Sage Journals: Discover world-class research

Abstract

BACKGROUND:

Prognostic biomarkers are promising targets for cancer prevention and treatment.

OBJECTIVE:

We try to filtrate survival-related genes for non-small cell lung cancer (NSCLC) via transcriptome analysis.

METHODS:

Transcriptome data and clinical information of Lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC), mainly subtypes of NSCLC, were obtained from The Cancer Genome Atlas (TCGA) program. Differentially expressed genes (DEGs) analyzed by DESeq2 package were regarded as candidate genes. For survival analysis, univariate and multivariate Cox regression were applied to select biomarkers for overall survival (OS) and progression-free survival (PFS), where univariate analysis was for preliminary filtration and multivariate analysis considering age, gender, TNM parameters and clinical stage was for ultimate determination. Gene ontology (GO) analysis and pathway enrichment were used for biological annotation.

RESULTS:

We ultimately acquired a series of genes closely related to prognosis. For LUAD, we determined 314 OS-related genes and 275 PFS-related genes, while 54 OS-related genes and 78 PFS-related genes were chosen for LUSC. The final biological analysis indicated important function of proliferative signaling in LUAD but for LUSC, only cornification process had statistical meaning.

CONCLUSIONS:

We strictly determined prognostic genes of NSCLC, which would contribute to its carcinogenesis investigation and therapeutic methods improvement.

Keywords

Lung adenocarcinoma lung squamous cell carcinoma transcriptome prognosis biomarker

1. Introduction

Lung cancer has the leading morbidity and mortality among cancers, with about 2.1 million new cases and 1.8 million deaths estimated in 2018 worldwide [1]. The most frequent histopathological type of lung cancer is non-small cell lung cancer (NSCLC), comprising mainly lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) subtypes, on which mounting researchers lay great emphasis [2, 3].

With the comprehension of carcinogenesis advancing, therapeutic methods for NSCLC have achieved breakthroughs. The glamorous success must be assigned to targeted therapies based on tumor driver mutations, which could distinctly prolong survival periods in individual with specific genetic alterations compared to traditional platinum-based regimens [4]. Recently, immune checkpoint blockers have demonstrated powerful efficacy to patients carrying high expression levels of PD-L1(programmed cell death 1 ligand 1) or high tumor mutational burden [5]. However, there are still problems restraining from defeating such nightmare. On the one hand, a large portion of cases lack actionable mutations for targeted medicines, while many patients suffer from “cold tumors”, which are insensitive to current immunotherapies [6, 7]. On the other hand, drug tolerance comes to almost all patients sooner or later, inherent or adaptive resistance immensely weakens longtime benefits from targeted or immune therapeutic approaches [3, 8, 9]. So, molecular mechanisms underlying NSCLC still require deeper investigation.

Cancer-related gene remains as an important aspect for comprehending malignant etiology, while the first-line candidates are genes linked closely with clinicopathological characteristics of tumors. The Cancer Genome Atlas (TCGA) program contains sufficient tumors information including genetic and clinical records, which contribute enormously to investigation on cancers of many kinds [10, 11]. Here, by univariate and multivariate filtration, we finally obtained a series of tumor survival- related genes especially exhibiting independent risk suggestion, which will greatly enlighten pathogenesis investigation into NSCLC.

2. Materials and methods

2.1 Transcriptomic data and clinical records

LUSC and LUAD RNA-sequencing data and corresponding clinicopathologic annotation were derived from TCGA program [10, 11]. For LUAD, differentially expressed genes (DEGs) were found by comparing 58 tumors and 58 normal tissues, and 253 individuals containing transcriptome profiling and complete clinicopathologic information (age, gender, TNM parameters, clinical stage, overall survival status and progression-free survival status) were applied for survival analysis. As to LUSC, 51 tumors and 51 normal tissues were used to search DEGs and 288 individuals with analogous intact information were applied for survival analysis. Clinical characteristics of individuals contained in this research are shown in Table 1.

