Paclitaxel (MCE, Shanghai, China), afatinib (MCE), rapamycin (RA; MCE), dorsomorphin (DO; MCE), and tunicamycin (TM; Selleck, Shanghai, China) were dissolved in dimethyl sulfoxide at stock concentrations of 10 mM, and aliquots were stored at -20 °C.

A549 (RRID: CVCL_0023) and H1299 (RRID: CVCL_0060) cells were obtained from the American Type Culture Collection (VA, USA) and cultured in RPMI-1640 (Gibco, MA, USA) supplemented with 10% fetal bovine serum (GeminiBio, CA, USA) at 37 °C with 5% CO₂ in a humidified incubator (Thermo Fisher Scientific, MA, USA). Paclitaxel-resistant cell lines were established through stepwise exposure to increasing concentrations of paclitaxel (12, 23, 59, and 117 nM) during an initial 28-day induction period, followed by continued selection with 117 nM paclitaxel for approximately three months. The cells were maintained at each concentration until tolerance was reached; if > 50% cell death occurred, paclitaxel was temporarily withdrawn to allow recovery. Cells exhibiting stable tolerance to 117 nM paclitaxel after selection for approximately four months were designated A549-TR and H1299-TR. Parental A549 and H1299 cells were derived from early-passage frozen stocks and used in experiments within 20 passages. Passage 1 resistant cell lines with confirmed paclitaxel resistance were maintained for stabilization and used within 15 passages after resistance was established. All human cell lines were authenticated via short tandem repeat (STR) profiling (Shanghai Fuheng Biotechnology Co., Ltd.) in November 2023. The STR profiles matched the reference profiles of the corresponding cell lines. Mycoplasma contamination was assessed using polymerase chain reaction (PCR) and monitored monthly.

The ST6GAL1 shRNA lentiviruses were purchased from GENEWIZ Company (Suzhou, China) with the following target sequences: shST6GAL1-1 (CGCTGCTCTATGAGAAGAATT) and shST6GAL1-2 (CCTAAGCATGAACAAGTACAA). The ST6GAL1 overexpression plasmid was constructed with the ST6GAL1 coding sequence (CCDS3285.1) and packaged into lentiviral particles by GENEWIZ. The cells were digested with 0.25% trypsin (Gibco, USA) and resuspended at 2-5 × 10⁴/mL. Next, 2 mL of the cell suspension was seeded into six-well plates. After overnight incubation, 20 μL of lentivirus and 2 μL of polybrene were added to the wells. Infected cells were selected using 2 μg/mL puromycin per well after 2-3 days.

3,000 cells were seeded in 96-well plates and incubated overnight. The medium was then replaced with TM-, RA-, DO-, or paclitaxel-containing media, followed by incubation for 48 h. Next, 10 μL of Cell Counting Kit-8 reagent (Selleck, Shanghai, China) was added to each well, followed by incubation for 2-4 h. Absorbance was then measured at 450 nm on a microplate reader (Agilent, VT, USA).

Cells were digested with 0.25% trypsin and resuspended in 1X phosphate-buffered saline (PBS) at a final density of 1 × 10⁵/mL. Next, 30 μL of cell suspension (3,000 cells per well) was seeded into Corning 24-well ultra-low-attachment plates and cultured in 500 μL of sphere-forming medium containing DMEM-F12 (Gibco), 1X B27 (Absin, Shanghai, China), 10 ng/mL epidermal growth factor (Absin), and 10 ng/mL basic fibroblast growth factor (Absin). These cells were then cultured at 37 °C for 5 days. In treatment assays, TM, RA, DO, and paclitaxel were added at the beginning of culture. At the fixed endpoint on day 5 after seeding, spheres with a diameter of > 50 μm were imaged and quantified using ImageJ. SFA was assessed by sphere count. The reported n values represent independent biological replicates.

Total RNA was extracted using TRIzol reagent (Sigma, MO, USA) and reverse-transcribed using a StarScript II first-strand cDNA synthesis kit (Yeason, Shanghai, China) following the manufacturer’s instructions. Reverse transcription quantitative polymerase chain reaction (RT-qPCR) was performed using ChamQ Universal SYBR qPCR Master Mix (Vazyme, Nanjing, China) with GAPDH as the reference gene. The RT-qPCR primer sequences are presented in Supplementary File 1.

