A signature of plasma exosomal miRNAs predict therapeutic efficacy to neoadjuvant immunotherapy in patients with non-small cell lung cancer

Patient characteristics

The study included a prospective discovery cohort (n = 20) and a retrospective training cohort (n = 48) receiving neoadjuvant immunochemotherapy (Supplementary Figure S1A). The details of patient characteristics were shown in Supplementary Table S1. The median follow-up duration for the training cohort was 26.6 months.

Exosmoal miRNA-seq in the discovery cohort

Exosomes were isolated using ultracentrifugation and characterised by transmission electron microscopy, particle size analysis, and western blotting for specific markers, including TSG101 and CD81 (Supplementary Figure S1B). Total RNA was extracted from the purified exosomes, and miRNA sequencing was performed on samples from the discovery cohort and healthy volunteers. Differentially expressed miRNAs (DEmiRNAs) were analysed across three comparisons: pre-treatment versus healthy samples, pre-treatment versus post-treatment samples in pCR patients, and pre-treatment samples between pCR and non-pCR patients. 103 DEmiRNAs (P < 0.05, fold change > 1.4) were identified, with 44 miRNAs upregulated and 59 downregulated in pre-treatment compared with healthy samples (Figure 1A). The top five upregulated and downregulated DEmiRNAs are listed in Supplementary Table S2. Forty-four DEmiRNAs were identified, including 16 upregulated and 28 downregulated in pre-treatment compared with post-treatment samples (Figure 1A). The top five upregulated and downregulated miRNAs are presented in Supplementary Table S3. Twenty DEmiRNAs were identified, with 18 upregulated and 2 downregulated in pCR compared with non-pCR samples (Figure 1A). The top five upregulated and two downregulated DEmiRNAs are shown in Supplementary Table S4. Principal component analysis (PCA) revealed that DEmiRNAs effectively distinguished pre-treatment from post-treatment samples and pCR from non-pCR samples (Figure 1B). Interestingly, PCA could differentiate healthy from pre-treatment non-pCR samples but not from pCR samples (Figure 1B; Supplementary Figure S1C). A heatmap of DEmiRNA expression profiles is shown in Supplementary Figure S1D. The intersection of DEmiRNAs between healthy versus pre-treatment and pCR versus non-pCR samples identified let-7e-5p, miR-181a-2-3p, and miR-1271-5p were upregulated in healthy and pCR samples, while miR-22-3p and miR-589-5p were downregulated in both groups (Figure 1C). The intersection between healthy versus pre-treatment samples and pre-treatment versus post-treatment samples revealed miR-27a-3p, which was upregulated in pre-treatment samples and downregulated post-treatment, and miR-30b-3p, which was downregulated in pre-treatment samples and upregulated post-treatment (Figure 1D). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was conducted to identify the potential functions of the identified DEmiRNAs. The top 20 enriched pathways for the three miRNAs upregulated in healthy samples, two miRNAs upregulated in non-pCR samples, and miR-27a-3p are displayed as bar plots (Figures 1E-G). Seven enriched pathways for miR-30b-3p are shown in Figure 1H. Notably, the fatty acid biosynthesis pathway emerged as a common enriched pathway across the four KEGG analyses, highlighting a metabolic axis where tumour cells, immune cells, and exosomal miRNAs intersect to shape immunotherapy responses. Mechanistically, this pathway may drive resistance by fostering an immunosuppressive lipid-rich tumour microenvironment or sensitize tumour via ferroptosis susceptibility. Further experimental exploration is needed to clarify these interactions. If validated, targeting fatty acid biosynthesis could emerge as a strategy to augment neoadjuvant immunotherapy in NSCLC.

