Noninvasive early diagnosis of laryngeal squamous cell carcinoma (LSCC) revealing the high-performance potential of the HOXD10 gene monomethylated locus: a case–control study Authors: Xiaoyang Gong, Zhaoyi Lu, Zihang Zhu, Qin Zhang, Haibing Chen, Han Zhou, Qianqian Duan, Chan Zhu, Dongsheng Chen, Lei Cheng, Xin Cao, Xi Chen Published: 2025 ABSTRACT AbstractLaryngeal squamous cell carcinomas (LSCCs) was a frequent malignancy in upper aerodigestive tract. The lack of non-invasive biomarkers for early diagnosis has led to unfavorable prognosis. We firstly identified methylation biomarkers from TCGA and GEO databases. We further validated the diagnostic performance by utilizing blood samples obtained from 43 LSCCs and 50 normal controls (NCs). The HOXD10 cg10364040 was hypermethylated in tumors compared with NCs (P < 0.01), with an area under the curve (AUC) greater than 0.90 in each public database. In clinical validation phase, HOXD10 cg10364040 methylation was able to discriminate LSCCs from NCs with high accuracy (AUC = 0.767, sensitivity: 72.1%, specificity: 82.0%), and exhibited greater sensitivity (85.0%) and specificity (86.0%) in stage I patients than in NCs (AUC = 0.864). In summary, we firstly identified cg10364040 as a novel diagnosis biomarker of LSCC and confirmed it as a noninvasive biomarker in clinical.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13148-025-02036-2.Keywords: DNA methylation, Laryngeal squamous cell carcinoma, Early diagnosis, HOXD10, Noninvasive diagnostic marker Introduction Laryngeal squamous cell carcinoma (LSCC) is one of the common malignant neoplasms of the head and neck squamous cell carcinoma (HNSCC) [1, 2]. Early signs of LSCC tend to be subtle, leading to a situation where over 50% of patients are diagnosed at intermediate or advanced stages. At these stages, the disease is often characterized by infiltration and lymphatic metastasis [3]. Early detection plays a pivotal role in increasing the survival rates of these patients; however, limited clinical achievements have been made in the development of efficacious and noninvasive diagnostic approaches on early detection. Traditional noninvasive serum tumor markers, such as carcinoembryonic antigen (CEA) quantification, squamous cell carcinoma-associated antigen (SCC), and cytokeratin 19 fragment antigen (CYFRA21-1), have shown good specificity, but their use has been limited by their low sensitivity, which ranges from 51.4 to 57% [4, 5]. Laryngoscopy is often considered the best method for early visual detection of laryngeal cancer, and it can be effectively complemented by CT or MRI examination to facilitate comprehensive tumor assessment and subsequent treatment planning [6]. However, this examination process is highly dependent on the personal habits, clinical experience and state of the examiner at the time of examination. If laryngoscopy is not standardized, it can lead to missed detection or misdiagnosis of key pathological changes, which will have a profound negative impact on the long-term health and functional recovery of patients with head and neck cancer. Therefore, we need to develop the biomarkers for effective supplement for laryngoscopy screening of LSCCs. Recent advancements in liquid biopsy technologies have paved the way for earlier cancer detection by analyzing cell-free DNA (cfDNA), including its methylation status [7, 8], mutations [9], and/or fragmentations [10], or other circulating analytes, such as proteins [11]. Among recent advancements, blood methylation detection technology has garnered considerable attention because of its advantages of being tumor specific, occurring in the early stages of tumors or even precancerous lesions, and not being interfered with by clonal hematopoiesis [12, 13]. DNA methylation markers are aberrant alterations in DNA methylation in cancer cells, and this aberrant methylation occurs throughout the entire process of cancer development and progression. Compared with traditional tumor markers, blood DNA methylation markers have the advantages of presenting earlier and being noninvasive and more accurate [14]. For instance, a multicenter prospective study involving 1381 patients indicated that RNF180 and SEPTIN9 methylation serve as promising markers for gastric adenocarcinoma, and their sensitivity (62.2%, 95% CI 58.3%–65.9%) for early gastric cancer (stage I and II) is much greater than that of traditional protein tumor markers, such as CEA, CA199, and CA125 [15]. The Circulating Cell-free Genome Atlas (CGGA) study demonstrated that a blood-based multicancer early detection (MCED) test utilizing cfDNA methylation sequencing in combination with machine learning could detect cancer signals across multiple cancer types and predict cancer signal origin (CSO) with high accuracy [16]. Although complex multigene models show great diagnostic potential, their clinical application remains challenging. Conversely, several studies highlight the remarkable performance of single-gene or site-specific blood methylation markers in early cancer detection because of their clinical feasibility and health economics. For example, a study by Luo et al. [17] demonstrated that the cg10673833 methylation test successfully identified 19 out of 21 colorectal cancer (CRC) patients and detected 7 out of 8 cases of CRC in situ, yielding a sensitivity of 89.7% (95% CI 0.727–0.978) and specificity of 86.8% (95% CI 0.849–0.884) among 16,890 participants. Thus, investigating high-performance, single-gene or site-specific blood methylation markers holds promise