317

Alain Goossens and Laurens Pauwels (eds.), Jasmonate Signaling: Methods and Protocols, Methods in Molecular Biology, vol. 1011, DOI 10.1007/978-1-62703-414-2_25, © Springer Science+Business Media, LLC 2013

Chapter 25

Transcriptome Coexpression Analysis Using ATTED-II for Integrated Transcriptomic/Metabolomic Analysis

Keiko Yonekura-Sakakibara and Kazuki Saito

Abstract

Transcriptome coexpression analysis is an excellent tool for predicting the physiological functions of genes. It is based on the “guilt-by-association” principle. Generally, genes involved in certain metabolic processes are coordinately regulated. In other words, coexpressed genes tend to be involved in common or closely related biological processes. Genes of which the metabolic functions have been identi fi ed are preselected as “guide” genes and are used to check the transcriptome coexpression fi delity to the pathway and to determine the threshold value of correlation coef fi cients to be used for subsequent analysis. The coexpres-sion analysis provides a network of the relationships between “guide” and candidate genes that serves to create the criteria by which gene functions can be predicted. Here we describe a procedure to narrow down the number of candidate genes by means of the publicly available database, designated Arabidopsis thaliana trans -factor and cis -element prediction database (ATTED-II).

Key words Coexpression analysis , ATTED-II , Correlation coef fi cients , Transcriptomics , Metabolomics , Arabidopsis thaliana , Plant

The proliferation of plant genome sequencing projects and the subsequent development of high-throughput technologies, includ-ing DNA microarrays, have generated a massive amount of bio-logical data sets. In an effort to make these rapidly increasing data sets publicly available, functional genomics data repositories were established, such as Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo ) and ArrayExpress ( http://www.ebi.ac.uk/arrayexpress ). As of April 2012, over 700,000 microarray data are available through GEO. As a secondary analytical tool, data-mining informatics provides a manner to develop gene coex-pression databases.

Gene coexpression databases have been established on publicly available microarray data sets. They provide a list of coexpressed

1 Introduction

318 Keiko Yonekura-Sakakibara and Kazuki Saito

genes and the degree of similarity between gene expression patterns, generally by Pearson’s correlation coef fi cients, Spearman’s correlation coef fi cients, or unique calculated scores between any two genes. Various gene coexpression databases for plants are accessible publicly, such as Arabidopsis co-expression tool (ACT) [ 1 ] , Arabidopsis Systems Interaction Database (ASIDB) [ 2 ] , ATTED-II, Botany Array Resource (BAR) Expression Angler [ 3 ] , Co-Expression analysis for Arabidopsis (CressExpress) [ 4 ] , Comprehensive Systems-Biology Database (CSB.DB) [ 5 ] , Gene Co-Expression Analysis Toolbox (GeneCAT) [ 6 ] , and Plant Gene Expression Database (PED) [ 7 ] . Details about these databases, including data sources, calculation methods, and data retrieval tools, are available [ 8 ] .

Transcriptome coexpression analysis is based on the hypothesis that genes in the same and/or nearby pathway are regulated in a coordinated manner. In other words, coexpressed genes tend to contribute to common or closely related biological processes. Genes of which the expression patterns are highly similar to that of genes with determined functions, the so-called guide genes, can be selected as potential target genes because the probability of their involvement in a common or related pathway is high (Fig. 1 ). The degree of gene coexpression similarity is measured by correlation coef fi cients (Table 1 ). So far, genes encoding enzymes, transcrip-tion factors, and complex-forming proteins have been identi fi ed based on transcriptome coexpression analysis [ 8 ] . Coordinate expression is especially pronounced in plant secondary metabolism. By means of ATTED-II, fl avonoid modi fi cation enzymes have been determined functionally from among 107 candidates and

known“guide”genes: transcription factors: enzymes

etc.

candidate genesGene families,homologues and etc.

CorrelatedGene Search

Based onATTED-II

CorrelatedGene Search

Based onATTED-II

1st-round analysis 2nd-round analysis

Fig. 1 Basic concept of transcriptome coexpression analysis. Genes that have been identi fi ed functionally are used as known “guide” genes. In a fi rst-round analysis, the threshold values are determined for the second-round analysis that is conducted to search candidate target genes. Positive correlations ( r > de fi ned value) are indicated by connecting lines

319ATTED-II-Based Transcriptome Coexpression Analysis

MYB and biosynthetic enzymes in the glucosinolate pathway as well [ 9– 11 ] . Many examples of coexpression-based identi fi cation with ATTED-II are shown on the ATTED-II homepage ( http://atted.jp/top_publication.shtml ). Here, we describe how to assess the effectiveness of transcriptome coexpression analysis for a given metabolic pathway and to narrow a fi eld of candidate genes by using ATTED-II [ 12– 14 ] .

