The AtGenExpress hormone and chemical treatment data set: experimental design, data evaluation, model data analysis and data access

Authors


*(fax +81 45 503 9492; e-mail shimada@postman.riken.go.jp).

Summary

We analyzed global gene expression in Arabidopsis in response to various hormones and in related experiments as part of the AtGenExpress project. The experimental agents included seven basic phytohormones (auxin, cytokinin, gibberellin, brassinosteroid, abscisic acid, jasmonate and ethylene) and their inhibitors. In addition, gene expression was investigated in hormone-related mutants and during seed germination and sulfate starvation. Hormone-inducible genes were identified from the hormone response data. The effects of each hormone and the relevance of the gene lists were verified by comparing expression profiles for the hormone treatments and related experiments using Pearson’s correlation coefficient. This approach was also used to analyze the relationships among expression profiles for hormone responses and those included in the AtGenExpress stress-response data set. The expected correlations were observed, indicating that this approach is useful to monitor the hormonal status in the stress-related samples. Global interactions among hormones-inducible genes were analyzed in a pairwise fashion, and several known and novel hormone interactions were detected. Genome-wide transcriptional gene-to-gene correlations, analyzed by hierarchical cluster analysis (HCA), indicated that our data set is useful for identification of clusters of co-expressed genes, and to predict the functions of unknown genes, even if a gene’s function is not directly related to the experiments included in AtGenExpress. Our data are available online from AtGenExpressJapan; the results of genome-wide HCA are available from PRIMe. The data set presented here will be a versatile resource for future hormone studies, and constitutes a reference for genome-wide gene expression in Arabidopsis.

Introduction

The sequencing of the Arabidopsis genome (Arabidopsis Genome Initiative, 2000) is the first step toward understanding its function. Genomic sequencing and the collection of large numbers of cDNAs have enabled researchers to predict genome-wide genetic structure and to design DNA microarrays to monitor genome-wide gene expression. The Affymetrix ATH1 array, an oligonucleotide-based DNA microarray consisting of 22 746 probe sets that cover approximately 23 700 genes (nearly the entire Arabidopsis genome) was designed based on computer-predicted genes (Redman et al., 2004). This system has been widely accepted by the plant science community because of its high sensitivity and reproducibility. Using this system, an international research effort, the AtGenExpress consortium, was organized to enhance the knowledge of gene function in Arabidopsis. Large-scale transcriptome data sets developed by the AtGenExpress consortium have already detailed the developmental process (Schmid et al., 2005) and stress responses (Kilian et al., 2007) in Arabidopsis. Here, we present a hormone-response data set based on treatment with phytohormones and hormone-related inhibitors, and analysis of hormone-related mutants, seed germination and sulfate starvation.

Seven major plant hormones play central roles in the regulation of plant growth, development and the stress response: auxin, cytokinin (CK), gibberellin (GA), brassinosteroid (BR), abscisic acid (ABA), jasmonate (JA) and ethylene. Numerous studies have investigated the functions of each hormone at the physiological and molecular levels; however, no study has considered all seven hormones at the same time. We conducted a comprehensive analysis of the plant responses to all seven phytohormones using defined conditions. Our results will enable members of the plant science community to compare hormonal effects and to study the complex interactions among hormone networks. In fact, the data presented here have already been used to analyze hormone-induced expression profiles (Nemhauser et al., 2006). We describe our experimental design and data validation method, how to access the data, and the results of a model analysis. In the model analysis, interactions among hormone-inducible genes were analyzed in a pairwise manner with respect to their time dependence and direction (up or down). We also demonstrate that the hormone-inducible genes of this data set may be used to monitor hormone status in an experiment performed using the ATH1 GeneChip.

Together with the other contributions from the AtGenExpress project, our data set forms a large-scale transcriptome database. One purpose of this study was to establish a comprehensive database to facilitate searching for gene expression patterns; however, another and more ambitious motivation was to analyze genome-wide co-expression in Arabidopsis so that the functions of unknown genes can be predicted based on similarities between their expression patterns and those of known genes. Thus, we performed a genome-wide co-expression analysis using HCA (Eisen et al., 1998) of data from the AtGenExpress project to demonstrate the usefulness of large-scale transcriptome data sets.

Results and discussion

Experimental design and data overview

The AtGenExpress data set presented here (referred to hereafter as the hormone series) includes treatment with seven phytohormones, their inhibitors, hormone-related mutants, and abiotic treatments, including seed imbibition and sulfate starvation. The data were collected in five laboratories at the RIKEN Plant Science Center and by colleagues outside RIKEN. All hormones and inhibitors used are summarized in Table 1. Seven hormones were applied to wild-type seedlings. Indole-3-acetic acid (IAA) was used as auxin, trans-zeatin was used as cytokinin, and 1-aminocycropropane-1-carboxylic acid (ACC) was used in place of ethylene, unless otherwise noted. In addition to the experiments on wild-type seedlings, CK was applied to an ARR22-overexpressing line (Kiba et al., 2004), GA3 was applied to GA-deficient ga1-5 seedlings, and GA4 was applied to ga1-3 seeds (Koornneef and Van Der Veen, 1980). In addition, ABA was applied to germinating wild-type seeds, and brassinolide (BL) was applied to BR-deficient det2-1 seedlings (Li et al., 1996); BR precursors were also applied to det2 seedlings. Moreover, wild-type seedlings were exposed to inhibitors of GA, BR, auxin and ethylene, other plant growth inhibitors, as well as to salicylic acid (Table 1). BR-, GA-, and CK-related mutants were also analyzed (Table 2). Germination was analyzed using imbibed seeds to study hormonal function at this developmental stage (Table 3). The effect of temperature on seed germination was also assessed (Table 4). Seedlings were transferred to sulfate-deficient medium to test sulfate-regulated gene expression (Table 4). Each experiment was performed twice or three times as biological replicates.

Table 1.   AtGenExpress: hormone and inhibitor treatments
No.ExperimentLaboratoryControlGenotypeTreatmentTime point Organ/tissueGrowth conditionAgeReplicatesAbbreviation
  1. aArabidopsis thaliana seedlings (genotypes Col-0, ga1-5, or det2-1) were grown in half-strength MS liquid medium for 7 days at 23°C and treated with IAA, zeatin, GA3, ABA, MJ, ACC or BL for 30 min, 1 h or 3 h.

  2. bArabidopsis thaliana seedlings (det2-1) were grown in half-strength MS liquid medium for 7 days at 23°C and treated with campestanol, 6-deoxocathasterone, cathasterone, 6-deoxoteasterone, teasterone, 3-dehydro-6-deoxoteasterone, 3-dehydroteasterone, 6-deoxotyphasterol, typhasterol, 6-deoxocastasterone, castasterone or brassinolide for 3 h.

