Genomewide identification of genes under directional selection: gene transcription Q(ST) scan in diverging Atlantic salmon subpopulations

Abstract

Evolutionary genomics has benefited from methods that allow identifying evolutionarily important genomic regions on a genomewide scale, including genome scans and QTL mapping. Recently, genomewide scanning by means of microarrays has permitted assessing gene transcription differences among species or populations. However, the identification of differentially transcribed genes does not in itself suffice to measure the role of selection in driving evolutionary changes in gene transcription. Here, we propose and apply a "transcriptome scan" approach to investigating the role of selection in shaping differential profiles of gene transcription among populations. We compared the genomewide transcription levels between two Atlantic salmon subpopulations that have been diverging for only six generations. Following assessment of normality and unimodality on a gene-per-gene basis, the additive genetic basis of gene transcription was estimated using the animal model. Gene transcription h(2) estimates were significant for 1044 (16%) of all detected cDNA clones. In an approach analogous to that of genome scans, we used the distribution of the Q(ST) values estimated from intra- and intersubpopulation additive genetic components of the transcription profiles to identify 16 outlier genes (average Q(ST) estimate = 0.11) whose transcription levels are likely to have evolved under the influence of directional selection within six generations only. Overall, this study contributes both empirically and methodologically to the quantitative genetic exploration of gene transcription data.

Figure 1.—

Assessment of normality and unimodality. (A) Relationship between Pearson's determination coefficient between a normal distribution and the observed distribution of normalized gene transcription data, on the one hand, and the inverse of the log of the P-value from the Shapiro–Wilk normality test of gene transcription data, on the other hand, for each of the 6484 cDNA clones corresponding to genes for which expression was detected. The horizontal line indicates the position of the Bonferroni-corrected significance threshold for the Shapiro–Wilk test. (B) Density curve of Hartigan's dip statistic testing for unimodality of the gene transcription data for all detected genes. Gene transcription data from genes at the right of the vertical line (dip > 0.054, P < 0.05) show significant departure from unimodality. (C) Density curves of the normalized gene transcription data of the 90 sampled individuals for the four genes with the highest dip statistic. The corresponding gene products are (i) an unknown gene product (GenBank accession number of the cDNA sequence: CA059553), (ii) ribosomal protein L35, (iii) ubiquitin, and (iv) α-actin 2.

Figure 2.—

Gene transcription heritability. (A) Distribution of heritability estimates of the normalized transcription profiles for the 6484 cDNA clones corresponding to genes for which expression was detected. (B) Relationship between h² and its standard error for each of the 6484 detected cDNA clones.

Figure 3.—

Gene transcription Q_ST. Distribution of Q_ST estimates of the normalized transcription profiles of the 1044 cDNA clones representing genes with significant gene transcription heritability (P ≤ 0.05)

Figure 4.—

Relationships between gene transcription Q_ST, heritability, and the significance from the ANOVA for the 1044 genes with significant heritability estimates. Scatterplots illustrating the relationships between (A) gene transcription Q_ST estimates and the inverse of the log of the P-values from an ANOVA testing for significant gene transcription-level differences between the progeny of fish from the two subpopulations, (B) gene transcription Q_ST and heritability estimates, and (C) gene transcription heritability estimates and the inverse of the log of the P-values from the above-mentioned ANOVA.