was performed by two independent institutional pathologists at each study site, blinded to the TCR sequencing data. Discrepancies were resolved through consensus review.

Statistical analysis

All statistical analyses were performed using R software (version 4.2.0; R Foundation for Statistical Computing, Vienna, Austria). Continuous variables, including TCR repertoire metrics, were expressed as median with interquartile range (IQR) and compared using non-parametric tests—Mann-Whitney U test for comparisons between two groups or Kruskal-Wallis test for multiple group comparisons, as appropriate. Categorical variables were summarized as frequencies and percentages and analyzed using Fisher’s exact test. Associations between TCR characteristics (diversity and clonality) and pathological stage were quantified using Spearman’s rank correlation coefficient (ρ). Associations between TCR metrics and categorical variables (stage, treatment) were assessed using Kruskal-Wallis tests for multiple groups and Mann-Whitney U tests for pairwise comparisons. Effect sizes were calculated using Cohen’s d for continuous outcomes. All correlation analyses used Spearman’s rank correlation for non-parametric data. Multicollinearity in multivariate Cox models was assessed using Variance Inflation Factors (VIF), with values ≥ 5 indicating concerning collinearity. Sensitivity analyses were performed by fitting models with different covariate combinations to assess the stability of hazard ratio estimates. Model stability was considered satisfactory if hazard ratios changed by < 20% when removing correlated variables. Stratified Cox proportional hazards models were fitted within stage subgroups (I-II and III-IV) to assess whether TCR metrics predict outcomes independently within homogeneous stage categories. Interaction between stage and TCR metrics was tested using product terms in multivariate models. Time-dependent receiver operating characteristic analysis was performed for 3-year overall survival prediction. Area under the curve values were calculated for individual predictors (TCR metrics, pathological stage) and combined models. Model performance was compared using AUC differences and concordance indices.

For survival analysis, disease-free survival (DFS) and overall survival (OS) were assessed using Kaplan-Meier estimates, and differences between TCR-based subgroups were compared with the log-rank test. Both univariate and multivariate Cox proportional hazards models were employed to calculate hazard ratios (HR) with 95% confidence intervals (CI); the proportional hazards assumption was verified using Schoenfeld residuals. Multivariate models were adjusted for clinically relevant covariates including age, sex, smoking status, tumor stage, and treatment modality. The prognostic accuracy of TCR metrics was evaluated via time-dependent receiver operating characteristic (ROC) analysis at 1, 2, and 3 years using the survival ROC package. Optimal cutpoints for TCR diversity and clonality were determined using maximally selected rank statistics via the survminer package to dichotomize patients into high- and low-risk groups. Results from hierarchical clustering of TCR repertoires were validated using silhouette width analysis. A two-sided p-value < 0.05 was considered statistically significant, and false discovery rate (FDR) correction was applied to multiple comparisons in clonotype-stage analyses.

Results

Cohort characteristics

The study cohort comprised 120 treatment-naïve NSCLC patients who underwent surgical resection between 2022 and 2025. Demographic and clinicopathological characteristics are summarized in Table 1. The median age was 62 years (range: 38–78), with a male predominance (65 patients, 54.2%). Histologically, adenocarcinoma accounted for 65.0% (n = 78) of cases, followed by squamous cell carcinoma (n = 36, 30.0%) and large cell carcinoma (n = 6, 5.0%). Pathological staging distribution revealed 32 patients (26.7%) with stage I disease, 36 (30.0%) with stage II, 42 (35.0%) with stage III, and 10 (8.3%) with stage IV. Current or former smokers represented 68.3% (n = 82) of the cohort. Adjuvant therapy was administered to 92 patients (76.7%), including chemotherapy (n = 64, 53.3%), targeted therapy (n = 20, 16.7%), and immunotherapy (n = 28, 23.3%). Treatment distribution across stage groups revealed expected patterns, with higher utilization of systemic therapy in advanced stages. While only 66.2% (45/68) of stage I-II patients received adjuvant therapy, this proportion increased to 88.1% (37/42) in stage III and 100% (10/10) in stage IV patients. Immunotherapy was predominantly administered to advanced-stage patients, comprising 42.9% (18/42) of stage III and 50.0% (5/10) of stage IV cases, compared to only 7.4% (5/68) of stage I-II patients. Chemotherapy was the most common modality across all stages, while targeted therapy was relatively evenly distributed (Table 1).

Table 1.

Baseline demographic and clinicopathological characteristics

Characteristic	Total (n = 120)	%	Stage I-II (n = 68)	Stage III (n = 42)	Stage IV (n = 10)
Age (years)
Median (range)	62 (38–78)	–	61 (42–75)	63 (38–78)	65 (51–72)
Sex
Male	65	54.2	35 (51.5%)	24 (57.1%)	6 (60.0%)
Female	55	45.8	33 (48.5%)	18 (42.9%)	4 (40.0%)
Histology
Adenocarcinoma	78	65.0	48 (70.6%)	25 (59.5%)	5 (50.0%)
Squamous cell	36	30.0	17 (25.0%)	15 (35.7%)	4 (40.0%)
Large cell	6	5.0	3 (4.4%)	2 (4.8%)	1 (10.0%)
Molecular Alterations (among adenocarcinoma, n = 78)	(n = 78)
EGFR mutation	15/78	19.2%	8/48 (16.7%)	6/25 (24%)	1/5 (20%)
ALK rearrangement	3/78	3.8%	1/48(2.1%)	2/25(8%)	0/5
KRAS mutation	12/78	15.4%	7/48(14.6%)	4/25(16%)	1/5(20%)
Other/Unknown	48	61.5%	32/48(66.7%)	13/25(52%)	3/5(60%)
Smoking Status
Never	38	31.7	24 (35.3%)	11 (26.2%)	3 (30.0%)
Current/Former	82	68.3	44 (64.7%)	31 (73.8%)	7 (70.0%)
Adjuvant Therapy
Total	92	76.7	45 (66.2%)	37 (88.1%)	10 (100%)
Chemotherapy	64	53.3	28 (41.2%)	28 (66.7%)	8 (80.0%)
Targeted Therapy	20	16.7	12 (17.6%)	6 (14.3%)	2 (20.0%)
Immunotherapy	28	23.3	5 (7.4%)	18 (42.9%)	5 (50.0%)
No Adjuvant Therapy	28	23.3	23 (33.8%)	5 (11.9%)	0 (0%)

