mRNA sequencing data from TCGA dataset was downloaded from the Genome Data General Database (GDC) data portal, which contains 553 patients with lung squamous cell carcinoma (LUSC) and 576 patients with lung adenocarcinoma (LUAD). Adjacent normal tissue samples were collected from patients with LUAD (n = 58) and LUSC (n = 51), and subsequently combined for downstream analyses. Clinical data were downloaded from the same source and matched to the processed lung TCGA data. Highly variable genes were selected based on the tool (http://pklab.med.harvard.edu/scw2014/subpop_tutorial.html). Upon doing calculations for estimates of variance and coefficient of variation of the bulk data, a total of 15252 genes were ranked based on the significance of deviation from the fit. The Wilcoxon test was used to identify differentially expressed genes (DEGs) in LUAD/LUSC compared to combined adjacent normal samples of LUAD and LUSC. DEGs were selected based on absolute binary logarithms of fold changes (Log₂FC) >0.8 and false discovery rate (FDR) < 10^-5. A previously published list of all glycosylation genes was used (10). Uniform Manifold Approximation and Projection (UMAP) was used for dimension reduction.

The integrated scRNA-seq atlas from Salcher et al. (11) was downloaded, which consists of 1,283,972 cells from 318 patients. Cells which were annotated as originating from primary tumor sites and either LUAD or LUSC were selected, resulting in a dataset of 345,260 cells from 163 LUAD patients and 128,423 cells from 72 LUSC patients. After selecting cells, the UMAP space was recomputed using the reprocess_adata_subset_scANVI function (https://github.com/icbi-lab/luca) with default settings for visualization. Coarse cell type annotations with 12 cell types were adopted to study cell type specific-gene expression. For UMAP visualizations, gene expression counts were log-transformed and library size was corrected to 10,000 counts per cell. The Wilcoxon test was used to identify DEGs between LUAD and LUSC epithelial cells. To increase sensitivity of DEG identification in scRNA-seq data (12), pseudobulk mixtures were generated by aggregating counts of epithelial cells from each patient. Subsequently, pseudobulk mixtures were library size-corrected to 10,000 counts per mixture. DEGs between LUSC and LUAD pseudobulk mixtures were identified based on absolute binary logarithms of fold changes (Log₂FC) >0.8 and false discovery rate (FDR) < 10^-2.

Immunohistochemical (IHC) staining of tissues was performed on FFPE sections (4 µm). Assisted by a pathologist, FFPE-tissue samples were selected from stage 3 LUAD (n=5) or LUSC (n=5) patients that underwent surgery prior to chemo, radio- and/or immunotherapy at the Vrije Universiteit medical center in Amsterdam. Paired adjacent non-malignant tissues were incorporated as reference controls. Ethical approval was not mandatory for this study due to the usage of leftover patient material, as stated in dossier number 2021.0063-VIP which was issued by the aforementioned medical center. Resected material was processed using conventional FFPE-tissue preservation techniques within clinical pathology labs.

Tissue slides were deparaffinized using xylene and rehydrated using ethanol, washed using demineralized water prior to heat induced antigen retrieval (DAKO Agilent, K800521–2 or K800421-2). Endogenous peroxidase activity was blocked by peroxidase-blocking solution (DAKO Agilent, S202386-2) for 10min and aspecific binding to tissue and Fc-receptors was blocked using protein block (Immunologic, VWRKBD09-999). Primary antibody ( Supplementary Table S1 ) was dissolved in aforementioned protein block solution and incubated for 1 hour at room temperature, except for ST6GALNAC6 which was incubated overnight (20 hours) at 4°C, further details are listed in Supplementary Table S1 . After incubation with the primary antibody, slides were incubated with BrightVision Poly-HRP-Anti Mouse/Rabbit IgG Biotin-free (Immunologic, VWRKDPVO55HRP) at room temperature for 30 mins. Antibody targets were visualized using DAB (3,3’-diaminobenzidine) for 10 minutes and slides were counterstained with hematoxylin (Sigma-Aldrich, 51275) and embedded with entellan (Sigma-Aldrich, 1079610500).

Stained slides were imaged at a 40x magnification on the Vectra Polaris Automated Quantitative Pathology Imaging System (Akoya Biosciences, software version 1.0.13). To remove anthracosis aspecific signal, acquired images were processed using inForm^® Tissue Analysis (Akoya Biosciences, software version 2.6.0) with a brightfield spectral library.

