Molecular insights into plant desiccation tolerance: transcriptomics, proteomics and targeted metabolite profiling in Craterostigma plantagineum

Summary

The resurrection plant Craterostigma plantagineum possesses an extraordinary capacity to survive long‐term desiccation. To enhance our understanding of this phenomenon, complementary transcriptome, soluble proteome and targeted metabolite profiling was carried out on leaves collected from different stages during a dehydration and rehydration cycle. A total of 7348 contigs, 611 proteins and 39 metabolites were differentially abundant across the different sampling points. Dynamic changes in transcript, protein and metabolite levels revealed a unique signature characterizing each stage. An overall low correlation between transcript and protein abundance suggests a prominent role for post‐transcriptional modification in metabolic reprogramming to prepare plants for desiccation and recovery. The integrative analysis of all three data sets was performed with an emphasis on photosynthesis, photorespiration, energy metabolism and amino acid metabolism. The results revealed a set of precise changes that modulate primary metabolism to confer plasticity to metabolic pathways, thus optimizing plant performance under stress. The maintenance of cyclic electron flow and photorespiration, and the switch from C₃ to crassulacean acid metabolism photosynthesis, may contribute to partially sustain photosynthesis and minimize oxidative damage during dehydration. Transcripts with a delayed translation, ATP‐independent bypasses, alternative respiratory pathway and 4‐aminobutyric acid shunt may all play a role in energy management, together conferring bioenergetic advantages to meet energy demands upon rehydration. This study provides a high‐resolution map of the changes occurring in primary metabolism during dehydration and rehydration and enriches our understanding of the molecular mechanisms underpinning plant desiccation tolerance. The data sets provided here will ultimately inspire biotechnological strategies for drought tolerance improvement in crops.

Significance Statement

This study provides a transcriptomic, proteomic and metabolic signature of Craterostigma plantagineum leaves during a dehydration and rehydration cycle. Integrative analysis of all three data sets reveals a set of precise changes that modulate primary metabolism to confer plasticity to metabolic pathways, thus optimizing plant performance under stress. The data provided here are a step towards a systems biology approach to understand desiccation tolerance and will ultimately inspire biotechnological strategies for drought tolerance improvement in crops.

​

INTRODUCTION

Drought is one of the most devastating constraints in crop productivity (Lobell et al., ²⁰¹¹; Zhu, ²⁰⁰²). Continuous research has been devoted to developing crops with improved drought tolerance to meet increasing food demands under climate change. In this respect, a comprehensive understanding of how plants adapt to water loss is a prerequisite. Resurrection plants serve as model organisms to study plant desiccation tolerance because of their ability to survive prolonged periods of desiccation and to recover upon rehydration.

Approximately 300 species of resurrection plants are known. They can be assigned to two types: poikilochlorophyllous and homoiochlorophyllous. Poikilochlorophyllous plants degrade chlorophyll and lose photosynthetic structures during dehydration, whereas homoiochlorophyllous plants retain most of the chlorophyll and retain intact thylakoid membranes (Tuba et al., ¹⁹⁹⁸). Craterostigma plantagineum, a dicotyledonous plant native to rocky outcrops of sub‐Saharan Africa, is poikilochlorophyllous. This plant has been studied extensively (Bartels and Salamini, 2001; Giarola and Bartels, ²⁰¹⁵) for two reasons: genetic transformation is feasible (Furini et al., ¹⁹⁹⁴); and desiccation tolerance is present in both differentiated plants and undifferentiated callus, thus allowing studies in two genetically identical systems. Despite these positive features, genome‐scale analyses have been largely hampered by its highly complex octoploid genome (1027 Mb) (VanBuren et al., ²⁰¹⁸).

Some high‐throughput and large‐scale studies have been carried out on resurrection plants to investigate the metabolic changes that occur during dehydration and rehydration. Transcriptomic analyses have been conducted on C. plantagineum (Rodriguez et al., ²⁰¹⁰) and Haberlea rhodopensis (Gechev et al., ²⁰¹³; Liu et al., ^2018a). Proteomic analyses have been performed on Boea hygrometrica (Jiang et al., ²⁰⁰⁷), Selaginella tamariscina (Wang et al., ²⁰¹⁰), Sporobolus stapfianus (Oliver et al., ²⁰¹¹) and Xerophyta viscosa (Ingle et al., ²⁰⁰⁷). Metabolomic analyses of H. rhodopensis (Gechev et al., ²⁰¹³) and Selaginella lepidophylla (Yobi et al., ²⁰¹³) have also been reported. So far, to our knowledge, only one integrative analysis of the transcriptome and metabolome has been reported for H. rhodopensis; other studies have only captured changes in transcripts, proteins or metabolites. Comparison across different species resulted in the identification of some genes, proteins and metabolites that are similarly affected (Challabathula and Bartels, 2013); however, the lack of phylogenetic linkage between resurrection plants means that these studies cannot be simply integrated. Therefore, a multi‐omics study on the same experimental set‐up is needed for an integrated understanding of dehydration‐induced physiological phenomena and a comprehensive picture of mechanisms underlying desiccation tolerance.

In the present study, temporal profiles of the transcriptome, soluble proteome and primary metabolites of C. plantagineum leaves were investigated at four stages across a dehydration and rehydration cycle. We provide a transcriptomic, proteomic and metabolic signature of the leaves at each stage. Correlation analysis of transcript and protein levels was performed and temporal dynamics between gene expression and protein abundance profiles are described. An integrative analysis of three data sets was performed, with an emphasis on primary metabolism.

RESULTS AND DISCUSSION

Morphological changes of leaves upon dehydration and rehydration

To better understand the morphological changes of C. plantagineum leaves and their accordion‐like folding upon dehydration and rehydration, leaves at different hydration stages were studied: untreated leaves (F, relative water content, RWC ~ 100%; Figure 1a), partially dried leaves (D1, RWC ~ 60%; Figure 1b), desiccated leaves (D2, RWC ~ 2%; Figure 1c) and 24‐h rehydrated leaves (R1; Figure 1d).

Figure 1.

Morphological changes of plants (a–d) and cross sections of leaves (e–h) during the dehydration and rehydration cycle. Cross sections were stained with FASGA dye and observed with light microscopy. The chloroplasts are indicated with arrows in the inset of (g). Abbreviations: is, intercellular space; lep, lower epidermis cell; pc, parenchyma cell; RWC, relative water content; uep, upper epidermis cell; v, vascular bundle.

Cross sections of samples were stained with FASGA and observed using light microscopy. Images were captured in the leaf midrib region. Untreated leaves contained round and turgid cells with large intercellular spaces (Figure 1e), whereas the cells in partially dried leaves appeared closer to each other, with an angular shape, thus leaving no intercellular spaces (Figure 1f). Severe shrinkage was observed in the desiccated sample (Figure 1c,g). Although individual cells, except epidermal cells, cannot be easily identified as a result of the cell shrinkage, chloroplasts with more rounded shapes were visible in the cells (Figure 1g). After 24 h of rehydration, cells displayed a similar morphology to those of the partially dried leaf (Figure 1d,h). The vascular tissue was visible at all stages, with unchanged cell shape (Figure 1e–h). These observations are consistent with those reported by Schneider et al. (1993).

Transcriptome sequencing, assembly and annotation

A de novo transcriptome assembly of C. plantagineum was performed to generate a reference for high‐throughput gene expression analysis and the identification of proteins. The transcriptome contained a total of 48 045 contigs with an N50 (the largest contig size at which at least 50% of bases are contained in contigs at least this length) of 868 bp and an average length of 730 bp (Data S1). About 79.4% of total reads were mapped back onto this transcriptome, which is good given the polyploidy level of C. plantagineum.

The BLAST analysis of the transcriptome against Viridiplantae resulted in the annotation of 40 906 contigs (85.14%). Among them, 47.92% of transcripts showed top‐hit BLAST results against Sesamum indicum, followed by Erythranthe guttata (16.47%) (Data S1), similar to previous results for the transcriptome of H. rhodopensis (Liu et al., ^2018a). To perform a Gene Ontology enrichment (GOE) analysis, a second annotation was performed against Arabidopsis thaliana resulting in 77.42% of annotated contigs. The transcriptome presented herein highly enriches the previous transcriptome generated using pyrosequencing (Rodriguez et al., ²⁰¹⁰) by obtaining 1.6‐fold more contigs and 2.0‐fold more contigs with annotation. This transcriptome thus provides a better reference for RNA‐Seq and proteome analyses.

Transcriptome profiles of the leaves upon dehydration and rehydration

A total of 7348 differentially expressed contigs in response to dehydration and rehydration were generated using a stringent threshold false discovery rate (FDR) of P < 0.001 and at least a twofold change between two or more time points (Data S1). To validate the RNA‐Seq data, real‐time quantitative reverse transcription PCR (qRT‐PCR) analysis was performed on 14 genes encoding enzymes and regulators involved in stress response and cell wall biogenesis. These genes displayed a large difference in the expression level among time points, ranging from 1 to 6000 reads per kilobase transcript per million reads (RPKM). The expression level of the genes determined by qRT‐PCR correlated well with those obtained from RNA‐Seq (R ² = 0.95) (Data S1).

The principal component analysis (PCA) of the RNA‐Seq data showed that the global gene expression profiles were distinct among samples, except for F and R1 samples (Figure 2a). D1 and D2 samples were separated from F/R1 based on both principal component 1 (PC1), explaining 63% of the variance, and PC2, explaining 25% of the variance. The clustering of F and R1 indicates that, at least at the transcriptional level, plants recovered from the desiccation after 24 h of rehydration and resumed the majority of physiological processes to a level comparable with that in F, thereby determining a resurrection cycle. This agrees with the findings of Bernacchia et al. (1996), who described events that occur during rehydration.

Figure 2.