Figure 1

Differential expression, KEGG analyses of miRNAs and construction of the 5-miRNA signature for predicting pCR in discovery cohort. (A) Volcano plot showing the differentially DEmiRNAs in pre-treatment versus healthy group (left), pre-treatment versus post-treatment group (middle), and pCR versus non-pCR group (right). (B) PCA plot showing that DEmiRNAs could clearly stratify samples into non-pCR versus healthy group (left), pre-treatment versus post-treatment group (middle), and pCR versus non-pCR group (right). (C) Boxplot showing that let-7e-5p, miR-181a-2-3p, and miR-1271-5p up-regulated in healthy and pCR samples, while miR-22-3p and miR-589-5p down-regulated in healthy and pCR samples. (D) Boxplot showing that miR-27a-3p up-regulated in pre-treatment samples and down-regulated after neoadjuvant therapy, while miR-30b-3p down-regulated in pre-treatment samples and up-regulated after neoadjuvant therapy. (E) The top 20 pathways in KEGG analyses of 3 miRNAs which up-regulated in healthy and pCR samples. (F) The top 20 pathways in KEGG analyses of 2 miRNAs which down-regulated in healthy and pCR samples. (G) The top 20 pathways in KEGG analyses of miR-27a-3p. (H) The 7 pathways in KEGG analyses of miR-30b-3p. (I) Boxplot showing the significantly higher score in pCR patients than in the non-pCR patients. (J) The receiver operating characteristic curve of pCR score. (K) The calibration curve of pCR score. Predicted and actual pCR probability were respectively plotted on the X-axis and the Y-axis. The 45-degree dashed lines through the coordinate origin represent the excellent calibration models. (L) The distribution of AUC for 1000 times bootstrapping. (M) The DCA of the clinical value for the pCR score. (N) Event-free survival for patients with pCR and non-pCR. (O) Overall survival for patients with pCR and non-pCR. KEGG: Kyoto Encyclopedia of Genes and Genomes; PCA: principal component analysis; DEmiRNAs: differential expression miRNAs; PCA: principal component analysis; AUC: areas under the ROC; DCA: decision curve analysis.

Construction of 5 exosomal miRNA signatures for predicting pCR and prognostic stratification

The expression levels of five miRNAs (let-7e-5p, miR-181a-2-3p, miR-1271-5p, miR-22-3p and miR-589-5p) were measured using RT-qPCR in the training cohort. Least absolute shrinkage and selection operator (LASSO) regression selected all five miRNAs with the optimal log lambda value of -5.830026 (Supplementary Figure S1E). A pCR signature, named as “pCR score”, was constructed. The specific formula for the pCR score = miR-1271-5p*0.879 + let-7e-5p*0.490 + miR-181a-2-3p*0.888 – miR-589-5p*1.308 – miR-22-3p*0.358. As shown in Figure 1I, pCR patients exhibited significantly higher pCR scores. With an area under the receiver operating characteristic curve (AUC) of 0.826, the pCR score demonstrated satisfactory predictive performance (Figure 1J). The calibration curve demonstrated good agreement between the predicted and actual probabilities of pCR (Figure 1K). Internal bootstrap validation, performed with 1000 repetitions, yielded a median AUC of 0.813 (95% confidence interval [CI]: 0.764–0.829; Figure 1L). Additionally, decision curve analysis (DCA) indicated that using the pCR score for clinical decision-making provides a net benefit for patients undergoing neoadjuvant immunotherapy (Figure 1M). Based on a cut-off value of 2.60, derived from maximally selected log-rank statistics (Supplementary Figure S1F), patients were stratified into two groups: high pCR score (14 patients) and low pCR score (34 patients). Among patients with a high pCR score, only one experienced event-free survival (EFS) event and no patients had overall survival (OS) event, while 18 patients with a low pCR score experienced EFS events and 14 patients experienced OS events. As shown in Figure 1N, patients with a high pCR score had significantly longer EFS compared to those with a low pCR score (log-rank test P= 0.005). Similarly, patients with a high pCR score also had better OS than those with a low pCR score (log-rank test P = 0.047, Figure 1O). Though, the median follow-up duration of 26.6 months is shorter than the median EFS (31.6 months) in checkmate 816 trial, historical precedents suggest that early survival advantages in biomarker-stratified NSCLC cohorts often persist, particularly when linked to pathological response. Therefore, the pCR score represents a promising but preliminary tool for risk stratification. However, late recurrences may modestly alter survival curve and we would concentrate prolonged follow-up outcomes.

This study highlights exosomal miRNAs as promising biomarkers for predicting pCR in NSCLC patients receiving neoadjuvant immunotherapy. We identified significant changes in the expression of two miRNAs (miR-27a-3p and miR-30b-3p) following neoadjuvant immunotherapy. Furthermore, five miRNAs (let-7e-5p, miR-181a-2-3p, miR-1271-5p, miR-22-3p and miR-589-5p) were found to correlate strongly with pCR outcomes. The developed model for predicting pCR shows high accuracy and reliability, offering a valuable tool for patient selection and personalized treatment.