Construction and validation of a prognostic model based on ten signature cell cycle-related genes for early-stage lung squamous cell carcinoma

Abstract

BACKGROUND:

We performed a bioinformatics analysis to screen for cell cycle-related differentially expressed genes (DEGs) and constructed a model for the prognostic prediction of patients with early-stage lung squamous cell carcinoma (LSCC).

METHODS:

From a gene expression omnibus (GEO) database, the GSE157011 dataset was randomly divided into an internal training group and an internal testing group at a 1:1 ratio, and the GSE30219, GSE37745, GSE42127, and GSE73403 datasets were merged as the external validation group. We performed single-sample gene set enrichment analysis (ssGSEA), univariate Cox analysis, and difference analysis, and identified 372 cell cycle-related genes. Additionally, we combined LASSO/Cox regression analysis to construct a prognostic model. Then, patients were divided into high-risk and low-risk groups according to risk scores. The internal testing group, discovery set, and external verification set were used to assess model reliability. We used a nomogram to predict patient prognoses based on clinical features and risk values. Clinical relevance analysis and the Human Protein Atlas (HPA) database were used to verify signature gene expression.

RESULTS:

Ten cell cycle-related DEGs (EIF2B1, FSD1L, FSTL3, ORC3, HMMR, SETD6, PRELP, PIGW, HSD17B6, and GNG7) were identified and a model based on the internal training group constructed. From this, patients in the low-risk group had a higher survival rate when compared with the high-risk group. Time-dependent receiver operating characteristic (tROC) and Cox regression analyses showed the model was efficient and accurate. Clinical relevance analysis and the HPA database showed that DEGs were significantly dysregulated in LSCC tissue.

CONCLUSION:

Our model predicted the prognosis of early-stage LSCC patients and demonstrated potential applications for clinical decision-making and individualized therapy.

Keywords

Lung squamous cell carcinoma cell cycle-related differentially expressed genes prognostic signature survival

1. Introduction

Globally, lung cancer has the highest incidence and mortality rate, with 1.8 million individuals annually diagnosed and approximately 1.6 million dying from the disease every year [1]. Non-small cell lung cancer (NSCLC) accounts for 85%–90% of all these cases [2]. As the second most common NSCLC type, lung squamous cell carcinoma (LSCC) is more likely to show a central type by imaging [3]. LSCC survival rates appear to be worse when compared with lung adenocarcinoma (LUAD) due to the fact that LSCC develops more frequently in smokers who suffer with chronic obstructive pulmonary disease and heart disease [4]. Equally, LSCC survival rates are also affected by several clinical factors, including the Tumor-Node-Metastasis (TNM) stage, age, gender, and health status. At the cell and molecular level, the expression of particular genes can robustly reflect LSCC diagnostics and prognostics. Additionally, when compared with late-stage NSCLC patients, early-stage patients have a better overall survival (OS) rate due to the presence of unique molecular mechanisms [5, 6]. In recent years, LSCC treatment has not been limited to traditional therapies; along with chemoradiotherapy and surgery, personalized treatment plans based on gene mutations and immunotherapy have increased survival rates in some LSCC patient groups [7].

Despite considerable progress in novel therapies for LSCC, disease verification and evaluation methodologies are relatively poor. In clinical settings, most clinicians use TNM staging and histological classification to evaluate carcinoma development. Previously, it was reported that several factors (e.g., age and gender) could be better disease predictors than TNM staging [8, 9, 10], thus, it is important to develop novel and robust models to estimate prognoses in LSCC patients.

Several studies have reported that the cell cycle is closely connected with LUAD cell growth and proliferation, with some genes having roles in these pathways. For example, the CUL4A ${}^{\text{knockdown}}$ but not the CUL4B ${}^{\text{knockdown}}$ model arrested cells in G1 phase, whereas the CUL4B ${}^{\text{knockdown}}$ promoted cell apoptosis by upregulating FOXO3A expression [11]. Recently, high-throughput techniques (RNA-seq and microarrays) have facilitated the convenient and large-scale screening of huge genome datasets [12]. Recent evidence has shown that NSCLC patients have unique gene profiles, while distinct mutual exclusivity and co-occurrence patterns exist between LUAD and LUSC [13]. Therefore, we identified and characterized several cell cycle-related differentially expressed genes (DEGs), and then constructed a novel and robust signature model by combining this information with genome data and patient clinical features to predict the prognosis of patients with early-stage LSCC.