Table 1
Clinical characteristics of individuals in TCGA-LUAD and TCGA-LUSC project

Variables	NSCLC
	LUAD (253)	LUSC (288)
Age
$>$ 60	165	225
$\leqslant$ 60	88	63
Gender
Male	121	204
Female	132	84
Tumor size
T12	226	240
T34	27	48
Lymph node invasion
No	162	184
Yes	91	104
Hematogenous metastasis
No	240	286
Yes	13	2
Clinical stage
I II	199	245
III IV	54	43
OS
1-year survival rate	89%	87%
3-year survival rate	62%	64%
5-year survival rate	39%	55%
PFS
1-year survival rate	78%	80%
3-year survival rate	44%	54%
5-year survival rate	30%	44%

Figure 1.

NSCLC-DEGs from TCGA RNA-sequencing data. (A) DEGs of LUAD; (B) DEGs of LUSC. DEGs, differentially expressed genes; NSCLC, non-small cell lung cancer; TCGA, The Cancer Genome Atlas; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma.

Figure 2.

Biological signatures of NSCLC-DEGs. (A) gene ontology analysis of LUAD-DEGs; (B) pathway enrichment of LUAD-DEGs; (C) gene ontology analysis of LUSC-DEGs; (D) pathway enrichment of LUSC-DEGs. DEGs, differentially expressed genes; NSCLC, non-small cell lung cancer; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma.

2.2 Enrichment analysis for biological properties

We first used BiomaRt and org.Hs.eg.db packages to map official genetic labels of these DEGs [12, 13]. Next we applied clusterProfiler package to execute gene ontology (GO) analysis involving molecular function, cellular component, and biological process [14]. Subsequently, we adopted clusterProfiler and ReactomePA packages to conduct pathway enrichment analysis based on Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome database [14, 15].

Figure 3.

Selection of OS-related genes of NSCLC. (A) flow chart shows selecting OS-related genes of LUAD; (B) coefficients and $p$ values for the OS-related genes of LUAD; (C) flow chart shows selecting OS-related genes of LUSC; (D) coefficients and $p$ values for the OS-related gene of LUSC. OS, overall survival; NSCLC, non-small cell lung cancer; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma.

2.3 Statistical methods

We used DESeq2 package to acquire DEGs from RNA-sequencing data (Adjusted $P$ -value $<$ 0.01, fold change $>$ 4 or $<$ 0.25) [16]. Z score was used for data normalization. Univariate and multivariate Cox proportional hazards regression were used to select prognosis-related genes. Likelihood ratio test, Wald test, scoring test were applied for statistical testing. $P<$ 0.05 was considered significant. R language offered packages for arithmetic functions [17].

3. Results

3.1 Acquiring DEGs of LUAD and LUSC from RNA-sequencing data

Malignant progression largely relies on gene expression alterations, indicating momentous effect of DEGs upon carcinogenesis. Thus, we first tried to filtrate DEGs of LUAD and LUSC from public TCGA transcriptome data. DESeq2 package was practiced under strict inclusion criteria (fold-change $>$ 4 or fold-change $<$ 0.25, adjusted $p$ value $<$ 0.01). We obtained 1412 up-regulated DEGs and 662 down-regulated DEGs in LUAD, while 2040 up-regulated genes and 1327 down-regulated DEGs in LUSC (tumor tissue versus normal tissue) (Fig. 1A and B).

Figure 4.

Filtration of PFS-related genes of NSCLC. (A) flow diagram shows selecting PFS-related genes of LUAD; (B) coefficients and $p$ values for the PFS-related gene of LUAD; (C) flow diagram shows selecting PFS-related genes of LUSC; (D) coefficients and $p$ values for the PFS-related gene of LUSC. PFS, progression-free survival; NSCLC, non-small cell lung cancer; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma.

Figure 5.

Biological annotation for prognostic genes of NSCLC. (A) Biological annotation for LUAD-related prognostic genes; (B) Biological annotation for LUSC-related prognostic genes. NSCLC, non-small cell lung cancer; LUAD, lung adenocarcinoma; LUSC, lung squamous cell carcinoma.