Relationship between TCR diversity and clonality metrics

We assessed the relationship between the two primary TCR repertoire metrics to determine their independence and complementary value. As expected, TCR diversity and clonality showed a strong inverse correlation (Spearman ρ = -0.82, P < 0.001; Pearson r = -0.79, P < 0.001), explaining 62% of shared variance (R² = 0.62) (Supplementary Fig. 1). This strong association reflects their mathematical relationship as complementary measures of repertoire architecture. However, despite this correlation, each metric provided unique clinical insights: diversity showed stronger associations with pathological stage (ρ = -0.62 vs. ρ = 0.41 for clonality), while clonality demonstrated superior prognostic performance for survival outcomes (AUC = 0.72 vs. 0.65 for diversity in 2-year OS prediction). The variance not explained by their correlation (38%) suggests each metric captures distinct aspects of TCR repertoire biology with differential clinical relevance.

Associations with pathological stage and treatment modality

Formal correlation analyses confirmed strong associations between pathological stage and TCR repertoire metrics. TCR diversity showed a significant inverse correlation with higher pathological stage (Spearman ρ = -0.62, P < 0.001), while clonality demonstrated a positive correlation (ρ = 0.41, P < 0.001). The effect size for stage on diversity was substantial (Cohen’s d = 0.85 for I-II vs. III-IV comparison), indicating clinically meaningful differences across stage groups (Supplementary Fig. 2A and B).

Treatment modality also significantly associated with TCR metrics (Kruskal-Wallis P = 0.002 for diversity). Immunotherapy recipients exhibited higher TCR diversity compared to other treatment groups (Cohen’s d = 0.72, P = 0.008), consistent with selection of patients with more robust immune profiles for immune checkpoint inhibition. These associations underscore the importance of multivariate adjustment for both stage and treatment in evaluating TCR metrics as independent prognostic biomarkers (Supplementary Fig. 2C and D).

Survival analysis by pathological stage

To validate the prognostic performance of established staging criteria in our cohort, we first analyzed survival outcomes stratified by AJCC 8th edition pathological stage. As expected, significant differences were observed across stage groups for both Disease-free survival (DFS; log-rank P < 0.001) and overall survival (OS; log-rank P < 0.001) (Fig. 1). Patients with stage I-II disease demonstrated the most favorable outcomes, with median DFS of 45.2 months and median OS of 70.8 months. Intermediate-stage (III) patients showed reduced survival (median DFS: 25.1 months; median OS: 40.3 months), while stage IV patients had the poorest prognosis (median DFS: 15.8 months; median OS: 25.6 months). These results confirm the expected prognostic gradient of pathological staging in our cohort and provide a benchmark against which to evaluate the performance of TCR-based biomarkers.

Fig. 1.

Survival outcomes by pathological stage. A Progression-free survival and (B) overall survival stratified by AJCC 8th edition pathological stage in 120 NSCLC patients. Stage I-II (green, n=68), stage III (orange, n=42), and stage IV (red, n=10) showed significant differences in both DFS (log-rank P < 0.001) and OS (log-rank P < 0.001). Tables show number of patients at risk at indicated time points. Number of events by stage: DFS - Stage I-II: 22/68, Stage III: 19/42, Stage IV: 7/10; OS - Stage I-II: 18/68, Stage III: 16/42, Stage IV: 6/10

TCR-stage correlation

Preoperative TCR repertoire diversity demonstrated a significant inverse association with pathological disease progression. The Shannon diversity index was markedly lower in advanced-stage (III–IV) patients (median 5.2, IQR: 4.1–6.0) compared to early-stage (I–II) cases (median 6.8, IQR: 5.9–7.4; P = 0.003, Wilcoxon test) (Fig. 2A). Conversely, TCR clonality exhibited a strong positive correlation with higher TNM stage (Spearman’s ρ = 0.41, 95% CI: 0.27–0.54, P < 0.001), with stage IV patients showing the highest median clonality (0.49 vs. 0.21 in stage I, P = 0.001) (Fig. 2B). These findings remained significant after adjusting for age, sex, and smoking status in multivariate linear regression (β = -0.33 for diversity, P = 0.007; β = 0.38 for clonality, P < 0.001).

Fig. 2.

Association between TCR repertoire features and pathological stage. A Comparison of TCR repertoire diversity (Shannon index) between early-stage (I-II, n = 68) and advanced-stage (III-IV, n = 52) NSCLC patients. Boxes represent interquartile range (IQR), horizontal lines indicate medians, whiskers extend to 1.5×IQR, and dots show individual values. Diversity was significantly reduced in advanced-stage disease (median 5.2 vs. 6.8; P = 0.003, Mann-Whitney U test). B Correlation between TCR clonality and pathological stage (I: n = 32, II: n = 36, III: n = 42, IV: n = 10). Dots represent individual patients, with the red line showing the linear regression fit. Clonality exhibited a significant positive correlation with higher pathological stage (Spearman’s ρ = 0.41, P < 0.001)

PD-L1 expression and association with TCR metrics

PD-L1 expression was evaluable in all 120 patients, with distribution of 40.0% (48/120) negative (< 1%), 35.0% (42/120) low (1–49%), and 25.0% (30/120) high (≥ 50%) expression. PD-L1 high expression was more frequent in advanced-stage (III-IV) compared to early-stage (I-II) disease (60.0% vs. 40.0%, P = 0.02). As expected, PD-L1 status significantly influenced treatment selection, with immunotherapy administered to 46.7% (14/30) of PD-L1 high patients compared to 8.3% (4/48) of PD-L1 negative patients (P < 0.001) (Table 2).

Table 2.