2. Materials and methods

2.1 Microarray data

We downloaded gene expression profiles from GSE157011, GSE30219, GSE37745, GSE42127, and GSE73403 datasets encompassing 678 early-stage LSCC tissue samples from the Gene Expression Omnibus (GEO) database of the National Center for Biotechnology Information (https://www.ncbi.nlm.nih.gov/geo/), based on the GPL570 platform. Five profiles, merged by gene symbols and the SVA software package in R (Version 3.6.3), were used to balance batch effects [14, 15]. Then, GSE157011 (containing 483 samples) was defined as the discovery set and the other four datasets (containing 195 samples) were defined as the validation group.

2.2 Acquisition of key cell cycle-related genes and difference analysis

The reference gene set was acquired from “KEGG_CELL_CYCLE” (http://www.genome.jp/kegg/pathway/hsa/hsa04110.html) and “GO_CELL_CYCLE” (http://amigo.geneontology.org/amigo/term/GO:0007049) pathways, both of which were screened from the MSigDB2 database [16]. From the discovery set, we calculated enrichment scores based on Spearman’s correlations in each patient using single-sample gene set enrichment analysis (ssGSEA) [17]. After this, genes whose $p$ -values were $<$ 0.01 and absolute correlation values were $>$ 0.3 were selected as key early-stage LSCC cell cycle-related genes. These steps were performed in the “GSVA” package in R (Version 3.6.3). Further gene filtering was conducted using univariate Cox regression analysis. RNA sequencing data from 501 LSCC samples and 49 normal samples were downloaded from The Cancer Genome Atlas (TCGA). Then, difference analysis of LSCC specific cell cycle-related genes was performed using the “limma” package in R (Version 3.6.3) [18]. $|$ Log ${}_{2}$ FC $|$ $>$ 1 and $P\leqslant$ 0.05 values were used as criteria for screening DEGs.

2.3 Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis

Biological functions and pathways of specific cell cycle-related DEGs were identified by GO and KEGG enrichment analyses, respectively. GO comprised three terms: molecular function (MF), biological process (BP), and cellular component (CC), and they were calculated using the “clusterProfilter” package in R (Version 3.6.3) [19]. A $P<$ 0.05 value indicated enrichment.

2.4 Model construction and validation

The GSE157011 discovery set was randomly grouped into two parts in a 1:1 scale, and were defined as the internal training group and the internal testing group. The least absolute shrinkage and selection operator (LASSO) was used to further predict candidate cell cycle-related DEGs, and was used to construct the prognosis prediction model in the internal training group [20]. From multivariate Cox regression, a signature was established, and the risk score of each patient calculated using the following formula:

$\displaystyle\text{Risk score}=\mathop{\sum}\limits_{k=1}^{n}\textit{Coef}_{k}% *X_{k},$

Where $\textit{Coef}_{k}=$ the risk coefficient of each factor and $X_{k}=$ the expression value of each candidate cell cycle-related gene.

Based on the median risk score, patients in the internal training group were divided into low-risk and high-risk groups. Kaplan-Meier (KM) survival curves and time-dependent receiver operating characteristic curves (tROC), evaluating the OS of both groups, were plotted to assess model efficiency. The model was also verified in the internal testing group, the external validation set, and all groups. To verify if risk scores were independent risk factors, risk score and clinical feature combinations were examined using univariate and multivariate Cox regression analysis. A $P<$ 0.05 value indicated statistical significance.

2.5 Developing and assessing a predictive nomogram based on clinical features

A prognostic nomogram was established to predict 1, 3, and 5 year patient survival rates based on clinical features and risks in the internal training set, while a calibration plot was used to assess the degree of fitness by comparing with facts. The “rms” and “survival” packages in R were used for this (Version 3.6.3).

2.6 Identifying molecular subtypes and validating prognostic genes

Using the “ConsensusClusterPlus” package in R (Version 3.6.3), the candidate gene matrix was used to identify molecular subtypes in early-stage LSCC Then, survival analysis was performed between these subtypes. $P<$ 0.05 indicated statistical significance. The Human Protein Atlas (HPA) database [21] was used to identify prognostic DEGs between tumor and adjacent tissue in early-stage LSCC.

2.7 Clinical relevance analysis

According to the integration of clinical information and expression of prognostic genes, the relevance between a series of integrated matrix parameters was plotted using the “beeswarm” package in R (Version 3.6.3). A $P<$ 0.05 value was regarded as significant.

