Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Subpathway Analysis based on Signaling-Pathway Impact Analysis of Signaling Pathway

  • Xianbin Li,

    Affiliation Department of Physics and Electronic information engineering, Wenzhou University,Wenzhou, Zhejiang, China

  • Liangzhong Shen,

    Affiliation Department of Physics and Electronic information engineering, Wenzhou University,Wenzhou, Zhejiang, China

  • Xuequn Shang,

    Affiliation School of Computer Science and Technology, Northwestern Polytechnical University, Xi'an, China

  • Wenbin Liu

    wbliu6910@126.com

    Affiliation Department of Physics and Electronic information engineering, Wenzhou University,Wenzhou, Zhejiang, China

Abstract

Pathway analysis is a common approach to gain insight from biological experiments. Signaling-pathway impact analysis (SPIA) is one such method and combines both the classical enrichment analysis and the actual perturbation on a given pathway. Because this method focuses on a single pathway, its resolution generally is not very high because the differentially expressed genes may be enriched in a local region of the pathway. In the present work, to identify cancer-related pathways, we incorporated a recent subpathway analysis method into the SPIA method to form the “sub-SPIA method.” The original subpathway analysis uses the k-clique structure to define a subpathway. However, it is not sufficiently flexible to capture subpathways with complex structure and usually results in many overlapping subpathways. We therefore propose using the minimal-spanning-tree structure to find a subpathway. We apply this approach to colorectal cancer and lung cancer datasets, and our results show that sub-SPIA can identify many significant pathways associated with each specific cancer that other methods miss. Based on the entire pathway network in the Kyoto Encyclopedia of Genes and Genomes, we find that the pathways identified by sub-SPIA not only have the largest average degree, but also are more closely connected than those identified by other methods. This result suggests that the abnormality signal propagating through them might be responsible for the specific cancer or disease.

Introduction

Various “omics” technologies, such as microarrays, RNAseq, and gas chromatography mass spectrometry, can help to identify potentially interesting (i.e., differential) genes and metabolites, especially those associated with specific diseases. However, using such information to better understand the underlying biological phenomena remains a challenge. Pathway analysis has become a popular approach for gaining insight into the underlying biology of differentially expressed genes (DEGs) and proteins. The evolution of knowledge-driven pathway analysis can be divided into four generations. The first-generation analysis method is called the overrepresentation approach (ORA) [1] and compares the number of differential genes expected to hit the given pathway by chance. If this number differs significantly from that expected by chance, the pathway is significant. Many tools are based on first-generation methods, such as Onto-Express [2, 3] or GOEASE [4]. The ORA assumes that each gene is independent of the other genes. However, biological processes form a complex web of interactions between gene products that constitute different pathways. Functional class scoring (FCS) is a second-generation method for detecting coordinated changes in the expression of genes in the same pathway. Gene-set enrichment analysis is an example of a second-generation method [5, 6]. Because upstream genes may have a larger impact than downstream genes, pathway-topology-based approaches, such as signaling-pathway impact analysis (SPIA) [7] and ScorePAGE [8], qualify as third-generation methods. In particular, SPIA combines classical enrichment analysis and the perturbation on a given pathway, which allows it to capture the influence of upstream genes. Following SPIA, Vaske et al. [9] proposed a method named PARADIGM, which integrates diverse high-throughput genomics information with known signaling pathways to provide patient-specific genomic inferences on the state of gene activities, complexes, and cellular processes. Recently, researchers proposed that key subpathway regions may represent the corresponding pathway and be more relevant for interpreting the associated biological phenomena. Moreover, several studies show that abnormalities in subpathway regions of metabolic pathways may contribute to the etiology of diseases [1012]. Subpathway analysis in signaling pathways has also been studied, resulting in approaches such as DEgraph [13], the clipper approach [14], and Pathiways [15]. These are qualified as fourth-generation methods.

In the present work, we combine the approaches of subpathway analysis and SPIA, which we call sub-SPIA, to identify biologically meaningful signaling pathways. One key problem in subpathway analysis is how to define a subpathway. Li et al. [11] used the k-clique concept to define a subpathway. However, pathways are usually sparsely connected and are composed of many linear structures. The k-clique concept has two limitations: (i) the relationship between DEGs in a subpathway may not exactly form a k-clique structure, and (ii) the k-clique algorithm usually results in many redundant and overlapped subpathways. The minimal-spanning tree (MST) is a simple data structure that is frequently used to represent tightly related nodes in a graph. It is more appropriate to represent various subpathways than the k-clique structure, especially in sparse-pathway networks.

We applied the sub-SPIA method to the colorectal cancer (CRC) and lung cancer datasets, and our results demonstrate that the proposed method can identify more disease-related pathways than SPIA, DEgraph, Clipper, and Pathiways. Furthermore, we find that most of the pathways identified by sub-SPIA have a high degree and are tightly connected within the entire pathway network. This result reveals that diseases (e.g., cancer) may result from the synergic interactions of a group of related pathways.

Results

There are 137 signaling pathways in KEGG. To deduce pathway significance, we used a significance threshold of 1% on the p-values corrected for false discovery rate (FDR). For both sub-SPIA and SPIA, the FDR-adjusted p-values PG(FDR) which combine the enrichment and perturbation p-values (see definition in the method section), were computed from the nominal p-values by using the R function “p.adjust.” We present herein the significantly enriched pathways by applying sub-SPIA and SPIA to CRC and lung cancer datasets. The significant pathways that were also identified by DEgraph, Clipper, and Pathiways and are given in Tables 1 and 2.

thumbnail
Table 1. Significantly enriched pathways identified by sub-SPIA and SPIA from the CRC dataset.

https://doi.org/10.1371/journal.pone.0132813.t001

thumbnail
Table 2. Significantly enriched pathways identified by sub-SPIA and SPIA from lung cancer dataset.

https://doi.org/10.1371/journal.pone.0132813.t002

Pathway Analysis

Colorectal-Cancer Dataset.

Sub-SPIA identified 18 potential pathways associated with CRC, while SPIA identified only 8 pathways. These pathways and their corresponding PG are listed in Table 1. Four common pathways were identified by both SPIA and sub-SPIA: focal adhesion, pathways in cancer, PPAR signaling pathway, and ECM-receptor interaction. Previous studies show that some common pathways are highly associated with various cancers, including CRC and lung cancer, such as focal adhesion [16, 17], pathways in cancer, regulation of actin cytoskeleton [18, 19], the MAPK signaling pathway [2027], ECM-receptor interaction, the Wnt signaling pathway [2832], and the p53 signaling pathway [33, 34]. Sub-SPIA identified all of these pathways, whereas SPIA only identified three of them. However, some pathways are not likely to be relevant to CRC, such as the pathways for Huntington’s disease, Alzheimer's disease, and Parkinson’s disease [7]. These are ranked first, second, and sixth in significance by SPIA, whereas sub-SPIA identifies them either as not significant or not found.

