GSE6631 dataset, an mRNA microarray comparing 44 HNC with paired normal tissue controls, was downloaded from the Gene Expression Omnibus (GEO; ncbi.nlm.nih.gov/geo/). To identify genes with enzyme annotations that were significantly differentially expressed, fold-change was determined using GEO2R (ncbi.nlm.nih.gov/geo/info/geo2r.html#how_to_use). Ingenuity Pathway Analysis (IPA) was used to identify the metabolic pathways significantly altered in HNC tumor tissue (digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/analysis-and-visualization/qiagen-ipa/).

ENCORI database (starbase.sysu.edu.cn/index.php) was used to analyze the correlation between ALDH2 and vascular endothelial growth factor C (VEGFC) mRNA expression. VEGFC mRNA levels in HNC tissue and its impact on overall survival of patients with varied ALDH2 and VEGFC expression were investigated using the OncoLnc platform (oncolnc.org/). Gene expression analyses were conducted using GEPIA (gepia.cancer-pku.cn/), based on data from The Cancer Genome Atlas (TCGA) (10).

Tissue sample were fixed in 10% buffered formalin for 24 h at room temperature and embedded in paraffin. Paraffin-embedded sections of malignant tissue and their matched adjacent non-cancerous tissue (distance, >1 cm) were obtained from 106 patients diagnosed with HNC. The cohort included 91 male patients and 15 female pateints (age range: 27 to 87). Specimens were from patients without distant metastasis at the initial presentation, with histologically confirmed squamous cell carcinoma, and primary tumors located in the oral cavity, oropharynx, hypopharynx, or larynx. All the patients underwent surgical intervention at the Kaohsiung Veterans General Hospital, Kaohsiung, Taiwan between 2010/01 and 2016/12. Relevant clinical variables were recorded, including age, sex, American Joint Committee on Cancer (AJCC) T and N classification 11, overall stage (11), substance misuse, treatment strategy and disease status. Each discrete position within the initial tissue microarray (TMA-1) was configured to accommodate two cancer specimens and one adjacent non-cancerous specimen for quantification of protein levels. For additional analyses, a second microarray (TMA-2) was created from the same cohort. Due to sample loss during TMA-2 preparation, 84 samples were available. Ethics approval, including a waiver of informed consent, was granted by the Ethics Committee of Kaohsiung Veterans General Hospital (approval no. KSVGH23-CT8-10).

The 4-µm slides of the TMA paraffin block were used for IHC. TMA slides were deparaffinized in xylene, dehydrated using graded ethanols at room temperature. Briflyantigen retrieval from the TMA slides was performed using retrieval solution (Tris-EDTA buffer, pH 9.0) at 95°C for 12 min and incuabated with peroxidase Blocking buffer (3% hydrogen peroxide) at room temperature for 10 min. Slides were incubated with blocking reagent (ready to use, BioTnA, TA00C2) for 30 min at room temperature. After blocking, slides were incubated with a primary antibody against ALDH2 (1:200; GTX101429, Genetex) and VEGFC (1:1,000, GTX113574, Genetex) for 1 h at room temperature. Expression level was determined by employing the HRP-conjugated secondary antibody (ready to use, TnAlink Polymer Detection System, cat. no. TACH02D, BioTnA,) for 30 min at room temperature, followed by 1 min of DAB (1:20) and 5 sec hematoxylin counterstaining at room temperature (TnAlink Polymer Detection System, TACH02D, BioTnA, Kaohsiung, Taiwan). The TMA slides were scanned by MoticEasyScan Pro (light microscope, 400×). A pathologist assessed the TMAs to exclude microarrays of suboptimal quality. Finally, quantification of ALDH2 and VEGFC IHC staining was performed using HistoQuest (TissueGnostics, version 7.1, Deep-learning nuclear segmentation program,). ALDH2 protein expression in cancer tissues was categorized into the high expression group, defined as >50% positive cells, and the low expression group (≤50% positive cells).

