This article is mentioned in:

Introduction

The budding yeast, Saccharomyces cerevisiae, has been used to make bread and beer for thousands of years. As a single-cell organism, budding yeast has been extensively investigated in genetics and physiology as a model system for eukaryotes, due to its well-annotated genome and short life cycle (1). Completion of the budding yeast genome sequencing project helped to determine a total of 6,275 genes on 16 chromosomes (12 million base pairs). Yeast possesses 23% homologous genes to humans; therefore, it is considered as a useful model for gene function studies (2). Although yeast and human diverged from a common ancestor ~1 billion years ago, lines of evidence demonstrate the strong conservation of gene function between yeast and humans (3).

The advancement of cDNA array technology and its low cost make genome-wide gene expression analysis possible. There are many transcriptome data of budding yeast in public databases, including gene expression data from different nutritional conditions, growth stages and gene knock-out models (4). A task for bioinformaticians is to reanalyze these large-scale microarray data and next-generation sequencing data, and identifying the hidden information within these databases. Systems biology studies use high-throughput data and mathematical models to construct yeast transcriptional, gene regulation and protein interaction networks (5,6). The current study used the Google Scholar search engine (https://scholar.google.com/) and, to the best of our knowledge, weighted gene co-expression network analysis (WGCNA) has not been applied to budding yeast. Using WGCNA to construct a yeast gene co-expression network has many advantages. Firstly, individual experiments may lose meaningful biological information due to relatively small sample sizes and different statistical methods. Pooling datasets from studies helps to strengthen statistical power. Secondly, the gene co-expression analysis reduces high-dimensional genome-wide gene expression to only tens of modules, which simplifies the depiction of yeast biology functions. Thirdly, functionally unknown genes are inferred from co-expressed gene modules by the principle of guilt by association (7). Finally, the current analysis may provide a paradigm for single-cell microorganism.

The aim of the present study was to construct a budding yeast gene co-expression network and identify functional modules that may represent different aspects of yeast biological function. It is hypothesized that certain highly connected genes may be involved in module function. The conservation of these modules is further validated in human tumor cells, some of which may differentiate the survival times of separate patients into long and short.

Materials and methods

Microarray data processing

Yeast microarray data were downloaded from Gene Expression Omnibus (GEO) (http://ncbi.nlm.nih.gov/gds) (8). Detailed information of the 218 gene expression datasets used in the current study are available at Online Resource 1 (http://bioinformatics.fafu.edu.cn/Downloads.html). To simplify the data analysis procedure and to enhance reproducibility, only 2,814 budding samples from experiments run on Affymetrix Yeast Genome 2.0 Array (Thermo Fisher Scientific, Inc., Waltham, MA, USA) were included. Datasets from fission yeast was excluded. Expression Console (version 1.2) (Affymetrix, Inc., Santa Clara, CA, USA) was used to process raw data. Probe level expression data were extracted using the MAS5.0 method and normalized using the DNAMR package (http://www.rci.rutgers.edu/~cabrera/DNAMR).

Weighted gene co-expression network construction and module detection

Signed network was constructed according to WGCNA protocols (9–11). The WGCNA R function was implemented using the following parameters: power=12, minModuleSize=30, deepSplit=4, networkType=‘signed’. Briefly, Pearson correlation coefficients were calculated for all pair-wise comparisons of the probes across all samples. The resulting correlation matrix was converted into a matrix of connection strengths (that is, an adjacency matrix) using the power function αij = [0.5+0.5 × cor(xi, xj)]β where xi and xj are the i th and j th probe expression, which resulted in a weighted network. Then topological overlap measure (TOM) was calculated as follows:

Calculation of 1-TOM was performed as a biologically meaningful measure of node similarity, representing how close the neighbors of probe 1 were to the neighbors of probe 2. Probes were hierarchically clustered using 1-TOM as the distance and modules were determined by choosing a height cutoff of 0.995 for the resulting dendrogram. Highly similar modules were identified by clustering and merged together using a dynamic tree-cutting algorithm (7). The module eigengene (ME) corresponds to the first principal component of a given module and is calculated as follows: ME = princomp(xil(q)), where i corresponds to module probes and l represents microarray samples in module q. This may be considered as the most representative probe expression in a module. Module membership (MM or kME) for each probe in each module refers to the Pearson correlation between the expression level of the probe and ME (7). MM was calculated by Kcor,i(q) = cor(xi,E(q)), where xi is the profile of node i and E(q) is the ME of module q. Hub genes were defined as corresponding probes that have high module membership values within a module (11). The stability of each module was evaluated by sampling 1,407 samples from all 2,814 samples 1,000 times. Correlation between connectivity calculated from the sampled samples and the original samples was represented as the mean ± standard deviation.

Gene ontology (GO) and pathway enrichment analysis

GO enrichment and Kyoto Encyclopedia of Genes and Genomes pathway analysis for network modules were performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID 6.7, https://david-d.ncifcrf.gov) with the background list of all genes on the array (12–14). In DAVID, the overrepresentation of a term is defined as a modified Fisher's exact P-value with an adjustment for multiple tests using the Benjamini-Hochberg method (15).

Survival analysis

A human breast cancer dataset, GSE31448, colorectal cancer dataset, GSE17536, and sarcoma dataset, GSE21050 were downloaded from the NCBI GEO database. Raw files were processed using the Affymetrix Expression Console. The yeast module genes were mapped to human genes using the NCBI HomoloGene system (https://www.ncbi.nlm.nih.gov/homologene). To obtain the module expression value MEs, the three disease gene expression datasets were projected to yeast modules according to homologous genes. Patients were then separated into two groups with high and low MEs. Survival analysis was conducted to compare the survival time difference in R using the survival package. P-values for survival curves were determined from the Kaplan-Meier survival curves by use of the log-rank test.

Results

WGCNA

A large budding yeast gene co-expression network was constructed in the present study, to the best of our knowledge, for the first time. WGCNA identified 17 modules, which contained genes with similar expression patterns. To evaluate the stability of modules, gene connectivity was correlated before and after sampling (Fig. 1). Module stability was expressed as the correlation of intramodule connectivity between the original and sampled module. The average correlation was >0.8. Module Cyan was the most stable module, and module Lightyellow was the least stable (Fig. 1). All gene modules and corresponding connectivity data are provided in Online Resource 2 (http://bioinformatics.fafu.edu.cn/Downloads.html).

Figure 1. Correlation of intramodule connectivity for each module after sampling 1,000 times (means ± standard deviation).

Modules are involved in different functions

To identify modular features, DAVID was used to characterize module genes (Table I). All the modules were associated with distinct biological functions, representing aspects of yeast cell function. Module yellow is enriched with genes associated with ncRNA processing, which predominantly localize in the nucleolus. Both modules cyan and green-yellow were associated with mitochondria, while module cyan encodes proteins involved with oxidative phosphorylation on mitochondrial membranes, and module green-yellow encodes mitochondrial ribosome proteins.

Table I. GO and KEGG annotation of the identified 17 gene co-expression modules in Saccharomyces cerevisiae.

Table I.

GO and KEGG annotation of the identified 17 gene co-expression modules in Saccharomyces cerevisiae.

