An Unmet Clinical Need. Non-small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancers and has a five-year survival rate of only about 15%. For patients lacking actionable mutations, first-line treatment typically consists of platinum-based chemotherapy combined with radiotherapy. However, no reliable biomarkers currently exist to predict which patients will respond to this treatment -- a critical gap in personalized cancer care.
The Role of lncRNAs in Treatment Resistance. Long non-coding RNAs (lncRNAs) are RNA transcripts over 200 nucleotides long that regulate gene expression without encoding proteins. Aberrant lncRNA expression has been closely linked to NSCLC development, invasion, and metastasis. Multiple lncRNAs have been shown to mediate chemo- and radio-resistance by regulating DNA damage repair, apoptosis, and tumor stem cell activity.
Blood-Based Biomarker Potential. Tumor-derived exosomal lncRNAs are consistently detectable in human blood, offering a non-invasive sampling approach. Studies have demonstrated that blood exosomal lncRNAs can serve as biomarkers for early cancer detection, disease progression monitoring, and prognosis assessment. Plasma lncRNA testing is simpler and safer than repeat tissue biopsies, making it an attractive target for clinical biomarker development.
Radiomics as a Complementary Approach. CT-based radiomics extracts quantitative features from tumor imaging including morphology, density, intensity, and texture. Machine learning models built from these features have shown potential to predict individual patient responses to cancer therapy without requiring tissue sampling. Integrating plasma biomarkers with radiomics may produce more accurate predictive models than either approach alone.
Discovery Pipeline. Researchers enrolled 130 NSCLC patients from Hefei Cancer Hospital between 2021 and 2023. For the discovery phase, plasma exosomal RNA from 6 patients (3 treatment-sensitive, 3 treatment-resistant) was sequenced using next-generation sequencing. Differentially expressed lncRNAs were filtered using a log fold change above 0.5 and false discovery rate below 0.01, yielding 5,189 candidate lncRNAs.
Three-Layer Convergence Strategy. To identify the most clinically relevant candidates, researchers applied three independent filters: differential expression in patient plasma (5,189 lncRNAs), association with NSCLC patient survival in the TCGA database (430 lncRNAs), and differential expression between NSCLC normal and tumor tissues (203 lncRNAs). These clinical filters were then crossed with in vitro results from NSCLC cell lines in the GDSC database, where 15 lncRNAs were differentially expressed between cisplatin-sensitive and cisplatin-resistant lines.
MIF-AS1 Emerges. The intersection of all these filtered lists converged on a single candidate: the lncRNA MIF-AS1. In the TCGA database, higher MIF-AS1 expression was significantly associated with both poorer overall survival and worse progression-free survival in NSCLC patients. MIF-AS1 was upregulated in tumor versus normal tissue, and elevated in cisplatin-resistant compared to cisplatin-sensitive cell lines.
Validation in 124 Patients. RT-qPCR was used to measure plasma MIF-AS1 levels in 124 NSCLC patients before chemo- and radio-therapy. This independent clinical validation confirmed that MIF-AS1 expression was significantly elevated in the treatment-resistant group compared to the treatment-sensitive group, providing robust clinical evidence for the biomarker's predictive value.
MIF-AS1 Promotes Cancer Cell Aggressiveness. In vitro knockdown of MIF-AS1 using siRNA in two NSCLC cell lines (PC9 and H1299) significantly reduced cell proliferation, inhibited invasion in Matrigel assays, and decreased cell motility in scratch assays. These results confirm that MIF-AS1 plays an oncogenic role in NSCLC biology, promoting the aggressive behaviors that characterize treatment-resistant tumors.
MIF-AS1 Knockdown Sensitizes Cells to Cisplatin. Treating MIF-AS1-knockdown NSCLC cells with cisplatin (10 micrograms/mL) resulted in significantly increased apoptosis compared to control cells, as measured by Annexin V flow cytometry. This demonstrates that reducing MIF-AS1 expression removes a key resistance mechanism, restoring the cells' sensitivity to the chemotherapy drug.
The MIF-AS1/RAD21 Resistance Axis. To identify the molecular target mediating MIF-AS1's effects, researchers constructed a competing endogenous RNA (ceRNA) network. Among 1,650 target mRNAs identified through miRNA-mediated interactions, RAD21 was the only gene both present in the ceRNA network and significantly positively correlated with MIF-AS1 expression in TCGA NSCLC data. Knockdown of MIF-AS1 reduced RAD21 mRNA and protein levels in NSCLC cells.
RAD21's Role in Resistance. RAD21 encodes a core subunit of the cohesin complex, which is critical for DNA double-strand break repair, chromosome segregation, and prevention of inappropriate recombination. Higher RAD21 expression was significantly associated with worse NSCLC survival in TCGA. Prior studies showed RAD21 overexpression in cisplatin-resistant NSCLC and ionizing radiation-resistant liver cancer. The MIF-AS1/RAD21 axis thus promotes chemo- and radio-resistance by enhancing the cancer cell's ability to repair treatment-induced DNA damage.