for early clinical screening and diagnosis. Homeobox D 10 (HOXD10) encodes a protein with a homeobox DNA-binding domain. It plays distinct roles in different types of cancer. Specifically, it functions as a tumor suppressor in colorectal and gastric cancers, whereas it exhibits a dual role in breast cancer [18]. Previous studies have shown that HOXD10 is closely related to the occurrence of HNSCC and may serve as a biomarker for HNSCC [19, 20]. The dysregulation of HOX genes, including HOXD10, exerts functional effects through various DNA and RNA epigenetic modifications that influence both transcriptional and post-transcriptional control mechanisms, thereby promoting the development and progression of oral cancer [20]. A research investigation conducted DNA amplification and analyzed methylation patterns in tissue specimens obtained from oral squamous cell carcinoma (OSCC) patients, including both tumor tissues and adjacent surgical margins. HOXD10 was found to be amplified in the surgical margin tissue and exhibited high levels of methylation in approximately half of the tumor samples, suggesting that HOXD10 expression is regulated by DNA methylation in OSCC [21]. However, whether HOXD10 can be used as an early diagnostic indicator in LSCC remains unclear. Currently, some methylation markers have also been identified in LSCC. Shen et al. [22]. found that hypermethylation of the SHISA3 promoter might increase the risk of LSCC by regulating gene expression and could serve as a potential diagnostic and prognostic biomarker for LSCC by comparing tumor tissues with adjacent normal tissues. The same team also discovered that elevated methylation of CMTM3 is a risk factor for male LSCC patients, especially those over 55 years old and with a history of smoking [23]. However, these studies did not further investigate the diagnostic performance or the clinical application potential of this methylation biomarker in LSCC. In our study, we aimed to evaluate the potential of HOXD10 methylation markers for early LSCC diagnosis. We constructed methylation marker models by screening public data via lasso, random forest, and logistic statistical methods. Additionally, we assessed the sensitivity and specificity of this methylation-based early detection approach for LSCCs in a prospectively recruited Chinese cohort. Materials and methods Study design The aim of this study was to identify methylation biomarkers in blood for early diagnosis of LSCC (Fig. 1A). In the database analysis phase, we identified methylation biomarkers from The Cancer Genome Atlas (TCGA) database and validated them from the TCGA and Gene Expression Omnibus (GEO) databases. In the clinical validation phase, we prospectively and continuously enrolled participants who came to the Department of Otorhinolaryngology, First Affiliated Hospital, Nanjing Medical University, from November 2023 to May 2024. This study has been registered on the Chinese Clinical Trial Registry (ChiCTR, https://www.chictr.org.cn/) with the registration number ChiCTR2300077259 and a registration date of November 2, 2023. The study was approved by the institutional ethics committee of the hospital (2023-SR-373). Participants provided written informed consent and a blood sample before any medical preparation for laryngoscopy. This work has been reported in line with the STARD criteria [24]. Flow diagram of the research process and model generation. A Enrollment and prospective screening cohort study. B Flow chart for constructing diagnostic models for methylation markers via public databases. C Unsupervised hierarchical heatmap clustering of 13 candidate methylation biomarkers differentially methylated between HNSCC and normal subject DNA in the training cohort and validation cohort (HNSCC:normal = 528:656. Each row represents one patient, and each column represents one CpG marker. D Unsupervised hierarchical heatmap clustering of 14 candidate methylation markers that are differentially methylated between HNSCC and normal subject DNA in the training cohort and validation cohort (HNSCC: normal = 528:656). The row represents the CpG marker and column represents the patient. E Venn diagram of HNSCC biomarker methylation sites identified via LASSO (left) and random forest (right) methods. AJCC, American Joint Committee on Cancer; HGSIN, high-grade squamous intraepithelial neoplasia; HNSCC, head and neck squamous cell carcinoma; TCGA, cancer genome atlas DNA methylation model construction DNA methylation data for HNSCC samples and normal control samples were downloaded from the TCGA (https://xenabrowser.net/) [25] and GEO databases (normal control samples: GSE40279 [26], HNSCC: GSE75537 [27]). The DNA methylation profiles provided by TCGA and GEO were generated using the 450k array. The R package ChAMP was used to analyze the 450 K methylation array data. First, we applied the champ.filter function to remove probes with a detection p-value exceeding 0.01, probes located at non-CpG sites, probes associated with single nucleotide polymorphisms (SNPs), multi-hit probes, and probes located on the X and Y chromosomes. Next, normalization was carried out using the champ.norm function to correct for signal bias between type-I and type-II probes. Batch effects were adjusted using the champ.runCombat function. Complete clinical, molecular, and histopathological datasets of all HNSCC patients are available at the TCGA website. The DNA methylation data of HNSCC patients from TCGA and all normal control individuals (GEO40279) were randomly divided into a training set