A Coagulation-Related Gene-Based Prognostic Model for Invasive Ductal Carcinoma

CEBPB-mediated upregulation of SERPINA1 promotes colorectal cancer progression by enhancing STAT3 signaling

Colorectal cancer (CRC) is a highly malignant carcinoma associated with poor prognosis, and metastasis is one of the most common causes of death in CRC. Serpin Family A Member 1 (SERPINA1) is a serine protease inhibitor from the Serpin family. Till now, the function and mechanism of SERPINA1 in CRC progression have not been fully illustrated. We established highly metastatic colorectal cancer cells named as RKO-H and Caco2-H by mice liver metastasis model. By integrative bioinformatic approaches, we analyzed the prognostic value and clinical significance of SERPINA1 in CRC, and predicted potential transcription factors. Colony formation, EDU, MTS, Transwell and wound healing assay were performed to evaluate the biological functions of SERPINA1 in CRC in vitro. Experiments in vivo were conducted to explore the effects of SERPINA1 on liver metastasis of CRC. ChIP and luciferase reporter gene assays were performed to identify the transcriptional regulatory mechanism of SERPINA1 by CEBPB. Our results show that SERPINA1 is highly expressed in CRC and correlated with poor clinical outcomes. SERPINA1 promotes the proliferation, migration by activating STAT3 pathway. Mechanistically, CEBPB binds SERPINA1 gene promoter sequence and promotes the transcription of SERPINA1. SERPINA1 drives CEBPB-induced tumor cell growth and migration via augmenting STAT3 signaling. Our results suggest that SERPINA1 is a potential prognostic marker and may serve as a novel treatment target for CRC.

Drug repositioning for immunotherapy in breast cancer using single-cell analysis

Immunomodulatory peptides, while exhibiting potential antimicrobial, antifungal, and/or antiviral properties, can play a role in stimulating or suppressing the immune system, especially in pathological conditions like breast cancer (BC). Thus, deregulation of these peptides may serve as an immunotherapeutic strategy to enhance the immune response. In this meta-analysis, we utilized single-cell RNA sequencing data and known therapeutic peptides to investigate the deregulation of these peptides in malignant versus normal human breast epithelial cells. We corroborated our findings at the chromatin level using ATAC-seq. Additionally, we assessed the protein levels in various BC cell lines. Moreover, our in-house drug repositioning approach was employed to identify potential drugs that could positively impact the relapse-free survival of BC patients. Considering significantly deregulated therapeutic peptides and their role in BC pathology, our approach aims to downregulate B2M and SLPI, while upregulating PIGR, DEFB1, LTF, CLU, S100A7, and SCGB2A1 in BC epithelial cells through our drug repositioning pipeline. Leveraging the LINCS L1000 database, we propose BRD-A06641369 for B2M downregulation and ST-4070043 and BRD-K97926541 for SLPI downregulation without negatively affecting the MHC complex as a significantly correlated pathway with these two genes. Furthermore, we have compiled a comprehensive list of drugs for the upregulation of other selected immunomodulatory peptides. Employing an immunotherapeutic approach by integrating our drug repositioning pipeline with single-cell analysis, we proposed potential drugs and drug targets to fortify the immune system against BC.

Correlation Between miR-497-5p Expression With Clinicopathological Characteristics and Prognosis in Patients With Breast Cancer