3.2 Investigating biological signatures of DEGs via enrichment analysis

For further comprehension of these DEGs, we then investigated biological annotation via enrichment analysis. GO analysis and pathway enrichment analysis by clusterProfiler and ReactomePA packages were practiced to DEGs of LUAD and LUSC respectively. For LUAD, biological processes like extracellular matrix organization, collagen formation and degradation, cornification, ligand binding and ligand-receptor interaction topped the list (adjusted $p$ value $<$ 0.01) (Fig. 2A and B)(Supplementary Table 1). As to LUSC, formation of the keratinization and cornified envelope, epidermis and skin development, and cell-cell junction were most statistical significant (adjusted $p$ value $<$ 0.01) (Fig. 2C and D)(Supplementary Table 2).

3.3 Finding genes associated tightly with overall survival

Overall survival (OS) is an undoubtedly prominent indicators for cancer prognosis. We tried to dig out OS-related biomarkers by univariate and multivariate Cox regression based on statistic significance and regression coefficient (coef). For LUAD, 359 genes were first chosen from 1412 up-regulated DEGs ( $p<$ 0.05, coef $>$ 0), then 296 genes were filtrated by multivariate analysis where confounding factors like age, gender, TNM parameters and clinical stage were all considered ( $p<$ 0.05, coef $>$ 0) (Fig. 3A). As to down-regulated DEGs of LUAD, we got 34 genes from 662 down-regulated DEGs by univariate analysis ( $p<$ 0.05, coef $<$ 0), and later 18 genes were obtained by multivariate analysis containing the above concomitant variables ( $p<$ 0.05, coef $<$ 0) (Fig. 3A). We also gave coefficients and $p$ values of these OS-related DEGs of LUAD in Fig. 3B and Supplementary Table 3. Similarly, as to LUSC, we finally obtained 53 up-regulated DEGs and 1 down-regulated DEGs by univariate and multivariate analysis (Fig. 3C). Coefficients and $p$ values of these selected DEGs of LUSC are presented in Fig. 3D and Supplementary Table 4.

3.4 Searching genes linked closely with progression-free survival

Progression-free survival (PFS) also imparts potent indications for cancer prognosis. We then applied the Cox proportional hazard regression to search biomarker of PFS, where univariate analysis was for primary filtration and multivariate analysis including factors like age, gender, TNM parameters and clinical stage was for ultimate selection. For LUAD, 283 up-regulated DEGs were chosen from univariate analysis and then 238 genes were filtrated by multivariate analysis ( $p<$ 0.05, coef $>$ 0) (Fig. 4A). Similarly, we got 64 down-regulated DEGs of LUAD preliminarily, and later 37 genes were identified by multivariate analysis ( $p<$ 0.05, coef $<$ 0) (Fig. 4A). Coefficients and $p$ values of these PFS-related DEGs of LUAD are exhibited in Fig. 4B and Supplementary Table 5. Analogously, we acquired 78 up-regulated DEGs but none down-regulated DEGs of LUSC by univariate and multivariate analysis (Fig. 4C). Coefficients and $p$ values of filtrated DEGs are shown in Fig. 4D and Supplementary Table 6.

3.5 Pathway enrichment analysis of prognostic DEGs

Subsequently, we investigated biological signatures of these prognostic DEGs by clusterProfiler and ReactomePA packages. And we found the prognostic DEGs of LUAD mostly converged on processes involving cell division and proliferation, while the prognostic DEGs of LUSC only had close association with cornification formation (Fig. 5A and B).

4. Discussion

Malignant progression is a dynamic biological process involving function of numerous genes. And rapid development in genomic and bioinformatic technology has provided much convenience to search cancer-related genes. In this research, we used bioinformatic tools to screen out genes closely linked to prognosis of NSCLC, providing much help for further investigation.

