LIM Homeobox Transcription Factors as Novel Prognosis Biomarkers in Head and Neck Squamous Cell Carcinoma

Background: LIM class homeobox (LHX) genes, an important subfamily of the homeobox genes, encode transcription factors that have a fundamental role during embryonic development. However, knowledge regarding the function and mechanism of LHXs in head and neck squamous cell carcinoma (HNSCC) is still lacking. Methods: We conduct a bioinformatic analysis to systematically explore the mRNA expression, clinical correlation, prognostic values, and underlying mechanisms of distinct LHXs in HNSCC. The differentially expressed mRNAs in the LIM homeobox gene family and their correlation with clinical variables were determined and veried with packages in software R. The prognosis values of LHXs expression levels were evaluated by Kaplan-Meier (KM) method and Cox proportional hazard model. Gene set enrichment analysis (GSEA) was conducted to understand the potential biological function of LHXs. The immune cell inltration patterns were estimated through CIBERSORT and TIMER, and the methylation levels of LHXs in HNSCC were explored in UALCAN and MEXPRESS. Results: We found that among 12 LHXs, 8 genes (ISL1, LHX1-3, 5, 9, LMX1A, and LMX1B) showed altered expression in HNSCC tissues and detected signicant correlations between their expression and clinical variables. Survival analysis revealed that LHX1, LHX5, LMX1A, LMX1B can serve as unfavorable prognosis predictors, and ISL1, LHX2, LHX9 can serve as favorable predictors in all HNSCC patients. Gene sets enrichment analysis and immune inltration analyses showed that the aberrant expression of LHXs was closely related to cancer-associated processes and immune cell inltration patterns. Finally, we observed hyper-methylation in the promoters of ISL1, LHX2, 5, 9, LMX1A, LMX1B and hypo-methylation in LHX3 promoter, suggesting the regulatory mechanism of LHXs abnormal expression may be related to aberrant DNA methylation. Conclusions: Our study found the oncogenic roles of LHX1,5 carcinoma; RFS: ccRCC: renal cell carcinoma; OTSCC: oral tongue squamous cell carcinoma; NHL: non-Hodgkin lymphoma; PDA: pancreatic ductal adenocarcinoma; ADC: adenocarcinoma;


Background
Head and neck squamous cell carcinoma (HNSCC) is a malignant tumor derived from the mucosal epithelium in the oral cavity, larynx and pharynx (1). It is the sixth most common cancer by incidence worldwide and survivors of HNSCC have the second-highest suicide rates due to depression and depressed quality of life after cancer treatment (2,3). Although continuous improvement of HNSCC treatment has been achieved by surgery, chemotherapy or radiotherapy, the survival rate of this cancer remains stagnant, especially for HPV-negative HNSCC (4). Thus, there is an urgent need to develop effective target therapeutic strategies and search for novel biomarkers to improve diagnosis, prognosis, and treatment for HNSCC.
Homeobox genes encode transcription factors that are critical for embryonic development. In human genomes, homeobox genes are divided into several subfamilies, one of which is the LIM class homeobox (LHX) genes. The proteins encoded by LHXs contain a centrally located homeodomain for binding DNA and two tandem cysteine-rich zinc-nger LIM domains for protein-protein interactions, which enable them to participate in a wide range of biological processes (5)(6)(7)(8). To date, 12 LHXs have been identi ed in human genomes and their roles in neoplastic transformation and tumor progression have attracted great interest in recent years (9). However, the scienti c role of distinct LHXs in HNSCC remained unknown.
Thus, a comprehensive study of LHX transcription factors will help to understand the molecular mechanisms behind HNSCC and develop novel biomarkers for this cancer.
In this study, we conducted a bioinformatic analysis to investigate the role of LHXs in HNSCC and their underlying mechanisms in tumor formation. We examined 8 LHXs with altered expression in HNSCC and con rmed their discriminatory power between normal and tumor samples using Receiver Operating Characteristic (ROC) curve analysis. Then their connection with clinical attributes and patient survival was evaluated. Subsequently, we conducted the gene set enrichment analysis (GSEA) as well as immune in ltration analysis to understand the potential biological function of LHXs. Finally, methylation analysis was performed to explore the epigenetic mechanism behind the abnormal expression of LHXs in HNSCC.