Breast cancer (BC) comprises multiple biological and histologic properties. MicroRNAs show key functions in cancer prognosis. This paper explored the relationship between miR-497-5p with clinicopathological characteristics and prognosis in BC. Cancer tissues and normal adjacent tissues (NATs) were collected from 140 included patients with BC. The clinical baseline data, including age, tumor size, pathologic grade, clinical stage, modified Scraff-Bloom-Richardson grade, and lymph node metastasis, were recorded. miR-497-5p expression in cancer tissues and NAT was determined by reverse transcription-quantitative polymerase chain reaction. Patients with BC were followed up for 5 years to record their survival. Patients were divided into the miR-497-5p low expression and high expression groups to assess the correlation between miR-497-5p expression with clinicopathological characteristics and overall survival of patients. The role of miR-497-5p as an independent risk factor for death was further analyzed by a multivariate Cox regression model. miR-497-5p was downregulated in BC tissues than NAT. Tumor size, clinical stage, and lymph node metastasis showed significant differences among patients with high and low miR-497-5p expression levels. Patients with BC with low miR-497-5p expression presented decreased survival. Lowly-expressed miR-497-5p was an independent risk factor for death in patients. Collectively, cancer tissue miR-497-5p low expression increases the risk of death and serves as an independent risk factor for death in patients with BC.

Transcriptome Analysis Reveals Distinct Patterns Between the Invasive and Noninvasive Pituitary Neuroendocrine Tumors

Although most pituitary neuroendocrine tumors (PitNETs)/pituitary adenomas remain intrasellar, a significant proportion of tumors show parasellar invasive growth and 6% to 8% infiltrate the bone structures, thus affecting the prognosis. There is an unmet need to identify novel markers that can predict the parasellar growth of PitNETs. Furthermore, mechanisms that regulate bone invasiveness of PitNETs and factors related to tumor vascularization are largely unknown. We used genome-wide mRNA analysis in a cohort of 77 patients with PitNETs of different types to explore the differences in gene expression patterns between invasive and noninvasive tumors with respect to the parasellar growth and regarding the rare phenomenon of bone invasiveness. Additionally, we studied the genes correlated to the contrast enhancement quotient, a novel radiological parameter of tumor vascularization. Most of the genes differentially expressed related to the parasellar growth were genes involved in tumor invasiveness. Differentially expressed genes associated with bone invasiveness are involved in NF-κB pathway and antitumoral immune response. Lack of clear clustering regarding the parasellar and bone invasiveness may be explained by the influence of the cell lineage-related genes in this heterogeneous cohort of PitNETs. Our transcriptomics analysis revealed differences in the molecular fingerprints between invasive, including bone invasive, and noninvasive PitNETs, although without clear clustering. The contrast enhancement quotient emerged as a radiological parameter of tumor vascularization, correlating with several angiogenesis-related genes. Several of the top genes related to the PitNET invasiveness and vascularization have potential prognostic and therapeutic application requiring further research.

Oral bacterium Streptococcus mutans promotes tumor metastasis through thrombosis formation

Thrombosis is a well-known cardiovascular disease (CVD) complication that has caused death in many patients with cancer. Oral bacteria have been reported to contribute to systemic diseases, including CVDs, and tumor metastasis. However, whether oral bacteria-induced thrombosis induces tumor metastasis remains poorly understood. In this study, the cariogenic oral bacterium Streptococcus mutans was used to examine thrombosis in vitro and in vivo. Investigation of tumor metastasis to the lungs was undertaken by intravenous S. mutans implantation using a murine breast cancer metastasis model. The results indicated that platelet activation, aggregation, and coagulation were significantly altered in S. mutans-stimulated endothelial cells (ECs), with elevated neutrophil migration, thereby inducing thrombosis formation. Streptococcus mutans stimulation significantly enhances platelet and tumor cell adhesion to the inflamed ECs. Furthermore, S. mutans-induced pulmonary thrombosis promotes breast cancer cell metastasis to the lungs in vivo, which can be reduced by using aspirin, an antiplatelet drug. Our findings indicate that oral bacteria promote tumor metastasis through thrombosis formation. Oral health management is important to prevent CVDs, tumor metastasis, and their associated death.

Prognostic characteristics of a six-gene signature based on ssGSEA in sarcoma

BACKGROUND

Sarcoma is a rare malignant tumor originating of the interstitial or connective tissue with a poor prognosis. Next-generation sequencing technology offers new opportunities for accurate diagnosis and treatment of sarcomas. There is an urgent need for new gene signature to predict prognosis and evaluate treatment outcomes.

