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

Open Access

Peer-reviewed

Research Article

A Hybrid Computational Method for the Discovery of Novel Reproduction-Related Genes

  • Lei Chen ,

    ‡ These authors contributed equally to this work.

    Affiliation College of Information Engineering, Shanghai Maritime University, Shanghai, 201306, People’s Republic of China

    ⨯
  • Chen Chu ,

    ‡ These authors contributed equally to this work.

    Affiliation State Key Laboratory of Molecular Biology, Shanghai Key Laboratory of Molecular Andrology, Institute of Biochemistry and Cell Biology, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai, 200031, People’s Republic of China

    ⨯
  • Xiangyin Kong,

    Affiliation Institute of Health Sciences, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai, 200025, People’s Republic of China

    ⨯
  • Guohua Huang,

    Affiliation Institute of Systems Biology, Shanghai University, Shanghai, 200444, People’s Republic of China

    ⨯
  • Tao Huang ,

    tohuangtao@126.com (TH); cai_yud@126.com (YDC)

    Affiliation Institute of Health Sciences, Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai, 200025, People’s Republic of China

    ⨯
  • Yu-Dong Cai

    tohuangtao@126.com (TH); cai_yud@126.com (YDC)

    Affiliation Institute of Systems Biology, Shanghai University, Shanghai, 200444, People’s Republic of China

    ⨯

Abstract

Uncovering the molecular mechanisms underlying reproduction is of great importance to infertility treatment and to the generation of healthy offspring. In this study, we discovered novel reproduction-related genes with a hybrid computational method, integrating three different types of method, which offered new clues for further reproduction research. This method was first executed on a weighted graph, constructed based on known protein-protein interactions, to search the shortest paths connecting any two known reproduction-related genes. Genes occurring in these paths were deemed to have a special relationship with reproduction. These newly discovered genes were filtered with a randomization test. Then, the remaining genes were further selected according to their associations with known reproduction-related genes measured by protein-protein interaction score and alignment score obtained by BLAST. The in-depth analysis of the high confidence novel reproduction genes revealed hidden mechanisms of reproduction and provided guidelines for further experimental validations.

Citation: Chen L, Chu C, Kong X, Huang G, Huang T, Cai Y-D (2015) A Hybrid Computational Method for the Discovery of Novel Reproduction-Related Genes. PLoS ONE 10(3): e0117090. https://doi.org/10.1371/journal.pone.0117090

Academic Editor: Peter Csermely, Semmelweis University, HUNGARY

Received: September 6, 2014; Accepted: December 13, 2014; Published: March 13, 2015

Copyright: © 2015 Chen et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited

Data Availability: All relevant data are within the paper and its Supporting Information files.

Funding: National Basic Research Program of China (2011CB510101, 2011CB510102), the National Natural Science Foundation of China (31371335, 61202021, 61373028, 11371008, 61303099), the Innovation Program of the Shanghai Municipal Education Commission (12YZ120, 12ZZ087) and the Shanghai Educational Development Foundation (12CG55). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

All living creatures generate healthy offspring and maintain population growth through reproduction. In mammals, this fundamental and complex process includes the development of male and female germ cells [1,2], fertilization and embryonic development [3]. Impairment in any of these stages can lead to severe consequences such as infertility, miscarriage and fetal defects. Among mammalian species, humans are more susceptible to reproductive problems. It has been reported that infertility affects approximately 15% of couples [4], and this percentage is increasing. Over the past few decades, mounting evidence has indicated that human fertility and reproduction may be jeopardized by genetic abnormalities [5,6], environmental chemicals [7,8], unhealthy diets and lifestyles [9–11]; however, the underlying molecular mechanisms are still largely unknown. Therefore, it is important to identify reproduction-related genes and pathways that may be used as biomarkers for early diagnosis and treatment.

With advancements in reproductive biology research, a number of reproduction-related genes have been identified and characterized. Their functions are enriched in different reproductive process stages, including gonad development [12,13], germ cell development [14], meiosis [15,16], sperm-egg binding [17,18] and embryo implantation and development [19]. For example, the nanos proteins function in primordial germ cell (PGC) migration into the gonads [14], the TDRP and TNP proteins are involved in spermatogenesis [20,21], and ZP family proteins facilitate the sperm acrosomal reaction and sperm-egg binding [18]. Additionally, several important pathways have proven to be directly involved in reproduction. The Wnt signaling pathway plays a crucial role in gonad development by patterning the sex-specific vasculature and regulating steroidogenic cell recruitment [12]. Not only have these studies promoted the understanding of human reproduction mechanisms but the resulting data have also served as useful resources for deducing new reproductive-related genes and predicting their functions [22,23].

One possible strategy for elucidating the molecular mechanisms underlying the entire reproductive process is to identify all reproduction-related genes and to test their biological roles in vitro and in vivo. However, such an approach is challenging because the research space, i.e., the number of human genes, is large, and confirming a reproduction-related gene is temporally and financially intensive.

There are already several network based disease gene identification methods. The basis of most methods is Guilt-by-association [24]. They assumed that the genes are similar with their neighbors. Therefore, the neighbors of the seed disease genes are very likely to be disease genes as well. In general, it is true due to modularized nature of network [25]. But when the seed disease genes are incomplete and scattered overall the whole network, the performance of such method will be poor.

Many other methods are based on Random Walk with Restart (RWR) [26–31]. RWR simulates a walker who randomly walks on the network. It starts from the seed disease genes and moves to its randomly chosen neighbors at each step [28]. After many steps, the probability of the walker walks to each node on the network will be steady. Based on these probabilities, the novel candidate disease genes can be identified.

Different variations of Guilt-by-association and RWR were developed for different research topics, such as neighbor counting [32], RWOAG (https://r-forge.r-project.org/R/?group_id=1126) developed by Kohler et al. [28], HumanNet developed by Lee et al. [33].

Shortest path based methods have less applications, but in yeast longevity study, it has been proven to be useful for identifying the genetic determinants [34]. And for disease gene identification, there are several successful applications based on shortest path [35–37].