PD-L1 expression distribution and association with clinicopathological features

Characteristic	Total (n = 120)	PD-L1 Negative (< 1%) (n = 48)	PD-L1 Low (1–49%) (n = 42)	PD-L1 High (≥ 50%) (n = 30)	P-value
Age, median (IQR)	62 (54–68)	63 (55–69)	61 (53–67)	60 (52–66)	0.45
Sex					0.32
Male	65 (54.2%)	28 (58.3%)	21 (50.0%)	16 (53.3%)
Female	55 (45.8%)	20 (41.7%)	21 (50.0%)	14 (46.7%)
Pathological Stage					0.02
I-II	68 (56.7%)	32 (66.7%)	24 (57.1%)	12 (40.0%)
III-IV	52 (43.3%)	16 (33.3%)	18 (42.9%)	18 (60.0%)
Histology					0.18
Adenocarcinoma	78 (65.0%)	28 (58.3%)	30 (71.4%)	20 (66.7%)
Squamous	36 (30.0%)	16 (33.3%)	10 (23.8%)	10 (33.3%)
Large cell	6 (5.0%)	4 (8.3%)	2 (4.8%)	0 (0%)
TCR Diversity, median (IQR)	6.1 (4.8–7.2)	5.8 (4.5–6.9)	6.2 (5.1–7.3)	6.5 (5.4–7.6)	0.04
TCR Clonality, median (IQR)	0.28 (0.18–0.42)	0.32 (0.22–0.45)	0.27 (0.17–0.40)	0.24 (0.15–0.36)	0.03
Treatment Received					< 0.001
Immunotherapy	28 (23.3%)	4 (8.3%)	10 (23.8%)	14 (46.7%)
Other/None	92 (76.7%)	44 (91.7%)	32 (76.2%)	16 (53.3%)

Notably, PD-L1 expression showed significant correlations with TCR repertoire metrics. Higher PD-L1 tumor proportion scores correlated with increased TCR diversity (Spearman’s ρ = 0.29, P = 0.002) and reduced clonality (ρ=-0.26, P = 0.005). PD-L1 expression categories alone did not show significant survival differences in our cohort (log-rank P = 0.12), though this may reflect the modest sample size and heterogeneous treatment approaches.

Survival analysis

With a median follow-up of 36 months (range: 6–48), 45 patients (37.5%) experienced disease progression, and 32 (26.7%) died. Kaplan-Meier analysis demonstrated significantly inferior outcomes in patients with high TCR clonality (≥ 0.35, optimal cutpoint):

• Disease-free survival (DFS): Median DFS 16 months (95% CI: 12–21) vs. 34 months (95% CI: 28–NR) in low-clonality group (HR = 2.1, 95% CI: 1.3–3.4; log-rank P = 0.002) (Fig. 3A).

• Overall survival (OS): Median OS 28 months (95% CI: 22–35) vs. not reached (HR = 2.4, 95% CI: 1.4–4.2; log-rank P = 0.001) (Fig. 3B).

Fig. 3.

Survival outcomes stratified by TCR clonality. A Disease-free survival (DFS) and (B) overall survival (OS) in 120 NSCLC patients stratified by preoperative TCR clonality (cutoff = 0.35). Patients with high TCR clonality (≥ 0.35, red curve, n = 60) showed significantly worse DFS (median 16.4 months vs. 34.2 months; HR = 2.1, 95% CI: 1.3–3.4; log-rank P < 0.001) and OS (median 28.7 months vs. 54.3 months; HR = 2.4, 95% CI: 1.4–4.2; log-rank P < 0.001) compared to patients with low TCR clonality (< 0.35, blue curve, n = 60). Number of events: DFS - High: 35/60, Low: 20/60; OS - High: 28/60, Low: 14/60. Tables below each plot show the percentage of patients surviving at indicated time points (0, 12, 24, 36, and 48 months post-surgery)

In multivariate Cox regression (Table 3), high TCR clonality remained an independent predictor of worse OS (adjusted HR = 1.9, 95% CI: 1.1–3.3, P = 0.02) after controlling for age, stage, and treatment modality. Similar results were observed for DFS (adjusted HR = 1.7, 95% CI: 1.0–2.8, P = 0.04).

Table 3.

Multivariate cox regression for overall survival

Variable	Adjusted HR	95% CI	P-value
High TCR clonality	1.9	1.1–3.3	0.02
Stage III–IV	2.8	1.6–4.9	< 0.001
Age > 65 years	1.5	0.8–2.7	0.21
Immunotherapy	0.6	0.3–1.1	0.08

Prognostic performance

Time-dependent ROC analysis evaluated the discriminatory capacity of TCR metrics for overall survival (Fig. 4). TCR clonality demonstrated superior prognostic performance compared to clinical variables:

Fig. 4.

Time-dependent ROC analysis for 2-year overall survival prediction. Receiver operating characteristic (ROC) curves comparing the prognostic performance of preoperative biomarkers for 2-year overall survival in 120 NSCLC patients. Analysis includes all patients with complete survival data (n = 120 for TCR clonality, n = 120 for pathological stage, n = 120 for PD-L1 status). TCR clonality (blue curve; AUC = 0.72, 95% CI: 0.63–0.81) demonstrated superior discrimination compared to pathological stage (orange curve; AUC = 0.65, 95% CI: 0.55–0.75) and PD-L1 status (green curve; AUC = 0.58, 95% CI: 0.47–0.69). The combined model integrating TCR clonality with pathological stage (red curve) achieved the highest predictive accuracy (AUC = 0.69, 95% CI: 0.71–0.87). Diagonal gray line represents reference performance of random guessing

• 2-year OS prediction: AUC = 0.72 (95% CI: 0.63–0.81).

• 3-year OS prediction: AUC = 0.68 (95% CI: 0.58–0.78).

TCR clonality outperformed both pathological stage (AUC 2-year = 0.65, 95% CI: 0.55–0.75) and PD-L1 status (AUC 2-year = 0.58, 95% CI: 0.47–0.69). The combination of TCR clonality with pathological stage further improved prognostic accuracy (AUC 2-year = 0.79, 95% CI: 0.71–0.87).