  3. cArabidopsis thaliana seedlings (Col-0 or ARR22-ox) were grown on MS agar plates for 21 days at 22°C and treated with t-zeatin for 3 h.

  4. dDry seeds of Arabidopsis thaliana (Col-0) were imbibed in water or 3 μm ABA for 24 h at 22°C under continuous light.

  5. eSeeds of Arabidopsis thaliana (ga1-3) were imbibed at 4°C in the dark for 48 h and then incubated for 24 h under white light at 22°C. The seeds were then incubated under the same conditions with 5 μm GA4 or water for 3, 6 or 9 h.

  6. fArabidopsis thaliana seedlings (Col-0) were grown in half-strength MS liquid medium for 7 days at 23°C and treated with propiconazole, uniconazole, paclobutrazol or prohexadione for 3 or 12 h.

  7. gArabidopsis thaliana seedlings (Col-0) were grown in half-strength MS liquid medium for 7 days at 23°C and treated with 2,4,6-trihydroxybenzamide, p-chlorophenoxyisobutyric acid, 2,3,5-triiodobenzoic acid or naphthylphthalamic acid for 3 h.

  8. hArabidopsis thaliana seedlings (Col-0) were grown in half-strength MS liquid medium for 7 days at 23°C and treated with brassinazole 220 or brassinazole 91 for 3 or 12 h.

  9. iArabidopsis thaliana seedlings (Col-0) were grown in half-strength MS liquid medium for 7 days at 23°C and treated with AgNO3 or aminoethoxyvinylglycine for 3 h.

  10. jArabidopsis thaliana seedlings (Col-0) were grown in half-strength MS liquid medium for 7 days at 23°C and treated with CHX (cycloheximide), MG132 (carbobenzoxyl-leucinyl-leucinyl-leucinal), PNO8 (N-octyl-3-nitro-2,4,6-trihydroxybenzamide), ibuprofen, daminozide or salicylic acid for 3 or 12 h.