For the systematic analysis of survival, patients were stratified in high (top 25%) and low (bottom 25%) according to the expression of each glycosylation-related gene. The hazard ratio (HR), calculated in a univariate Cox proportional regression model analysis, was used to select genes that affect the prognosis, which were plotted in Kaplan Meier curves and significance studied using log-rank test.

Plasma from patients with lung adenocarcinoma (n=20) and Squamous cell carcinoma (n=16) was obtained from the Liquid Biopsy Center of the Amsterdam UMC. The studies involving human participants were reviewed and approved by the Medical Ethical Committee, Amsterdam UMC. Written consent was obtained from all the donors. A Galectin-7 ELISA kit (R&D Systems, DY1339) was used to measure its concentration in the plasma of patients. A standard curve based on recombinant Galectin-7 was used for quantification, with a blank control included to subtract background signal.

R v4.3.0 software (https://mirror.lyrahosting.com/CRAN/) as used for statistical analysis and figure drawing. Significance was called when the adjusted p-value < 0.05. Key clinical characteristics of the patient cohort (TCGA datasets, single-cell datasets, IHC staining tissues, and ELISA sample sets) are summarized in Supplementary Table S5 .

The study’s flow chart is shown in Figure 1 . In order to investigate the expression of glycosylation-related genes, we started by analyzing differential gene expression between the different lung cancer subtypes and adjacent normal tissue ( Supplementary Table S2 , Supplementary Table S3 ). To facilitate the interpretation of the results, we grouped the results based on their involvement in different glycosylation pathways, including mucins, GalNAc-initiation, elongation, fucosylation, and sialylation ( Figure 2A ).

We observed that LUAD was particularly enriched in genes encoding for mucins or mucin-like proteins, in particular MUC4, MUC5B, MUC13, MUC16, MUC20, MUC21, and OVGP1 (Oviductal glycoprotein 1) that were highly expressed in LUAD compared to normal lung, while EMCN was downregulated in LUAD. In LUSC, only MUC6 and MUC20 were upregulated, while EMCN, MUC15, and MUC21 were downregulated.

Regarding initiation of GalNAc-type O-glycosylation, GALNT2, GALNT3, GALNT6, GALNT7, and GALNT14 were highly expressed in both cancer types, LUAD and LUSC. In addition, GALNT4 was also highly expressed in LUAD and GALNT1 was also highly expressed in LUSC.

As for elongation, B3GNT3, B3GNT4, and GCNT3 were highly expressed in both cancer types (LUAD and LUSC). B3GNT6 and C1GALT1 were highly expressed, and GCNT4 was downregulated in LUAD. B3GNT5 was upregulated, while B3GNT7, B3GNT8, and GCNT4 were downregulated in LUSC. These results suggest that LUAD and LUSC both present an increase of O-glycans. We found several glycosylation elongation genes essential in the progress of synthesizing core 3 (GlcNAcβ1–3GalNAc) and core 4 (GlcNAcβ1–6[GlcNAcβ1–3]GalNAc) structures highly expressed in LUAD, including C1GALT1, B3GNT6, and GCNT3.

After aggregating all sialylation associated genes, we found CMAS and NANP were upregulated in LUSC. In the context of α2,6-GalNAc-sialylation, ST6GALNAC1 exhibited upregulation in LUAD, while ST6GALNAC2 showed downregulation in LUAD but upregulation in LUSC. Additionally, for α2,8-sialyltransferases, we found ST8SIA2 was elevated in these two primary subtypes of NSCLC, with much higher expression in LUSC. However, we did not identify any upregulated genes compared to adjacent normal tissue among other sialyltransferases ( Supplementary Table S2 , Supplementary Table S3 ).

Genes associated with fucosylation, FUT2, FUT3, FUT6, FUT8, FUT9, and GMDS were upregulated in both LUAD and LUSC, although more significantly in the former. Especially in LUSC, we found an upregulation of POFUT1, encoding for an enzyme involved with O-fucosylation. In both LUAD and LUSC, the expression of fucosyltransferase gene FUT2 was increased, facilitating the attachment of α1,2-fucosylation to Gal-residues. Moreover, within the same fucosyltransferase family, FUT3, FUT6, and FUT9—which facilitate α1,3- and α1,4-fucosylation of GlcNAc—along with FUT8, responsible for α1,6 core fucosylation of N-glycans, also exhibited upregulation. Interesting for the Galectin family, LGALS4 was highly expressed in LUAD. Oppositely LGALS7 and LGALS7B were upregulated, and LGALS2, LGALS3, and LGALS4 were downregulated in LUSC ( Figure 2C ).

