Genomic survey of gene expression diversity in Arabidopsis thaliana

Abstract

Differential gene expression controls variation in numerous plant traits, such as flowering time and plant/pest interactions, but little is known about the genomic distribution of the determinants of transcript levels and their associated variation. Affymetrix ATH1 GeneChip microarrays representing 22,810 genes were used to survey the transcriptome of seven Arabidopsis thaliana accessions in the presence and absence of exogenously applied salicylic acid (SA). These accessions encompassed approximately 80% of the moderate- to high-frequency nucleotide polymorphisms in Arabidopsis. A factorial design, consisting of three biological replicates per accession for the two treatments at three time points (4, 28, and 52 hr post-treatment), and a total of 126 microarrays were used. Between any pair of Arabidopsis accessions, we detected on average 2234 genes (ranging from 1428 to 3334) that were significantly differentially expressed under the conditions of this experiment, using a split-plot analysis of variance. Upward of 6433 genes were differentially expressed between at least one pair of accessions. These results suggest that analysis of additional genetic, developmental, and environmental conditions may show that a significant fraction of the Arabidopsis genome is differentially expressed. Examination of sequence diversity demonstrated a significant positive association with diversity in gene expression.

Figure 1.

Sampling of species variation. The percentage of polymorphisms in 96 Arabidopsis accessions that are polymorphic in the six accessions in common with this data set (Col-0, Cvi-1, Kin-0, Mt-0, Tsu-1, and Van-0) are shown (Nordborg et al. 2005). The alleles are binned on the basis of the minor allele percentage in the 96-accession data set (shown on the x-axis). The smooth line represents the expected random-sampling distribution for 6 accessions for each allele percentage class.

Figure 2.

Differential gene expression per pairs of accessions. The number of genes showing differential expression per accession pair is shown. The left y-axis and bars show the number of genes showing polymorphic expression, as determined by a significant H₀₁ (differentially expressed between two accessions), with significance determined by Holm's test (see materials and methods for details). The right y-axis and diamonds show the number of genes with expression polymorphisms that satisfy H₀₁ with the significance level determined by the FDR test. The accession pair is listed on the x-axis: C, Col-0; I, Cvi-1; E, Est; K, Kin-0; M, Mt-0; T, Tsu-1; and V, Van-0. Avg is the average across all pairwise comparisons.

Figure 3.

Frequency of pairwise accession gene expression differences. The horizontal axis indicates the number of accession pairs (n = 21 pairs) for which the genes are detected as differentially expressed. Solid bars denote the number of genes identified with Holm's procedure; open bars denote the number of genes identified with the FDR procedure.

Figure 4.

GO biological annotation comparison for differentially expressed genes. Solid bars show the percentage of total annotations in each GO biological annotation classification using the completely sequenced Arabidopsis genome from Col-0. Shaded bars (with standard errors) show the average percentage of total annotations for each of the above GO biological classifications using all 21 comparisons between pairs of accessions. Asterisks indicate those classifications that are significantly different from the whole-genome analysis as determined by χ²-tests, with Bonferroni correction. GO biological classifications that describe unknown proteins were not included (i.e., other physiological processes, other metabolic processes, other cellular processes, and biological processes unknown) since they were uninformative.

Figure 5.

Sequence divergence and gene expression differences are associated. The relationship between π and number of ELPs between pairs of six accessions (Col-0, Cvi-1, Kin-0, Mt-0, Tsu-1, and Van-0), as determined by linear regression and FDR gene list (n = 15, P < 0.001, r² = 0.83), is shown. Pairwise comparisons containing Col-0 are shown as open circles; solid circles denote pairs that do not include Col-0. A similar linear relationship was identified with Holm's gene list (n = 15, P < 0.001, r = 0.80).

Figure 6.

Local chromosomal association between nucleotide and differential gene expression variation. Sliding-window comparison of nucleotide variation (π) vs. differential gene expression in the 6 accessions (Col-0, Cvi-1, Kin-0, Mt-0, Tsu-1, and Van-0) in common with the 96 accessions from the sequence database is shown (Nordborg et al. 2005). The black line is the average number of ELPs per locus measured in a 500-gene-wide sliding window. The x-axis of each graph is in megabase pairs as marked. The horizontal lines represent the P = 0.05 thresholds for the differentially expressed sliding window as obtained via the permutation analysis (0.056 is the P = 0.05 minimum and 0.132 is the P = 0.05 maximum). The gray line represents a 5-Mbp sliding-window analysis of average π per loci.