Cells were washed with ice-cold phosphate-buffered saline and lysed in immunoprecipitation buffer [50 mM Tris-HCl (pH 7.4; Sangon Biotech, Shanghai, China), 150 mM NaCl (Sangon Biotech), 1% Triton X-100 (Sinopharm Chemical Reagent Co., Ltd., Beijing, China), and 1 mM EDTA (Sinopharm Chemical Reagent Co., Ltd.)], supplemented with protease inhibitors (Selleck). Lysates were incubated on ice for 45 min with gentle mixing, then centrifuged at 14,000 rpm (4 °C) for 10 min. The supernatants were collected, and protein concentrations were measured using a bicinchoninic acid assay (Beyotime, Shanghai, China). Equal amounts of protein were adjusted to the same volume with lysis buffer. Lysates were then precleared with 10 μL of protein A/G agarose beads (PR40025; Proteintech, Wuhan, China) for 30 min at 4 °C with rotation. After centrifugation, 100 μL of each supernatant was retained as input. 1 mg of each lysate was incubated overnight with 2 μg of anti-epidermal growth factor receptor (EGFR) antibody at 4 °C. Species-matched IgG (98136-1-RR; Proteintech) was used as the negative control. Washed protein A/G agarose beads were then added to the samples, followed by incubation for 4 h at 4 °C with rotation. The beads were collected via centrifugation, washed three times with ice-cold immunoprecipitation buffer, and resuspended in protein loading buffer. Proteins were eluted by heating, followed by Western blot analysis. Sambucus nigra agglutinin (SNA) lectin blotting was used to assess α2,6-linked sialylation of immunoprecipitated EGFR. Briefly, the membranes were incubated with biotinylated SNA (B-1305-2; Vector Laboratories, CA, USA) at 4 μg/mL, followed by incubation with Streptavidin-HRP Conjugate (HRP-PF00030; Proteintech) for signal detection. Band intensities were quantified using Lane 1D software (Beijing Sage Creation Science Co., Ltd., Beijing, China), and the SNA signal was normalized to the corresponding immunoprecipitated EGFR signal.

Total protein was extracted using a lysis buffer (X-Blot, Suzhou, China). The lysates were subsequently centrifuged at 14,000 rpm for 10 min at 4 °C. Proteins were then denatured, resolved using 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis, and transferred onto polyvinylidene difluoride membranes. Membranes were incubated overnight at 4 °C with primary antibodies, followed by incubation with horseradish peroxidase (HRP)-conjugated secondary antibodies for one hour at room temperature. Signal was developed using an enhanced chemiluminescence solution (X-blot), and bands were imaged on a Tanon Chemi Dog 5200T system (Tanon, Shanghai, China). Band intensities were quantified using Lane 1D software (SAGE, Beijing, China). Relative protein expression was determined by normalizing target signal intensities to GAPDH, β-actin, and β-tubulin. Because total mechanistic target of rapamycin (mTOR) and AMP-activated protein kinase (AMPK) levels varied in some assays, phosphorylated mTOR and AMPK were normalized to the corresponding loading controls, as described in previous studies^[24,25]. The antibodies used in this study are provided in Supplementary File 2.

Cells were dissociated into single-cell suspensions, resuspended in 1X PBS containing 2% fetal bovine serum, and 200 μL of the cell suspension was incubated with 4 μL of CD104-FITC (BioLegend, CA, USA), CD166-PE (BioLegend), or CD49f-APC (BioLegend) for 15 min at 4 °C in the dark. After centrifugation at 350 × g at 4 °C for 5 min, the cells were suspended in PBS, transferred to a flow cytometry tube, and analyzed on a CytoFLEX LX flow cytometer (Beckman Coulter, Germany). Gating analysis was conducted with FlowJo (BD, OR, USA).

Total RNA was extracted using TRIzol reagent (Sigma), and sequencing was outsourced to GENEWIZ. RNA sequencing was performed using one independent biological sample per group, comprising adherent A549-TR cells and A549-TR spheres. Differential gene expression analysis between adherent A549-TR cells and A549-TR spheres was conducted using the R package edgeR. Genes with counts per million > 2 in at least one sample were retained for differential expression analysis. Differentially expressed genes were defined as those with |log2 fold change| ≥ 1 and a Benjamini–Hochberg-adjusted false discovery rate (FDR) < 0.05 and were visualized on volcano plots (the complete gene list is provided in Supplementary File 3). Gene Ontology biological process (GO-BP), molecular function (GO-MF), and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were used to characterize the functional and pathway-level features of the differentially expressed genes, with the results provided in Supplementary File 3. Heatmap visualization and gene set enrichment analysis (GSEA) were conducted in RStudio.