	GO term (Benjamini-Hochberg adjusted P-value)
Module (no. genes)	Biological process	Cellular component	Molecule function	KEGG (Benjamini-Hochberg adjusted P-value)
Black (277)	Response to heat (1E-45)	Plasma membrane (6E-11)	Aldehyde reductase activity (9E-4)	Sucrose metabolism (7E-7)
Blue (935)	Cell cycle phase (8E-10)	Chromosome (1E-8)	Protein kinase activity (5E-2)	Cell cycle (4E-12)
Brown (796)	Regulation of translation (4E-27)	Cytosolic ribosome (1E-80)	Structural constituent of ribosome (5E-39)	Ribosome (5E-66)
Cyan (111)	Oxidative phosphorylation (2E-38)	Mitochondrial (4E-41)	Hydrogen ion transmembrane transporter activity(4E-29)	Oxidative phosphorylation (4E-31)
Dark green (37)	M phase of meiotic cell cycle (2E-22)	Condensed nuclear chromosome (3E-10)	Structure-specific DNA binding (1E-04)	Meiosis (7E-7)
Dark red (49)	Amino acid biosynthesis (2E-23)	Glycine cleavage complex (7E-3)	Oxidoreductase activity, acting on the CH-NH2 group of donors, disulfide as acceptor (3E-2)	Arginine and proline metabolism (3E-3)
Green (736)	Autophagy (1E-19)	Vacuole (5E-4)	Transcription regulator activity (9E-4)
Green yellow (135)	Mitochondrial translation (7E-98)	Mitochondrial ribosome (7-E101)	Structural constituent of ribosome (8E-69)	Aminoacyl-tRNA biosynthesis (8E-7)
Grey 60 (84)	Steroid metabolism (1E-5)	Intrinsic to membrane (5E-7)	O-acyltransferase activity (1E-3)
Light cyan (85)	Ubiquitin-dependent protein metabolism (1E-38)	Proteasome (2E-58)	Threonine-type endopeptidase activity (3E-21)	Proteasome (2E-50)
Light green (187)	Protein amino acid glycosylation (2E-13)	Intrinsic to membrane (4E-34)	Mannosyltransferase activity(1E-3)	N-Glycan biosynthesis (5E-9)
Light yellow (57)	Oligosaccharide metabolic process (3E-4)
Pink (194)	Sporulation (5E-4)
Purple (143)	Organic acid catabolic process (4E-11)	Microbody (3E-17)	Ligase activity, forming carbon-sulfur bonds (2E-2)	Glyoxylate and dicarboxylate metabolism (5E-6)
Royal blue (52)	Sulfur metabolism (2E-25)		Sulfur amino acid transmembrane transporter activity (6E-5)	Sulfur metabolism (5E-10)
Tan (1851)	Monosaccharide catabolic process (3E-2)	Cytosolic ribosome (2E-10)	Structural constituent of ribosome (9E-5)	Ribosome (3E-10)
Yellow (666)	ncRNA processing (1E-100)	Nucleolus (2E-128)	DNA-directed RNA polymerase activity (2E-14)	Pyrimidine metabolism (9E-16)

[i] GO, gene ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Functions of module hub genes

In a network biology, genes do not contribute equally to a module. Those genes with higher connectivity may exert more significant roles in a module. To establish the functional annotation of those genes, the top hub genes were screened to examine their functions (16). For example, module blue was enriched in cell cycle-associated genes. The top hub gene of the module was YDR506C, which contributes to nuclear division and genome integrity (17,18). The results are presented in Table II.

Table II. Hub genes and their encoding proteins of the gene co-expression modules in Saccharomyces cerevisiae.

Table II.

Hub genes and their encoding proteins of the gene co-expression modules in Saccharomyces cerevisiae.

Module	Gene	Encoding protein
Black	DCS2	Protein DCS2
Blue	YDR506C	Putative multicopper oxidase YDR506C
Brown	RPL22A	60S ribosomal protein L22-A
Cyan	SDH2	Succinate dehydrogenase [ubiquinone] iron-sulfur subunit, mitochondrial
Dark green	IME2	Meiosis induction protein kinase IME2/SME1
Dark red	ARG7	Arginine biosynthesis bifunctional protein ARG7, mitochondrial
Green	GPX1	Glutathione peroxidase 1
Green yellow	MRPL49	54S ribosomal protein L49, mitochondrial
Grey 60	PAU5	Seripauperin-5; Seripauperin-7
Light cyan	RPT4	26S protease subunit RPT4
Light green	ALG8	Dolichyl pyrophosphate Glc1Man9GlcNAc2 alpha-1,3-glucosyltransferase
Light yellow	PRP39	Pre-mRNA-processing factor 39
Pink	MER1	Meiotic recombination 1 protein
Purple	FOX2	Peroxisomal hydratase-dehydrogenase-epimerase
Royal blue	MET16	Phosphoadenosinephosphosulfate reductase
Tan	SPBC947.09	Uncharacterized protein C947.09
Yellow	RPF2	Ribosome biogenesis protein RPF2

Module conservation in human cancer cell lines

WGCNA ModulePreservation function was used to evaluate the preservation of the 17 yeast modules in the human cancer cell lines dataset, GSE36133 (Fig. 2). Five modules, yellow, green-yellow, light cyan, cyan and brown were identified to be well preserved from high to low in human cancer cell lines. They are associated with ncRNA processing, mitochondrial translation, ubiquitin-dependent protein metabolism, oxidative phosphorylation and regulation of translation.

