gb-2005-6-4-r32 GBJ Research <p>Contribution of transcriptional regulation to natural variations in <it>Arabidopsis</it></p> Chen J Wenqiong wenchen@diversa.com Chang H Sherman hchang@diversa.com Hudson E Matthew mhudson@uiuc.edu Kwan Wai-King wkwan@diversa.com Li Jingqiu jili@diversa.com Estes Bram bram.estes@syngenta.com Knoll Daniel danielknoll@gmx.de Shi Liang liang.shi@syngenta.com Zhu Tong tong.zhu@syngenta.com

Torrey Mesa Research Institute, Syngenta Research and Technology, 3115 Merryfield Row, San Diego, CA 92121, USA

Diversa Corporation, 4955 Directors Place, San Diego, CA 92121, USA

Department of Crop Sciences, University of Illinois, 1101 W. Peabody, Urbana, IL 61801, USA

Syngenta Biotechnology, 3054 Cornwallis Road, Research Triangle Park, NC 27709, USA

Institut für Allgemeine Botanik, Universität Hamburg, Ohnhorststrasse 18, 22609 Hamburg, Germany

 Genome Biology 1465-6906 2005 6 4 R32 http://genomebiology.com/2005/6/4/R32 1583311910.1186/gb-2005-6-4-r32 4 6 2004 16 11 2004 9 2 2005 15 3 2005 2005 Chen et al.; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Differential transcriptional regulation among Arabidopsis accessions

Among five accessions 7,508 probe sets with no detectable genomic sequence variations were identified on the basis of the comparative genomic hybridization to the Arabidopsis GeneChip microarray, and used for accession-specific transcriptome analysis, identifying 60 genes that were differentially expressed in different accession backgrounds in an organ-dependent manner. Correlation analysis of expression patterns of these 7,508 genes between pairs of accessions identified a group of 65 highly plastic genes with distinct expression patterns in each accession.

Abstract

Background

Genetic control of gene transcription is a key component in genome evolution. To understand the transcriptional basis of natural variation, we have studied genome-wide variations in transcription and characterized the genetic variations in regulatory elements among Arabidopsis accessions.

Results

Among five accessions (Col-0, C24, Ler, WS-2, and NO-0) 7,508 probe sets with no detectable genomic sequence variations were identified on the basis of the comparative genomic hybridization to the Arabidopsis GeneChip microarray, and used for accession-specific transcriptome analysis. Two-way ANOVA analysis has identified 60 genes whose mRNA levels differed in different accession backgrounds in an organ-dependent manner. Most of these genes were involved in stress responses and late stages of plant development, such as seed development. Correlation analysis of expression patterns of these 7,508 genes between pairs of accessions identified a group of 65 highly plastic genes with distinct expression patterns in each accession.

Conclusion

Genes that show substantial genetic variation in mRNA level are those with functions in signal transduction, transcription and stress response, suggesting the existence of variations in the regulatory mechanisms for these genes among different accessions. This is in contrast to those genes with significant polymorphisms in the coding regions identified by genomic hybridization, which include genes encoding transposon-related proteins, kinases and disease-resistance proteins. While relatively fewer sequence variations were detected on average in the coding regions of these genes, a number of differences were identified from the upstream regions, several of which alter potential cis-regulatory elements. Our results suggest that nucleotide polymorphisms in regulatory elements of genes encoding controlling factors could be primary targets of natural selection and a driving force behind the evolution of Arabidopsis accessions.

 Genetics Genome studies Evolution Plant biology

Background

Transcription of mRNA from DNA and subsequent translation of mRNA into protein transform genetic blueprints into cellular functions. This process of gene expression and regulation plays a key role in determining the fitness of the genome, through the production of different proteins in different cells and at different times. Therefore, in addition to genome composition and structure, regulation of gene expression is also a key component in development and evolution 1.

The importance of regulatory genes during evolution is well recognized 2. For example, major differences in axial morphology consistently correlate with a difference in spatial regulation of Hox gene expression 34. In addition, a cis-regulatory element has functionally diverged during the course of bird and mammal evolution and has resulted in different gene-expression patterns between these two taxa 34. Recently, many studies have suggested that cis-regulatory regions of regulatory genes and their downstream target genes might be a major driving force behind evolutionary changes in humans 5. In plants, evidence for the importance of variations in upstream regulatory regions in the evolution of plant form have also been described. Polymorphisms in an upstream regulatory region of the teosinte branched1 gene have been implicated in the domestication of maize 6, and changes in the promoter region of ORFX may associate with increases in fruit size during tomato domestication 78.

Despite its potential importance, the genetic basis of cis-regulatory evolution is poorly understood. Stone and Wray 1 suggested the following reasons: first, the lack of information on sequence variations in the regulatory regions, and lack of association between the degree of coding sequence divergence and the change in gene expression 9; second, the lack of experimental data from gene-expression analyses to support sequence variation analyses; and third, the lack of a conceptual framework for understanding regulatory evolution that could guide empirical studies. Therefore, to better understand cis-regulatory evolution and its implications for genome stability and dynamics, an essential step is to identify sequence variations in the regulatory regions of regulatory genes and downstream target genes on a genome-wide scale, and establish the correlations between gene-expression variations and regulatory sequence divergence. However, few studies have attempted to correlate molecular studies of the evolution of cis-regulatory genotype with that of phenotype 10.

Naturally occurring phenotypic differences such as leaf shape or biomass among different Arabidopsis accessions 11 have recently become used as resources to study gene function, which traditionally has been studied through mutagenesis and phenotypic characterization of genetic variants 12. Differences in transcriptional regulation have the potential to contribute substantially to such phenotypic differences among accessions. Thus, it is important to understand the extent to which evolutionary differences between accessions are the result of regulatory polymorphisms causing alterations in transcription, as opposed to coding-region polymorphisms that alter the function of gene products. Although transcriptional profiling has been applied to study the transcriptome differences within or among species using both Affymetrix oligonucleotide GeneChip microarrays and cDNA microarrays 131415, a recent study from Hsieh et al. 16 showed a strong species-by-probe interaction effect when using Affymetrix GeneChip microarray for such inter-species transcriptome analysis. Species differences in hybridization signal strength from a probe set can reflect both sequence differences between probes and their hybridizing targets, and differences in abundance of the mRNA. Therefore, comparative transcriptome analysis of different species or accessions is difficult to interpret without controlling for the effect of coding DNA polymorphism before assaying for differences in transcript abundance.