RNA-seq data from 870 lung cancer patient samples were retrieved from The Cancer Genome Atlas (TCGA) database via the Genomic Data Commons Data Portal. The quantitative protein levels in lung cancer patients were downloaded from the cBio Cancer Genomics Portal^[26], while the protein levels in various cancer cell lines were obtained from the DepMap portal^[27]. RNA-seq data and normalized protein expression data were processed on RStudio. Signaling pathways and N-glycosylation indices derived from the 19 upregulated N-glycosylation-related genes were analyzed using the gene set variation analysis (GSVA) package, as shown in Supplementary File 4. Correlation analysis was performed using the Pearson method.

A total of 46 clinical lung tumor tissues were stained for ST6GAL1. The use of these clinical samples was approved by the Institutional Ethics Committee of Huadong Hospital (approval No. 20180046), and written informed consent was obtained from all patients. Antigen retrieval was conducted using the microwave-based heat-induced epitope retrieval method. Tissue microarrays were incubated overnight with primary antibodies against ST6GAL1 (Proteintech) at 4 °C, followed by incubation with HRP-conjugated secondary antibodies at 37 °C for 1 h. Signal was developed using HRP substrate 3,3′-diaminobenzidine for 2-5 min, followed by counterstaining with hematoxylin and mounting. Slides were digitized using a high-resolution whole-slide scanner (NanoZoomer series; Hamamatsu, Japan). Staining intensity was quantified based on the mean grayscale value using ImageJ and then converted into optical density values using the following formula: optical density = log₁₀$$ \frac{255}{gray\ value} $$. The ST6GAL1 staining intensity and corresponding cancer tissue clinical information are presented in Supplementary File 5.

Overall survival (OS) and disease-free survival (DFS) analyses of genes of interest and the seven-gene glycosylation signature were performed using GEPIA2. The “Signatures” option in GEPIA2 was selected for signature analysis, and ALG1, HEXA, MAN1C1, MGAT5, MOGS, ST6GAL1, and MGAT4A were entered as Gene Set A. The combined signature score was generated using GEPIA2’s signature module. Quartile-based expression cut-offs were applied, and hazard ratios (HRs) were estimated within the lung adenocarcinoma and lung squamous cell carcinoma datasets. Patients with signature scores above the 75th percentile were classified into the high-signature group, while those with scores below the 25th percentile were classified into the low-signature group. Patients with intermediate scores were not included in the high vs. low survival comparison. HRs and 95% confidence intervals (CIs) were calculated using the Cox proportional hazards model.

For the survival analysis of chemotherapy-treated patients, 14 patients were included and stratified into high- and low-ST6GAL1 expression groups based on the median ST6GAL1 OD value of the 46 tumor samples (high, n = 8; low, n = 6; Supplementary File 5). Patients received different chemotherapy regimens according to standard clinical treatment protocols, and the specific regimens are provided in Supplementary File 5. OS was defined as the time from surgery to death or last follow-up. Kaplan-Meier analysis was performed in R (version 4.5.2) using RStudio (version 2025.9.2.418), and survival differences between groups were assessed using the log-rank test. HRs and 95%CIs were estimated using Cox proportional hazards regression.

Gene-specific HRs, p(HRs), and 95%CIs for DFS were obtained from GEPIA2 and presented as forest plots generated in STATA version 16.0 (STATA Corporation, TX, USA). An HR > 1 indicated an association between higher gene expression and shorter DFS. For RT-qPCR data, gene-specific standardized mean differences (SMDs) and corresponding 95%CIs were calculated to quantify changes in the expression of stemness-associated genes. An SMD > 0 indicated increased expression relative to the reference group, whereas an SMD < 0 indicated decreased expression.

Data are presented as the mean ± standard error of the mean (SEM) from at least three independent replicates. Differences between two groups were compared using a two-sided Student’s t-test. Differences between multiple groups were compared using one-way analysis of variance (ANOVA), followed by Dunnett’s multiple comparisons. A P value < 0.05 indicated statistically significant differences. Error bars indicate mean ± SEM. n.s., *, and ** indicate P > 0.05, < 0.05, and < 0.01, respectively.