Concerning the significant pathways identified by sub-SPIA but not by SPIA, existing evidence indicates that they may be associated with CRC. The dysregulation of the regulation of actin cytoskeleton pathway plays a key role in the progression of CRC [18, 19]; this pathway contains a differential gene LIMK2 that promotes tumor-cell invasion and metastasis. LIMK2 expression is associated with CRC progression, and its deletion affects gastrointestinal stem cell regulation and tumor development.

The complement and coagulation cascades pathway involves 14 DEGs and also plays a key role in the progression of CRC. For example, in a comparison of radiosensitive and radio-resistant lines of CRC cells, the result of five lines of CRC cells shows that 30 up-regulated genes were identified as being involved in DNA damage response pathways, immune response pathways, and the complement and coagulation cascades pathway [35, 36].

The deregulation of the notch signaling pathway was observed in colorectal and other forms of cancer [37]. NOTCH is the center of the subpathway identified by sub-SPIA (see Fig 1). The abnormal expression of NOTCH and its upstream gene NUMB would lead to the dysfunction of many downstream genes. The Notch signaling pathway is involved in regulating stem-cell hierarchy and determining cell fate [38]. A recent study [39] indicates that inhibiting prolactin can completely abrogate the Notch signaling pathway and may provide a novel target for therapeutic intervention.

thumbnail
Fig 1. Subpathways in the Wnt signaling pathway identified by Sub-SPIA on CRC dataset.

(A) The insulin-signal pathway with DEGs highlighted. (B) Gene network obtained by graphite package. (C) Sub Gene network corresponding to the MST for Minimal-spanning tree for ns = 4. (D) Sub Gene networks corresponding to the MST for Minimal-spanning tree for ns = 2. Red indicates an up-regulated gene and blue indicates a down-regulated gene. Reprinted from http://www.kegg.jp/kegg/kegg1.html under a CC BY license, with permission from Miwako Karikomi, original copyright 2013.

https://doi.org/10.1371/journal.pone.0132813.g001

Elafin is a protease inhibitor with antibacterial effects against bacteria such as Pseudomonas aeruginosa and Staphylococcus aureus, and which modulates inflammation through its antiprotease activity. A delicate relationship between proteases and antiproteases is central in determining how inflammation develops during colitis. Proteases damage tissues during inflammation whereas protease inhibitors minimize tissue damage and facilitate healing. In a microarray study of human colonic biopsies, patients with ulcerative colitis expressed 30-fold more elafin mRNA than healthy controls, which indirectly indicates that the Staphylococcus aureus infection pathway is associated with CRC [40].

The T cell receptor signaling pathway is responsible for impaired immune responsiveness of T cells in cancer patients. In such patients, the TCR-β gene in lymphocytes is less expressed in colon cancer, renal cell carcinoma, melanoma, and cervical cancer, which has important clinical implications for monitoring the patient immune status during therapy [41].

The human epidermal growth factor receptor family contains four members that belong to the ErbB lineage of proteins (ErbB1-4)[4245] [4245]. Downstream ErbB signaling modules include the phosphatidylinositol 3-kinase/Akt pathway, the Ras/Raf/MEK/ERK1/2 pathway, and the phospholipase C pathway. Several malignancies are associated with the mutation or increased expression of members of the ErbB family, including lung, breast, and stomach cancer [42]. By immunohistochemistry, Yao et al. found that receptor tyrosine kinase (RTK) members ErbB2, ErbB3, and c-Met were indeed differentially overexpressed in samples from CRC patients, leading to constitutive activation of RTK signaling pathways. By using the ErbB2-specific inhibitor Lapatinib and the c-Met-specific inhibitor PHA-665752, they further demonstrated that this constitutive activation of RTK signaling is necessary to the survival of CRC cells [46].

As of 2008, CRC is the third most commonly diagnosed cancer in males and the second in females [47]. Worldwide, mycobacterium tuberculosis (MTB) is the second leading cause of death from an infectious disease [48]. Significant evidence shows that active MTB could be present in patients with metastatic CRC [49]. Although the samples we used were not reported to be infected with these pathogens, it is possible that the tuberculosis pathway plays an important role in them and thus is identified by sub-SPIA as being significant in CRC.

Lung Cancer Dataset

Sub-SPIA identified 19 potential pathways associated with lung cancer, whereas SPIA only identified 5 pathways (see S2 Table). These pathways and their corresponding PG are listed in Table 2. Both methods identified three common pathways: ECM-receptor interaction, cell cycle, and focal adhesion. Qiu et al.[50] found that DEGs may promote metastasis of lung cancer cells through complicated networks, including pathways in cancer, focal adhesion [51, 52], regulation of actin cytoskeleton [53], the p53 signaling pathway, the MAPK signaling pathway [54], ECM-receptor interaction, and the Wnt signaling pathway [55]. These pathways are also identified by sub-SPIA, whereas SPIA only identifies five of them with P<0.06.

Gene anomalies in cell-cycle pathways have been frequently observed in a variety of human malignancies, including lung cancer [5658]. Dysfunctions of proto-oncogenes, such as CCND1 and STK15, and tumor-suppressor genes, such as p53, p21, and p27, are commonly associated with increased cell proliferation, defective apoptosis, elevated cancer risk, and poor survival rates [50, 59].

Constitutive activation of NF-κB was detected in non-small-cell lung carcinoma (NSCLC) and was implicated in imparting resistance to CDDP [60, 61]. Therefore, inhibiting NF-κB signaling may be a critical target for enhancing the efficacy of CDDP against NSCLC. Wang et al. [62] found that the inhibition of NF-κB by geldanamycin (GA) could be responsible for the synergistic apoptosis-inducing effect of GA and CDDP in NSCLC cells and tumor xenografts.

By using methylation-specific PCR, Marsit et al. [63] found the epigenetic alterations in the fanconi anemia pathway in NSCLC. They demonstrated that inactivation of the FANC-BRCA pathway is relatively common in solid tumors and may be related to tobacco and alcohol exposure and to the survival of these patients.

Aberrant activities of the vascular smooth muscle contraction and the focal adhesion pathways may play key roles in the initiation and development of NSCLC. Fang et al. [64] demonstrated the indispensable roles of these two signal pathways in the carcinogenesis of NSCLC.

The comorbidity of lung cancer and pulmonary tuberculosis (TB) is a clinical problem whose diagnosis and treatment presents a challenge [65]. In a review of 36 patients with salmonella pneumonia, or lung abscess, Cohen et al. noted that thirteen of them (36%) had prior abnormalities of the lung or pleura. From among these thirteen, seven had lung malignancies [66]. Although the samples we used were not reported to be infected with salmonella pathogens, it is possible that the tuberculosis pathway plays an important role in them and thus is identified by sub-SPIA.

Subpathway Analysis

According to the KEGG pathway hsa05200 (also called pathways in cancer), the signals from the outside pathways, such as Wnt signaling and MAPK signaling, are common driving forces during carcinogenesis [15]. The sub-SPIA not only improves the identification resolution of cancer-related pathways, but also helps biologists to understand their underlying mechanisms.