For glycolipid biosynthesis ( Figure 2B ), in both LUAD and LUSC, upregulation of B4GALT2, B4GALT3, and B4GALT4, pivotal genes integral to the synthesis of glycoproteins and glycolipids, was observed. These genes play a crucial role in facilitating the transfer of galactose during the growth of carbohydrate chains. GalNAc transferase B4GALNT1 was upregulated both in LUAD and LUSC during the synthesis of ganglioside sugar structure GA2. Especially in LUSC, B4GALT6 and B3GNT5 were upregulated, which involved in the transfer of galactose and GlcNAc during the synthesis of glycoproteins and glycolipids. Given that there is a general increase of fucosylation in tumor, these glycolipids may also be fucosylated to generate Lewis antigens. ST3GAL5, which participates in the transfer of sialic acid (Neu5Ac) to galactose-containing substrates and catalyzes the formation of ganglioside GM3 using lactosylceramide (LacCer) as the substrate, was downregulated in LUSC.

As our results showed that LUAD and LUSC present a dissimilar regulation of glycosylation-related genes when compared to adjacent normal samples, we next investigated further the differences between these 2 main subtypes of NSCLC ( Figure 3A ). We observed that high expression of LGALS4, MUC21, B3GNT6, MUC5B, MUC13, MUC1, GAL3ST1, B3GAT1, B3GNT7, and ST3GAL5 was associated with LUAD, while high expression of LGALS7, LGALS7B, ST6GALNAC2, B3GNT5, HS6ST1, TMTC3, and ALG3 was associated with LUSC. In Figures 4A, B , we also compared DEGs from LUAD and LUSC versus adjacent normal tissue, respectively. In LUAD, FUT9, MUC13, B3GNT6, B4GALNT4, GCNT3, B3GNT3, LGALS4, GALNT14, MUC21, and HS6ST2 were found as top 10 most highly overexpressed genes. In LUSC, the top 10 most highly overexpressed genes were LGALS7, LGALS7B, B4GALNT4, UGT1A1, FUT9, ST8SIA2, GALNT14, B4GALNT1, B3GNT4, and HS6ST2.

After reducing dimensions of UMAP, adjacent normal, LUAD, and LUSC were clustered into different groups based on their biological properties ( Figure 3B ). Then we selected a few significant DEGs from the comparison of LUAD and LUSC and subjected them to further analysis (LGALS4, LGALS7, LGALS7B, MUC21, ST6GALNAC1 and ST6GALNAC2). In the UMAP plots shown in Figure 3C , LGALS4 and MUC21 were consistently found to be highly expressed in LUAD, whereas they were downregulated in LUSC compared to the control group. ST6GALNAC1 exhibited high expression levels in LUAD, with less expression in the LUSC cluster. In contrast, we detected higher levels of LGALS7, LGALS7B, and ST6GALNAC2 expression in LUSC than in LUAD ( Figures 3C , 5A ).

To evaluate the differential expression of LGALS4, LGALS7, MUC21, ST6GALNAC1, and ST6GALNAC2 in human tissue, IHC staining was employed in both LUSC and LUAD subtypes of lung cancer. As shown in Figure 3D , results confirm that Galectin-4 (LGALS4) is highly expressed in epithelial cells of LUAD, while Galectin-7 (LGALS7) is highly expressed in epithelial cells of LUSC. MUC21 exhibits high expression in the epithelial cells of LUAD, contrasting with dim staining observed in LUSC. ST6GALNAC1 demonstrates high expression in the epithelial cells of LUAD compared to LUSC. Furthermore, it was weakly stained in the stroma area of LUAD and LUSC tissue. For ST6GALNAC2, we discovered that it was also weakly stained in the stroma region, although there were no appreciable variations between LUAD and LUSC. As shown in Figure 4D , adjacent non-malignant tissues exhibited absent or markedly weaker expression for Galectin-4, Galectin-7, MUC21, ST6GALNAC1, and ST6GALNAC2 compared to the corresponding cancer tissues.