The objectives of this study are to develop a reliable method for comparing transcriptomes among samples with different genetic backgrounds, to identify differences in transcriptomes among different genetic lines, and to understand the regulatory mechanisms responsible for gene-expression differences by analyzing their predicted promoters. To accomplish these goals, we have adopted a new analysis strategy to analyze the transcriptome variations in five Arabidopsis accessions. Our results suggest that genes with functions involved in signal transduction, transcription and stress response are the primary targets for natural selection. This study should shed light on the field of plant evolutionary genomics by furthering our understanding of how the two-way evolutionary interactions between genomic polymorphisms and transcriptional regulatory mechanisms contribute to shaping the evolution of genome.

Results

Strategy for comparing gene expression among accessions

The GeneChip microarray used in this study contains approximately 8,700 probe sets for 8,300 Arabidopsis genes, which covers about one-third of the genome of accession Col-0 (ecotype Columbia) 17. Both perfect match (PM) and mismatch probes of the majority of the probe sets on this GeneChip microarray are able to cross-hybridize to genomic targets from other accessions; however, the hybridization signals are affected by any sequence polymorphisms between the probes and the targets 18. With the standard Affymetrix algorithms (MAS4.0 or MAS5.0) polymorphisms between the hybridizing mRNA samples are likely to invalidate the assumptions underlying the perfect-match mismatch signal subtraction step, leading to inaccurate measurements of the transcript levels, and thus preventing accurate comparisons of the transcriptomes among different accessions.

To address these issues, we selected for the comparative transcriptome analysis PM probes that hybridize similarly to the genomic targets of test accessions (Figure 1). Briefly, genomic DNAs from different accessions were fragmented, labeled and hybridized to the Arabidopsis GeneChip microarrays 19. The hybridization signals from the PM probes were summarized into genomic DNA hybridization indices (gDHI) using the PM-only model 20 to avoid the complication of the array mismatch probes. The coefficient of variance (CV) of the gDHI among the five accessions used in this study for each probe set was used to determine whether there was sufficient genomic sequence difference among the different accessions to substantially alter hybridization to the oligonucleotide probes. Probe sets were ranked on the basis of their CV and those with the largest CV (CV ≥ 0.20) were eliminated (see Additional data files 1 and 8). The cutoff value was chosen on the basis of the overall mean and standard deviation of the CV from genomic DNA hybridization (mean + standard deviation). For the further comparative transcriptome analysis, 7,736 probe sets with CV less than 0.20 were selected.

<p>Figure 1</p>

Schematic diagram of the data analysis process

Schematic diagram of the data analysis process. A genome scan (left panel) was used to identify probe sets corresponding to the genes that were highly polymorphic or less polymorphic in gene coding regions among the five accessions. Genes with polymorphic sequences were functionally categorized. Probe sets corresponding to the less polymorphic genes were used for a transcriptome scan of various accessions (right panel). Genes transcribed at different levels in different accessions were identified and analyzed.

To measure the consistency of our probe set selection in this procedure, the reproducibility of the comparative genomic hybridization experiments was determined by labeling and hybridizing the same genomic DNA onto two different microarrays in parallel. The results were highly reproducible and only a small fraction of genes showed twofold or greater difference in hybridization signals between the two replicated experiments: 0.1% between the Col-0 replicates, 0.02% between the Ler replicates, 0.2% between the C24 replicates, 0.01% between the NO-0 replicates, and 0% between the WS-2 replicates. These results are consistent with the average reproducibility for other genomic DNA labeling and hybridization experiments in Arabidopsis, and similar to the results from reproducibility studies for RNA detection using the same GeneChip microarray 17.

Comparative analysis of transcriptome of different accessions and its validation

Transcription profiles of different organs at different developmental stages (see Additional data file 2) were compared among the five accessions using the following strategy. First, the PM-only model was used to estimate the raw RNA hybridization index (rRHI), to reduce the complication of the array mismatch probes. Second, gDHIs were used to normalize rRHI to remove contributions from sequence variations due to undetected single feature polymorphisms (SFPs) in probe sets. The normalized RNA hybridization index (nRHI), calculated by dividing the rRHI of each probe set by the corresponding gDHI of a particular accession, is used to represent the relative transcript level of the target gene. Third, all the genes were ranked on the basis of their nRHI values, and the lowest 5% were chosen as the cutoff value for background. Genes with an nRHI value less than the cutoff value across all the RNA samples from at least one accession were eliminated from further analysis. By this method, genes whose transcripts could not be detected or were close to the background level were excluded. Fourth, the nRHI values of the 7,508 genes after step 3 were used for statistical analyses, for calculating the Pearson correlation coefficient between all possible pairs of accessions (10 pairs from pairwise comparison of five different accessions) for each gene, and for cluster analysis 21.

To validate variations in transcript abundance detected by the GeneChip microarray through heterologous hybridization using our strategy, quantitative reverse transcription PCR (RT-PCR) using accession-specific primers and probes was performed. Table 1 compares nRHI of 13903_at (At3g54050) and 17392_s_at (At3g53260), measured by the GeneChip microarray and the quantitative RT-PCR in 18 different samples. In general, the quantitative RT-PCR results agreed with the GeneChip microarray results, and confirmed the expression differences of these two genes between accessions Col-0 and C-24. The correlation coefficient between the results of the GeneChip microarray and quantitative RT-PCR is 0.93 for 13903_at, and 0.82 for 17392_s at. As expected, those probe sets with probes cross-hybridizing with genes in a family, such as 17392_s_at, correlated less strongly with accession-specific quantitative RT-PCR.