3. Results

3.1 Acquisition and screening cell cycle-related DEGs

After balancing for batch effects, GSE157011 raw data, including gene expression and clinical information were processed in ssGSEA. Using Spearman’s correlations and $P$ values, 3737 cell cycle-related genes were identified based on reference sets. Additionally, univariate Cox analysis was used to filter out 467 genes whose $P$ values were $<$ 0.05. In the TCGA LSCC dataset, we identified 372 DEGs based on these 467 genes (Fig. 1A and B).

Figure 1.

The acquisition of key cell cycle-related differentially expressed genes (DEGs). (A) A heatmap showing cell cycle-related DEGs. (B) A volcano map showing cell cycle-related DEGs. (C, D) Gene Ontology (GO) terms in enrichment analysis. (E) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways in enrichment analysis.

Figure 2.

Prognostic model of the internal training group and risk signatures with 10 differentially expressed genes (DEGs).(A, B) Cell cycle-related DEGs were screened using log ( $\lambda$ ) with the minimal deviation and the coefficient other than 0. (C) Survival analysis in high and low patient groups. (D) The AUC of the ROC curve for overall survival (OS). (E) Risk scores in high and low patient groups and a heatmap showing the 10 DEGs. ROC, Receiver Operating Characteristic; AUC, Area under Curve.

Figure 3.

Model validation. Risk scores and survival analysis of high and low patient groups. The AUC of the ROC curve for OS and a risk plot and heatmap of the 10 DEGs in the (A) internal testing group, (B) discovery set, (C) external validation set, and (D) all samples. OS, overall survival; ROC, Receiver Operating Characteristic; AUC, Area under Curve.

Figure 4.

The AUC of the ROC for OS was calculated in the model using different clinical features. Survival analysis of high and low patient groups and the AUC of the ROC for (A, B) age, (C, D) gender, (E, F) T1a $+$ b and T2a $+$ b stages, (G) patients with no relapse, and (H) patients before stage IV. OS, overall survival; ROC, Receiver Operating Characteristic; AUC, Area under Curve.

Figure 5.

Univariate and multivariate Cox analysis of signatures combined with clinical features and nomogram construction. (A) Univariate Cox regression analysis. (B) Multivariate Cox regression analysis. (C) The AUC of the ROC. (D, E) Nomogram construction and the degree of fitness from calibration analysis. ROC, Receiver Operating Characteristic; AUC, Area under Curve.

Figure 6.

Molecular subtype identification. (A–C) Molecular subtypes based on 10 DEGs: Elbow and gap plots for different numbers of clusters; Consensus heatmap of clusters. (D) Survival analysis of the five clusters. DEGs, differentially expressed genes.

Figure 7.

Clinical relevance analysis. EIF2B1, FSD1L, FSTL3, ORC3, HMMR, PRELP, and SETD6 expression (A–G), PIGW, HSD17B6, and GNG7 expression (H–J) between high-risk and low-risk patient groups. The relevance between risk and stage T is shown (K–L).

Figure 8.

Protein levels of cell cycle-related DEGs in LSCC tumor and normal tissue from the HPA database.

3.2 GO and KEGG enrichment analysis

From GO and KEGG enrichment analysis on the 372 cell cycle-related DEGs, we identified several functional enrichment pathways (Fig. 1D and E). GO analyses showed the largest number of terms were involved in organelle fission, mitotic nuclear division, regulation of lymphocyte proliferation, and the regulation of mononuclear and nuclear division. Genes enriched for each term are shown (Fig. 1C). KEGG results showed that most enrichment pathways were involved in cell adhesion molecules and the phagosome. The most common GO function terms were in division and proliferation which were closely related to the cell cycle and reflected our hypothesis.