The Wnt signaling pathway has a canonical Wnt/β-catenin cascade and two non-canonical pathways named the Wnt/Planar cell-polarity (Wnt/PCP) pathway and the Wnt/Ca2+ pathway. Sub-SPIA identified that both the canonical Wnt/β-catenin cascade and the Wnt/Planar cell-polarity (Wnt/PCP) pathways are significantly enriched with DEGs relating with CRC and lung cancer. Figs 1A and 2A show the Wnt signaling pathway with the differentially expressed genes highlighted on the two datasets. The identified subpathways are circled by the red dashed lines for ns = 4 and the blue dashed lines for ns = 2. β-catenin was observed highly expressed in the CRC patients. According to the identified Wnt/β-catenin cascade, the upstream signal of Wnt triggers the activation of gene Frizzled and Dv1, and the downregulation of the inhibitor GSK-3β by Dv1 activates the expression of β-catenin. Genes belonging to the Wnt/PCP pathway, such as JNK are known to be up-regulated in cancer [67]. Based on the results given in Figs 1A and 2A, we can see that gene expression of various activators and inhibitors related to Wnt signaling activation is consistent with the regulation flows.

thumbnail
Fig 2. Subpathways in the Wnt signaling pathway identified by Sub-SPIA on lung cancer dataset.

(A) The insulin-signal pathway with DEGs highlighted. (B) Gene network obtained by graphite package. (C) Sub Gene network corresponding to the MST for Minimal-spanning tree for ns = 4. (D) Sub Gene network corresponding to the MST for Minimal-spanning tree for ns = 2. Red indicates an up-regulated gene and blue indicates a down-regulated gene. Reprinted from http://www.kegg.jp/kegg/kegg1.html under a CC BY license, with permission from Miwako Karikomi, original copyright 2013.

https://doi.org/10.1371/journal.pone.0132813.g002

Figs 1C and 2C show the gene networks corresponding to the identified subpathway for ns = 4. We see that, not only do the number of DEGs differ for the two datasets, but also some DEGs exist that are highly expressed on the CRC dataset but lowly expressed on the lung dataset. this demonstrates that the Wnt signaling pathway may function differently in the CRC and lung cancer. For example, the downstream of the Wnt/PCP pathway which leads to the gene transcription in CRC is down-regulated whereas it is upregulated in lung cancer.

Comparison with other Approaches

As mentioned in introduction, several subpathway analysis methods have been proposed to exploit the various functions of the pathway. For example, DEgraph [13] uses multivariate analysis to identify differential-expression patterns that are coherent with a given subgraph structure. The clipper approach [14] applies a Gaussian graphical model to deconstruct the pathway into smaller subgraphs (cliques). Pathiways aims to identify subpathways that have significant differences in the probability of activation of the individual stimulus-response signaling circuits. The results of applying these methods to the CRC data set and to the lung cancer dataset are included in S1 and S2 Tables, respectively.

It seems that, in comparison with sub-SPIA, the three methods mentioned above are very sensitive to the dataset. With a p-value of 0.01, Clipper, and DEgraph identified 13 and 21 significant pathways, respectively, from the CRC dataset. However, they identified many significant pathways with dubious biological meaning in the lung cancer dataset (107 and 89, respectively). Pathiways identified 10 significant pathways from the colorectal- cancer dataset but no significant pathways at all from the lung cancer data set. Therefore, in comparison, the proposed sub-SPIA generally performs more stably with both data sets.

From Tables 1 and 2, we see that some of the potential significant pathways identified by sup-SPIA were not found by these three methods, especially in the CRC data set. This result is attributed to the fact that these methods have different motivations. Although the number of significant pathways identified by the recently proposed Pathiways is relative small, but it can analyze the activity probability of a subpathway, which provides a tool with which to understand the biological mechanisms of diseases. However, these methods, including the proposed sub-SPIA method, mainly focus on identifying subpathways.

Relationship between Identified Pathways

From the perspective of systems biology, the emergence and development of a cancer or disease may be due to abnormal changes in some related pathways instead of in individual pathways. Based on the connection between pathways, we construct the entire pathway network in the KEGG. Fig 3A is the pathway network that consists of 239 nodes and 818 edges. Fig 3B is its corresponding degree distribution. Each node represents a specific pathway and the edge represents the connection between nodes. Obviously, most of the pathways are sparsely connected and only a few pathways which may serve some important biological function for living systems are highly connected with other pathways. The core region in Fig 3A is the 12 pathways which connected more 20 other pathways.

thumbnail
Fig 3. Pathway network in KEGG.

(A) The connection between Pathways. The core nodes represent pathways connected with more than 20 other pathways. (B) The degree distribution of pathways in (A).

https://doi.org/10.1371/journal.pone.0132813.g003

To investigate the topological relationship of the significant pathways identified by each method in the entire pathway network, we extracted from Fig 3 the identified pathways and their directly connected neighbors. The average degree, clustering coefficient, and betweenness of the identified pathways are presented in Table 3. The significance of the average degree is obtained from generating 10 000 random degree distributions. The average degree of those pathways identified by sub-SPIA and Pathiways is larger than that identified by other methods, and the p-value of their average degree further indicates that they are very significant. The average betweenness of the pathways identified by sub-SPIA and Pathiways is also the largest. Both the average degree and betweenness actually reflect the hub characteristic of the identified pathways; the two observations just mentioned demonstrate that the pathways identified by sub-SPIA and Pathiways generally play important roles in the entire system.

thumbnail
Table 3. The topological characteristics of the significantly enriched pathways identified by five methods.

https://doi.org/10.1371/journal.pone.0132813.t003

The average clustering coefficient generally reflects the degree of the closeness of the identified pathways. For both datasets, the average clustering coefficient of pathways identified by sub-SPIA is about 0.38, which is the largest obtained of the five methods. Considering the relative sparseness of the entire pathway network, this indicates that the pathways identified by sub-SPIA are more closely related than those identified by other methods. In other words, they interact closely with each other to fulfill various biological functions.

Discussion

To identify significant pathways, we combine the subpathway analysis and the SPIA in a method we call sub-SPIA. Compared with the original SPIA, sub-SPIA dramatically improves the resolution for identifying significant pathways because subpathway analysis focuses on a local region in a pathway. From Tables 1 and 2, we see that the p-value of pathways identified by sub-SPIA is much smaller than for SPIA. Furthermore, the flexibility of the minimal-spanning tree makes it possible to capture various subpathways with complicated topologies. These two factors make sub-SPIA more sensitive than SPIA, allowing sub-SPIA to identify more potential pathways associated with specific cancers or diseases. Note that sub-SPIA misses a few pathways identified by SPIA that are related to the corresponding cancer, such as the Epstein-Barr virus infection [68, 69]. However, upon investigating this pathway, we find that most of the DEGs on it are scattered over the entire pathway instead of being clustered in a local region, which is why the spanning-tree algorithm fails to group them as a cluster or subpathway.

We also compared sub-SPIA with three other subpathway-analysis methods: DEgraph, Clipper, and Pathiways. Based on the number of the identified significant pathways, sub-SPIA is more stable than these three methods because they either find many dubious pathways or no significant pathways at all on the lung cancer dataset. Analyzing the relationship between the significant pathways identified by those methods further reveals that those identified by sub-SPIA not only play a more important role in the entire network but also are highly connected. Because most of these pathways are known to relate to various cancers, we may conclude that the abnormality signal propagating through them may be responsible for the specific cancer or disease.