Differential Expression Analysis
First, the differential expression of 251 homeobox genes between HNSCC tissues and normal samples in 546 patients was examined using the R package "limma" (12).72 differentially expressed genes (DEGs) in homeobox gene family were selected with p-values < 0.05 and |log2 fold change (FC)| > 1.5 as thresholds.
The results were visualized using a volcano plot by R package "ggplot2", and the classi cation of these differentially expressed homeobox genes was displayed as a heatmap using the R package "pheatmap". According to the classi cation, 8 DEGs belong to the LIM homeobox gene family and were kept for subsequent analysis. We also performed a Student's paired t-test to validate the differential expression of selected LHXs between tumor and corresponding normal samples. Moreover, to assess whether the differentially expressed LHXs can distinguish between normal and tumor samples, we performed a receiver operating characteristic curve (ROC) analysis and calculated the area under the curve (AUC) using the R package "pROC" (13). In addition, the co-expression relationships of selected LHXs were analyzed and visualized using the R package "ggcorrplot" (14).

Clinical and Survival Data Analysis
The relationships between LHXs expression and the following clinical characteristics were assessed: clinical stage (I, II, III, IV), pathological grade (G1, G2, G3-4), primary tumor site (oral cavity, pharynx, larynx), and HPV status (Positive, Negative) using the R package "ggstatsplot" (15). In survival analysis, patients were divided into high and low expression groups based on the cutoff value of selected LHXs calculated by the R Package "survminer" (16). After that, the overall survival (OS), disease-speci c survival (DSS), and Disease-free survival (DFS) with a 5-year follow-up were estimated with the Kaplan-Meier (KM) method and the Cox proportional hazard model.

Gene Set Enrichment Analysis and Immune In ltration Analysis
To identify the potential pathways that are associated with selected LHXs, we rst divided the HNSCC patients into two groups by the median expression value of LHXs. DESeq2 was then applied to nd the signi cant differential gene sets between these two groups. After that, we used "fgsea" (32) and "clusterPro ler" (34) in R software to run gene set enrichment analysis (GSEA) and visualize the results.
The in ltrating levels of 22 immune cell types in each HNSCC sample were calculated using the CIBERSORT algorithm (https://cibersort. stanford.edu/) (18). Samples with CIBERSORT p < 0.05 were considered eligible for subsequent analysis and divided into high and low groups according to the median expression value of LHXs. Then, the tumor-in ltrating immune cell levels between these two groups were compared. Moreover, another immune in ltration data was analyzed and visualized by the TIMER database (Tumor Immune Estimation Resource, https://cistrome. shinyapps.io/timer/) (19).

Analysis of DNA Methylation and mRNA Expression
In order to investigate the epigenetic mechanism behind the abnormal expression of LHXs in HNSCC, the methylation levels of LHXs promoters between HNSCC tissues and normal samples were assessed using the UALCAN database (http://ualcan.path.uab.edu/) (20). After that, we utilized the MEXPRESS database (https://mexpress.be) (21) to investigate the association between LHXs expression and their DNA methylation status.

