An integrative statistical method to explore herbivory-specific responses in plants

Abstract

Spatial-temporal coordination between multiple processes/pathways is a key determinant of whole-organism transcriptome and metabolome reconfigurations in plant’s response to biotic stresses. To explore tissue-based interdependencies in Nicotiana attenuata’s resistance to insect attack, we performed time course analyses of the plant’s transcriptome and metabolome in herbivory-elicited source leaves and unelicited sink leaves and roots. To dissect the multidimensionality of these responses, we have recently designed a novel approach of constructing interactive motifs by combining an extended self-organizing maps (SOM) based dimensionality reduction method with bootstrap-based non-parametric ANOVA models. In this previous study, we used this method to study nonlinearities in gene-metabolite associations involved in the acyclic diterpene glucoside pathway. Here, we extend the application of this method to the extraction of genes showing herbivory-elicitation specifically in systemic (distal from the treatment sites) tissues using motif analysis for different combinations of treatment applied to Nicotiana attenuata.

Plants have evolved efficient defense strategies which involve rapid changes in intricately connected signaling and metabolic networks.^1-4 Experiments designed to study such intricate networks often have a complex factorial structure, obtained by assessing plant responses in different conditions/treatments, tissue types or genetic contexts. Among the major problems associated with the statistical analysis of multifactorial experimental designs are those of gene prioritization and of large numbers of false positives.^5,6 Since signaling pathways underlying major stress responses are generally affected only by a subset of the experimental conditions⁷ and are based on transient gene associations, therefore these cannot be captured using collective information studies. Although bioinformatic approaches such as mutual information⁸ and biclustering⁹ have been developed to address this limitation, gene networks assemble dynamically as the organism adapts to external stimuli and therefore their analysis necessitates the mining of multifactorial time series experiments. Several efforts including those by Park et al.,¹⁰ Wang and Kim¹¹ and Tai and Speed¹² have been published to assess single factorial effect on gene expression in a time course experiment. Zhou et al.¹³ have developed a method to simultaneously analyze experiments involving more than one factor measured across time series by finding the significant direction in the time course across different conditions.

Nicotiana attenuata is an annual fire-chasing plant native to the Great Basin Desert of the southwestern United States of America which has evolved a large number of specific induced responses against generalist and specialist herbivores.^14,15 Some of the essential nodes in the plant’s transcriptome and metabolome responses to attack from larvae of the specialist lepidopteran herbivore, Manduca sexta, have been functionally characterized.¹⁶ Feeding by this specialist herbivore or its simulation by the application of its oral secretions (OS) into puncture wounds produced by mechanical wounding activate rapid changes in the plant’s metabolic and growth processes in order to facilitate de novo production of defense compounds.

In a recent study,¹⁷ we profiled the transcriptome and metabolome of identically treated wild type Nicotiana attenuata plants for 3 tissues and 2 stress conditions (mechanical wounding and simulated herbivory) with a regular time series of 6 time points. To investigate the dynamics of activation in time and space of herbivory-induced changes in gene-to-metabolite networks, we employed a bootstrap-based non-parametric ANOVA (NANOVA) model designed to find gene/metabolite-specific responses across the time series based on their dependency on experimental factors used for comparison.¹³ We conducted dynamic response analyses taking control (Ctrl) and OS treated samples (W+OS) for 2 tissue comparisons: treated (source) leaf vs. untreated systemic (sink) leaf (TvS comparison) to explore differential gene expression patterns activated during shoot systemic signaling, and treated leaf vs. untreated roots (TvR comparison) to obtain novel insights into root specific responses. Using a series of statistical tests on factor effects,¹³ we divided the transcriptome/metabolome into 4 mutually exclusive groups showing their best ANOVA structure along the estimated optimal direction in the time series. The 4 resulting structures represent interactive (tissues behaving differently in response to OS-elicitation across the time series), additive (herbivore responses independent of tissue type), or corresponding main effects on gene expression (major treatment effects in both treated and untreated tissue or significant differences in tissue type with no response to treatment). Set of genes and metabolites displaying interactive response patterns were further studied. With this approach, we captured the dynamic response of a gene in more than one tissue in terms of a single metric which was then used to delineate elements of signaling pathway and to analyze activation transition points between different sub-branches of a single pathway. Next, we imposed structure on the data using batch learning self-organizing maps (BL-SOM)¹⁸ to obtain interactive motifs which are defined as patterns of interconnections between genes and metabolites that are differentially perturbed in local and systemic tissues in response to stress, additional information of their time of action having been obtained from projected data on time series termed as ANOVA directions. Since dynamic responses for signaling/metabolic pathways are considered highly coordinated, we hypothesized that the nodes in these spatio-temporally resolved motifs with similar ANOVA directions along the time series may reflect biological organization. We isolated interactive motifs from the SOM grids for genes involved in the biosynthesis of defense metabolites of the 17-hydroxygeranyllinalool diterpene glycoside (17-HGL-DTG) class^19,20 and analyzed their dynamic behaviors using network analysis. Metabolic analyses conducted in parallel further supported the advantage of constructing dynamic correlation network based on response features captured by the factorial analysis. Specifically, we identified multiple clusters of biochemically-connected metabolites that shared similar time response metric and grouped according to their inferred compound-family-wise grouping of m/z ions for 17-HGL-DTG, O-acyl sugars, shikimate pathway-derived amino acids and downstream metabolites produced within the phenylpropanoid pathway.

