Study of the co-expression gene modules of non-small cell lung cancer metastases

Abstract

BACKGROUND:

Metastasis regularly is a marker of the disease development of cancers. Some metastatic sites significantly showed more serious clinical outcomes in non-small cell lung cancer (NSCLC). Whether they are caused by tissue-specific (TS) or non-tissue-specific (NTS) mechanisms is still unclear.

OBJECTIVE:

Explore co-expression gene modules of non-small cell lung cancer metastases.

METHODS:

Weighted Correlation Network Analysis (WGCNA) was used to identify the gene modules among the metastases of NSCLC. The clinical significance of those gene modules was evaluated with the Cox hazard proportional model with another independent dataset. Functions of each gene module were analyzed with gene ontology. Typical genes were further studied.

RESULTS:

There were two TS gene modules and two NTS gene modules identified. One TS gene module (green module) and one NTS gene module (purple module) significantly correlated with survival. This NTS gene module (purple module) was significantly enriched in the epithelial-to-mesenchymal transition (EMT) process. Higher expression of the typical genes (CA14, SOX10, TWIST1, and ALX1) from EMT process was significantly associated with a worse survival.

CONCLUSION:

The lethality of NSCLC metastases was caused by TS gene modules and NTS gene modules, among which the EMT-related gene module was critical for a worse clinical outcome.

Keywords

Non-small cell lung cancer metastases gene module weighted correlation network analysis co-expression

Abbreviations

NSCLC:	non-small cell lung cancer;
TS:	tissue-specific;
NTS:	non-tissue-specific;
WGCNA:	Weighted Correlation Network Analysis;
EMT:	epithelial-to-mesenchymal transition;

GEO:	Gene Expression Omnibus;
RMA:	Robust Multi-array Average;
CCL4:	C–C motif ligand 4;
OSCC:	oral squamous cell carcinoma.

Figure 1.

The workflow of this study.

1. Introduction

Lung cancer is the most deadly cause of cancer-related deaths worldwide. Among them, non-small cell lung cancer (NSCLCs) accounts for 80%. Along with disease development, metastasis of lung cancer is one of the most significant markers [1, 2]. It is reasonable to hypothesize that there could be some shared mechanism which is responsible for the lethality from the metastases of NSCLC. It has been reported that multiple mechanisms are involved in the process of angiogenesis, hypoxia, circulation, and establishment of a metastatic focus [3].

Figure 2.

WGCNA analysis of the metastasis tissues. The upper panel is the hierarchical clustering tree. The lower panel are two heatmaps of the gene module membership from “dynamic tree cut” and “merged” analysis in WGCNA.

Figure 3.

The relationship between modules and tissues. The left color labels on the left indicate the gene modules. The lower labels indicate the tissues of metastatic sites (ly: lymphoid; ad: adrenal; sk: skin; in: intestine; om: omentum). In the heatmap, red denotes a positive correlation and blue denotes a negative correlation between the eigengene and the tissue type vector. The two numbers in each heatmap blocks are the Pearson’s correlation coefficient and the $p$ -value (in the parenthesis), respectively.

However, tumors are highly heterogeneous. It has long been realized that patients with the same metastatic site showed different survival time [4, 5]. Recently, the metastatic site was recognized as an important factor that affects clinical outcome [6, 7]. Riihimaki et al. [6] have found that liver and bone metastases caused a worse survival time, and the respiratory and nervous system metastases showed a better survival time. Gibson et al. [8] have found that liver metastasis had worse survival and adrenal metastasis had the best survival. Tamura et al. [9] have found that liver and adrenal gland metastasis had bad survival and brain and bone metastases have better survival. Those studies conflicted with each other for some metastatic sites. Commonly, liver metastasis had bad survival.