METHODS

We used transcriptome data from the Cancer Genome Atlas (TCGA) database and single sample gene set enrichment analysis (ssGSEA) to explore the cancer hallmarks most associated with prognosis in sarcoma patients. Then, weighted gene coexpression network analysis, univariate COX regression analysis and random forest algorithm were used to construct prognostic gene characteristics. Finally, the prognostic value of gene markers was validated in the TCGA and Integrated Gene Expression (GEO) (GSE17118) datasets, respectively.

RESULTS

MYC targets V1 and V2 are the main cancer hallmarks affecting the overall survival (OS) of sarcoma patients. A six-gene signature including VEGFA, HMGB3, FASN, RCC1, NETO2 and BIRC5 were constructed. Kaplan-Meier analysis suggested that higher risk scores based on the six-gene signature associated with poorer OS (P < 0.001). The receiver Operating characteristic curve showed that the risk score based on the six-gene signature was a good predictor of sarcoma, with an area under the curve (AUC) greater than 0.73. In addition, the prognostic value of the six-gene signature was validated in GSE17118 with an AUC greater than 0.72.

CONCLUSION

This six-gene signature is an independent prognostic factor in patients with sarcoma and is expected to be a potential therapeutic target for sarcoma.

Identification of a coagulation-related signature correlated with immune infiltration and their prognostic implications in lung adenocarcinoma

BACKGROUND

Lung adenocarcinoma (LUAD) is a fatal form of lung cancer with a poor prognosis. Coagulation system had been confirmed closely related to tumor progression and the hypercoagulable state encouraged the immune infiltration and development of tumor cells, leading to a poor prognosis in cancer patients. However, the use of the coagulation-related genes (CRGs) for prognosis in LUAD has yet to be determined. In this study, we constructed an immune-related signature (CRRS) and identified a potential coagulation-related biomarker (P2RX1).

METHODS

We obtained a total of 209 CRGs based on two coagulation-related KEGG pathways, then developed the CRRS signature by using the TCGA-LUAD RNA-seq data via the procedure of LASSO-Cox regression, stepwise-Cox regression, univariate and multivariate Cox regression. Grouped by the CRRS, Kaplan-Meier survival curves and receiver operating characteristic curves were drawn for the training and validation sets, respectively. In addition, single-sample gene set enrichment analysis was exploited to explore immune infiltration level. Moreover, immunophenotypes and immunotherapy grouped by CRRS were further analyzed.

RESULTS

We developed an immune-related signature (CRRS) composed of COL1A2, F2, PLAUR, C4BPA, and P2RX1 in LUAD. CRRS was an independent risk factor for overall survival and displayed stable and powerful performance. Additionally, CRRS possessed distinctly superior accuracy than traditional clinical variables and molecular features. Functional analysis indicated that the differentially high expressed genes in the low-risk group significantly enriched in T cell and B cell receptor signaling pathways. The low-risk group was sensitive to anti-PD-1/PD-L1 immunotherapy and displayed abundant immune infiltration and immune checkpoint gene expression. Finally, we identified an independent prognostic gene P2RX1. Low expression of P2RX1 associated with poor overall survival and decreased immune infiltration.

CONCLUSIONS

Our study revealed a significant correlation between CRRS and immune infiltration. CRRS could serve as a promising tool to improve the clinical outcomes for individual LUAD patients.

Computational identification and clinical validation of a novel risk signature based on coagulation-related lncRNAs for predicting prognosis, immunotherapy response, and chemosensitivity in colorectal cancer patients

Background

Coagulation is critically involved in the tumor microenvironment, cancer progression, and prognosis assessment. Nevertheless, the roles of coagulation-related long noncoding RNAs (CRLs) in colorectal cancer (CRC) remain unclear. In this study, an integrated computational framework was constructed to develop a novel coagulation-related lncRNA signature (CRLncSig) to stratify the prognosis of CRC patients, predict response to immunotherapy and chemotherapy in CRC, and explore the potential molecular mechanism.