Basic hormone treatment of seedlingsa
1AShimada & YoshidaTrueCol-0Mock30 minSeedlingMS liquid7 days2Mock 30m (A1)
2AShimada & Yoshida Col-01 μm IAA30 minSeedlingMS liquid7 days2IAA 30m (A1)
3AShimada & Yoshida Col-01 μm zeatin30 minSeedlingMS liquid7 days2Zeatin 30m (A1)
4AShimada & Yoshida Col-01 μm GA330 minSeedlingMS liquid7 days2GA 30m (A1)
5AShimada & Yoshida Col-010 μm ABA30 minSeedlingMS liquid7 days2ABA 30m (A1)
6AShimada & Yoshida Col-010 μm MJ30 minSeedlingMS liquid7 days2MJ 30m (A1)
7AShimada & Yoshida Col-010 μm ACC30 minSeedlingMS liquid7 days2ACC 30m (A1)
8AShimada & Yoshida Col-010 nm BL30 minSeedlingMS liquid7 days2BL 30m (A1)
9AShimada & YoshidaTrueCol-0Mock1 hSeedlingMS liquid7 days2Mock 1h (A2)
10AShimada & Yoshida Col-01 μm IAA1 hSeedlingMS liquid7 days2IAA 1h (A2)
11AShimada & Yoshida Col-01 μm zeatin1 hSeedlingMS liquid7 days2Zeatin 1h (A2)
12AShimada & Yoshida Col-01 μm GA31 hSeedlingMS liquid7 days2GA 1h (A2)
13AShimada & Yoshida Col-010 μm ABA1 hSeedlingMS liquid7 days2ABA 1h (A2)
14AShimada & Yoshida Col-010 μm MJ1 hSeedlingMS liquid7 days2MJ 1h (A2)
15AShimada & Yoshida Col-010 μm ACC1 hSeedlingMS liquid7 days2ACC 1h (A2)
16AShimada & Yoshida Col-010 nm BL1 hSeedlingMS liquid7 days2BL 1h (A2)
17AShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h (A3)
18AShimada & Yoshida Col-01 μm IAA3 hSeedlingMS liquid7 days2IAA 3h (A3)
19AShimada & Yoshida Col-01 μm zeatin3 hSeedlingMS liquid7 days2Zeatin 3h (A3)
20AShimada & Yoshida Col-01 μm GA33 hSeedlingMS liquid7 days2GA 3h (A3)
21AShimada & Yoshida Col-010 μm ABA3 hSeedlingMS liquid7 days2ABA 3h (A3)
22AShimada & Yoshida Col-010 μm MJ3 hSeedlingMS liquid7 days2MJ 3h (A3)
23AShimada & Yoshida Col-010 μm ACC3 hSeedlingMS liquid7 days2ACC 3h (A3)
24AShimada & Yoshida Col-010 nm BL3 hSeedlingMS liquid7 days2BL 3h (A3)
25AShimada & YoshidaTruega1-5Mock30 minSeedlingMS liquid7 days2ga1mock 30m (A4)
26AShimada & Yoshida ga1-51 μm GA330 minSeedlingMS liquid7 days2ga1 + GA 30m (A4)
27AShimada & YoshidaTruega1-5Mock1 hSeedlingMS liquid7 days2ga1mock 1h (A5)
28AShimada & Yoshida ga1-51 μm GA31 hSeedlingMS liquid7 days2ga1 + GA 1h (A5)
29AShimada & YoshidaTruega1-5Mock3 hSeedlingMS liquid7 days2ga1mock 3h (A6)
30AShimada & Yoshida ga1-51 μm GA33 hSeedlingMS liquid7 days2ga1 + GA 3h (A6)
31AShimada & YoshidaTruedet2-1Mock30 minSeedlingMS liquid7 days2det2mock 30m (A7)
32AShimada & Yoshida det2-110 nm BL30 minSeedlingMS liquid7 days2det2 + BL 30m (A7)
33AShimada & YoshidaTruedet2-1Mock1 hSeedlingMS liquid7 days2det2mock 1h (A8)
34AShimada & Yoshida det2-110 nm BL1 hSeedlingMS liquid7 days2det2 + BL 1h (A8)
35AShimada & YoshidaTruedet2-1Mock3 hSeedlingMS liquid7 days2det2mock 3h (A9)
36AShimada & Yoshida det2-110 nm BL3 hSeedlingMS liquid7 days2det2 + BL 3h (A9)
Brassinosteroid treatment of seedlingsb
37BShimada & YoshidaTruedet2-1Mock3 hSeedlingMS liquid7 days2mock 3h (B)
38BShimada & Yoshida det2-110 μm campestanol3 hSeedlingMS liquid7 days2CN 3h (B)
39BShimada & Yoshida det2-11 μm 6-deoxocathasterone3 hSeedlingMS liquid7 days26DCT 3h (B)
40BShimada & Yoshida det2-11 μm cathasterone3 hSeedlingMS liquid7 days2CT 3h (B)
41BShimada & Yoshida det2-11 μm 6-deoxoteasterone3 hSeedlingMS liquid7 days26DTE 3h (B)
42BShimada & Yoshida det2-11 μm teasterone3 hSeedlingMS liquid7 days2TE 3h (B)
43BShimada & Yoshida det2-11 μm 3-dehydro-6-deoxoteasterone3 hSeedlingMS liquid7 days26D3DT 3h (B)
44BShimada & Yoshida det2-11 μm 3-dehydroteasterone3 hSeedlingMS liquid7 days23DT 3h (B)
45BShimada & Yoshida det2-11 μm 6-deoxotyphasterol3 hSeedlingMS liquid7 days26DTY 3h (B)
46BShimada & Yoshida det2-11 μm typhasterol3 hSeedlingMS liquid7 days2TY 3h (B)
47BShimada & Yoshida det2-11 μm 6-deoxocastasterone3 hSeedlingMS liquid7 days26DCS 3h (B)
48BShimada & Yoshida det2-1100 nm castasterone3 hSeedlingMS liquid7 days2CS 3h (B)
49BShimada & Yoshida det2-110 nm brassinolide3 hSeedlingMS liquid7 days2BL 3h (B)
Cytokinin treatment of seedlingsc
50CSakakibara & MizunoTrueCol-0No treatment0 hSeedlingMS agar21 days3Cont (C1)
51CSakakibara & Mizuno Col-020 μmt-zeatin 3 hSeedlingMS agar21 days3Zeatin20 μm (C1)
52CSakakibara & MizunoTrueARR22-oxNo treatment0 hSeedlingMS agar21 days3ARR22-ox (C2)
53CSakakibara & Mizuno ARR22-ox20 μmt-zeatin3 hSeedlingMS agar21 days3ARR22-ox+Zeatin20 μm (C2)
Effect of ABA during seed imbibitiond
54DNambara & KamiyaTrueCol-0Dry seeds0 hSeedsN/A0 day2Dry seed (D)
55DNambara & KamiyaTrueCol-0Imbibed seeds (water)24 hSeedsN/A0 day2Imbibed 24h (D2)
56DNambara & Kamiya Col-0Imbibed seeds (3 μm ABA)24 hSeedsN/A0 day2Imbibed 24h+ABA3 μm (D2)
57DNambara & Kamiya Col-0Imbibed seeds (30 μm ABA)24 hSeedsN/A0 day2Imbibed 24h+ABA30 μm (D2)
Effect of gibberellin during seed imbibitione
58EYamaguchi & KamiyaTruega1-3Mock3 hSeedsWater24 h2ga1-3 imbibed 3h (E1)
59EYamaguchi & KamiyaTruega1-3Mock6 hSeedsWater24 h2ga1-3 imbibed 6h (E2)
60EYamaguchi & KamiyaTruega1-3Mock9 hSeedsWater24 h2ga1-3 imbibed 9h (E3)
61EYamaguchi & Kamiya ga1-35 μm GA43 hSeedsWater24 h2ga1-3 imbibed 3h+GA (E1)
62EYamaguchi & Kamiya ga1-35 μm GA46 hSeedsWater24 h2ga1-3 imbibed 6h+GA (E2)
63EYamaguchi & Kamiya ga1-35 μm GA49 hSeedsWater24 h2ga1-3 imbibed 9h+GA (E3)
Effect of GA inhibitors on seedlingsf.
64FShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h (F1)
65FShimada & YoshidaTrueCol-0Mock12 hSeedlingMS liquid7 days2Mock 12h (F2)
66FShimada & Yoshida Col-010 μm propiconazole3 hSeedlingMS liquid7 days2PPI 3h (F1)
67FShimada & Yoshida Col-010 μm propiconazole12 hSeedlingMS liquid7 days2PPI 12h (F2)
68FShimada & Yoshida Col-010 μm uniconazole3 hSeedlingMS liquid7 days2Unic 3h (F1)
69FShimada & Yoshida Col-010 μm uniconazole12 hSeedlingMS liquid7 days2Unic 12h (F2)
70FShimada & Yoshida Col-010 μm paclobutrazol3 hSeedlingMS liquid7 days2Pac 3h (F1)
71FShimada & Yoshida Col-010 μm paclobutrazol12 hSeedlingMS liquid7 days2Pac 12h (F2)
72FShimada & Yoshida Col-010 μm prohexadione3 hSeedlingMS liquid7 days2PX 3h (F1)
73FShimada & Yoshida Col-010 μm prohexadione12 hSeedlingMS liquid7 days2PX 12h (F2)
Effect of auxin inhibitors on seedlingsg
74FShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h (F1)
75FShimada & Yoshida Col-010 μm 2,4,6T3 hSeedlingMS liquid7 days2246T 3h (F1)
76FShimada & Yoshida Col-010 μm PCIB3 hSeedlingMS liquid7 days2PCIB 3h (F1)
77FShimada & Yoshida Col-010 μm TIBA3 hSeedlingMS liquid7 days2TIBA 3h (F1)
78FShimada & Yoshida Col-010 μm NPA3 hSeedlingMS liquid7 days2NPA 3h (F1)
Effect of brassinosteroid inhibitors on seedlingsh
79FShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h (F1)
80FShimada & YoshidaTrueCol-0Mock12 hSeedlingMS liquid7 days2Mock 12h (F2)
81FShimada & Yoshida Col-010 μm Brz2203 hSeedlingMS liquid7 days2Brz220 10 μm 3h (F1)
82FShimada & Yoshida Col-010 μm Brz22012 hSeedlingMS liquid7 days2Brz220 10 μm 12h (F2)
83FShimada & Yoshida Col-010 μm Brz913 hSeedlingMS liquid7 days2Brz91 3h (F1)
84FShimada & Yoshida Col-010 μm Brz9112 hSeedlingMS liquid7 days2Brz91 12h (F2)
85GShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h (G)
86GShimada & Yoshida Col-03 μm Brz2203 hSeedlingMS liquid7 days2Brz220 3 μm 3h (G)
Effect of ethylene inhibitors on seedlingsi.
87FShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h(F1)
88FShimada & Yoshida Col-010 μm AgNO33 hSeedlingMS liquid7 days2AgNO3 3h (F1)
89FShimada & Yoshida Col-010 μm AVG3 hSeedlingMS liquid7 days2AVG 3h (F1)
Effect of other inhibitors on seedlingsj
90FShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h (F1)
91FShimada & Yoshida Col-010 μm CHX3 hSeedlingMS liquid7 days2CHX 3h (F1)
92FShimada & Yoshida Col-010 μm MG1323 hSeedlingMS liquid7 days2MG132 3h (F1)
93FShimada & YoshidaTrueCol-0Mock12 hSeedlingMS liquid7 days2Mock 12h (F2)
94FShimada & Yoshida Col-01 μm PNO83 hSeedlingMS liquid7 days2PNO8 1 μm 3h (F1)
95FShimada & Yoshida Col-01 μm PNO812 hSeedlingMS liquid7 days2PNO8 1 μm 12h (F2)
96GShimada & YoshidaTrueCol-0Mock3 hSeedlingMS liquid7 days2Mock 3h (G)
97GShimada & Yoshida Col-010 μm PNO83 hSeedlingMS liquid7 days2PNO8 10 μm 3h (G)
98FShimada & Yoshida Col-010 μm ibuprofen3 hSeedlingMS liquid7 days2Ibup 3h (F1)
99FShimada & Yoshida Col-010 μm daminozide3 hSeedlingMS liquid7 days2B9 3h (F1)
100FShimada & Yoshida Col-010 μm salicylic acid3 hSeedlingMS liquid7 days2Sal 3h (F1)
Table 2.   AtGenExpress: various genotypes
No.ExperimentLaboratoryControlGenotypeTreatmentTime point Organ/tissueGrowth conditionAge (days)ReplicatesAbbreviation
  1. aSeedlings were grown in half-strength MS liquid medium for 7 days at 23°C.