and a validation set at a 2:1 ratio. DNA methylation data from HNSCC patients in the GSE75537 and GSE178218 cohort were used as external independent verification set. The detailed study flowchart is shown in Fig. 1B. Data analysis, including filtering, imputing and normalization, was performed via the R package ChAMP from Bioconductor with the default settings [28]. We performed differential methylation analysis on the training dataset, which included patients with HNSCC tumors and healthy donors. Differential methylation probes (DMPs) between tumors and normal controls were run in ChAMP. Differentially methylated probes met the following conditions: Benjamini‒Hochberg (BH) and false discovery rate (FDR) < 0.05 and delta beta < 0.4. We applied two variable selection methods suitable for high dimensionality on the above intersection marker list: the least absolute shrinkage and selection operator (LASSO) [29] and Random Forest [30]-based variable selection methods using OOB error. For LASSO regression, the dataset was subsampled without replacement 500 times, and LASSO regression was applied to each subsample. A biomarker was retained in the final model if it exhibited a non-zero coefficient in more than 90% of the subsamples (i.e., in over 450 iterations), thereby ensuring high selection stability. The regularization parameter (lambda) was determined via 10-fold cross-validation. To favor model parsimony, we selected the lambda.1se value—the largest lambda within one standard error of the minimum cross-validation error—thus promoting a simpler yet effective model. For random forest analysis, the random forest model was trained using 3000 trees to minimize variance in variable importance estimates. Out-of-Bag (OOB) error was utilized as an unbiased internal estimate of classification performance. Variable importance was assessed using the mean decrease in Gini impurity, calculated from OOB observations. To optimize the feature set, we conducted an iterative backward elimination procedure: in each of 2,000 iterations, the least important 30% of variables—based on current importance scores—were excluded, and a new forest was constructed. The OOB error of the resulting model served as the objective criterion for further refinement, ensuring that only variables contributing to improved predictive accuracy were retained. We then selected one overlapping methylation marker by the two methods to construct a binary prediction. We fitted a logistic regression model using this marker and obtained a diagnosis score in the TCGA training datasets. Collection of blood samples All samples were collected using blood collection tubes with a procoagulant evenly coated on the inner wall. For the validation cohort, 120 blood consecutive samples were prospectively collected. Patients with HNSCC were selected based on the following criteria: (I) age ≥ 18 years, (II) pathological diagnosis of primary HNSCC with no history of other malignancies, (III) availability of peripheral blood for testing, (IV) ability of subjects to understand and provide written informed consent to participate in the study, and (V) ability to follow the study protocol and follow-up procedures. Normal control individuals were free of malignant disease. Finally, 16 samples were excluded from analysis due to (1) head and neck tumors that were not squamous carcinomas or (2) samples without sufficient blood DNA extraction or failed quality control. The remaining 104 blood samples (from 43 laryngeal squamous cell carcinoma patients, 11 high-grade squamous intraepithelial neoplasia (HGSIN) patients and 50 normal control individuals) were used for DNA methylation detection and validation. DNA preparation and methylation detection Plasma samples were frozen at − 80 °C until use. The plasma DNA was extracted using the blood genomic DNA Extraction kit (Concertbio, Xiamen, China) with optimized manufacturer’s protocols. After DNA was extracted, the DNA concentration and quality were determined. The quality of genomic DNA was assessed using the NanoDrop 2000 (NanoDrop Technologies, Wilmington, DE, USA). The criteria for acceptable DNA quality included a concentration of at least 20 ng/µL, a total amount of no less than 400 ng, an OD260/280 ratio ranging from 1.7 to 1.9, and an OD260/230 ratio of 2.0 or higher. 20 ng of DNA was subjected to bisulfite conversion according to the EZ DNA Methylation-Gold™ Kit. PCR amplification was performed using Methylation FastTarget V4.1 software (GENESKY, Shanghai) to design sequencing primers for the target region of the human HOXD10 gene (Supplemental Table S1). Using primers with Index sequences, specific tag sequences compatible with the Illumina platform were introduced to the end of the library by PCR amplification. After quantification of the library, high-throughput sequencing was performed using the Illumina HiSeq platform in 2 × 150 bp double-end sequencing mode to obtain FastQ data. Methylation data analysis After evaluating the raw sequencing data of each sample, R1/R2 reads were merged using Flash software (FLASH: fast length adjustment of short reads to improve genome assemblies). The valid reads were then filtered using BLAST+ (Camacho C, (2009) “BLAST+: architecture and applications”) for comparison with the reference cohort. Subsequently, the methylation levels of the CpG sites in the amplicons were calculated, and a higher β value suggested greater methylation of the target gene sites. Traditional serum tumor biomarker and HPV test The levels of the traditional HNSCC biomarkers