<p>Table 1</p>

Quantitative RT-PCR confirmation of GeneChip Microarray data for genes 13903_at (At3g54050) and 17392_s_at (At3g53260) in Col-0 and C24

Samples

13903_at

17392_s_at

log2(rRHI)

log2(nRHI)

Taqman

log2(rRHI)

log2(nRHI)

Taqman

Col-0-4 day seedlings

10.11940591

0.911529477

1.348 ± 0.262

10.38351776

0.658285681

0.362 ± 0.024

Col-0-2 week leaf

11.80337083

2.595494397

4.652 ± 0.389

10.56878747

0.84355539

0.299 ± 0.050

Col-0-11 week leaf

10.77324577

1.565369327

1.415 ± 0.336

10.33789612

0.612664042

0.163 ± 0.052

Col-0-2 week root

7.674725423

-1.533151014

0.134 ± 0.014

11.26384894

1.538616864

1.313 ± 0.324

Col-0-5 week root

7.873250697

-1.334625741

0.590 ± 0.064

10.99787749

1.272645415

0.648 ± 0.246

Col-0-influorescence

10.09145865

0.883582211

1.320 ± 0.247

11.01034472

1.285112643

0.519 ± 0.104

Col-0-flower

10.42134176

1.213465325

2.093 ± 0.658

10.62442631

0.899194238

0.263 ± 0.053

Col-0-young siliques

10.65287316

1.444996723

1.999 ± 2.885

10.57630495

0.851072873

0.430 ± 0.197

Col-0-mature siliques

9.475076913

0.267200476

1.432 ± 2.345

10.80990555

1.084673476

0.473 ± 0.113

C24-4 day seedlings

10.90593269

1.883371001

3.690 ± 0.482

10.20742445

0.596353845

0.321 ± 0.059

C24-2 week leaf

12.29789156

3.275329874

6.819 ± 3.507

10.65702025

1.04594965

0.299 ± 0.044

C24-11 week leaf

12.09006973

3.067508045

6.073 ± 1.283

9.19787898

-0.413191622

0.071 ± 0.037

C24-2 week root

7.550943148

-1.471618541

0.069 ± 0.022

10.89199181

1.280921209

0.790 ± 0.133

C24-5 week root

7.945743693

-1.076817996

0.317 ± 0.087

11.16598953

1.554918929

1.122 ± 0.324

C24-influorescence

10.72350042

1.700938727

2.397 ± 0.304

11.10540542

1.494334819

0.743 ± 0.105

C24-flower

10.71423996

1.691678266

1.054 ± 0.167

9.761854806

0.150784204

0.153 ± 0.048

C24-young siliques

11.01401689

1.991455197

1.885 ± 0.726

10.61478826

1.00371766

0.365 ± 0.058

C24-mature siliques

11.21144986

2.188888168

3.808 ± 0.569

11.24013223

1.629061624

1.002 ± 0.151

Correlation with log2 (Taqman assay)

0.925

0.933

0.801

0.821

gDHI for 13903_at is 591.35 and 520.07 for Col-0 and C24, respectively. gDHI for 17392_s_at is 846.42 and 782.02 for Col-0 and C24, respectively.

In addition, nRHI of 12 randomly selected genes with various expression patterns was also validated by quantitative RT-PCR. Some of them did not show different expression levels, and others did show a difference between the flowers of Col-0 and those of Ler. As shown in Table 2, the results from the quantitative RT-PCR analysis were generally consistent with the nRHI regarding the trend of the change for each gene between Col-0 flower and Ler flower. There are two exceptions (16892_at and 20545_at), which showed slightly reduced expression in Ler flower as compared to Col-0 from the GeneChip microarray experiments, but showed an opposite trend of expression from Taqman data. In addition there are a few examples (14172_at and 17860_at), which showed a less than twofold difference from the GeneChip microarray experiments, but slightly higher than twofold differences (14172_at: 2.05-fold, 17860_at: 2.26-fold) from RT-PCR. The slight inconsistency between the GeneChip microarray results and the RT-PCR results may result from the difference in detection technology, and associated sensitivities, between the two methods. It also indicates that definition of significance using twofold change is not appropriate for this experiment. Nevertheless, the results from this extensive validation study using accession-specific primers and probes support our analysis strategy used for transcription analysis of different accessions in both sensitivity and specificity aspects.

<p>Table 2</p>

Quantitative RT-PCR confirmation of GeneChip microarray data for genes expressed in Col-0 and Ler flowers

Probe set ID

Col-flower

Ler-flower

Fold changes

rRHI

gDHI

nRHI

Taqman

RHI

gDHI

nRHI

Taqman

Ler/Col (nRHI)

Ler/Col (Taqman)