To discover novel reproduction-related genes, we compared these methods and developed a hybrid computational method which integrates the network topology, sequence similarity and protein interaction confidence score. The biological significance of the identified novel reproduction genes were evaluated by enrichment analysis and manually literature survey. Many of the reproduction gene candidates looked very promising.

Materials and Methods

2.1 Materials

The known 115 human reproduction-related genes with experimental evidence in the Gene Ontology database (GO:0000003) were downloaded from the following website: http://amigo.geneontology.org/amigo/term/GO:0000003 (May 10, 2014) [38]. These 115 genes are listed in the S1 Information.

According to the methods in [35–37], to conduct our assessment, we also required data from a protein-protein interaction (PPI) network. We downloaded the file (protein.links.detailed.v9.1.txt.gz) containing the PPI information from STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) (http://www.string-db.org/) [39], a large database containing known and predicted protein interactions. These protein interactions are derived from the following sources: (I) genomic context, (II) high-throughput experiments, (III) (conserved) co-expression and (IV) previous knowledge. Thus, these interactions include physical and functional associations of proteins, which therefore widely measure the relationship between proteins and are different from experimentally determined PPIs provided in some other databases, such as DIP (Database of Interaction Proteins) database [40], BioGRID [41], etc. This kind of PPIs has been applied to investigate some protein-related problems [35–37,42–44]. We extracted 1,640,707 PPIs of human from the obtained file, where 86,854 are validated by solid experiments. Each obtained PPI contains two proteins (represented by ensembl IDs) and eight types of score entitled by ‘neighborhood’, ‘fusion’, ‘cooccurence’, ‘coexpression’, ‘experimental’, ‘database’, ‘textmining’, and ‘combined_score’. Since the last type score (with a range of 150–999) integrated the information of other seven types of score, it was used in this study to measure the interaction strength between two proteins. For convenience, it was termed as confidence score and denoted by S(p1, p2) for a certain PPI containing proteins p1 and p2. And proteins p1 and p2 were deemed to be interacting proteins if the interaction between them was a member of 1,640,707 PPIs.

2.2 Graph-based method to select candidate genes and further selection

Some studies have shown that two interacting proteins are more likely to share similar functions than those that do not interact with each other [45,46]. The interacting proteins of the reproduction-related genes may share some reproductive functions. The interaction confidence scores should also be considered, i.e., proteins that can interact with known reproduction-related genes with high confidence scores are more likely to possess reproduction-related functions. To test this hypothesis, a weighted graph G = (V, E) was built according to the information of PPIs as follows: V consisted of all human proteins occurring in 1,640,707 PPIs and two nodes were adjacent if and only if the corresponding proteins can interact with each other, i.e., they were interacting proteins. As the range of confidence score is between 150 and 999, a weight was assigned to each edge e = (n1, n2) by w(e) = 1000-S(p1, p2), where p1 and p2 are corresponding proteins of nodes n1 and n2, respectively. The graph-based method was fully based on this weighted graph G. Please refer to our previous studies for detailed information on this method [35–37]. A brief procedure of the method follows:

  • Step 1. Apply Dijkstra’s algorithm [47] to search all shortest paths in G such that their endpoints were known reproduction-related genes.
  • Step 2. For each node n in G, count the number of paths obtained in Step 1 which contained n as an inner node. This value was termed as “betweenness” in this study. In fact, the betweenness indicates the direct and indirect influences of proteins at distant network [48]. Here, it suggested the direct and indirect association with reproduction-related genes.
  • Step 3. Select nodes, i.e., corresponding genes, with betweenness larger than zero as shortest path genes.
  • Step 4. Randomly produce 500 gene/node sets in G. Each produced set had the same size of the set consisting of known reproduction-related genes but was different from it.
  • Step 5. For each gene set, use the Dijkstra’s algorithm to search all shortest paths in G connecting any pair of genes in the set and calculate the betweenness for each shortest path gene obtained in Step 3 based on these shortest paths (betweenness of some shortest path genes may be zero).
  • Step 6. For each shortest path gene, count the number of gene sets on which the betweenness was larger than that on the set consisting of known reproduction-related genes, thereby calculating the permutation FDR (False Discovery Rate) defined as this value divided by 500.
  • Step 7. Select shortest path genes whose permutation FDRs were smaller than 0.05 as candidate genes.

2.3 Random walk with restart

Random Walk with Restart (RWR) [26–31,49] simulates a random walker starting from m known reproduction genes on the network and moves to its randomly chosen interaction neighbors at each step [28]. In each step, the state probabilities Pt + 1 at time t + 1 is (1) where Pt + 1 is state probabilities at time t, r is the restart probability, 0.7 as suggested by previous literatures [28], A is the column-wise normalized adjacency matrix of the protein interaction network, P0 is the initial state probabilities which is a column vector with 1/ m for the m known reproduction genes and to 0 for other genes on the protein interaction network.

This process is repeated until the difference between two states is smaller than 1e-6, as suggested by previous literatures [28]. At last, each protein on the network will be assigned with a probability of being novel reproduction gene.

R package RWOAG [28] from https://r-forge.r-project.org/R/?group_id=1126 was used to apply RWR.

2.4 Similarity-based method to select candidate genes

Using the properties of protein sequences to study various protein-related problems is a classic approach. Blast (basic local alignment search tool), proposed by Altschul et al. [50], is a well-known tool that can search local similarity between two protein sequences. For two proteins p1 and p2, the alignment score of their sequences, denoted by Sb(p1, p2), can be obtained by BLAST.

It is known that two proteins with high alignment score have similar structures, thereby sharing similar functions. Thus, using the alignment scores to identify novel reproduction-related genes is an alternative method. For formulation, let S be a training gene set consisting of two parts: SR and SNR, where SR comprised known reproduction-related genes, SNR was composed of other genes that had not been validated to be reproduction-related genes. For a query gene p, we can calculate two values as follows: (2) (3)

It is easy to observe that measures the structure associations between p and genes in SR, whereas measures the structure associations between p and genes in SNR. Specifically, the high values of and indicate strong structure associations. In view of this, if , then p is identified to be a candidate gene for reproduction.

