Genomic Basis of Transcriptome Dynamics in Rice under Field Conditions

Abstract

How genetic variations affect gene expression dynamics of field-grown plants remains unclear. Expression quantitative trait loci (eQTL) analysis is frequently used to find genomic regions underlying gene expression polymorphisms. This approach requires transcriptome data for the complete set of the QTL mapping population under the given conditions. Therefore, only a limited range of environmental conditions is covered by a conventional eQTL analysis. We sampled sparse time series of field-grown rice from chromosome segment substitution lines (CSSLs) and conducted RNA sequencing (RNA-Seq). Then, by using statistical analysis integrating meteorological data and the RNA-Seq data, we identified 1,675 eQTLs leading to polymorphisms in expression dynamics under field conditions. A genomic region on chromosome 11 influences the expression of several defense-related genes in a time-of-day- and scaled-age-dependent manner. This includes the eQTLs that possibly influence the time-of-day- and scaled-age-dependent differences in the innate immunity between Koshihikari and Takanari. Based on the eQTL and meteorological data, we successfully predicted gene expression under environments different from training environments and in rice cultivars with more complex genotypes than the CSSLs. Our novel approach of eQTL identification facilitated the understanding of the genetic architecture of expression dynamics under field conditions, which is difficult to assess by conventional eQTL studies. The prediction of expression based on eQTLs and environmental information could contribute to the understanding of plant traits under diverse field conditions.

Keywords: Environmental response; Oryza sativa; RNA-Seq; Statistical modeling; eQT.

Fig. 1

Concept and workflow of eQTLs detected in this study. (A) Conceptual differences in the cover range of eQTLs identified with the conventional and our novel approach. (B) Workflow of eQTL detection and its evaluation using CSSLs. (C) Summary of the sampling design for eQTL detection. The left panels show plots of meteorological data [air temperature (Temp., °C) and global solar radiation (Rad., kJ m⁻² min⁻¹)] in Takatsuki from May to September in 2015. Vertical red lines represent the sampling time points. (D) Pearson’s correlations among the 854 samples used for developing the prediction model based on expression data of 23,294 expressed genes. White lines indicate the border of each bihourly sampling set.

Fig. 2

eQTL detection in this study. (A) Predictive models of gene expression for ‘Koshihikari’ and ‘Takanari’ were developed with ‘FIT’ based on RNA-Seq data (observed log₂rpm), the corresponding precision weights, meteorological data and scaled age. Then, based on the models with input of meteorological data and sample attributes, predicted log₂rpm of ‘Koshihikari’ and ‘Takanari’ can be obtained. (B) The association between genetic variations and gene expression polymorphisms was evaluated by calculating the sum of residual errors in gene expression prediction. It was assumed that the type of each gene expression dynamics is determined by SSR markers. The color of the enclosing lines of the circles, triangles and quadrangles indicate the BGs of each line. The fill color of the circles, triangles and quadrangles indicate which ‘Koshihikari’ or ‘Takanari’ model is used to predict gene expression. In the example, eQTL affecting gene i exist around SSR marker 3. In this case, the sum of residual errors on the assumption that eQTL is SSR marker 3 is smaller than the other residual errors.

Fig. 3

eQTL detection revealed cis- and trans-eQTLs that were involved in environmental responses. A, C, E, Observed and predicted expressions (log₂rpm) of Os09g0343200 (A, C) and Os01g0537250 (E) in Takatsuki in 2015. The predicted expressions were calculated using ‘FIT’ based on scaled age and environmental information (time, air temperature and global solar radiation). Blue and red/pink lines indicate predicted expression levels in ‘Koshihikari’ and ‘Takanari’ in transplant sets 1, 2, 3 and 4, respectively. Blue and red/pink points indicate the expression level obtained by RNA-Seq for samples in transplant sets 1, 2, 3 and 4 of individuals with ‘Koshihikari’ BG and ‘Takanari’ BG in (a) or ‘Koshihikari’- and ‘Takanari’-type eQTL in (C, E) respectively. In (C, E) strong colored points emphasized by the arrowheads indicate samples harboring eQTLs different from their BGs, and light-colored points indicate the samples harboring eQTLs identical to their BGs. The upper gray bar indicates dark periods (global solar radiation < 0.3 kJ m⁻² min⁻¹). (B, F) eQTLs regulating Os09g0343200 (B) and Os01g0537250 (F). (D) Position of the 1,675 eQTLs are shown as red bars (false discovery rate = 0.05). X-axis and Y-axis represent the positions of markers with eQTLs and the positions of genes influenced by eQTLs, respectively.

Fig. 4

eQTL-based prediction of gene expression dynamics under different environments. (A) Examples of prediction of expression dynamics in Kizugawa in 2016 based on environmental information and eQTLs in transplant set 1. Points in intense colors emphasized by arrowheads indicate samples harboring eQTLs different from their BGs and light colors indicate samples harboring eQTLs identical to their BGs. The upper gray bars indicate dark periods (global solar radiation < 0.3 kJ m⁻² min⁻¹). (B) Effects of eQTLs on gene expression prediction dynamics in Kizugawa in 2016. Genes influenced by eQTLs are in the order of the positions on the chromosomes along the horizontal axis.

Fig. 5

eQTL-based prediction in cultivars with more complex genotypes than the CSSLs. (A–D) Prediction accuracy of gene expression regulated by all eQTLs (A, B) and trans-eQTLs (C, D) based on eQTL model for HP-a (A, C) and HP-b (B, D). The blue, red and orange vertical lines indicate the sums of prediction errors based on ‘Koshihikari’, ‘Takanari’ and eQTL models. The histogram shows the distribution of the sums of prediction errors based on the eQTL model in 10,000 permutations of markers in HP-a or HP-b genomes. The dashed vertical line indicates the 0.1% percentile of the distribution. (E) eQTL for Os03g0388300 and genotypes of HP-a and HP-b. Dark blue points indicate significant ‘Koshihikari’-type markers. (F) Prediction of Os03g0388300 expression in HP-a and HP-b. Intense color lines are applied models for HP-a and HP-b.