Figure 2. Module preservation analysis between bake yeast and human cancer cell lines. Z summary represents the summary preservation statistics, where the y-axis demonstrates the preservation statistics and the x-axis is the gene numbers in each module. The dashed blue and green lines indicate the thresholds Z=2 and Z=10, respectively. Z summary <2 implies no evidence for module preservation, Z summary ≤10 indicates weak to moderate evidence, while Z summary >10 suggests strong evidence for module preservation. The statistics are derived from 1,000 permutations.

Co-expression modules differentiate between patients with different survival times

To further validate the importance of these modules, human breast cancer, colorectal cancer and sarcoma microarray datasets were used to plot survival curves. For example, module blue differentiates between breast cancer patients (Fig. 3A); module royal blue separates colorectal cancer patients (Fig. 3B); and module light green distinguishes sarcoma patients (Fig. 3C).

Figure 3. Identified modules differentiate between patients with different survival/metastasis-free survival time. (A) Breast cancer patients (dataset, GSE31448). (B) Colorectal cancer patients (dataset, GSE17536). (C) Sarcoma patients (dataset, GSE21050). P-values were computed using Log-rank test. Blue line, low ME group; green line, high ME group. ME, module eigengene.

Discussion

WGCNA has been extensively applied to gene co-expression network construction and analysis in various species. For example, in human brain transcriptome analysis, Oldham et al (10) identified the gene co-expression corresponding to different brain regions. In plants, Zhan et al (19) identified cell-type specific gene co-expression modules, and observed regulatory modules that were associated with endosperm cell differentiation. WGCNA was previously used to depict functional gene co-expression modules in mouse liver and human cancer cell lines (20,21). In the present study, a gene network of budding yeast was successfully constructed using WGCNA analysis. All of the identified 17 modules were associated with specific functional categories. As a single cell organism, the results are easier to interpret. Therefore, WGCNA has an advantage over differential gene expression analysis or ANOVA, which compare two or more experimental groups. When there are many different biological groups, it is more complicated to analyze these data. WGCNA surmount these disadvantages, as it simplifies thousands of genes into tens of functional modules. Finally, the method does not require prior knowledge, so novel gene functions may be identified. WGCNA has previously been used as a gene annotation method (22).

The 17 identified modules represent different aspects of budding yeast functions, including substance and energy metabolism, cell proliferation and stimulus response (Table I). Module black contains genes associated with heat response, which is an important trait of yeast function (23). Recent studies indicate that yeast has an adaptation for environmental stress, such as high temperature (24). Substance metabolism modules, include amino acid metabolism (dark red), steroid metabolism (grey 60), organic acid metabolism (purple) and sulfur metabolism (royal blue). Each module has a distinct function, indicating the robustness of WGCNA.

Only 1,944 module genes were projected to human homologous genes due to the limited number of yeast genes on microarray. Thus, there is no definitive conclusion that modules with a low preservation Z summary value are not preserved within humans as a result of fewer genes in these modules (Fig. 2). However, the five preserved modules identified in the present study are consistent with a previous study that demonstrated that genes within these modules are replaceable (3).

The significance of cancer cell line gene co-expression modules in tumors has previously been reported (21). In the current study, yeast modules were observed in various human cancer datasets. For example, certain modules differentiate between patients with long and short survival times, indicating their importance from yeast to humans. Those modules may be crucial in cancer biology and provide information for human tumor research within yeast cells.

Acknowledgements

The present study was supported in part by the National Natural Science Foundation of China (grant nos. 31270454 and 81502091), and of Zhejiang Provincial Natural Science Foundation (grant no. LQ14H030001) and a Ningbo Natural Science Foundation Grant (grant no. 2013A610232).

References