CDR3 length distribution

Analysis of CDR3 amino acid length distribution revealed a Gaussian-like pattern in early-stage (I–II) patients, with a modal length of 14–15 residues accounting for 35.2% ± 4.1% of sequences. In contrast, advanced-stage (III–IV) patients exhibited a skewed distribution toward shorter CDR3 lengths (modal 12–13 residues, 28.7% ± 3.8%), indicative of potential antigen-driven clonal selection (P = 0.002, Kolmogorov-Smirnov test). No significant differences were observed in overall CDR3 length variance between stages (F-test, P = 0.12). These patterns correlated moderately with clonality scores (Spearman’s ρ = 0.42, P = 0.008), suggesting that restricted CDR3 lengths may reflect oligoclonal expansions in progressive disease (Fig. 5).

Fig. 5.

CDR3β length distribution patterns in NSCLC. A Density distributions of CDR3β amino acid lengths in early-stage (I-II, blue, n = 68) and advanced-stage (III-IV, red, n = 52) NSCLC patients. Early-stage patients show a Gaussian-like distribution with a mode at 14–15 amino acids (aa), accounting for 35.2% ± 4.1% of sequences. Advanced-stage patients exhibit a left-skewed distribution with a mode at 12–13 aa (28.7% ± 3.8%). Statistical comparisons: Kolmogorov-Smirnov test, P = 0.002; F-test for variance, P = 0.12. B Correlation between modal CDR3β length and clonality score across all patients (n = 120). Each point represents one patient, colored by CDR3β length. A significant negative correlation was observed (Spearman’s ρ = -0.42, P = 0.008). The dashed line indicates linear regression fit

Kaplan-Meier analysis by CDR3 length tertiles

Patients were stratified into tertiles based on mean CDR3 amino acid length: low (< 13 residues, n = 40), medium (13–15 residues, n = 40), and high (> 15 residues, n = 40). Kaplan-Meier analysis revealed significant differences in survival outcomes across these groups. The low CDR3 length tertile was associated with inferior progression-free survival (median DFS: 18 months) compared to medium (median: 28 months) and high tertiles (median: not reached; log-rank P = 0.004). Similarly, overall survival was poorest in the low tertile (median OS: 32 months) versus medium (median: 45 months) and high (median: not reached; log-rank P = 0.007). Multivariate Cox regression confirmed the independent prognostic value of low CDR3 length (adjusted HR for DFS: 2.1, 95% CI: 1.3–3.4, P = 0.002; adjusted HR for OS: 1.9, 95% CI: 1.2–3.1, P = 0.006), after controlling for age, sex, smoking status, stage, and treatment (Fig. 5). These findings suggest that restricted CDR3 lengths, potentially indicative of antigen-driven clonal selection and T-cell exhaustion, correlate with worse postoperative outcomes in NSCLC (Fig. 6).

Fig. 6.

Kaplan-Meier survival analysis by CDR3 length tertiles. A Disease-free survival (DFS) and (B) overall survival (OS) curves stratified by CDR3β amino acid length tertiles in 120 NSCLC patients. Patients were categorized into: Low (< 13 residues, n = 40, red), Medium (13–15 residues, n = 40, green), and High (> 15 residues, n = 40, blue) groups based on mean CDR3 length. The low CDR3 length group showed significantly inferior outcomes, with median DFS of 18 months (vs. 28 months for medium and not reached for high tertile; log-rank P = 0.004) and median OS of 32 months (vs. 45 months for medium and not reached for high tertile; log-rank P = 0.007). Shaded areas represent 95% confidence intervals. NR = not reached. Multivariate Cox regression confirmed CDR3 length as an independent prognostic factor after adjusting for age, sex, smoking status, and tumor stage (DFS: adjusted HR = 2.1, 95% CI:1.3–3.4, P = 0.002; OS: adjusted HR = 1.9, 95% CI:1.2–3.1, P = 0.006). Number of events: DFS - Low: 28/40, Medium: 19/40, High: 12/40; OS - Low: 24/40, Medium: 16/40, High: 10/40

Clonotype and clustering analysis

Hierarchical clustering of TCRβ repertoires segregated patients into two distinct subgroups (Fig. 7A). Cluster 1 (n = 73, 60.8%) was significantly enriched for early-stage (I–II) disease (82.2% vs. 31.9% in Cluster 2, P < 0.001) and exhibited higher TCR diversity (median Shannon index 6.9 vs. 4.8, P < 0.001). Conversely, Cluster 2 (n = 47, 39.2%) demonstrated dominant oligoclonal expansions and contained 89.4% of advanced-stage (III–IV) cases.

Fig. 7.

TCR repertoire clustering and stage-specific clonotypes. A Hierarchical clustering of preoperative TCRβ repertoires based on clonotype presence/absence (Euclidean distance, Ward’s linkage). The heatmap and dendrogram reveal two distinct patient clusters: Cluster 1 (blue sidebar, n = 73) was predominantly early-stage (I-II) cases with higher repertoire diversity (median Shannon index 6.9 vs. 4.8, P < 0.001), while Cluster 2 (orange sidebar, n = 47) was enriched for advanced-stage (III-IV) disease (89.4% of cases) and exhibited oligoclonal dominance. B Frequency distribution of three TCRβ clonotypes significantly enriched in advanced-stage patients: TRBV12-3/J2-7 (24.1% vs. 2.9% in early-stage, P < 0.001), TRBV19/J1-2 (19.0% vs. 3.7%, P = 0.002), and TRBV28/J1-1 (21.4% vs. 5.9%, P = 0.004). These public clonotypes showed minimal representation in healthy controls (≤ 0.1%)

Analysis of stage-enriched clonotypes identified three public TCRβ sequences recurrently expanded in stage III–IV patients (Fig. 7B):

1. TRBV12-3/J2-7 (CASSYVGNTGELFF): 24.1% frequency in stage III–IV vs. 2.9% in stage I–II (P < 0.001)

2. TRBV19/J1-2 (CASSLAGGPYEQYF): 19.0% vs. 3.7% (P = 0.002)

3. TRBV28/J1-1 (CASSESGWDTQYF): 21.4% vs. 5.9% (P = 0.004)

These clonotypes showed minimal overlap with healthy control repertoires (≤ 0.1% frequency in 50 age-matched donors).