Methods

CRC samples from The Cancer Genome Atlas (TCGA) were used as the training set, while the substantial bulk or single-cell RNA transcriptomics from Gene Expression Omnibus (GEO) datasets and real-time quantitative PCR (RT-qPCR) data from CRC cell lines and paired frozen tissues were used for validation. We performed unsupervised consensus clustering of CRLs to classify patients into distinct molecular subtypes. We then used stepwise regression to establish the CRLncSig risk model, which stratified patients into high- and low-risk groups. Subsequently, diversified bioinformatics algorithms were used to explore prognosis, biological pathway alteration, immune microenvironment, immunotherapy response, and drug sensitivity across patient subgroups. In addition, weighted gene coexpression network analysis was used to construct an lncRNA-miRNA-mRNA competitive endogenous network. Expression levels of CRLncSig, immune checkpoints, and immunosuppressors were determined using RT-qPCR.

Results

We identified two coagulation subclusters and constructed a risk score model using CRLncSig in CRC, where the patients in cluster 2 and the low-risk group had a better prognosis. The cluster and CRLncSig were confirmed as the independent risk factors, and a CRLncSig-based nomogram exhibited a robust prognostic performance. Notably, the cluster and CRLncSig were identified as the indicators of immune cell infiltration, immunoreactivity phenotype, and immunotherapy efficiency. In addition, we identified a new endogenous network of competing CRLs with microRNA/mRNA, which will provide a foundation for future mechanistic studies of CRLs in the malignant progression of CRC. Moreover, CRLncSig strongly correlated with drug susceptibility.

Conclusion

We developed a reliable CRLncSig to predict the prognosis, immune landscape, immunotherapy response, and drug sensitivity in patients with CRC, which might facilitate optimizing risk stratification, guiding the applications of immunotherapy, and individualized treatments for CRC.

Live-imaging studies reveal how microclots and the associated inflammatory response enhance cancer cell extravasation

Previous clinical studies and work in mouse models have indicated that platelets and microclots might enable the recruitment of immune cells to the pre-metastatic cancer niche, leading to efficacious extravasation of cancer cells through the vessel wall. Here, we investigated the interaction between platelets, endothelial cells, inflammatory cells, and engrafted human and zebrafish cancer cells by live-imaging studies in translucent zebrafish larvae, and show how clotting (and clot resolution) act as foci and as triggers for extravasation. Fluorescent tagging in each lineage revealed their dynamic behaviour and potential roles in these events, and we tested function by genetic and drug knockdown of the contributing players. Morpholino knockdown of fibrinogen subunit α (fga) and warfarin treatment to inhibit clotting both abrogated extravasation of cancer cells. The inflammatory phenotype appeared fundamental, and we show that forcing a pro-inflammatory, tnfa-positive phenotype is inhibitory to extravasation of cancer cells.

Establishment of a N1-methyladenosine-related risk signature for breast carcinoma by bioinformatics analysis and experimental validation

OBJECTIVES

Breast carcinoma (BRCA) has resulted in a huge health burden globally. N1-methyladenosine (m¹A) RNA methylation has been proven to play key roles in tumorigenesis. Nevertheless, the function of m¹A RNA methylation-related genes in BRCA is indistinct.