  2. bSeedlings were grown on MS agar plates for 21 days at 22°C.

Comparison of plant hormone-related mutantsa
101HShimada & YoshidaTrueWs-2No treatmentN/ASeedlingMS liquid72WT-WS (H)
102HShimada & Yoshida bri1-5No treatmentN/ASeedlingMS liquid72bri1 (H)
103KShimada & YoshidaTrueLer-1No treatmentN/ASeedlingMS liquid72WT-Ler (K)
104KShimada & Yoshida ga1-5No treatmentN/ASeedlingMS liquid72ga1 (K)
ARR21C overexpressionb
105LSakakibara & MizunoTrueCol-0No treatmentN/ASeedlingMS agar213Cont (L)
106LSakakibara & Mizuno ARR21C-oxNo treatmentN/ASeedlingMS agar213ARR21C-ox (L)
Table 3.   AtGenExpress: development (time course of early seed germination)a
No.ExperimentLaboratoryControlGenotypeTreatmentTime point (h) Organ/tissueGrowth conditionAge (days)ReplicatesAbbreviation
  1. aDry seeds of Arabidopsis thaliana (Col-0) were imbibed in water for 1 or 3 h at 22°C under continuous light.

107MNambara & KamiyaTrueCol-0No treatment0SeedsN/A02Dry seed (M)
108MNambara & Kamiya Col-0Imbibed seeds (water)1SeedsN/A02Imbibed 1 h (M)
109MNambara & Kamiya Col-0Imbibed seeds (water)3SeedsN/A02Imbibed 3 h (M)
Table 4.   AtGenExpress: abiotic treatments
No.ExperimentLaboratoryControlGenotypeTreatmentTime point (h) Organ/tissueGrowth conditionAge (days)ReplicatesAbbreviation
  1. aSeeds of Arabidopsis thaliana (Ler-0), were irradiated with a far-red light pulse 1 h after imbibition, and then incubated for 96 h in the dark at 22 or 4°C.

  2. bArabidopsis thaliana seedlings (Col-0) were grown vertically for 10 days on MGRL agar medium containing 1500 μm sulfate at 22°C under 16 h/8 h light/dark cycles and then transferred to medium containing or lacking sulfate.

Differential temperature treatment of seedsa
110NYamaguchi & KamiyaTrueLer-0Incubated at 22°C96SeedsN/A02Imbibed 96h22dg. (N)
111NYamaguchi & Kamiya Ler-0Incubated at 4°C96SeedsN/A02Imbibed 96h4dg. (N)
Response to sulfate limitationb
112PTakahashiTrueCol-0Mock0RootMGRL agar102+S 0h (P)
113PTakahashiTrueCol-0Mock2RootMGRL agar102+S 2h (P2)
114PTakahashiTrueCol-0Mock4RootMGRL agar102+S 4h (P3)
115PTakahashiTrueCol-0Mock8RootMGRL agar102+S 8h (P4)
116PTakahashiTrueCol-0Mock12RootMGRL agar102+S 12h (P5)
117PTakahashiTrueCol-0Mock24RootMGRL agar102+S 24h (P6)
118PTakahashi Col-00 μm sulfate2RootMGRL agar102−S 2h (P2)
119PTakahashi Col-00 μm sulfate4RootMGRL agar102−S 4h (P3)
120PTakahashi Col-00 μm sulfate8RootMGRL agar102−S 8h (P4)
121PTakahashi Col-00 μm sulfate12RootMGRL agar102−S 12h (P5)
122PTakahashi Col-00 μm sulfate24RootMGRL agar102−S 24h (P6)

The number of detected genes (defined by detection< 0.05) ranged from 51 to 72% of total genes in all experiments (Figure S1). Dry seeds and germinating seeds expressed smaller numbers of genes compared to the other samples. The total numbers of detected genes were compared among the three major experiments conducted by AtGenExpress: the development, stress and hormone series. As shown in Figure 1, 21 095 genes were detected from 236 GeneChips in the hormone series, compared with 20 499 genes from 250 GeneChips in the stress series (Kilian et al., 2007) and 21 268 genes from 237 GeneChips in the development series (Schmid et al., 2005). In total, 476 genes were unique to the development data set, 121 were unique to the stress data set, and 359 genes were unique to the hormone series. Newly detected genes from our data set are listed in Table S1.

Figure 1.

 Comparison of the numbers of detected genes in the AtGenExpress project.
The numbers of genes detected at a significant level (detection level P < 0.05) were compared among the hormone, stress and development data sets.