Treatment subgroup analysis

Stratified analysis by adjuvant therapy modality revealed differential associations between TCR repertoire features and survival outcomes (Fig. 8). Among patients receiving adjuvant immunotherapy (n = 28), a non-significant trend toward improved progression-free survival was observed in patients with high TCR diversity (Shannon index ≥ 6.0) compared to those with low diversity (median DFS: not reached vs. 18 months; HR = 1.00, 95% CI: 0.38–2.63, log-rank P = 0.967). This result should be interpreted with caution due to the small subgroup size (n = 28). In a multivariate analysis adjusted for pathological stage, the association remained non-significant (adjusted HR = 0.98, 95% CI: 0.36–2.64, P = 0.96), suggesting that the apparent effect may be confounded by the more advanced disease stage in the low-diversity group.

Fig. 8.

Association between TCR diversity and immunotherapy benefit. A Forest plot of progression-free survival hazard ratios (high vs. low TCR diversity) across adjuvant treatment subgroups. High TCR diversity (Shannon index ≥ 6.0) was significantly associated with improved Disease-free survival (DFS) in patients receiving immunotherapy (n = 28; HR = 0.32, 95% CI: 0.11–0.89; P = 0.03) but not in chemotherapy (n = 64, HR = 0.75, 95% CI: 0.47–1.20; P = 0.21) or targeted therapy (n = 20, HR = 0.92, 95% CI: 0.42–2.03; P = 0.84) subgroups. B Kaplan-Meier analysis for immunotherapy recipients stratified by TCR diversity. Patients with high diversity (blue curve, ≥ 6.0, n = 15) showed a median DFS that was not reached, compared to 18 months for patients with low diversity (red curve, < 6.0, n = 13). The difference was not statistically significant (HR = 1.00, 95% CI: 0.38–2.63; log-rank P = 0.967). Number of events: High diversity group: 7/15; Low diversity group: 9/13. The lack of significant association after adjustment for stage (adjusted HR = 0.98, P = 0.96) suggests the observed trend may be influenced by the distribution of pathological stages between the groups

Collinearity diagnostics and model stability

To assess potential multicollinearity between TCR metrics and established prognostic factors, we performed comprehensive collinearity diagnostics. Variance Inflation Factors (VIF) for all covariates in our multivariate Cox models were below the concerning threshold of 5 (range: 1.2–3.8), indicating no problematic multicollinearity (Supplementary Fig. 3). Sensitivity analyses demonstrated stable hazard ratio estimates for TCR clonality across different model specifications: the hazard ratio changed by only + 12.3% when pathological stage was removed from the model (from HR = 1.9 to HR = 2.1) and by + 8.4% in the TCR-only model (HR = 2.0). Similarly, stage hazard ratios showed minimal change (-6.7%) when TCR clonality was excluded. This stability confirms the robustness of our findings and supports the inclusion of both TCR metrics and traditional prognostic factors in multivariate models.

Stratified analysis by pathological stage

To determine whether TCR clonality predicts outcomes independently within stage groups, we performed stratified Cox analyses. TCR clonality remained a significant predictor of progression-free survival in both early-stage (I-II: HR = 2.1, 95% CI: 1.3–3.4, P = 0.002) and advanced-stage (III-IV: HR = 2.4, 95% CI: 1.5–3.8, P = 0.001) disease (Supplementary Fig. 4). The effect size was slightly larger in advanced stages, though the stage × TCR clonality interaction term was not statistically significant (P = 0.18). These findings demonstrate that the prognostic value of TCR clonality is maintained within stage subgroups, supporting its utility across the disease spectrum rather than being driven solely by stage association.

CDR3 length correlation and stratified analyses

CDR3 length demonstrated significant correlation with pathological stage (Spearman ρ = -0.48, P < 0.001), with shorter lengths observed in advanced-stage disease (13.2 ± 1.4 vs. 14.5 ± 1.2 amino acids in early stage, P < 0.001). Sensitivity analyses confirmed the stability of CDR3 length hazard ratios across different model specifications, with HRs for short CDR3 length ranging from 1.8 to 2.1 (Supplementary Fig. 5). Stratified analyses revealed that short CDR3 length predicted worse progression-free survival in both early-stage (HR = 1.9, 95% CI: 1.3–2.8, P = 0.002) and advanced-stage (HR = 2.2, 95% CI: 1.5–3.2, P = 0.001) disease. Collinearity diagnostics showed no concerning multicollinearity (all VIF < 5), supporting CDR3 length as an independent prognostic factor complementary to TCR clonality.

Three-year survival predictive performance

Time-dependent ROC analysis for 3-year overall survival demonstrated superior discriminatory performance for combined models integrating TCR metrics with pathological staging. The comprehensive model achieved the highest predictive accuracy (AUC = 0.79, 95% CI: 0.71–0.87), outperforming pathological stage alone (AUC = 0.65, ΔAUC = + 0.14) and TCR clonality alone (AUC = 0.72, ΔAUC = + 0.07). TCR clonality showed better discrimination than traditional staging (AUC = 0.72 vs. 0.65), while TCR diversity (AUC = 0.68) and CDR3 length (AUC = 0.66) provided moderate predictive value (Supplementary Fig. 6). At optimal cutpoints, the combined model achieved 78% sensitivity and 75% specificity for 3-year survival prediction, supporting the integration of TCR repertoire profiling with conventional staging for enhanced risk stratification.

Comprehensive multivariable cox regression analyses

Multivariable Cox regression analyses confirmed the independent prognostic value of TCR repertoire metrics after comprehensive adjustment for clinical covariates (Table 4). TCR clonality remained significantly associated with worse progression-free survival (HR = 1.79, 95% CI: 1.38–2.32, P < 0.001) and overall survival (HR = 1.88, 95% CI: 1.44–2.45, P < 0.001) in fully adjusted models. Complete regression tables for all analyses, including CDR3 length tertile models, stage-stratified analyses, and sensitivity models are provided in Supplementary Tables 1–4.