Global gene expression profiling of leaves during a dehydration and rehydration cycle. (a) Principal component analysis (PCA) of RNA‐Seq data obtained from leaves of an untreated (F), partially dried (D1), desiccated (D2) and 24‐h rehydrated (R1) plant. (b) Hierarchical clustering heat map of RNA‐Seq data. Seven clusters, C1–C7, were generated using a Pearson correlation coefficient of >0.47. The expression intensity is indicated as the scale bar. (c) Gene Ontology enrichment analysis of the seven clusters. Values in brackets indicate the number of genes present in each cluster. The identified clusters were further classified into four groups: 1, high gene expression at early dehydration stage; group 2, expression progressively increasing upon dehydration; group 3, expression peaking at the desiccated stage; group 4, expression progressively decreasing during the dehydration process then increasing to a level similar to untreated samples. The values are expressed as the rescaled RPKM value ± SD.

Hierarchical clustering was performed on the differentially expressed contigs to identify clusters of contigs with similar gene expression patterns. Seven distinct clusters, namely C1–C7, were revealed using a Pearson correlation threshold of >0.47 (Figure 2b). These seven clusters were assigned to four groups based on their expression profiles (Figure 2c). Group 1 was composed of C1 and C2, containing early dehydration‐responsive genes. The expression level of these genes increased during early dehydration and decreased to control levels at the desiccated stage. The second group was characterized by C3 with increased gene expression in response to mild and severe dehydration. C4 and C5 constituted the third group, characterized by contigs displaying a peak in expression at D2, namely desiccation‐responsive genes. The fourth group was represented by contigs in C6 and C7, which were abundant in both F and R1 samples.

GOE and mapman analysis

To obtain an overview of major biological processes associated with each expression pattern, the genes in each cluster were subjected to GOE analysis (Data S1; Figure 2c). The distribution of contigs in each group was visualized using mapman to summarize the transcriptomic changes with respect to the metabolic pathways (Figure S1). A close correspondence was observed between the results obtained from both analyses.

Contigs annotated as involved in secondary metabolism, lipid metabolism and the ascorbate–glutathione (GSH) cycle were equally represented, regardless of the state (Figure S1). Contigs involved in photosynthesis (i.e. light reactions and the Calvin cycle), mainly belong to group 4 (Figures 2c and S1), suggesting a shutdown of these metabolic processes upon dehydration to minimize the damage from photo‐oxidation. The changes in the expression level of photosynthesis‐related genes corroborate the changes in the photosynthetic CO₂ assimilation rates of leaves (see details below) and previous investigations on the photosynthesis of C. plantagineum (Dinakar et al., ²⁰¹²; Schwab et al., ¹⁹⁸⁹).

In contrast, cellular respiration, i.e. glycolysis, the tricarboxylic acid (TCA) cycle and the electron transport chain, was elevated in the early response to dehydration, which was reflected in the dominant presence of genes derived from group 1 (Figures 2c and S1). This phenomenon is most likely associated with a high energy demand for establishing protective mechanisms to prepare C. plantagineum for desiccation survival and recovery upon rehydration. Genes involved in amino acid synthesis mainly belonged to groups 3 and 4 (Figures 2c and S1). In particular, genes linked to the biosynthesis of branched‐chain amino acids (BCAAs; valine, leucine and isoleucine) were upregulated during desiccation (group 3).

In addition to metabolic pathways, GOE analysis revealed other crucial players (Figure 2c). During early dehydration, C. plantagineum may launch multifaceted defense/protective mechanisms, including sucrose biosynthesis, membrane remodeling, protein localization, cellular homeostasis and stomatal closure, as well as ABA‐activated signaling (group 1, Figure 2c). ABA signaling during early dehydration is crucial for the acquisition of desiccation tolerance (Bartels et al., ¹⁹⁹⁰; Furini et al., ¹⁹⁹⁷). Five out of seven genes involved in ‘protein dephosphorylation’ processes act as a negative regulator of ABA signaling, namely protein phosphatases of the PP2C family (i.e. HAB1, HAI2, HAI3 and PP2CA) (Kim et al., ²⁰¹³; Yoshida et al., ²⁰⁰⁶) and CPL3 (Koiwa et al., ²⁰⁰²) (Data S1). The upregulation of these phosphatase genes suggests that a tight regulation of ABA signaling occurs simultaneously with the activation of the ABA signaling pathway during early dehydration. These PP2Cs participate in regulating seed dormancy via double‐negative feedback loops controlling ABA biosynthesis, signaling and the action of PP2Cs in A. thaliana (Kim et al., ²⁰¹³). It is tempting to speculate that a similar mechanism could be present in C. plantagineum, especially considering the analogy between seed dormancy and desiccation tolerance in vegetative tissues (Costa et al., ²⁰¹⁷; Giarola et al., ²⁰¹⁷; Rodriguez et al., ²⁰¹⁰).

When approaching desiccation, GO terms linked with RNA processing, regulation, catabolism and translation, such as post‐transcriptional regulation of gene expression, mRNA splicing and RNA interference via siRNA (C4, Figure 2c) were enriched. Earlier studies have shown that post‐transcriptional modifications contribute to plant drought tolerance (Deak et al., ²⁰¹⁰). In C. plantagineum there is evidence for the role of short interfering RNA synthesized from an ABA‐inducible gene (i.e. CDT‐1) in the acquisition of desiccation tolerance (Hilbricht et al., ²⁰⁰⁸). Our findings support this notion and suggest that extensive transcriptional reprogramming may occur at desiccation, aimed at triggering metabolic pathways for desiccation tolerance and at accumulating transcripts and enzymes to rapidly restore cell activities upon rehydration. The latter perspective is supported by studies showing that the application of transcription and translation inhibitors during rehydration had no effect on cell recovery in dried resurrection plants (Cooper and Farrant, 2002; Dace et al., ¹⁹⁹⁸).

Genes associated with intracellular trafficking were also observed, namely the translocation of metabolites, ions and/or proteins into the vacuole (C3, Figure 2c), mitochondria (C4 and C5, Figure 2c) and chloroplast (C5, Figure 2c). Several transcripts encoding vacuolar‐type H⁺‐ATPases (V‐ATPases) were present in C3 (i.e. AVA‐P2, TUF and VHA‐A3), confirming their role in regulating osmotic pressure during desiccation by promoting the accumulation of solutes and ions in the vacuole, similarly to what is reported in other plant species (Baisakh et al., ²⁰¹²; Liu et al., ^2018b; Zhang et al., ²⁰¹³). Multiple genes encoding subunits of the TIM23 (translocase of the inner membrane) complex were observed in C5, namely TIM17‐2, TIM17‐3, TIM23‐1 and TIM23‐2 (Data S1). This complex is located in the inner membrane of mitochondria and mediates the import of precursor proteins into the mitochondrial matrix (Neupert, 1997). Concerning protein import into the chloroplast, we observed translocon genes TOC90 and TOC120 and genes involved in assisting protein translocation, such as CLPC1 (Zhang et al., ²⁰¹⁸), and transcription factor CIA2 (Sun et al., ²⁰⁰⁹). This finding agrees with studies in which several highly expressed nuclear‐encoded, chloroplast‐localized protective proteins were identified (Bartels et al., ¹⁹⁹²; Giarola et al., ²⁰¹⁵; Phillips et al., ²⁰⁰²; Schneider et al., ¹⁹⁹³). The rapid import of these proteins is very likely to be essential to protect chloroplasts during dehydration/rehydration. These results indicate that active import processes from the cytosol to subcellular compartments are required for cellular adjustment to water deficit. As this enhancement in mitochondrial and chloroplast transport only occurs at the desiccated stage, it is more likely to contribute to repair damaged organelles and to accumulate molecules for recovery upon rehydration, rather than an adaptation to early dehydration.

Studies demonstrated that, to survive long‐term exposure to adverse conditions, plants need to remove aberrant proteins to alleviate their toxicity (Lu et al., ²⁰¹⁸). Our analysis suggests that the ubiquitin–proteasome system could be the major pathway for protein degradation during dehydration. This is evidenced by the upregulation of genes encoding ubiquitin‐activating enzyme E1 (UBA1 and UBA2), ubiquitin‐conjugating enzyme E2 (UBC28 and UBC2) and components of the E3 ubiquitin ligase complex (i.e. SKP2A, EBF1, RDUF1, CUL1, CPR1, RMA2, CER9, RBX1, RHF2A and ZTL) at the desiccated stage (C5; Data S1; Figure S1c). The increased expression of E3 ubiquitin ligase during dehydration was also observed for another resurrection plant: H. rhodopensis (Liu et al., ^2018a).

Proteome profiles of leaves in response to dehydration and rehydration

To complement the transcriptome analysis, the soluble proteome of leaves was analyzed. The same samples were used as for transcriptomics plus an additional time point, namely 48 h of rehydration (R2). A later time point during rehydration could provide a more complete view concerning protein accumulation at full recovery.

By interrogating our C. plantagineum transcriptome, a total of 1430 proteins were identified after filtering with at least two peptides and one unique peptide per protein (Data S2). Of these proteins, 611 proteins (43%) showed statistically significant differences in abundance among time points (DAPs; P < 0.05 and fold change > 1.5) (Data S2). Prediction of protein subcellular localization revealed that about 54% of DAPs are cytoplasmic and plastid‐localized proteins, whereas 10% are mitochondrion‐ and nuclear‐localized proteins (Data S2). The functional classification of DAPs using the TAIR codes was performed using panther. The result showed that DAPs are mainly associated with cellular processes (36.5% of DAPs with TAIR code) and metabolic processes (30.8%), cellular component organization or biogenesis (7.5%), localization (6.3%), biological regulation (6.9%) and response to the stimulus (9.6%) (Data S2). In the metabolic process category, 84.1% of DAPs are involved in primary metabolic processes (Data S2). PCA of all DAPs resulted in a clear discrimination of F, D1, D2 and the two rehydration stages (R1 and R2), respectively (Figure 3a). The grouping of R1 and R2 is indicative of high similarity in their proteomic signature.

Figure 3.