In order to further explore the mRNA expression of these candidate genes in NSCLC, we examined Single-cell RNA-seq (scRNA-seq) data of 345,260 cells from 163 LUAD patients and 128,423 cells from LUSC patients retrieved from Salcher et al. (11) ( Figure 6A ). The scRNA-seq atlas consisted of 12 major cell types: epithelial cells, stromal cells, endothelial cells, macrophages/monocytes, T cells, natural killer (NK) cells, neutrophils, B cells, plasma cells, plasmacytoid DCs (pDCs), conventional or classical DCs (cDCs) and mast cells ( Figure 6B ). From all annotations of various cell types, we found that the epithelial cell component of LUAD expressed more LGALS4 and MUC21 than that of LUSC ( Figure 6C ). In contrast, there was increased expression of LGALS7 and LGALS7B in LUSC. Furthermore, in the tumor microenvironment of NSCLC, ST6GALNAC1 and ST6GALNAC2 showed elevated expression levels not only in epithelial cells but also in T cells and macrophages/monocytes ( Figure 4C ).

Since malignant epithelial cells were the main component that contributes to the signature of LGALS4, LGALS7, MUC21, and ST6GALNAC1, we further analyzed DEGs in epithelial cells of Single-cell data ( Figure 6D ). Our analysis revealed a consistent trend with our findings from TCGA data, showing higher expression of LGALS4, MUC21, and ST6GALNAC1 in LUAD compared to LUSC, while LGALS7 and LGALS7B exhibited elevated expression in LUSC compared to LUAD. These results further validate our exploration from tissue staining.

Furthermore, we wanted to confirm that Galectin-7 expression can be used to distinguish LUSC from LUAD. Since Galectins can be secreted, we set out to detect its presence in serum of LUSC and LUAD patients. An ELISA kit was used to detect the secretion level of Galectin-7 in the plasma of lung cancer patients (Adenocarcinoma (n=20) vs. Squamous cell carcinoma (n=16)). We measured significantly higher levels of Galectin-7 in plasma from squamous cell carcinoma patients than in plasma from adenocarcinoma patients, suggesting that it could serve as a biomarker for LUSC ( Figure 6E , P < 0.0001).

Univariate Cox regression analysis was conducted to evaluate the survival implications of all glyco-associated genes in LUAD based on Hazard ratio values (Figure 7A). Our findings revealed that elevated expression of 36 genes was linked to unfavorable survival outcomes, whereas 31 genes were correlated with better survival outcomes. Patients were divided into two groups based on the expression level of each gene, specifically the top 25% with high expression and the bottom 25% with low expression. Subsequently, survival curves were plotted, comparing the impact of individual genes on survival. Notably, CMAS, SLC35A1, ST3GAL4, MUC16, and FUT4 were identified as genes linked to poor survival (Figure 8A).

In parallel, the survival implications of all glyco-associated genes were also assessed in LUSC. We found that 14 genes with high expression were associated with bad survival of squamous cell carcinoma, and 9 genes were associated with better survival ( Figure 7B ). SLC35A1, ST6GALNAC4, and ST6GALNAC6 were selected and validated, demonstrating that their elevated expression is associated with poor survival of LUSC patients ( Figure 8B ). No significant differences in survival were observed in LUAD and LUSC for high and low expression of LGALS4, LGALS7, MUC21, ST6GALNAC1, and ST6GALNAC2 ( Figures 7C, D ).

UMAP projection was utilized to visualize features of these seven survival-relevant genes in both LUAD and LUSC ( Figure 5A ). SLC35A1, MUC16, FUT4, ST3GAL4, ST6GALNAC4, and ST6GALNAC6 showed more expression in LUAD than in LUSC ( Supplementary Table S4 ). In contrast, CMAS and LGALS7B were highly expressed in LUSC.

Intriguingly, it was also observed that the expression of ST6GALNAC6 in the adjacent normal cluster was higher than that in the clusters of LUAD and LUSC; and SLC35A1 was also highly expressed in adjacent normal tissue ( Figure 5A ). Furthermore, tissue staining revealed that ST6GALNAC6 expression levels were comparable between LUAD and LUSC, and neither subtype showed significant differences relative to adjacent non-malignant tissues ( Figure 5B ). High ST6GALNAC6 expression in LUSC was associated with bad survival, while the opposite was observed in LUAD ( Figures 8B , 7A, B ).

We found that CMAS and SLC35A1 correlated with worse survival outcomes in lung adenocarcinoma. The activity of CMAS, responsible for converting Neu5Ac to CMP-Neu5Ac, has been described to be significantly associated with decreased survival of breast cancer (13). Previously, we found that CMAS KO in a murine model resulted in enhanced infiltration of CD4⁺ and CD8⁺ T cells within the tumor microenvironment of pancreatic ductal carcinoma and improved survival outcomes (14). In contrast, upregulation of CMAS in pancreatic tumors increased sialylation and promoted immune suppression (9, 15).