HNC cell lines SAS (Japanese Collection of Research Bioresources; cat. no. JCRB0260) and MTCQ1 (Bioresource Collection and Research Center, cat. no. 60620) were maintained in DMEM (HyClone; Cytiva; cat. no. SH30003.02) supplemented with 10% FBS (Cytiva, SH30396.03) and 1% penicillin-streptomycin-amphotericin (PSA; Biological Industries, 03-033-1B) at 37°C in 5% CO₂. DOK (oral dysplasia), TW1.5 and TW2.6 (buccal mucosa) cell lines were maintained in DMEM/F12 (HyClone, SH30004.04) supplemented with 10% FBS and 1% PSA. DOK, TW1.5 (12,13) and TW2.6 were kindly provided by Dr Michael Hsiao (Academia Sinica, Taipei, Taiwan).

Lentivir us vector control (pLKO-1-shLuc) and short hairpin (sh)ALDH2 viral supernatants (TRCN0000026452 and TRCN0000026486) were obtained from the National RNAi Core Facility (Taipei, Taiwan). RNAi core based on 3rd Generation lentiviral guide, shRNA lentiviruses were produced by co-transfecting with hairpin-pLKO.1 vector (1 μg), VSV-G/pMDG2.G (0.1 μg), and pΔ8.91 (0.9 μg) constructs into 293T (American Type Culture Collection, CRL-3216™) cells using TransIT-LT1 transfection reagent (Mirus Bio, LLC; cat. no. #MIR 2300) and DNA complex incubated for 30 min at room temperature. After 40 h transfection, the viral supernatants were then harvested. Target sequences are listed in Table SI. Polybrene (8 μg/ml) was used to infect DOK or TW2.6 cells (1×10⁵/well) using viral supernatants (5 multiplicity of infection). Following 72 h infection, cells were selected and maintained using 2 μg/ml puromycin (InvivoGen, ant-pr-5). Using 3rd Generation lentiviral guide, overexpression lentiviruses were produced by co-transfecting with cDNA-expressing vector (10 μg), pMDG (10 μg), and pΔ8.91 (1 μg) constructs into 293T cells using calcium phosphate (DNA complex incubated for 30 min at room temperature). After 48 h transfection, the viral supernatants were then harvested and concentrated by the Lentivirus Concentration kit (TopGen Biotechnology Co., Ltd., GM801-02). The lentiviral expression vector carrying human ALDH2 (NM_000690.4, pLVX-ALDH2-IRES-Neo) and control vector (pLVX-IRES-Neo) were acquired from TopGen Biotechnology Co., Ltd. Human ALDH2 cDNA was overexpressed in SAS and MTCQ1 cell lines. A total of 8 μg/ml polybrene was used to infect SAS or MTCQ1 cells (1×10⁵/well) using viral supernatants (10MOI). After 72 h infection, the cells were selected and maintained using 400 μg/ml G418 Sulfate (Gibco^TM, 10131035). After 48 h selected and cultured for 48 h for subsequent analyses.

Total RNA was extracted from HNC cells using TRIzol^® (Thermo Fisher Scientific, Inc.; cat. no. #15596018) following the manufacturer's instructions. cDNA was then synthesized using a PrimeScript™ RT Reagent kit (cat. no. #RR037A; Takara Biotechnology Co., Ltd.) according to the manufacturer's protocol. qPCR was performed using a SYBR Green PCR Master Mix (PCR Biosystems Ltd. qPCRBIO SyGreen Mix Lo-ROX) to evaluate target gene expression. Thermocycling conditions were as follows: Initial denaturation at 95°C for 3 min; followed by 40 cycles of 95°C for 10 sec and 60°C for 30 sec. The 2^−ΔΔCq was used to calculate the results, and ACTB was used for normalization (14). The primers are listed in Table SII.

Total RNA was isolated from TW2.6/shluc and TW2.6/shALDH2 cells using TRIzol^® (Thermo Fisher Scientific, Inc.; cat. no. #15596018). Microarray analysis was performed on isolated RNA using a Clariom S Array, human (Applied Biosystems; Thermo Fisher Scientific, Inc.; cat. no. 902916) and scanned using an Affymetrix GeneChip Scanner 3000. To analyze the effects of ALDH2 knockdown, an interaction network was generated using Qiagen GmbH IPA (QIAGEN 2000-2023) from TW2.6/shALDH2 cells, with the threshold set at 2-fold change. The microarray data were deposited in the GEO under accession no. GSE253622.