As for now, the understanding of the mechanism of the lethality difference within the metastases of lung cancer is still unclear. Only a limited number of studies in genomic variation analysis, especially in some driver genes, were reported. For examples, EGFR mutation could be an important marker for clinical outcome [10, 11]. The median survival time for the EGFR ${}^{+}$ group and EGFR ${}^{-}$ was 22.4 months and 7.9 months, respectively [10]. However, in the EGFR ${}^{+}$ cohort, the brain showed the highest risk ratio, which was contrary to a previous study [6]. Besides, there were only 22.3% of the patients tested as EGFR ${}^{+}$ [10], indicating that there must be other mechanisms involved.

In spite of the success in studying such driver genes, there are many unresolved questions. For examples, which mechanism is most important for the survival “Is there shared mechanism among these metastases” Is there a tissue-specific mechanism among these metastases?” In order to solve these problems, a global transcriptome study would be necessary.

In this study, the Weighted Correlation Network Analysis (WGCNA) was used to identify the gene modules in NSCLC. The tissue-specificity and function of those gene modules were studied. Further using an independant dataset, the clinical outcomes of these modules were evaluated. Specifically, some typical genes were carefully examined.

2. Result

2.1 Prepare data for WGCNA analysis

By searching in the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/) with keywords of “lung cancer AND metastasis” and further checking manually, only one dataset, GSE12630, which contains 12 metastatic samples of lung cancer, was screened (Suppl. Table 1) and then the CEL files for those samples were downloaded. The expression was called with the affy package [12] from Bioconductor on R platform and normalized with the Robust Multi-array Average (RMA) method [13]. Low variable probes were removed according to the interquartile range ( $<$ 0.5). In total, 3533 probes were kept for further analysis.

2.2 WGCNA analysis for the gene modules

In order to get the co-expression gene modules, the WGCNA analysis was conducted with the 12 samples. As denoted by the analysis of the scale-free topology model fit, a soft power threshold at 37 showed a better performance. Unsigned Pearson’s correlation coefficient between genes was used to generate the adjacency matrix. The topological overlap was calculated to construct the co-expression network. Each gene was then hierarchically clustered according to the topological distance in the co-expression network (Fig. 2). The hierarchical tree was dynamically cut. Nine gene modules were inferred. The gene modules with high eigengene similarity ( $>$ 0.75) were merged. Finally, they were merged into four gene modules. For convenience, they were named as black, green, yellow and purple modules, which contained 157, 749, 131, and 47 genes respectively.

Figure 4.

Biological processes enriched in purple module. The $x$ -axis is the “ $-\log 10$ ”-transformed false discovery rate (FDR). The $y$ -axis is the biological processes of gene ontology. The length of the horizontal blue bar indicates the transformed FDR for each biological processes.

2.3 Tissue specificity and functions of the gene modules

To characterize the gene modules, the eigengene was used as a surrogate measure for each gene module. A Pearson’s correlation coefficient between those eigengenes for each gene modules and the tissue type vector was calculated. As shown in (Fig. 3), the green module was highly expressed in the lymph node ( $r=$ 0.67, $p$ -value $=$ 0.02) and expressed lower in the adrenal gland ( $r=-$ 0.52, $p$ -value $=$ 0.08); the yellow module was highly expressed in the intestine tissue ( $r=$ 0.87, $p$ -value $=$ 0.0001). The purple and black gene modules showed no significant correlation with any tissue types. Those results showed that there were tissue-specific and non-tissue-specific gene modules in NSCLC metastases.

Table 1
Study of module eigengenes by survival analysis

	Coef	Exp (coef)	Se (coef)	$z$	$P(>\|z\|)$
MEgreen	0.7889	2.2010	0.3754	2.102	0.03559
MEpurple	$-$ 1.0137	0.3629	0.3750	$-$ 2.703	0.00686
MEblack	$-$ 0.5291	0.5892	0.2973	$-$ 1.779	0.07516
MEyellow	0.1942	1.2144	0.2989	0.650	0.51579

The study was conducted in a dataset consisting of 133 samples. The log-likelihood ratio test showed a significant correlation between the gene modules and the survival time. The green and purple gene modules were the two significant gene modules with hazard ratios, 0.45 and 2.76, respectively.