12222_s_at

1407.33

700.57

2.01

0.48 ± 0.16

1440.54

557.60

2.58

0.78 ± 0.13

1.29

1.62

14097_at

610.06

1822.91

0.34

0.13 ± 0.03

899.39

1762.56

0.51

0.70 ± 0.23

1.52

5.56

20561_at

760.62

648.27

1.17

0.90 ± 0.14

625.43

719.12

0.87

0.88 ± 0.24

0.74

0.97

14634_s_at

2914.65

1050.64

2.77

0.31 ± 0.05

4304.12

871.65

4.94

0.88 ± 0.05

1.78

2.85

15290_at

701.80

679.74

1.03

0.35 ± 0.03

965.63

583.78

1.65

1.04 ± 0.06

1.60

2.94

14072_at

2034.34

957.24

1.57

0.85 ± 0.13

2285.99

948.01

1.68

1.08 ± 0.20

1.07

1.27

14172_at

894.36

1042.33

0.86

0.44 ± 0.06

1114.93

1107.46

1.01

0.91 ± 0.08

1.17

2.04

14947_at

1888.06

1250.42

1.51

0.98 ± 0.22

1754.25

981.19

1.79

1.24 ± 0.12

1.18

1.26

16892_at

2688.88

836.69

3.22

0.51 ± 0.05

2798.26

1061.25

2.64

1.10 ± 0.11

0.82

2.17

17860_at

959.84

1263.46

0.76

0.49 ± 0.06

1209.50

1322.29

0.92

1.11 ± 0.13

1.20

2.26

20545_at

2183.17

724.58

3.02

0.59 ± 0.09

1971.40

668.92

2.95

0.99 ± 0.09

0.98

1.686

To assess the residual interference from sequence variations between targets and probes within the probe sets used for comparative transcriptome analysis, for each sample, we compared the overall transcriptome profiles by calculating Pearson correlation coefficient between rRHI and nRHI for selected probe sets and all probe sets including those probe sets detecting significant difference in genomic hybridization. A general consistency for each sample was observed (see Additional data files 3 and 9). However, the inclusion of the probe sets detecting difference in genomic hybridization reduces the Pearson correlation coefficients between rRHI and nRHI (see Additional data file 3), demonstrating a greater degree of interference from sequence variation in those probe sets. Data from Tables 1 and 2 also showed examples of high correlation between the rRHI and nRHI. When these data were compared to the data from accession-specific quantitative RT-PCR, the correlation coefficients were slightly different: 0.92 (rRHI) and 0.93 (nRHI) for 13903_at, and 0.80 (rRHI) and 0.82 (nRHI) for 17392_at. These results indicate that the probe sets selected for the comparative transcriptome analysis have a low level of interference, and can be utilized to measure the transcript abundance in the five accessions.

General similarities of transcriptional profiles among accessions from various organs at different stages

As shown in Table 3 and Figure 2, among the 7,508 genes whose expression was above the cutoff value in at least one of the RNA samples, the expression patterns of most of the genes (5,985) were correlated (r > 0.5) in at least five pairwise comparisons (gray bars), indicating that the expression patterns for most genes from different accessions share some similarity. To test whether the high correlation in expression patterns among different accessions was likely to be obtained by chance, we randomly permuted the RNA samples from the same organs of five different accessions (see Materials and methods for details). The number of genes whose expression did not correlate at r > 0.5 for any pair of accession comparisons increased significantly (Figure 2, white bars) from a total of 65 in the original data to 130 (group 10 in Figure 2), and the number of genes whose expression did correlate for all pairs of accession comparisons decreased significantly, from 3,532 in the original data to 1,266 in the permuted data. Because of the close relationship of the five accessions chosen in this study, these data suggest, as expected, that the tissue-specific gene-expression patterns are more consistent between accessions of a single species than any accession-specific patterns between organs.

<p>Table 3</p>

Correlation analysis of expression patterns of genes among the five accessions

Per 1

Per 2

Per 3

Per 4

Per 5

Per 6

Per 7

Per 8

Per 9

Per 10

Average of Per

Observed

10

141

133

127

129

129

130

139

121

130

125

130.4

65

9

263

263

246

268

281

285

271

295

273

259

270.4

271

8

324

324

341

323

336

320

307

324

319

359

327.7

376

7

1555

1555

1542

1539

1499

1508

1541

1524

1521

1505

1528.9

399

6

1523

1523

1575

1505

1515

1547

1557

1495

1524

1539

1530.3

412

5

603

603

590

617

607

629

609

642

632

577

610.9

416

4

692

692

661

725

724

660

668

669

715

719

692.5

471

3

441

441

441

436

440

457

461

457

424

441

443.9

438

2

345

345

375

334

341

351

340

355

357

359

350.2

528

1

360

360

365

382

362

329

345

350

358

350

356.1

600

0

1261

1261

1245

1250

1274

1292

1270

1276

1255

1275

1265.9

3532

7508

7500

7508

7508

7508

7508

7508

7508

7508

7508

7508

For each gene, the Pearson correlation coefficient was calculated for all the 10 pairwise comparisons among the five accessions, as described in Materials and methods. Genes were then grouped into 11 groups (0-10) according to the number of comparisons having correlation coefficients less than 0.5 (group 10 corresponds to the genes with r < 0.5 from all 10 pairwise comparisons, whereas group 0 corresponds to genes with r ≥ 0.5 from all 10 pairwise comparisons). These results are given in the Observed column. Columns Per 1 to Per 10 show the numbers of genes from the 10 permuted datasets, as described in Materials and methods. These results are visualized in Figure 2.

 <p>Figure 2</p>

Correlation analysis of expression patterns of genes among the five accessions

Correlation analysis of expression patterns of genes among the five accessions. A histogram based on the number of genes in each of the 11 groups in Table 3 that have Pearson correlation coefficients less than 0.5 in a given number of pairwise comparisons (see Table 3 for explanation). The white bars indicate the numbers of genes from the experimental datasets, and the gray bars indicate the average numbers of genes from the 10 permuted datasets, as described in Materials and methods.

We used by cluster analysis of the nRHI data to further analyze relationships among the accessions on the basis of the transcriptome profiles (Figure 3). The overall relationships among all samples confirmed that the expression differences among the accessions were small, as the gene-expression differences were greater across different organs of the same accession than that across different accessions in the same organ (Figure 3). Two clusters emerge from the experimental tree: a cluster of axis-origin organs, including roots and young seedlings, and a cluster of auxiliary organs, including vegetative leaves, flowers and siliques (reproductive leaves) and the associated inflorescences (Figure 3). The axis cluster consisted of roots from two different developmental stages - 2 weeks and 5 weeks - as well as 4-day-old seedlings, which are mainly composed of root tissues. The cluster of auxiliary organs could be further divided into two subclusters, one for the vegetative leaves, and one composed of organs originating from the reproductive leaves. Within an organ, especially for leaves, however, variations were contributed by both developmental differences and accession differences. These relationships, as illustrated in Figure 3, were supported by bootstrap analysis 22. One hundred datasets, each containing the same number of genes, were generated from the original dataset by random sampling with replacement. The bootstrap results confirmed the robustness of the cluster results at the top two levels of the dendrogram (Figure 3).