Hormone-inducible genes and hormone response validation

The effects of each hormone were confirmed using the marker genes listed in Table 5. Each marker gene was up or downregulated as reported in the studies cited for each experiment. Differentially expressed genes between the mock- and hormone-treated samples were identified as hormone-inducible genes (see Experimental procedures for details). Genes were first filtered by the detection P-value, calculated using Affymetrix Microarray Suite software. Of the 22 746 genes (probe sets) represented on the GeneChip, 18 775 passed this filtering. The signal intensities were then analyzed by Welch’s t-test at each of three time points, and further filtered based on the false discovery rate (FDR, q value < 0.1; Storey and Tibshirani, 2003). The numbers of hormone-inducible genes (at low stringency) are shown in Table 6. Their probe ID, AGI code, P value, q value, signal ratio, and annotation are shown in Table S2. ABA-inducible genes formed the largest group. When GA or BR were applied to wild-type seedlings, the gene expression responses were smaller than in similarly treated GA- or BR-deficient mutants (data not shown). This was probably because the hormone levels or responses were saturated in the wild-type plants. We therefore adopted the results for the hormone-deficient mutants as the typical hormone response (low-stringency lists, Table 6 and Table S2). These gene lists were further used to analyze their overlap (Figure 5).

Table 5.   Marker genes used to confirm the hormone responses
Gene symbolAGI locusReference
Auxin
 IAA1At4g14560Abel et al. (1995)
 IAA2At3g23030Abel et al. (1995)
Cytokinin
 ARR5At3g48100Taniguchi et al. (1998)
 ARR6At5g62920Taniguchi et al. (1998)
 ARR15At1g74890Kiba et al. (2002)
Gibberellin
 At-EXP1At1g69530Ogawa et al. (2003)
 GAIAt1g14920Ogawa et al. (2003)
 SCL3At1g50420Ogawa et al. (2003)
Abscisic acid
 RD29AAt5g52310Yamaguchi-Shinozaki and Shinozaki (1993)
 COR15AAt2g42540Wilhelm and Thomashow (1993)
Jasmonate
 JMTAt1g19640Seo et al. (2001)
 OPR3At2g06050Müssig et al. (2000)
 JIN1/MYC2/RAP1At1g32640Lorenzo et al. (2004)
Ethylene
 ERS2At1g04310Hua et al. (1998)
 EFEAt1g05010Gomezlim et al. (1993)
Brassinosteroid
 BR6ox2/CYP85A2At3g30180Shimada et al. (2003)
 CPD/CYP90AAt5g05690Mathur et al. (1998)
 DWF4/CYP90BAt3g50660Noguchi et al. (2000)
Table 6.   Number of genes differentially expressed following hormone treatment
 Time
30 min1 h3 h
WT + ABA (up)142671965
WT + ABA (down)172251661
WT + ACC (up)16414529
WT + ACC (down)49549160
det2 + BL (up)076923
det2 + BL (down)043870
ga1 + GA (up)039183
ga1 + GA (down)177145
WT + IAA (up)3361250
WT + IAA (down)2328388
WT + MJ (up)231415576
WT + MJ (down)328464455
WT + zeatin (up)2039104
WT + zeatin (down)637121
Figure 5.

 Overlap between the hormone-inducible genes and hormone networks of Arabidopsis seedlings.
The relationships among the hormone-inducible genes were analyzed using Fisher’s exact test. The hormone-inducible genes (Table S2) were classified as up or downregulated for each time point (30 min, 1 h or 3 h). The groups (7 × 3 × 2 = 42 groups in total) were then analyzed using Fisher’s exact test in a pairwise manner.
The resulting relationships were classified as complete positive (a), complete negative (b), partial positive (c, d) or partial negative (e, f), depending on the type of overlap. Circles indicate a significantly higher frequency of overlap than expected to occur at random. A cross indicates that the overlap was not significant. Each complete positive interaction was further tested to assess whether or not the interaction was time-dependent (g). If the interaction did not fit the above criteria, it was defined as a complex interaction. (h) Summary of the results for all seven hormones.

We assessed the relevance of the hormone responses and hormone-inducible gene lists by comparing the results of the hormone treatment experiments to those of additional experiments or previous experiments using Pearson’s correlation coefficient. In the BR experiment, we compared BL-induced gene expression with that induced by a BR-biosynthesis inhibitor, brassinazole (Brz). The BR-deficient det2 mutant was treated with BL or given mock treatment for 3 h, while wild-type seedlings were treated with Brz or given mock treatment for 3 h. The resulting BL signal ratios (BL/mock on the x axis) were compared with those of Brz (Brz/mock on the y axis) for the BR-inducible genes listed in Table S2. The correlation coefficient between the two experiments was −0.58. Although a negative correlation was observed, it was less significant than that observed in a previous report (−0.79, Goda et al., 2002). Based on the FDR q value threshold, we estimated that 10% of the identified genes were false positives, which could create noise in the Pearson’s correlation analysis. Thus, we optimized the correlation coefficients by adjusting the threshold of gene selection for each hormone (see Experimental procedures) to further exclude false-positive genes. This made the correlation coefficients more significant for each hormone (Figure 2): −0.78 for BL versus Brz in the BR experiment (41 genes), −0.52 for GA versus prohexadione in the GA experiment (49 genes), −0.86 for ACC versus ethylene inhibitor aminoethoxy-vinylglycine (AVG) in the ethylene experiment (41 genes), 0.70 for ABA versus mannitol treatment in the ABA experiment (472 genes), 0.43 for zeatin versus an ARR21C-overexpressing mutant in the CK experiment (52 genes), 0.61 for MJ versus Botrytis cinerea infection (Ferrari et al., 2007) in the JA experiment (758 genes), and 0.94 for IAA versus previous IAA treatment data (Goda et al., 2004) in the auxin experiment (143 genes). These data indicate that the gene expression profiles for each hormone are well correlated with those in additional or previous experiments, as were the direction of the correlations (positive or negative). The number of hormone-inducible genes decreased under these highly stringent conditions because the number of false-negative genes increased. These gene lists are considered highly stringent lists (Table S5) and were used to analyze and estimate hormone status as described below.

Figure 2.

 Confirmation of the hormone responses and hormone-inducible genes.
Gene expression in response to hormone exposure was confirmed as follows. A scatterplot of the hormone-inducible genes (Table S2) was drawn for each hormone. The x axis indicates transcript responses as a log signal ratio (hormone treatment/mock treatment), and the y axis indicates transcript responses as a log signal ratio for separate experiments. (a) ABA treatment versus osmotic stress (mannitol treatment). (b) ACC treatment versus ethylene biosynthesis inhibitor treatment (AVG). (c) BL treatment versus BR biosynthesis inhibitor treatment (Brz220 for 3 h). (d) Zeatin treatment versus an ARR21C-overexpressing mutant. (e) GA treatment versus GA the biosynthesis inhibitor treatment (prohexadione). (f) IAA treatment (x) versus previous IAA experiments (y). (g) MJ treatment versus infection with Botrytis cinerea.