Next, the function of the four gene modules in the biological process was examined with gene ontology. The yellow module was enriched with “extracellular matrix organization” and “cell adhesion” (Suppl. Fig. 1A); the black module was enriched with “immune response”, “defense response”, “innate immune response” and “regulation of T cell activation” (Suppl. Fig. 1B); the green module was enriched with “hormone-mediated signaling pathway”, “cell communication” and “cellular response to endogenous stimulus” (Suppl. Fig. 1C); and the purple module was enriched with “neural crest cell migration”, “mesenchyme development” and “stem cell differentiation” (Fig. 4).

2.4 EMT process was critical for lethality caused by metastasis

Except for the biological function of gene modules identified by WGCNA analysis, we also studied their clinical association. Without loss of generality, another dataset, GSE14814, was used to validate the gene modules. That dataset had recorded survival/censor time, which was used as a dependent variable in a Cox hazard proportional model (Table 1). The eigengene expressions of those gene modules was dichotomized by the median. They were used as independent variables in the model. As denoted by the log-likelihood ratio test, the model was significant with $p$ -value $=$ 0.04. Specifically, the green and purple modules were significantly correlated with survival. The $p$ -values were 0.0069 and 0.036, respectively. The purple module had a hazard ratio equal to 2.20 and a 95% confient interval within [1.06–4.59]. The purple module had a hazard ratio equal to 0.36 and a 95% confient interval within [0.17–0.76]. Higher expression of the purple module suggested a lower risk of lethality.

2.5 Clinical significance of the typical genes from the purple module

Many studies [14, 15, 16] had shown that EMT was important for the metastasis process, but there were no studies about the clinical association with metastasis. In order to study the clinical significance of the EMT process, the hub genes and EMT related genes from the purple module were further validated in a wider dataset provided by the web service [17] (Fig. 5). The hub gene was defined as the gene which had the highest correlation with the eigengene of the gene modules. CA14 was the hub gene of the purple module. SOX10, ALX1, and TWIST1 were the genes involved in the EMT in the purple module. Using the median as the cutoff, their Kaplan-Meier curves were plotted as in Fig. 5. Higher expression of these genes showed a significantly shorter survival time with the $p$ -values of 0.0023, 0.037, 3.7e-5 and 3.7e-3, respectively. The results further proved the lethality of the purple module and the EMT process for the lung cancer metastasis.

Figure 5.

Validation of typical genes. CA14 is the hub gene of the purple module. SOX10, ALX1, and TWIST1 are three genes enriched in the EMT process. The four plots demonstrate the clinical significance of the four genes by Kaplan-Meier survival analysis. The Cox hazard proportional model and the log-rank tests gave the hazard ratio and $p$ -value for each gene.

2.6 Discussion and conclusion

Lung metastasis could severely affect the clinical outcome [6]. Different metastatic sites may be correlated to significantly different survival [6, 8, 9, 10]. Those studies have found both common and different results. To resolve those discrepancies, it is urgent to determine the background molecular mechanisms. Compared to the genomic variations, the expression profile may reflect the biological process in those metastasis tissues more directly when considering the difficulties in the evaluation of the effects of genomic variations. Due to the difficult availability of metastatic tissues, there have been only a few studies on the expression profiles of lung cancer metastasis up to now. We search for relevant data in the GEO database (https://www.ncbi.nlm.nih.gov/geo/). GSE42076 explored lung cancer metastasis at the cellular level. GSE83950 and GSE124619 studied the relationship between non-coding RNA and lung cancer metastasis. GSE57492 analyzed metastasis-related suppressor genes in the formation of lung cancer brain metastases.