Finally, The MST structure overcomes the disadvantage of the k-clique method, which generally results in many overlapped subpathways. For example, in the pathways identified by sub-SPIA on the two datasets with p-values of 5%, most pathways contain just one significant subpathway whereas only seven pathways contain two to five subpathways (see supplemental files). Additionally, the cross talk between pathways had been found to play an important role in cancer [70]. The flexibility of MST has the potential to find cross talk if we apply it to connected pathway networks.

Materials and Methods

Dataset

The first is the CRC dataset, which compares 12 CRC samples with 10 normal samples [71] by using the Affymetrix HG-U133 Plus 2.0 microarray platform (ID = GSE4107). The second dataset, which was initially analyzed by Landi et al. [72], is a lung cancer dataset and is publicly available at the GEO database (accession number GSE10072).This dataset contains 58 tumor samples and 49 normal samples and uses the Affymetrix Human Genome U133A Array.

Minimal-Spanning Tree

A spanning tree is a subgraph that is a tree and connects all the vertices of the parent graph. Minimal-spanning trees (MSTs) have many applications in telecommunication and transportation-route design. As mentioned in the introduction, genes in a pathway are generally sparsely connected and the DEGs mapped in it may not be connected directly, so we search for a minimal-spanning tree that includes both the maximum number of signature nodes and the minimum number of non-signature nodes. This concept is more flexible than the k-clique concept for representing a subpathway. The Kruskal algorithm is one commonly used algorithm to find the minimal-spanning tree.

Methods

Sub-SPIA is implemented by using the statistical programming language R and can be freely downloaded from https://github.com/eshinesimida/subpathway-analysis. The main steps to identify significantly enriched subpathways include (i) reconstruct the gene network from the signaling pathways, (ii) map the DEGs in the constructed gene network, and (iii) locate subpathways and evaluate their statistical and perturbation significance. In this work, a pathway confers significantly enriched pathways if and only if it contains at least one significant-enrichment subpathway.

Map DEGs to graphs of pathways

We downloaded the signaling pathways from KEGG. These are directed graphs based on biochemical-reaction information in the KGML file (an XML representation of the KEGG-pathway information, see http://www.kegg.jp/kegg/xml/). The KEGG database provides one xml file for each pathway. In the KGML format, nodes in pathways often correspond to multiple gene products and compounds. Gene products can be divided into protein complexes and groups containing alternative members (gene families). We applied the graphite [73] package to reconstruct the gene network from the pathway. Next, the DEGs were mapped to the gene network.

Locate subpathways by DEGs

DEGs within a pathway provide important signatures to locate subpathways associated with diseases of interest. Because DEGs in a subpathway are generally closely connected in the converted gene network, we first find all node sets in a pathway which include closely connected signature nodes. The main process is as follows: (i) we define a node set S = ϕ and add a randomly selected signature node to it. (ii) if the shortest path between a signature nodes u∈S and v∉S is less than ns + 1 (parameter ns is the maximum number of permitted non-signature nodes in the shortest path between two signature nodes)ns+1, then we add the non-signature nodes in their shortest path and node v to S. (iii) We repeat step (ii) until no other signature nodes can be added to S. (iv) If there exist some signature nodes in the pathway not included in the node set S, we repeat steps (i)–(iii) on them to find other node sets S′.

Obviously, genes in the node set S forms a connected sub-gene-network. However, it may include some redundant non-signature nodes that are not necessarily needed to make the signature nodes form a connected subregion. According to the definition of the MST, these redundant non-signature nodes should locate in the leaves of MST. Therefore, we first use the Kruskal algorithm to find the MSTs and then remove these non-signature nodes in the leaves of MST. We convert the subgraph of nodes in set S into an undirected weighted graph. The weight w of an edge connecting two nodes u and v is defined as where ku and kv are the number of signature nodes connected with nodes u and v. The Kruskal algorithm first sorts the edges in a graph according to their weight and then iteratively selects |S|-1 minimal-weighted edges that cannot form a loop with the previously selected edges at each step. After trimming these non-signature nodes in the leaves, we obtain the MST that includes the maximum number of signature nodes and the minimum number of non-signature nodes.

Flexibility can be introduced to this subpathway strategy by varying the parameter ns. A smaller value of ns means that only those nodes meeting stricter distance similarities will be added to the corresponding subpathway, and the subpathways thus identified become smaller compared with what happens with larger values of ns. A smaller number of permitted non-signature nodes helps to increase the ratio of signature nodes in the located subpathway regions. In this paper, we use ns = 4 to search the minimal-spanning tree based on the DEGs.

In Figs 1 and 2, we show the subpathways obtained in the Wnt signaling pathway from the CRC and lung cancer datasets by Sub-SPIA. Figs 1A and 2A show the Wnt-signal pathway with the differentially expressed genes highlighted. Figs 1B and 2B show the gene network obtained by the graphite package. Figs 1C and 2C show the subnetwork corresponding to the MST obtained by ns = 4, and Figs 1D and 2D show the subnetwork corresponding to the MST obtained by ns = 2. The subpathway circled by the red dashed lines in Figs 1A and 2A correspond to the subnetwork in Figs 1C and 2C. The subpathway circled by the blue dashed lines in Figs 1A and 2A correspond to the subnetwork in Figs 1D and 2D.

We now give a brief description of the MST by referring to the node set in Fig 1B. Assuming ns = 2 and starting from signature node 16, we reach the node set S = {16,17,19,20,21,22,24}, which includes the two non-signature nodes 19 and 21. Fig 4A shows the sub-gene network of the nodes in Fig 1B. Fig 4B shows the converted undirected weighted graph from it. Fig 3C shows the MST obtained from the Kruskal algorithm. The final MST is obtained by removing leaf node 19, because it is a non-signature node.

thumbnail
Fig 4. An example of the MST.

(A) A sub gene network extracted from Fig 1B for ns = 2. (B) The converted undirected weighted graph. (C) The resulted MST.

https://doi.org/10.1371/journal.pone.0132813.g004

The statistical significance of subpathways

We used the hypergeometric test and abnormal perturbation to calculate the statistical significance of each subpathway. This process contains two types of evidence: the overrepresentation of DEGs and the abnormal perturbation in a given subpathway. The first probability PNDE = P(XNde|H0) captures the significance of the given subpathway Pi by an over-representation analysis of the number of DE genes (NDE) observed on the pathway.H0 stands for the null hypothesis where random DEGs appear on a given subpathway. From a biological perspective, this would mean that the subpathway is not relevant to the condition under study. The value of PNDE represents the probability of obtaining a number of DEGs on the given subpathway that is at least as large as the observed number Nde. The probability PNDE is obtained by assuming that NDE follows a hypergeometric distribution. If the whole genome has a total of m genes of which t are involved in the pathway under investigation, and the set of genes submitted for analysis has a total of n genes of which r are involved in the same pathway, then the p-value can be calculated to evaluate enrichment significance for that pathway as follows:

The second probability PPERT is calculated based on the amount of perturbation measured in each pathway. A gene perturbation factor is defined as: where the term ΔE(gi) represents the signed normalized measured expression change of the gene gi (log fold-change if two conditions are compared). The second term in Equation is the sum of perturbation factors of the genes gj directly upstream of the target gene gi, normalized by the number of downstream genes of each such gene Nds(gj). The absolute value of βij quantifies the strength of the interaction between genes gi and gj. Other detailed information can be referred in Ref. [7].