CEA, CYFRA21-1, AFP, CA19-9, CA724, NSE and SCC were analyzed using a chemical immunoassay (Roche, Switzerland) [31–35]. The cutoff values for CEA and CYFRA21-1 were 5 ng/mL and 3.3 ng/mL, respectively. The HPV status was determined using the p16 immunohistochemistry (IHC) method [36]. Statistical analysis The diagnostic efficacy of the model was evaluated by receiver operating characteristic (ROC) curves, areas under the curve (AUCs), specificity and sensitivity [37]. Confusion tables were generated using an optimized diagnosis score cutoff with a maximum Youden’s index (specificity + sensitivity − 1) [38]. The Mann‒Whitney U test (false discovery rate < 0.05; fold change >2) was used to assess the differential expression of HOXD10 cg10364040 in tumor and normal samples, and p < 0.05 was considered to indicate a significant difference. In addition, odds ratio (OR) and its 95% confidence interval was also computed in clinical validation phase. Results Identified cg10364040 as a diagnostic methylation marker The samples from the TCGA and GEO cohorts were randomly assigned to the training and validation sets at a 2:1 ratio, as shown in Fig. 1B. A total of 1139 differentially expressed methylation sites were identified between HNSCC patients and normal control individuals (P < 0.05). LASSO Cox regression analysis further identified 13 key methylation sites (Fig. 1C, Supplemental Figure S1A and S1B), and the random forest algorithm identified 14 key methylation sites (Fig. 1D, Supplemental Figure S1C). Finally, one methylation biomarker (cg10364040) overlapped between these two screening criteria (Fig. 1E). Performance of cg10364040 methylation As the primary aim of this study was to explore early screening markers for HNSCC, we validated them in public databases. Based on LASSO and random forest analysis via a Venn diagram, we identified only one methylation site, HOXD10 (cg10364040), and evaluated its performance as a biomarker for distinguishing HNSCC tissue from normal tissue (Fig. 1E). According to the TCGA training database, the β value of HOXD10 (cg10364040) in tumors was significantly greater than that in normal control individuals (p < 0.01, Fig. 2A), and the AUC was 0.999 (95% CI 0.998–1.000, specificity: 0.993, sensitivity: 0.989; Fig. 2B). According to the validation database, the β value of the cg10364040 site in tumors was also significantly greater than that in normal controls (p < 0.01, Fig. 2C), and the AUC was 1.000 (95% CI 0.999–1.000, specificity: 1.000, sensitivity: 0.994; Fig. 2D). Moreover, two independent cohort was used to validated the performance of the cg1036404. In independent validation cohort 1, the β value of the cg10364040 site was significantly greater in the tumor group than that in the normal control group (p < 0.01, Fig. 2E), and the AUC was 0.923 (95% CI 0.864–0.981, specificity: 0.926, sensitivity: 0.852; Fig. 2F). In independent validation cohort 2, the β value of the cg10364040 site was also significantly greater in the tumor group (p < 0.01, Fig. 2G), with AUC = 0.950 (95% CI 0.852–1.000, specificity: 1.000, sensitivity: 0.950; Fig. 2H). Furthermore, the methylation level was significantly lower in normal control individuals than in HNSCC patients (Fig. 2I). These results strongly suggest that this tissue-based biomarker, hypermethylated HOXD10 cg10364040, can serve as a novel biomarker for the early detection of HNSCC. Expression and diagnostic value analysis of cg10364040 methylation of HOXD10 in the training and validation cohorts. A Methylation level of the HOXD10 cg10364040 site in the normal control and tumor groups (HNSCC:normal = 352:437). B ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between HNSCC and normal samples (HNSCC:normal = 352:437). AUC = 0.999 (95% CI 0.998-1–000). C Methylation level of HOXD10 cg10364040 in the normal and tumor groups in the validation cohort (HNSCC:normal = 176:219). D ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate HNSCC patients from normal controls in the validation cohort (HNSCC:normal = 1176:219). AUC = 1.000 (95% CI 0.999–1.000). E Methylation level of HOXD10 cg10364040 in the normal and tumor groups in the dependent validation cohort (HNSCC:normal = 54:54). F ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate HNSCC patients from normal controls in the dependent validation cohort 1 (HNSCC:normal = 54:54). AUC = 0.923 (95% CI 0.864–0.981). G Methylation level of HOXD10 cg10364040 in the normal and tumor groups in the dependent validation cohort 2 (HNSCC:normal = 20:11). H. ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate HNSCC patients from normal controls in the dependent validation cohort 2 (HNSCC:normal = 20:11). AUC = 0.950 (95% CI 0.852–1.000). I Box plots presenting the β value distribution of HOXD10 cg10364040 methylation at different clinical stages (Ⅰ:Ⅱ:Ⅲ:Ⅳ = 27:74:82:270) in the database Clinical participant information In the clinical validation phase, we collected 104 eligible blood samples for testing (Supplemental Figure S2). Normal control (NC) participants were diagnosed mainly by CT and laryngoscopy. They were followed up 1–2 months after diagnosis, and no abnormalities were found. The final diagnoses of high-grade squamous intraepithelial neoplasia (HG) and HNSCC were based on postoperative pathological findings. The demographics and clinical characteristics of the 104 participants are presented in Table 1. The patients with LSCC in our cohort were older than the NCs were (66 vs. 58 years) and were more likely to be male (100% vs. 60%), have a smoking