Statistical Analyses
All statistical analyses were performed using R software 4.1.1(2021-08-10). Normality distribution was analyzed using the Shapiro-Wilk test and the equality of variances was calculated with Levene's test. Expression correlation between selected LHXs was tested by Pearson's correlation analysis. To compare the expression levels of LHXs according to clinical variables, a Welch's unequal variances t-test (for 2 groups) or a Welch's ANOVA (3 or more groups) was used based on the number of levels in the grouping parameters. The post-hoc test was performed using the Games-Howell test and the p-value was adjusted using the False Discovery Rate (FDR) method. For the KM curve, we used the log-rank test to evaluate the differences between groups. The Cox proportional hazards regression was applied with the Wald test and the tied event times were handled with the Efron approximation method. For GSEA analysis, p-values were calculated based on the adaptive multilevel splitting Monte Carlo approach (22). |Normalized enrichment scores (NES)| > 1, adjusted p-value < 0.05 and FDR q-value < 0.25 were considered signi cant.
Next, we validated the altered expression of 8 LHXs in tumor and corresponding normal tissue using paired t-test. Similarly, signi cant up-regulation of LHX1, LHX2, LHX3, LHX5, LHX9, LMX1B and downregulated of ISL1 and LMX1A were examined in HNSCC tissues (all p<0.0001, Figure  3C). Furthermore, the ability to distinguish between normal and tumor samples was tested using the ROC curve. As shown in Figure 3D

Association Between LHXs Expression and Clinicopathological Features
To understand the clinical signi cance of the selected LHXs, we analyzed the relationships between clinical variables and LHXs levels in patients with HNSCC. As shown in Figure 4, the expression of LHX1 and LHX2 was correlated with clinical stages (p= 0.041, p= 0.019, respectively) and patients with more advanced clinical stages have higher LHX2 expression levels ( Figure 4A). As for LHX2 and LMX1A, high mRNA expression was correlated with advanced tumor grade (p= 7.23e-5, p= 0.01, respectively), in contrast to LHX1 which was highly expressed in G1 tumors (p= 0.015) ( Figure 4B). Figure 4C displayed the signi cant correlations between LHX1, LHX2, LHX5 expression and primary tumor sites (all p< 0.0001). The highest expression of LHX1, LHX2, and LHX5 was evaluated in the oral cavity, pharynx and larynx, respectively. For HPV status, Figure 4D showed a remarkable higher expression for LHX1 and LHX5 in HPV-negative patients (p = 8.98e-13 and 1.57e-5). Conversely, LHX2 was found to express highly in HPV-positive patients(p=1.22e-10). The ndings were further veri ed using Cox regression analysis. As shown in Figure 6