For invasion experiments, a polycarbonate membrane was pre-coated with fibronectin (10 μg, Thermo Fisher, 33016015) on the lower side and Matrigel (Corning, Inc.; cat. no. 354234) on the upper side at room temperature, 3 min. Migration assay used membranes without Matrigel pre-coating. The lower chamber was filled with complete culture medium (containing DMEM, 10% FBS). The cells (3×10⁵/ml) were suspended in a serum-free DMEM, added to the 50 μl cells into upper chamber of each well and incubated at 37°C in 5% CO₂ for 15-18 h. NF-κB inhibitor (Bay11-7082) was included in cells to perform the assay. To block VEGFC, we incubated TW2.6/shALDH2 cells with a VEGFC-neutralizing antibody (10-20 μg, GTX52393, GeneTex) for 30 min. After incubation, cells were trypsinized for migration and invasion assays, and a VEGFC-neutralizing antibody was added to the cells. After incubation, the cells were stained with crystal violet for 30 min at room temperature and counted under a light microscope (16×).

Cells (5,000 cells/well for DOK/shluc, DOK/shALDH2, TW2.6/shluc, TW2.6/shALDH2, SAS/Vector, and SAS/ALDH2) were plated in a 96 well plate with complete DMEM (10% FBS) mediaum for 24-72 h at 37°C in 5% CO₂.

Using MTT (Sigma-Aldrich, 10 μl) for 4 h, the cell proliferation was assessed, and the signals were measured using an ELIAS reader (Thermo Fisher Scientific; Multiskan FC).

After extracting protein from the cell lines (DOK/shluc, DOK/shALDH2, TW2.6/shluc, TW2.6/shALDH2, SAS/Vector, and SAS/ALDH2) using RIPA lysis buffer [50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% NP40, 1% sodium deoxycholate, 0.1% SDS, PMSF (0.1 mM), Na₃VO₄ (2 mM, and NaF (1 mM)] and protein concenteration were determined using Pierce™ BCA Protein Assay kits (ThermoFisher, 23225). Protein (50 μg) were loaded onto 10% SDS-PAGE for electrophoresis and subsequently transferred to a PVDF membrane and blocked with 5% non-fat dry milk in 1xPBST for 1 h at room temperature. The primary antibodies were incubated overnight at 4°C. Table SIII lists the primary antibodies used for immunoblotting. The membranes were incubated with second antibodies [1:5,000, mouse (Jackson ImmunoResearch, 115-035-003), 1:5,000, rabbit (Jackson ImmunoResearch, 111-035-0031)] for 1 h at room temperature. The signals were visualised using an ECL™ Western Blotting Reagents (Cytiva, XR-IGE-RPN2106). ImageJ software (version 1.53t, National Institutes of Health) was used to quantify the densitometry.

NF-κB p65 (RELA) siRNA, locus 11q13.1.; cat. no. #sc-29410 and control siRNA-A (cat. no. #sc-37007) were purchased from Santa Cruz Biotechnology, Inc. Next, 1×10⁶ cells with complete media were plated in 6-cm plates for 24 h at 37°C and subsequently transfected with 100 nM NF-κB p65 or control siRNA using TransIT-X2 (Mirus Bio, LLC; cat. no. #MIR 6000) and incubacted room temperature 30 min. Subsequent experiments were performed after 48 h.

Briefly, 2.5×10⁴ cells/well were plated on a 96-well plate and incubated overnight at 37°C in 5% CO₂. DCFDA (Abcam; cat. no. ab113851) was added to the cells and incubated at 37°C in 5% CO₂ for 45 min. After removing the DCFDA, cells were examined using a fluorescence plate reader (FLUOstar Omega; BMG Labtech GmbH).