history (90.7% vs. 32%), and have a drinking history (67.4% vs. 34%). Table 1Clinicopathologic characteristic of participantsLSCCHGNormal control Total 431150 Gender Male (%)43 (100%)11 (100%)30 (60%)Female (%)0 (0%)0 (0%)20 (40%) Age Median (range)66 (41–88)66.5 (42–85)58 (34–81) ECOG score 0 (%)38 (88.4%)9 (81.8%)49 (98%)1 (%)4 (9.3%)2 (18.2%)1 (2%)2 (%)1 (2.3%)0 (0%)0 (0%) Smoking Yes (%)39 (90.7%)11 (0%)16 (32%)No (%)4 (9.3%)0 (0%)34 (68%) Drinking Yes (%)29 (67.4%)9 (81.8%)17 (34%)No (%)14 (32.6%)2 (18.2%)33 (66%) HPV status Positive12 (27.9%)2 (18.2%)1 (2.0%)Negative31 (72.1%)9 (81.8%)49 (98.0%) Tumor Stage T1 (%)20 (46.5%)––T2 (%)9 (20.9%)––T3 (%)6 (14.0%)––T4 (%)8 (18.6%)––LSCC laryngeal squamous cell carcinoma, HG high-grade squamous intraepithelial neoplasia Clinicopathologic characteristic of participants LSCC laryngeal squamous cell carcinoma, HG high-grade squamous intraepithelial neoplasia Diagnostic performance of HOXD10 cg10364040 methylation in the clinical validation cohort Next, to establish the noninvasive diagnostic potential of HOXD10 (cg10364040) for HNSCC, its performance in blood samples (from 43 patients with LSCC, 11 patients with HGSIN and 50 normal control individuals, Supplemental Table S2) obtained from the clinical cohort was determined. As shown in Fig. 3A, the β value of the cg10364040 site in the blood of LSCC patients was significantly greater than that in the blood of normal control individuals (p < 0.01) and those with HGSIN (p = 0.036). Since the aim of this study was to distinguish tumors from non-tumors, we further validated the diagnostic performance of the cg10364040 site in LSCC patients (n = 43) and normal control individuals (n = 50) and demonstrated excellent diagnostic performance, with an AUC of 0.767 (95% CI 0.664–0.870, specificity: 82%, sensitivity: 72.1%, OR = 9.92, 95% CI 3.90-27.41; Fig. 3B). However, cg10364040 had relatively poor diagnostic performance in identifying HGSIN and LSCC, with an AUC of 0.706 (95% CI 0.532–0.880, specificity: 63.6%, sensitivity: 76.7%, Supplemental Figure S3). HGSIN is a precancerous lesion. The diagnostic performance of methylation detection in differentiating normal patients from precancerous lesions is indeed somewhat inferior [15]. Expression and diagnostic value analysis of cg10364040 methylation of HOXD10 in the clinical validation cohort. A Methylation level of HOXD10 cg10364040 in the blood of the normal control (NC), high-grade squamous intraepithelial neoplasia (HG) and LSCC groups in the clinical validation cohort (NC:HG:LSCC = 50:11:43). B ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between the blood of LSCC patients and normal controls in the clinical validation cohort (normal:LSCC = 50:43). AUC = 0.767 (95% CI 0.664–0.870). C ROC curve corresponding to HOXD10 cg10364040 methylation plus CYFRA21-1 when used to differentiate between LSCC patients and normal controls in the clinical validation cohort (normal:LSCC = 20:29). AUC = 0.824 (95% CI 0.708–0.941) Further, according to the clinical status of patients, we compared the differences in methylation levels of HOXD10 cg10364040 in patients with HPV infection status (Supplemental Figure S4). The analysis indicated that there was also no statistical difference in the methylation level of HOXD10 cg10364040 (P = 0.31) and diagnostic performance between HPV negative and HPV positive patients (Supplemental Figure S5). We also tested traditional serum biomarkers (Supplemental Figure S6, Supplemental Table S2), including CEA (AUC, 0.663; NC:LSCC = 21:29), CYFRA21-1 (AUC, 0.722; NC:LSCC = 20:29), AFP (AUC, 0.633; NC:LSCC = 19:22), CA19-9 (AUC, 0.538; NC:LSCC = 19:22), CA724 (AUC, 0.539; NC:LSCC = 18:22), NSE (AUC, 0.617; NC:LSCC = 21:29) and SCC (AUC, 0.844; NC:LSCC = 2:8), in which the AUC value of CYFRA21-1 was superior to that of other biomarkers but was inferior to that of cg10364040 in the clinical cohort. Thus, cg10364040 methylation has superior diagnostic performance to traditional serum markers in LSCC. Diagnostic performance of HOXD10 cg10364040 methylation plus CYFRA21-1 Given the good diagnostic ability of CYFRA21-1, we sought to combine cg10364040 methylation and CYFRA21-1 to determine whether there was an improvement in the diagnosis of LSCC patients and normal control individuals. As shown in Fig. 3C, the results showed that cg10364040 methylation plus CYFRA21-1 had a better AUC (AUC, 0.824, 95% CI 0.708–0.941, sensitivity: 72.4%, specificity: 85.0%; OR = 13.47, 95% CI 3.40–73.68). Hence, this blood-based methylation biomarker combined with traditional biomarkers for detection has better diagnostic performance. Diagnostic performance of HOXD10 cg10364040 methylation at different clinical stages Interestingly, we found that the methylation level of cg10364040 was negatively correlated with clinical stage, with significantly greater methylation at earlier stages. The methylation level of cg10364040 was significantly greater in stage I LSCC patients than in normal control individuals, and the difference between stage II/III/IV patients and normal control individuals was not significant (Fig. 4A). While in other cancers, similar situations have been reported in the literature [39–41]. Next, to characterize the potential of this methylation marker for detecting early-stage LSCC tumors, we generated ROC curves. As expected, the cg10364040 methylation marker could distinguish stage I LSCC patients from normal control individuals (AUC, 0.864; 95% CI 0.750–0.978; OR = 31.02, 95% CI 8.00–168.94), with good specificity (85%) and sensitivity (86%) (Fig. 4B). Furthermore, we validated the prognostic performance of cg10364040 methylation combined with CYFRA21-1, and the AUC (0.887, 95% CI 0.691–1.000; OR = 43.98, 95% CI 4.84–1470.74), specificity (90%) and sensitivity (87.5%) were greater than those of cg10364040 methylation alone (Fig. 4C). Identification of cg10364040 methylation of HOXD10 in different clinical stages. A Box plots presenting the β value distribution of cg10364040 methylation of HOXD10 in blood samples from different clinical stages (Ⅰ:Ⅱ:Ⅲ:Ⅳ = 20:9:6:8). B ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between the blood of LSCC patients (stage Ⅰ) and normal controls in the clinical validation cohort (normal:LSCC stage Ⅰ = 50:20). AUC = 0.864 (95% CI 0.750–0.978). C The ROC curve corresponding to the HOXD10 cg10364040 site combined with CYFRA21-1 when used to differentiate between the blood of LSCC patients (stage Ⅰ) and normal controls in the clinical validation cohort (normal:LSCC stage Ⅰ = 20:8). AUC = 0.887 (95% CI 0.691–1.000) Discussion Laryngeal cancer is a type of malignancy that originates in the larynx, with the majority of these cancers being well-differentiated squamous cell carcinomas [3]. Unfortunately, despite some symptoms, such as hoarseness, painful swallowing, can develop early in patients with laryngeal malignancies, the majority of patients receive their diagnosis at more advanced stages, with over 75% being diagnosed at stage III or IV [42, 43]. For patients who can be treated, the 5-year survival rate is approximately 80% for glottic cancer and about 50% for supraglottic cancer [44]. Early detection of head and neck squamous cell carcinoma (HNSCC) is critical for improving treatment outcomes. However, current screening strategies, apart from those for nasopharyngeal carcinoma, have not shown significant efficacy, with meticulous physical examinations remaining the primary method for early detection [2]. For asymptomatic individuals at risk of nasopharyngeal carcinoma, plasma EBV DNA analysis has demonstrated impressive sensitivity (97.1%) and specificity (98.6%), suggesting that innovative, noninvasive screening methods are essential [45]. Usually, standard clinical practices often include the assessment of specific serological markers, such as carcinoembryonic antigen (CEA) and CYFRA21-1, in patients suspected of having HNSCC. However, our study revealed that the sensitivities of these markers were low, at 10.34% and 24.13%, respectively (Supplemental Table S3). This underscores the need for better biomarkers, especially as these traditional markers lack sufficient diagnostic efficacy. Therefore, in this study, we aimed to identify a high diagnostic performance early detection marker and first identified and validated the HOXD10 methylation site (cg10364040) as a promising noninvasive biomarker for early LSCC diagnosis. The present study is the first to successfully identify and validate a blood-based biomarker for the early screening of LSCC through a single methylation site. These findings provide crucial guidance for precise diagnosis and treatment strategies. In recent years, DNA methylation-based sequencing platforms, particularly noninvasive plasma-derived cfDNA methylation analysis, have emerged as promising early pancancer screening strategies [7]. The methylation patterns of oncogenes and tumor suppressor genes may be detectable at the early stages of malignant transformation, indicating their potential as reliable discriminatory markers for malignancy detection and diagnosis [46]. The multigene-methylated MCDBT-1 panel had excellent detection performance (sensitivity 69.1%, specificity 98.9%) in multiple cancers (all stages). It has demonstrated excellent detection performance in early-stage (I + II) colorectal cancer, esophageal cancer, liver cancer, lung cancer, ovarian cancer and pancreatic cancer, with sensitivities of 50.0%, 32.0%, 79.5%, 28.3%, 55.6% and 26.5%, respectively [47]. Accordingly, the corresponding “OverC™ Multicancer Early Detection Kit” has also obtained a CE registration license for in vitro diagnostic reagents in the European Union (identification number:DE/CA20/01-IVD-Luxuslebenswelt-523/22, DE/CA20/01-IVD-Luxuslebenswelt-524/22). The multicancer early detection (MCED) test of the CGGA study, which integrates blood cfDNA multisite methylation with machine learning to establish an early screening model, demonstrated a sensitivity of 85.7% in detecting patients with head and neck cancer (n = 105, all stages). The sensitivity of cancer detection increases proportionally with advanced stage (stage I: 63.2%, stage II: 82.4%, stage III: 84.2%, and stage Ⅳ: 96.0%) [16]. Although our sensitivity in LSCC samples was lower than that reported in the CGGA study (72.1% vs. 85.7%), we achieved excellent sensitivity (86%) in diagnosing stage I LSCC. Furthermore, our study employed a one-site methylation detection approach, which not only simplifies the model building and detection methods but also significantly reduces costs. Thus, it is highly feasible for clinical applications. Dong et al. [48] explored single-gene methylation of multiple cancer early screening markers. These authors found that PCDHGB7 was hypermethylated not only in all 17 cancer types in the TCGA database but also in 13 clinical tumor tissue samples (n = 727), with an AUC greater than 0.85, particularly in biliary tract (AUC = 0.98) and esophageal cancers (AUC = 0.99), and in HNSCC, with an AUC greater than 0.946. However, it should be noted that the methylation