Cancer-related genes always exhibit aberrant expression in tumors compared to normal tissues. We first obtained DEGs from TCGA transcriptome data with critical standards set for high precision. Then we investigated biological characteristics of these DEGs, and we found different signatures of LUAD and LUSC. For LUAD, extracellular matrix (ECM) associated processes headed the list, suggesting that tumor microenvironment may contribute much to carcinogenesis of such histological type. And in fact, ECM alterations in the malignant stroma could function in kinds of aspects to facilitate neoplastic progression like promoting angiogenesis, assisting migration and invasion, providing sustainable signal for proliferation and survival, and so on [18, 19]. As to LUSC, most DEGs got enriched in biological sets of cornification and cell junction, which conforms to morphological characteristics of squamous cell carcinoma. Besides, cornification and keratins could not only been applied as diagnostic malignant markers especially about tissue origin, but also mediate diverse malignant properties such as metastasis, treatment responsiveness and the like [20, 21].

OS and PFS both are pivotal indictors for tumor prognosis, so DEGs linked closely with them always imply valuable information for handling cancer. Faced with quantities of DEGs, we conducted preliminary selection by univariate cox regression analysis, where admittance criterion was determined by $p$ value and coef (for the upregulated gene, the potential oncogene with hazardous effect, the entry condition was $p$ value $<$ 0.05 and coef $>$ 0, while for the downregulated gene, the possible tumor suppressor gene with protective potency, the entry condition was $p$ value $<$ 0.05 and coef $<$ 0). In fact, confounding factors could severely hide or exaggerate actual function of genes selected in univariate analysis. So, we practiced optimal filtration by multivariate Cox regression analysis with a view to canonical factors like age, gender, TNM parameters and clinical stage. Based on p value and coef, we finally obtained a group of candidates, which would possibly promote or prohibit malignant progression of LUAD. Notably, we acquired many potential oncogenes but comparatively less tumor suppressor genes, which was in accordance with the current tendency for finding cancer-related genes [6, 22].

Cancer is a highly heterogeneous disorder and different histological type always exhibits some peculiar biological properties while possessing general malignant attributes [22, 23]. Noticeably, biological attributes of cancer-related genes we ultimately chose presented high consistence with known features of NSCLC. For LUAD, these genes were enriched in cell cycle and cell division activities. And as is well known, uncontrolled proliferation has always been a core hallmark of malignancy like LUAD, and many frequent driver mutations responsible for “oncogene addiction” like epidermal growth factor receptor (EGFR), Kirsten rat sarcoma (KRAS), tumor protein p53 (TP53) in LUAD are always involved in proliferative path-ways [24, 25, 26, 27]. As to LUSC, only the cornification process, a typical morphological feature of squamous cell carcinoma, but no signaling pathways proved to be significant. The result again revealed weakness in finding driving mechanisms of LUSC, and indicated molecular underpinnings facilitating LUSC might have big difference with canonical biological pathways contained in current public database [28, 29].

Certainly, our research has some drawbacks needed to be remedied in the future work. First, sample capacity should be enlarged to enhance precision for the high heterogeneity of NSCLC [23, 30, 31]. Second, posttranscriptional and posttranslational modification should be both considered for not just transcriptional regulation contributing to carcinogenesis [32, 33].

In conclusion, by screening out the whole transcriptome and considering typical confounding factors, we obtained a series of genes tightly associated with prognosis of NSCLC. And what we found would function as potential targets for etiology investigation and possibly be used as biomarkers for assisting clinical approaches of NSCLC.

Footnotes

Acknowledgments

We sincerely thanked Prof. Lei Shang and Prof. Lintao Jia for their professional guidance. This study was funded by the National Natural Science Foundation of China (81772462).

Conflict of interest

The authors have no conflicts of interest to declare.

Supplementary data

The supplementary files are available to download from http://dx.doi.org/10.3233/CBM-190222.

References

Bray

Ferlay

Soerjomataram

Siegel

R.L.

Torre

L.A.

and Jemal

, Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries, CA Cancer J Clin 68 (2018), 394–424.