Proteomic changes of leaves in response to dehydration and rehydration. (a) Principal component analysis (PCA) plot of proteomic data of leaves at different hydration stages. (b) Hierarchical clustering analysis of proteins with significantly differential abundance. Six clusters, P1–P6, were identified. (c) Protein abundance profiles of clusters. The intensity of protein abundance is indicated as the scale bar. Values in brackets indicate the number of proteins present in each cluster. (d) Dot plot of Gene Ontology (GO) enrichment analysis showing the 15 most over‐represented biological processes at a GO level of 4. The points in each dot plot are sized by the proportion of all proteins within the cluster annotated with the GO term and colored by enrichment confidence (P‐adjusted value). No GO terms were enriched in P3. For the complete results of GO enrichment analysis, see Data S2. Abbreviations: F, untreated leaves; D1, partially dried leaves; D2, desiccated leaves; R1, 24‐h rehydrated leaves; R2, 48‐h rehydrated leaves.

The hierarchical clustering of DAPs resulted in the identification of six clusters, namely P1–P6 (Figure 3b). All clusters were subjected to GOE analysis (Data S2) and the 15 most significantly over‐represented biological processes are shown in Figure 3(c). P1 is represented by the proteins with the highest abundance at D2. These proteins are mainly involved in metabolic compound salvage, monosaccharide and cellular aldehyde metabolic processes, lipid oxidation, ribonucleoprotein complex biogenesis and response to incorrectly folded proteins (Figure 3d). The response to incorrectly folded proteins coincides with our transcriptomic evidence depicting the induced transcript levels of genes involved in the ubiquitin–proteasome system at desiccation (Figure 2c).

Proteins in P2 displayed an unchanged or higher abundance upon dehydration, with the highest level observed at R1 (Figure 3b,c). The majority of these proteins participate in photosynthesis, responses to various stresses (i.e. oxidative and heat stress) and the metabolic processing of monosaccharide, cellular aldehyde and reactive oxygen species (Figure 3d). It is interesting to note that the significantly enriched GO terms of this cluster span a broader range of biological processes compared with those of other clusters (Figure 3d). This observation agrees with the transcriptomic data (Figure 2), supporting the notion that C. plantagineum reactivates many metabolic processes after 24 h of rewatering.

P4, P5 and P6 consist of proteins with the lowest abundances at D2, yet with differing abundances at other time points. More specifically: P4 shows a higher protein abundance at R2; proteins belonging to P5 have a sharply increased abundance, to a comparable level with F and D1, upon rewatering; and the abundance of proteins in P6 decreases during dehydration and remains unchanged upon rehydration (Figure 3b,c). All three clusters contain proteins that are associated with photosynthesis, cellular metabolic compound salvage and monosaccharide metabolic processes, indicating a tight temporal coordination in the abundance of proteins participating in the same process.

Variation in temporal dynamics of transcript and protein levels

Transcriptome and proteome data sets were compared to investigate whether changes in the abundance of transcripts and proteins correlate. We focused on a total of 421 transcript–protein (TP) pairs, which is the intersection obtained between the transcriptome and the proteome, filtered with analysis of variance (anova), P < 0.05 (Data S2). To examine the monotonic relationship between these paired data, Spearman’s correlation coefficients were calculated globally to relate transcript and protein abundances at the same time point. A poor correlation between transcript and protein level of the same time point was observed (ρ = 0.20–0.23; Data S2). Likewise, limited correlations (ρ = 0.14–0.29; Data S2) were found when the abundance of transcripts was compared with the corresponding protein abundance at later time points, thereby suggesting that the delay associated with the processing of transcripts is not the only factor determining the discrepancy between transcript level and protein abundance.

To examine the changes in the expression level of transcripts and their corresponding proteins throughout dehydration and rehydration, a k‐means clustering analysis was performed on these 421 TP pairs. The temporal behavior of these TP pairs was grouped across nine major clusters (i.e. TP1–TP9) with distinct dynamics (Figure 4a). Based on the protein profile, these clusters can be classified into three categories: TP1–TP4 showed the lowest protein abundance at the desiccated stage; the opposite trend with the highest abundance at the D2 stage was observed for TP5–TP6; and TP7–TP8 displayed the highest protein abundance after rehydration (Figure 4a,b). Various gene expression patterns were present within each category.

Figure 4.

Integrative analysis of paired transcriptome and proteome. (a) The k‐means clustering heat map of intersection data between transcriptome and proteome depicting nine clusters (TP1–TP9). The intersection containing 421 TP pairs was obtained using transcriptome and proteome data filtered with analysis of variance (anova)P < 0.05. The average gene expression and corresponding protein abundance obtained from three or four biological replicates were rescaled to range from −2 to 2 (by subtracting the mean value of all stages and dividing by the standard deviation), and are indicated by the intensity of the blue and orange shading, respectively. Three sections with roughly similar protein profiles are divided by dashed lines: lowest abundance at the desiccated stage (D2; TP1–TP4); highest abundance at the desiccated stage (D2; TP5–TP6); and highest abundance after 24 h of rehydration (R1; TP7–TP9). (b) The profile of the transcripts (in blue) and the corresponding profile of the proteins (in orange) of each cluster across different stages: untreated (F), partially dried (D1), desiccated (D2), 24‐h rehydrated (R1) and 48‐h rehydrated (R2). The values were expressed as the mean value of each stage ± standard deviation. The number of proteins/genes belonging to each cluster is indicated between brackets. The summarized result of GO enrichment analysis of each cluster is given. For a complete result of GO enrichment analysis, see Data S2.

A few clusters, such as TP1 and TP2, showed roughly analogous changes in transcript and protein abundance, whereas others (approx. 68% of TP pairs) displayed a low correlation. This result agrees with Spearman’s correlation analysis, indicating no direct link between the temporal dynamics of most transcripts and their encoded proteins. This discrepancy is likely to be attributed to post‐transcriptional/‐translational modifications. Indeed, the transcriptomic data highlight an induction of transcripts involved in post‐transcriptional regulation at the desiccated state (C4 and C5 in Figure 2c). It has been demonstrated that the abundance of transcripts and their corresponding proteins generally show a weak correlation, which is often weaker under stress and developmental changes (Vogel and Marcotte, 2012). In C. plantagineum, a lack of correlation between transcript and protein abundances was previously shown for transketolases (Bernacchia et al., ¹⁹⁹⁵), sucrose phosphate synthase (SPS; Ingram et al., ¹⁹⁹⁷) or sucrose synthase (Kleines et al., ¹⁹⁹⁹). Transcripts with a delayed translation were observed, as protein abundance increased at later stages with respect to the corresponding transcript level, such as the pairs in TP4, TP6, TP8 and TP9. In dry seeds, many mRNAs are highly accumulated, thereby preparing the seed for protein synthesis during early germination (Holdsworth et al., ²⁰⁰⁸). Similarly, some transcripts required for recovery may be accumulated and stored during the quiescent/desiccated stage, thus preparing plants for a fast recovery. Intriguingly, late embryogenesis abundant proteins (LEAs), LEA‐like proteins, heat‐shock proteins (HSPs) and low temperature‐induced 65‐kDa protein (‐like) (LTI65) displayed an earlier accumulation of the relative transcripts, as compared with the respective proteins (Data S2). The abundances of these hydrophilic proteins are generally high in hydrated control samples and further increase under stress as co‐ordinately regulated sets, comprising two main patterns. One is represented by transcript levels that are abundant at D1, whereas their corresponding proteins accumulated substantially at D2 and/or R1, and another is given by genes with high expression levels at D2 and increased protein abundance at R1 and/or R2. The protective role of these proteins in desiccation tolerance is well documented (Bartels, 2005; Gechev et al., ^{2012; 2013}; Liu et al., ^2018a; Rodriguez, et al., ²⁰¹⁰). The accumulation of transcripts to ensure the synthesis of sufficient protein, when conditions require it, may be important in the energy management of C. plantagineum upon dehydration and at the onset of rehydration. Such a strategy may allow plants to allocate more energy to the most urgent metabolic needs.

The GOE analysis of clusters revealed that photosynthetic light reactions, glucose metabolism and glycolysis are over‐represented in TP1 and chlorophyll biosynthetic processes are enriched in TP2 (Figure 4b), thus showing a correlation between transcript and protein levels across the dehydration and rehydration cycle. This indicates that transcription and translation of these genes are closely linked. It should be noted that some processes related to photosynthesis (i.e. electron transport chain and carbon fixation) as well as glycolysis are also present in TP7, which shows a substantial decrease of the related genes during dehydration that is not accompanied by a decrease of the corresponding proteins (Figure 4b).

These analyses indicate that, to a large extent, transcript levels by themselves are insufficient to predict protein levels in this highly dynamic system, and thus the simultaneous investigation of the proteome and metabolites is indispensable for understanding desiccation tolerance.

Alteration of metabolite abundance upon dehydration and rehydration

To correlate the transcriptome and proteome analyses with metabolic pathways, changes were investigated in the accumulation of primary metabolites in leaves exposed to dehydration and rehydration (i.e. F, D1, D2 and R1) using GC‐MS. A total of 39 metabolites showed significant changes in abundance between at least two different experimental conditions examined (P < 0.05; Data S3), including 15 sugars and sugar alcohols, 13 amino acids, 10 organic acids and one additional metabolite (i.e. putrescine). PCA showed that F and D1 were closely clustered but were clearly discriminated from D2 and R1 by PC1, explaining 57.7% of the variance, and PC2, explaining 18.2% of the variance. This result indicates that the drastic changes in the primary metabolite profile occurred at the late stage of dehydration (Figure S2a). PC1 mainly consists of contributors with positive loading (i.e. positive contributors), except for d‐mannitol, galactitol, 2‐oxoglutaric acid, pipecolic acid and meso‐erythritol, whereas the top contributors for PC2 are galactitol, glutamine, glucose/fructose‐6‐phosphate, as positive contributors, and sucrose, myo‐inositol, 4‐aminobutanoic acid, meso‐erythritol and malic acid, as negative contributors (Figure S2b). D2 and R1 both fell in the positive scale of PC1, indicating the accumulation of most of the metabolites examined in the desiccated and rehydrated leaves.

Clustering based on k‐means revealed six accumulation patterns across the experiment (Figure S2c–h). In general, the levels of most metabolites peaked at the desiccated (D2) and/or 24‐h rehydration (R1) stage, as reflected in three big clusters characterized by 36 metabolites (Figure S2f–h), indicating the involvement of these metabolites in coping with desiccation and in ensuring a fast recovery. Some of these metabolites may form a unique mixture, termed natural deep eutectic solvents (NADES), in which the enzymes, membranes and metabolites are protected and remain stable during desiccation, as suggested by Choi et al. (2011).