LHXs Correlated Genes Are Involved in Numerous Biological Processes
After indicated selected LHXs as prognosis factors for patient survival, we conducted the GSEA analysis to assess these genes at a functional level (Additional le 2, Table 3). We found that high-LHX1 expression exhibited enrichment of genes associated with environmental information processing pathway (ECM-receptor interaction and ) and immune system pathway (RIG-I-like receptor signaling pathway, cytosolic DNA-sensing pathway ). Gene sets that are involved in metabolism processes like metabolism of xenobiotics by cytochrome P450 were associated with low expression of LHX1 and LMX1B. As for LHX5, GSEA analysis revealed that pathways like focal adhesion and ribosome were upregulated and pathways related to immune response (T cell receptor signaling pathway, primary immunode ciency, and Natural killer cell mediated cytotoxicity) were down-regulated. Pathways implicated in cancer invasion and metastases like focal adhension and tight junction were activated in the high-expression groups of LHX5, LMX1A, LMX1B and low-expression groups of LHX2 and LHX9. The enrichment of post-transcriptional control related pathways like ribosome, spliceosome, proteasome, and RNA degradation was correlated to high LHX3 and LHX5 expression while low ISL1, LHX2 and LMX1A expression. Moreover, LHXs have a close relationship with immune-related pathways, which were enriched among the low-expression groups of ISL1 (cytosolic DNA sensing pathway, antigen processing and presentation, RIG-I-like receptor signaling pathway), LHX2 (cytosolic DNA sensing pathway), LHX9(intestinal immune network for IgA production, T-cell receptor signaling pathway), and LMX1B (primary immunode ciency), Figure 7A.
Next, differential gene sets associated with high and low LHXs expression phenotypes with regard to hallmark gene sets (h.all.v7.4) were analyzed (Additional le 2, Table S4). High LHX1 expression was related to the process of IL6 JAK STAT3 signaling, TGF beta signaling and angiogenesis. High expression of LHX3 was a connection with G2M checkpoint, MYC family targets, DNA repair, TGF alpha response, and E2F targets. LHX5 and LMX1B high expression was associated with angiogenesis and epithelialmesenchymal transition (EMT). Low ISL1 expression corresponded with the process of reactive oxygen species (ROS) pathway, TGF beta signaling and MYC targets v2. Low LHX2 expression corresponded with angiogenesis, epithelial-mesenchymal transition, P53 pathway, TNF signaling via NF-kB, stress reaction pathways like unfolded protein response and UV response, immune response pathways like in ammatory response, interferon-alpha response, and interferon-gamma response. Low LHX9 expression corresponded with the process of IL2 STAT5 signaling, KRAS signaling and immune response pathways like allograft rejection and in ammatory response, Figure 7B.
Immune In ltration Is associated with LHXs Expression Figure 8 displayed the CIBERSORT analysis of the relationship between tumor-in ltration immune cells and LHXs expression. ISL1 expression level had negative associations with the ratio of resting memory CD4+ T cells, M0 macrophages, M1 macrophages, and positive relationships with the ratio of plasma cells. The expression levels of LHX1 and LMX1B showed positive correlations with the proportion of resting memory CD4+ T cells, resting NK cells, CD4+ T cells, M0 macrophages, as well as M1 macrophages, and negative association with the ratio of plasma cells and CD8+ T cells. LHX2 expression was positively correlated with the in ltration of plasma and CD4+ T cells, but negatively correlated with M0 macrophages and activated dendritic cells. A higher proportion of resting NK cells and M1 macrophages, while a lower proportion of activated dendritic cells was observed in the LHX3 highexpression group. LHX5, as well as LMX1A, had high M0 macrophages frequency and low plasma cells, CD4+ T cells, M1 macrophages frequencies in their high-expression group. As for LHX9, the CIBERSORT output revealed that the M1 macrophages and activated dendritic cells proportion were signi cantly higher in LHX9 high-expression group.
The tumor-in ltration immune cells according to LHXs expression evaluated in Timer database was shown in Figure 9.
We further explored the relationship between DNA methylation and the mRNA expression levels of LHXs that are down-regulated in HNSCC tissues using MEXPRESS. As shown in Figure 10B, low ISL1 expression was found to be signi cantly associated with the hyper-methylation for all promoter regions probes in HNSCC tissues (Pearson correlation coe cients: −0.329 ~ −0.537, all p < 0.001). As for LMX1A, we found the 3 signi cant methylation sites (cg09248345, cg04351049, and cg09069138) in the DNA sequences of LMX1A that were negatively associated with its expression levels ( Figure S1).

