A Novel Role of Medicago truncatula KNAT3/4/5-like Class 2 KNOX Transcription Factors in Drought Stress Tolerance

Abstract

Class 2 KNOX homeobox transcription factors (KNOX2) play a role in promoting cell differentiation in several plant developmental processes. In Arabidopsis, they antagonize the meristematic KNOX1 function during leaf development through the modulation of phytohormones. In Medicago truncatula, three KNOX2 genes belonging to the KNAT3/4/5-like subclass (Mt KNAT3/4/5-like or MtKNOX3-like) redundantly works upstream of a cytokinin-signaling module to control the symbiotic root nodule formation. Their possible role in the response to abiotic stress is as-of-yet unknown. We produced transgenic M. truncatula lines, in which the expression of four MtKNOX3-like genes was knocked down by RNA interference. When tested for response to water withdrawal in the soil, RNAi lines displayed a lower tolerance to drought conditions compared to the control lines, measured as increased leaf water loss, accelerated leaf wilting time, and faster chlorophyll loss. Reanalysis of a transcriptomic M. truncatula drought stress experiment via cluster analysis and gene co-expression networks pointed to a possible role of MtKNOX3-like transcription factors in repressing a proline dehydrogenase gene (MtPDH), specifically at 4 days after water withdrawal. Proline measurement and gene expression analysis of transgenic RNAi plants compared to the controls confirmed the role of KNOX3-like genes in inhibiting proline degradation through the regulation of the MtPDH gene.

1. Introduction

Climate change typically results in prevalent drought stress conditions over vast areas at a global scale, threatening agricultural production world-wide. Drought stress affects the plant growth rate and development by affecting several physiological and biochemical functions, ultimately leading to reduced productivity [1,2]. Drought stress alters plant growth and development and severely affects photosynthesis, chlorophyll content, nutrient metabolism, ion uptake, and carbohydrates metabolism [1,3,4]. Plants have evolved several morphological, physiological, biochemical, and molecular mechanisms to overcome drought stress effects [1]. Since cellular water loss marks the first event of drought stress, the first plant response aims at increasing water uptake and maintaining cellular water content. This is achieved by enhancing root growth and accumulating organic and inorganic solutes in the cytosol, such as proline and sugars, to lower cellular osmotic potential, thereby maintaining cell turgor and improving water uptake from the drying soil [5,6]. In particular, proline acts as an excellent osmolyte, but also as a metal chelator, an antioxidant, and a signalling molecule [7,8]. Mechanisms enhancing stomatal closure, the reduction in shoot growth, and the acceleration of leaf senescence are also triggered to limit water loss [2,9]. The regulation of plant stress responses is inherently coordinated with other essential biological processes, especially growth-related pathways [2]. A balance between plant growth and survival to stress must be achieved [9], and hormones play a key role in this by integrating developmental signals with those related to stress response [10]. Drought response also involves genome-wide transcriptional reprogramming that leads to additional protective mechanisms, such as osmotic adjustment, detoxification, the mitigation of stress-induced damage, and the amplification or attenuation of stress signalling [11]. Stress-specific transcription pathways are connected to upstream signalling and developmental pathways via transcription factors (TFs). Several families of the TFs have been identified as important players in orchestrating the response to drought [11,12,13,14,15,16], but only few in the delicate balance between growth and survival. Amongst the hormones involved in the drought stress response, cytokinins (CKs) are good candidates in the trade-off between plant growth and stress resilience [17]. CKs mediate drought resilience by modifying root architecture and improving root fitness, influencing the photosynthetic machinery, enhancing antioxidant mechanisms, modulating the water balance, and directly affecting the expression of drought-responsive genes [18].