Table 4.

Key multivariable cox regression models

Model	Endpoint	TCR Clonality HR (95% CI)	P-value	Stage HR (95% CI)	P-value	Concordance
Primary	DFS	1.79 (1.38–2.32)	< 0.001	1.73 (1.32–2.27)	0.008	0.712
Primary	OS	1.88 (1.44–2.45)	< 0.001	1.82 (1.38–2.40)	0.006	0.705

Treatment-stratified analyses

Treatment-stratified analyses revealed differential prognostic value of TCR metrics across therapeutic modalities. TCR clonality demonstrated the strongest association with outcomes in immunotherapy recipients (HR = 2.35, 95% CI: 1.52–3.63, P < 0.001), moderate association in chemotherapy patients (HR = 1.82, 95% CI: 1.28–2.59, P = 0.002), and weaker association in targeted therapy (HR = 1.45, 95% CI: 0.92–2.29, P = 0.08) (Supplementary Fig. 7). The TCR clonality × immunotherapy interaction term approached but did not reach statistical significance (HR = 1.29, P = 0.07). Immunotherapy patients exhibited more favorable TCR profiles with lower clonality (0.28 ± 0.12 vs. 0.32 ± 0.14 in chemotherapy, P = 0.03) and higher diversity (6.8 ± 1.0 vs. 6.1 ± 1.2, P = 0.02), consistent with selection of patients with robust immune profiles for immune checkpoint inhibition.

Discussion

This prospective cohort study establishes that preoperative T cell receptor (TCR) repertoire profiling provides clinically significant biomarkers for non-small cell lung cancer (NSCLC). We observed a significant inverse correlation between TCR diversity and higher pathological stage, with stage III-IV patients exhibiting 32% lower Shannon diversity compared to stage I-II counterparts (median 5.2 vs. 6.8, P = 0.003). More critically, elevated TCR clonality (≥ 0.35) independently predicted worse progression-free survival (adjusted HR = 1.7, 95% CI: 1.0–2.8) and overall survival (adjusted HR = 1.9, 95% CI: 1.1–3.3) after controlling for conventional prognostic factors including AJCC stage. The time-dependent ROC analysis further demonstrated TCR clonality’s superior prognostic accuracy for 2-year survival prediction (AUC = 0.72 vs. 0.65 for pathological stage). Notably, preserved TCR diversity identified immunotherapy beneficiaries who showed significantly prolonged DFS (median not reached vs. 18 months, P = 0.03), while hierarchical clustering revealed distinct repertoire architectures between early and advanced-stage disease, including stage-enriched public clonotypes (e.g., TRBV12-3/J2-7) present in > 20% of stage III-IV patients. These findings collectively suggest TCR profiling may have potential as a prognostic indicator for disease stratification and possibly as a predictive biomarker for therapy selection. However, these observations require confirmation in larger, prospective studies before clinical implementation. The observed correlations between PD-L1 expression and TCR repertoire metrics provide important biological context for our findings. Higher PD-L1 expression associated with increased TCR diversity and reduced clonality suggests that tumors with greater immune activation may support more polyclonal T-cell responses. While PD-L1 influenced treatment selection in our cohort (particularly the preferential use of immunotherapy in PD-L1 high patients), our multivariate analyses demonstrated that TCR metrics retained independent prognostic value after adjusting for both PD-L1 status and treatment modality. This suggests TCR profiling provides complementary information to PD-L1 testing for patient stratification. The strong inverse correlation between TCR diversity and clonality (ρ = -0.82) confirms their mathematical relationship as complementary measures of repertoire architecture. However, their differential performance in clinical associations suggests each metric captures distinct biological information. Diversity, reflecting the breadth of the T-cell repertoire, showed stronger correlations with pathological stage, potentially indicating global immune competence decline with disease progression. In contrast, clonality, measuring the dominance of expanded clones, demonstrated superior prognostic accuracy for survival outcomes, possibly reflecting the balance between effective anti-tumor responses and pathological clonal exhaustion. The 38% unexplained variance between these metrics supports their use as complementary rather than redundant biomarkers, with diversity potentially better reflecting immune reservoir capacity and clonality more sensitive to active antigen-driven expansions. The formal correlation analyses confirm robust associations between TCR repertoire metrics and both pathological stage and treatment modality. The strong stage-dependent patterns (ρ = -0.62 for diversity, ρ = 0.41 for clonality) validate TCR profiling as a sensitive indicator of disease progression. Importantly, the association between treatment modality and TCR metrics (particularly the higher diversity in immunotherapy recipients) highlights both potential confounding and biological plausibility, as patients selected for immunotherapy likely possess more favorable immune profiles. Our multivariate analyses that adjust for these factors demonstrate that TCR metrics provide independent prognostic value beyond these established associations. The stability of our multivariate models was confirmed through comprehensive collinearity diagnostics and sensitivity analyses. The low Variance Inflation Factors (all VIF < 5) and minimal changes in hazard ratios across different model specifications (< 15% variation) demonstrate that our findings are robust and not unduly influenced by correlations between TCR metrics and traditional prognostic factors. This methodological rigor strengthens the conclusion that TCR clonality provides independent prognostic information beyond established clinical variables. The stratified analyses confirming TCR clonality’s prognostic value within both early-stage (HR = 2.1, P = 0.002) and advanced-stage (HR = 2.4, P = 0.001) subgroups provide compelling evidence for its stage-independent utility. This finding is particularly noteworthy given the different biological contexts and treatment patterns across stage groups. The consistent performance suggests TCR clonality captures fundamental aspects of anti-tumor immunity that transcend traditional staging categories, potentially reflecting underlying immune competence rather than simply correlating with tumor burden. The comprehensive CDR3 length analyses revealed significant stage-dependent patterns and independent prognostic value. The correlation between shorter CDR3 lengths and advanced disease (ρ = -0.48) suggests potential biological mechanisms involving antigen-driven selection or T-cell exhaustion signatures. The consistent prognostic performance across stage subgroups and model specifications, coupled with low collinearity (all VIF < 5), supports CDR3 length as a robust biomarker that complements TCR clonality in risk stratification. These findings align with emerging evidence that CDR3 length distributions reflect underlying T-cell repertoire dynamics and may capture distinct aspects of anti-tumor immunity. The superior 3-year survival prediction achieved by combined models integrating TCR metrics with pathological staging (AUC = 0.79 vs. 0.65 for stage alone) demonstrates the complementary value of immune profiling in prognostic assessment. The consistent outperformance of TCR clonality over traditional staging (AUC = 0.72 vs. 0.65) across multiple timepoints underscores its robustness as a prognostic biomarker. These findings suggest that preoperative TCR repertoire analysis substantially enhances conventional risk stratification, potentially enabling more personalized adjuvant therapy decisions based on integrated clinical and immunological risk assessment.