METHODS

The RNA sequencing (RNA-seq), copy-number variation (CNV), single-nucleotide variant (SNV), and clinical data of BRCA were acquired via The Cancer Genome Atlas (TCGA) database. In addition, the GSE20685 dataset, the external validation set, was acquired from the Gene Expression Omnibus (GEO) database. 10 m¹A RNA methylation regulators were obtained from the previous literature, and further analyzed through differential expression analysis by rank-sum test, mutation by SNV data, and mutual correlation by Pearson Correlation Analysis. Furthermore, the differentially expressed m¹A-related genes were selected through overlapping m¹A-related module genes obtained by weighted gene co-expression network analysis (WGCNA), differentially expressed genes (DEGs) in BRCA and DEGs between high- and low- m¹A score subgroups. The m¹A-related model genes in the risk signature were derived by univariate Cox and least absolute shrinkage and selection operator (LASSO) regression analyses. In addition, a nomogram was built through univariate and multivariate Cox analyses. After that, the immune infiltration between the high- and low-risk groups was investigated through ESTIMATE and CIBERSORT. Finally, the expression trends of model genes in clinical BRCA samples were further confirmed by quantitative real-time PCR (RT‒qPCR).

RESULTS

Eighty-five differentially expressed m¹A-related genes were obtained. Among them, six genes were selected as prognostic biomarkers to build the risk model. The validation results of the risk model showed that its prediction was reliable. In addition, Cox independent prognosis analysis revealed that age, risk score, and stage were independent prognostic factors for BRCA. Moreover, 13 types of immune cells were different between the high- and low-risk groups and the immune checkpoint molecules TIGIT, IDO1, LAG3, ICOS, PDCD1LG2, PDCD1, CD27, and CD274 were significantly different between the two risk groups. Ultimately, RT-qPCR results confirmed that the model genes MEOX1, COL17A1, FREM1, TNN, and SLIT3 were significantly up-regulated in BRCA tissues versus normal tissues.

CONCLUSIONS

An m¹A RNA methylation regulator-related prognostic model was constructed, and a nomogram based on the prognostic model was constructed to provide a theoretical reference for individual counseling and clinical preventive intervention in BRCA.

Identification of genetic and immune signatures for the recurrence of HER2-positive breast cancer after trastuzumab-based treatment

PURPOSE

To determine the genetic and immune features associated with the recurrence of human epidermal growth factor receptor2-positive (HER2 +) breast cancer (BC) after trastuzumab-based treatment.

METHODS

A retrospective cohort study of 48 patients who received trastuzumab-based treatment was divided into recurrent and non-recurrent groups according to clinical follow-up. Baseline samples from all 48 patients were analyzed for genetic variation, HLA allele type, gene expression, and immune features, which were linked to HER2 + BC recurrence. Statistics included logistic regression models, Kaplan-Meier plots, and Univariate Cox proportional hazards models.

RESULTS

Compared with the non-recurrent group, the extracellular matrix-related pathway and 3 Hallmark gene sets were enriched in the recurrent group. The infiltration levels of immature B cells and activated B cells were significantly increased in the non-recurrent group, which correlated remarkably with improved overall survival (OS) in two other published gene expression datasets, including TCGA and METABRIC. In the TCGA cohort (n = 275), activated B cells (HR 0.23, 95%CI 0.13-0.43, p < 0.0001), and immature B cells (HR 0.26, 95%CI 0.12-0.59, p < 0.0001). In the METABRIC cohort (n = 236), activated B cells (HR 0.60, 95%CI 0.43-0.83, p = 0.002), and immature B cells (HR 0.65, 95%CI 0.47-0.91, p = 0.011). Cox regression suggested that immature B cells and activated B cells were protective factors for outcome OS.

CONCLUSIONS

Aberrant activation of multiple pathways and low baseline tumor-infiltrating B cells are related to HER2 + BC trastuzumab-based recurrence, which primarily affects the antitumor activity of trastuzumab.