Discussion
HNSCC is a common malignancy associated with high aggressiveness and many di culties in successful treatment. Given the signi cant morbidities and high recurrence rates associated with conventional treatments such as surgery, radiation therapy, and chemotherapy, innovative approaches are urgently needed to effectively treat HNSCC(4). Although targeted therapy and immunotherapy have shown great promise in recent years, the long-term e cacy is still not ideal due to the lack of speci c targets (23,24). Homeobox genes comprise one of the largest groups of annotated oncogenes and have been reported to participate in the development and progression of various cancers. However, as an important group in homeobox genes, there are only sporadic studies on the roles of LHX genes in HNSCC.
Here, we conduct a bioinformatic analysis of TCGA data to determine the roles of LHXs in patients with HNSCC. Results from our study demonstrated that among 12 LHXs, 8 genes had altered expression levels in HNSCC tumor tissues compared to normal samples and their expression levels were signi cantly correlated with clinic-pathological characters such as tumor grade, tumor localizations as well as HPV status. Besides, LHX1, LHX5, LMX1A, and LMX1B were indicated as unfavorable prognosis predictors and ISL1, LHX2, as well as LHX9 were indicated as favorable predictors through KM analysis and Cox regression analysis. Moreover, the GSEA and immune in ltration analysis con rmed their anticipation in various biological processes and association with immune cell in ltration patterns. Finally, we determined the hype-methylation and hype-methylation of ISL1, LHX2, LHX5, LHX9, LMX1A, LMX1B promoter and hypo-methylation of LHX3 promoter in HNSCC tissues, then veri ed the relationship between the abnormal DNA methylation and aberrant mRNA expression of ISL1 as well as LMX1A in HNSCC. To our knowledge, this is the rst comprehensive analysis to explore the mRNA expression, clinical correlation, prognostic values and underlying mechanisms of distinct LHXs in HNSCC.
By reviewing the literature, we observed that only LMX1B and LHX1 have been previously described in HNSCC. Smith et al. (25) found that LMX1B was signi cantly over-expressed in laryngeal squamous cell carcinoma (LSCC) tissues and was an independent prognostic indicator in terms of OS and (Recurrence Free Survival) RFS. Besides, they suggested that elevated gene body methylation cloud be responsible for its upregulation. Consistent with the literature, our current research found that LMX1B mRNA was more highly expressed in HNSCC tissues, and was correlated with poor OS and DSS. Besides, we found its expression to be signi cantly related to tumor sites and HPV states. LMX1B is highly expressed in the oral cavity and HPV-negative patients. Interestingly, in terms of tumor grade, the expression level of LMX1B tended to be lower in higher tumor grades.
Previous studies have found LHX1 to be highly expressed in gastric, pancreatic cancers, chronic leukemia, human clear cell renal cell carcinoma (ccRCC), both primary and metastasis medulloblastoma, as well as hepatitis B virus-related hepatocellular carcinoma (26-30). Dormoy et al. found that LHX1 knockdown decreased ccRCC cell motility, migration, invasion, and clonogenicity. Functional Genomics identi ed LHX1 as one of the drivers of Medulloblastoma Dissemination. In terms of HNSCC, studies showed over-expression of LHX1 was found in oral tongue squamous cell carcinoma, and KM analysis indicated high expression of LHX1 with poor OS in OTSCC (31). In our study, LHX1 was signi cantly upregulated in HNSCC tissues compared to normal samples, and the ROC curve con rmed that LHX1 was capable of distinguishing between normal and tumor samples (with AUC = 0.919). In addition, patients with high LHX1 expression had poor OS, DSS, and DFS, which indicates LHX1 as a diagnosis and prognostic biomarker for HNSCC. Furthermore, GSEA analysis in HNSCC by up-regulating ECM-receptor interaction and genes participated in IL6 JAK STAT3 signaling, TGF beta signaling and angiogenesis. Notably, the effects of LHX1 on cell movements and metastasis development are consistent with earlier ndings in the literature, although in other tumor types. Dormoy et al. (32) showed that LHX1 knockdown decreased ccRCC cell motility, migration, invasion, and clonogenicity, and Mumert's (33) functional genomics experiment identi ed LHX1 as one of the drivers of Medulloblastoma Dissemination.