Additionally, SLC35A1 transports cytidine 5’ monophosphate (CMP)-sialic acid, the donor substrate for a range of sialyltransferases, thereby modulating sialylation within the Golgi apparatus (16). Notably, SLC35A1 was also seen to be associated with worse survival outcomes in lung squamous cell carcinoma. SLC35A1 knock-down in B16 melanoma reduced tumor growth due to the reduction of sialylation and enhanced effector T cell response (17). These studies demonstrated that the reduction of sialylation lowers the engagement of Siglecs (Sialic acid-binding lectin receptors) and their immune inhibitory function; and showed that sialylation as glyco-immune checkpoint modulates tumor growth (18).

The mucin family comprises a large group of heavily O-glycosylated proteins, classified into membrane-bound and secretory types. Among membrane-bound mucins, MUC1, MUC16, and MUC21 attach to cell surfaces via their transmembrane domains (19). In lung cancer, high-grade polarized expression of MUC1 is observed in well-differentiated adenocarcinoma, while depolarized MUC1—extending from the apex to the entire surface—is associated with advanced stages, lymph node metastasis, and disruption of cell–cell and cell–matrix interactions (20, 21). MUC16 was overexpressed in both human primary lung carcinoma and associated lymph node metastases, potentially playing a role in the epithelial-to-mesenchymal transition during lung cancer cell metastasis (22). In addition, the highly glycosylated tandem repeat domain of MUC21 on the cell surface impairs cell–cell and cell–matrix adhesion via steric hindrance, potentially contributing to tumor metastasis through enhanced cell migration and invasion (23).

Our results on mRNA differential expression, both in bulk data and in malignant cells from the scRNA-seq data, demonstrated that MUC1 and MUC16 were substantially overexpressed in LUAD compared to LUSC, with MUC1’s expression validated by IHC (24). Retrospective studies align these findings with poor survival in NSCLC, though our data did not confirm MUC1’s correlation with poor survival, despite previous affirmations (24–26). MUC21 showed higher mRNA expression in LUAD than in adjacent normal or LUSC tissues, consistent with GEPIA database findings (26). IHC revealed greater MUC21 expression in certain cancer cell patterns, particularly micropapillary, papillary, and lepidic, compared to cohesive tumor components in LUAD patients (27).

An increasing number of clinical trials have focused on targeting mucins such as MUC1, employing modalities including monoclonal antibodies (e.g., PankoMab-GEX for TA-MUC1), liposomal vaccines (e.g., tecemotide), and CAR-T cell therapy, demonstrating clinical feasibility of MUC1-targeted therapy (28, 29). MUC21 is highly expressed in micropapillary structures and may contribute to the transition from pure lepidic to micropapillary pattern, suggesting its involvement in LUAD progression and potential as a biomarker for predicting disease progression (30). Specific glycosylated forms of MUC21 may contribute to the development of LUAD with EGFR mutations, which are strongly associated with a dominant micropapillary growth pattern (30, 31). Therapeutic strategies targeting MUC21 may offer a promising approach, particularly in LUAD subtypes characterized by micropapillary architecture and EGFR mutations, where its expression is implicated in tumor progression.

Galectins, a family of carbohydrate-binding proteins, are classified into prototypical, tandem-repeat, and chimeric types based on their carbohydrate recognition domains (CRDs) (32). These proteins serve as diagnostic biomarkers for detecting malignant tumors (33). Among the galectins that emerged as most relevant in our research, Galectin-2 and Galectin-7 are prototypical galectins, Galectin-4 is a kind of tandem−repeat galectin, and Galectin−3 is the only chimeric galectin, which contains a single CRD and a large amino−terminal domain (32).

Galectin-4 was strongly expressed in LUAD patients with lymph node metastasis and was associated with aggressive cancer traits such as lymphatic and venous invasion, although it did not correlate with overall or recurrence-free survival (34).

Conversely, Galectin-7, significantly upregulated in lung squamous cell carcinoma (LUSC) versus LUAD, is validated by increased serum levels and higher immunohistochemistry (IHC) expression in LUSC, indicating its potential as a biomarker. In a syngeneic mouse squamous cell carcinoma (SCC) model, Galectin-7 has been recognized as a mediator of metastasis linked to immunosuppression, exhibiting significant induction in the tumor microenvironment during tumorigenesis, and is released extracellularly at advanced stages of tumor growth (35). Consistent with this, a previous study of breast carcinoma has demonstrated that Galectin-7, absent in low-grade but upregulated in high-grade, is associated with increased metastasis to the lungs and bones (36).