The global probability value PG, which tests whether the subpathway is significantly perturbed by the condition being studied, combines PNDE and PPERT to rank the pathways. When the null hypothesis is true, the probability of observing a pair of p-values whose product, ci = PNDE(i)*PPERT(i) is at least as low as that observed for a given pathway Pi is When several tens of subpathways are tested simultaneously, small PG values can occur by chance. Therefore, we control the FDR of the subpathway 1% by applying the commonly used FDR algorithm [74].

Supporting Information

S1 Table. Results obtained by the five methods in CRC dataset (XLS).

https://doi.org/10.1371/journal.pone.0132813.s001

(XLSX)

S2 Table. Results obtained by the five methods in lung cancer dataset (XLS).

https://doi.org/10.1371/journal.pone.0132813.s002

(XLSX)

Author Contributions

Conceived and designed the experiments: XL WL. Performed the experiments: XL LS. Analyzed the data: XL LS XS WL. Contributed reagents/materials/analysis tools: LS XS. Wrote the paper: XL WL.

References

  1. 1. Khatri P, Sirota M, Butte AJ. Ten years of pathway analysis: current approaches and outstanding challenges. PLoS Comput Biol. 2012;8(2):e1002375. Epub 2012/03/03. pmid:22383865; PubMed Central PMCID: PMCPmc3285573.
  2. 2. Khatri P, Draghici S, Ostermeier GC, Krawetz SA. Profiling gene expression using onto-express. Genomics. 2002;79(2):266–70. Epub 2002/02/07. pmid:11829497.
  3. 3. Draghici S, Khatri P, Martins RP, Ostermeier GC, Krawetz SA. Global functional profiling of gene expression. Genomics. 2003;81(2):98–104. Epub 2003/03/07. pmid:12620386.
  4. 4. Zheng Q, Wang XJ. GOEAST: a web-based software toolkit for Gene Ontology enrichment analysis. Nucleic Acids Res. 2008;36(Web Server issue):W358–63. Epub 2008/05/20. pmid:18487275; PubMed Central PMCID: PMCPmc2447756.
  5. 5. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545–50. Epub 2005/10/04. pmid:16199517; PubMed Central PMCID: PMCPmc1239896.
  6. 6. Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J, et al. PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet. 2003;34(3):267–73. Epub 2003/06/17. pmid:12808457.
  7. 7. Tarca AL, Draghici S, Khatri P, Hassan SS, Mittal P, Kim JS, et al. A novel signaling pathway impact analysis. Bioinformatics. 2009;25(1):75–82. Epub 2008/11/08. pmid:18990722; PubMed Central PMCID: PMCPmc2732297.
  8. 8. Rahnenfuhrer J, Domingues FS, Maydt J, Lengauer T. Calculating the statistical significance of changes in pathway activity from gene expression data. Stat Appl Genet Mol Biol. 2004;3:Article16. Epub 2006/05/02. pmid:16646794.
  9. 9. Vaske CJ, Benz SC, Sanborn JZ, Earl D, Szeto C, Zhu J, et al. Inference of patient-specific pathway activities from multi-dimensional cancer genomics data using PARADIGM. Bioinformatics. 26. England 2010. p. i237–45. pmid:20529912
  10. 10. Li C, Shang D, Wang Y, Li J, Han J, Wang S, et al. Characterizing the network of drugs and their affected metabolic subpathways. PLoS One. 2012;7(10):e47326. Epub 2012/11/01. pmid:23112813; PubMed Central PMCID: PMCPmc3480395.
  11. 11. Li C, Li X, Miao Y, Wang Q, Jiang W, Xu C, et al. SubpathwayMiner: a software package for flexible identification of pathways. Nucleic Acids Res. 2009;37(19):e131. Epub 2009/08/27. pmid:19706733; PubMed Central PMCID: PMCPmc2770656.
  12. 12. Li X, Li C, Shang D, Li J, Han J, Miao Y, et al. The implications of relationships between human diseases and metabolic subpathways. PLoS One. 2011;6(6):e21131. Epub 2011/06/23. pmid:21695054; PubMed Central PMCID: PMCPmc3117879.
  13. 13. Jacob L, Neuvial P, Dudoit S. More power via graph-structured tests for differential expression of gene networks. 2012:561–600.
  14. 14. Martini P, Sales G, Massa MS, Chiogna M, Romualdi C. Along signal paths: an empirical gene set approach exploiting pathway topology. Nucleic Acids Res. 2013;41(1):e19. Epub 2012/09/25. pmid:23002139; PubMed Central PMCID: PMCPmc3592432.
  15. 15. Sebastian-Leon P, Vidal E, Minguez P, Conesa A, Tarazona S, Amadoz A, et al. Understanding disease mechanisms with models of signaling pathway activities. BMC Syst Biol. 82014. p. 121.
  16. 16. Albasri A, Fadhil W, Scholefield JH, Durrant LG, Ilyas M. Nuclear expression of phosphorylated focal adhesion kinase is associated with poor prognosis in human colorectal cancer. Anticancer Res. 2014;34(8):3969–74. Epub 2014/07/31. pmid:25075018.
  17. 17. Heffler M, Golubovskaya VM, Dunn KM, Cance W. Focal adhesion kinase autophosphorylation inhibition decreases colon cancer cell growth and enhances the efficacy of chemotherapy. Cancer Biol Ther. 2013;14(8):761–72. Epub 2013/06/25. pmid:23792569; PubMed Central PMCID: PMCPmc3841216.
  18. 18. Lourenco FC, Munro J, Brown J, Cordero J, Stefanatos R, Strathdee K, et al. Reduced LIMK2 expression in colorectal cancer reflects its role in limiting stem cell proliferation. Gut. 2014;63(3):480–93. Epub 2013/04/16. pmid:23585469; PubMed Central PMCID: PMCPmc3932979.
  19. 19. Kanaan Z, Qadan M, Eichenberger MR, Galandiuk S. The actin-cytoskeleton pathway and its potential role in inflammatory bowel disease-associated human colorectal cancer. Genet Test Mol Biomarkers. 2010;14(3):347–53. Epub 2010/04/22. pmid:20406101; PubMed Central PMCID: PMCPmc2938833.
  20. 20. Grossi V, Peserico A, Tezil T, Simone C. p38alpha MAPK pathway: A key factor in colorectal cancer therapy and chemoresistance. World J Gastroenterol. 2014;20(29):9744–58. pmid:25110412.