Paclitaxel-resistant NSCLC cell lines, A549-TR and H1299-TR, were established to investigate the molecular mechanism of acquired chemoresistance. The relative cell growth of the parental cells decreased in a dose-dependent manner upon exposure to paclitaxel, while resistant cells exhibited significantly less growth inhibition under the same treatment conditions (P < 0.05; Figure 1A and B), confirming the successful establishment of paclitaxel resistance. Notably, spheres derived from paclitaxel-resistant cells exhibited markedly higher chemoresistance compared with those derived from parental controls, with increased SFAs during paclitaxel treatment [Figure 1C-F]. The expression of stemness-associated transcription factors, including SOX2, NANOG, BMI1, KLF4, and c-MYC, was evaluated because of their established roles in cancer stemness^[28-30]. RT-qPCR analysis showed that the five genes were significantly more highly expressed in resistant spheres, and the consistently positive gene-specific effect estimates further supported this upregulation trend [Figure 1G and H]. Flow cytometric analysis revealed a marked increase in the stem-like CD104^-CD166⁺CD49f^hi subpopulation in resistant cells [Figure 1I and J]. Collectively, these results indicate that acquired paclitaxel resistance is characterized by an enrichment of CSLC subpopulations.

Consequently, RNA-seq analysis of A549-TR adherent cells and spheres was conducted to elucidate the molecular mechanisms underlying CSLC enrichment in paclitaxel-resistant NSCLC. Transcriptomic profiling revealed extensive gene expression changes in resistant spheres [Figure 2A]. Gene Ontology highlighted glycosylation-related pathways, including glycoprotein metabolism, protein glycosylation, and glycosyltransferase activity [Figure 2B and C]. KEGG and GSEA also revealed enrichment of N-glycan biosynthesis pathways, including those involving hsa00510 and hsa00513 [Figure 2D-F]. Hierarchical clustering analysis identified 19 N-glycosylation-associated differentially expressed genes in resistant spheres, all of which were upregulated [Figure 2G]. The upregulated 19-gene N-glycosylation signature was also elevated in lung tumor tissues compared with normal tissues [Figure 2H]. RT-qPCR validation analysis of the top 14 upregulated N-glycosylation-related genes based on RNA-seq found that ALG1, HEXA, MAN1C1, MGAT5, MOGS, ST6GAL1, and MGAT4A were consistently upregulated in A549-TR cells compared with parental cells, with further increases in A549-TR spheres [Figure 2I]. These results revealed a CSLC maintenance-associated seven-gene N-glycosylation signature in chemoresistant NSCLC.

Next, we explored the functional relevance of the seven N-glycosylation-related genes in paclitaxel resistance. Kaplan-Meier analysis revealed that the seven-gene signature was significantly associated with shorter DFS, but not with OS [Figure 3A and B]. The consistently higher gene-specific HR estimates for DFS further supported the association of these genes with poorer DFS [Figure 3C and Supplementary Figure 1]. To assess the functional contribution of N-glycosylation to resistance, we treated parental and paclitaxel-resistant cells with TM, a selective N-glycosylation inhibitor. At 0.1 μM, TM reduced relative cell growth by > 10% in parental cells (P < 0.01) but had a minimal effect (< 2%) on paclitaxel-resistant cells [Supplementary Figure 2A and B]. Notably, at 0.1 μM, TM significantly restored paclitaxel sensitivity in both A549-TR and H1299-TR cells (P < 0.01), which was enhanced at 0.5 μM TM, indicating a dose-dependent chemoresistance reversal [Figure 3D and E]. These findings indicated that N-glycosylation contributed functionally to chemoresistance in NSCLC.

Analysis of the effects of inhibiting N-glycosylation in spheres revealed that cotreatment with paclitaxel and 0.1 μM TM significantly restored the paclitaxel sensitivity of resistant spheres, whereas 0.5 μM TM alone nearly abolished SFA [Figure 3F-I]. Mechanistic analysis revealed that TM treatment did not significantly alter the mRNA levels of stemness-associated genes, consistent with the gene-specific effect estimates at both 0.1 and 0.5 μM TM [Figure 3J and K]. In contrast, the protein levels of SOX2, c-MYC, and BMI1 were markedly reduced following TM exposure (P < 0.05; Figure 3L and M). Flow cytometric analysis revealed a significant reduction in the CD104^-CD166⁺CD49f^hi subpopulation in TM-treated H1299-TR cells [Figure 3N and Supplementary Figure 2C]. These findings indicated that inhibiting N-glycosylation downregulated stemness-associated proteins and reduced CSLC populations, thereby restoring paclitaxel sensitivity in NSCLC cells.