Clonality serves as a key indicator of T-cell exhaustion in advanced stages NSCLC, where persistent tumor antigen exposure drives the expansion of specific T-cell clones that progressively lose effector functions [26]. This clonal expansion is evident in NSCLC, where greater T cell clonality correlates with advanced disease, reflecting antigenic responses and impaired T-cell dynamics [26]. In advanced NSCLC, high clonality of tumor-infiltrating lymphocytes (TILs) is associated with terminally exhausted CD8 + T cells, as shown by TCR overlap analysis suggesting origins from circulating precursors with reduced functionality [27]. Increased clonality in peripheral PD-1 + CD8+ T cells predicts poorer clinical outcomes after immunotherapy, marking exhaustion in progressive disease [28]. Deep sequencing confirms the presence of CD8 + PD-1high T cells with an exhaustion phenotype in Chinese NSCLC patients, where high clonality signals advanced immunosuppressive states [28]. Tumor-associated Tregs obstruct antitumor immunity by promoting T cell dysfunction and clonal constriction of tumor-infiltrating CD8 + T cells, further mediating exhaustion in advanced NSCLC [29]. Overall, in advanced NSCLC, elevated clonality reflects T-cell exhaustion, with CD39 expression enriching for CD8 + T cells showing features of exhaustion, tumor reactivity, and clonal expansion [27].

Public clonotypes, which are shared TCR sequences across NSCLC patients, may represent expansions driven by common neoantigens, particularly in advanced disease involving chronic tumor stimulation [30]. In neoadjuvant anti-PD-1-treated NSCLC, MANA-specific T cell clones identified via functional expansion assays highlight neoantigen-driven public clonotypes [31]. Neoantigen-specific peripheral T cell clonotypes observed in high abundance post-treatment in NSCLC suggest shared public expansions from tumor neoantigens [32]. The PIK3CA public neoantigen is immunogenic and capable of driving T cell clonal expansion, leading to recurrent public clonotypes in NSCLC [30]. Neoantigenic stimulation achieves selective expansion of neoantigen-reactive TILs, broadening CD4 + and CD8 + public clonotypes in NSCLC [33]. Public TCR clonotypes associated with lung cancer, derived from shared CDR3β-Vβ sequences, indicate neoantigen-driven expansions across patients [34]. Enrichment of neoantigen-reactive TCR clonotypes in MSI-H NSCLC patients points to public expansions driven by tumor mutational heterogeneity [35]. NeoTCR signatures identify TCRs targeting driver neoantigens, suggesting common metastatic TIL exhaustion and public clonotype persistence in NSCLC [32]. Endogenous CD4 + T cells recognize neoantigens in NSCLC, with public clonotypes derived from patient blood reflecting neoantigen-driven responses [35].

Preoperative TCR profiling enables effective risk stratification in NSCLC by identifying patients at higher risk for advanced disease progression based on circulating TCR repertoire metrics [36]. In treatment-naïve NSCLC patients, pre-treatment peripheral blood TCR diversity correlates inversely with pathological stage, allowing for early identification of high-risk cases prior to surgical resection [36]. Circulating TCR repertoire analysis, including clonality and diversity indices, serves as a non-invasive tool to predict postoperative recurrence and stratify patients for intensified adjuvant therapy [37]. Dynamic changes in TCR repertoire post-concurrent chemoradiotherapy further refine risk stratification, identifying patients likely to benefit from consolidative immunotherapy based on preoperative baselines [38].

Pretreatment tissue TCR evenness associates with complete pathologic response to neoadjuvant immunotherapy, enabling preoperative selection of high-risk patients for alternative regimens [39]. Comprehensive T cell repertoire characterization in early-stage NSCLC reveals clinicopathologic associations, positioning preoperative profiling as a key tool for personalized risk-adapted management [40]. In a recent study conducted by Shirasawa et al. [41], post-chemoradiotherapy TCR diversity in peripheral PD-1 + CD8+ T-cells (measured by DE50 score) significantly predicted recurrence risk in unresectable NSCLC patients receiving durvalumab consolidation. Patients with high post-CRT DE50 (≥ 0.00137) had markedly longer progression-free survival (median NR vs. 9.6 months; HR = 0.11, *p*=0.02) and reduced recurrence compared to low-DE50 counterparts [41]. Specific TCR profiles predict clinical outcomes of adjuvant EGFR-TKI therapy, enabling risk stratification for EGFR-mutant NSCLC patients preoperatively [42]. Based on Nakahara et al. [19], in EGFR/ALK wild-type NSCLC, a significant reduction in peripheral TCR diversity 6 weeks post-anti-PD-1 therapy correlated with treatment response, longer DFS, and immune-related adverse events, while no such association existed in mutant subgroups. Predictive value of TCR Vβ-Jβ profile for adjuvant gefitinib supports its use in preoperative risk assessment for resected NSCLC [42].