TW2.6/shluc and TW2.6/shALDH2 cells were plated on 6 cm plates with complete DMEM/F12 (10% FBS) at a density of 7.5×10⁵ cells/plate for 24 h at 37°C in 5% CO₂. Subsequently, 0.5 pCMV6-AV-GFP (OriGene; cat. no. #PS100010) and 4.5 μg pGL4.32 (luc2P/NF-κB-RE/Hygro) [Promega Corporation; E8491, NF-κ B response element (33-84), 5′-GGGAATTTCCGGGGACTTTCCGGGAATTTCCGGGGACTTTCCGGGAATTTCC-3′] plasmid were co-transfected into cells using TransIT-X2 (Mirus Bio, LLC; cat. no. #MIR 6000). Following 48 h incubation, 1×10⁴ cells with DMEM/F12 (10% FBS) were plated in 96-well plates for 6 h at 37°C in 5% CO₂ and subsequent analyses. TW2.6/shALDH2 cells were treated with N-acetyl-L-cysteine (NAC, 10 mM; Table SIV) for 12 h at 37°C in 5% CO₂. A ONE-Glo Luciferase assay kit (Promega Corporation; cat. no. #E6120) and a luminometer (FLUOstar Omega; BMG Labtech GmbH) were used to measure the green fluorescence and luciferase signals. pCMV6-AV-GFP was used as internal control.

SAS/ALDH2 cells (5×10⁵ cells/well) were plated in a 6-well plate and incubated overnight at 37°C in 5% CO₂. H₂O₂ (80 μM; Table SIV) was added at 37°C in 5% CO₂ overnight. After 24 h, cells were examined using RT-qPCR to examine VEGFC expression.

TW2.6/shluc, TW2.6/shALDH2, SAS/Vector, SAS/ALDH2, MTCQ1/Vector, and MTCQ1/ALDH2 HNC cells (4,000 cells/well) were plated in a 6-well plate. After being incubated for 7 days at 37°C and 5% CO₂, the resulting cell colonies were fixed with 1% formalin for 30 min at room tempature and stained with crystal violet for 30 min at room tempature. ImageJ software (version 1.53t, National Institutes of Health) was used to count the number of colonies (diameter >0.1 cm).

Data are shown as the mean ± standard deviation of ≥3 independent repeats. The data analyses used the Student's t- and χ² parametric tests, while Mann-Whitney U and Fisher's exact test were employed for non-parametric tests. One-way ANOVA followed by Tukey's post hoc test was used used for multiple group comparisons. Correlation analysis was determined by Spearman's correlation. Prognostic importance of covariates was evaluated through Cox proportional hazards regression in both univariate and multivariate models. Kaplan-Meier method was applied to examine survival curves and the log-rank test was used to compare survival rates between groups. All statistical tests were two-tailed. P<0.05 was considered to indicate a statistically significant difference.

To identify the metabolic pathways involved in the initiation and progression of HNC, genes with enzyme annotations that showed differential expression in tumor compared with normal tissues from patients with HNC in the GSE6631 dataset were analyzed (Fig. 1A; Table SV). Based on 150 genes that showed differential expression in HNC compared with normal tissue (Table SV), pathways were filtered based on the z-scores with z-score <-2.0 indicating inhibition (Table SVI). IPA revealed that three of the top 15 enriched pathways were 'ethanol degradation II' and 'Ethanol Degradation IV' (Fig. 1B; Table SVI); the remaining 12 pathways predominantly involved enzymes from the ALDH and ADH families. Notably, all of these genes consistently showed significant downregulation, with the commonly downregulated genes being ADH1B, ALDH2, ALDH3A2 and ALDH9A1. These enzymes were also found to be downregulated in tumor compared with normal tissues in TCGA/HNC cohort (Fig. 1C). Survival analysis using TCGA/HNC cohort identified ALDH2 as the only gene significantly associated with overall survival (Fig. 1D).

To investigate the differential protein expression of ALDH2 in HNC, the present study analyzed 60 tumor specimens and their paired adjacent normal tissues using IHC. ALDH2 was significantly downregulated in tumor specimens compared with adjacent normal tissue (Fig. 2A and B), To assess the clinical role of ALDH2 expression in HNC, TMA-1 from 106 patients was evaluated. Low ALDH2 expression in HNC tissue was significantly correlated with higher AJCC T classification, advanced overall stage and increased recurrence rates (Table I). Survival analysis using log-rank test revealed that patients with high ALDH2 expression (n=50) had significantly better overall survival (OS) and disease-free survival (DFS) rates than those with low expression (n=56) (Fig. 2C). Univariate analyses identified low ALDH2 expression and high AJCC T and N stage as poor prognostic factors for OS, whereas multivariate analysis showed the no significance of ALDH2 after adjusting for other variables (Fig. 2D; Table SVII). High ALDH2 expression was an independent favorable factor for DFS (Fig. 2D; Table SVIII).