<p>Figure 3</p>

Relationships among the five Arabidopsis accessions based on their expression patterns in different organs at various developmental stages

Relationships among the five Arabidopsis accessions based on their expression patterns in different organs at various developmental stages. The normalized expression values, obtained by dividing the mRNA expression indices of each organ of one accession by the intensity indices in genomic DNA hybridization for that particular accession, were log2-transformed and subjected to cluster analysis. The yellow vertical lines separate the whole cluster into three subclusters, the root cluster, the vegetative leaf cluster, and the reproductive organ cluster.

Accession-specific gene expression during development

Although in general, the gene-expression patterns from the same organs of different accessions were similar, the correlation tends to get worse towards late development (Figure 4). The differences observed among the five accessions in late development could be due to the following reasons: biological noise (individual variation) within each accession during the sampling of biological materials; developmental differences among different accessions; and accession-specific differences due to default regulatory programming. It is unlikely that the differences are due to the sampling noise, as these noises will become undetectable by extensive pooling of biological materials in this study.

<p>Figure 4</p>

Correlations in transcription among five accessions during leaf and silique development

Correlations in transcription among five accessions during leaf and silique development. (a) The Pearson correlation coefficient for a given sample was calculated with nRHI for all the genes from each accession and the reference accession Col-0. Each bar represents the correlation of a particular accession as compared to Col-0 in the sample group. Note the common trend in reduction of the correlation during leaf and silique development for each organ. (b) The regression coefficient for a given sample was calculated with nRHI for all the genes from each accession (Y-values, regressor) and the reference accession Col-0 (X-values, predictor). Each bar represents the regression coefficient of a particular accession as compared to Col-0 in the sample group. The regression coefficient (b) was calculated as b = (ΣXiYi - (ΣXi)(ΣYi)/n)/(ΣXi2 - (ΣXi)2/n), where n is the total number of genes in either Col-0 or the sample to be compared (7,508 in this case). The error bar indicates the upper or lower limit of the 95% confidence interval for each of the given regression coefficients. The 95% confidence interval was calculated as b ± tα(2), (n-2)Sb, where tα(2), (n-2) is the t critical value at α = 0.05, two-tail, df = 7,506, and Sb is the standard deviation of b.

The phenotypic differences, especially during late plant development, such as leaf shape, size and flowering time, prompted us to search for genes whose expression is different among different accessions. To identify genes that represent accession-specific difference, and to differentiate them from the genes which could possibly reflect the developmental differences of these five accession plants at the same age grown under the same conditions, we used the one-way analysis of variance (ANOVA) to analyze nRHI data of 2-, 5-, and 11-week-old leaves from the five accessions. Here we treated samples from 2-, 5-, and 11-week-old leaves as three leaf replicates for each accession, thus the only factor we are analyzing is 'accession' which has five levels in this study (see Additional data file 4).

On the basis of ANOVA, 1,525 genes were found to have p-values less than 0.01 (false discovery rate or FDR = (7,508 × 0.01)/1,525 = 4.9%). Bonferroni correction was further applied for the strong control of family-wise type I error rate (FWER). As shown in Table 4, 58 genes were thus selected, which potentially represent the genes with differential expression among the leaves from the five accessions (p < 0.05). These genes were then functionally classified according to the Munich Information Centre for Protein Sequences (MIPS) functional classification. As shown in Figure 5, these 58 genes encode products with diverse functions. Besides those proteins with unknown function, the top five categories contained genes with possible functions in transcription (18% vs 9% for all the genes on the chip), subcellular localization (18% vs 11% overall), stress/defense response (15% vs 6% overall), metabolism (9% vs 18% overall) and signal transduction (9% vs 9% overall). Compared to the overall distribution for all the genes on the chip among different functional categories, genes involved in transcription, subcellular localization and stress/defense response are enriched in this group (p ≤ 0.008, p ≤ 0.018, and p ≤ 0.004, respectively). Eight genes encoding putative transcriptional regulators, including Dof zinc-finger transcription factors, HD-zip transcription factor Athb-8, and MADS-box containing proteins, were included within this group of 58 genes. Genes involved in stress/defense responses include ones that encode disease-resistant proteins such as those of the TIR-NBS-LRR class, enzymes involved in secondary metabolism, and proteins involved in detoxification.

<p>Table 4</p>

Genes whose expression is different in leaves of the five accessions by one-way ANOVA analysis

Functional category

ATH1 hits

rawp

Bonferroni correction

GenBank ID

Description

01 METABOLISM

17946_s_at

At1g03410

2.84132E-06

0.0213326

gb|AAB97721.1|

2-Oxoglutarate-dependent dioxygenase, putative

19689_at

At5g24140

7.0739E-07

0.0053111

emb|CAA06771.1|

Squalene monooxygenase 2 (squalene epoxidase 2) (SQP2) (SE2)

12277_at

At1g47600

6.47203E-07

0.0048592

gb|AAD46026.1|

Glycosyl hydrolase family 1, similar to thioglucosidase

18836_at

At2g24710

5.09671E-06

0.0382661

gb|AAD26894.1|

Plant glutamate receptor family (GLR2.3)

17620_s_at

At2g42990

3.79611E-06

0.0285012

gb|AAD21711.1|

GDSL-motif lipase/hydrolase protein similar to family II lipase EXL3

20514_i_at

At2g15370

1.35384E-08

0.0001016

gb|AAD22287.1|

Similar to xyloglucan fucosyltransferase

02 ENERGY

12277_at

At1g47600

6.47203E-07

0.0048592

gb|AAD46026.1|

Glycosyl hydrolase family 1, similar to thioglucosidase

10 CELL CYCLE AND DNA PROCESSING

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

15785_g_at

At1g08840

2.79162E-06

0.0209595

gb|AAB70418.1|

Hypothetical protein gene overlaps Sp6 end of F7G19

11 TRANSCRIPTION

12869_s_at

At4g11880

3.45683E-06

0.0259539

gb|AAC49082.1|

MADS-box protein AGL14

16072_s_at

At5g65790

2.97265E-06

0.0223187

gb|AAC83623.1|

Identical to putative transcription factor (MYB68)