Limitations and future directions

Our study establishes preoperative TCR repertoire profiling as a promising prognostic and predictive tool in NSCLC, however, several limitations warrant consideration. First, the single-country (Chinese) cohort may limit the generalizability of our findings across diverse ethnic populations, given known variations in TCR repertoire architecture influenced by HLA polymorphisms and regional environmental exposures. Second, although our sample size (n = 120) provided sufficient statistical power for primary endpoint analyses, it constrained robust subgroup evaluations particularly in the immunotherapy cohort (n = 28) and precluded stratified molecular analyses (e.g., by EGFR/ALK status). Third, heterogeneity in adjuvant treatment modalities (chemotherapy, targeted therapy, and immunotherapy) introduces potential confounding in survival analyses, despite multivariate adjustments.

Future research should prioritize validation in large, multi-center international cohorts to confirm the universality of our TCR-based risk stratification model. Functional characterization of the stage-enriched public clonotypes (e.g., TRBV12-3/J2-7) is essential; experiments using peptide-MHC tetramers or tumor-infiltrating lymphocyte coculture systems could elucidate their specificity toward shared tumor neoantigens or viral antigens. Additionally, longitudinal tracking of TCR dynamics during therapy may clarify causal relationships between repertoire restoration and treatment efficacy, particularly in immunotherapy contexts. Integrating TCR metrics with emerging biomarkers such as tumor mutational burden, microbiome signatures, or spatial transcriptomics could further refine predictive algorithms for personalized adjuvant strategies. While our study demonstrates a clear association between preoperative TCR repertoire and outcomes, we acknowledge the conceptual question regarding its immediate clinical utility in the specific context of a single pre-resection blood draw. For patients already scheduled for curative resection, a preoperative biomarker adds limited value if it merely mirrors information provided by established clinical staging. The true clinical potential of this approach likely lies in two key applications beyond a one-off preoperative test. First, the preoperative TCR profile could serve as a baseline for longitudinal monitoring during adjuvant therapy and follow-up. Dynamic changes in TCR diversity and clonality post-surgery may provide an early, sensitive readout of minimal residual disease or treatment efficacy, surpassing the resolution of imaging [41]. Second, this baseline profiling could be integrated with other biomarkers to create a composite risk score that more accurately identifies high-risk stage I-II patients who might benefit from intensified adjuvant therapy (e.g., immunotherapy in addition to chemotherapy), or conversely, de-escalation in low-risk patients. Furthermore, in the evolving landscape of lung cancer care, where neoadjuvant immunotherapy has become a standard for resectable stages, a preoperative liquid biopsy could be highly valuable for predicting response to neoadjuvant treatment and tailoring subsequent surgical and adjuvant strategies. Our study provides the essential foundational evidence that the preoperative TCR repertoire contains meaningful prognostic information; future work must now focus on validating its utility in these dynamic and integrated clinical scenarios.

Conclusion

This prospective cohort study provides evidence suggesting that preoperative peripheral blood TCRβ repertoire profiling provides robust prognostic and predictive biomarkers for NSCLC management. While these findings are promising, they should be considered exploratory. Our findings suggest that reduced TCR diversity and elevated clonality correlate strongly with higher pathological stage and poorer survival outcomes, independent of conventional clinical factors. The identification of stage-enriched public clonotypes suggests antigen-driven expansions in advanced disease, while the superior predictive accuracy of TCR clonality over traditional staging (AUC = 0.72 vs. 0.65 for 2-year OS) underscores its clinical utility. While an initial analysis suggested that preserved TCR diversity might identify patients who derive benefit from immunotherapy, this association was not significant after adjusting for pathological stage in our cohort. This highlights the challenges of analyzing small subgroups and the strong confounding effect of disease stage. The potential of TCR diversity as a predictive biomarker for immunotherapy in the adjuvant setting requires further validation in larger, prospective studies. These results establish that preoperative TCR repertoire profiling provides robust prognostic information. This supports its future development as a component of integrated risk models and, crucially, as a baseline for serial liquid biopsy monitoring after surgery. Implementing TCR profiling in this manner could ultimately optimize postoperative management and improve outcomes in NSCLC patients.

Abbreviations

AJCC

American Joint Committee on Cancer

AUC

Area Under the Curve

CDR3

Complementarity-Determining Region 3

CI

Confidence Interval

CT

Computed Tomography

DNA

Deoxyribonucleic Acid

FDR

False Discovery Rate

HR

Hazard Ratio

ICI

Immune Checkpoint Inhibition

IQR

Interquartile Range

MHC

Major Histocompatibility Complex

NSCLC

Non-Small Cell Lung Cancer

OS

Overall Survival

PBMCs

Peripheral Blood Mononuclear Cells

PCR

Polymerase Chain Reaction

PD-1

Programmed Cell Death Protein 1

PD-L1

Programmed Death-Ligand 1

DFS

Disease-Free Survival

pCR

Pathological Complete Response

RECIST

Response Evaluation Criteria in Solid Tumors

ROC

Receiver Operating Characteristic

TCR

T-Cell Receptor

TILs

Tumor-Infiltrating Lymphocytes

TNM

Tumor-Node-Metastasis

Authors’ contributions

All the authors prepared and analyzed the data, interpreted the patient data, wrote the main manuscript text, and created the tables and figure. All authors read and approved of the final manuscript.

Funding

The authors acknowledge the financial support of the Second Hospital of Shanxi.

Data availability

The processed TCR sequencing datasets generated and analyzed during the current study—including clonotype tables, diversity metrics, and all related analysis files—are available in the Figshare repository under accession 10.6084/m9.figshare.30978886. The anonymized clinical metadata are also included in this repository. The TCRβ repertoire sequencing was performed using the ImmunoSEQ^® hsTCRB Assay (Adaptive Biotechnologies). The raw FASTQ sequence files are proprietary. As confirmed by the data provider (see Supplementary Documentation), our licensing agreement does not permit public deposition or redistribution of these raw files. For independent verification, qualified researchers may submit a request to Adaptive Biotechnologies Client Services at customercare@adaptivebiotech.com; such inquiries are subject to review and a Data Transfer Agreement. Any other data supporting the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This prospective study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of the Second Hospital of Shanxi Medical University (Protocol No. 2025YX314). Written informed consent was obtained from all individual participants prior to their enrollment in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.