1. ATTED-II ( http://atted.jp/ ). 2. PRIMe, Correlated Gene Search ( http://prime.psc.riken.

jp/?action=coexpression_index ).

1. Pajek ( http://vlado.fmf.uni-lj.si/pub/networks/pajek/ ). 2. BioLayout ( http://www.biolayout.org/ ). 3. Cytoscape ( http://www.cytoscape.org/ ).

ATTED-II can be used to obtain a quick overview of gene coex-pression for a given biological process.

1. Open ATTED-II ( http://atted.jp/ ). 2. Enter keyword, GO ID, gene alias, or gene ID and press the

search button ( see Note 1 ). Functional categories and/or loci matching the search are shown.

3. In the Gene Ontology (GO) term results, click the hyperlinks “list” ( see Note 2 ) and/or “network” ( see Note 3 , Fig. 2 ) for further information.

2 Materials

2.1 Software for Transcriptome Coexpression Analysis

2.2 Software for Analysis and Visualization of Networks

3 Methods

3.1 Coexpression Analysis to Generate Hypothesis

Table 1 Example of a general interpretation of Pearson’s correlation coef fi cients [ 15 ]

Pearson’s correlation coef fi cients, r Degree of correlation a

| r | = 1.0 Perfect

1 > | r | ≥ 0.7 Strong

0.7 > | r | ≥ 0.3 Moderate

0.3 > | r | > 0 Weak

| r | = 0 None

a The range may vary slightly according to the references

320 Keiko Yonekura-Sakakibara and Kazuki Saito

4. Within the obtained Locus search results, click the “locus” hyperlink to get more details regarding functional annotations, gene coexpression, gene expression, and predicted cis -elements for genes of interest. If there are coexpressed gene networks, transcriptome coexpression analysis may be applicable ( see Subheading 3.2 ).

1. Collect “guide” genes. These are functionally identi fi ed genes known to be in the pathway of interest. Having more guide genes may increase the reliability of the pool of candidate genes. As examples, genes involved in fl avonol or jasmonate biosynthesis are listed in Table 2 .

2. Check (1) the presence of probes corresponding to guide genes on the microarray chip and (2) the possibility of cross hybrid-ization with these guide genes ( see Note 4 ).

3.2 Transcriptome Coexpression Analysis to Narrow Down Candidate Genes Involved in the Pathway of Interest

Fig. 2 ATTED-II Network and gene list output of the GO term jasmonic acid metabolic process (GO:0009694) as of August 2012

321ATTED-II-Based Transcriptome Coexpression Analysis

3. Remove the genes from the list if either the probes corresponding to the guide genes are absent on the microar-ray chip or cross hybridization to the guide genes is highly possible.

4. Open “Correlated Gene Search” in PRIMe ( http://prime.psc.riken.jp/?action=coexpression_index ).

5. Enter the Arabidopsis Genome Initiative (AGI) codes of guide genes in the Locus ID column.

6. Set parameters, such as Matrix, Methods, and Threshold value, and press Search button ( see Note 5 ).

7. To check all correlation coef fi cient values between guide genes, set Matrix at “data sets v.3,” Methods at “interconnection of sets,” and Threshold value to −1 ( see Note 6 ).

8. Set Format to HTML. 9. An Order and Display limit can be selected. 10. Check all correlation coef fi cient values between guide genes

and determine the threshold value used for further analysis ( see Notes 7 – 9 ).

Table 2 “Guide” genes in fl avonol and jasmonate metabolism

Function Abbreviation AGI

Flavonol metabolism

4-Coumarate:CoA ligase 4CL3 At1g65060

Chalcone synthase CHS At5g13930

Chalcone isomerase CHI At3g55120

Flavanone 3-hydroxylase F3H At3g51240

Flavonoid 3 ′ -hydroxylase F3 H At5g07990

Flavonol synthase FLS At5g08640

Jasmonate metabolism

Lipoxygenasea LOX3 At1g17420

Lipoxygenasea LOX4 At1g72520

Allene oxide synthasea AOS At5g42650

Allene oxide cyclasea AOC3 At3g25780

Oxophytodienoic acid reductasea OPR3 At2g06050

OPC-8:0 CoA ligase1a OPCL1 At1g20510

a Genes linked to the Kyoto Encyclopedia of Genes and Genomes (KEGG) map in the gene list of GO:0009694 (jasmonic acid metabolic process) in ATTED-II (Fig. 2 ) were used as “guide” genes for jasmonate metabolism

322 Keiko Yonekura-Sakakibara and Kazuki Saito

11. Prepare a list of the candidate genes. Gene families, paralogs, homologs, and/or pathway-related genes are frequently used as candidate genes ( see Notes 10 and 11 ).