2.5 Interaction-based method to select candidate genes

As mentioned in Section 2.2, interacting proteins of proteins encoded by reproduction-related gene may share some reproductive functions, thereby inferring that genes encoding these proteins may be reproduction-related gene. Thus, we can directly use the information concerning PPIs to identify possible reproduction-related genes. We still used the notations defined in Section 2.4. For a query gene p, we can calculate two values as follows: (4) (5)

With the similar argument in Section 2.4, we can identify p as a candidate gene for reproduction if .

2.6 Hybrid method to select candidate genes

The hybrid method partly combined the methods described in Section 2.2, 2.4 and 2.5. Since graph-based method is similar with RWR and graph-based method has better performance than RWR (see Section 3.1), we chose graph-based method to represent network method and then integrated with similarity-based method and interaction-based method.

The purpose of this study is to discover new candidate reproduction-related genes, it is difficult to completely integrate the similarity-based method and interaction-based method because the set SNR is difficult to be well-defined. In view of this, we only used Eq. 2 and Eq. 4 and set thresholds to select candidate genes. The detailed procedures were as follows:

  1. The graph-based method was first applied to identify candidate genes for reproduction.
  2. For each obtained candidate gene p in (I), calculate in the similarity-based method by Eq. 2, where SR consisted of all reproduction-related genes. Then, exclude the candidate genes with less than 90.
  3. For each remaining candidate gene p in (II), calculate in the interaction-based method by Eq. 4, where SR consisted of all reproduction-related genes. Then, exclude the candidate genes with less than 900.

The workflow of the hybrid method was shown in Fig. 1.

thumbnail
Download:
Fig 1. The workflow of hybrid method for novel reproduction gene identification.

(A-E) were the steps of graph-based method, (F) was to filter candidates of the graph based method with similarity-based method and interaction-based method. (A) The known reproduction genes (red nodes) were mapped onto network. (B) The shortest path genes (green nodes) on shortest paths (dash line) were identified. (C) The known reproduction genes were permuted. (D) The shortest path genes on the shortest path between permuted reproduction genes were identified. (E) The actual betweenness of shortest path genes were compared with permuted betweenness and the genes that were not specific to reproduction were removed. (F) The candidates of the graph based method were further filtered by checking alignment score and interaction confidence score with known reproduction genes and novel candidate reproduction genes were selected if they were selected by graph-based method, similarity-based method and interaction-based method.

https://doi.org/10.1371/journal.pone.0117090.g001

Results and Discussions

3.1 Comparison of the four methods

This section gave the performance of the four methods described in Section 2.2–2.5 evaluated by the jackknife test, i.e., one reproduction-related gene was singled out to check whether it can be identified by the rest reproduction-related genes.

To compare the performance of the four methods in a fair circumstance, proteins occurring in PPIs were all considered. 129 shortest path genes with FDR smaller than 0.05 were discovered by graph-based method (see Section 3.2). To make a fair comparison, we considered a gene to be a candidate if its probability of being novel reproduction genes ranks on the top 129th in RWR. For similarity-based and interaction-based methods, the criteria were described in Section 2.4 and 2.5. As a result, the identified reproduction-related genes are listed in Table 1, from which we can observe that fourteen, eleven, thirteen and eight reproduction-related genes were identified by graph-based method, RWR method, similarity-based method and interaction-based method, respectively. It is clear that the graph-based method gave the best performance, followed by similarity-based method, RWR method and interaction-based method. Since RWR method and graph-based method were both network method and graph-based method had better performance than RWR, we chose graph-based method over RWR. And we arranged the graph-based method as the first choice in the hybrid method, the similarity-based method as the second choice and the interaction-based method as the last choice. The graph-based method is more likely to find global long distance candidate genes while the similarity-based method and interaction-based method are exploring the local candidates. Therefore, the graph-based candidates may cross several pathways and are scattered overall the network. They may be not significantly enriched onto single pathway, but may reveal novel mechanisms in complex biological systems, such as cross-talks and synergy effects. The false positive rates of similarity-based candidates and interaction-based candidates could be lower since they only explore limited number of local genes. Integrating these methods will balance the novelty and reliability of discovered candidate genes.

thumbnail
Download:
Table 1. Reproduction-related genes that can be identified by four methods.

https://doi.org/10.1371/journal.pone.0117090.t001

3.2 Candidate genes obtained by the hybrid method

According to the procedures of the hybrid method, the graph-based method was first applied to discover candidate genes for reproduction. By the graph-based method, of the 115 known reproduction-related genes, the shortest paths connecting any two were searched in G. The betweenness of each node was calculated, thereby obtaining 406 shortest path genes, which are listed in the S2 Information. According to Steps 4–6 of the method, the permutation FDR was calculated for each shortest path gene, which is also listed in the S2 Information. The purpose of this procedure is to exclude some genes with both high betweenness and permutation FDRs. If a certain candidate gene can always receive high betweenness for randomly produced gene sets, i.e., this gene always have strong direct and indirect associations with randomly selected genes, resulting in high permutation FDR, it cannot be deemed to be related to reproduction even if its betweenness was very high. In view of this, this kind of genes should be excluded. By setting the threshold of permutation FDR to 0.05, 129 candidate genes whose permutation FDRs were smaller than 0.05 were obtained, which are listed in the S3 Information. These genes would be further filtered by the following procedures of the hybrid method.

By the second step of hybrid method, for each candidate gene p, was computed according to Eq. 2. These values for 129 candidate genes obtained by the graph-based method are listed in S3 Information. Clearly, by setting 90 as a threshold, 27 candidate genes remained, which are listed in S4 Information.