After searching for the GEO database, only one dataset containing multiple metastatic tissues has been collected. That dataset only contains metastatic sites of the adrenal gland, lymphoid, skin and omentum tissues. There are no expression profiles from the liver, which is the most lethal metastasis site [6, 8, 9]. Limited by the data availability, it is hard to know whether the liver has a tissue-specific module or non-tissue-specific module. Brain metastasis is one of the most common metastatic sites of advanced lung cancer, and it is also one of the common metastatic sites leading to the death of lung cancer patients [18, 19, 20]. Although there is a study which contains only a single metastatic site of brain tissue [21], to avoid possible error from multiple experiments, it was not included in this study. Though the overall incidence of bone metastasis is unknown, bone metastasis is a frequent complication in patients with advanced cancer [22, 23]. However, there is also a lack of relevant data in this GEO data. We look forward to more efforts in this field of research.

In order to infer the gene modules that are tissue-specific and non-tissue-specific, unsupervised clustering is needed. In fact, in considering the high heterogeneity of lung cancer metastasis, Student’s t-test in combination with supervised clustering could always be futile. WGCNA has the advantage to cluster all genes into separate gene modules without the knowledge of the tissue types. In this study, we have successfully identified four gene modules. The function of those gene modules was studied with the tissue-specificity and the gene ontology of biological process. As we have hypothesized, two gene modules (green and yellow) are correlated with the tissue types and two other gene modules (black and purple) were not correlated with any tissue types.

The green module was correlated with lymphoid metastasis tissue. It was enriched with “cell communication”. The typical gene was Wnt10B, which was reported to be associated with lymph node metastasis of gastric cancer [24]. However, there is also a report in endometrial cancer that high expression of Wnt10B is prone to not have lymphoid metastasis [25]. Those results suggest that the expression of Wnt10B is critical for lymphoid metastasis. The green gene module could be tissue-specific for lymphoid metastasis of pan-cancer.

Similarly, the yellow module was correlated with intestine metastasis. It was enriched with “extracellular matrix organization” and “cell adhesion”. The typical genes were MMP2, COL1A1 $,$ and ECM2. Currently, there are rarely studies in the intestine metastasis of lung cancer, but they have appeared in the metastasis of other cancers, such as, breast cancer [26], ovarian cancer [27] and lung cancer [28].

The black gene module was non-tissue-specific and enriched with“immune response”, “defense response”, “innate immune response” and “regulation of T cell activation”. The black module reflects the function of immune microenvironment. It is a general mechanism in all types of cancers. For example, the inflammatory chemokine C–C motif ligand 4 (CCL4) is an “innate immune response”-related gene. CCL4 can promote the lymphangiogenesis in oral squamous cell carcinoma (OSCC), which is a critical step in tumor metastasis [29]. In breast cancer, CCL4 can contribute to the metastasis to the bone by binding to CC chemokine receptor type 5 [30]. The genes in the “regulation of T cell activation” category, HLA-DR or CD4, also play a critical role in the metastasis in the metastasis of melanoma [31], breast cancer [32], and prostate cancer [33].

The most interesting module was the purple module, which was also a non-tissue-specific gene module and had a significant effect on the survival. The purple module was enriched with“neural crest cell migration”, “mesenchyme development” and “stem cell differentiation”. Generally, for cancer metastasis, the nervous system plays a big part of metastatic initiation and clonal expansion, angiogenesis to extravasation and colonization [34]. The typical genes from it are SOX10, TWIST1 $,$ and ALX1. Those genes also take part in the EMT process. EMT is a process whereby epithelial cells are transformed into mesenchymal cells [35]. After the EMT process, the transited cells gain migratory/invasive properties [36]. This study proves that EMT could be the most important signature for the severity of lung cancer metastasis.

In summary, in this article, we have studied the gene co-expression modules in the metastatic tissues of lung cancer. Four gene modules were identified, from which two gene modules were tissue-specific and the other two gene modules were non-tissue-specific. Specifically, one of the non-tissue-specific gene modules was significantly correlated with survival. The function of that gene module demonstrates that the EMT process plays a big role in the metastasis process of lung cancer and survival.