12. Enter the combined list of AGI codes of the guide and candidate genes in the Locus ID column of the Correlated Gene Search. Set the threshold value of correlation coef fi cients determined above using the guide genes ( see Subheading 3.2 , step 10 ).

13. Set other parameters and press the Search button. “Interconnection of sets” should be chosen in Methods, and the same Matrix data sets should be used as when the threshold value was chosen ( see Note 8 ). As examples, the networks using guide genes in fl avonol metabolism (Table 2 ) and 120 family-1 glycosyltransferase genes as candidate genes (Matrix: all data set v.3 (1388 data), Methods: interconnection of sets, thresh-old value 0.667) and guide genes in jasmonate metabolism (Table 2 ) and all Arabidopsis genes (Matrix: all data set v.3 (1388 data), Methods: union of sets, threshold value 0.669) are shown in Fig. 3 .

14. Look at an alternative perspective of candidate genes, i.e., gene annotations, primary sequences, gene expression pro fi les, etc. ( see Note 12 ).

15. Select target genes for further analysis ( see Note 13 ).

1. If there is no hit with a GO term search, check the word used in GO released by The Arabidopsis Information Resource (TAIR) that was used in ATTED. For example, input of “jasmonic acid” should yield nine functional categories and 34 loci and input of “jasmonate” should yield one functional cat-egory and 39 loci.

2. “List” shows the list of all the genes with the GO term. In addition, coexpressed genes among the selected genes can be searched by selecting them and pressing “search coexpressed genes.” The identical network shown in “network” ( see Note 3 ) and unconnected genes are shown.

3. “Network” shows the network of all coexpressed genes in the GO term and the list of genes in a coexpressed gene network. Unconnected genes are omitted from the network. As exam-ple, network of “GO:0009694” (jasmonic acid metabolic pro-cess) is shown in Fig. 2 .

4. Not all genes are represented on the ATH1 microarray. The possibility of cross hybridization of guide genes can be checked in the Affymetrix NetAffx™ Analysis Center ( http://www.affymetrix.com/analysis/index.affx ). If the guide genes cross-hybridize with other genes, their expression patterns are not