The changes of metabolites involved in the primary carbon and nitrogen metabolism were visualized within the context of the metabolic pathways (Figure 5). Most of the sugars decreased in abundance at D1, then accumulated during further desiccation and/or rehydration, such as mannitol, maltose, myo‐inositol, galactinol and trehalose. In contrast, levels of sucrose, fructose and glucose, the three most abundant sugars, were initially elevated during early dehydration, and later changed towards an opposite pattern, that is, sucrose was massively accumulated at D2 and then diminished at R1, whereas glucose and fructose were decreased in abundance at D2 and then substantially accumulated at R1. These results indicate that the higher level of sucrose at D1 is likely to be mainly derived from the mobilization of 2‐octulose, as reported previously (Bianchi et al., ¹⁹⁹¹), and sucrose synthesized from fructose and glucose (UDP‐glucose) via SPS may contribute to the sucrose accumulation during late phases of dehydration. This is in line with the sharp increase in SPS abundance at D2 (contig_4687; Data S2) and increased enzyme activity measured previously (Ingram et al., ¹⁹⁹⁷). The interconversion between sucrose and glucose/fructose at different stages strengthens the crucial role of sucrose in the glassy state in osmotic protection to survive desiccation and as a carbon source providing energy through glycolysis during dehydration and rehydration (Ingram and Bartels, 1996; Kuroki et al., ²⁰¹⁹; Scott, ²⁰⁰⁰).

Figure 5.

Simplified pathways showing the changes observed in the accumulation of metabolites involved in primary carbon and nitrogen metabolism upon dehydration and rehydration. Metabolites were classified as sugars (blue), amino acids (orange), organic acids (green) and others (hyacinth). The intensity of metabolite abundance is indicated in the scale bar. Colored boxes represent the normalized intensity of the corresponding metabolite across four stages: untreated (F), partially dried (D1), desiccated (D2) and rehydrated (R1). Dotted arrows indicate multiple enzyme‐catalyzed reaction steps. Metabolites with no significant changes among samples are shown in blue letters. The full list of identified metabolites and their normalized abundances is given in Data S3.

The aromatic amino acids (tryptophan and phenylalanine) and BCAAs (leucine, isoleucine and valine) were highly accumulated during early dehydration (Data S3; Figures 5 and S2). Tryptophan and phenylalanine increased further when desiccated and remained elevated during rehydration (D1 versus F, >9 and 25; D2 versus F, >20 and 87; R1 versus F, >25 and 92, respectively), pointing to a role in desiccation tolerance and an active shikimate pathway, which leads to the synthesis of antioxidants. Following the initial increase at D1, BCAAs decreased in abundance at D2 and later increased again at R1. BCAAs can be consumed as alternative carbon sources for sucrose synthesis at the late stage of dehydration, as suggested by studies in A. thaliana (Diebold et al., ²⁰⁰²; Schuster and Binder, ²⁰⁰⁵) and H. rhodopensis (Mladenov et al., ²⁰¹⁵).

The TCA intermediates examined showed a high abundance across all samples, especially malate and 2‐oxoglutarate (2‐OG; Data S3). These intermediates displayed an increased level throughout dehydration, with the exception of 2‐OG, which was reduced (Data S3; Figure 5). This indicates that cellular respiration is maintained during dehydration, consistent with our transcriptomic and proteomic data showing some respiratory genes with elevated abundances during dehydration (Data S1 and S2; Figure 2c).

Some photorespiratory cycle intermediates changed their levels across different stages, such as serine with a high accumulation at D1 and R1 and glycerate with a high abundance at D2 and R1, indicating changes in photorespiration during dehydration and rehydration. Sugar alcohols such as meso‐erythritol, mannitol and myo‐inositol showed a higher abundance either at D1 or at D2, supporting their role as osmoprotectants by stabilizing protein structures, as reported for S. lepidophylla (Yobi et al., ²⁰¹³).

To investigate the correlation between the transcriptome/proteome data and targeted metabolite profiling, we analyzed the changes in the abundance of metabolites and their related transcripts/proteins containing TAIR codes (e.g. synthetase and dehydrogenase) across time points (Data S3) using a Pearson correlation analysis. Among the 158 transcripts/proteins examined, the majority were not positively correlated with the corresponding metabolite levels. The discrepancy between transcript and protein levels was also reflected in this analysis, which resulted in the abundance of metabolites only correlated either with transcripts or proteins. For example, fructose and glucose contents were positively correlated with the protein abundance of sucrose synthase 3 (contig_1064), whereas the transcript level of hexokinase 1 and 2 (contig_17704 and 37764, respectively) correlated with glucose‐6‐phosphate levels. Positive correlations were observed for organic acids and their associated transcripts/proteins. Citrate levels were correlated with the protein level of citrate synthase 4 (contig_219). The transcript abundances of two succinate dehydrogenases (contig_41694 and 22096) were correlated with fumarate levels. The protein abundance of succinyl‐CoA ligase (SUCL; contig_1371) correlated with succinate content. The transcript levels of four glutamate (Glu) dehydrogenases (contig_167, 17656, 3398 and 5062) correlated with 2‐OG content. The transcript level of two isocitrate dehydrogenases (IDHs; contig_9261 and 8903) and the protein abundance of one IDH (contig_13569) correlated with 2‐OG level. For amino acids, a positive correlation was only observed for serine contents and the transcript level of phosphoserine phosphatase 1 (contig_30495). The low correlation in the level of amino acids and their associated transcripts/proteins indicates that, besides biosynthesis, other factors may account for amino acid levels during dehydration, such as ubiquitin‐mediated proteolysis and the autophagy process (see description below).

The analysis of the transcriptomic, proteomic and metabolite changes allowed us to evaluate the temporal dynamics of key metabolic pathways, including photosynthesis and photorespiration, cell respiration and amino acid biosynthesis upon dehydration and rehydration.

Molecular dynamics related to photosynthesis and photorespiration

As a homoiochlorophyllous plant, C. plantagineum preserves most of its chlorophyll and the integrity of thylakoid membranes during desiccation (Tuba et al., ¹⁹⁹⁸). This permits rapid recovery during rehydration but increases photooxidative stress, through the excessive production of reactive oxygen species (Bernacchia et al., ¹⁹⁹⁶). Therefore, the inhibition of photosynthesis during dehydration is an adaptive strategy employed by homoiochlorophyllous plants, as suggested by previous studies (Challabathula et al., ²⁰¹⁶; Dinakar et al., ²⁰¹²). To study the changes in the photosynthesis of C. plantagineum, we measured CO₂ assimilation rates, transpiration rates and stomatal conductance in leaves at different hydration stages (Figure S3). The results showed that the rates of photosynthetic carbon assimilation decreased progressively during dehydration, and a complete shutdown of photosynthesis was observed in desiccated leaves. The rehydration of plants for 48 h restored the photosynthetic carbon assimilation rates by 71% (Figure S3a). In correlation with the photosynthetic response, the transpiration rates and stomatal conductance were also significantly decreased during dehydration, and a complete recovery was observed 48 h after rehydration (Figure S3b,c). These phenomena are in line with the decline in transcript and protein abundance of most photosynthesis‐related genes during dehydration, including chlorophyll a/b binding proteins (CBPs), oxygen‐evolving enhancer proteins (PSBO, PSBP and PSBQ), subunits of the photosystem I and II (PSI and PSII) complexes, phosphoglycerate kinase (PGK), RuBisCO and RuBisCO activase (Data S3; Figure 6a).

Figure 6.

Schematic diagram showing the changes in transcript and protein abundances of key enzymes involved in (a) the photosynthetic electron transfer chain and (b) the Calvin cycle and photorespiration during a dehydration and rehydration cycle. Colored boxes represent the normalized intensity of transcript (in blue) and protein (in orange) levels across different stages: untreated (F), partially dried (D1), desiccated (D2), 24‐h rehydrated (R1) and 48‐h rehydrated (R2). Enzymes displaying an overall increased transcript level and/or protein level during the dehydration process are indicated in red. Missing values are left as blank, indicating that changes in transcript or protein levels are not significant across stages or that the protein was not detected. The changes in levels of metabolites (glycolate, serine and glycerate) at different stages are also shown with colored boxes. LHCB, light‐harvesting chlorophyll a/b (CAB)‐binding proteins; PSI/II, photosystem I/II; FNR, chloroplast ferredoxin‐NADP+ oxidoreductase; PGRL1, proton gradient regulation 5 (PGR5)‐like 1; NDH, NAD(P)H dehydrogenase; 3PGA, 3‐phosphoglycerate; 1,3‐PGA, 1,3‐bisphosphoglycerate; GAP, glyceraldehyde‐3‐phosphate; DHAP, dihydroxy‐acetone‐phosphate; F6P, fructose‐6‐phosphate; FBP, fructose‐1,6‐bisphosphate; E4P, erythrose‐4‐phosphate; X5P, xylulose‐5‐phosphate; SDP, sedoheptulose‐1,7‐bisphosphate; S7P, sedoheptulose‐7‐phosphate; R5P, ribose‐5‐phosphate; Ru5P, ribulose‐5‐phosphate; RuBP, ribulose‐1,5‐bisphosphate; RuBisCO, ribulose‐1,5‐bisphosphate carboxylase/oxygenase; RCA, RuBisCO activase; RBCL, RuBisCO large subunit‐binding protein subunit; RBCS, RuBisCO small chain; PGK, phosphoglycerate kinase; GAPDH, glyceraldehyde‐3‐phosphate dehydrogenase; FBPase, fructose‐1,6‐bisphosphatase; FBA, fructose‐bisphosphate aldolase; TKT, transketolase; SBPase, sedoheptulose‐1,7‐bisphosphatase; PRK, phosphoribulokinase; RPI, ribose‐5‐phosphate isomerase; PGLP, phosphoglycolate phosphatase; GOX, glycolate oxidase; GGAT, glutamate‐glyoxylate aminotransferase; AMT, aminomethyltransferase; GDC, glycine dehydrogenase; SHMT, serine hydroxymethyltransferase; HPR, hydroxyphenylpyruvate reductase; SGAT, serine‐glyoxylate aminotransferase. The normalized values for the abundances of transcripts and proteins are given in Data S3.