3. Method

3.1 Data collections

Two datasets, GSE12630 [17] and GSE14814 [37], were downloaded from the Gene Expression Omnibus (GEO) website (http://www.ncbi.nlm.nih.gov/geo/). For better comparison, only the sample data generated on the Affymetrix U133A microarray platform were chosen as candidates. The samples for primary or metastatic lung cancer were manually curated from dataset GSE12630. The raw CEL files from both datasets were processed by affy package on R platform. The expression was further normalized with the Robust Multi-array Average (RMA) method [13]. The suboptimal probes with the name suffix “s_at” and “x_at” were removed from the analysis. To better fit the normal distribution, the expression was log2-transformed. The annotation package for U133A was downloaded from Bioconductor (http://bioconductor.org). For dataset GSE12630, the low variant probes were filtered according to the interquartile range ( $<$ 0.5).

3.2 WGCNA analysis

WGCNA analysis was conducted on the R platform [38]. Briefly, the scale-free topology model fit was used to choose the soft power. An unsigned correlation distance defined as “ $(1-r)/2$ ” was calculated, where $r$ is the Pearson’s correlation coefficient. The adjacency matrix was generated according to the unsigned correlation distance matrix with the soft power. The adjacency correlation was then used to construct the topological network. The genes were hierarchically clustered with the topological distance. Then, the tree was dynamically cut into multiple gene modules with the parameter “minClusterSize $=$ 30, deepSplit $=$ 2, pamRespectsDendro $=$ FALSE”. Those gene modules with high eigengene similarity ( $>$ 0.75) were merged.

3.3 Gene function analysis

The biological processes of gene ontology were used to enrich the gene modules with the DAVID web service (https://david.ncifcrf.gov/) [39]. The bar plot is generated on the R platform.

3.4 Survival analysis

The WGCNA modules were selected by survival analysis of their eigengenes. The eigengenes were dichotomized by the median. The expression over the median will be defined as high expression; the expression below the median will be defined as low expression. The gene modules were regressed against the survival time with Cox proportional hazards model with R package “survival” [40]. The log-likelihood test was used to test the significance. The modules with higher survival probability were further enriched with gene ontology. The typical genes were validated in a large dataset with the web service [17]. The median value was used as a cutoff to categorize the gene expression into high and low. The survival time was plotted against survival probability with the Kaplan-Meier curve. The log-likelihood ratio test was used to compare their different significance.

Supplementary data

The supplementary files are available to download from http://dx.doi.org/10.3233/CBM-201605.

sj-docx-1-cbm-10.3233_CBM-201605.docx - Supplemental material

Supplemental material, sj-docx-1-cbm-10.3233_CBM-201605.docx

Funding

This work was supported by the National Natural Science Foundation of China, grant nos. 81602009 (funder: Guanghui Wang) and 81672288 (funder: Jiajun Du) and The Joint Research Funds for Shandong University and Karolinska Institute, grant no. SDU-KI-2019-16 (funder: Jiajun Du).

Footnotes

Conflict of interest

Conflict of interest: Guanghui Wang, Fenglong Bie, Guangxu Li, Junping Shi, Yanwu Zeng and Jiajun Du declared no conflicts of interest in this work.

References

Taylor

Bekele

B.N.

Debnam

J.M.

Allen

P.K.

Suki

Sawaya

Komaki

Stewart

D.J.

and Karp

D.D.

, Number of metastatic sites is a strong predictor of survival in patients with nonsmall cell lung cancer with or without brain metastases, Cancer 115(13) (2009), 2930–2938.

Chadha

A.S.

Ganti

A.K.

Sohi

J.S.

Sahmoun

A.E.

and Mehdi

S.A.

, Survival in untreated early stage non-small cell lung cancer, Anticancer Res 25(5) (2005), 3517–3520.