Goldstraw

Ball

Jett

J.R.

Le Chevalier

Lim

Nicholson

A.G.

and Shepherd

F.A.

, Non-small-cell lung cancer, Lancet 378 (2011), 1727–40.

Herbst

R.S.

Morgensztern

and Boshoff

, The biology and management of non-small cell lung cancer, Nature 553 (2018), 446–454.

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.

Prelaj

Tay

Ferrara

Chaput

Besse

and Califano

, Predictive biomarkers of response for immune checkpoint inhibitors in non-small-cell lung cancer, Eur. J. Cancer 106 (2018), 144–159.

Pao

and Girard

, New driver mutations in non-small-cell lung cancer, Lancet Oncol 12 (2011), 175–80.

Cyriac

and Gandhi

, Emerging biomarkers for immune checkpoint inhibition in lung cancer, Semin Cancer Biol 52 (2018), 269–277.

Reck

and Rabe

K.F.

, Precision Diagnosis and Treatment for Advanced Non-Small-Cell Lung Cancer, N Engl J Med 377 (2017), 849–861.

Sharma

Hu-Lieskovan

Wargo

J.A.

and Ribas

, Primary, Adaptive, and Acquired Resistance to Cancer Immunotherapy, Cell 168 (2017), 707–723.

10.

Weinstein

J.N.

Collisson

E.A.

Mills

G.B.

Shaw

K.R.

Ozenberger

B.A.

Ellrott

Shmulevich

Sander

and Stuart

J.M.

, The Cancer Genome Atlas Pan-Cancer analysis project, Nat Genet 45 (2013), 1113–20.

11.

Tomczak

Czerwińska

and Wiznerowicz

, The Cancer Genome Atlas (TCGA): an immeasurable source of knowledge, Contemp Oncol (Pozn) 19 (2015), A68–77.

12.

Durinck

Spellman

P.T.

Birney

and Huber

, Mapping identifiers for the integration of genomic datasets with the R/Bioconductor package biomaRt, Nat Protoc 4 (2009), 1184–91.

13.

Lawrence

Huber

Pages

Aboyoun

Carlson

Gentleman

Morgan

M.T.

and Carey

V.J.

, Software for computing and annotating genomic ranges, PLoS Comput Biol 9 (2013), e1003118.

14.

Wang

L.G.

Han

and He

Q.Y.

, clusterProfiler: an R package for comparing biological themes among gene clusters, Omics 16 (2012), 284–7.

15.

and He

Q.Y.

, ReactomePA: an R/Bioconductor package for reactome pathway analysis and visualization, Mol Biosyst 12 (2016), 477–9.

16.

Love

M.I.

Huber

and Anders

, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2, Genome Biol 15 (2014), 550.

17.

Chan

B.K.C.

, Data Analysis Using R Programming, Adv Exp Med Biol 1082 (2018), 47–122.

18.

Gotte

and Kovalszky

, Extracellular matrix functions in lung cancer, Matrix Biol 73 (2018), 105–121.

19.

Pickup

M.W.

Mouw

J.K.

and Weaver

V.M.

, The extracellular matrix modulates the hallmarks of cancer, EMBO Rep 15 (2014), 1243–53.

20.

Werner

Keller

and Pantel

, Epithelial keratins: Biology and implications as diagnostic markers for liquid biopsies, Mol Aspects Med (2019).

21.

Karantza

, Keratins in health and cancer: more than mere epithelial cell markers, Oncogene 30 (2011), 127–38.

22.

Chen

Fillmore

C.M.

Hammerman

P.S.

Kim

C.F.

and Wong

K.K.

, Non-small-cell lung cancers: a heterogeneous set of diseases, Nat Rev Cancer 14 (2014), 535–46.

23.

Thomas

Liu

S.V.

Subramaniam

D.S.

and Giaccone

, Refining the treatment of NSCLC according to histological and molecular subtypes, Nat Rev Clin Oncol 12 (2015), 511–26.

24.

Ding

Getz

Wheeler

D.A.