4 Notes

323ATTED-II-Based Transcriptome Coexpression Analysis

4CL3 At1g65060

At1g06000

F3H At3g51240

At5g17050

FLS At5g08640F3’H At5g07990

CHIAt3g55120

CHS At5g13930

At4g15480

At1g32640

At3g09830At3g44860At4g14680At4g36500At5g47220At3g51450

At1g17380 At3g25760

At3g23250

At2g22880

At3g01830

At5g13220

At3g02840At3g16860At3g44260At3g50930At3g55980At3g57450

At2g21120At2g22500At2g32140At2g34600At2g46150

At1g29690At1g30135At1g56060At1g61890

At2g44840At4g17230At4g31800At4g34410At5g22250At5g66210

At1g28370At1g28380At1g72450At1g74950At1g80840At2g26530

At1g19180At1g27770

At5g13190At5g41740At5g42380At5g59550At5g64660At5g64870

LOX4 At1g72520

LOX3 At1g17420

AOC3 At3g25780

OPR3 At2g06050 AOS At5g42650

OPLC1 At1g20510

At4g24380At4g24570At4g29780At4g30210At4g34150At4g34390

a

b

Fig. 3 Coexpression relationships of genes in fl avonol and jasmonate synthesis pathways. White and black circles indicate “guide” and candidate genes, respectively. ( a ) Genes in fl avonol metabolism (Table 2 ) and 120 family-1 glycosyltransferase genes [ 19 ] are used for analysis with Matrix (“all data set v.3 (1388 data)”), Methods (“interconnection of sets”), and threshold value (0.667). The genes surrounded by circles were identi fi ed as fl avonoid 3- O -glucosyltransferase (At5g17050) and fl avonol 7- O -rhamnosyltransferase (At1g06000) based on transcriptomics and transcriptome coexpression analyses [ 4, 20 ] . ( b ) Genes in jas-monate metabolism (Table 2 ) are used with Matrix (“all data set v.3 (1388 data)”), Methods (“union of sets”), and threshold value (0.669). The genes surrounded by circles were allene oxide cyclase ( AOC1 , At3g25760), MYC2 (At1g32640), and jasmonate ZIM-domain (JAZ) proteins ( JAZ1 , At1g19180; JAZ2 , At1g74950; JAZ5 , At1g17380; JAZ6 , At1g72450; JAZ7 , At2g34600; JAZ8 , At1g30135; and JAZ10 , At5g13220)

324 Keiko Yonekura-Sakakibara and Kazuki Saito

suitable for analysis. If needed, the contribution of each gene can be estimated by tiling data through TileViz ( http://jsp.weigelworld.org/tileviz/tileviz.jsp ).

5. A matrix can be chosen among “All data sets v.3 (1388 data),” “All data sets v.1 (771 data),” “Hormone treatments v.1 (236 data),” “Tissue and development v.1 (237 data),” and “Stress treatments v.1 (298data).” At this point, the best Matrix is determined by trial and error.

6. “Interconnection of sets” reports on correlated gene pairs amongst the queried genes only.

7. Normally, the fi rst choice is the lowest correlation coef fi cient value that is enough to minimally connect all guide genes. General statistical descriptions about relationship and correlation coef fi cient values are shown in Table 1 [ 15 ] . Biologically signi fi cant relationships are expected to be above a threshold value from 0.55 to 0.66 [ 16 ] . The lowest values that minimally connect all guide genes involved in fl avonol and jasmonate metabolism (Table 2 ) are 0.667 and 0.669, respectively. These values will be used for later analyses. If the lowest correlation coef fi cient value that is enough to minimally connect all guide genes is quite lower that the above values, omit some of the caus-ative guide genes that are not as likely to be coregulated.

8. The threshold value should be adjusted as the coexpression network is consistent with a known regulatory system for the pathway of interest. The threshold value is dependent on the metabolic pathways and Matrix (data set) used. If gene coex-pression databases were used with smaller data sets, the least correlations considered statistically signi fi cant, depending on sample size, should be taken into account [ 17 ] . If all correla-tion coef fi cient values between guide genes are too low, either the Matrix (data set) used for analysis is unsuitable for the pathway of interest or the target pathway is not regulated at the transcriptional level.

9. To understand the relationships between genes at a glance, results can be saved and visualized with network visualization programs, such as Pajek ( http://pajek.imfm.si/doku.php ) and BioLayout ( http://www.biolayout.org/ ). Instructions on how to use Pajek are given in http://vlado.fmf.uni-lj.si/pub/networks/pajek/howto.htm . Also useful is Cytoscape ( http://www.cytoscape.org/ ).

10. For a nontargeted analysis, the entire Arabidopsis genes can be used as the candidate list ( see Note 11 ).

11. To use all Arabidopsis genes as the candidates, input AGI codes of guide genes only in Locus IDs column and set Methods at “union of sets.” The latter method searches for all genes cor-related with any of the queried genes.

325ATTED-II-Based Transcriptome Coexpression Analysis

12. In the case of a nontargeted gene search, check the full locus detail information, especially the publication, in TAIR. Annotations released by TAIR are shortened in ATTED-II.

13. For proof of concept of transcriptome coexpression analyses for functional genomics and pitfalls and limitations of this method, see Saito et al. [ 18 ] .

Acknowledgments

We would like to thank Drs. A. Fukushima and Y. Sasaki-Sekimoto for their helpful comments and Dr. T. Obayashi for kind permis-sion of the use of fi gures on ATTED-II.

References

1. Man fi eld IW, Jen C-H, Pinney JW, Michalopoulos I, Bradford JR, Gilmartin PM, Westhead DR (2006) Arabidopsis Co-expression Tool (ACT): web server tools for microarray-based gene expression analysis. Nucleic Acids Res 34:W504–W509

2. Rawat A, Seifert GJ, Deng Y (2008) Novel implementation of conditional co-regulation by graph theory to derive co-expressed genes from microarray data. BMC Bioinformatics 9:S7

3. Tou fi ghi K, Brady SM, Austin R, Ly E, Provart NJ (2005) The Botany Array Resource: e-Northerns, Expression Angling, and pro-moter analyses. Plant J 43:153–163