  21. 21. Lu X, Li C, Wang YK, Jiang K, Gai XD. Sorbitol induces apoptosis of human colorectal cancer cells via p38 MAPK signal transduction. Oncol Lett. 2014;7(6):1992–6. Epub 2014/06/17. pmid:24932277; PubMed Central PMCID: PMCPmc4049712.
  22. 22. Gupta J, del Barco Barrantes I, Igea A, Sakellariou S, Pateras IS, Gorgoulis VG, et al. Dual function of p38alpha MAPK in colon cancer: suppression of colitis-associated tumor initiation but requirement for cancer cell survival. Cancer Cell. 2014;25(4):484–500. Epub 2014/04/02. pmid:24684847.
  23. 23. Zuo L, Lu M, Zhou Q, Wei W, Wang Y. Butyrate suppresses proliferation and migration of RKO colon cancer cells though regulating endocan expression by MAPK signaling pathway. Food Chem Toxicol. 2013. Epub 2014/02/12. pmid:24513060.
  24. 24. Waterston RH, Lindblad-Toh K, Birney E, Rogers J, Abril JF, Agarwal P, et al. Initial sequencing and comparative analysis of the mouse genome. Nature. 2002;420(6915):520–62. Epub 2002/12/06. pmid:12466850.
  25. 25. Haigis KM, Kendall KR, Wang Y, Cheung A, Haigis MC, Glickman JN, et al. Differential effects of oncogenic K-Ras and N-Ras on proliferation, differentiation and tumor progression in the colon. Nat Genet. 2008;40(5):600–8. Epub 2008/04/01. pmid:18372904; PubMed Central PMCID: PMCPmc2410301.
  26. 26. Wang S, El-Deiry WS. TRAIL and apoptosis induction by TNF-family death receptors. Oncogene. 2003;22(53):8628–33. Epub 2003/11/25. pmid:14634624.
  27. 27. De Robertis M, Massi E, Poeta ML, Carotti S, Morini S, Cecchetelli L, et al. The AOM/DSS murine model for the study of colon carcinogenesis: From pathways to diagnosis and therapy studies. J Carcinog. 2011;10:9. Epub 2011/04/13. pmid:21483655; PubMed Central PMCID: PMCPmc3072657.
  28. 28. Pooja T, Karunagaran D. Emodin suppresses Wnt signaling in human colorectal cancer cells SW480 and SW620. Eur J Pharmacol. 2014. Epub 2014/09/11. pmid:25205133.
  29. 29. Hu TH, Yao Y, Yu S, Han LL, Wang WJ, Guo H, et al. SDF-1/CXCR4 promotes epithelial-mesenchymal transition and progression of colorectal cancer by activation of the Wnt/beta-catenin signaling pathway. Cancer Lett. 2014. Epub 2014/08/26. pmid:25150783.
  30. 30. Amado NG, Predes D, Moreno MM, Carvalho IO, Mendes FA, Abreu JG. Flavonoids and Wnt/beta-catenin signaling: potential role in colorectal cancer therapies. Int J Mol Sci. 2014;15(7):12094–106. Epub 2014/07/10. pmid:25007066; PubMed Central PMCID: PMCPmc4139831.
  31. 31. Vermeulen L, De Sousa EMF, van der Heijden M, Cameron K, de Jong JH, Borovski T, et al. Wnt activity defines colon cancer stem cells and is regulated by the microenvironment. Nat Cell Biol. 2010;12(5):468–76. Epub 2010/04/27. pmid:20418870.
  32. 32. Astin M, Griffin T, Neal RD, Rose P, Hamilton W. The diagnostic value of symptoms for colorectal cancer in primary care: a systematic review. Br J Gen Pract. 2011;61(586):e231–43. Epub 2011/05/31. pmid:21619747; PubMed Central PMCID: PMCPmc3080228.
  33. 33. Bunney TD, Katan M. Phosphoinositide signalling in cancer: beyond PI3K and PTEN. Nat Rev Cancer. 2010;10(5):342–52. Epub 2010/04/24. pmid:20414202.
  34. 34. Sauer K, Cooke MP. Regulation of immune cell development through soluble inositol-1,3,4,5-tetrakisphosphate. Nat Rev Immunol. 2010;10(4):257–71. Epub 2010/03/26. pmid:20336153; PubMed Central PMCID: PMCPmc2922113.
  35. 35. Pesson M, Volant A, Uguen A, Trillet K, De La Grange P, Aubry M, et al. A gene expression and pre-mRNA splicing signature that marks the adenoma-adenocarcinoma progression in colorectal cancer. PLoS One. 2014;9(2):e87761. Epub 2014/02/12. pmid:24516561; PubMed Central PMCID: PMCPmc3916340.
  36. 36. Huang MY, Wang JY, Chang HJ, Kuo CW, Tok TS, Lin SR. CDC25A, VAV1, TP73, BRCA1 and ZAP70 gene overexpression correlates with radiation response in colorectal cancer. Oncol Rep. 2011;25(5):1297–306. Epub 2011/02/24. pmid:21344162.
  37. 37. Nickoloff BJ, Osborne BA, Miele L. Notch signaling as a therapeutic target in cancer: a new approach to the development of cell fate modifying agents. Oncogene. 2003;22(42):6598–608. Epub 2003/10/07. pmid:14528285.
  38. 38. Kopan R, Turner DL. The Notch pathway: democracy and aristocracy in the selection of cell fate. Curr Opin Neurobiol. 1996;6(5):594–601. Epub 1996/10/01. pmid:8937822.
  39. 39. Neradugomma NK, Subramaniam D, Tawfik OW, Goffin V, Kumar TR, Jensen RA, et al. Prolactin signaling enhances colon cancer stemness by modulating Notch signaling in a Jak2-STAT3/ERK manner. Carcinogenesis. 2014;35(4):795–806. Epub 2013/11/23. pmid:24265293; PubMed Central PMCID: PMCPmc3977144.
  40. 40. Ishibashi K, Kuwabara K, Ishiguro T, Ohsawa T, Okada N, Miyazaki T, et al. Short-term intravenous antimicrobial prophylaxis in combination with preoperative oral antibiotics on surgical site infection and methicillin-resistant Staphylococcus aureus infection in elective colon cancer surgery: results of a prospective randomized trial. Surg Today. 2009;39(12):1032–9. Epub 2009/12/10. pmid:19997797.
  41. 41. Kono K, Ichihara F, Iizuka H, Sekikawa T, Matsumoto Y. Expression of signal transducing T-cell receptor zeta molecules after adoptive immunotherapy in patients with gastric and colon cancer. Int J Cancer. 1998;78(3):301–5. Epub 1998/10/10. pmid:9766562.
  42. 42. Roskoski R Jr. The ErbB/HER family of protein-tyrosine kinases and cancer. Pharmacol Res. 2014;79:34–74. Epub 2013/11/26. pmid:24269963.
  43. 43. Rubinson DA, Hochster HS, Ryan DP, Wolpin BM, McCleary NJ, Abrams TA, et al. Multi-drug inhibition of the HER pathway in metastatic colorectal cancer: results of a phase I study of pertuzumab plus cetuximab in cetuximab-refractory patients. Invest New Drugs. 2014;32(1):113–22. Epub 2013/04/10. pmid:23568716; PubMed Central PMCID: PMCPmc3775976.