ISL1 is a LIM homeodomain transcription factor that plays a critical role in the development of function in the development of the nervous system, heart, and pancreas (34)(35)(36). Previous researches have indicated ISL1 as a good marker for well-differentiated pancreatic neuroendocrine neoplasms (37,38). Zhang et al. (39) demonstrated that ISL1 was signi cantly over-expressed in non-Hodgkin lymphoma(NHL), and could contribute to NHL proliferation and progression by promoting the transcription of c-Myc. The in vitro and in vivo expriments conducted by Guo et al. (40) revealed that ISL1 was frequently overexpressed in gastric cancer (GC) samples and associated with worse clinical outcomes. Patients with positive ISL1 staining tended to have poor prognosis and Cox regression analysis proved ISL1 expression as an independent prognostic factor. In our present study, the expression level of ISL1 was lower in HNSCC tissues compared to normal tissues, and low ISL1 level was correlated with poor OS and DSS. The KEGG pathway analysis and hallmark analysis for ISL1 con rmed its role as an anti-tumor gene in HNSCC by down-regulating post-transcriptional pathways such as ribosome and proteasome and genes participated in reactive oxygen species pathway, TGF-beta signaling, as well as MYC targets. Using UALCAN, we found hype-methylation of ISL1 promoter in HNSCC tissues, and con rmed that ISL1 expression and promoter methylation level are negatively correlated using MEXPESS, suggesting that DNA methylation may be one of the main mechanisms of downregulation of ISL1 gene expression in HNSCC.
LHX2 is an evolutionarily conserved, versatile and powerful regulator of multiple developmental phenomena such as tissue patterning, control of cell differentiation, and axon path nding (41). Recently, the role of LHX2 in carcinogenesis has gained increasing attention. The high expression level of LHX2 was observed in non-small cell lung cancer (42), cervical cancer (43), chronic myelogenous leukemia (44) pancreatic ductal adenocarcinoma (PDA) (45) and osteosarcoma (46). Besides, previous studies have discovered the functions of LHX2 as an oncogenic factor enhancing proliferation, metastasis and chemoresistance in many cancers (42,43). Zhou et al. (45) reported that LHX2 promoted PDA cells proliferation in vitro/vivo and knockdown of LHX2 reduced cell viability. Functional experiments in vivo and in vitro demonstrated that LHX2 plays an important role during EMT and tumor progression in breast carcinogenesis (47). Moreover, LHX2 has been identi ed as an autophagy inhibitor via mTOR signaling in osteosarcoma (46). On the other hand, LHX2 exhibited low expression in extranodal diffuse large B-cell lymphoma and can be used as independent predictors for the prognosis of glioblastoma (48,49). In this study, LHX2 was highly expressed in HNSCC tissues and capable of discriminating between normal and tumor samples (AUC = 0.913). Besides, LHX2 was highly expressed in advanced stages and higher tumor grades. Interestingly, KM curve analysis found that patients in LHX2 high-expression group had more favorable OS and DFS. In contrast to LHX1, KEGG analysis found that LHX2 was associated with the upregulation of pathways related to cytochrome P450 metabolism and downregulation of pathways like focal adhesion and ribosome. The further hallmarks analysis for LHX2 con rmed its function in deregulating multiple biological processes associated with tumorigenesis such as angiogenesis, epithelial-mesenchymal transition, P53 pathway, TNF signaling via NF-kB, unfolded protein response and UV response, interferon-alpha and gamma response.
LHX3, located at 9q34.3, plays a fundamental role in spinal cord motor neurons and pituitary development (50)(51)(52). Only recently, the relationship between LHX3 and cancer has been reported. Dietrich et al (53) detected improved methylation of LHX3 in ductal carcinoma compared with none-tumor tissue. One study by Lin (54) examined the expression level, prognostic value and biological role of LHX3 in lung adenocarcinoma (ADC), the results showed that LHX3 is highly expressed in tumor tissues and signi cantly associated with clinical factors of ADC patients. Besides, survival analyses indicate that upregulation of LHX3 is related to adverse survival and multivariate regression analysis veri ed LHX3 as an independent prognostic factor. In our current study, high LHX3 expression was found in HNSCC tissues, and KEGG analysis found LHX3 to be associated with the upregulation of post-transcriptional pathways such as ribosome and spliceosome. Hallmarks analysis revealed that genes involved in the G2M checkpoint, MYC targets, E2F targets, and interferon-alpha response are up-regulated in LHX3 highexpression group. Moreover, we found that promoter methylation of LHX3 was lower in HNSCC tissues compared to normal tissues, suggesting that the hypo-methylation of LHX3 promotor may be responsible for the upregulation of LHX3 in HNSCC.
