An autophagy-related long non-coding RNA signature in tongue squamous cell carcinoma

We obtained the RNA-sequencing data sets in HTSeq FPKM format including 15 normal samples and 147 tumor samples from the TCGA database (https://​portal.​gdc.​cancer.​gov/​). Considering the lack of certain clinical traits, such as 7th edition AJCC stage, TMN staging, etc., the samples of 134 patients with complete clinical information are retained. Since the data of this study are obtained from public database, the approval of the ethics committee is not necessary. The data of all patients are provided in the Additional files 1, 2, 3, 4, 5 and 6.

The 232 autophagy-associated genes are collected through the Human Autophagy Database (HADb, http://​www.​autophagy.​lu/​clustering/​index.​html), which is a public database containing human genetic information related to autophagy that has been described so far [16]. All the lncRNA was isolated and recognized from RNA-sequencing data sets by Perl programming language. Pearson correlation coefficient is served to describe the pertinence between the expression of autophagy-related genes and lncRNAs by R. These are the criteria used to shortlist autophagy-related lncRNAs: p < 0.05 and |R²| > 0.3 [17].

We identified autophagy-related lncRNAs via a univariate Cox regression model by R. The lncRNAs that were significant (p < 0.05) with overall survival of TSCC patients were chosen. The multivariate Cox stepwise regression model was performed to evaluate the potential of autophagy-related lncRNAs as independent prognostic factors. Based on the minimum Akaike information criterion (AIC) value, we selected the first-rank autophagy-related lncRNA to construct a risk score for each patient with TSCC. The risk score formula of autophagy-related lncRNAs prognosis signature was shown as the following: Risk score = \({\sum }_{\theta =1}^{n}Coef(\theta )\times Expr(\theta )\). The n represents the number of lncRNAs used to construct the model. \(Coef(\theta )\) represents the coefficient in the regression model, and \(Expr(\theta )\) is defined as the expression of each lncRNA in this formula [17]. 146 patients of TCGA-TSCC were divided into two groups with low-risk or high-risk score according to the median value of risk score.

The construction of lncRNAs and mRNA co-expression network on TSCC was to gain insight into the correlation between autophagy-related lncRNAs and co-expression of autophagy-related mRNA. Pearson’s coefficients > 0.3 was used to shortlist mRNA that was related to lncRNA. We applied Cytoscape software (version 3.7.1) to showcase the relationship between mRNA and autophagy-related lncRNA for visualization.

We established a nomogram to predict the 1-year and 3-year survival rates of TSCC by integrating the corresponding clinical information (including age, gender, grade, AJCC stage, T, and N stage) and risk score. Subsequently, we utilized the concordance index (C-index) to appraise the reliability of this nomogram. The C-index refers to the consistency between the actual probability of the outcome and the predicted probability. Generally, the closer the value of the C-index is to 1, the higher the accuracy of the predictive ability of the nomogram. Both the nomogram and calibration curves are based on the RMS package [18] in the R software (version 4.0.3).

To explore the differences in biological annotations and pathways between the high-risk and low-risk groups, we applied the gene set enrichment analysis (GSEA) software (version 4.1.0, https://​www.​gsea-msigdb.​org) to obtain the results [19]. The Molecular Signatures Database (MSigDB) of c2 (c2.cp.kegg.v7.2.symbols.gmt) was selected as the reference gene sets database, and the number of permutations was 1000. False discovery rate (FDR) < 0.25 and p < 0.05 were considered significant.

The Cox regression analysis was performed to estimate the hazard ratio (HR) and 95% confidence interval (CI) values using R software (version 4.0.3). The correlation of the autophagy-related lncRNAs was calculated by Pearson correlation. p < 0.05 were regarded as statistically significant. All data and code can be available in Additional file 1, 2, 3, 4, 5, 6 and 7 related to this study.

We identified 14,142 lncRNAs by analyzing the RNA-seq of 162 TSCC patients samples obtained from TCGA. A total of 232 autophagy-related genes were downloaded from the HADb database and we acquired the corresponding expression levels of the patient samples. By setting the threshold of Pearson coefficient and p value, 941 lncRNAs related to autophagy were shortlisted out. The univariate Cox regression analysis was used to obtain 25 autophagy-related lncRNAs that were significantly related to the overall survival time of TSCC patients. Based on the minimum AIC (Table 1), multivariate Cox analysis revealed that ten lncRNAs AC010326.3, AL160006.1, AL122010.1, AC139530.1, AC092747.4, AL139287.1, MIR503HG, AC009318.2, LINC01711, and LINC02560 were the best candidates for constructing a prognostic model (Table 2). Among the ten autophagy-related lncRNAs, risk factors of AC010326.3 and AL139287.1 with HR values were greater than 1, while the rest were considered as protective factors with HR values that were less than 1.