Here, we illustrate the extension of this statistical method to study herbivory-specific plant responses in treated and untreated leaves. Additionally to the time course transcriptome data collected from the W+OS treatment type, we considered transcriptome data for leaf tissues (treated and untreated leaves) that had been wounded and treated with water (W+W) and collected at 3 time points (2 early time points, 1 h and 5 h after treatment; one late time point, 17 h after treatment; Fig. 1A). By applying this novel method we differentiated OS-specific systemic response (W+OS) from those inherent to the mechanical wounding (W+W) by extracting the set of genes showing an interactive effect to the treatment type along the time series for the comparison between treated and untreated leaf tissues. OS-specific responses dependent on tissue type are more clearly visible for early time points (1 h and 5 h after treatment) in a hierarchical cluster analysis of genes showing interactive effect (Fig. 1A). To assess the functional significance of these genes, we computed the enrichment of GO terms using hyper-geometric tests (f < 0.05) (Fig. 1A). As expected from previously characterized mechanisms of biotic stress adaptation in this plant species, the group of genes showing OS-specific systemic responses were highly enriched for processes associated with stress responses, hormone metabolism, secondary metabolism, photosynthetic pathway and amino-acid metabolism. We next mapped the set of genes showing an interactive effect to the W+OS treatment type when compared with the W+W condition onto the SOM map which was generated earlier for the analysis of the TvS (treated vs systemic leaf response) comparison between Control and W+OS treatment type¹⁷ and detected these genes of interest being localized into the nodes from interactive motifs 1a, 1b, 5a, and 5c which have been shown earlier to be enriched for stress, signaling and secondary metabolic pathways. Since we found OS-specific systemic responses being more pronounced at 1 h and 5 h, we resolved the map by filtering genes based on their time response metrics (> = 0.5) for these 2 time points. Genes showing large effects at 1 h are in motifs 1a and 1b with many of them being up-regulated and few down-regulated in treated leaf tissues while those showing large effects at 5 h are in motifs 5c and only few from motif 1a. The pathway for diterpene glycoside, present in motif 5c, is well studied and has been presented in an earlier study.¹⁷ Figure 2 presents the comparative time response behavior of genes involved in the phenylpropanoid pathway for 2 comparisons: (A) Control vs. W+OS (combined herbivory and mechanical wounding responses), (B) W+OS vs. W+W (OS-specific responses). Activation of gene expression in response to OS in treated leaves 1 h post elicitation is supported by the high value of the response metric at 1 h while the differential activation of these genes specifically to OS at 5 h in untreated leaf tissue in response to application of OS is reflected by the high value of response metric obtained from W+OS vs. W+W comparison at 5 h.

Figure 1.(A) Implementation of a statistical method designed to study OS-elicitation specific responses in untreated tissues in Nicotiana attenuata: Replicated transcriptomic and metabolomic data were analyzed using multi-factorial analysis with both factors (tissue and treatment) taken together across the time series to identify modules showing differential OS-elicitation. Heatmap represents hierarchical cluster analysis of genes showing interactive effect for W+OS and W+W condition when compared for treated and untreated leaf tissue. Temporal profile of NaPAL1 represents the specific pattern, OS specific response at 5 h after elicitation, extracted using this method. Pie chart represents the distribution of enriched GO terms for genes showing interactive effect. (B) Localization of genes showing OS-specific responses dependent on tissue type in different interactive motifs: Probes from interactive bin for the analysis comparing 2 factors (treatment, W+OS vs. W+W; tissue, Treated leaf vs. systemic leaf) were mapped onto SOM grids and found overrepresented in motifs 1a, 1b, 5a and 5c. This set is further divided into 2 sub-groups based on their time response. (A) greater than 0.5 at 1h after treatment; (B) greater than 0.5 at 5h after treatment. Set (B) is represented by 2 well-studied secondary metabolic pathways—the acyclic diterpene glycoside and phenylpropanoid pathways.

Figure 2. OS elicitation selectively activates genes in the phenylpropanoid pathway at 5 h after elicitation. Schematic representation of the phenylpropanoid pathway with the color coded representation of interactive effect response metric for two different comparisons: (A) Control vs. W+OS, (B) W+OS vs. W+W. Activation of gene expression in response to OS in treated leaves 1 h post elicitation is supported by the high value of the response metric at 1 h while the differential activation of these genes specifically to OS at 5 h in untreated leaf tissue in response to application of OS is reflected by the high value of response metric obtained from W+OS vs. W+W comparison at 5 h.

This broadly applicable approach allows identifying complex interdependencies between metabolites and transcripts with a high level of accuracy and robustness. The application presented here for the identification of herbivory-specific responses in leaves distal to the plant treatment sites provides an additional support for the importance of finely-tuned changes in metabolism throughout the complete plant during stress adaptation.

Glossary

Abbreviations:

OS

oral-secretion

SOM

self-organizing maps

ANOVA

analysis of variance

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.