3.3 Constructing and validating a prognostic genetic risk score model

According to a random standard, we generated an internal training group and internal testing group from the discovery set. LASSO regression was used for cross-validation, after which we selected 27 candidate genes with the log ( $\lambda$ ) with the minimal deviation and the coefficient other than 0 (Fig. 2A and B). These genes were: CARD16, ABCC6P1, CALHM5, CLNS1A, COL5A3, DCSTAMP, DDX51, EIF2B1, FSD1L, FSTL3, GNG7, HMMR, HSD17B6, MKS1, NIP7, OCRL, ORC3, PIGW, PRELP, RORC, RPAP1, SETD6, SSRP1, TIMP3, TTC27, TTC4, and TUBB3. Finally, 10 signature genes were identified by multivariate Cox regression and a prognostic model established using the following formula: risk score $=$ (expression of EIF2B1*0.72) $+$ (expression of FSD1L*1.593) $+$ (expression of FSTL3*1.772) $+$ (expression of GNG7* $-$ 1.286) $+$ (expression of HMMR*0.615) $+$ (expression of HSD17B6* $-$ 0.8) $+$ (expression of ORC3*0.387) $+$ (expression of PIGW* $-$ 1.058) $+$ (expression of PRELP*0.923) $+$ (expression of SETD6*0.545). Finally, the risk score of each patient was calculated and sorted by low to high; the median was 1.065 and was assigned as the cut-off value. All patients were divided into high-risk and low-risk groups based on this value (Fig. 2E). From survival analysis, the OS of patients in the high-risk group was obviously lower than in the low-risk group (Fig. 2C). Additionally, the AUC of the ROC curve for OS was 0.813, 0.816, and 0.816 at 1, 3, and 5 years, respectively (Fig. 2D).

The same design formula was used for the internal testing group (Fig. 3A), discovery set (Fig. 3B), external validation set (Fig. 3C), and all groups (Fig. 3D). Datasets showed that patients in the high-risk group had significantly lower OS than those in the low-risk group, consistent with the internal training group. Also, the AUC of the ROC curve for OS in different datasets verified that the signature was robust. The AUC of the internal testing group was 0.660, 0.637 and 0.634 at 1, 3, and 5 years, respectively. The AUC in the discovery set was 0.747, 0.728, and 0.725 at 1, 3, and 5 years, respectively. The AUC in the external validation set was 0.710, 0.664, and 0.672 at 1, 3, and 5 years, respectively, and in all groups, it was 0.649, 0.623, and 0.634 at 1, 3, and 5 years, respectively.

Using different patient clinical classifications, we also performed survival analysis. When classified for age and gender, OS results were the same as our model (Fig. 4A–D). In particular for T stages, whether in T1 or T2, the OS of patients in the high-risk group was lower than in the low-risk group (Fig. 4E and F). Additionally, patients who did not relapse and patients before stage IV in OS also met our expectations (Fig. 4G and H). Values above the AUC in ROC curves for OS in different clinical classifications were desirable.

3.4 Risk scores are strong independent prognostic factors for early-stage LSCC

To further assess the prognostic value of risk scores, we collected different clinical features from early-stage LSCC patients for univariate and multivariate Cox regression analysis, including age, gender, and T stage (Fig. 5A and B). The AUC of the ROC curve for the risk score was 0.907 which was the best comparing with other clinical features (Fig. 5C). These results showed that the risk score was a strong independent prognostic factor for each patient. Additionally, a nomogram was executed by combining these clinical features and risk values (Fig. 5D). Calibration analysis showed the degree of fitness by comparing with facts (Fig. 5E), thus the nomogram was efficient and accurate.

3.5 Identifying molecular subtypes and clinical relevance analysis

To identify novel tumor molecular subtypes [22], we used consensus clustering to generate five potential clusters due to their stability when the $k$ value $=$ 5 (Fig. 6A–C). The survival analysis curve of the five clusters suggested a potential clinical contributions when the classification method was performed (Fig. 6D). Moreover, we assessed the expression of 10 signature genes between patients in high-risk and low-risk groups and showed that EIF2B1, FSD1L, FSTL3, ORC3, HMMR, PRELP, and SETD6 expression in the high-risk group was higher than in the low-risk group (Fig. 7A–G), while PIGW, HSD17B6, and GNG7 expression in the high-risk group was lower than in the low-risk group (Fig. 7H–J). The relevance between risk and T stage is shown (Fig. 7K and L). Our analyses also suggested that these signature genes could function as early-stage LSCC biomarkers and may have important roles in clinical therapy with respect to different clinical features.

3.6 Verifying prognostic genes

Validation results from the HPA database suggested that EIF2B1, FSD1L, FSTL3, and ORC3 expression in tumor tissue was higher than in normal tissue (Fig. 8A–D), and that HMMR, SETD6, HSD17B6, and GNG7 expression was not detected in both tumor and normal tissue (Fig. 8E–H). PRELP and PIGW expression was not detected in the HPA database, in either tumor or normal tissue. In terms of validation, our analyses were reliable.