Popper

H.H.

, Progression and metastasis of lung cancer, Cancer Metastasis Rev 35(1) (2016), 75–91.

Pop

Nadeemy

A.S.

Venissac

Guiraudet

Otto

Poudenx

and Mouroux

, Skeletal muscle metastasis from non-small cell lung cancer, Journal of Thoracic Oncology: Official Publication of the International Association for the Study of Lung Cancer 4(10) (2009), 1236–1241.

Kochhar

Frytak

and Shaw

E.G.

, Survival of patients with extensive small-cell lung cancer who have only brain metastases at initial diagnosis, American Journal of Clinical Oncology 20(2) (1997), 125–127.

Riihimaki

Hemminki

Fallah

Thomsen

Sundquist

and Hemminki

, Metastatic sites and survival in lung cancer, Lung Cancer 86(1) (2014), 78–84.

Bauml

Mick

Zhang

Watt

C.D.

Vachani

Aggarwal

Evans

and Langer

, Determinants of survival in advanced non-small-cell lung cancer in the era of targeted therapies, Clinical Lung Cancer 14(5) (2013), 581–591.

Gibson

A.J.W.

D’Silva

Tudor

R.A.

Elegbede

A.A.

Otsuka

S.M.

Bebb

D.G.

and Cheung

W.Y.

, Impact of number versus location of metastases on survival in stage IV M1b non-small cell lung cancer, Med Oncol 35(9) (2018), 117.

Tamura

Kurishima

Nakazawa

Kagohashi

Ishikawa

Satoh

and Hizawa

, Specific organ metastases and survival in metastatic non-small-cell lung cancer, Mol Clin Oncol 3(1) (2015), 217–221.

10.

Hsu

De Caluwe

Anderson

Nichol

Toriumi

and Ho

, Patterns of spread and prognostic implications of lung cancer metastasis in an era of driver mutations, Current oncology (Toronto, Ont) 24(4) (2017), 228–233.

11.

W.Y.

Zhao

T.T.

H.M.

Wang

Z.N.

Y.Y.

Han

Song

Y.X.

J.H.

Yin

S.C.

Liu

X.Y.

and Miao

Z.F.

, The role of EGFR mutation as a prognostic factor in survival after diagnosis of brain metastasis in non-small cell lung cancer: A systematic review and meta-analysis, BMC Cancer 19(1) (2019), 145.

12.

Gautier

Cope

Bolstad

B.M.

and Irizarry

R.A.

, Affy-analysis of affymetrix GeneChip data at the probe level, Bioinformatics 20(3) (2004), 307–315.

13.

Bolstad

B.M.

Irizarry

R.A.

Astrand

and Speed

T.P.

, A comparison of normalization methods for high density oligonucleotide array data based on variance and bias, Bioinformatics 19(2) (2003), 185–193.

14.

Shen

Jiang

Zhang

Ding

and Chen

, Inhibition of ATM reverses EMT and decreases metastatic potential of cisplatin-resistant lung cancer cells through JAK/STAT3/PD-L1 pathway, J Exp Clin Cancer Res 38(1) (2019), 149.

15.

Nouri

Ratther

Stylianou

Nelson

C.C.

Hollier

B.G.

and Williams

E.D.

, Androgen-targeted therapy-induced epithelial mesenchymal plasticity and neuroendocrine transdifferentiation in prostate cancer: An opportunity for intervention, Front Oncol 4 (2014), 370.

16.

Puisieux

Brabletz

and Caramel

, Oncogenic roles of EMT-inducing transcription factors, Nat Cell Biol 16(6) (2014), 488–494.

17.

Monzon

F.A.

Lyons-Weiler

Buturovic

L.J.

Rigl

C.T.

Henner

W.D.

Sciulli

Dumur

C.I.

Medeiros

and Anderson

G.G.