Despite the decreased transcript levels of genes encoding PSI reaction center subunit N (PSAN) during dehydration, the protein level increased throughout the whole treatment (D1, D2, R1 and R2 versus F, >3, 13, 32 and 12, respectively). PSAN has been identified to be the only subunit of PSI located entirely in the thylakoid lumen (Nielsen et al., ¹⁹⁹⁴) and to participate in docking plastocyanin to the PSI complex (Haldrup et al., ¹⁹⁹⁹). This protein has been suggested to fine‐tune PSI function as it has only been found in eukaryotes (Haldrup et al., ¹⁹⁹⁹). Our results indicate that PSAN may be important in regulating electron transfer flow during dehydration and rehydration.

The protein abundance of proton gradient regulation 5 (PGR5)‐like 1B (PGRL1B) increased 12‐fold in response to early dehydration, although its transcript largely decreased during drying. Chloroplast NADH dehydrogenase subunit 2 (NDH) maintained a high protein abundance at D1. The transcript of FAD/NAD(P)‐binding oxidoreductase (FNR) and ferredoxin 3 (FDX3) increased throughout the dehydration process (Data S3; Figure 6a). All of these proteins have been demonstrated to be involved in the cyclic electron flow (CEF) that generates ATP without the accumulation of NADPH in chloroplasts. It is thought to be essential for balancing the ratio of ATP/NADPH and for protecting both photosystems from damage caused by stromal over‐reduction (Nikkanen et al., ²⁰¹⁸). Therefore, the upregulation of these proteins at D1 indicates the presence of dehydration‐induced CEF around PSI in C. plantagineum. This is likely to protect PSII against excessive excitation pressure and maintain some photosynthetic ATP synthesis to fuel a partially functional Calvin cycle (see description below) and energy‐dependent processes at late stages of dehydration (Huang et al., ²⁰¹²). This result is further strengthened by the accumulation of various subunits of ATP synthase (e.g. ATPO, ATPC and ATPF) during dehydration, although the abundances of the corresponding transcripts were lower (Figure 6a). The maintenance/induction of CEF has also been suggested for other resurrection plants, such as B. hygrometrica (Tan et al., ²⁰¹⁷), Craterostigma pumilum, which belongs to the same genus as C. plantagineum (Zia et al., ²⁰¹⁶), H. rhodopensis (Mladenov et al., ²⁰¹⁵) and Paraboea rufescens (Huang et al., ²⁰¹²).

Four proteins involved in the Calvin cycle, glyceraldehyde‐3‐phosphate dehydrogenase (GAPDH), phosphoribulokinase (PRK), sedoheptulose‐1,7‐bisphosphatase (SBPase) and transketolase 3 (TKT3), showed an increased abundance during the dehydration process. Nonetheless, the corresponding transcripts were sharply decreased, except for TKT3 (Data S2; Figure 6b). Similar findings have been reported for a Malus domestica variety demonstrating high water‐use efficiency under moderate drought stress (Zhou et al., ²⁰¹⁵). The authors proposed that upregulating key enzymes involved in the Calvin cycle could counterbalance the decrease in intercellular CO₂ concentration, thus contributing to the partial maintenance of photosynthesis. Therefore, our data suggest that the activity of the Calvin cycle could be partially maintained during early dehydration (Figure S3).

Malate was the most abundant organic acid at all stages, with a peak at D2, although no significant changes were observed across different stages (Data S3). This was accompanied by an increase in several transcripts/proteins associated with malate synthesis and mobilization during drying. For instance, phosphoenolpyruvate carboxylase (PEPC, contig_1285) accumulated more than fivefold during the whole dehydration and rehydration process. Similarly, malate dehydrogenase (MDH, contig_7027) showed a twofold increase in protein abundance at D1 and remained elevated at later stages. The gene expression of six NADP‐dependent malic enzymes (NADP‐MEs) was on average six times higher in response to early dehydration and returned to the initial level at late dehydration (Data S3). In crassulacean acid metabolism (CAM)‐type photosynthesis, PEPC serves as the key enzyme in CO₂ fixation, leading to the synthesis of malate, the pivotal metabolite in CAM metabolism (Igamberdiev and Eprintsev, 2016). CAM has been demonstrated to be induced in C3/CAM‐intermediate plants upon water limitation (Cushman and Borland, 2002). The transition from C3 photosynthesis to CAM has been shown to coincide with the induction of enzymes involved in malate synthesis and mobilization, as well as enzymes associated with glycolysis/gluconeogenesis (Häusler et al., ²⁰⁰⁰; Höfner et al., ¹⁹⁸⁷; Winter et al., ¹⁹⁸²). The upregulation of genes involved in glycolysis is observed here (C1 in Figure 2c). It is, therefore, likely that there is a shift from C3 to CAM‐type photosynthesis in C. plantagineum during dehydration. The upregulation of PEPC has also been observed in another resurrection plant, H. rhodopensis, and this C3‐to‐CAM shift hypothesis has been suggested (Gechev et al., ²⁰¹³). Nonetheless, we cannot rule out the possibility that the induction of PEPC and malate metabolism‐related proteins is involved in the anaplerotic replenishment of TCA cycle intermediates consumed during N‐assimilation and biosynthesis (O'Leary and Plaxton, ²⁰¹⁶). It would be interesting to make further investigations focusing on malate metabolism across diurnal time courses in C. plantagineum and H. rhodopensis, thereby shedding light on whether CAM is involved in desiccation tolerance.

Serine showed accumulation at D1, whereas glycolate and glycerate displayed a lower abundance relative to serine, but with a progressive increase throughout the dehydration and rehydration cycle (Data S3; Figure 6b). All these metabolites are linked to photorespiration; their increased abundance at the early phase of dehydration suggests a possible increase in photorespiratory activity at this stage. This is supported by a concomitant increase in the level of proteins participating in the photorespiration pathway, such as glycolate oxidase (GOX), glutamate‐glyoxylate aminotransferase (GGAT) and serine‐glyoxylate aminotransferase (SGAT) (Data S3; Figure 6b). Other photorespiration‐related proteins at the D1 stage generally remained at the same level as F before their level dropped at D2, including phosphoglycolate phosphatase 1B (PGLP), aminomethyltransferase (AMT), glycine dehydrogenase (GDC), hydroxyphenylpyruvate reductase (HPR) and serine hydroxymethyltransferase (SHMT) (Data S3; Figure 6b). The photorespiratory pathway serves as an electron sink to protect the photosynthetic apparatus from electron‐induced photo‐damage in drought‐stressed C3 plants (Guan et al., ²⁰⁰⁴; Wingler et al., ¹⁹⁹⁹). Future work is needed to dissect which role photorespiration plays in the acquisition of desiccation tolerance.

Molecular dynamics related to cellular respiration

Transcriptomic and proteomic data suggest an overall increase in the accumulation of transcripts and/or proteins associated with the glycolytic process during dehydration, such as β‐fructofuranosidase/cell wall invertase (CWINV), hexokinase (HXK), sucrose synthase (SUS), UDP‐glucose pyrophosphorylase (UDPGP), phosphoglyceromutase (PGAM), ATP‐dependent 6‐phosphofructokinase (PFK), fructose‐bisphosphate aldolase (FBPA), GAPDH, phosphoglucomutase (PGM), PGK, pyrophosphate fructose 6‐phosphate 1‐phosphotransferase subunit alpha (PFP), pyruvate kinase (PK) and PEPC. Among them, FBPA and GADPH displayed a peak at the early phase of rehydration (Data S3; Figure 7a). These results indicate that there is an increased energy demand during stress and the early phase of recovery. It is very likely that C. plantagineum employs UDPGP and PFP, pyrophosphate (PPi)‐dependent enzymes, to partly circumvent ATP‐dependent reactions, thereby optimizing the energetic performance of glycolysis (O'Leary et al., ²⁰¹⁹; O'Leary and Plaxton, ²⁰¹⁶).

Figure 7.

Schematic diagram showing changes in transcript and protein levels of key enzymabundances involved in (a) glycolysis, (b) the TCA cycle and (c) the electron transport chain in Craterostigma plantagineum during the dehydration and rehydration cycle. Colored boxes represent the normalized intensity of transcript levels (in blue), protein levels (in orange), metabolites levels (sugars in blue and organic acids in green) across different stages: untreated (F), partially dried (D1), desiccated (D2) and 24‐h rehydrated (R1) and 48‐h rehydrated (R2). Enzymes displaying an overall increased transcript level and/or protein level during the dehydration process are indicated in red. Missing values are left blank, indicating that changes in transcript levels or protein levels were not significant across stages or that the protein was not detected. Fru(glu)‐6‐P, fructose(glucose)‐6‐phosphate; Fru‐1,6‐P, fructose‐1, 6‐biphosphate; GAP, glyceraldehyde‐3‐phosphate; 1,3‐DPGA, 1,3‐diphosphoglycerate; 2(3)‐PGA, 2(3)‐phosphoglycerate; PEP, phosphoenolpyruvate; SUS, sucrose synthase; CWINV, cell wall invertase; UDPGP, UDP‐glucose pyrophosphorylase; PGM, phosphoglucomutase; HXK, hexokinase, PFP, pyrophosphate fructose 6‐phosphate 1‐phosphotransferase; FBA, fructose bisphosphate aldolase; PFK, ATP‐dependent 6‐phosphofructokinase; GAPDH, glyceraldehyde‐3‐phosphate dehydrogenase; PGK, phosphoglycerate kinase; PGAM, 2,3‐bisphosphoglycerate‐independent phosphoglycerate mutase; ENO, enolase; PK, pyruvate kinase; PEPC, phosphoenolpyruvate carboxylase; CS, citrate synthase; PDC, pyruvate dehydrogenase complex subunits; AH, aconitate hydratase; IDH, isocitrate dehydrogenase; FH, fumarate hydratase; MDH, mitochondrial malate dehydrogenase; OGDC, 2‐oxoglutarate dehydrogenase complex; SDHC, succinate dehydrogenase complex subunits; SUCL, succinyl‐CoA ligase; SSADH, succinate‐semialdehyde dehydrogenase; GABA, 4‐aminobutyric acid; NDB1/2, external alternative NAD(P)H‐ubiquinone oxidoreductase; NDA1, internal alternative NAD(P)H‐ubiquinone oxidoreductase A1; AOX, ubiquinol oxidase; UCP, uncoupling protein. The normalized values for the abundances of transcripts and proteins are given in Data S3.