For the remaining 27 candidate genes, the third step of hybrid method was finally used to make selection. The value was calculated by Eq. 4 for each candidate gene p. Similarly, we set a threshold of 900 to filter these candidate genes, resulting in 21 candidate genes. These genes were deemed to be significant for reproduction and were analyzed for their likelihood to be novel reproduction-related genes in the following sections. The detailed information of these 21 candidate genes are listed in Table 2.

thumbnail
Download:
Table 2. Detailed information of 21 candidate genes obtained by hybrid method.

https://doi.org/10.1371/journal.pone.0117090.t002

3.3 Analysis of the PPIs used to identify candidate genes in graph-based method

As mentioned in Section 2.1, the PPIs used in this study are not all validated by experiments, i.e., they are not very reliable. However, for wide selection of candidate genes of reproduction, some interactions which are not validated by experiments but can be found evidences in other ways should also be considered, thereby finding additional clues on the identification of novel reproduction-related genes. This section gave the statistical results of the PPIs used in the graph-based method to discover candidate genes of reproduction.

According to the graph-based method, all shortest paths connecting any pair of reproduction-related genes were searched in G. Since the graph-based method finally produced 129 genes, we extracted the paths among the aforementioned shortest paths such that each of them contained at least one member of the 129 candidate genes as inner nodes. 877 PPIs were involved in these paths and are provided in S5 Information. It is surprising that 639 (639/877 = 72.86%) interactions have been validated by experiments, which was much higher than the ratio of experimentally verified human PPIs and total human PPIs reported in STRING (86,854/1,640,707 = 5.30%). It is indicated that the candidate genes obtained by the graph-based method are quite reliable. Besides, a same number of PPIs that were not verified by experiments also gave contribution to discover new candidate genes for reproduction, which may provide new clues to study reproduction.

3.4 The functional difference between novel and known reproduction genes

To fairly compare the functions of the 21 novel reproduction genes and 115 known reproduction genes and avoid the effect of GO hierarchical structures, we analyzed the count distribution of the 21 novel genes and 115 known genes on level 3, 4 and 5 GO BP (Biological Process), respectively. On each level, the 21 novel genes and 115 known genes were mapped onto the same level GO BP terms and therefore, the GO hierarchical structures will be the same. Then, we used R package goProfiles [51,52] to calculate the significant p value of the functional annotation distributions of the 21 novel genes and 115 known genes that were the same. The p values of GO BP level 3, 4 and 5 terms were 0.0011, 0.0007 and 0.0008, respectively. The results are provided in the S6 Information. This means that the function annotations of the 21 novel genes and 115 known genes were different. The 21 genes include novel information that was not represented by 115 known genes.

We also performed Gene Ontology (GO) term and KEGG pathway analyses of the 21 significant candidate genes using DAVID (Database for Annotation, Visualization and Integrated Discovery) [53]. The enrichment results of 21 novel reproduction genes and 115 known reproduction genes can be found in the S7 Information and S8 Information, respectively.

GO analysis revealed that the 21 novel reproduction genes have significantly enriched functions in cell proliferation, cell differentiation, pattern specification and development. Comparatively, the known reproduction genes were also enriched in differentiation and development functions, but more specific to reproduction (e.g. reproductive developmental process, gamete generation and male gonad development). Furthermore, the novel reproduction genes and the known reproduction genes share several significant GO terms, including cellular process involved in reproduction, developmental process involved in reproduction and single organism reproductive process in the level 3; organ development, anatomical structure morphogenesis and regulation of cell differentiation in the level 4; embryo development, nervous system development and pattern specification process in the level 5. These results suggested the potential roles of novel reproduction genes in the reproduction processes such as gamete generation and embryonic development.

KEGG pathway analysis revealed that the 21 novel reproduction genes were enriched in TGF-β signaling (hsa04350) and cytokine-cytokine receptor interaction pathway (hsa04060). TGF-β (transforming growth factor β) superfamily members, such as bone morphogenetic proteins (BMPs), growth and differentiation factors (GDFs), anti-Müllerian hormone (AMH), Activin, Nodal and TGFβs, were secreted cytokines that involved in a number of important physiological processes in reproduction including the maintenance of stem cell pluripotency [54,55], germ cell development [56,57] and embryonic development [58–60]. Significant amount of activated TGF-β family member proteins were detected in both testis and placenta, and they were reported to regulate male spermatogenesis [56,61] as well as female pregnancy [62,63]. The BMP / Noggin signaling is powerful in controlling ES cell differentiation. BMP2 was reported to control the differentiation of embryonic stem cells into cells with the properties of extra-embryonic endoderm, and Noggin was the antagonist of BMP and blocked this form of differentiation and induced the appearance of a novel cell type that could give rise to neural precursors [64]. In our study, both the reference and the candidate genes show significant enrichment in the TGF-β signaling pathway (9 significant candidate genes shared this pathway: ACVR1, ACVR2A, INHBE, TGFBR1, TGFBR2, BMPR2, BMP4, BMP7, BMPR1B).