KNOTTED1-LIKE homeobox (KNOX) TFs control several aspects of plant development by modulating hormonal homeostasis and integrating the signalling of cytokinins, gibberellin, brassinosteroids, and auxin [19,20,21,22]. Due to their involvement in controlling multiple hormonal pathways during plant development, they may constitute novel good candidates in the balance between growth and stress response. KNOXs belong to the Three Amino acid Loop Extension (TALE) protein family of the TFs (KNOX and BLH) which are found in all eukaryote lineages [23] and are classified into two subfamilies, class 1 (KNOX1) and class 2 (KNOX2). KNOX1 and KNOX2 were proposed to have antagonistic functions in development, with KNOX1 promoting meristematic cell fate and KNOX2 promoting cell differentiation [24]. In Arabidopsis, there are four KNOX2 genes, KNAT3, KNAT4, KNAT5 that are closely related to each other, and KNAT7 [21]. So far, there are no studies that directly associate members of KNOX2 TFs with stress response, although there are some indirect indications that this could be the case [25,26]. KNAT3, together with its TALE interactor BLH1, was shown to mediate ABA responses during seed germination and post-germinative growth [25]. Additionally, KNAT3 and KNAT4 were amongst the 2000 drought-responsive genes identified in Arabidopsis thaliana under progressive soil drought stress using whole-genome oligonucleotide microarrays [26].

We decided to investigate the possible role of KNOX2 TFs in the response to drought stress in the model legume Medicago truncatula. In the M. truncatula genome, there are five KNOX2 genes, of which four more are related to the Arabidopsis KNAT3, KNAT4, KNAT5 (Mt KNAT3/4/5-like genes, also known as MtKNOX3-like genes), and one more is related to KNAT7 [27,28]. M. truncatula is a plant that occurs naturally in the arid and semi-arid environments of the Mediterranean Basin area, and which has also been developed into an annual legume forage crop in Australia. It can be considered as a drought-adapted species and an ideal experimental model to advance our understanding of the underlying molecular mechanisms of drought adaptation and tolerance in legumes [29]. Legumes is the second family of crops after cereals in terms of cultivated areas worldwide, represents a valuable source of proteins for food and feed, and is vital to sustainable agriculture due to its ability to fix atmospheric nitrogen by establishing symbiotic interactions with rhizobia [29]. As observed for other crops, legume production is threatened by the extreme environmental conditions, with water scarcity being the main constrain [30]. Legume species vary in tolerance/sensitivity to drought, but in all cases, the final yield is significantly decreased [30].

Characterization of drought stress response in M. truncatula have been addressed, although functional studies to identify the important regulators are still limited. Global reprogramming of transcription and metabolism during progressive drought and after rewatering have been assessed in M. truncatula [29]. In this study, a detailed overview of the transcriptome and metabolome changes associated with the progression of drought stress, linked to the physiological measurements of the plant water status, was achieved in shoots and roots [29]. These analyses confirmed a role of abscisic acid (ABA) as one of the early signal molecules that modulate physiological responses to drought [31]. Among the metabolites detected in drought-stressed plants, myo-inositol and proline had striking regulatory profiles indicating involvement in M. truncatula’s drought tolerance [29]. The genes encoding Δ1-Pyrroline-5-carboxylate synthase (P5CS), a key enzyme in proline synthesis, and those coding for proline degrading enzymes, such as proline dehydrogenase (PDH), were up- and down-regulated, respectively. A comparison of molecular and physiological responses to drought stress in the two widely used M. truncatula genotypes, Jemalong A17 and R108, highlighted important differences between the two genotypes, with Jamalong A17 showing a greater tolerance to drought stress [32].

To explore the possible role of KNOX2 genes in the M. truncatula response to drought, we used a previously obtained RNA interference (RNAi) construct [28] to produce transgenic lines in which four MtKNOX3-like genes (MtKNOX3, MtKNOX5, MtKNOX9, and MtKNOX10) were simultaneously knocked down. The MtKNOX3-like genes were previously shown to redundantly work upstream of a CK-signalling module to control the symbiotic root nodule formation [28]. They are homologous to the Arabidopsis KNAT3/4/5 TFs and their expression is found in both root and leaf tissues [27,28]. When tested for response to water withdrawal in the soil, RNAi lines displayed a lower tolerance to drought conditions compared to controls. Reanalysis of the transcriptomic data from a drought stress time course experiment [29] using cluster analysis and Gene Co-expression Network (GCN) pointed for the first time to a possible role of KNOX3-like TFs in repressing a proline dehydrogenase gene at early stages of stress. Proline measurement and gene expression analysis of transgenic RNAi plants compared to the controls confirmed the role of KNOX3-like genes in increasing proline content by inhibiting its degradation through the repression of a PDH gene.