Functional roles of ALDH2 in HNC cell proliferation, migration and invasion were assessed using MTT and Boyden chamber assay. ALDH2 was most highly expressed in DOK cells, an oral dysplasia cell line (15) and TW2.6 cells, with the lowest expression observed in SAS cells (Fig. S1A). Based on these expression levels, ALDH2 was knocked down in DOK and TW2.6 cells and ALDH2 was overexpressed in SAS cells. As an oral dysplasia cell line derived from a pre-malignant lesion, DOK cells exhibit higher ALDH2 expression levels and lower invasive and migratory ability compared with other HNC cell lines, making them an ideal model for investigating the early stages of HNC progression and ALDH2 tumor-suppressive function (15,16). Following lentivirus-mediated knockdown of ALDH2 via shRNA, western blot and RT-qPCR showed that endogenous ALDH2 was downregulated in DOK and TW2.6 cells (Fig. 3A). ALDH2 knockdown in DOK and TW2.6 cells significantly increased migration and invasion abilities compared with shluc-infected control cells (Fig. 3B). Conversely, overexpression of ALDH2 in SAS cell significantly reduced their ability to invade and migrate (Fig. 3C and D). Similarly, colony formation assays showed a significant increase in colony-forming ability in ALDH2-knockdown HNC cells (Fig. 3E). Conversely, ALDH2 overexpression significantly reduced colony formation (Fig. 3F). However, neither overexpression nor knockdown of ALDH2 significantly affected the proliferation of HNC cells (Fig. 3G).

MTCQ1 cells (derived from 4-nitroquinoline 1-oxide-induced tongue carcinoma in C57BL/6 mice) were used to validate the aforementioned results. MTCQ1 cells are characterized by higher invasiveness and clonogenic potential but lower proliferation rate than SAS cells (17). ALDH2 overexpression in MTCQ1 cells significantly decreased migration, invasion and colony formation without significantly affecting cell proliferation (Fig. S1B-E). These results indicate that ALDH2 influences migration, invasion and colony formation in HNC cells, including DOK cell.

To elucidate the molecular mechanism mediated by ALDH2, gene expression microarray analysis of ALDH2-knockdown (TW2.6/shALDH2) and control cells (TW2.6/shluc) was performed. ALDH2 knockdown resulted in the upregulation of 980 and the downregulation of 748 genes, with an absolute 2-fold change as the threshold. IPA on these differentially expressed genes, identified tumor necrosis factor (TNF) as the top upstream regulator (Fig. S2A; Table SIX). Pathway analysis showed that 'NF-κB signaling' pathway was significantly upregulated following ALDH2 knockdown, with a z-score of 2.324 (Table SX).

Effect of ALDH2 on NF-κB signaling was assessed in ALDH2-knockdown cells. Knockdown of ALDH2 in TW2.6 cells resulted in increased levels of phosphorylated (p)NF-κB (Fig. 4A). Conversely, ALDH2 overexpression in SAS cells led to a significant decrease in pNF-κB expression (Fig. 4B). Silencing RELA) using siRNA markedly inhibited the migration and invasion activity of ALDH2-knockdown cells (Figs. 4B and S2B and C). These results suggested that ALDH2 mediated cell migration and invasion through NF-κB.

Analysis of microarray data from ALDH2-knockdown TW2.6 cells suggests that VEGFC is a key factor in cancer progression and is regulated by NF-κB (18). RT-qPCR confirmed that VEGFC was significantly upregulated in ALDH2-knockdown cells (Fig. 4D), whereas its expression was notably decreased in ALDH2-overexpressing cells (Figs. 4E and S2D). ALDH2 knockdown enhanced cell migration and invasion, which were significantly suppressed following treatment with a VEGFC-neutralizing antibody (Fig. 4F). To determine whether NF-κB mediates VEGFC expression in ALDH2-knockdown cells, NF-κB inhibitor Bay11-7082 was used, which inhibits IκBα phosphorylation (19). The present results showed a significant decrease in VEGFC levels and migration and invasion of TW2.6/shALDH2 cells upon NF-κB inhibition (Fig. 4G and H). These findings confirm that ALDH2 suppression activates the NF-κB/VEGFC axis, leading to aggressiveness of HNC cells.