13575_at

At4g03430

6.50883E-06

0.0488683

gb|AAD11585.1|

Similar to yeast pre-mRNA splicing factors

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

12366_s_at

At4g11880

2.5787E-06

0.0193608

emb|CAB44326.1|

MADS-box protein AGL14

14885_at

At4g21340

2.22259E-06

0.0166872

emb|CAA20199.1|

Expressed protein, putative bHLH transcription factor (bHLH103)

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

19244_s_at

At2g04230

2.84687E-06

0.0213743

gb|AAD27915.1|

F-box protein family, contains F-box domain

19279_i_at

At4g21040

7.07742E-07

0.0053137

emb|CAB45899.1|

Dof zinc finger protein, finger protein rolB

13306_at

At2g41070

4.38217E-06

0.0329013

gb|AAD12004.1|

bZIP family transcription factor, contains a bZIP transcription factor basic domain signature

13343_at

At1g34310

4.60448E-08

0.0003457

gb|AAD39615.1|

Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related

15224_at

At1g61540

8.21522E-08

0.0006168

gb|AAD25554.1|

Kelch repeat containing F-box protein family low similarity to SKP1 interacting partner 6

15227_at

At2g01280

5.89076E-06

0.0442278

gb|AAD14528.1|

Transcription factor -related, putative transcription factor IIIB 70 KD subunit (TFIIIB)

16263_at

At2g02320

3.12005E-06

0.0234253

gb|AAC78515.1|

F-box protein (SKP1 interacting partner 3-related)

17145_at

At1g10110

6.9462E-08

0.0005215

gb|AAC34337.1|

Contains Pfam PF00646: F-box domain; similar to F-box protein family, AtFBX7

13863_at

At2g21470

9.36899E-07

0.0070342

gb|AAD23691.1|

Nearly identical to SUMO activating enzyme 2 (SAE2)

12599_at

At2g29910

2.33543E-10

0.0000018

gb|AAD23631.1|

F-box protein family contains F-box domain Pfam:PF00646

12913_at

At4g32880

1.90072E-06

0.0142706

emb|CAA90703.1|

Identical to HD-zip transcription factor (athb-8)

13216_s_at

At1g26310

8.05827E-07

0.0060501

gb|AAA64789.1|

Floral regulatory gene CAULIFLOWER

12863_r_at

At4g18960

1.05463E-06

0.0079181

emb|X53579.1|

Floral homeotic protein agamous (AGAMOUS)

14 PROTEIN FATE (folding, modification, destination)

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

Pseudogene, putative GTP-binding protein

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

13863_at

At2g21470

9.36899E-07

0.0070342

gb|AAD23691.1|

Nearly identical to SUMO activating enzyme 2 (SAE2)

16 PROTEIN WITH BINDING FUNCTION OR COFACTOR REQUIREMENT (structural or catalytic)

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

18836_at

At2g24710

5.09671E-06

0.0382661

gb|AAD26894.1|

Plant glutamate receptor family (GLR2.3)

16262_at

At2g46850

4.75288E-06

0.0356846

gb|AAC34215.2|

Ser/Thr protein kinase -related

20 CELLULAR TRANSPORT, TRANSPORT FACILITATION AND TRANSPORT ROUTES

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

Pseudogene, putative GTP-binding protein

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

18836_at

At2g24710

5.09671E-06

0.0382661

gb|AAD26894.1|

Plant glutamate receptor family (GLR2.3)

17618_at

At2g31910

3.44193E-06

0.0258420

gb|AAD32281.1|

Similar to monovalent cation:proton antiporter family 2

30 CELLULAR COMMUNICATION/SIGNAL TRANSDUCTION MECHANISM

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

Pseudogene, putative GTP-binding protein

16816_at

At1g19230

5.65137E-06

0.0424305

gb|AAC39478.1|

Respiratory burst oxidase protein E (NADPH oxidase) (RbohE)

18836_at

At2g24710

5.09671E-06

0.0382661

gb|AAD26894.1|

Plant glutamate receptor family (GLR2.3)

19311_g_at

At2g41210

1.643E-06

0.0123356

gb|AAC78530.2|

Phosphatidylinositol-4-phosphate 5-kinase -related

13343_at

At1g34310

4.60448E-08

0.0003457

gb|AAD39615.1|

Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related

15787_s_at

At1g09090

3.64297E-07

0.0027351

gb|AAB70399.1|

Respiratory burst oxidase protein B (NADPH oxidase) (RbohB)

16262_at

At2g46850

4.75288E-06

0.0356846

gb|AAC34215.2|

Ser/Thr protein kinase -related

32 CELL RESCUE, DEFENSE AND VIRULENCE

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

Pseudogene, putative GTP-binding protein

12111_s_at

At4g19240

3.30499E-07

0.0024814

emb|CAA18611.1|

Expressed protein

12258_s_at

At4g14370

6.60533E-06

0.0495928

emb|CAB10216.1|

Disease resistance protein (TIR-NBS-LRR class)

12277_at

At1g47600

6.47203E-07

0.0048592

gb|AAD46026.1|

Glycosyl hydrolase family 1, similar to thioglucosidase

12956_i_at

At1g05170

3.64708E-06

0.0273823

gb|AAB71461.1|

Galactosyltransferase family

16375_at

At1g54480

6.28683E-06

0.0472015

gb|AAD25626.1|

Leucine rich repeat protein family contains leucine rich-repeat (LRR) domains

16816_at

At1g19230

5.65137E-06

0.0424305

gb|AAC39478.1|

Respiratory burst oxidase protein E (NADPH oxidase) (RbohE)

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

19244_s_at

At2g04230

2.84687E-06

0.0213743

gb|AAD27915.1|

F-box protein family, contains F-box domain

15224_at

At1g61540

8.21522E-08

0.0006168

gb|AAD25554.1|

Kelch repeat containing F-box protein family low similarity to SKP1 interacting partner 6

15787_s_at

At1g09090

3.64297E-07

0.0027351

gb|AAB70399.1|

Respiratory burst oxidase protein B (NADPH oxidase) (RbohB)