4. Discussion

Globally, LSCC represents approximately one third of all NSCLC cases [23]. In terms of LUAD, therapy options are varied and include surgery, targeted therapy, and immunotherapy [24, 25]. However, very limited personalized therapies are performed for advanced LSCC [26], thus their survival rates are poor. However, patients with early-stage LSCC (I–IIIA) are treated by surgery, often in combination with neoadjuvant treatment before the operation, therefore their OS is higher than for advanced disease [27, 28]. Consequently, the construction of a model for patients in early-stage LSCC is highly significant for predicting disease prognoses.

The cell cycle is key process for cell proliferation and division, and consists of five consecutive phases: G0, G1, S, G2, and M [29]. Most carcinoma cells break down in phase G1 and G2, due to more mutations occurring on G1 checkpoint easier and DNA damage and degradation occurring on G2 checkpoint mainly via the ATM/CHK2/p53 pathway [30]. Using gene data from high-throughput techniques, we collected cell cycle-related gene information and used it to systematically investigate associations between tumors and genes. Critically, more studies are now constructing disease signatures using such methods [31, 32].

In our study, we screened 372 genes from 467 cell cycle-related genes and performed LASSO regression and multivariate Cox regression analysis. From this, 10 signature DEGs were identified. Based on this approach, we established a predictive model; the AUC was 0.813, 0.816, and 0.816 at 1, 3, and 5 years in the internal training group, which showed good robustness. Additionally, similar data were observed for the internal testing group, discovery set, or external validation set. In a short, on the premise of rare models based on early-stage LSCC, the model is reliable to assess the prognosis.

From these gene signatures, FSTL3 is a member of the FSTL family and regulates testicular aging and size [33] and also glucose and fat homeostasis [34]. In breast and hepatocellular carcinoma, FSTL3 was reported as a potential biomarker due to its abnormal expression [35, 36]. Recently, FSTL3 overexpression was shown to promote NSCLC proliferation and metastasis, which was regulated by the long non-coding RNA DSCAM-AS1/miR-122-5p axis [37]. EIF2B1 is a guanine nucleotide exchange factor for its GTP-binding protein partner EIF2, and is involved in protein biosynthesis [38]. During cell cycle S phase, OBI1 catalyzes the multi-mono-ubiquitylation of a subset of chromatin-bound ORC3 and ORC5 proteins [39]. As an intracellular, microtubule-associated, spindle assembly factor, HMMR increases the activity of mitotic kinases [40]. HMMR was also reported to inhibit LUAD cell proliferation, migration, invasion, and promote cell apoptosis by down-regulating HMMR [41]. From laboratory studies, SETD6-deficient cells progressed faster through the different mitotic steps toward cytokinesis stages [42]. GNG7 induces autophagy and cell death by inhibiting the mTOR pathway, which is also related to actin cytoskeleton and cell division [43].

However, our study had some limitations. Model construction was based on the internal training group, and further validation in multi-institution seemed to be a great choice. Additionally, functional verification studies are required to assess the relevance of these cell cycle DEGs in lung cancer etiology.

5. Conclusions

Based on cell cycle-related genes, we constructed a prognostic model for early-stage LSCC patients which could significantly guide clinical decision-making and individualized therapy. Also, these specific cell cycle-related genes could be used as new targets to treat LSCC.

Data availability statement

The datasets generated for this study can be found in the online repositories. The names of the repository/repositories and accession number(s) can be found in the article.

Ethical approval and consent to participate

This study has been approved by the Ethics Committee of the First Affiliated Hospital of Soochow University and was performed in accordance with the provisions of the Ethics Committee of Soochow University. Written informed consent was obtained from the patient for publication of this manuscript.

Author contributions

Conception: FY and JW.

Interpretation or analysis of data: LYK.

Preparation of the manuscript: ZCP, HY and FC.

Revision for important intellectual content: JD, LJX and MHT.

Supervision: FY and JW.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

This work was supported by the Jiangsu Provincial Research Foundation for Basic Research (Grant no. BK20191174), Suzhou Medical and Health Science and Technology Innovation Project (Grant no. SKY2022142), and Suzhou Science, Education and Health Youth Project (Grant no. KJXW2022006).

Supplementary data

The supplementary files are available to download from https://dx-doi-org.web.bisu.edu.cn/10.3233/CBM-220227.

sj-zip-1-cbm-10.3233_CBM-220227.zip - Supplemental material

Supplemental material, sj-zip-1-cbm-10.3233_CBM-220227.zip

References

Hirsch

F.R.