Proteomics-Based Identification of Dysregulated Proteins and Biomarker Discovery in Invasive Ductal Carcinoma, the Most Common Breast Cancer Subtype

Invasive ductal carcinoma (IDC) is the most common histological subtype of malignant breast cancer (BC), and accounts for 70–80% of all invasive BCs. IDC demonstrates great heterogeneity in clinical and histopathological characteristics, prognoses, treatment strategies, gene expressions, and proteomic profiles. Significant proteomic determinants of the progression from intraductal pre-invasive malignant lesions of the breast, which characterize a ductal carcinoma in situ (DCIS), to IDC, are still poorly identified, validated, and clinically applied. In the era of “6P” medicine, it remains a great challenge to determine which patients should be over-treated versus which need to be actively monitored without aggressive treatment. The major difficulties for designating DCIS to IDC progression may be solved by understanding the integrated genomic, transcriptomic, and proteomic bases of invasion. In this review, we showed that multiple proteomics-based techniques, such as LC–MS/MS, MALDI-ToF MS, SELDI-ToF-MS, MALDI-ToF/ToF MS, MALDI-MSI or MasSpec Pen, applied to in-tissue, off-tissue, BC cell lines and liquid biopsies, improve the diagnosis of IDC, as well as its prognosis and treatment monitoring. Classic proteomics strategies that allow the identification of dysregulated protein expressions, biological processes, and interrelated pathway analyses based on aberrant protein–protein interaction (PPI) networks have been improved to perform non-invasive/minimally invasive biomarker detection of early-stage IDC. Thus, in modern surgical oncology, highly sensitive, rapid, and accurate MS-based detection has been coupled with “proteome point sampling” methods that allow for proteomic profiling by in vivo “proteome point characterization”, or by minimal tissue removal, for ex vivo accurate differentiation and delimitation of IDC. For the detection of low-molecular-weight proteins and protein fragments in bodily fluids, LC–MS/MS and MALDI-MS techniques may be coupled to enrich and capture methods which allow for the identification of early-stage IDC protein biomarkers that were previously invisible for MS-based techniques. Moreover, the detection and characterization of protein isoforms, including posttranslational modifications of proteins (PTMs), is also essential to emphasize specific molecular mechanisms, and to assure the early-stage detection of IDC of the breast.

Identification of new co-diagnostic genes for sepsis and metabolic syndrome using single-cell data analysis and machine learning algorithms

Sepsis, a serious inflammatory response that can be fatal, has a poorly understood pathophysiology. The Metabolic syndrome (MetS), however, is associated with many cardiometabolic risk factors, many of which are highly prevalent in adults. It has been suggested that Sepsis may be associated with MetS in several studies. Therefore, this study investigated diagnostic genes and metabolic pathways associated with both diseases. In addition to microarray data for Sepsis, PBMC single cell RNA sequencing data for Sepsis and microarray data for MetS were downloaded from the GEO database. Limma differential analysis identified 122 upregulated genes and 90 downregulated genes in Sepsis and MetS. WGCNA identified brown co-expression modules as Sepsis and MetS core modules. Two machine learning algorithms, RF and LASSO, were used to screen seven candidate genes, namely, STOM, BATF, CASP4, MAP3K14, MT1F, CFLAR and UROD, all with an AUC greater than 0.9. XGBoost assessed the co-diagnostic efficacy of Hub genes in Sepsis and MetS. The immune infiltration results show that Hub genes were expressed at high levels in all immune cells. After performing Seurat analysis on PBMC from normal and Sepsis patients, six immune subpopulations were identified. The metabolic pathways of each cell were scored and visualized using ssGSEA, and the results show that CFLAR plays an important role in the glycolytic pathway. Our study identified seven Hub genes that serve as co-diagnostic markers for Sepsis and MetS and revealed that diagnostic genes play an important role in immune cell metabolic pathway.

A universal co-expression gene network and prognostic model for hepatic-biliary-pancreatic cancers identified by integrative analyses

Hepatic, biliary and pancreatic cancers are a diverse set of malignancies with poor prognoses. It is possible that common molecular mechanisms are involved in the carcinogenesis of these cancers. Here, we identified LINC01537 and seven protein-coding genes by integrative analysis of transcriptomes of mRNAs, microRNAs and long non-coding RNAs from cholangiocarcinoma, hepatocellular carcinoma and pancreatic adenocarcinoma cohorts in TCGA. A predictive model constructed from seven biomarkers was established to successfully predict the survival rate of patients, which was then further verified in external cohorts. Additionally, patients with high-risk scores in our model were prone to epithelial-mesenchymal transition. Finally, activation of the biomarker PDE2A significantly attenuated migration and epithelial-mesenchymal transition in the HepG2 liver cancer cell line.