Other candidate genes included two Notch proteins (NOTCH1 and NOTCH2), forkhead box protein A1 (FOXA1) and H1 (FOXH1), fibroblast growth factor 4 (FGF4) and receptor 1 (FGFR1), T-box 5 (TBX5), Indian hedgehog (IHH), slit homolog 1 (SLIT1), calcium channel, voltage-dependent, L type, alpha 1S subunit (CACNA1S), nuclear receptor subfamily 0, group B, member 2 (NR0B2) and coagulation factor X (F10). Previous studies revealed part of their roles in reproduction and embryoic development. First, the Notch pathway was shown to play important roles in controlling stem cell proliferation and differentiation [65–67], which is essential in embryonic development. Other studies indicated that Notch pathway were also important for male spermatogenesis [68,69] and female oogenesis [70,71]. Then, both FOXA1 and FOXH1 are important transcription factors. FOXA1 was reported to regulate the differentiation and development of epithelial cells and ducts [72–74], while Foxh1 was shown to control the expression of many genes including Smad and Mixl1 in mouse and xenopus, and functions in patterning the early embryo [75,76]. The FGF signaling pathway is responsible for multiple events during embryo development, such as axial elongation [77] and somitogenesis [78]. Furthermore, they play diverse roles in the male reproductive system. For example, FGFR1 was also shown to be highly expressed in the male testis and maintain the spermatogonia in the undifferentiated state [79]. TBX5, this T-box transcription factor was reported to be closely related to embryonic heart and limb development and mutation of TBX5 could lead to Holt-Oram syndrome [80,81]. IHH, this Indian hedgehog signaling molecule is mainly involved in the chondcrocyte proliferation, differentiation and maturation process [82,83], which is crucial to the bone development and morphogenesis. SLIT1, this gene is highly expressed during embryonic development, mainly in the midline, hypochord, telencephalon and hindbrain with the established roles in axon guidance and cell migration [84,85]. These lines of evidence are consistent with our prediction. Other three genes predicted in our study were CACNA1S, NR0B2 and F10. CACNA1S encodes one subunits of voltage-dependent calcium channel in skeletal muscle cells and its mutation have been associated with malignant hyperthermia [86] and periodic paralysis [87]. NR0B2 is a unique nuclear receptor, which only has ligand-binding domain but no DNA-binding domain. Previous studies showed that NR0B2 interacted with several transcription factors and inhibited their function [88], and it was tightly linked to human diseases such cancer, diabetes and obesity [89]. F10 is the vitamin K-dependent coagulation factor X of the blood coagulation cascade and it also play roles in host defense and innate immunity activation [90]. So far no experimental evidence indicated that these three genes had reproduction-related functions, and further investigations are required to explore their roles in human reproduction.

Conclusion

This work contributed to the elucidation of the reproductive process by discovering novel human reproduction-related genes. Based on the known reproduction-related genes, PPIs, sequence similarity and interaction confidence score, new candidate genes were identified. Many of these newly identified genes were supported by latest literatures. It is hoped that these findings will guide investigators to confirm novel reproduction-related genes with in vivo and in vitro experiments.

Supporting Information

S1 Information. The 115 known reproduction-related genes.

https://doi.org/10.1371/journal.pone.0117090.s001

(DOCX)

S2 Information. 406 shortest path genes with betweenness greater than zero and their Permutation FDRs.

https://doi.org/10.1371/journal.pone.0117090.s002

(DOCX)

S3 Information. 129 candidate genes obtained by graph-based method and their maximum alignment scores to reproduction-related gene.

https://doi.org/10.1371/journal.pone.0117090.s003

(DOCX)

S4 Information. 27 candidate genes filtered by the second step of hybrid method and their maximum interaction score to reproduction-related gene.

https://doi.org/10.1371/journal.pone.0117090.s004

(DOCX)

S5 Information. The protein-protein interactions used to identify 129 candidate genes for reproduction.

https://doi.org/10.1371/journal.pone.0117090.s005

(XLSX)

S6 Information. The functional annotation distributions of the 21 novel reproduction genes and 115 known reproduction genes on GO BP level 3, 4 and 5.

https://doi.org/10.1371/journal.pone.0117090.s006

(XLSX)

S7 Information. GO and KEGG enrichment results of 21 novel reproduction genes.

https://doi.org/10.1371/journal.pone.0117090.s007

(XLSX)

S8 Information. GO and KEGG enrichment results of 115 known reproduction genes.

https://doi.org/10.1371/journal.pone.0117090.s008

(XLSX)

Author Contributions

Conceived and designed the experiments: LC GH TH YDC. Performed the experiments: LC GH TH. Analyzed the data: LC CC XK. Contributed reagents/materials/analysis tools: CC XK TH YDC. Wrote the paper: LC CC TH.