The activity of the TCA cycle is likely to be maintained or induced during early dehydration. An accumulation of TCA cycle intermediates was observed during dehydration and early rehydration (Data S3; Figure 7b), which is in agreement with the induced transcript and/or protein level of pyruvate dehydrogenase complex subunits (PDCs), aconitate hydratase (AH), citrate synthase (CS), IDH, mitochondrial MDH, 2‐OG dehydrogenase complex components (OGDCs), SUCL and succinate dehydrogenase complex subunits (SDHCs) upon dehydration. Unlike most of the quantified TCA cycle intermediates, the abundance of 2‐OG decreased during dehydration (Data S3; Figure 7b). Given the unique role of 2‐OG as the major carbon skeleton for nitrogen‐assimilatory reactions, this decreased accumulation could suggest an increase in carbon flux into amino acids in response to dehydration. This is supported by the increased level of Glu during dehydration with concomitant accumulation of glutamate synthase (GOGAT).

For the mitochondrial electron transport chain, most associated transcripts and proteins increased or remained the same during dehydration. For example, ATP synthase subunit β displayed a two‐ to sevenfold increase upon water loss, compared with untreated samples. Some subunits of the cytochrome bc ₁ complex, cytochrome c oxidase and NADH dehydrogenase electron transport chain‐related proteins exhibited higher gene expression, either at early dehydration or at the desiccated stage. The upregulation of genes involved in the alternative respiratory pathway was observed in response to dehydration, namely alternative oxidase/ubiquinol oxidase (AOX), internal and external alternative dehydrogenases (NDA1 and NDB1) and the uncoupling protein (UCP) (Data S3; Figure 7c). These enzymes do not pump protons and consequently cannot contribute to energy conservation via ATP synthesis, but this non‐energy‐conserving bypass allows respiration NAD(P)H oxidation to be uncoupled from ATP synthesis, thereby regulating redox balance independently of ATP synthesis (Millar et al., ²⁰¹¹). Previous studies have observed the activation of this bypass during dehydration and other abiotic stresses, presumably to counteract ROS production, thus preventing oxidative damage (Dahal and Vanlerberghe, 2017; Van Aken et al., ²⁰⁰⁹; Vanlerberghe, ²⁰¹³). Our results imply that the alternative respiratory pathway may play a role in desiccation tolerance of C. plantagineum through regulating the redox status to avoid oxidative damage.

Molecular dynamics related to nitrogen assimilation and amino acid metabolism

Chloroplast degradation mediated by autophagy during dehydration has been described in C. pumilum, a homoichlorophyllous plant belonging to the same genus as C. plantagineum (Charuvi et al., ²⁰¹⁹). Furthermore, in this study, the observed amino acid profile up to the dehydration stage (Data S3) corresponds to that observed in the early stage of sugar starvation in A. thaliana, where protein degradation via autophagy was suggested as an adaptive response (Hirota et al., ²⁰¹⁸). Mining the transcriptome for genes involved in autophagy reveals that most contigs associated with autophagy peak in desiccated samples. The recently described constitutively stressed 1 protein (COST1), identified as a negative regulator of drought responses in A. thaliana (Bao et al., ²⁰²⁰), and annotated here as myosin‐4 protein (DUF641), is also more expressed in desiccated samples, indicating that different pathways may be involved in autophagy while preparing the plants for resurrection once favorable conditions occur. Ubiquitin ligases and components of the SKP/cullin/F‐box complex, a marker for autophagy (Rao et al., ²⁰¹⁸), likewise attain their highest expression in desiccated plants. A noteworthy exception to this is contigs annotated as NBR1, a peroxisomal receptor for ubiquitinated proteins (Kirkin et al., ²⁰⁰⁹) that peak during rehydration. These data indicate that the activation of autophagy, denoted as critical in drought responses (Liu et al., ²⁰⁰⁹), is involved in the recovery of primary metabolites during the desiccation of C. plantagineum.

Like other resurrection plants, C. plantagineum accumulates amino acids upon dehydration (Figure 5). They serve as protective osmolytes (e.g. proline), as well as nitrogen and carbon reservoirs for resuming metabolism upon rehydration (Martinelli et al., ²⁰⁰⁷). Among the amino acids examined, Glu showed overall the highest abundance across all stages, which is consistent with its role as a primary nitrogen donor for the synthesis of nitrogen‐containing compounds (Lea and Ireland, 1999). Glu accumulated during dehydration, with the concomitant increase in the abundance of ferredoxin‐dependent glutamate synthase (FdGOGAT), glutamine synthetase (GS), and nitrate reductase (NIR) transcripts, indicating a possible induction of nitrogen assimilation in response to dehydration. Further accumulation of Glu upon rehydration was observed in this study, whereas a decreased level has been reported for H. rhodopensis (Moyankova et al., ²⁰¹⁴). This discrepancy could be related to the timing of rewatering. Glutamine (Gln) decreased during the whole dehydration process and increased during rehydration. This decrease in glutamine and the increase in Glu, interconverted by GSs, results in a declining Gln/Glu ratio. The higher expression of other enzymes using Gln as nitrogen donor, such as GMP synthetase (contig_1432 and 1431), might contribute to this decreasing ratio.

In contrast to the observations in H. rhodopensis (Moyankova et al., ²⁰¹⁴), asparagine (Asn) increased throughout dehydration and remained stable during rehydration. This is consistent with the expression of glutamine hydrolyzing asparagine synthetase (DIN6, contig_20580), which increased during early dehydration and returned to control values in desiccated and rehydrated samples. The conversion of Gln to Asn is thought to be beneficial through the higher N‐to‐C ratio of Asn and its lower metabolic activity in carbon‐limited conditions (Lam et al., ¹⁹⁹⁴). Although in the current experiment there is no carbon shortage, Asn remains the nitrogen storage form with the highest N‐to‐C ratio.

As pointed out before, an increased carbon flux towards amino acid synthesis during dehydration is also evidenced by the simultaneous decrease in 2‐OG, which provides carbon skeletons for Glu synthesis (Data S3; Figure 8). Maintaining the carbon flux towards amino acid synthesis during the early phase of dehydration has also been reported for Sporobolus stapfianus (Whittaker et al., ²⁰⁰⁷). In contrast, inactivation of the nitrogen assimilatory pathway has been suggested for H. rhodopensis based on reduced levels of amino acids, including Glu (Gechev et al., ²⁰¹³). The response of nitrogen assimilation upon dehydration may vary among species and may depend on environmental factors.

Figure 8.

A summary detailing the main events that occur in leaves upon a dehydration and rehydration cycle. Hypotheses for the changes in primary metabolism, conferring desiccation tolerance, are in blue.

The level of 4‐aminobutyric acid (GABA) increased upon dehydration, which is a common response to abiotic stress (Fait et al., ²⁰⁰⁸). GABA substantially accumulated during dehydration (D2 versus F, >3), although this was not statistically significant. Genes involved in the GABA shunt were all drastically upregulated at the early phase of dehydration, including Glu decarboxylase (GAD, D1 versus F, >44), GABA transaminase (GABA‐T, D1 versus F, >17) and succinate‐semialdehyde dehydrogenase (SSADH, D1 versus F, >8) (Data S3). The induction of the GABA shunt may be used to replenish succinate for the TCA cycle and mitochondrial electron transport chain (Fait et al., ²⁰⁰⁸), thus maintaining energy production upon water deficit.

The levels of BCAAs showed a significant increase in response to early dehydration (D1) and rehydration (R1) (Data S3). However, accumulation of most of the transcripts associated with the biosynthesis of BCAAs occurred at the desiccated stage (D2), such as transcripts encoding acetolactate synthase (ALS) and 2‐isopropylmalate synthase B (IPMS2), except for branched‐chain‐amino‐acid aminotransferase 2 (BCAT2), which showed an elevated transcript level at D1 (Data S3; Figure S1). This discrepancy suggests that the elevated accumulation of BCAAs during early dehydration could result from protein degradation or that the biosynthetic genes of BCAAs are transcribed before the onset of desiccation, leading to a rapid increase in BCAAs during rehydration through de novo biosynthesis. Protein degradation is the primary source for the accumulation of BCAAs in osmotically stressed Arabidopsis (Huang and Jander, 2017). This is also supported by our transcriptomic and proteomic results, showing an induced ubiquitin–proteasome system for protein degradation, and the breakdown of damaged proteins (Figures 4d and S1c).

The accumulation of 5‐oxo‐proline diminished slightly at D1, but remained high overall, peaking during rehydration. In plants, 5‐oxo‐proline is synthesized by γ‐glutamylcyclotransferase as part of GSH catabolism. GSH is degraded via two pathways, with the pathway over γ‐glutamyl cyclotransferase (GGCT) and 5‐oxo‐prolinase (5OPase) dominating (Ohkama‐Ohtsu et al., ²⁰⁰⁸). The expression of GGCT peaked in desiccated plants, whereas that of 5OPase peaked during rehydration, suggesting that the accumulation of 5‐oxo‐proline may contribute to the osmotic adjustment of the cytosol. The increased expression of genes involved in the degradation of GSH coincides with a decreased expression of GSH synthase. No data on the GSH content was acquired in this study, but a decrease in GSH was observed in desiccated Myrothamnus flabellifolia (Kranner et al., ²⁰⁰²).