, Multicenter validation of a 1,550-gene expression profile for identification of tumor tissue of origin, J Clin Oncol 27(15) (2009), 2503–2508.

18.

Waqar

S.N.

Samson

P.P.

Robinson

C.G.

Bradley

Devarakonda

Govindan

Gao

Puri

and Morgensztern

, Non-small-cell lung cancer with brain metastasis at presentation, Clinical Lung Cancer 19(4) (2018), e373–e379.

19.

Goldberg

S.B.

Schalper

K.A.

Gettinger

S.N.

Mahajan

Herbst

R.S.

Chiang

A.C.

Lilenbaum

Wilson

F.H.

Omay

S.B.

J.B.

Jilaveanu

Tran

Pavlik

Rowen

Gerrish

Komlo

Gupta

Wyatt

Ribeiro

Kluger

Zhou

Wei

Chiang

V.L.

and Kluger

H.M.

, Pembrolizumab for management of patients with NSCLC and brain metastases: Long-term results and biomarker analysis from a non-randomised, open-label, phase 2 trial, The Lancet Oncology 21(5) (2020), 655–663.

20.

Chen

Wang

and Xue

, Leptomeningeal metastases in non-small-cell lung cancer, The Lancet Oncology 19(1) (2018), e43–e55.

21.

Luke

Blazquez

Yamaci

R.F.

Pregler

Hannus

Menhart

Hellwig

Wester

H.J.

Kropf

Heudobler

Grosse

Moosbauer

Hutterer

Hau

Riemenschneider

M.J.

Bayerlova

Bleckmann

Polzer

Beissbarth

Klein

C.A.

and Pukrop

, Isolated metastasis of an EGFR-L858R-mutated NSCLC of the meninges: The potential impact of CXCL12/CXCR4 axis in EGFRmut NSCLC in diagnosis, follow-up and treatment, Oncotarget 9(27) (2018), 18844–18857.

22.

Ajona

Zandueta

Corrales

Moreno

Pajares

M.J.

Ortiz-Espinosa

Martínez-Terroba

Perurena

de Miguel

F.J.

Jantus-Lewintre

Camps

Vicent

Agorreta

Montuenga

L.M.

Pio

and Lecanda

, Blockade of the complement C5a/C5aR1 axis impairs lung cancer bone metastasis by CXCL16-mediated effects, American Journal of Respiratory and Critical Care Medicine 197(9) (2018), 1164–1176.

23.

Katakami

Kunikane

Takeda

Takayama

Sawa

Saito

Harada

Yokota

Ando

Saito

Yokota

Ohashi

and Eguchi

, Prospective study on the incidence of bone metastasis (BM) and skeletal-related events (SREs) in patients (pts) with stage IIIB and IV lung cancer-CSP-HOR 13, Journal of Thoracic Oncology: Official Publication of the International Association for the Study of Lung Cancer 9(2) (2014), 231–238.

24.

X.D.

Bie

Q.L.

Zhang

Yan

Z.H.

and Han

Z.J.

, Wnt10B is critical for the progression of gastric cancer, Oncol Lett 13(6) (2017), 4231–4237.

25.

Chen

Wang

and Xue

, Expression and the clinical significance of Wnt10a and Wnt10b in endometrial cancer are associated with the Wnt/beta-catenin pathway, Oncol Rep 29(2) (2013), 507–514.

26.

Mendes

Kim

H.T.

Lungu

and Stoica

, MMP2 role in breast cancer brain metastasis development and its regulation by TIMP2 and ERK1/2, Clin Exp Metastasis 24(5) (2007), 341–351.

27.

Kenny

H.A.

Kaur

Coussens

L.M.

and Lengyel

, The initial steps of ovarian cancer cell metastasis are mediated by MMP-2 cleavage of vitronectin and fibronectin, J Clin Invest 118(4) (2008), 1367–1379.

28.

Rojiani

M.V.

Alidina

Esposito

and Rojiani

A.M.