  44. 44. Pan T, Sun J, Zhou J, Fu Z, Hu Y, Zheng S, et al. Function and mode of action of cytohesins in the epidermal growth factor pathway in colorectal cancer cells. Oncol Lett. 2013;5(2):521–6. Epub 2013/02/20. pmid:23420529; PubMed Central PMCID: PMCPmc3573086.
  45. 45. Khelwatty SA, Essapen S, Seddon AM, Modjtahedi H. Prognostic significance and targeting of HER family in colorectal cancer. Front Biosci (Landmark Ed). 2013;18:394–421. Epub 2013/01/02. pmid:23276932.
  46. 46. Yao YL, Shao J, Zhang C, Wu JH, Zhang QH, Wang JJ, et al. Proliferation of colorectal cancer is promoted by two signaling transduction expression patterns: ErbB2/ErbB3/AKT and MET/ErbB3/MAPK. PLoS One. 2013;8(10):e78086. Epub 2013/11/10. pmid:24205104; PubMed Central PMCID: PMCPmc3813539.
  47. 47. Jemal A, Bray F, Center MM, Ferlay J, Ward E, Forman D. Global cancer statistics. CA Cancer J Clin. 2011;61(2):69–90. Epub 2011/02/08. pmid:21296855.
  48. 48. WHO publishes Global tuberculosis report 2013. Euro Surveill. 2013;18(43). Epub 2013/11/02. pmid:24176622.
  49. 49. Hirashima T, Nagai T, Shigeoka H, Tamura Y, Yoshida H, Kawahara K, et al. Comparison of the clinical courses and chemotherapy outcomes in metastatic colorectal cancer patients with and without active Mycobacterium tuberculosis or Mycobacterium kansasii infection: a retrospective study. BMC Cancer. 2014;14(1):770. Epub 2014/10/19. pmid:25326267; PubMed Central PMCID: PMCPmc4210613.
  50. 50. Dobashi Y. Cell cycle regulation and its aberrations in human lung carcinoma. Pathol Int. 2005;55(3):95–105. Epub 2005/03/04. pmid:15743317.
  51. 51. Dy GK, Ylagan L, Pokharel S, Miller A, Brese E, Bshara W, et al. The prognostic significance of focal adhesion kinase expression in stage I non-small-cell lung cancer. J Thorac Oncol. 2014;9(9):1278–84. Epub 2014/08/15. pmid:25122425; PubMed Central PMCID: PMCPmc4133746.
  52. 52. Havel LS, Kline ER, Salgueiro AM, Marcus AI. Vimentin regulates lung cancer cell adhesion through a VAV2-Rac1 pathway to control focal adhesion kinase activity. Oncogene. 2014. Epub 2014/05/27. pmid:24858039.
  53. 53. Chen QY, Xu LQ, Jiao DM, Yao QH, Wang YY, Hu HZ, et al. Silencing of Rac1 modifies lung cancer cell migration, invasion and actin cytoskeleton rearrangements and enhances chemosensitivity to antitumor drugs. Int J Mol Med. 2011;28(5):769–76. Epub 2011/08/13. pmid:21837360.
  54. 54. Zheng F, Tang Q, Wu J, Zhao S, Liang Z, Li L, et al. p38alpha MAPK-mediated induction and interaction of FOXO3a and p53 contribute to the inhibited-growth and induced-apoptosis of human lung adenocarcinoma cells by berberine. J Exp Clin Cancer Res. 2014;33:36. Epub 2014/04/29. pmid:24766860; PubMed Central PMCID: PMCPmc4013801.
  55. 55. Qiu X, Guo S, Wu H, Chen J, Zhou Q. Identification of Wnt pathway, uPA, PAI-1, MT1-MMP, S100A4 and CXCR4 associated with enhanced metastasis of human large cell lung cancer by DNA microarray. Minerva Med. 2012;103(3):151–64. Epub 2012/06/02. pmid:22653095.
  56. 56. Vincenzi B, Schiavon G, Silletta M, Santini D, Perrone G, Di Marino M, et al. Cell cycle alterations and lung cancer. Histol Histopathol. 2006;21(4):423–35. Epub 2006/01/27. pmid:16437388.
  57. 57. Au NH, Cheang M, Huntsman DG, Yorida E, Coldman A, Elliott WM, et al. Evaluation of immunohistochemical markers in non-small cell lung cancer by unsupervised hierarchical clustering analysis: a tissue microarray study of 284 cases and 18 markers. J Pathol. 2004;204(1):101–9. Epub 2004/08/13. pmid:15307143.
  58. 58. Abdulkader I, Sanchez L, Cameselle-Teijeiro J, Gude F, Chavez JE, Lopez-Lopez R, et al. Cell-cycle-associated markers and clinical outcome in human epithelial cancers: a tissue microarray study. Oncol Rep. 2005;14(6):1527–31. Epub 2005/11/08. pmid:16273250.
  59. 59. Malusecka E, Zborek A, Krzyzowska-Gruca S. Changes in expression of pRb, p16 and cyclin D1 in non-small cell lung cancer: an immunohistochemical study. Folia Histochem Cytobiol. 1999;37(1):19–24. Epub 1999/03/26. pmid:10091946.
  60. 60. Baby J, Pickering BF, Vashisht Gopal YN, Van Dyke MW. Constitutive and inducible nuclear factor-kappaB in immortalized normal human bronchial epithelial and non-small cell lung cancer cell lines. Cancer Lett. 2007;255(1):85–94. Epub 2007/05/12. pmid:17493745.
  61. 61. Chen W, Li Z, Bai L, Lin Y. NF-kappaB in lung cancer, a carcinogenesis mediator and a prevention and therapy target. Front Biosci (Landmark Ed). 2011;16:1172–85. Epub 2011/01/05. pmid:21196225; PubMed Central PMCID: PMCPmc3032584.
  62. 62. Wang LH, Li Y, Yang SN, Wang FY, Hou Y, Cui W, et al. Gambogic acid synergistically potentiates cisplatin-induced apoptosis in non-small-cell lung cancer through suppressing NF-kappaB and MAPK/HO-1 signalling. Br J Cancer. 2014;110(2):341–52. Epub 2013/12/05. pmid:24300974; PubMed Central PMCID: PMCPmc3899775.
  63. 63. Marsit CJ, Liu M, Nelson HH, Posner M, Suzuki M, Kelsey KT. Inactivation of the Fanconi anemia/BRCA pathway in lung and oral cancers: implications for treatment and survival. Oncogene. 2004;23(4):1000–4. Epub 2003/12/03. pmid:14647419.
  64. 64. Fang X, Netzer M, Baumgartner C, Bai C, Wang X. Genetic network and gene set enrichment analysis to identify biomarkers related to cigarette smoking and lung cancer. Cancer Treat Rev. 2013;39(1):77–88. Epub 2012/07/14. pmid:22789435.
  65. 65. Varol Y, Varol U, Unlu M, Kayaalp I, Ayranci A, Dereli MS, et al. Primary lung cancer coexisting with active pulmonary tuberculosis. Int J Tuberc Lung Dis. 2014;18(9):1121–5. Epub 2014/09/06. pmid:25189563.