To investigate whether autophagy-related lncRNAs are correlated to the prognosis of TSCC, we have established a risk scoring model. The model is a summation of the expression of lncRNA multiplied by the corresponding coefficient, namely risk score = (0.3919 × expression of AC010326.3) + (− 0.3971 × expression of AL160006.1) + (− 0.4357 × expression of AL122010.1) + (− 0.9133 × expression of AC139530.1) + (− 0.1894 × expression of AC092747.4) + (0.2274 × expression of AL139287.1) + (− 0.2733 × expression of MIR503HG) + (− 0.7450 × expression of AC009318.2) + (− 1.3486 × expression of LINC01711) + (− 0.0652 × expression of LINC02560). TSCC patients were divided into high-risk and low-risk groups based on the median risk score of 1.607. In order to verify the significance of the high-risk and low-risk groups in the OS of TSCC patients, Kaplan–Meier survival curve analysis was used to create the risk survival curve. As the survival time increases, the survival rate of the high-risk group drops sharply, while the result of the low-risk group is better than the high-risk group (Fig. 1A). The areas under the time-dependent receiver operating characteristic (ROC) curve of the autophagy-related lncRNAs prognostic model were 0.782 (Fig. 1B), which suggests the prognosis signature has the potential in predicting survival. Figure 1C shows the survival status between different groups, and Fig. 1D depicts the classification of TSCC patients into high-risk and low-risk groups based on risk scores. These results indicate that the prognosis model based on autophagy-related lncRNAs could accurately classify the survival status and the risk of TSCC patients. The heat map displays the expression levels of ten autophagy-related lncRNAs in the high-risk and low-risk groups (Fig. 1E). The color changes from red to green, indicating a downward trend from high to low expression levels.

We gained the clinical data of TSCC patients from the TCGA database and analyzed the correlation between the risk score and clinicopathological characteristics. The results show that there is no significant statistical difference between the risk score and the risk score and age, gender, grade, AJCC stage, and N stage of TSCC patients (Fig. 2A–E). Meanwhile, in the T stage, patients with T3–T4 had a higher risk score compared with T1–T2 (Fig. 2F, p < 0.05). These analysis results imply that the autophagy-related lncRNA risk score is related to the T stage of TSCC patients.

We further stratified the clinical information of TSCC patients to estimate the autophagy-related lncRNA prognostic model. As shown in Fig. 3, we draw survival curves of this prognostic models based on age (age ≥ 65: p = 7.05e−04; age < 65: p = 1.925e−03), gender (male: p = 1.234e−02; female: p = 1.234e−02), tumor grade (G1–2: p = 2.152e−05; G3–4: p = 6.335e−02), AJCC stage (Stage I–II: p = 5.429e−02; Stage III–IV: p = 2.527e−05), T stage (T1–2: p = 6.708e−04; T3–4: p = 5.056e−03) and N stage (N0: p = 3.141e−03; N1–2: p = 4.263e−04), respectively. Those suggest that the patients with low-risk scores have significantly longer OS times than high-risk scores.

By way of determining whether the prognostic signature we constructed could be an independent prognostic factor for TSCC, we applied univariate and multivariate Cox regression analysis. The results of univariate Cox regression analysis indicated that the Hazard Ratio of AJCC stage (p = 0.018), T stage (p = 0.001), N stage (p = 0.022), and risk score (p < 0.001) in TSCC patients were greater than 1. It shows that these factors are significantly related to the OS of patients (Fig. 4A). In the results of multivariate Cox regression analysis, T stage (p = 0.001) and risk score (p < 0.001) were significantly correlated with OS (Fig. 4B). We further plotted the AUC curve, as shown in Fig. 4C, the risk score of the autophagy-related lncRNA prognostic signature has an AUC of 0.782, which is higher than age (AUC = 0.567), gender (AUC = 0.497), grade (AUC = 0.559), stage (AUC = 0.585), T stage (AUC = 0.681), and N stage (AUC = 0.545). These data suggest that the autophagy-related lncRNA prognostic signature is an independent prognostic factor for TSCC patients.

We constructed a nomogram with a concordance index (C-index) value of 0.81 composed of clinicopathological characteristics, including age, gender, grade, AJCC stage, T stage, N stage, and risk score of prognostic signatures (Fig. 5A). Figure 5B and C show the results of the 1-year and 3-year calibration curve analysis respectively. As shown in the figures, compared with the reference line, the actual survival time is the same as the predicted survival time. This result suggests that the autophagy-related lncRNA prognostic signature that we constructed can accurately predict the prognostic survival time of TSCC patients.

In order to further clarify the relationship between autophagy-related mRNA in the TSCC patients and the ten autophagy-related lncRNAs selected, we obtained 97 autophagy-related mRNAs through the threshold of the Pearson correlation coefficient (|R²| > 0.3) to construct the mRNA-lncRNA co-expression network via Cytoscape (version 3.7.1) (Fig. 6A). Then, the sankey diagram (Fig. 6B) shows the correspondence between autophagy-related lncRNAs and risk factors (risk or protective factors) in the co-expression network. Figure 6C depicted the top ten enriched terms for biological processes (BP), cellular components (CC), and molecular functions (MF) in gene ontology (GO) enrichment analysis. The top three terms in BP are autophagy, process utilizing autophagic mechanism, and macroautophagy. Autophagosome, vacuum membrane, and late endosome are the top three terms in CC. Protein serine/threonine kinase activity, heat shock protein binding, and phosphatase binding are the top three terms in CC. In the KEGG pathway analysis, we showed the top thirty enriched pathways. As shown in Fig. 6D, autophagy, PI3K-Akt signaling pathway, protein processing in endoplasmic reticulum pathway played an indispensable role in the co-expression of mRNA and autophagy-related lncRNAs.

Further functional GSEA revealed that the altered genes (including mRNA and lncRNA) were significantly enriched in autophagy, cancer, and immune-related pathways, for example, pathways in cancer (p = 0.032), mTOR signaling pathway (p = 0.008), p53 signaling pathway (p = 0.045), NOD-like receptor signaling pathway (p = 0.006), Toll-like receptor signaling pathway (p = 0.021), regulation of actin cytoskeleton (p = 0.01), chemokine signaling pathway (p = 0.019) in patients with high risk of TSCC (Fig. 7). This indicates that changes in these pathways may affect the prognosis of TSCC patients. These results provide new insights into the pathogenesis of TSCC and cancer-targeted treatment strategies (such as immunotherapy).