Analysis of hormone actions

Existing statistical analyses or clustering methods for microarray data often fail to answer the specific questions of interest to biologists (Dhollander et al., 2007; Hibbs et al., 2007; Ivakhno and Armstrong, 2007). Moreover, comparisons of data across multiple laboratories have shown that the data sets for each laboratory tend to be more similar to each other than they are to those of other laboratories (data not shown). Likewise, data on similar tissues from different laboratories tend to be similar even though they focus on different responses. Thus, developing methods to elucidate clear biological relationships from high-dimensional biological data is very important. As shown in Figure 2, we detected similarities between two independent experiments using hormone-inducible genes selected by hormone treatment, even when conducted in different laboratories using samples of different tissues at different developmental stages. We applied this approach to estimate hormone status in multiple expression profiles from the AtGenExpress stress series data set (Kilian et al., 2007). The experimental treatments in this data set include cold, drought, UV-B, high salt, high osmolarity, heat and wounding. Signal ratio values (hormone treatment/mock, or stress treatment/control) were calculated for both experiments and then transformed to a log2 scale to adjust the origin (i.e. zero point). Then Pearson’s correlations were calculated for each combination of stressed sample and hormone-treated sample using the log signal ratios for each hormone-inducible gene (highly stringent list in Table S5). Significant correlations with ABA treatment were observed soon after osmotic stress (Figure 4b). Interestingly, the correlation was observed at an early stage in the roots, beginning at 30 min and reaching a plateau after 1 h, whereas it was observed in the shoot after 1 h and did not reach a plateau until 3 h, suggesting that the stress response begins in the roots and then spreads to the shoot. The result was consistent with the design of the experiments, as the osmotic stress was applied to the roots (Kilian et al., 2007). Similarly, a correlation with ABA was also observed for cold stress; however, it was delayed and only became significant 6 h after treatment (Figure 4a). The correlation with ABA was even weaker for drought treatment (data not shown), suggesting that the conditions used were mild. This is consistent with the observations by Kilian et al. (2007), because the drought-induced transcript response was small compared to other stress-induced responses. Correlations were also detected between ACC treatments and wounding or between MJ treatments and wounding (Figure 4c), consistent with previous studies showing that ethylene and JA function in the response to wounding. An additional correlation was detected between ACC treatment and UV-B irradiation (Figure 4d). In contrast to the strong interaction between stress and stress-related hormones, we detected little interaction with growth-related hormones. Nevertheless, we detected correlations that probably indicate novel hormone actions, as follows. GA was inactivated by various stressors, including cold, high osmolarity, wounding and UV light. In contrast, CK was activated transiently by cold and osmotic stress, especially in the roots. Thus, this approach allowed us to simultaneously predict the status of all seven hormones from a single experiment. Formerly, it was necessary to rely on marker genes and endogenous hormones to study hormone function because there was no conclusive means to monitor the in vivo status of multiple hormones at the same time.

Figure 4.

 Correlations between the expression profiles for the hormone and stress treatment data.
Hormone status in the AtGenExpress stress series data (Kilian et al., 2007) was analyzed using the expression profile of hormone-responsive genes (high-stringency list). The stress series data were collected from shoots (gray bars) and roots (white bars) exposed to stress for 0.5–24 h. Pearson’s correlation coefficients were calculated for cold (a), osmotic (b), wounding (c) and UV-B (d) stress data, and are shown as bar graphs for each time point.

Analysis of the hormone network

To obtain insight into the relationships among phytohormones, we investigated whether there was significant overlap between each set of hormone-inducible genes. The hormone-inducible genes (Table S2) were classified into groups based on whether they were up or downregulated at each of three time points (30 min, 1 h and 3 h), and the groups were compared in a pairwise manner using Fisher’s exact test. A significant number of overlapped genes were detected (pink in Table S4), more than would be expected by chance, indicating that hormones regulate shared target genes at the transcriptional level. This result was clearer when the overlap was considered in terms of the direction and/or time dependence of the response. For example, if there was significant overlap between genes that were inducible by hormone A and those that were inducible by hormone B genes, and if the overlap was observed in the same direction (i.e. the genes upregulated by hormone A overlapped with those upregulated by hormone B and the genes downregulated by hormone A overlapped with those downregulated by hormone B), the two hormones regulate common target genes in the same direction. Such interactions were termed complete positive interactions (Figure 5a). In contrast, if the overlap occurred in opposite directions (i.e. the genes upregulated by hormone A overlapped with those downregulated by hormone B, and vice versa), the interaction was termed a complete negative interaction (Figure 5b). Finally, if the overlap was incomplete (i.e. there was significant overlap between the genes upregulated by hormone A and those upregulated by hormone B, but not between the genes downregulated by hormone A and those downregulated by hormone B), the interaction was designated a partial positive interaction (Figure 5c,d) or a partial negative interaction (Figure 5e,f). In partial interactions, it is unlikely that the interaction can be accounted for by regulation at the endogenous hormone level, even if the amount of change is controlled by its transport. Partial interactions are probably the result of limited overlap between signaling pathways (e.g. two hormones share a transcriptional regulator that may function as either an activator or repressor). A partial positive interaction was observed between auxin and BR (Figure 5h and Table S4), which is consistent with previous reports showing that these two hormones act synergistically in monocots and dicots. Cross-talk between the two signaling pathways has been suggested in recent molecular and genetic studies (Nakamura et al., 2006; Nemhauser et al., 2004).

A complete interaction can be accounted for by one of two mechanisms: either one of the hormones promotes the active level of the other or they share the same signaling pathway to promote sensitivities of each other. These possibilities may be distinguished by considering the timing of the interaction as follows. If hormone A induces the biosynthesis of hormone B, induction of gene expression by hormone B will be observed at a later time point after hormone A treatment, and no early induction of gene expression by hormone A will be observed following hormone B treatment (Figure 5g). Consistent with this model, early ethylene-inducible gene expression was observed 3 h after auxin treatment (Figure 5h and Table S4), whereas no early auxin-inducible gene expression was observed following ACC (ethylene) treatment. This is consistent with the fact that exogenous auxin induces ethylene biosynthesis in various organs in many species. Given that a similar interaction was found between auxin and ABA (Figure 5h and Table S4), auxin probably induces the biosynthesis of ABA. In fact, NCED5/At1g30100, an ABA biosynthetic gene, was induced by IAA treatment (signal ratio = 4.6, P-value = 0.064, q value = 0.12, t-test at 3 h). Positive interactions were also identified among the stress-related hormones ABA, ethylene and JA (Figure 5h and Table S4). Interestingly, because no time-dependent interaction was observed, stress-related hormones are more likely to share signaling pathways than to activate the biosynthesis of one another. The hormone network presented in Figure 5(h) is based on expression profiles from Arabidopsis seedlings; it may not be applicable to other tissues or species. Nevertheless, our results demonstrate the usefulness of this type of analysis in studying hormone networks and functions. This approach will be applicable to other tissues and species once additional experiments have been conducted. Nemhauser et al. (2006) used our data to analyze hormone networks from the point of view of hormone metabolism. The data presented here provide a starting point from which to advance the understanding of plant hormone networks using a systems-based approach.