16263_at

At2g02320

3.12005E-06

0.0234253

gb|AAC78515.1|

F-box protein (SKP1 interacting partner 3-related)

17145_at

At1g10110

6.9462E-08

0.0005215

gb|AAC34337.1|

Contains Pfam PF00646: F-box domain; similar to F-box protein family, AtFBX7

12599_at

At2g29910

2.33543E-10

0.0000018

gb|AAD23631.1|

F-box protein family contains F-box domain Pfam:PF00646

34 INTERACTION WITH THE CELLULAR ENVIRONMENT

16816_at

At1g19230

5.65137E-06

0.0424305

gb|AAC39478.1|

Respiratory burst oxidase protein E (NADPH oxidase) (RbohE)

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

15787_s_at

At1g09090

3.64297E-07

0.0027351

gb|AAB70399.1|

Respiratory burst oxidase protein B (NADPH oxidase) (RbohB)

36 INTERACTION WITH THE ENVIRONMENT (systemic)

17946_s_at

At1g03410

2.84132E-06

0.0213326

gb|AAB97721.1|

2-Oxoglutarate-dependent dioxygenase, putative

38 TRANSPOSABLE ELEMENTS, VIRAL AND PLASMID PROTEINS

16731_at

At2g11690

1.06284E-06

0.0079798

gb|AAD28679.1|

Pseudogene

18340_at

At4g07700

2.79501E-06

0.0209849

gb|AAD29786.1|

Athila transposon protein -related

40 CELL FATE

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

41 DEVELOPMENT (systemic)

17677_at

At1g03910

1.54447E-06

0.0115959

gb|AAD10685.1|

Hypothetical protein

13216_s_at

At1g26310

8.05827E-07

0.0060501

gb|AAA64789.1|

Floral regulatory gene CAULIFLOWER

12863_r_at

At4g18960

1.05463E-06

0.0079181

emb|X53579.1|

Floral homeotic protein agamous (AGAMOUS)

42 BIOGENESIS OF CELLULAR COMPONENTS

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

Pseudogene, putative GTP-binding protein

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

13343_at

At1g34310

4.60448E-08

0.0003457

gb|AAD39615.1|

Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related

43 CELL TYPE DIFFERENTIATION

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

Pseudogene, putative GTP-binding protein

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

70 SUBCELLULAR LOCALIZATION

19689_at

At5g24140

7.0739E-07

0.0053111

emb|CAA06771.1|

Squalene monooxygenase 2 (squalene epoxidase 2) (SQP2) (SE2)

20254_at

At2g22390

2.41214E-06

0.0181104

gb|AAD22360.1|

Pseudogene, putative GTP-binding protein

18830_at

At2g32790

1.27302E-06

0.0095579

gb|AAC04484.1|

Ubiquitin-conjugating enzyme

18836_at

At2g24710

5.09671E-06

0.0382661

gb|AAD26894.1|

Plant glutamate receptor family (GLR2.3)

13343_at

At1g34310

4.60448E-08

0.0003457

gb|AAD39615.1|

Transcriptional factor B3 family protein / auxin-responsive factor AUX/IAA-related

13863_at

At2g21470

9.36899E-07

0.0070342

gb|AAD23691.1|

Nearly identical to SUMO activating enzyme 2 (SAE2)

15785_g_at

At1g08840

2.79162E-06

0.0209595

gb|AAB70418.1|

Hypothetical protein gene overlaps Sp6 end of F7G19

13216_s_at

At1g26310

8.05827E-07

0.0060501

gb|AAA64789.1|

Floral regulatory gene CAULIFLOWER

14356_at

At5g59370

3.6446E-08

0.0002736

gb|AAB39403.1|

Identical to SP|P53494 Actin 4

12863_r_at

At4g18960

1.05463E-06

0.0079181

emb|X53579.1|

Floral homeotic protein agamous (AGAMOUS)

No hits to TIGR gene prediction

20512_at

4.2379E-07

0.0031818

gb|AC002336.3|

Arabidopsis thaliana chromosome 2 clone T2P4 map CIC10A06, complete

18049_s_at

5.37206E-07

0.0040333

emb|AJ132404.1|

Arabidopsis thaliana anti-sense transcript, AKL kinase-like gene

<p>Figure 5</p>

Functional distribution of genes that are differentially regulated in leaves of the five accessions

Functional distribution of genes that are differentially regulated in leaves of the five accessions. Fifty-eight genes, identified by one-way ANOVA analysis, were subjected to MIPS functional classification based on their annotations.

Organ-specific gene expression in different accessions

In addition to identifying accession-specific genes, we were also interested in determining if there were genes whose expression is regulated by accession-by-organ interaction. In other words, we tried to test if the accession effect on gene expression is organ/development dependent. To address this question, two-way ANOVA analysis was performed. In one case, two samples from 2- and 5-week-old leaves, and two samples from 2- and 5-week-old roots were treated as replicates. In this two-way ANOVA study, the two factors are 'accessions' and 'organs'. For the 'accession' factor, there are five levels. For the 'organ' factors, there are eight levels (see Additional data file 4). The total mean squares for all the genes due to organ difference was 13,182.91 (df = 7), much greater than the total mean squares due to accession difference, which was equal to 2,936.21 (df = 4), consistent with our previous observation from the cluster analysis (Figure 3). The total mean square due to accession-by-organ interaction was only 436.00 (df = 28), suggesting that the effect of accession-by-organ interaction on gene expression might be small. Among the 296 genes that were found to have p-values less than 0.01 (FDR = 25.36%), 60 were further selected following Bonferroni correction to control the type-I error rate (Table 5), and subjected to functional classification.