References

  1. 1. Clermont Y (1972) Kinetics of spermatogenesis in mammals: seminiferous epithelium cycle and spermatogonial renewal. Physiol Rev 52: 198–236. pmid:4621362
  2. 2. Pepling ME (2006) From primordial germ cell to primordial follicle: mammalian female germ cell development. Genesis 44: 622–632. pmid:17146778
  3. 3. Christians E, Davis AA, Thomas SD, Benjamin IJ (2000) Maternal effect of Hsf1 on reproductive success. Nature 407: 693–694. pmid:11048707
  4. 4. de Kretser DM (1997) Male infertility. Lancet 349: 787–790. pmid:9074589
  5. 5. Morris RS, Gleicher N (1996) Genetic abnormalities, male infertility, and ICSI. Lancet 347: 1277. pmid:8622500
  6. 6. Aston KI (2014) Genetic susceptibility to male infertility: news from genome-wide association studies. Andrology 2: 315–321. pmid:24574159
  7. 7. Sharpe RM, Franks S (2002) Environment, lifestyle and infertility—an inter-generational issue. Nat Cell Biol 4 Suppl: s33–40. pmid:12479613
  8. 8. Cabaton NJ, Wadia PR, Rubin BS, Zalko D, Schaeberle CM, et al. (2011) Perinatal exposure to environmentally relevant levels of bisphenol A decreases fertility and fecundity in CD-1 mice. Environ Health Perspect 119: 547–552. pmid:21126938
  9. 9. Ouvrier A, Alves G, Damon-Soubeyrand C, Marceau G, Cadet R, et al. (2011) Dietary cholesterol-induced post-testicular infertility. PLoS One 6: e26966. pmid:22073227
  10. 10. Attaman JA, Toth TL, Furtado J, Campos H, Hauser R, et al. (2012) Dietary fat and semen quality among men attending a fertility clinic. Hum Reprod 27: 1466–1474. pmid:22416013
  11. 11. Dechanet C, Anahory T, Mathieu Daude JC, Quantin X, Reyftmann L, et al. (2011) Effects of cigarette smoking on reproduction. Hum Reprod Update 17: 76–95. pmid:20685716
  12. 12. Jeays-Ward K, Hoyle C, Brennan J, Dandonneau M, Alldus G, et al. (2003) Endothelial and steroidogenic cell migration are regulated by WNT4 in the developing mammalian gonad. Development 130: 3663–3670. pmid:12835383
  13. 13. Morais da Silva S, Hacker A, Harley V, Goodfellow P, Swain A, et al. (1996) Sox9 expression during gonadal development implies a conserved role for the gene in testis differentiation in mammals and birds. Nat Genet 14: 62–68. pmid:8782821
  14. 14. Tsuda M, Sasaoka Y, Kiso M, Abe K, Haraguchi S, et al. (2003) Conserved role of nanos proteins in germ cell development. Science 301: 1239–1241. pmid:12947200
  15. 15. Ortega S, Prieto I, Odajima J, Martin A, Dubus P, et al. (2003) Cyclin-dependent kinase 2 is essential for meiosis but not for mitotic cell division in mice. Nat Genet 35: 25–31. pmid:12923533
  16. 16. Llano E, Gomez R, Gutierrez-Caballero C, Herran Y, Sanchez-Martin M, et al. (2008) Shugoshin-2 is essential for the completion of meiosis but not for mitotic cell division in mice. Genes Dev 22: 2400–2413. pmid:18765791
  17. 17. Leyton L, Saling P (1989) 95 kd sperm proteins bind ZP3 and serve as tyrosine kinase substrates in response to zona binding. Cell 57: 1123–1130. pmid:2472220
  18. 18. van Duin M, Polman JE, De Breet IT, van Ginneken K, Bunschoten H, et al. (1994) Recombinant human zona pellucida protein ZP3 produced by chinese hamster ovary cells induces the human sperm acrosome reaction and promotes sperm-egg fusion. Biol Reprod 51: 607–617. pmid:7819440
  19. 19. Dong YL, Reddy DM, Green KE, Chauhan MS, Wang HQ, et al. (2007) Calcitonin gene-related peptide (CALCA) is a proangiogenic growth factor in the human placental development. Biol Reprod 76: 892–899. pmid:17267696
  20. 20. Wang X, Jiang H, Zhou W, Zhang Z, Yang Z, et al. (2010) Molecular cloning of a novel nuclear factor, TDRP1, in spermatogenic cells of testis and its relationship with spermatogenesis. Biochem Biophys Res Commun 394: 29–35. pmid:20170638
  21. 21. Yu YE, Zhang Y, Unni E, Shirley CR, Deng JM, et al. (2000) Abnormal spermatogenesis and reduced fertility in transition nuclear protein 1-deficient mice. Proc Natl Acad Sci U S A 97: 4683–4688. pmid:10781074
  22. 22. Huang T, Cui W, Hu L, Feng K, Li YX, et al. (2009) Prediction of pharmacological and xenobiotic responses to drugs based on time course gene expression profiles. PLoS One 4: e8126. pmid:19956587
  23. 23. Huang T, Shi XH, Wang P, He Z, Feng KY, et al. (2010) Analysis and prediction of the metabolic stability of proteins based on their sequential features, subcellular locations and interaction networks. PLoS ONE 5: e10972. pmid:20532046
  24. 24. Oliver S (2000) Guilt-by-association goes global. Nature 403: 601–603. pmid:10688178
  25. 25. Barabasi AL, Gulbahce N, Loscalzo J (2011) Network medicine: a network-based approach to human disease. Nat Rev Genet 12: 56–68. pmid:21164525
  26. 26. Macropol K, Can T, Singh AK (2009) RRW: repeated random walks on genome-scale protein networks for local cluster discovery. BMC Bioinformatics 10: 283. pmid:19740439
  27. 27. Li Y, Patra JC (2010) Genome-wide inferring gene-phenotype relationship by walking on the heterogeneous network. Bioinformatics 26: 1219–1224. pmid:20215462
  28. 28. Kohler S, Bauer S, Horn D, Robinson PN (2008) Walking the interactome for prioritization of candidate disease genes. Am J Hum Genet 82: 949–958. pmid:18371930
  29. 29. Jiang R, Gan M, He P (2011) Constructing a gene semantic similarity network for the inference of disease genes. BMC Syst Biol 5 Suppl 2: S2. pmid:22784573
  30. 30. Chen X, Liu MX, Yan GY (2012) Drug-target interaction prediction by random walk on the heterogeneous network. Mol Biosyst 8: 1970–1978. pmid:22538619
  31. 31. Shi H, Xu J, Zhang G, Xu L, Li C, et al. (2013) Walking the interactome to identify human miRNA-disease associations through the functional link between miRNA targets and disease genes. BMC Syst Biol 7: 101. pmid:24103777
  32. 32. Schwikowski B, Uetz P, Fields S (2000) A network of protein-protein interactions in yeast. Nat Biotechnol 18: 1257–1261. pmid:11101803