Radiomics Feature Extraction. CT chest images from 38 NSCLC patients in the TCGA Cancer Imaging Archive were used for radiomics development. Tumor regions were manually delineated by a pulmonologist and thoracic surgeon, reviewed by a radiologist, and 1,409 radiomic features were extracted using AccuContour software. All features were standardized using Z-scores before feature selection.
Two Key Radiomic Predictors. LASSO regression identified two radiomic features associated with chemo- and radio-therapy response: 'correlation' (a texture feature) and 'mean' (an intensity feature). The correlation feature Z-score was significantly higher in treatment-sensitive patients, while the mean feature Z-score was significantly higher in treatment-resistant patients, indicating they capture distinct aspects of tumor biology relevant to treatment response.
Integrated Model Performance. A multivariate logistic regression model incorporating MIF-AS1 expression plus the two radiomic features was constructed and presented as a nomogram. In the TCGA training dataset, this integrated model achieved an AUC of 0.886. When validated in the independent cohort of 124 patients from Hefei Cancer Hospital, the model achieved an AUC of 0.808 -- substantially outperforming MIF-AS1 alone (AUC 0.724) and far exceeding the conventional biomarker CEA (AUC 0.556).
Clinical Utility Confirmed. Decision curve analysis (DCA) showed that the nomogram consistently yielded greater net clinical benefit than either treating all patients or treating no patients across a wide range of threshold probabilities, in both training and validation datasets. This confirms the model's practical value in guiding clinical treatment decisions beyond mere statistical discrimination.
MIF-AS1 Across Cancer Types. MIF-AS1 has been implicated in multiple cancer types beyond NSCLC. It was found upregulated in ovarian cancer and associated with poor prognosis; a MIF-AS1 polymorphism was linked to gastric cancer survival; and a MIF-AS1/miR-624-3p/HOXB8 ceRNA network was identified in breast cancer. This cross-cancer oncogenic significance reinforces MIF-AS1's potential as both a biomarker and a therapeutic target.
The Radiomics-Biomarker Integration Trend. This study joins a growing literature integrating plasma biomarkers with CT radiomics for cancer prediction. Prior work has shown that combining plasma miRNAs with radiological features improved lung cancer identification in indeterminate pulmonary nodules, and integrating DNA methylation biomarkers with CT characteristics improved malignant versus benign nodule classification. The MIF-AS1 plus radiomics model extends this paradigm to treatment response prediction specifically.
Superiority Over CEA. CEA (carcinoembryonic antigen) is a conventional tumor marker used to assess therapy response in NSCLC, but achieved an AUC of only 0.556 in this study -- essentially no better than random. MIF-AS1 alone achieved AUC 0.724, and the integrated model reached AUC 0.808. This demonstrates that lncRNA biomarkers combined with imaging data represent a meaningful advance over legacy blood tests for this clinical application.
MIF-AS1 Is Not Confounded by Clinical Features. Importantly, plasma MIF-AS1 levels were not significantly associated with age, sex, smoking status, tumor size, clinical stage, lymph node metastasis, distal metastasis, or tumor histology (adenocarcinoma versus squamous cell carcinoma). This clinical independence suggests MIF-AS1 captures a distinct biological signal -- likely related to DNA damage repair capacity -- that is not already captured by standard clinicopathological variables.
Single-Center and Sample Size Constraints. The study's primary limitation is that sample collection was confined to a single institution with a relatively small cohort. The TCGA radiomics training set contained only 38 patients, and the clinical validation cohort had 124 patients. While these numbers are sufficient for initial model development and validation, multicenter studies with larger samples are needed to confirm generalizability across different institutions, patient populations, and CT scanner protocols.
In Vitro Functional Validation Gap. The functional experiments demonstrating MIF-AS1's role in resistance were conducted in lung adenocarcinoma cell lines (PC9 and H1299) only. The study lacks in vitro validation in lung squamous cell carcinoma cell lines, which represent a major NSCLC subtype. Since the clinical cohort included both adenocarcinoma (74 patients) and squamous cell carcinoma (50 patients), functional validation in squamous cell lines is needed.
Molecular Mechanisms Still Incomplete. While the MIF-AS1/RAD21 axis has been identified as a key mechanism, the precise molecular pathways through which MIF-AS1 regulates RAD21 and promotes treatment resistance remain to be fully elucidated. Understanding whether this involves direct miRNA sponging, epigenetic regulation, or other mechanisms will be important for determining whether MIF-AS1 or RAD21 could serve as therapeutic targets.
A Promising Integrated Framework. Despite these limitations, this study establishes proof-of-concept for a non-invasive integrated prediction model combining plasma lncRNA measurement with CT radiomics. The model is particularly appealing because both components (blood draw and standard CT) are already part of routine NSCLC clinical care, making clinical implementation feasible once prospective multicenter validation is complete.