Scagliotti

G.V.

Mulshine

J.L.

Kwon

Curran

W.J.

, Jr. Wu

Y.L.

and Paz-Ares

, Lung cancer: Current therapies and new targeted treatments, Lancet 389 (2017), 299–311.

Siegel

R.L.

Miller

K.D.

and Jemal

, Cancer statistics, 2020, CA Cancer J Clin 70 (2020), 7–30.

Travis

W.D.

Brambilla

Noguchi

Nicholson

A.G.

Geisinger

K.R.

Yatabe

Beer

D.G.

Powell

C.A.

Riely

G.J.

Van Schil

P.E.

Garg

Austin

J.H.

Asamura

Rusch

V.W.

Hirsch

F.R.

Scagliotti

Mitsudomi

Huber

R.M.

Ishikawa

Jett

Sanchez-Cespedes

Sculier

J.P.

Takahashi

Tsuboi

Vansteenkiste

Wistuba

Yang

P.C.

Aberle

Brambilla

Flieder

Franklin

Gazdar

Gould

Hasleton

Henderson

Johnson

Kerr

Kuriyama

Lee

J.S.

Miller

V.A.

Petersen

Roggli

Rosell

Saijo

Thunnissen

Tsao

and Yankelewitz

, International association for the study of lung cancer/american thoracic society/european respiratory society international multidisciplinary classification of lung adenocarcinoma, J Thorac Oncol 6 (2011), 244–285.

Kawase

Yoshida

Ishii

Nakao

Aokage

Hishida

Nishimura

and Nagai

, Differences between squamous cell carcinoma and adenocarcinoma of the lung: Are adenocarcinoma and squamous cell carcinoma prognostically equal? Jpn J Clin Oncol 42 (2012), 189–195.

Rami-Porta

Asamura

Travis

W.D.

and Rusch

V.W.

, Lung cancer – major changes in the American Joint Committee on Cancer eighth edition cancer staging manual, CA Cancer J Clin 67 (2017), 138–155.

Tang

Pan

Wang

Cao

Chu

Dai

Shu

Chen

Jin

and Shen

, Genome-wide association study of survival in early-stage non-small cell lung cancer, Ann Surg Oncol 22 (2015), 630–635.

Marentakis

Karaiskos

Kouloulias

Kelekis

Argentos

Oikonomopoulos

and Loukas

, Lung cancer histology classification from CT images based on radiomics and deep learning models, Med Biol Eng Comput 59 (2021), 215–226.

Chansky

Sculier

J.P.

Crowley

J.J.

Giroux

Van Meerbeeck

and Goldstraw

, The international association for the study of lung cancer staging project: Prognostic factors and pathologic TNM stage in surgically managed non-small cell lung cancer, J Thorac Oncol 4 (2009), 792–801.

Sculier

J.P.

Chansky

Crowley

J.J.

Van Meerbeeck

and Goldstraw

, The impact of additional prognostic factors on survival and their relationship with the anatomical extent of disease expressed by the 6th Edition of the TNM Classification of Malignant Tumors and the proposals for the 7th Edition, J Thorac Oncol 3 (2008), 457–466.

10.

Kawaguchi

Takada

Kubo

Matsumura

Fukai

Tamura

Saito

Maruyama

Kawahara

and Ignatius Ou

S.H.

, Performance status and smoking status are independent favorable prognostic factors for survival in non-small cell lung cancer: A comprehensive analysis of 26,957 patients with NSCLC, J Thorac Oncol 5 (2010), 620–630.

11.

Jia

Cao

Yao

Zhao

and Li

, CUL4 E3 ligase regulates the proliferation and apoptosis of lung squamous cell carcinoma and small cell lung carcinoma, Cancer Biol Med 17 (2020), 357–370.

12.

Zhang

Huang

Peng

Qian

Wang

Zhang

Wang

and Sun

, Upregulated miR-1258 regulates cell cycle and inhibits cell proliferation by directly targeting E2F8 in CRC, Cell Prolif 51 (2018), e12505.

13.

Zhang

Yue

Zhang

Gao

Zhao

Chen

Huo

Pang

Chen

Chuai

Zhang

Giaccone

and Wang

, Genomic characteristics in Chinese non-small cell lung cancer patients and its value in prediction of postoperative prognosis, Transl Lung Cancer Res 9 (2020), 1187–1201.