level of tumor tissue generally exhibits a higher magnitude compared to blood samples. Nevertheless, this study did not provide further evidence regarding the potential utility of gene methylation analysis as a noninvasive method for early screening through blood-based testing. Our study builds on the results of tissue samples from a public database and further utilizes clinical blood samples to confirm the potential of cg10364040 as a noninvasive early screening marker for LSCC. Therefore, we developed and validated a single-gene unit methylation site marker (cg10364040) for the early screening of LSCC, which has the advantages of operational simplicity, low testing costs, and high specificity. The homeobox (HOX) family of genes has been demonstrated in several previous studies to play a crucial role in the pathogenesis of HNSCC [49, 50]. HOX genes encode highly conserved homeotic transcription factors that play crucial roles in organogenesis and tissue homeostasis and play roles in both organogenesis and oncogenesis [51]. Lleras et al. [49] conducted a genome-wide scan using the Illumina HumanMethylation27 beadchip to identify aberrant DNA methylation in 118 HNSCC DNA samples. The results revealed 44 CpG loci that exhibited significantly altered DNA methylation across all anatomic sites of HNSCC. The most overrepresented group was the Kruppel family of ZNF genes on chromosome 19q13. Notably, the Hox genes HOXD9, HOXD10, and HOXD12 also presented hypermethylation across anatomic sites in the present study. However, the HumanMethylation27 beadchip does not contain the cg10364040 locus that we studied. In this study, we incorporated additional loci via 450 K chip data and observed cg10364040 relevance in the early detection of HNSCC. These findings further indicate that HOX genes may play a significant role in the pathogenesis of HNSCC. Bueno et al. employed cDNA microarray sequencing to identify significant upregulation of HOX genes in laryngeal squamous cell carcinoma (LSCC) [50]. Subsequent cellular assays validated the crucial role of the HOXC8, HOXD10, and HOXD11 genes in promoting cell proliferation and migration. Studies based on DNA methylation or gene expression profiles suggest that HOXD10 has the potential to distinguish HNSCC patients from normal control individuals. Rodrigues et al. demonstrated that HOXD10 was amplified in oral squamous cell carcinoma (OSCC) surgical margin tissues and displayed hypermethylation in approximately half of the tumor samples, indicating that HOXD10 expression is regulated by DNA methylation in OSCC [21]. Our study further applied single-site methylation of HOXD10 for noninvasive early clinical screening, thus addressing the gap in noninvasive early screening of LSCC. Interestingly, the cg10364040 marker performed less well for precancerous lesions (high-grade squamous intraepithelial neoplasia), with 76.7% sensitivity and 63.6% specificity, as would be expected. However, the methylation levels of cg10364040 were greater in stage I patients than in stage II–IV patients, and the ROC curve indicated that this locus exhibited superior diagnostic performance specifically for stage I LSCCs (AUC = 0.864, sensitivity = 85%, specificity = 86%) rather than encompassing all patients across stages I-IV (AUC = 0.767, sensitivity = 72.1%, specificity = 82.0%). However, there are several limitations that need to be emphasized. The clinical validation phase of our study had a limited sample size, with the all of enrolled patients diagnosed with laryngeal squamous cell carcinoma. Small sample sizes may limit the ability to detect associations between the identified biomarkers and various clinical factors, as robust estimation of such relationships typically requires larger datasets. Therefore, the diagnostic efficacy of this biomarker in other subtypes of HNSCC requires further validation through additional clinical trials. Second, this was not a randomized controlled study, and thus, some amount of selection bias may exist. Third, the methylation levels of cg10364040 are negatively correlated with clinical stage, indicating that this locus may be crucial in the initial phases of LSCC. With respect to the underlying mechanisms of this phenomenon, exploring its biological significance is crucial. However, the current data do not fully elucidate the detailed physiological mechanisms involved. In future studies, we would allocate greater attention to conducting larger-scale multiple center validation on diverse ethnic populations. And we could further explore the specific effects of methylation changes at cg10364040 on the development of LSCC and other HNSCC disease and its biological mechanism by adding an experimental design. Conclusion Taken together, the results of this study demonstrate the value of HOXD10 cg10364040 methylation in diagnosing LSCC, especially in the detection of early-stage LSCC. These findings support the potential of this blood-based test to complement existing LSCC screening methods, thereby potentially reducing cancer mortality rates. Supplementary Information Below is the link to the electronic supplementary material. Supplemental Figure S1 LASSO regression and random forest analysis A. Cross-validation for tuning parameter selection in the LASSO model via minimum criteria; B. LASSO coefficient profiles of the 1139 methylation sites. A coefficient profile plot was produced against the log(lambda) sequence. C. Variable importance evaluation. Supplemental Figure S2 Study flow diagram. Supplemental Figure S3 ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between the blood of LSCC patients and HG-SIN patients. Supplemental Figure S4 Representative immunohistochemical images of LSCC patients with different HPV statuses. A. HE and IHC images of HPV-negative LSCC Patient-1; B. HE and IHC images of HPV-positive LSCC Patient-2; C. HE and IHC images of HPV-positive LSCC Patient-7. Supplemental Figure S5 Methylation levels and ROC curves of patients with different HPV status. A. Box plots presenting the β value distribution of cg10364040 methylation of HOXD10 in HPV negative (HPV-) LSCC, HPV positive (HPV+) LSCC and normal control; B. ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between HPV- LSCC and normal controls; C. ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between HPV+ LSCC and normal controls. Supplemental Figure S6 ROC curve corresponding to the serum biomarkers used to differentiate between the blood of LSCC patients and NC. A. CEA; B. CYFRA211; C. AFP; D. CA199; E. CA724; F. NSE; G. SCC. Supplementary Material 7 HOXD10 primer information. Supplementary Material 8 Clinical information, serum tumor marker levels, and HOXD10 cg10364040 β-values across all clinical samples. Supplementary Material 9 The diagnostic performance of serum tumor markers (LSCC vs. normal control). Acknowledgements Not applicable. Abbreviations American Cancer Society Urea under curve Circulating cell-free genome atlas Carcinoembryonic antigen Cell-free DNA Colorectal cancer Cancer signal origin Cytokeratin 19 fragment antigen Differential methylation probe False discovery rate Gene expression omnibus High-grade squamous intraepithelial neoplasia Head and neck squamous cell carcinoma Homeobox D 10 Least absolute shrinkage and selection operator Laryngeal squamous cell carcinoma Multi-cancer early detection Receiver operating characteristic Squamous cell carcinoma The Cancer Genome Atlas Human papillomavirus Author contributions All authors read and approved the final manuscript. Sample preparation was carried out by Gong X., Lu Z. and Zhu Z.; Chen H., Zhou H. and Cheng L. performed the experiments; Zhang Q., Cao X. and Chen X. contributed to designing research, analyzing data, and discussing findings; Duan Q., Zhu C. and Chen D. wrote the manuscript; Chen X. supervised the work. All authors reviewed the manuscript. Funding This work was supported by 2023 Clinical Capacity Enhancement Project Medical Project of Jiangsu Provincial People’s Hospital (JSPH-MA-2023-1), and the Natural Science Foundation of Jiangsu Province (BK20230739). Data availability The original contributions presented in the study are included in the article/supplementary materials, further inquiries can be directed to the corresponding author. In addition, we have uploaded analysis scripts (R/Python scripts) on GitHub website (https://github.com/qinzhang0824/monomethylated-locus-for-LSCC-early-diagnosis). Declarations Ethics approval and consent to participate The study protocol and all amendments were approved by the ethics committee of First Affiliated Hospital, Nanjing Medical University (Ethical number: 2023-SR-373). All participants provided written informed consent. The study was done in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Xiaoyang Gong, Zhaoyi Lu, Zihang Zhu and Qin Zhang contributed equally in this work. Contributor Information Xin Cao, Email: caoxin@njmu.edu.cn. Xi Chen, Email: chenxi2882@jsph.org.cn. References Associated Data Supplementary Materials Supplemental Figure S1 LASSO regression and random forest analysis A. Cross-validation for tuning parameter selection in the LASSO model via minimum criteria; B. LASSO coefficient profiles of the 1139 methylation sites. A coefficient profile plot was produced against the log(lambda) sequence. C. Variable importance evaluation. Supplemental Figure S2 Study flow diagram. Supplemental Figure S3 ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between the blood of LSCC patients and HG-SIN patients. Supplemental Figure S4 Representative immunohistochemical images of LSCC patients with different HPV statuses. A. HE and IHC images of HPV-negative LSCC Patient-1; B. HE and IHC images of HPV-positive LSCC Patient-2; C. HE and IHC images of HPV-positive LSCC Patient-7. Supplemental Figure S5 Methylation levels and ROC curves of patients with different HPV status. A. Box plots presenting the β value distribution of cg10364040 methylation of HOXD10 in HPV negative (HPV-) LSCC, HPV positive (HPV+) LSCC and normal control; B. ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between HPV- LSCC and normal controls; C. ROC curve corresponding to HOXD10 cg10364040 methylation when used to differentiate between HPV+ LSCC and normal controls. Supplemental Figure S6 ROC curve corresponding to the serum biomarkers used to differentiate between the blood of LSCC patients and NC. A. CEA; B. CYFRA211; C. AFP; D. CA199; E. CA724; F. NSE; G. SCC. Supplementary Material 7 HOXD10 primer information. Supplementary Material 8 Clinical information, serum tumor marker levels, and HOXD10 cg10364040 β-values across all clinical samples. Supplementary Material 9 The diagnostic performance of serum tumor markers (LSCC vs. normal control). Data Availability Statement The original contributions presented in the study are included in the article/supplementary materials, further inquiries can be directed to the corresponding author. In addition, we have uploaded analysis scripts (R/Python scripts) on GitHub website (https://github.com/qinzhang0824/monomethylated-locus-for-LSCC-early-diagnosis).