As an important paralog of LHX1, LHX5 functions as a transcriptional regulator in the control of forebrain differentiation and development (55,56). So far, LHX5 was only found to be associated with Breast cancer. A genome-wide methylation pro ling by Kim et al. (57) demonstrated that LHX5 showed signi cantly higher frequencies of hypermethylation in breast cancer tumor tissues, and had an independent prognostic value for this cancer. In our study, LHX5 was highly expressed in HNSCC tissues, and the ROC curve con rmed its discriminability between normal and tumor samples (AUC = 0.956). In survival analysis, we found patients with high LHX5 expression tend to have unfavorable OS and DSS. KEGG analysis for LHX5 con rmed its role as an oncogene gene in HNSCC by up-regulating focal adhesion, glutathione metabolism as well as the ribosome, and down-regulating immune response pathways like primary immunode ciency and T cell receptor signaling. Further hallmarks analysis revealed that gene sets implicated in EMT, apical junction formation, hedgehog signaling and angiogenesis were activated in LHX5 high-expression group.
LHX9 is a LIM homeodomain transcription factor functions in the control of neural cell de nition (58). It has been demonstrated that hypermethylation of LHX9 leads to gene silence in high-grade gliomas of childhood (59) and follicular lymphoma (60). Moreover, an in vitro functional study by Vladimirova et al. (59) found the epigenetic alteration of LHX9 may be responsible for glioma cell migration and invasiveness. Our results showed that LHX9 was up-regulated in HNSCC tissues, and the expression levels of LHX3, LHX5, LHX2 and LHX1 were positively correlated with LHX9. GSEA analysis found that LHX9 was related to the downregulation of cytochrome P450 metabolism as well as Cytokine-cytokine receptor interaction and genes positively regulated by LHX9 are involved in MYC targets and interferonalpha response.
LMX1A was found to play an important role in the development of the nervous system(60). Immunohistochemical analyses found LMX1A had a higher expression in meningiomas and gliomas tissues, and LMX1A high-expression was correlated with more advanced tumor grade(61). Moreover, the methylation level of LMX1A was also found to be closely related to various cancers. Dong et al (62) found that LMX1A was silenced due to DNA hypermethylation in gastric cancer and restoration of LMX1A can improve cell apoptosis and inhibit anchorage-independent growth. Here, we found that mRNA expression of LMX1A was lower in HNSCC tumor tissues, while patients with lower LMX1A expression tended to have better DFS. The KEGG pathway analysis and hallmark analysis found LMX1A was related to the down-regulation of post-transcriptional pathways such as ribosome and spliceosome and genes that participated in DNA repair, interferon-alpha signaling, as well as MYC targets. Using UALCAN, we found that the promoter methylation level of LMX1A was signi cantly improved in HNSCC tissues, suggesting that DNA methylation may be one of the mechanisms of downregulation of LMX1A gene expression in HNSCC.

Conclusions
In the present study, the exact roles of LIM homeobox genes in HNSCC were systematically investigated for the rst time. The ndings suggested the oncogenic roles of LHX1,5 and LMX1B as well as the tumorsuppressor roles of ISL1 and LHX2 in patients with HNSCC, indicating these LHXs as potential diagnostic and prognostic biomarkers for the improvement of HNSCC survival and prognostic accuracy.

Consent for publication
Not applicable.

Competing interests
All nancial and non-nancial competing interests must be declared in this section. Authors' contributions CL analyzed and interpreted the relevant data of TCGA, and was a major contributor in writing the manuscript. CCD assisted in organizing data and editing. BH assisted in reviewing and editing the manuscript. All authors read and approved the nal manuscript. gastric cancer. Cancer Sci. 2011;102(2):361-6. Figure 1 Flow diagram of our study on the transcriptional expression, clinical correlation, prognostic value, and underlying mechanisms of LHXs in HNSCC.         The Pearson correlation coe cients, P values and probes ID are shown on the right; Probes in the promoter region are marked with red font and were all signi cantly negatively correlated with ISL1 expression. Note: *p < 0.05, **p < 0.01, ***p < 0.001.