14.

Varma

, Blind estimation and correction of microarray batch effect, PLoS One 15 (2020), e0231446.

15.

Irizarry

R.A.

Hobbs

Collin

Beazer-Barclay

Y.D.

Antonellis

K.J.

Scherf

and Speed

T.P.

, Exploration, normalization, and summaries of high density oligonucleotide array probe level data, Biostatistics 4 (2003), 249–264.

16.

Liberzon

Subramanian

Pinchback

Thorvaldsdóttir

Tamayo

and Mesirov

J.P.

, Molecular signatures database (MSigDB) 3.0, Bioinformatics 27 (2011), 1739–1740.

17.

Barbie

D.A.

Tamayo

Boehm

J.S.

Kim

S.Y.

Moody

S.E.

Dunn

I.F.

Schinzel

A.C.

Sandy

Meylan

Scholl

Fröhling

Chan

E.M.

Sos

M.L.

Michel

Mermel

Silver

S.J.

Weir

B.A.

Reiling

J.H.

Sheng

Gupta

P.B.

Wadlow

R.C.

Hoersch

Wittner

B.S.

Ramaswamy

Livingston

D.M.

Sabatini

D.M.

Meyerson

Thomas

R.K.

Lander

E.S.

Mesirov

J.P.

Root

D.E.

Gilliland

D.G.

Jacks

and Hahn

W.C.

, Systematic RNA interference reveals that oncogenic KRAS-driven cancers require TBK1, Nature 462 (2009), 108–112.

18.

Ritchie

M.E.

Phipson

Law

C.W.

Shi

and Smyth

G.K.

, limma powers differential expression analyses for RNA-sequencing and microarray studies, Nucleic Acids Res 43 (2015), e47.

19.

Wang

L.G.

Han

and He

Q.Y.

, clusterProfiler: An R package for comparing biological themes among gene clusters, Omics 16 (2012), 284–287.

20.

Jiang

Zhang

Zhao

Deng

Mou

Cai

Zhou

Liu

Chen

and Qi

, ImmunoScore Signature: A Prognostic and Predictive Tool in Gastric Cancer, Ann Surg 267 (2018), 504–513.

21.

Uhlen

Zhang

Lee

Sjöstedt

Fagerberg

Bidkhori

Benfeitas

Arif

Liu

Edfors

Sanli

von Feilitzen

Oksvold

Lundberg

Hober

Nilsson

Mattsson

Schwenk

J.M.

Brunnström

Glimelius

Sjöblom

Edqvist

P.H.

Djureinovic

Micke

Lindskog

Mardinoglu

and Ponten

, A pathology atlas of the human cancer transcriptome, Science 357 (2017).

22.

Lancichinetti

and Fortunato

, Consensus clustering in complex networks, Sci Rep 2 (2012), 336.

23.

Jemal

Siegel

Ward

Hao

and Thun

M.J.

, Cancer statistics, 2009, CA Cancer J Clin 59 (2009), 225–249.

24.

Wang

Yang

Zhang

Han

Zhu

Feng

Chen

Lizaso

Qin

Liu

and He

, Effective Treatment of Lung Adenocarcinoma Harboring EGFR-Activating Mutation, T790M, and cis-C797S Triple Mutations by Brigatinib and Cetuximab Combination Therapy, J Thorac Oncol 15 (2020), 1369–1375.

25.

Schoenfeld

A.J.

Rizvi

Bandlamudi

Sauter

J.L.

Travis

W.D.

Rekhtman

Plodkowski

A.J.

Perez-Johnston

Sawan

Beras

Egger

J.V.

Ladanyi

Arbour

K.C.

Rudin

C.M.

Riely

G.J.

Taylor

B.S.

Donoghue

M.T.A.

and Hellmann

M.D.

, Clinical and molecular correlates of PD-L1 expression in patients with lung adenocarcinomas, Ann Oncol 31 (2020), 599–608.

26.

Derman

B.A.

Mileham

K.F.

Bonomi

P.D.

Batus

and Fidler

M.J.

, Treatment of advanced squamous cell carcinoma of the lung: A review, Transl Lung Cancer Res 4 (2015), 524–532.

27.

Santos

E.S.

and Hart

, Advanced squamous cell carcinoma of the lung: Current treatment approaches and the role of afatinib, Onco Targets Ther 13 (2020), 9305–9321.

28.