Co-expression analysis of Arabidopsis genes by genome-wide hierarchical clustering

One of the purposes of this study was to establish a comprehensive database to facilitate gene expression profile analysis. Our project allows access to gene expression profiles on an unprecedented scale, which allows more precise analysis. Using this data set, we attempted to calculate the correlation between all genes in Arabidopsis and to describe their relationships in one network. To this end, we used the HCA method (Eisen et al., 1998). Genome-wide gene-to-gene correlations were calculated as described in Experimental procedures. Example clusters were extracted from the global clustering results. The BR biosynthetic genes At5g05690/CPD (Szekeres et al., 1996), At3g50660/DWF4 (Choe et al., 1998) and At3g30180/BR6ox2 (Kim et al., 2005; Nomura et al., 2005), and the BR signaling components, At3g61460/BRH1 (Molnar et al., 2002) and At4g36780 [a homolog of BZR1 (Wang et al., 2002)], are clustered together with functionally unknown genes that encode a helix-loop-helix DNA-binding protein (At5g57780) and a wound-responsive protein homolog (At5G01740) (Figure 3). These genes were generally upregulated in the BR-deficient mutant det2 and in the BR-insensitive mutant bri1. They were upregulated by triazole inhibitors such as Brz. As these genes are expressed in a strictly coordinated manner, they probably have coordinated biological roles. Many genes encoding ribosomal proteins were clustered with the ubiquitin genes UBQ1, UBQ2 and UBQ6, as well as with other unknown genes (Figure S2a). The ubiquitin genes have been used as stable, internal standard genes in gene expression analysis. Interestingly, several other ubiquitin genes were not clustered closely with these genes (data not shown), indicating that not all ubiquitin genes are expressed in the same manner. The genes in this cluster were downregulated in dry seeds as well as in seeds in the early stages of germination, but they were expressed relatively stably throughout all experiments. Thus they may be appropriate as internal controls for gene expression studies. The data set provided useful information beyond the scope of the experimental design of the AtGenExpress. For example, although HCA analysis did not include data from experiments focusing on the cell cycle, we observed a cluster of cell cycle-related genes. Many histones, cyclins and cell division-related genes were clustered together among many unknown genes (Figure S2b). They were expressed predominantly in the shoot apex. Similarly, although no experiments concerning anthocyanin biosynthesis were conducted, we observed a cluster of genes related to anthocyanin metabolism. This cluster included PAP1, ANS, DFR/TT3, AtGST12/TT19, TTG2 (Lepiniec et al., 2006), UGT75C1 (Tohge et al., 2005) and UGT79B1 (Yonekura-Sakakibara et al., 2007), which are predominantly expressed in GA-deficient mutants (Figure S2c). These results indicate that the AtGenExpress data set is sufficiently broad and diverse to detect a large proportion of possible gene-to-gene correlations.

Figure 3.

 Genome-wide hierarchical cluster analysis.
All genes represented on the ATH1 chip were clustered hierarchically using AtGenExpress data sets. A cluster related to BR biosynthesis was extracted from the genome-wide results. The colors represent the relative expression level of each experimental group, where red is higher expression and green is lower expression. The AGI code is shown for each gene.

Data presentation and its utility

Our hormone series data has been available to the public on the AtGenExpress Japan Web site (Table 7) since 2004. The data were also incorporated into the gene expression resource home page of the Arabidopsis Information Resource (TAIR; Table 7), NASCArrays (Table 7), and the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/), as recommended in MIAME (http://www.mged.org/Workgroups/MIAME/miame.html). The data have also been used actively by informatics biologists to develop various tools or to conduct specific analyses. The amount and format of the original data were not user-friendly for non-informatics biologists. Therefore, demand developed for a biologist-oriented data presentation to facilitate the use of such a large-scale data set. The results presented above, based on HCA, represent our response to this demand. Genome-wide gene-to-gene correlations calculated by HCA can be used to show relationships between genes based on their expression patterns and to visualize gene expression levels at the same time. Although the results of HCA are useful when used on a genome-wide scale, the amount of information produced is too large to be visualized on a personal computer. Therefore, we have developed a web-based system called Cluster Cutting that visualizes portions of our global data (Figure S3A). By inputting a gene’s AGI code into the Locus ID window (Figure S3B), Cluster Cutting will extract the HCA results for that gene together with its co-expressed genes (default of 100 correlated genes). The results can then be opened by clicking ‘View AtXgXXXXX 100-node’ using JAVA Treeview software (Saldanha, 2004). In this way, the genes that are correlated with a gene of interest can be visualized with their expression patterns. By downloading the result files (result table file, gene tree file and array tree file), the results can be opened locally using JAVA Treeview (available at http://sourceforge.net/projects/jtreeview/) for further data presentation. As far as we are aware, this system is unique because it presents transcriptional gene-to-gene correlations for the entire Arabidopsis genome in a single network.

The data set from the AtGenExpress project has been incorporated and used in web-based tools and databases. For example, the Arabidopsis Information Resource (Rhee et al., 2003) presents a microarray expression search site that offers various search methods for genes and experiments, and presents gene expression patterns both in terms of the signal value and percentile. Genevestigator (Zimmermann et al., 2004, 2005) was one of the first web sites to offer AtGenExpress data, and provides the most versatile tools for analyzing and visualizing gene expression data. The Botany Array Resource provides electronic Northerns (Toufighi et al., 2005), electronic fluorescent pictographs (Winter et al., 2007) and other tools by means of a user-friendly interface that presents gene expression patterns in a visually appealing manner. ATTED-II provides a co-expression analytical tool that is ready to draw gene-to-gene networks, and the co-expression results are also used to predict cis-regulatory elements (Obayashi et al., 2007).