Conclusions

Plant desiccation tolerance is an intricate process orchestrated by many metabolic pathways and complex regulatory circuits. To better understand this phenomenon, we conducted a comprehensive analysis of temporal profiles of the transcriptome, soluble proteome and primary metabolites of C. plantagineum leaves across a dehydration and rehydration cycle. A summary detailing the main events in leaves and hypotheses for the regulation of desiccation tolerance focusing on primary metabolism is given in Figure 8. The correlation analysis of transcript and protein levels across the dehydration and rehydration cycle shows in general a low correlation, indicating sophisticated regulation at the post‐transcriptional level. This discrepancy should be taken into account in the future to interpret transcriptomic data. Furthermore, our integrative analysis of transcriptomes, proteomes and metabolites not only confirms some previous studies but also reveals the remarkable intrinsic flexibility of the primary metabolism, such as ATP‐independent bypasses, possible C3‐to‐CAM photosynthesis switch, enhanced CEF, alternative respiratory pathway and GABA shunt, to maximize the physiological performance of plants for their survival of desiccation. This knowledge provides a higher resolution concerning the changes in the primary metabolism during the dehydration and rehydration process in C. plantagineum. The data provided here are a step towards a systems biology approach to understand desiccation tolerance and will be a rich source for further in‐depth investigations of the physiological and genetic responses of C. plantagineum to dehydration.

EXPERIMENTAL PROCEDURES

Plant materials, growth conditions and sampling

Craterostigma plantagineum plants were grown as described by Bartels et al. (1990). Briefly, plants were multiplied vegetatively and grown under 60% relative humidity and a cycle of 16 h of light (60 000 lx) at 24°C and 8 h of dark at 20°C. For dehydration, untreated plants (F) were gradually dried in pots until reaching partial dehydration (D1, 3–4 days) or complete desiccation (D2, 15 days). Desiccated plants were rehydrated by submerging the plants in water for 24 h, then leaves were collected for 24‐h rehydrated samples (R1). For the 48‐h rehydrated samples (R2), plants were removed from the water after 24 h and kept under normal conditions before collection. The relative water content (RWC) of leaves was determined as described by Bernacchia et al. (1996). At least three independent biological replicates were used for each analysis (i.e. three for both transcriptomic and targeted metabolic analysis, and four for proteomic studies). For each biological replicate, leaves from between six and nine plants from each treatment were pooled, ground in liquid nitrogen and stored at −80°C. To enrich the sequence pool for the assembly of a reference transcriptome, additional leaf, root and flower samples were collected.

Measurements of CO₂ gas exchange rates

The photosynthetic CO₂ assimilation rates, transpiration rates and stomatal conductance were measured in leaves of C. plantagineum plants using a portable gas‐exchange analyzer (GFS‐3000; Walz, https://www.walz.com) at a photosynthetic photon flux density of 1000 μmol m⁻² sec⁻¹ and a CO₂ concentration of 350 μmol mol⁻¹, relative humidity of 60% and temperature of 25 ± 1°C. Leaves of four individual plants from each of the following hydration stages were analyzed: F, D1, D2, R1 and R2, with an additional time point, namely the late dehydrated stage (LD, 25–30% RWC). Stable photosynthetic gas exchange measurements were performed in leaves that were uncurled physically without damage, because it is difficult to perform measurements at low RWCs because of the curling of the leaves.

Library preparation and transcriptome assembly

Total RNA extraction was carried out using the RNeasy Plant Mini Kit (Qiagen, https://www.qiagen.com) according to the manufacturer’s instructions. RNA concentration and quality were determined using a Nanodrop ND‐1000 spectrophotometer (ThermoFisher Scientific, https://www.thermofisher.com) and a 2100 Bioanalyzer (Agilent, https://www.agilent.com), respectively. The RNA integrity number values of all extracted RNAs were above 8. The SMARTER stranded RNA‐Seq Kit was used to prepare libraries according to the manufacturer’s instructions (TaKaRa, https://www.takarabio.com). Twelve libraries collected from the time‐course experiment (three biological replicates for each time point) were pooled and sequenced to generate 75‐bp paired‐end reads (approx. 20 million per library) using five consecutive runs of MiSeq 150 cycles reagent kit (Illumina, https://www.illumina.com). Additional RNAs from root, leaf and flower samples were sequenced using a MiSeq 600 cycles reagent kit to generate 300 bp paired‐end reads (~50 M per library).

Polysomal mRNAs of F, D1, D2 and R2 were isolated according to the method described by Juszczak and Bartels (2017). About 150 mg of freeze‐dried sample was extracted with 1 ml of extraction buffer. After centrifugation for 5 min at 13 000 g at 4°C, the supernatant supplemented with 10% (w/v) sodium deoxycholate (1/20 volume) was loaded on continuous sucrose gradients (15–56%, w/v). Ultracentrifugation was performed for 80 min at 45 000 g at 4°C. Ten fractions (410 μl each) were collected and RNAs were precipitated using 2.5 m lithium chloride. Control gradients supplemented with 20 mm puromycin were used to identify fractions containing polysomes. The polysome‐containing fractions were pooled. RNAs were purified using RNA binding columns from Gene Matrix Universal RNA Purification Kit (EURx, https://eurx.com.pl). Polysomal mRNAs isolated from at least three independent biological replicates were pooled and used for library construction. RNA sequencing (RNA‐Seq) was performed by GATC Biotech (now Eurofins, https://eurofinsgenomics.eu) using the Illumina Hiseq platform and 50‐bp single‐end reads (approx. 30–40 million per library) were generated. As a control, RNA‐Seq analysis was carried out for libraries generated from total RNAs of the same samples.

FASTQ files were uploaded in clc genomics workbench 8.5, discarding reads with poor quality (<Q30). Reads with nucleotide ambiguity were filtered from the remaining sequences. A TrueSeq Illumina adaptor trimming was performed. Sequences with sizes below 35 bp were removed. A preliminary assembly was constructed from the 300‐bp paired‐end reads to generate guidance for the main assembly. Different word sizes (19–64 bp) and bubble sizes (50–400 bp) were evaluated using the percentage of remapped reads, the broken pairs rate, the N50 and the total number of contigs as criteria for the selection of the optimal assembly. The final guidance assembly was obtained using a word size equal to 19 and a bubble size equal to 150. A similar strategy was used to generate the main assembly by integrating the reads from time‐course samples of total and polysomal RNAs. The best combination of word size (45 bp) and bubble size (300 bp) was applied. An annotation was generated against the NCBI Viridiplantae and Arabidopsis thaliana databases using blast2go pro.

RNA‐Seq analysis and qRT‐PCR validation

The libraries were mapped to the reference with a maximum hit number of 10, a minimum threshold of 80% length coverage and 80% identity, a mismatch cost of 2 and a deletion/insertion cost of 3. Expression values were calculated using the RPKM method (Mortazavi et al., ²⁰⁰⁸). Contigs/genes with less than 10 specifically mapped reads, and/or with a mean RPKM value of the biological replicates lower than 0.1 in at least one of the libraries, were also removed from the data set. Contigs displaying a significant and differential expression were determined using an anova test with four groups (F, D1, D2 and R1). An FDR correction was applied with a cut‐off at 0.001 (Benjamini–Hochberg correction). Contigs with a significant difference in expression (FDR P < 0.001 and fold change > |2|) for at least one time point were retained in the analysis.

The hierarchical clustering of RNA‐Seq data was performed using cluster 3.0 software with Pearson correlation distance and complete linkage (Eisen et al., ¹⁹⁹⁸). Output files were visualized using java treeview 1.1.6r4 (https://sourceforge.net/projects/jtreeview/files/jtreeview/).

The GOE analysis was performed using the gluego 2.5.3 plugin implemented in cytoscape 3.7.1 with the following parameters: Benjamini–Hochberg enrichment, P < 0.05; kappa score set at 0.4; and GO level set from level 3 to level 8 (Bindea et al., ²⁰⁰⁹). Different gene expression profiles containing differentially expressed genes were mapped into metabolic pathways using mapman 3.6.0RC1 (Thimm et al., ²⁰⁰⁴). A mapping file was generated using mercator 3.6 (https://www.plabipd.de/portal/mercator‐sequence‐annotation). The PCA of RNA‐Seq data was performed using IDEP (Ge et al., ²⁰¹⁸).

Reverse transcription of RNA was performed using the ProtoScript II Reverse Transcriptase (NEB, https://international.neb.com). Fourteen genes were used to validate RNA‐Seq data via qRT‐PCR analysis. Primers were designed using primer3plus (Untergasser et al., ²⁰⁰⁷) and primary efficiency was calculated using fivefold serial dilutions of cDNA (from 12.5 to 0.0008 ng µl⁻¹). Reference genes reported previously were used (Giarola et al., ²⁰¹⁵) and data normalization was carried out using two genes (EF1α and GAPDH) that were recognized as the best reference genes by geNormTM. Relative gene expression levels were analyzed with qbase⁺ (Biogazelle, https://services.biogazelle.com). Primer sequences, primer efficiency and the correlation analysis between gene expression levels are provided in Data S1.

Analysis of soluble proteomes

The soluble proteome of F, D1, D2, R1 and R2 with four biological replicates of each were analyzed using LC‐MS/MS. Approximately 500 mg of sample was extracted with ice‐cold extraction buffer (10% TCA and 0.07% DTT in acetone) and allowed to precipitate at −20°C for 1 h. The pellet obtained after centrifugation (10 000 g, 5 min at 4°C) was washed twice with ice‐cold acetone. Dried pellets were resuspended in a solution of equal volumes of Tris‐buffered phenol (pH 7.5) and buffer with 2% SDS, 30% sucrose (0.1 m Tris‐HCl; pH 8) with vigorous vortexing. After centrifugation, 300 µl of the upper phenol phase was collected and proteins were precipitated by adding 1.5 ml of 0.1 m ammonium acetate in cold methanol for 1 h at −20°C. After centrifugation, the supernatant was discarded and the pellet was washed twice with 1.5 ml of cold precipitation solution, followed by two washes with 80% cold acetone. The pellet was air‐dried and dissolved in 150 µl of labeling buffer: 2 m thiourea, 7 m urea, 4% 3‐[(3‐cholamidopropyl)dimethylammonio]‐1‐propanesulfonate detergent and 30 mm Tris. A Bradford assay (Bio‐Rad, https://www.bio‐rad.com) was used for protein quantification.