<p>Table 5</p>

Genes whose expression is affected by accession-by-organ interaction, identified through two-way ANOVA analysis

Functional category

ATH1 hits

Pr(F)-accessions

Pr(F)-Organs

Pr(F)-accessions: organs

Bonferroni corrected Pr(F)-accessions: organs

Description

41 DEVELOPMENT (systemic)

18715_at

At1g14930

1.0285E-05

1.3523E-13

1.2253E-08

9.1998E-05

Major latex protein (MLP)-related low similarity to major latex protein

18229_at

At1g14940

5.4018E-07

3.5527E-15

3.3317E-10

2.5014E-06

Major latex protein (MLP)-related low similarity to major latex protein

18717_at

At1g14950

8.6683E-04

2.3537E-14

3.8173E-07

2.8661E-03

Major latex protein (MLP)-related low similarity to major latex protein

17893_at

At2g23110

2.2913E-06

1.7710E-10

3.0508E-07

2.2906E-03

Late embryogenesis abundant proteins -related

12731_f_at

At2g26960

1.1209E-09

4.2244E-09

1.6659E-09

1.2507E-05

MYB family transcription factor

20004_s_at

At2g35300

3.0731E-06

1.2479E-13

1.4412E-06

1.0820E-02

Late embryogenesis abundant proteins -related identical to GB:X91917

13674_s_at

At2g36640

9.1097E-07

1.4619E-11

8.2287E-07

6.1781E-03

Nearly identical to LEA protein in group 3

17038_s_at

At2g36640

6.4830E-06

1.4944E-10

5.8337E-06

4.3799E-02

Nearly identical to LEA protein in group 3

16896_s_at

At2g41260

3.9707E-11

1.1213E-14

3.9237E-10

2.9459E-06

Glycine-rich, identical to late-embryogenesis abundant M17 protein GI:3342551

19355_s_at

At2g41280

3.7790E-09

6.5988E-10

3.4337E-07

2.5780E-03

Late embryogenesis abundant M10 protein identical to GB:AF076979

15747_at

At2g42560

5.1854E-08

6.4206E-12

3.2085E-07

2.4090E-03

Late embryogenesis abundant (LEA) domain-containing protein

15604_s_at

At3g15400

4.6444E-07

4.5835E-09

3.5477E-06

2.6636E-02

Identical to anther development protein ATA20 GB:AAC50042

19918_at

At3g15670

3.7835E-08

1.5210E-14

4.4004E-08

3.3038E-04

Similar to SP|P13934 Late embryogenesis abundant protein 76 (LEA 76)

18872_at

At3g17520

4.2978E-10

1.1102E-16

2.5048E-09

1.8806E-05

Low similarity to PIR|S04045|S04045 embryonic abundant protein D-29

17282_s_at

At3g51810

1.4069E-08

1.5599E-13

6.4057E-10

4.8094E-06

Embryonic abundant protein AtEm1

20682_g_at

At4g26740

6.4621E-04

2.8866E-15

1.0571E-06

7.9364E-03

Embryo-specific protein 1 (ATS1) putative Ca2+-binding EF-hand protein

13675_s_at

At3g22500

8.1351E-08

7.9450E-09

7.7592E-07

5.8256E-03

LEA protein, putative

04 STORAGE PROTEIN

18295_s_at

At1g03880

3.5027E-08

0.0000E+00

1.9892E-10

1.4935E-06

12S seed storage protein (CRB)

13200_s_at

At1g03880

2.0300E-05

2.4425E-15

1.7983E-07

1.3502E-03

12S seed storage protein (CRB)

20221_at

At1g03890

8.7822E-06

5.1070E-15

2.5720E-07

1.9311E-03

Globulin (seed storage protein) family similar to Arabidopsis thaliana 12S seed storage proteins SP|P15455

20222_g_at

At1g03890

2.3617E-05

2.7756E-15

2.5729E-07

1.9317E-03

globulin (seed storage protein) family similar to Arabidopsis thaliana 12S seed storage proteins SP|P15455

20535_s_at

At2g28490

2.4914E-03

1.1269E-13

3.8367E-06

2.8806E-02

Cupin domain-containing protein similar to preproMP27-MP32 [Cucurbita cv. Kurokawa Amakuri]

15983_s_at

At4g27140

3.1858E-04

1.4433E-15

2.6036E-07

1.9547E-03

2S seed storage protein 1 (NWMU1 - 2S albumin 1) identical to SP|P15457

15984_s_at

At4g27170

8.4937E-06

0.0000E+00

6.1932E-09

4.6498E-05

2S seed storage protein 4 (NWMU2-2S albumin 4) identical to SP|P15460

13449_at

At4g36700

1.5016E-05

2.9865E-14

3.3621E-06

2.5242E-02

Cupin domain-containing protein low similarity to preproMP27-MP32 from Cucurbita cv. Kurokawa Amakuri

16025_s_at

At4g28520

6.5162E-09

0.0000E+00

2.2615E-10

1.6980E-06

12S seed storage protein (cruciferin), putative

16425_s_at

At5g44120

2.4424E-08

6.1062E-15

3.4512E-07

2.5912E-03

12S seed storage protein (CRA1)

13201_at

At5g54740

3.4456E-08

0.0000E+00

1.8704E-11

1.4043E-07

2S seed storage protein family protein

13194_at

At4g27160

1.0828E-06

5.7732E-15

2.4480E-07

1.8380E-03

NWMU3 - 2S albumin 3 precursor, seed storage protein AT2S3

13198_i_at

At4g28520

4.1773E-07

4.8295E-14

8.3466E-08

6.2666E-04

12S cruciferin seed storage protein

13199_r_at

At4g28520

9.8653E-08

1.0880E-14

1.8093E-08

1.3585E-04

12S cruciferin seed storage protein

32 CELL RESCUE, DEFENSE AND VIRULENCE

14789_at

At2g15010

1.0120E-04

1.2166E-12

4.2917E-06

3.2222E-02

Similar to thionin [Arabidopsis thaliana] gi|1181533|gb|AAC41679

18715_at

At1g14930

1.0285E-05

1.3523E-13

1.2253E-08

9.1998E-05

Low similarity to major latex protein {Papaver somniferum}

18229_at

At1g14940

5.4018E-07

3.5527E-15

3.3317E-10

2.5014E-06

Low similarity to major latex protein {Papaver somniferum}

18717_at

At1g14950

8.6683E-04

2.3537E-14

3.8173E-07

2.8661E-03