Identification of miRNA biomarkers for stomach adenocarcinoma

BACKGROUND

Stomach adenocarcinoma (STAD) is a common malignant tumor in the world and its prognosis is poor, miRNA plays a role mainly by influencing the expression of mRNAs, and participates in the occurrence and development of tumors. However, reliable miRNA prognostic models for stomach adenocarcinoma remain to be identified.

RESULTS

Using the data from the Cancer Genome Atlas (TCGA), a prognostic model of stomach adenocarcinoma was established including tumor stage and expression levels of 4 miRNAs (hsa-miR-379-3p, hsa-miR-2681-3p, hsa-miR-6499-5p and hsa-miR-6807-3p). A total of 50 ultimate target genes of these miRNAs were obtained through prediction. Enrichment analysis revealed that target genes were mainly concentrated in neural function and TGF-β and FoxO signaling pathways. Survival analysis showed that three model miRNAs (hsa-miR-379-3p, hsa-miR-2681-3p and hsa-miR-6807-3p) and five final target genes (DLC1, LRFN5, NOVA1, POU3F2 and PRICKLE2) were associated with the patient's overall survival outcome.

CONCLUSIONS

We used bioinformatics methods to screen new prognostic miRNA markers from TCGA and established a prognostic model of STAD, so as to provide a basis for the diagnosis, prognosis, and treatment of STAD in the future.

Transcriptome analysis reveals the prognostic and immune infiltration characteristics of glycolysis and hypoxia in head and neck squamous cell carcinoma

BACKGROUND

This study aims to construct a new prognostic gene signature in survival prediction and risk stratification for patients with Head and neck squamous cell carcinoma (HNSCC).

METHOD

The transcriptome profiling data and hallmark gene sets in the Molecular Signatures Database was used to explore the cancer hallmarks most relevant to the prognosis of HNSCC patients. Differential gene expression analysis, weighted gene co-expression network analysis, univariate COX regression analysis, random forest algorithm and multiple combinatorial screening were used to construct the prognostic gene signature. The predictive ability of gene signature was verified in the TCGA HNSCC cohort as the training set and the GEO HNSCC cohorts (GSE41613 and GSE42743) as the validation sets, respectively. Moreover, the correlations between risk scores and immune infiltration patterns, as well as risk scores and genomic changes were explored.

RESULTS

A total of 3391 differentially expressed genes in HNSCC were screened. Glycolysis and hypoxia were screened as the main risk factors for OS in HNSCC. Using univariate Cox analysis, 97 prognostic candidates were identified (P < 0.05). Top 10 important genes were then screened out by random forest. Using multiple combinatorial screening, a combination with less genes and more significant P value was used to construct the prognostic gene signature (RNF144A, STC1, P4HA1, FMNL3, ANO1, BASP1, MME, PLEKHG2 and DKK1). Kaplan-Meier analysis showed that patients with higher risk scores had worse overall survival (p < 0.001). The ROC curve showed that the risk score had a good predictive efficiency (AUC > 0.66). Subsequently, the predictive ability of the risk score was verified in the validation sets. Moreover, the two-factor survival analysis combining the cancer hallmarks and risk scores suggested that HNSCC patients with the high hypoxia or glycolysis & high risk-score showed the worst prognosis. Besides, a nomogram based on the nine-gene signature was established for clinical practice. Furthermore, the risk score was significantly related to tumor immune infiltration profiles and genome changes.

CONCLUSION

This nine-gene signature associated with glycolysis and hypoxia can not only be used for prognosis prediction and risk stratification, but also may be a potential therapeutic target for patients with HNSCC.