  66. 66. Cohen JI, Bartlett JA, Corey GR. Extra-intestinal manifestations of salmonella infections. Medicine (Baltimore). 1987;66(5):349–88. Epub 1987/09/01. pmid:3306260.
  67. 67. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell. 144. United States: 2011 Elsevier Inc; 2011. p. 646–74. https://doi.org/10.1016/j.cell.2011.02.013 pmid:21376230
  68. 68. Bouvard V, Baan R, Straif K, Grosse Y, Secretan B, El Ghissassi F, et al. A review of human carcinogens—Part B: biological agents. Lancet Oncol. 2009;10(4):321–2. Epub 2009/04/08. pmid:19350698.
  69. 69. Koshiol J, Gulley ML, Zhao Y, Rubagotti M, Marincola FM, Rotunno M, et al. Epstein-Barr virus microRNAs and lung cancer. Br J Cancer. 2011;105(2):320–6. Epub 2011/06/10. pmid:21654679; PubMed Central PMCID: PMCPmc3142804.
  70. 70. Han J, Li C, Yang H, Xu Y, Zhang C, Ma J, et al. A novel dysregulated pathway-identification analysis based on global influence of within-pathway effects and crosstalk between pathways. J R Soc Interface. 2015;12(102):20140937. Epub 2015/01/01. pmid:25551156; PubMed Central PMCID: PMCPMC4277084.
  71. 71. Hong Y, Ho KS, Eu KW, Cheah PY. A susceptibility gene set for early onset colorectal cancer that integrates diverse signaling pathways: implication for tumorigenesis. Clin Cancer Res. 2007;13(4):1107–14. Epub 2007/02/24. pmid:17317818.
  72. 72. Landi MT, Dracheva T, Rotunno M, Figueroa JD, Liu H, Dasgupta A, et al. Gene expression signature of cigarette smoking and its role in lung adenocarcinoma development and survival. PLoS One. 2008;3(2):e1651. Epub 2008/02/26. pmid:18297132; PubMed Central PMCID: PMCPmc2249927.
  73. 73. Sales G, Calura E, Cavalieri D, Romualdi C. graphite—a Bioconductor package to convert pathway topology to gene network. BMC Bioinformatics. 2012;13:20. Epub 2012/02/02. pmid:22292714; PubMed Central PMCID: PMCPmc3296647.
  74. 74. Benjamini Y, Yekutieli D. The control of the false discovery rate in multiple testing under dependency. 2001:1165–88.
  75. 75. Pasto A, Serafin V, Pilotto G, Lago C, Bellio C, Trusolino L, et al. NOTCH3 signaling regulates MUSASHI-1 expression in metastatic colorectal cancer cells. Cancer Res. 2014;74(7):2106–18. Epub 2014/02/15. pmid:24525742.
  76. 76. Serafin V, Persano L, Moserle L, Esposito G, Ghisi M, Curtarello M, et al. Notch3 signalling promotes tumour growth in colorectal cancer. J Pathol. 2011;224(4):448–60. Epub 2011/05/21. pmid:21598247.
  77. 77. Liu X, Cao L, Ni J, Liu N, Zhao X, Wang Y, et al. Differential BCCIP gene expression in primary human ovarian cancer, renal cell carcinoma and colorectal cancer tissues. Int J Oncol. 2013;43(6):1925–34. Epub 2013/10/09. pmid:24101097.
  78. 78. Cao L, Zhu L, Yang J, Su J, Ni J, Du Y, et al. Correlation of low expression of hMOF with clinicopathological features of colorectal carcinoma, gastric cancer and renal cell carcinoma. Int J Oncol. 2014;44(4):1207–14. Epub 2014/01/24. pmid:24452485.
  79. 79. Xu B, Goldman JS, Rymar VV, Forget C, Lo PS, Bull SJ, et al. Critical roles for the netrin receptor deleted in colorectal cancer in dopaminergic neuronal precursor migration, axon guidance, and axon arborization. Neuroscience. 2010;169(2):932–49. Epub 2010/05/25. pmid:20493932.
  80. 80. Je EM, Gwak M, Oh H, Choi MR, Choi YJ, Lee SH, et al. Frameshift mutations of axon guidance genes ROBO1 and ROBO2 in gastric and colorectal cancers with microsatellite instability. Pathology. 2013;45(7):645–50. Epub 2013/11/20. pmid:24247621.
  81. 81. Argaw A, Duff G, Zabouri N, Cecyre B, Chaine N, Cherif H, et al. Concerted action of CB1 cannabinoid receptor and deleted in colorectal cancer in axon guidance. J Neurosci. 2011;31(4):1489–99. Epub 2011/01/29. pmid:21273433.
  82. 82. Li VS, Yuen ST, Chan TL, Yan HH, Law WL, Yeung BH, et al. Frequent inactivation of axon guidance molecule RGMA in human colon cancer through genetic and epigenetic mechanisms. Gastroenterology. 2009;137(1):176–87. Epub 2009/03/24. pmid:19303019.
  83. 83. Wang W, Spitz MR, Yang H, Lu C, Stewart DJ, Wu X. Genetic variants in cell cycle control pathway confer susceptibility to lung cancer. Clin Cancer Res. 2007;13(19):5974–81. Epub 2007/10/03. pmid:17908995.
  84. 84. You DJ, Park CR, Lee HB, Moon MJ, Kang JH, Lee C, et al. A Splicing Variant of NME1 Negatively Regulates NF-kappaB Signaling and Inhibits Cancer Metastasis by Interacting with IKKbeta. J Biol Chem. 2014;289(25):17709–20. Epub 2014/05/09. pmid:24811176; PubMed Central PMCID: PMCPmc4067205.
  85. 85. Warin RF, Chen H, Soroka DN, Zhu Y, Sang S. Induction of lung cancer cell apoptosis through a p53 pathway by [6]-shogaol and its cysteine-conjugated metabolite M2. J Agric Food Chem. 2014;62(6):1352–62. Epub 2014/01/23. pmid:24446736; PubMed Central PMCID: PMCPmc3983336.
  86. 86. Liu ZH, Wang MH, Ren HJ, Qu W, Sun LM, Zhang QF, et al. Interleukin 7 signaling prevents apoptosis by regulating bcl-2 and bax via the p53 pathway in human non-small cell lung cancer cells. Int J Clin Exp Pathol. 2014;7(3):870–81. Epub 2014/04/04. pmid:24695377; PubMed Central PMCID: PMCPmc3971289.
  87. 87. Tan X, Chen M. MYLK and MYL9 expression in non-small cell lung cancer identified by bioinformatics analysis of public expression data. Tumour Biol. 2014. Epub 2014/09/03. pmid:25179839.
  88. 88. Samonis G, Maraki S, Kouroussis C, Mavroudis D, Georgoulias V. Salmonella enterica pneumonia in a patient with lung cancer. J Clin Microbiol. 2003;41(12):5820–2. Epub 2003/12/10. pmid:14662992; PubMed Central PMCID: PMCPmc309020.