Identification and Validation of a Prognostic Signature Derived from the Cancer Stem Cells for Oral Squamous Cell Carcinoma

Abstract

The progression and metastasis of oral squamous cell carcinoma (OSCC) are highly influenced by cancer stem cells (CSCs) due to their unique self-renewal and plasticity. In this study, data were obtained from a single-cell RNA-sequencing dataset (GSE172577) in the GEO database, and LASSO-Cox regression analysis was performed on 1344 CSCs-related genes to establish a six-gene prognostic signature (6-GPS) consisting of ADM, POLR1D, PTGR1, RPL35A, PGK1, and P4HA1. High-risk scores were significantly associated with unfavorable survival outcomes, and these features were thoroughly validated in the ICGC. The results of nomograms, calibration plots, and ROC curves confirmed the good prognostic accuracy of 6-GPS for OSCC. Additionally, the knockdown of ADM or POLR1D genes may significantly inhibit the proliferation, migration, and invasion of OSCC cells through the JAK/HIF-1 pathway. Furthermore, cell-cycle arrest occurred in the G1 phase by suppressing Cyclin D1. In summary, 6-GPS may play a crucial role in the occurrence and development of OSCC and has the potential to be developed further as a diagnostic, therapeutic, and prognostic tool for OSCC.

Keywords: ADM; POLR1D; RNA-sequencing; cancer stem cells; oral squamous cell carcinoma; prognosis signature.

Figure 1

Cancer stem cell clusters were discovered using scRNA-seq analysis. (A) The UMAP map is colored by various cell clusters; (B) Cell types recognized by marker genes; (C) The expression level of TACSTD2 in each cell cluster; (D) The expression level of KRT19 in each cell cluster; (E) TACSTD2 expression details; (F) KRT19 expression details.

Figure 2

Univariate Cox regression analysis was used to screen the genes related to the prognosis of OSCC (p < 0.05 was statistically significant).

Figure 3

Establishment of a prognosis-associated signature for OSCC. (A) 6-GPS-based LASSO cross-validation plot. (B) LASSO coefficient of 6-GPS in OSCC. (C) Visualization of risk scores of OSCC patients. (D) Visualization of dead and alive OSCC patients with high- and low-risk scores. (E) Heatmap of the expression of the 6-GPS in specimens with high-risk and low-risk scores. (F) Survival analyses of OSCC subjects with high- and low-risk scores with K-M curves. (G) Estimation of the predictive ability of the nomogram in OSCC prognosis using the ROC curves.

Figure 4

Estimated prognostic accuracy of 6-GPS in patients with OSCC. (A) Nomogram shows that the age, gender, clinical stage, and risk score were associated with 1-, 3-, and 5-year OS. (B) Calibration plots for showing the deviation between model-estimated and observed 1-, 3-, and 5-year survival.

Figure 5

Risk score analysis of 6-GPS in the ICGC. (A–D) The risk score, survival status of OSCC patients, heatmap of the 6-GPS expression, and survival curves between low and high-risk groups are shown. (E) Time-independent ROC analysis of risk score for predicting the OS in the ICGC cohort.

Figure 6

Expression levels of 6-GPS (A) Comparing the 6-GPS expression levels in all OSCC and normal epithelium tissues. (B) Comparison of 6-GPS expression levels between cancer and adjacent normal tissues from the same patient. (C) Negative expression of PTGR1 was observed in normal oral mucosa (HPA036724). (D) Negative expression of PGK1 was observed in normal oral mucosa (HPA073644). (E) Medium expression of POLR1D was observed in normal skin (25–75%, HPA039337). (F) Medium expression of P4HA1 was observed in normal oral mucosa (>75%, HPA026593). (G) Low expression of PTGR1 was observed in OSCC (HPA036724). (H) High expression of PGK1 was observed in OSCC (>75%, CAB010065). (I) Medium expression of POLR1D was observed in OSCC (>75%, HPA039337). (J) High expression of P4HA1 was observed in OSCC (>75%, HPA026593). (K–P) K-M survival curves for the 6-GPS. (*: p < 0.05; ***: p < 0.001; ns: No significance).

Figure 7

6-GPS co-expression network and functional enrichment. (A) PPI network and hub clustering genes of 6-GPS. (B) GeneMANIA demonstrates the gene interaction network of 6-GPS. (C) GO/KEGG enrichment analysis for 6-GPS. (D) Network visualization of GO/KEGG enrichment analysis for 6-GPS.

Figure 8

Correlations between ADM/POLR1D knockdown and biological characterizations of CAL-27 and SAS cells. The expression of ADM (A) and POLR1D (B) was interfered with using siRNAs. The interfering efficiency was examined by RT-qPCR. CCK-8 assay was conducted to detect the proliferation of CAL-27 (C) and SAS (D) cells after the targeted gene silence. Cell migration (E–F) and Invasion (G–H) were detected in CAL-27 and SAS cells transfected with RNAi-ADM or RNAi-POLR1D by Transwell assay. mRNA expression levels of MMP2 in CAL-27 and SAS cells with the ADM or POLR1D knockdown (I). Cell-cycle analysis demonstrated an increase in the proportions of the G1 phase upon knockdown of ADM and POLR1D (J). Data are shown as mean and SD of triplicates (mean ± SD). (*: p < 0.05; **: p < 0.01; ***: p < 0.001; ns: No significance; vs. RNAi-NC).

Figure 9

Mechanistic role of ADM and POLR1D in OSCC. Flow cytometry analysis of CD133 cell surface expression as a marker of cancer stem cells ((A) Flow cytometry images; (B) statistical analysis). The networking genes of ADM and POLR1D expression levels in CAL-27 (C) and SAS (D) cells were checked by RT-qPCR. Spearman’s correlation analysis between ADM (E) and POLR1D (F) and pathway scores. Expression of the protein in SAS cell was determined by Western blot ((G) Western blot images; (H) statistical analysis). Data are shown as mean and SD of triplicates (mean ± SD). (*: p < 0.05; **: p < 0.01; ***: p < 0.001; vs. RNAi-NC).