Several studies have been conducted using AtGenExpress data. For example, Nemhauser et al. (2006) used the data pertaining to the basic hormone treatments, and reported that hormones regulate non-overlapping transcriptional responses. Czechowski et al. (2005) identified genes with very stable expression patterns as reference genes for gene expression studies. Many other studies have utilized the gene expression data available on the web sites. Genevestigator has helped researchers to develop new hypotheses (Grennan, 2006), and this reference has been cited >400 times in the last two years. While some of these reports simply refer to gene expression patterns based on the data from AtGenExpress, others present more advanced analyses. For example, Nafisi et al. (2007) identified the function of CYP71A13, a P450 enzyme involved in the biosynthesis of the phytoalexin camalexin. They used Genevestigator to identify similarities in the gene expression patterns of CYP71A13 and CYP71B15, another enzyme involved in camalexin biosynthesis. Yonekura-Sakakibara et al. (2007) used co-expression analysis (ATTED-II) to identify the gene encoding flavonol 7-O-rhamnosyltransferase. Co-expression analysis using large-scale expression profile data is becoming an essential tool in plant biology (see Aoki et al., 2007; Saito et al., 2008). The AtGenExpress data presented here will be an indispensable resource for both plant biologists and informaticians.

Experimental procedures

Plant material and growth conditions

Our plant materials, growth conditions and sample treatment protocols are described in Tables 1–4. Further details on these protocols can be obtained from the references cited: treatment of seedlings with hormones and inhibitors (Goda et al., 2002, 2004; Sawa et al., 2002), CK mutants (Kiba et al., 2004, 2005), sulfate starvation (Maruyama-Nakashita et al., 2005), GA treatment during imbibition (Ogawa et al., 2003), temperature variation during imbibition (Yamauchi et al., 2004), and ABA treatment during imbibition (Nakabayashi et al., 2005).

GeneChip analysis

RNA preparation and GeneChip (Affymetrix) analysis were conducted as described previously (Goda et al., 2002, 2004). Shimada’s, Nambara’s and Yamaguchi’s groups performed the hybridization and scanning steps at a high level of sensitivity to focus on those genes that function in hormone signaling, because their transcripts are relatively less abundant. Under these conditions, signals of highly abundant transcripts may reach saturation. Other groups performed these steps under standard conditions following manufacturers’ instructions. The data were analyzed using Microarray Suite version 5 (MAS5; Affymetrix) to calculate the signal values and detection P-values. The 50th percentile of all measurements was used as the positive control for each sample (i.e. each GeneChip). The signal for each probe set on a particular chip was divided by the synthetic positive control (per-chip normalization). The quality of each repeated experiment was controlled by r2 and slope using linear regression analysis (Table S3). This analysis revealed that the data collected from germinating seeds at 22°C (experiment 110 in Table 4) were not reproducible. Therefore, the data should not be treated as replicates.

Data analysis

The signal values obtained using MAS5 were transformed to a log2 scale and further analyzed using R (http://www.r-project.org). The hormone-inducible genes (low-stringency list; Table S2) were detected as follows. The genes that were significantly expressed (detection level < 0.05) in at least one of the basic hormone treatments (60 GeneChips) were used for the subsequent analysis (18 775 in total), and were analyzed using Welch’s t-test for each hormone treatment at 30 min, 1 h and 3 h. The data were further filtered based on the FDR (q value < 0.1; Storey and Tibshirani, 2003). The genes were then classified as up or downregulated at each time point and analyzed in a pairwise manner using Fisher’s exact test (threshold = 0.001). Comparisons of data at the 3 h time point were excluded from this analysis, as we observed conflicting relationships that were probably caused by secondary responses. To create the high-stringency gene list (Table S5), the following analysis and thresholds were applied to each hormone. The genes that were significantly expressed (detection level < 0.05) in at least one of 12 GeneChip experiments (including mock treatment) were included in the analysis. Then, four groups of data were analyzed using standard one-way anova: mock, 30 min, 1 h and 3 h. The genes were then filtered by their FDR q values (<0.05) and further filtered based on the absolute value of their signal ratios (SR). The threshold SR and recovered gene number for each hormone were as follows: ABA, SR = 2.25, = 472; ACC, SR = 0.25, = 41; BL, SR = 2.25, = 41; CK, SR = 0.25, = 143; GA, SR = 0, = 49; IAA, SR = 1.25, = 143; MJ, SR = 0.75, = 758. Our conditions (> 40) correspond to a Pearson’s correlation coefficient of 0.4 (< 0.01).

Global hierarchical clustering

Data from all or a proportion of the following groups were included in our analysis. Blaesing (diurnal), the Nottingham Arabidopsis Stock Centre (light and pathogen), RIKEN: Shimada’s group (basic hormone treatments, inhibitors, etc), RIKEN: Kamiya’s group (seed germination, imbibition, ABA, low temperature, GA experiments), Mizuno and Sakakibara’s group (zeatin and ARRs), RIKEN: Takahashi’s group (sulfur starvation), the German Resource Centre for Genome Research (stress series), Scheible’s group (nitrogen starvation), Weigel’s group (floral transition), Weigel’s group (organs and development). The data from the Nottingham Arabidopsis Stock Centre, the German Resource Centre for Genome Research and Weigel’s group were subdivided into smaller groups. The Nottingham Arabidopsis Stock Centre data were divided into a light series and pathogen series. The data from the German Resource Centre for Genome Research were divided into three parts: root experiments, shoot experiments, and cell-culture experiments. Weigel’s organ-development experiments were divided into the following three subgroups: floral stage 12, shoot apices, and other. All signal values were imported into GeneSpring (Agilent/SiliconGenetics) and normalized per chip, assuming that the signal was at least 10 (signals lower than 10 were corrected to 10). Per-gene normalization was also applied in the above 15 experimental subgroups. The expression levels of each gene were normalized to their own median as a synthetic positive control. The data sets were then imported into Cluster 3 (de Hoon et al., 2004), transformed to a log2 scale. The genes and arrays were clustered using centroid linkage. Similarity metrics were defined using Pearson’s correlation.

Acknowledgements

We thank the Arabidopsis Information Resource for providing the AtGenExpress web site, and Ms Chitose Takahashi, Mr Hiroaki Yuasa and Mr Narumasa Miyauchi for technical assistance with data processing, and construction and the maintenance of the web system. We thank Dr Thomas Altmann (University of Potsdam) for coordinating the AtGenExpress project, Dr Alok Saldanha (Stanford University) for providing JAVA Treeview, the RIKEN Super Combined Cluster (RSCC) for supplying computational resources, and the data visualization service of Advanced Center for Computing and Communication at RIKEN for assisting with data visualization.

Ancillary