To investigate intracellular signaling pathways potentially involved in N-glycosylation-related signaling, transcriptomic correlation analysis was performed using TCGA lung cancer datasets. N-glycosylation signaling was positively correlated with several oncogenic signaling pathways [Supplementary Figure 3A]. Pearson correlation analysis revealed that N-glycosylation was positively associated with mTOR (R = 0.46) and AMPK (R = 0.38) [Figure 4A and B]. N-glycosylation was also positively associated with the Notch, sphingolipid, HIF, ErbB, VEGF, and Foxa pathways, although the association was weaker [Supplementary Figure 3B-G]. Moreover, the levels of phospho-mTOR and phospho-AMPK in paclitaxel-resistant cells were elevated by approximately 50% higher than those in parental cells [Figure 4C and D, Supplementary Figure 3H and I] and significantly lower following TM treatment [Figure 4E and F]. These findings indicated that mTOR and AMPK signaling may be downstream of the N-glycosylation process in acquired paclitaxel chemoresistance.

Paclitaxel-resistant cells were treated with RA (mTOR inhibitor) or DO (AMPK inhibitor) alone or in combination with paclitaxel to evaluate the function of mTOR and AMPK signaling in paclitaxel resistance. RA reduced relative cell growth of A549-TR and H1299-TR cells, and only 10 μM RA significantly enhanced paclitaxel-induced growth inhibition, resulting in an additional reduction of approximately 20% relative to paclitaxel treatment alone [Figure 4G and H]. In contrast, 1 µM DO alone had no inhibitory effect, but it significantly restored paclitaxel sensitivity in resistant cells [Figure 4G and H]. In sphere assays, treatment with RA at 1 µM significantly impaired sphere formation and restored paclitaxel sensitivity in resistant spheres, enhancing SFA reduction by 15% [Figure 4I and J, Supplementary Figure 4A and B]. In contrast, treatment with DO at 1 µM selectively restored paclitaxel sensitivity, with a 25% decrease in SFA, without inhibiting SFA alone [Figure 4I and J, Supplementary Figure 4A and B]. We assessed the expression of stemness-associated genes to determine whether these effects involved CSLC-associated mechanisms. RT-qPCR analysis revealed no significant changes in mRNA levels after RA or DO treatment, consistent with the gene-specific effect estimates [Figure 4K and L]. However, Western blot analysis revealed that DO significantly reduced the levels of BMI1 and SOX2 [Figure 4M and N], whereas RA suppressed the expression of stemness markers, including c-MYC, BMI1, and SOX2 [Figure 4O and P]. These findings suggested that inhibition of mTOR or AMPK restored the sensitivity of paclitaxel-resistant spheres to paclitaxel by downregulating key stemness-associated proteins such as SOX2 and BMI1.

Several receptors implicated in cancer progression undergo glycosylation^[31,32], including EGFR, which is extensively N-glycosylated on the extracellular domain^[33]. Our analysis revealed that EGFR tyrosine kinase inhibitor resistance signaling (hsa01521) was enriched in spheres [Supplementary Figure 5A]. Moreover, N-glycosylation was positively associated with EGFR-related signaling at the transcriptomic and proteomic levels [Figure 5A and B], suggesting that EGFR signaling is involved in paclitaxel resistance. To explore whether EGFR is involved in paclitaxel resistance, EGFR expression was analyzed in CSLCs from paclitaxel-resistant cells. EGFR levels were higher in the CSLC subpopulation than in the non-CSLC subpopulation of paclitaxel-resistant cells [Figure 5C and Supplementary Figure 5B]. Notably, inhibiting N-glycosylation using TM significantly reduced EGFR expression in A549-TR cells, suggesting that N-glycosylation regulates EGFR expression in paclitaxel-resistant cells [Figure 5D and Supplementary Figure 5C]. The role of EGFR in paclitaxel resistance was further assessed by pharmacologically inhibiting EGFR using afatinib, which markedly reversed paclitaxel resistance in adherent cells [Figure 5E and F] and resensitized paclitaxel-resistant spheres to paclitaxel [Figure 5G-J]. However, at 1 μM, afatinib alone did not significantly inhibit spheres [Figure 5G-J]. Subsequent EGFR inhibition significantly decreased phospho-mTOR levels but did not significantly affect phospho-AMPK levels [Figure 5K and L], indicating that mTOR signaling rather than AMPK signaling was downstream of EGFR, which was consistent with the correlation analysis [Supplementary Figure 5D]. Furthermore, 10 μM afatinib significantly downregulated BMI1 and SOX2 [Figure 5M and N]. These findings indicated that N-glycosylation inhibition restored paclitaxel sensitivity by modulating the EGFR-mTOR-SOX2/BMI1 axis.