Mardis

E.R.

McLellan

M.D.

Cibulskis

Sougnez

Greulich

Muzny

D.M.

Morgan

M.B.

Fulton

R.S.

Zhang

Wendl

M.C.

Lawrence

M.S.

Larson

D.E.

Chen

Dooling

D.J.

Sabo

Hawes

A.C.

Shen

Jhangiani

S.N.

Lewis

L.R.

Hall

Zhu

Mathew

Ren

Yao

Scherer

S.E.

Clerc

Metcalf

G.A.

Milosavljevic

Gonzalez-Garay

M.L.

Osborne

J.R.

Meyer

Shi

Tang

Koboldt

D.C.

Lin

Abbott

Miner

T.L.

Pohl

Fewell

Haipek

Schmidt

Dunford-Shore

B.H.

Kraja

Crosby

S.D.

Sawyer

C.S.

Vickery

Sander

Robinson

Winckler

Baldwin

Chirieac

L.R.

Dutt

Fennell

Hanna

Johnson

B.E.

Onofrio

R.C.

Thomas

R.K.

Tonon

Weir

B.A.

Zhao

Ziaugra

Zody

M.C.

Giordano

Orringer

M.B.

Roth

J.A.

Spitz Wistuba

M.R.

, II. Ozenberger

Good

P.J.

Chang

A.C.

Beer

D.G.

Watson

M.A.

Ladanyi

Broderick

Yoshizawa

Travis

W.D.

Pao

Province

M.A.

Weinstock

G.M.

Varmus

H.E.

Gabriel

S.B.

Lander

E.S.

Gibbs

R.A.

Meyerson

and Wilson

R.K.

, Somatic mutations affect key pathways in lung adenocarcinoma, Nature 455 (2008), 1069–75.

25.

Rom

W.N.

Hay

J.G.

Lee

T.C.

Jiang

and Tchou-Wong

K.M.

, Molecular and genetic aspects of lung cancer, Am J Respir Crit Care Med 161 (2000), 1355–67.

26.

Salgia

and Skarin

A.T.

, Molecular abnormalities in lung cancer, J Clin Oncol 16 (1998), 1207–17.

27.

Comprehensive molecular profiling of lung adenocarcinoma, Nature 511 (2014), 543–50.

28.

Oliver

T.G.

Patel

and Akerley

, Squamous non-small cell lung cancer as a distinct clinical entity, Am J Clin Oncol 38 (2015), 220–6.

29.

Comprehensive genomic characterization of squamous cell lung cancers, Nature 489 (2012), 519–25.

30.

Zhang

Wang

D.C.

Shi

Zhu

Min

and Jin

, Genome analyses identify the genetic modification of lung cancer subtypes, Semin Cancer Biol 42 (2017), 20–30.

31.

Marusyk

and Polyak

, Tumor heterogeneity: Causes and consequences, Biochim Biophys Acta 1805 (2010), 105–17.

32.

Krueger

K.E.

and Srivastava

, Posttranslational protein modifications: current implications for cancer detection, prevention, and therapeutics, Mol Cell Proteomics 5 (2006), 1799–810.

33.

Audic

and Hartley

R.S.

, Post-transcriptional regulation in cancer, Biol Cell 96 (2004), 479–98.

Identifying prognostic biomarkers of non-small cell lung cancer by transcriptome analysis

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSIONS:

Keywords

1. Introduction

2. Materials and methods

2.1 Transcriptomic data and clinical records

Table 1 Clinical characteristics of individuals in TCGA-LUAD and TCGA-LUSC project

3. Results

3.1 Acquiring DEGs of LUAD and LUSC from RNA-sequencing data

3.3 Finding genes associated tightly with overall survival

3.4 Searching genes linked closely with progression-free survival

3.5 Pathway enrichment analysis of prognostic DEGs

4. Discussion

Footnotes

Acknowledgments

Conflict of interest

Supplementary data

References

Table 1
Clinical characteristics of individuals in TCGA-LUAD and TCGA-LUSC project