Twenty micrograms of protein was used for one‐dimensional SDS‐PAGE using a pre‐cast Criterion™ XT gel (4–12% Bis‐Tris gels; Bio‐Rad). After migration, gels were stained with Instant Blue™ (BVBA; Gentaur, https://gentaur.com). Each gel lane was divided into two groups of bands, corresponding to low and high molecular weight proteins, and each band was cut into cubes for in‐gel digestion. After reduction with 10 mm DTT in 100 mm ammonium bicarbonate (NH₄HCO₃) and alkylation with 55 mm iodoacetamide in 100 mm NH₄HCO₃, the proteins were digested with trypsin (sequencing mass grade; Promega, https://www.promega.com).

The extracted peptides were analyzed with a NanoLC 425 Eksigent System coupled to a TripleTOF^® 6600 MS (Sciex, https://sciex.com) in two technical replicates. Peptides were loaded and desalted onto the trap column (C18 PepMap™, 5 µm, 5 mm × 300 µm; ThermoFisher Scientific) for 10 min at a flow rate of 2 µl min⁻¹ with 2% acetonitrile (ACN) and 0.05% trifluoroacetic acid (TFA) in MQ (Milli‐Q®; Merck, https://www.merckmillipore.com). Peptides were separated onto a C18 reverse‐phase column (PepMap™ 100, 3 µm, 100 Å, 75 µm × 15 cm; ThermoFisher Scientific) using a long binary gradient (solvent A, H₂O LC‐MS 0.1% formic acid and solvent B, ACN, 0.1% FA) at a flow rate of 300 nl min⁻¹. The gradient was a linear increase from 3 to 30% solvent B over 90 min, an increase to 40% solvent B over 10 min, followed by a 5‐min wash step to 80% solvent B. Prior to a new injection, the column was re‐equilibrated for 25 min at 3% B.

Time‐of‐flight mass spectrometry (TOF‐MS) scans from 300 to 1250 m/z were acquired in positive mode. From each TOF‐MS scan, the 30 most intense precursors were selected for product ion scans (100–1500 m/z) using the automatically adjusted rolling collision energy voltage in high sensitivity mode.

The data were imported into progenesis qi 4.1 (Nonlinear Dynamics, https://www.nonlinear.com). Peptide identification was carried out by searching the Craterostigma in‐house database containing 288,270 sequences via the mascot daemon 2.6.0 interface (Matrix Science, https://www.matrixscience.com) using the following parameters: fragment mass tolerance of 0.5 Da, peptide tolerance of 20 ppm, carbamido‐methylation of cysteine as fixed modification and oxidation of methionine, N‐terminal protein acetylation and pyroGlu from N‐terminal Glu as variable modifications. Only proteins identified with a significance mascot‐calculated confidence of 95% were kept for analysis.

Data from each group were treated independently and the results from these analyses combined in progenesis qip. Proteins with a fold change of >1.5, P < 0.05 and a minimum of two significant sequences per protein and one unique sequence per protein were considered as differentially abundant proteins (Data S2).

The hierarchical clustering of proteomic data was performed using the same approach that was used for RNA‐Seq data. Proteins showing TAIR identifiers of each cluster were subjected to GOE analysis using r libraries. The PCA of proteomic data was carried out using past 3 (Hammer et al., ²⁰⁰¹). The prediction of protein localization was carried out using the web server DeepLoc with BLOSUM62 settings (http://www.cbs.dtu.dk/services/DeepLoc/) (Almagro Armenteros et al., ²⁰¹⁷). The functional classification of proteins using the TAIR codes was performed using panther 14 (http://pantherdb.org) (Mi et al., ²⁰¹⁹).

Integrative analysis of transcriptomic and proteomic data

The intersection between transcriptomic and proteomic data was obtained by filtering both data sets with anova P < 0.05 (Data S2). The k‐means clustering of TP pairs was carried out using r libraries. Data of each contig were centered and rescaled to the standard deviation. The k‐means clustering was performed after determining the optimal number of clusters using the gap statistic method with the r library ComplexHeatmap. The uncentered correlation was used as a distance measure. The GOE analysis of all clusters was performed using the gluego 2.5.3 plugin implemented in cytoscape 3.7.1, with the same settings used for RNA‐Seq data. The relationships between transcript and protein levels were analyzed using Spearman’s rank correlation (Zar, 2005).

Targeted metabolite analysis

Three biological replicates of samples collected at different sampling stages were used. Methanol (400 µl) containing 2 µg ml⁻¹ adonitol (internal standard) was added to 12.5 mg of ground lyophilized sample and gently agitated for 15 min at room temperature (RT, 23–25°C). Subsequently, 200 µl of chloroform was added, followed by a 10‐min agitation. Finally, 400 µl of water was added to the mixture and centrifuged at 12 000 g for 5 min after vortexing. The upper phase (100 µl) was collected and vacuum‐dried. For derivatization, the dried residue was dissolved in 50 µl methoxyamine in pyridine (20 mg ml⁻¹), vortexed and incubated at 30°C, with agitation, for 90 min. Then 50 µl of N‐methyl‐N‐(trimethylsilyl) trifluoroacetamide was added. After 30 min at 37°C with agitation, the samples were transferred to a glass vial and left at RT for at least 4 h before injection.

Each sample was analyzed twice with a 7890B gas chromatograph (Agilent, https://www.agilent.com) with a multi‐purpose sampler (Gerstel GmbH & Co, http://www.gerstel.com), coupled to a 5977A quadrupole mass spectrometer (Agilent). The chromatograph was equipped with a Restek Rxi 5MS, 30 m × 0.25 mm × 0.25 µm column (Restek GmbH, https://www.restekgmbh.de). The derivatized sample (1 µl) was injected in splitless mode. The temperatures of the interface, injector and the ion source were maintained at 300, 250 and 230°C, respectively. The helium flow rate was 1.2 ml min⁻¹. After 4 min at 70°C, the temperature was increased to 310°C at 5°C min⁻¹, maintained for 5 min at 310°C, then cooled down to 70°C and kept at this temperature for 1 min. The filament delay was set at 6 min and the recording of mass spectra was carried out from 50 to 600 m/z at 5.5 scans per second.

Raw data were pre‐treated on XCMS ( https://xcmsonline.scripps.edu) (Tautenhahn et al., ²⁰¹²) to generate a list of differentially present ions among the experimental modalities. The identification of the metabolites was performed manually in the masshunter B.06.00 qualitative browser (Agilent), based on the retention time and fragmentation pattern of standards, as well as on the fragmentation pattern of the National Institute of Standards and Technology (NIST) library (https://www.nist.gov). Metabolites were quantified by normalizing the peak area to the dry weight of the sample and the internal standard. When a compound was identified as different trimethylsilyl derivatives or identified in different peaks, the coherence was checked and the data of the different peaks accumulated for statistical analysis. Significantly differentially present metabolites were determined by anova with P < 0.05 and a fold change of >|1.5| between at least two sampling stages. The same k‐means clustering approach described above was applied to metabolic data across different stages (Figure S2).

Author contributions

XX performed the qRT‐PCR analysis and bioinformatics analyses, interpreted all data and wrote the article. SL and GG performed library construction, transcriptome assembly and RNA‐Seq analysis. KS and SC conducted the targeted metabolic analysis. KS and CCL performed soluble proteome analysis. SZ contributed to the bioinformatics analyses. VG and XL prepared plant materials and isolated polysomal mRNAs for sequencing. DC made the photosynthesis measurements. GG, JR, J‐FH and DB conceived this study and designed the experiments. All authors contributed to data interpretation and manuscript revision.

Conflict of Interest

The authors declare that they have no conflicts of interest associated with this work.

Supporting information

Figure S1. mapman visualization of the transcriptomic changes during a dehydration and rehydration cycle. Significantly differentially expressed contigs were assigned to four groups based on their expression profiles across experimental conditions, namely untreated (F), partially dried (D1), desiccated (D2) and 24‐h rehydrated (R1) stage. Four different colors represent, respectively, the four groups with distinct expression profiles, namely higher gene expression at the partially dried stage (group 1), at both dehydrated and desiccated stages (group 2), the desiccated stage (group 3) and both untreated and 24‐h rehydrated stages (group 4). CHO: carbohydrate; OPP: oxidative pentose phosphate pathway; TCA: tricarboxylic acid cycle; mito.: mitochondria; ‐misc: miscellaneous.

Figure S2. Metabolic profiling of leaves in response to dehydration and rehydration. (a) Principal components analysis (PCA) of metabolite abundances obtained from leaves of untreated (F), partially dried (D1), desiccated (D2) and 24‐h rehydrated plant (R1). (b) Bar plot represents the PCA loadings of the metabolites for the main principal components (PC1 and PC2). (c–h) The k means clustering heatmap of metabolic data. The color scale reflects the changes in the metabolite abundance from low (red) to high (blue) levels.

Figure S3. The rates of photosynthetic CO₂ assimilation (a), transpiration (b) and stomatal conductance (c) in leaves of Craterostigma plantagineum upon dehydration and rehydration. Asterisk (*) indicates a statistically significant difference to the control (P ≤ 0.001). Error bars represent means ± SD (n = 4). F, untreated leaves; D1, partially dried leaves; LD, late dehydrated leaves; D2, desiccated leaves; R1, 24‐h rehydrated leaves and R2, 48‐h rehydrated leaves.

Data S1. Transcriptomic analysis of leaves during dehydration and rehydration.

Data S2. Proteomic analysis of leaves during dehydration and rehydration.

Data S3. Targeted metabolic analysis of leaves during dehydration and rehydration.

DATA AVAILABILITY STATEMENT

Raw sequences of transcriptome have been deposited at the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus website (GEO, http://www.ncbi.nlm.nih.gov/geo, accession number: GSE157098). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (Vizcaíno et al., ²⁰¹⁶) partner repository with the data set identifier PXD020923 and 10.6019/PXD020923.