Chen

Ballman

K.V.

Watson

M.A.

Govindan

Lanc

Beer

D.G.

Bueno

Chirieac

L.R.

Chui

M.H.

Chen

Franklin

W.A.

Gandara

D.R.

Genova

Brovsky

K.A.

Joshi

M.M.

Merrick

D.T.

Richards

W.G.

Rivard

C.J.

Harpole

D.H.

Tsao

M.S.

van Bokhoven

Shepherd

F.A.

and Hirsch

F.R.

, Correlation of PD-L1 expression with tumor mutation burden and gene signatures for prognosis in early-stage squamous cell lung carcinoma, J Thorac Oncol 14 (2019), 25–36.

29.

Beck

Horikawa

and Harris

, Cellular senescence: Mechanisms, morphology, and mouse models, Vet Pathol 57 (2020), 747–757.

30.

Zhang

Chen

Zhu

and He

, Construction and validation of a cell cycle-related robust prognostic signature in colon cancer, Front Cell Dev Biol 8 (2020), 611222.

31.

Gao

L.N.

and You

C.G.

, Integrated analysis of the functions and prognostic values of RNA binding proteins in lung squamous cell carcinoma, Front Genet 11 (2020), 185.

32.

Liu

Zhao

Leng

Liu

and Zhu

, Prognostic implications of autophagy-associated gene signatures in non-small cell lung cancer, Aging (Albany NY) 11 (2019), 11440–11462.

33.

Oldknow

K.J.

Seebacher

Goswami

Villen

Pitsillides

A.A.

O’Shaughnessy

P.J.

Gygi

S.P.

Schneyer

A.L.

and Mukherjee

, Follistatin-like 3 (FSTL3) mediated silencing of transforming growth factor β (TGFβ) signaling is essential for testicular aging and regulating testis size, Endocrinology 154 (2013), 1310–1320.

34.

Mukherjee

Sidis

Mahan

Raher

M.J.

Xia

Rosen

E.D.

Bloch

K.D.

Thomas

M.K.

and Schneyer

A.L.

, FSTL3 deletion reveals roles for TGF-beta family ligands in glucose and fat homeostasis in adults, Proc Natl Acad Sci U S A 104 (2007), 1348–1353.

35.

Zawadzka

A.M.

Schilling

Cusack

M.P.

Sahu

A.K.

Drake

Fisher

S.J.

Benz

C.C.

and Gibson

B.W.

, Phosphoprotein secretome of tumor cells as a source of candidates for breast cancer biomarkers in plasma, Mol Cell Proteomics 13 (2014), 1034–1049.

36.

Behnke

Reimers

and Fisher

, The expression of embryonic liver development genes in hepatitis C induced cirrhosis and hepatocellular carcinoma, Cancers (Basel) 4 (2012), 945–968.

37.

Gao

Chen

Wang

and Zhang

, Up-Regulation of FSTL3, Regulated by lncRNA DSCAM-AS1/miR-122-5p Axis, Promotes Proliferation and Migration of Non-Small Cell Lung Cancer Cells, Onco Targets Ther 13 (2020), 2725–2738.

38.

Jennings

M.D.

and Pavitt

G.D.

, A new function and complexity for protein translation initiation factor eIF2B, Cell Cycle 13 (2014), 2660–2665.

39.

Coulombe

Nassar

Peiffer

Stanojcic

Sterkers

Delamarre

Bocquet

and Méchali

, The ORC ubiquitin ligase OBI1 promotes DNA replication origin firing, Nat Commun 10 (2019), 2426.

40.

Mei

Connell

and Maxwell

C.A.

, Hyaluronan Mediated Motility Receptor (HMMR) Encodes an Evolutionarily Conserved Homeostasis, Mitosis, and Meiosis Regulator Rather than a Hyaluronan Receptor, Cells 9 (2020).

41.

Pan

Jiang

and Zhao

, HCG18/miR-34a-5p/HMMR axis accelerates the progression of lung adenocarcinoma, Biomed Pharmacother 129 (2020), 110217.

42.

Feldman

Vershinin

Goliand

Elia

and Levy

, The methyltransferase SETD6 regulates Mitotic progression through PLK1 methylation, Proc Natl Acad Sci U S A 116 (2019), 1235–1240.

43.

Liu

Yang

Wang

and Zhang

, G protein γ subunit 7 induces autophagy and inhibits cell division, Oncotarget 7 (2016), 24832–24847.