  33. 33. Lee I, Blom UM, Wang PI, Shim JE, Marcotte EM (2011) Prioritizing candidate disease genes by network-based boosting of genome-wide association data. Genome Res 21: 1109–1121. pmid:21536720
  34. 34. Managbanag JR, Witten TM, Bonchev D, Fox LA, Tsuchiya M, et al. (2008) Shortest-path network analysis is a useful approach toward identifying genetic determinants of longevity. PLoS One 3: e3802. pmid:19030232
  35. 35. Zhang J, Jiang M, Yuan F, Feng KY, Cai YD, et al. (2013) Identification of age-related macular degeneration related genes by applying shortest path algorithm in protein-protein interaction network. BioMed Research International 2013: 523415. pmid:24455700
  36. 36. Li B-Q, You J, Chen L, Zhang J, Zhang N, et al. (2013) Identification of Lung-Cancer-Related Genes with the Shortest Path Approach in a Protein-Protein Interaction Network. BioMed Research International 2013: 267375. pmid:23762832
  37. 37. Jiang M, Chen Y, Zhang Y, Chen L, Zhang N, et al. (2013) Identification of hepatocellular carcinoma related genes with k-th shortest paths in a protein–protein interaction network. Mol BioSyst 9: 2720–2728. pmid:24056857
  38. 38. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, et al. (2000) Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25: 25–29. pmid:10802651
  39. 39. Jensen LJ, Kuhn M, Stark M, Chaffron S, Creevey C, et al. (2009) STRING 8-a global view on proteins and their functional interactions in 630 organisms. Nucleic acids research 37: D412–416. pmid:18940858
  40. 40. Xenarios I, Rice DW, Salwinski L, Baron MK, Marcotte EM, et al. (2000) DIP: the database of interacting proteins. Nucleic Acids Research 28: 289–291. pmid:10592249
  41. 41. Stark C, Breitkreutz B-J, Reguly T, Boucher L, Breitkreutz A, et al. (2006) BioGRID: a general repository for interaction datasets. Nucleic acids research 34: D535–D539. pmid:16381927
  42. 42. Smedley D, Köhler S, Czeschik JC, Amberger J, Bocchini C, et al. (2014) Walking the interactome for candidate prioritization in exome sequencing studies of Mendelian diseases. Bioinformatics 30: 3215–3222. pmid:25078397
  43. 43. Nusinow DP, Kiezun A, O’Connell DJ, Chick JM, Yue Y, et al. (2012) Network-based inference from complex proteomic mixtures using SNIPE. Bioinformatics 28: 3115–3122. pmid:23060611
  44. 44. Moulos P, Klein J, Jupp S, Stevens R, Bascands J-L, et al. (2013) The KUPNetViz: a biological network viewer for multiple-omics datasets in kidney diseases. BMC Bioinformatics 14: 235. pmid:23883183
  45. 45. Hu LL, Huang T, Shi X, Lu WC, Cai YD, et al. (2011) Predicting functions of proteins in mouse based on weighted protein-protein interaction network and protein hybrid properties. PLoS ONE 6: e14556. pmid:21283518
  46. 46. Ng KL, Ciou JS, Huang CH (2010) Prediction of protein functions based on function-function correlation relations. Computers in Biology and Medicine 40: 300–305. pmid:20089249
  47. 47. Gormen TH, Leiserson CE, Rivest RL, Stein C, editors (1990) Introduction to algorithms: MIT press Cambridge, MA.
    • 48. Craven JBM (2005) Markov networks for detecting overlapping elements in sequence data. The MIT Press. pp. 193.
      • 49. Wang J, Zhang S, Wang Y, Chen L, Zhang XS (2009) Disease-aging network reveals significant roles of aging genes in connecting genetic diseases. PLoS Comput Biol 5: e1000521. pmid:19779549
      • 50. Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ (1990) Basic local alignment search tool. J Mol Biol 215: 403–410. pmid:2231712
      • 51. Sánchez A, Ocaña J, Salicrú M (2010) goProfiles: an R package for the Statistical Analysis of Functional Profiles.
        • 52. Sánchez A, Salicrú M, Ocaña J (2007) Statistical methods for the analysis of high-throughput data based on functional profiles derived from the Gene Ontology. Journal of Statistical Planning and Inference 137: 3975–3989.
        • 53. Huang da W, Sherman BT, Lempicki RA (2009) Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources. Nat Protoc 4: 44–57. pmid:19131956
        • 54. Attari F, Sepehri H, Ansari H, Hassani SN, Esfandiari F, et al. (2014) Efficient Induction of Pluripotency in Primordial Germ Cells by Dual Inhibition of TGF-beta and ERK Signaling Pathways. Stem Cells Dev 23: 1050–1061. pmid:24382167
        • 55. Beyer TA, Weiss A, Khomchuk Y, Huang K, Ogunjimi AA, et al. (2013) Switch enhancers interpret TGF-beta and Hippo signaling to control cell fate in human embryonic stem cells. Cell Rep 5: 1611–1624. pmid:24332857
        • 56. Shivdasani AA, Ingham PW (2003) Regulation of stem cell maintenance and transit amplifying cell proliferation by tgf-beta signaling in Drosophila spermatogenesis. Curr Biol 13: 2065–2072. pmid:14653996
        • 57. Miles DC, Wakeling SI, Stringer JM, van den Bergen JA, Wilhelm D, et al. (2013) Signaling through the TGF beta-activin receptors ALK4/5/7 regulates testis formation and male germ cell development. PLoS One 8: e54606. pmid:23342175
        • 58. Pelton RW, Saxena B, Jones M, Moses HL, Gold LI (1991) Immunohistochemical localization of TGF beta 1, TGF beta 2, and TGF beta 3 in the mouse embryo: expression patterns suggest multiple roles during embryonic development. J Cell Biol 115: 1091–1105. pmid:1955457
        • 59. Wu MY, Hill CS (2009) Tgf-beta superfamily signaling in embryonic development and homeostasis. Dev Cell 16: 329–343. pmid:19289080
        • 60. Tulachan SS, Tei E, Hembree M, Crisera C, Prasadan K, et al. (2007) TGF-beta isoform signaling regulates secondary transition and mesenchymal-induced endocrine development in the embryonic mouse pancreas. Dev Biol 305: 508–521. pmid:17418116
        • 61. Dobashi M, Fujisawa M, Yamazaki T, Okada H, Kamidono S (2002) Distribution of intracellular and extracellular expression of transforming growth factor-beta1 (TGF-beta1) in human testis and their association with spermatogenesis. Asian J Androl 4: 105–109. pmid:12085100