2. Results

2.1. Generation and Selection of MtKNOX3-like RNAi Transgenic Lines

To investigate a possible role of M. truncatula KNOX3-like TFs in drought response, two previously obtained Mt KNAT3/4/5 RNAi constructs (Mt KNAT3/4/5-like RNAi and Mt KNAT3/4/5-like hox RNAi) [28] were used to transform leaves and petals of M. truncatula R108 [33]. After selection on kanamycin, nine T0 plantlets for the Mt KNAT3/4/5-like RNAi construct (Figure 1a) and five from construct Mt KNAT3/4/5-like hox RNAi were analyzed for the presence of the kanamycin resistance gene (nptII). All the generated T0 plants resulted positively and were grown to obtain T1 seeds. However, no seeds were obtained from T0 plants transformed with Mt KNAT3/4/5-like hox RNAi, as leaves from these plantlets accumulated anthocyanin, lost chlorophyll, and died at an early developmental stage before producing seeds (Figure 1b). Plants transgenic for the Mt KNAT3/4/5-like RNAi construct and those transformed with the empty vector reached maturity and set seeds. The T1 plants were genotyped for kanamycin resistance and those that resulted positive were grown and analyzed for the gene expression of MtKNOX3, the most abundant KNAT3/4/5-like transcript [28], using quantitative real-time PCR (Figure 1c).

Amongst the T1 lines harboring the resistance gene, some plants displayed a reduction of more than 60% in the expression of MtKNOX3. Further gene expression analysis showed that the Mt KNAT3/4/5-like RNAi construct was effective in down-regulating the expression of the four tagged Mt KNAT3/4/5-like genes (an example is shown in Figure 1d). T1 plants (p.23, p78 and p.88) from three independent T0 transformants, showing the lowest expression of MtKNOX3 (CI38, CI73, and CI76, respectively), were selected for further analysis on the T2 seeds.

2.2. Physiological Characterization of MtKNAT3/4/5-like RNAi Transgenic Lines

To have a preliminary evaluation on whether the knock down of the MtKNOX3-like genes in M. truncatula affects drought response, leaves from the T1 progeny of the CI38, CI73, and CI76 T0 lines were also tested for water loss. Mt KNAT3/4/5-like RNAi leaves lost water more quickly than the empty vector controls, and the differences were already detectable after 16 h, with some lines showing 50% water loss compared to the 30% of control leaves (Figure 2a). After 40 h, leaves from all transgenic RNAi plants had lost more than 60% of water, some even reaching 80% of water loss, compared to the control plants that displayed no more than 50% water loss. These data pointed to a decreased drought tolerance of the MtKNAT3/4/5-like RNAi lines. To physiologically characterize the response of RNAi plants to oxidative stress, we took advantage of paraquat/methyl viologen. We observed a lower reduction in electrolyte leakage (60%) from CTR leaves exposed for 24 h to methyl viologen (Figure 2b) compared to the leaves of the RNAi lines (85%), thus indicating that the RNAi lines were less tolerant to the herbicide.

We then analyzed total chlorophyll content in detached leaves during a dark-induced leaf senescence assay at different time points (0 d, 6 d, and 10 d) (Figure 2c, left panel). Both empty vector plants (CTR) and MtKNAT3/4/5-like RNAi (RNAi) leaves showed a significant reduction in the content of total chlorophyll over time, but transgenic RNAi leaves lost pigment faster than the control plants. RNAi plants had lower total chlorophyll content already at time 0. The percentage of leaves undergoing senescence was significantly higher in the RNAi leaves compared to the controls at early time points (Figure 2c, right panel). The survival strategy behind the dark-induced senescence confirmed the reduction in the RNAi plants’ tolerance to stress, since its regulation paths are closely related [34,35]. Altogether, the responses of RNAi transgenic lines to different abiotic stress treatments indicated that the knock down of the MtKNOX3-like genes contributed to the increase in oxidative stress sensitivity.