ALDH2 and VEGFC expression in HNC tissue was examined via immunohistochemistry in 84 HNC samples from TMA-2. This revealed a negative correlation (Spearman correlation coefficient=-0.173; Fig. 5A), suggesting a potential suppressive effect of ALDH2 on VEGFC. Although VEGFC expression alone did not significantly influence OS (Fig. 5B), the combination of high ALDH2 and low VEGFC levels was associated with greater survival probability compared with the low ALDH2 and high VEGFC (Fig. 5C). In transcriptomic data of 502 patients from TCGA/HNC cohort, there was a consistent inverse relationship between ALDH2 and VEGFC (r=-0.245, Fig. 5D), supporting the protein-level findings. Additionally, GEPIA2 platform showed significantly higher VEGFC levels in tumor tissue (n=519) compared with normal (n=44; Fig. 5E). Survival analysis further demonstrated that patients with low VEGFC levels (VEGFC^low, n=248) had improved OS (Fig. 5F). Further analysis of the combined expression levels of ALDH2 and VEGFC revealed that ALDH2^high/VEGFC^low (n=134) had the most favorable survival outcome, whereas patients with ALDH2^low/VEGFC^high (n=135) exhibited the poorest prognosis (Fig. 5G).

Based on enhancement of malignant traits in HNC cells following ALDH2 knockdown, the present study aimed to determine the effect of modulating the enzyme activity of ALDH2 using an antagonist and agonist, particularly focusing on its effects mediated by NF-κB/VEGFC signaling pathway. Daidzin (isoflavone glycoside), a specific ALDH2 antagonist, was used to treat TW2.6 cells. Daidzin resulted in marked increase in pNF-κB expression in TW2.6 cells (Fig. 6A), accompanied by enhanced VEGFC levels, however this was not significant (Fig. 6B). In ALDH2-overexpressing SAS cells, daidzin treatment not only increased VEGFC levels but also restored the migration and invasion capacities of SAS/ALDH2 cells (Figs. 6C and D and S3A). To investigate the potential of ALDH2 agonist Alda-1 for decreasing cancer aggressiveness, its effect on viability in SAS cells was assessed at various concentrations. Alda-1 treatment showed minimal cytotoxicity in SAS cells, with <25 μM showing no significant decrease in cell viability across the tested range.Hovever, 50 μM of Alda-1 significantly reduced cell viability in SAS cells (Fig. S3B). Furthermore, a previous study showed that treatment with 20 μM Alda-1 increased ALDH2-mediated VISTA (V-domain Ig suppressor of T-cell activation) expression in breast cancer cells (20). Based on this evidence, we used a concentration of 20 μM of Alda-1 for subsequent experiments. At this concentration, Alda-1 significantly reduced VEGFC levels and markedly decreased the migratory and invasive ability of SAS cells (Figs. 6E and S3C).

Since pharmacological modulation of ALDH2 activity regulated VEGFC-mediated HNC migration and invasion similarly to ALDH2 expression modulation, the present study investigated whether these effects were associated with alcohol metabolism. pNF-κB and VEGFC expression were significantly upregulated in TW2.6 cells following treatment with acetaldehyde, the primary metabolite of ethanol (21) (Fig. 6F and G). Furthermore, VEGFC expression levels in the groups treated with both acetaldehyde and Alda-1 were comparable with those treated with Alda-1 alone (Fig. 6G).

Acetaldehyde may promote ROS production to enhance oxidative stress (22). NF-κB is a redox-sensitive transcription factor that can be either activated or deactivated by ROS (23). The present study investigated whether ALDH2 regulates NF-κB through ROS modulation. ALDH2 overexpression significantly reduced ROS levels, while ALDH2 knockdown increased ROS levels in HNC cells (Fig. 7A and B). NAC), a free radical scavenger, in ALDH2-knockdown TW2.6 cells significantly suppressed NF-κB activity (Fig. 7C). H₂O₂ led to a marked increase in VEGFC levels in ALDH2-overexpressing SAS cells compared with those without H₂O₂ treatment (Fig. 7D). These results collectively suggested that ALDH2 modulates the NF-κB/VEGFC axis by regulating ROS production in HNC cells (Fig. 7E).