        • 62. Moussad EE, Rageh MA, Wilson AK, Geisert RD, Brigstock DR (2002) Temporal and spatial expression of connective tissue growth factor (CCN2; CTGF) and transforming growth factor beta type 1 (TGF-beta1) at the utero-placental interface during early pregnancy in the pig. Mol Pathol 55: 186–192. pmid:12032230
        • 63. Shooner C, Caron PL, Frechette-Frigon G, Leblanc V, Dery MC, et al. (2005) TGF-beta expression during rat pregnancy and activity on decidual cell survival. Reprod Biol Endocrinol 3: 20. pmid:15927076
        • 64. Pera MF, Andrade J, Houssami S, Reubinoff B, Trounson A, et al. (2004) Regulation of human embryonic stem cell differentiation by BMP-2 and its antagonist noggin. J Cell Sci 117: 1269–1280. pmid:14996946
        • 65. Ben-Shushan E, Feldman E, Reubinoff BE (2014) Notch Signaling Regulates Motor Neuron Differentiation of Human Embryonic Stem Cells. Stem Cells.
          • 66. Wang C, Guo X, Xi R (2014) EGFR and Notch signaling respectively regulate proliferative activity and multiple cell lineage differentiation of Drosophila gastric stem cells. Cell Res 24: 610–627. pmid:24603358
          • 67. Woodhoo A, Alonso MB, Droggiti A, Turmaine M, D'Antonio M, et al. (2009) Notch controls embryonic Schwann cell differentiation, postnatal myelination and adult plasticity. Nat Neurosci 12: 839–847. pmid:19525946
          • 68. Hasegawa K, Okamura Y, Saga Y (2012) Notch signaling in Sertoli cells regulates cyclical gene expression of Hes1 but is dispensable for mouse spermatogenesis. Mol Cell Biol 32: 206–215. pmid:22037762
          • 69. Mori S, Kadokawa Y, Hoshinaga K, Marunouchi T (2003) Sequential activation of Notch family receptors during mouse spermatogenesis. Dev Growth Differ 45: 7–13. pmid:12630942
          • 70. Feng YM, Liang GJ, Pan B, Qin XS, Zhang XF, et al. (2014) Notch pathway regulates female germ cell meiosis progression and early oogenesis events in fetal mouse. Cell Cycle 13: 782–791. pmid:24398584
          • 71. Vachias C, Couderc JL, Grammont M (2010) A two-step Notch-dependant mechanism controls the selection of the polar cell pair in Drosophila oogenesis. Development 137: 2703–2711. pmid:20630949
          • 72. Strazzabosco M (2010) Foxa1 and Foxa2 regulate bile duct development in mice. J Hepatol 52: 765–767. pmid:20347503
          • 73. Besnard V, Wert SE, Kaestner KH, Whitsett JA (2005) Stage-specific regulation of respiratory epithelial cell differentiation by Foxa1. Am J Physiol Lung Cell Mol Physiol 289: L750–759. pmid:16214823
          • 74. Bernardo GM, Lozada KL, Miedler JD, Harburg G, Hewitt SC, et al. (2010) FOXA1 is an essential determinant of ERalpha expression and mammary ductal morphogenesis. Development 137: 2045–2054. pmid:20501593
          • 75. Kofron M, Puck H, Standley H, Wylie C, Old R, et al. (2004) New roles for FoxH1 in patterning the early embryo. Development 131: 5065–5078. pmid:15459100
          • 76. Izzi L, Silvestri C, von Both I, Labbe E, Zakin L, et al. (2007) Foxh1 recruits Gsc to negatively regulate Mixl1 expression during early mouse development. EMBO J 26: 3132–3143. pmid:17568773
          • 77. Boulet AM, Capecchi MR (2012) Signaling by FGF4 and FGF8 is required for axial elongation of the mouse embryo. Dev Biol 371: 235–245. pmid:22954964
          • 78. Naiche LA, Holder N, Lewandoski M (2011) FGF4 and FGF8 comprise the wavefront activity that controls somitogenesis. Proc Natl Acad Sci U S A 108: 4018–4023. pmid:21368122
          • 79. Hasegawa K, Saga Y (2014) FGF8-FGFR1 Signaling Acts as a Niche Factor for Maintaining Undifferentiated Spermatogonia in the Mouse. Biol Reprod.
            • 80. Basson CT, Huang T, Lin RC, Bachinsky DR, Weremowicz S, et al. (1999) Different TBX5 interactions in heart and limb defined by Holt-Oram syndrome mutations. Proc Natl Acad Sci U S A 96: 2919–2924. pmid:10077612
            • 81. Basson CT, Bachinsky DR, Lin RC, Levi T, Elkins JA, et al. (1997) Mutations in human TBX5 [corrected] cause limb and cardiac malformation in Holt-Oram syndrome. Nat Genet 15: 30–35. pmid:8988165
            • 82. Piao J, Tsuji K, Ochi H, Iwata M, Koga D, et al. (2013) Sirt6 regulates postnatal growth plate differentiation and proliferation via Ihh signaling. Sci Rep 3: 3022. pmid:24149372
            • 83. Wang W, Lian N, Ma Y, Li L, Gallant RC, et al. (2012) Chondrocytic Atf4 regulates osteoblast differentiation and function via Ihh. Development 139: 601–611. pmid:22190639
            • 84. Plump AS, Erskine L, Sabatier C, Brose K, Epstein CJ, et al. (2002) Slit1 and Slit2 cooperate to prevent premature midline crossing of retinal axons in the mouse visual system. Neuron 33: 219–232. pmid:11804570
            • 85. Hutson LD, Jurynec MJ, Yeo SY, Okamoto H, Chien CB (2003) Two divergent slit1 genes in zebrafish. Dev Dyn 228: 358–369. pmid:14579375
            • 86. Carpenter D, Ringrose C, Leo V, Morris A, Robinson RL, et al. (2009) The role of CACNA1S in predisposition to malignant hyperthermia. BMC Med Genet 10: 104. pmid:19825159
            • 87. Kawamura S, Ikeda Y, Tomita K, Watanabe N, Seki K (2004) A family of hypokalemic periodic paralysis with CACNA1S gene mutation showing incomplete penetrance in women. Intern Med 43: 218–222. pmid:15098604
            • 88. Li G, Thomas AM, Hart SN, Zhong X, Wu D, et al. (2010) Farnesoid X receptor activation mediates head-to-tail chromatin looping in the Nr0b2 gene encoding small heterodimer partner. Mol Endocrinol 24: 1404–1412. pmid:20444884
            • 89. Park YY, Choi HS, Lee JS (2010) Systems-level analysis of gene expression data revealed NR0B2/SHP as potential tumor suppressor in human liver cancer. Mol Cells 30: 485–491. pmid:20853064
            • 90. Doronin K, Flatt JW, Di Paolo NC, Khare R, Kalyuzhniy O, et al. (2012) Coagulation factor X activates innate immunity to human species C adenovirus. Science 338: 795–798. pmid:23019612