Identification and Validation of a Novel Genomic Instability-Associated Long Non-Coding RNA Prognostic Signature in Head and Neck Squamous Cell Carcinoma

Abstract

Background: Head and neck squamous cell carcinoma (HNSCC) is one of the most aggressive malignant cancers worldwide, and accurate prognostic models are urgently needed. Emerging evidence revealed that long non-coding RNAs (lncRNAs) are related to genomic instability. We sought to identify and validate a genomic instability-associated lncRNA prognostic signature to assess HNSCC patient survival outcomes. Methods: RNA-sequencing data, somatic mutation files, and patient clinical data were downloaded from The Cancer Genome Atlas database. A total of 491 patients with completely clinical files were randomly divided into training and testing sets. In the training set, genomic instability-associated lncRNAs were screened through univariate Cox regression analyses and least absolute shrinkage and selection operator regression analyses to build a genomic instability-associated lncRNA signature (GILncSig). In addition, time-dependent receiver operating characteristic (ROC) curve, Kaplan-Meier survival curve, and clinical stratification analyses were used to evaluate the signature's reliability. Finally, in situ hybridization experiments were performed to validate GILncSig expression levels between adjacent non-tumor tissues and tumor tissues from HNSCC patients. Results: Four genomic instability-associated lncRNAs (AC023310.4, AC091729.1, LINC01564, and MIR3142HG) were selected for the prognostic signature. The model was successfully validated using the testing cohort. ROC analysis demonstrated its strong predictive ability for HNSCC prognosis. Univariate and multivariate Cox analyses revealed that the GILncSig was an independent predictor of prognosis. HNSCC patients with a low-risk score showed a substantially better prognosis than the high-risk groups. The in situ hybridization experiments using human HNSCC tissue revealed high GILncSig expression in HNSCC tissues compared with adjacent non-tumor tissues. Conclusion: We developed a novel GILncSig for prognosis prediction in HNSCC patients, and the components of that signature might be therapeutic targets for HNSCC.

Keywords: genomic instability; head and neck squamous cell carcinoma; long non-coding RNA; prognostic signature; survival.

FIGURE 1

Graphical abstract of genomic instability-associated lncRNAs signature establishment in head and neck squamous cell carcinoma.

FIGURE 2

Screening and identifying of genomic instability-related lncRNAs and their functional annotation in patients with HNSCC. (A) Volcano plot of 103 differential expressed lncRNAs between the GU-like and GS-like groups. Upregulated lncRNAs are shown in red on the right, whereas downregulated lncRNAs are shown in green on the left. (B) Heatmap of expression of the top 50 differential lncRNAs, divided into the GS-like groups (blue) and GU-like group (red). (C) An unsupervised clustering of 492 patients with HNSCC was performed based on the expression patterns of 103 candidate genomic instability-associated lncRNAs. The GS-like groups is shown in blue on the right, the GU-like groups is shown in red on the left. (D) Boxplots of somatic mutation levels in the GU-like and GS-like groups. (E) Boxplots of UBQLN4 expression level in the GU-like and GS-like groups. (F) Co-expression networks of differential lncRNAs (blue) and their related mRNAs (red) based on the Pearson correlation coefficients. (G) GO enrichment analysis and KEGG pathway analysis of the lncRNA-mRNA network. GO, Gene Ontology; HNSCC, head and neck squamous cell carcinoma; KEGG, Kyoto Encyclopedia of Genes and Genomes.

FIGURE 3

Construction of a prognostic model related to overall survival of HNSCC patients based on genome instability-related lncRNAs in the training set. (A) Seven prognostic relevant lncRNAs based on univariate Cox regression analysis. (B) Screening the Log Lambda value corresponding to the minimum cross-validation error point. (C) The distribution plot of the LASSO coefficient. Selecting genome instability-related lncRNAs with a non-zero coefficient corresponding to the same Log Lambda value. (D) Multivariate Cox regression analysis revealed four independent genome instability-related lncRNAs related to patient prognosis. Two lncRNAs were protective (AC023310.4 and LINC01564), and two were risk factors for shorter survival (AC091729.1 and MIR3142HG). (E) Kaplan-Meier survival curves for HNSCC patients in the high- and low-risk groups grouped by the GILncSig score in the training set (F) Time-independent receiver operating characteristic curves of the GILncSig in the training set. (G) LncRNA expression patterns and the distributions of somatic mutations and UBQLN4 expression with increasing GILncSig scores in the training set. GILncSig, genomic instability-associated lncRNAs signature; HNSCC, head and neck squamous cell carcinoma.

FIGURE 4

Validation of the predictive performance of the genome instability-related lncRNAs signature in the testing and TCGA sets. (A) Kaplan-Meier survival curves for HNSCC patients in the high- and low-risk groups grouped by the GILncSig score in the testing set. (B) Time-independent receiver operating characteristic curves of the GILncSig in the testing set. (C) lncRNA expression patterns and the distributions of somatic mutation and UBQLN4 expression with increasing GILncSig score in the testing set. (D) Kaplan-Meier survival curves for HNSCC patients in the high- and low-risk groups grouped by the GILncSig score in the entire TCGA-HNSC set. (E) Time-independent receiver operating characteristic curves of the GILncSig in the entire TCGA-HNSC set. (F) LncRNA expression patterns and the distributions of somatic mutation and UBQLN4 expression with increasing GILncSig score in the entire TCGA-HNSC set. GILncSig, genomic instability-associated lncRNAs signature; HNSCC, head and neck squamous cell carcinoma.

FIGURE 5

Stratification analysis of the genome instability-related lncRNAs signature. (A–M) Kaplan-Meier analysis of clinical subgroups based on the GILncSig scores. The clinical characteristics including: male (A), female (B), age ≤65 (C), age >65 (D), stage I–II (E), stage III–IV (F), T1–2 (G), T3–4 (H), N0 (I), N1–3 (J), G1 (K), G2 (L), and G3 (M).

FIGURE 6

Construction and evaluation of a nomogram based on the genome instability-related lncRNAs signature in the TCGA-HNSC cohort. (A) Development of a nomogram based on the GILncSig score. (B–D) Calibration plots for the signature at 1, 3, and 5 years. GILncSig, genomic instability-associated lncRNAs signature.

FIGURE 7

Relationship between the GILncSig and DNAH5 somatic mutation and model comparison. (A) Kaplan-Meier curve analysis of overall survival of patients with DNAH5 mutant or wild-type status for the combined GS-like and GU-like groups. (B) Time-independent receiver operating characteristic curves of overall survival for GILncSig, Jiang’s LncSig, and Ji’s LncSig. GILncSig, genomic instability-associated lncRNAs signature; DNAH5, dynein axonemal heavy chain 5.

FIGURE 8

Verification of the expression levels of the genomic instability-associated lncRNAs signature in clinical samples. (A) Representative images of in situ hybridization experiments in HNSCC patients. Nucleus stained with hematoxylin appear blue, and positive expression of DAB is brownish yellow. (B) Relative expression of the four lncRNAs in HNSCC patients.