, Expression of MMP-2 correlates with increased angiogenesis in CNS metastasis of lung carcinoma, Int J Clin Exp Pathol 3(8) (2010), 775–781.

29.

Lien

M.Y.

Tsai

H.C.

Chang

A.C.

Tsai

M.H.

Hua

C.H.

Wang

S.W.

and Tang

C.H.

, Chemokine CCL4 induces vascular endothelial growth factor C expression and lymphangiogenesis by miR-195-3p in oral squamous cell carcinoma, Front Immunol 9 (2018), 412.

30.

Sasaki

Baba

Nishimura

Hayakawa

Hashimoto

Gotoh

and Mukaida

, Essential roles of the interaction between cancer cell-derived chemokine, CCL4, and intra-bone CCR5-expressing fibroblasts in breast cancer bone metastasis, Cancer Lett 378(1) (2016), 23–32.

31.

Costantini

and Barbieri

, The HLA-DR mediated signalling increases the migration and invasion of melanoma cells, the expression and lipid raft recruitment of adhesion receptors, PD-L1 and signal transduction proteins, Cell Signal 36 (2017), 189–203.

32.

DeNardo

D.G.

Barreto

J.B.

Andreu

Vasquez

Tawfik

Kolhatkar

and Coussens

L.M.

, CD4(+) T cells regulate pulmonary metastasis of mammary carcinomas by enhancing protumor properties of macrophages, Cancer Cell 16(2) (2009), 91–102.

33.

Liu

Guo

Liang

and Wang

, Identification of candidate genes that may contribute to the metastasis of prostate cancer by bioinformatics analysis, Oncol Lett 15(1) (2018), 1220–1228.

34.

Kuol

Stojanovska

Apostolopoulos

and Nurgali

, Role of the nervous system in cancer metastasis, J Exp Clin Cancer Res 37(1) (2018), 5.

35.

Kalluri

and Weinberg

R.A.

, The basics of epithelial-mesenchymal transition, J Clin Invest 119(6) (2009), 1420–1428.

36.

Bonnomet

Syne

Brysse

Feyereisen

Thompson

E.W.

Noel

Foidart

J.M.

Birembaut

Polette

and Gilles

, A dynamic in vivo model of epithelial-to-mesenchymal transitions in circulating tumor cells and metastases of breast cancer, Oncogene 31(33) (2012), 3741–3753.

37.

Zhu

C.Q.

Ding

Strumpf

Weir

B.A.

Meyerson

Pennell

Thomas

R.K.

Naoki

Ladd-Acosta

Liu

Pintilie

Der

Seymour

Jurisica

Shepherd

F.A.

and Tsao

M.S.

, Prognostic and predictive gene signature for adjuvant chemotherapy in resected non-small-cell lung cancer, J Clin Oncol 28(29) (2010), 4417–4424.

38.

Langfelder

and Horvath

, WGCNA: An R package for weighted correlation network analysis, BMC Bioinformatics 9 (2008), 559.

39.

Huang da

Sherman

B.T.

and Lempicki

R.A.

, Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources, Nat Protoc 4(1) (2009), 44–57.

40.

Therneau

, A package for survival analysis in S, 2015.

Study of the co-expression gene modules of non-small cell lung cancer metastases

Abstract

BACKGROUND:

OBJECTIVE:

METHODS:

RESULTS:

CONCLUSION:

Keywords

Abbreviations

2.1 Prepare data for WGCNA analysis

2.2 WGCNA analysis for the gene modules

Table 1 Study of module eigengenes by survival analysis

2.5 Clinical significance of the typical genes from the purple module

3. Method

3.1 Data collections

3.2 WGCNA analysis

3.3 Gene function analysis

3.4 Survival analysis

Supplementary data

sj-docx-1-cbm-10.3233_CBM-201605.docx - Supplemental material

Funding

Footnotes

Conflict of interest

References

Table 1
Study of module eigengenes by survival analysis