4. Srinivasasainagendra V, Page GP, Mehta T, Coulibaly I, Loraine AE (2008) CressExpress: a tool for large-scale mining of expression data from Arabidopsis. Plant Physiol 147:1004–1016

5. Steinhauser D, Usadel B, Luedemann A, Thimm O, Kopka J (2004) CSB.DB: a com-prehensive systems-biology database. Bioinformatics 20:3647–3651

6. Mutwil M, Øbro J, Willats WGT, Persson S (2008) GeneCAT–novel webtools that com-bine BLAST and co-expression analyses. Nucleic Acids Res 36:W320–W326

7. Horan K, Jang C, Bailey-Serres J, Mittler R, Shelton C, Harper JF, Zhu J-K, Cushman JC, Gollery M, Girke T (2008) Annotating genes of known and unknown function by large-scale coexpression analysis. Plant Physiol 147:41–57

8. Usadel B, Obayashi T, Mutwil M, Giorgi FM, Bassel GW, Tanimoto M, Chow A, Steinhauser D, Persson S, Provart NJ (2009) Co-expression tools for plant biology: opportunities for

hypothesis generation and caveats. Plant Cell Environ 32:1633–1651

9. Hirai MY, Sugiyama K, Sawada Y, Tohge T, Obayashi T, Suzuki A, Araki R, Sakurai N, Suzuki H, Aoki K, Goda H, Nishizawa OI, Shibata D, Saito K (2007) Omics-based identi fi cation of Arabidopsis Myb transcription factors regulating aliphatic glucosinolate bio-synthesis. Proc Natl Acad Sci USA 104:6478–6483

10. Yonekura-Sakakibara K, Tohge T, Matsuda F, Nakabayashi R, Takayama H, Niida R, Watanabe-Takahashi A, Inoue E, Saito K (2008) Comprehensive fl avonol pro fi ling and transcriptome coexpression analysis leading to decoding gene-metabolite correlations in Arabidopsis . Plant Cell 20:2160–2176

11. Yonekura-Sakakibara K, Tohge T, Niida R, Saito K (2007) Identi fi cation of a fl avonol 7- O -rhamnosyltransferase gene determining fl avonoid pattern in Arabidopsis by transcrip-tome coexpression analysis and reverse genet-ics. J Biol Chem 282:14932–14941

12. Obayashi T, Hayashi S, Saeki M, Ohta H, Kinoshita K (2009) ATTED-II provides coex-pressed gene networks for Arabidopsis. Nucleic Acids Res 37:D987–D991

13. Obayashi T, Kinoshita K, Nakai K, Shibaoka M, Hayashi S, Saeki M, Shibata D, Saito K, Ohta H (2007) ATTED-II: a database of co-expressed genes and cis elements for identify-ing co-regulated gene groups in Arabidopsis . Nucleic Acids Res 35:D863–D869

14. Obayashi T, Nishida K, Kasahara K, Kinoshita K (2011) ATTED-II updates: condition-speci fi c gene coexpression to extend coexpression analy-ses and applications to a broad range of fl owering plants. Plant Cell Physiol 52:213–219

326 Keiko Yonekura-Sakakibara and Kazuki Saito

15. Jackson SL (2011) Correlational methods and statistics. In: Jackson SL (ed) Research Methods and Statistics : A Critical Thinking Approach . Belmont, CA, Wadsworth, pp 147–170

16. Aoki K, Ogata Y, Shibata D (2007) Approaches for extracting practical information from gene co-expression networks in plant biology. Plant Cell Physiol 48:381–390

17. Frey B (2006) Discovering relationships. In: Frey B (ed) Statistical hacks: tips and tools for measuring the world and beating the Odds. O’Reilly Media, Sebastopol, CA, pp 41–95

18. Saito K, Hirai MY, Yonekura-Sakakibara K (2008) Decoding genes with coexpression

networks and metabolomics—“majority report by precogs”. Trends Plant Sci 13:36–43

19. Paquette S, Møller BL, Bak S (2003) On the origin of family 1 plant glycosyltransferases. Phytochemistry 62:399–413

20. Tohge T, Nishiyama Y, Hirai MY, Yano M, J-i N, Awazuhara M, Inoue E, Takahashi H, Goodenowe DB, Kitayama M, Noji M, Yamazaki M, Saito K (2005) Functional genomics by integrated analysis of metabo-lome and transcriptome of Arabidopsis plants over-expressing an MYB transcription factor. Plant J 42:218–235