To identify the specific N-glycosylation-related genes involved in paclitaxel resistance, we screened for genes that were downregulated upon TM treatment and observed a significant decrease in HEXA, MGAT5, and ST6GAL1 levels [Figure 6A]. Among these candidates, ST6GAL1 showed the most pronounced enrichment in paclitaxel-resistant spheres and was decreased upon treatment with TM in a concentration-dependent manner, suggesting it is closely associated with the N-glycosylation-associated paclitaxel-resistant phenotype. Previous studies have shown that ST6GAL1 is involved in the regulation of chemoresistance and stemness in ovarian^[34] and colorectal^[22] cancers and is highly expressed in NSCLC^[35]. Based on our previous findings and other evidence, ST6GAL1 was selected for further functional validation. Notably, ST6GAL1 overexpression was confirmed by western blotting [Figure 6B], and it significantly increased phospho-mTOR levels [Figure 6C] and the expression of BMI1 and SOX2 [Figure 6D]. Based on functional assays, overexpressing ST6GAL1 did not markedly increase paclitaxel resistance in adherent parental cells [Supplementary Figure 6A and B]. However, ST6GAL1-overexpressing spheres exhibited significantly higher tolerance to paclitaxel [Figure 6E and F], indicating that the effect of ST6GAL1 was more evident in CSLC-enriched models. Furthermore, ST6GAL1 knockdown [Figure 6G] decreased phospho-mTOR levels [Figure 6H] and the expression of BMI1 and SOX2 [Figure 6I]. Consistent with the overexpression results, knockdown of ST6GAL1 did not reverse the paclitaxel resistance in adherent cells [Supplementary Figure 6C and D]. However, ST6GAL1 silencing significantly sensitized resistant spheres to paclitaxel, indicating that shST6GAL1 could reverse resistance of paclitaxel-resistant spheres [Figure 6J and K]. Moreover, ST6GAL1 knockdown significantly decreased EGFR expression in paclitaxel-resistant cells [Figure 6L and M]. SNA lectin blotting of immunoprecipitated EGFR further showed that ST6GAL1 overexpression increased EGFR α2,6-sialylation in H1299-ST6GAL1 cells [Figure 6N], whereas ST6GAL1 knockdown reduced EGFR α2,6-sialylation in A549-TR-shST6GAL1 cells [Figure 6O], supporting the role of ST6GAL1 in regulating EGFR α2,6-sialylation. Flow cytometric analysis further demonstrated a significant decrease in the CD104^-CD166⁺CD49f^hi population [Figure 6P and Q], indicating a reduced CSLC subpopulation upon ST6GAL1 knockdown. These findings suggested that ST6GAL1 contributes to CSLC-mediated paclitaxel resistance by promoting EGFR α2,6-sialylation and sustaining downstream mTOR-SOX2/BMI1 signaling, rather than directly inducing broad resistance in adherent bulk cells.

To evaluate the clinical relevance of ST6GAL1, we analyzed its mRNA levels in the TCGA lung cancer dataset and found that it was significantly elevated in patients with stage IV disease (P = 0.000194; Figure 7A). Immunohistochemical analysis revealed that ST6GAL1 expression was significantly higher in clinical lung cancer tissues than in adjacent non-tumor tissues [Figure 7B and C]. Moreover, ST6GAL1 levels were significantly higher in tumors from patients who had received chemotherapy than in those from untreated patients, indicating an association between ST6GAL1 expression and previous chemotherapy exposure [Figure 7D and E]. Survival analysis showed a lower survival probability among chemotherapy-treated patients with high ST6GAL1 expression (HR = 1.84, 95%CI: 0.57-5.90), although the difference was not statistically significant (log-rank P = 0.3; Figure 7F). To evaluate the broader clinical relevance of N-glycosylation-associated genes, DFS analysis was conducted across multiple cancer types. The seven-gene signature was significantly associated with DFS in patients with lung cancer (HR = 1.50, 95%CI: 1.07-2.11), whereas no significant associations were observed in patients with other cancer types [Figure 7G and Supplementary Figure 7]. These findings indicated that N-glycosylation-associated genes, particularly ST6GAL1, may have prognostic value in NSCLC.