Enzymes from the ST6GALNAC sialyltransferase family are involved in α2,6-sialylation on glycolipids and O-glycosylated proteins, via the addition of sialic acid to GalNAc residue (5). Specifically, ST6GALNAC1 and ST6GALNAC2 have been shown to enhance metastatic potential in various cancers (37, 38).

High expression of ST6GALNAC1 induces the synthesis of the sialyl-Tn antigen via α2,6-linkage by promoting the sialylation of MUC5AC, thereby facilitating liver metastasis in LUAD (37). It shows high expression in LUAD with prominent localization on tumor cell membranes, distinguishing it from poorly differentiated squamous cell carcinoma (PDSCC) (39).

ST6GALNAC2 mainly sialylates T antigens, contributing to the formation of disialyl-T antigen. ST6GALNAC2 expression pattern and its potential implications in human lung cancer tissues remain unexplored in prior research. Notably, in other studies, disialyl-T antigen has been defined as a ligand for Siglec-7, an immune-inhibitory glycan-binding receptor expressed on NK cells and myeloid cells, which suppresses immune function upon ligand engagement (40). In addition, ST6GALNAC2 impacts Galectin-3 binding, with high expression correlating with reduced lung metastasis and improved survival in ER- breast cancers (38). Furthermore, ST6GALNAC2 has been shown to promote the invasive capabilities of breast carcinoma cells, potentially through activation of the PI3K/Akt/NF-κB signaling pathway (41). In contrast, elevated expression of ST6GALNAC2 in colorectal cancer is linked to poorer survival outcomes (42).

Based on our existing scRNA-seq data, T cells and macrophages/monocytes are prominent contributors to higher ST6GALNAC2 mRNA expression, which further supported our IHC staining result. This fact might help to explain our findings that it affected the analysis result of bulk data, and did not significantly differentiate between LUAD and LUSC in the malignant cell compartment.

ST6GALNAC4 is involved in the synthesis of disialyl-T antigen from sialyl-T antigen and GD1α from GM1b (43). ST6GALNAC4 is tied to adverse outcomes in LUSC per TCGA data. It increases T antigen expression and Galectin-3+ macrophage recruitment, which supports tumor invasion and immune suppression (44–46). It could be interpreted that in the absence of core 2 O-glycans, the increase of disialyl-T antigen portion is facilitated by ST6GALNAC4 prevents glycan elongation on the cell surface. Consequently, this situation mediates the adhesion of Galectin-3 to interact with residual T antigens (43).

ST6GALNAC6 is involved in the synthesis of ganglioside GD1α (representative 0-series gangliosides), GT1aα (a-series), and GQ1bα (b-series) from GM1b, GD1a, and GT1b, respectively (47). Downregulation of ST6GALNAC6 mRNA was detected in human colon cancer compared with non-malignant epithelium, accompanied by a concomitant decrease in disialyl-Lewis^a and an increase in sialyl Lewis^a during malignant transformation (48). Another discovery from a renal cancer study revealed that silencing ST6GALNAC6 in cancer cells resulted in a reduction of metastatic ability (49). Despite notable expression differences in non-malignant versus malignant lung tissues, the molecular mechanisms linking ST6GALNAC4 and ST6GALNAC6 to clinical outcomes in NSCLC remain underexplored.

Carcinoembryonic antigen (CEA) is an acidic glycoprotein associated with human embryonic antigen, characterized by extensive N-glycosylation, and has been utilized as a diagnostic or prognostic marker in various cancer types, including lung cancer (50–52). However, its clinical utility is limited by relatively low specificity, making it more suitable for disease monitoring rather than early detection. In contrast, our study identifies several glycosylation-associated genes with potential diagnostic value in NSCLC. These novel biomarkers may offer improved tumor subtype discrimination. Further functional validation and clinical correlation studies are warranted to elucidate their potential utility in the diagnosis and therapeutic intervention of lung cancer.

Although glycosylation plays a crucial biological role, its study remains challenging due to the structural complexity and dynamic biosynthesis of the glycan chain, as well as the lack of a direct structure–function relationship (53). Furthermore, the intrinsic heterogeneity of tumor tissues adds another layer of difficulty in translating glycosylation-based findings into effective clinical therapies.