2.3. Silencing of Mt KNAT3/4/5-like Transcription Factors Decreases Tolerance to Drought Stress

To validate these initial physiological and biochemical observations, we tested the behavior of T2 RNAi plants compared to the control plants for drought stress in soil at the emergence of the seventh leaf. The progeny of the T1 plants containing the vector only (from now on “control” plants) or the Mt KNAT3/4/5-like RNAi construct (from now on “RNAi” plants) were sown in soil and genotyped for the nptII kanamycin resistance gene. Plants harboring the constructs were grown in soil for five weeks under the regular watering regime. Progressive drought was induced by water withdrawing for 18 days in order to mimic what plants experience in their natural environment. During the experiment, soil moisture and soil electric conductivity were monitored with a sensor (as schematized in Figure 3a). Mild and severe stress timing was defined based on Zhang et al. 2014 [29]. At day 12, corresponding to mild stress, the leaves of RNAi plants started to wilt, whereas those of the control maintained a good turgor (Figure 3b). At day 18, corresponding to severe stress, RNAi T2 lines displayed rolled and dried up leaves, whereas the control plants were still green and only moderately wilted (Figure 3c). All these findings pointed to an increased sensitivity to drought stress of the RNAi lines, thus to a role of the MtKNOX3-like genes in drought stress tolerance.

2.4. Meta-Analysis of a M. truncatula Drought Stress Transcriptomic Experiment Identifies MtKNOX3-like Genes as Rapidly Induced during Drought Stress

In order to identify the regulatory networks that involve MtKNOX3-like genes, we explored publicly available transcriptomic data. We took advantage of a high-quality drought stress kinetics carried out and described in Zhang et al. 2014 [29], where transcriptomic and metabolomic response was analyzed at 2, 3, 4, 7, and 14 days after water withdrawal in M. truncatula shoots and roots. The experiment used the Affymetrix Medicago Gene Chip from which we extract functional modules of interest (all TFs modules [36]; CK, ABA, and gibberellin (GA) hormonal modules; and oxidative stress and proline metabolism modules). In fact, for the analysis, we applied a targeted knowledge-based bioinformatic pipeline that we previously described and successfully used to partition the data and focus on specific functional genetic modules [14,16,37]. Overall, 2468 probes, as listed in Supplementary Table S1, were considered and used for the K-means clusterization of gene expression to explore the main regulatory transcriptional networks acting during induction of drought stress in the shoot. The clustering gene expression data allows to identify substructures in the data and identify groups of genes that may share regulatory relationships. Co-expression and anticorrelation are commonly used in bioinformatics pipelines to identify potential TF targets [38,39,40,41], as target genes are often co-expressed with the TFs that positively regulate them, or are anticorrelated with the TFs acting as negative regulators.

The MtKNOX3-like genes were represented by the following Affy probes: Mtr.37567.1.S1_at (MtKNOX3); Mtr.6737.1.S1_s_at (MtKNOX5); Mtr.8750.1.S1_at (MtKNOX10); and Mtr.8842.1.S1_at (MtKNOX9). We identified ten main gene expression clusters that are shown in Figure S1 (K1–K10). Three out of the four MtKNOX3-like genes (MtKNOX3, MtKNOX5, and MtKNOX10) were placed in a specific cluster of 180 genes that are upregulated at 4 d of drought stress only (Cluster 3, Supplementary Figure S1). This cluster was highly anticorrelated (r = −0.84) with a cluster of 175 genes (Cluster 6, Supplementary Figure S1) that were specifically downregulated at 4 d of drought stress. In Clusters 3 and 6, 58 and 61 genes, respectively, had a cluster score ≥0.8 that identifies those genes whose expression closely match the core and were considered for further analysis (Supplementary Table S2). Most of the genes placed in two pairs of the anticorrelated clusters that contained the genes either gradually increased (Cluster 1 and Cluster 9) or gradually decreased (Cluster 7 and Cluster 10) their expression during drought stress (Supplementary Figure S1). MtKNOX9 was placed in Cluster 9, which contains the genes that are initially induced after 4 days of water withdrawal and steadily increase during stress.

We focused on Clusters 3 and 6 to explore the Gene Co-expression Networks (GCN) underlying the early expression changes occurring at day 4 in drought-stressed shoots that may involve MtKNOX3-like genes (Figure 4a). To do this, we used the transcript expression values from the 27-microarray data-related shoot samples to perform a robust pairwise correlation analysis and to identify those genes that resulted highly co-expressed or highly anticorrelated with the MtKNOX3-like genes (the blue and orange lines are shown in Figure 4b, respectively). Since KNOX TFs work as heterodimers with BLH TFs, we were also interested to explore KNOX and BLH genetic interactions during drought response. Hence, we selected all the transcripts representing both KNOX and BELL/BLH (superfamily of TALE TFs) for the construction and analysis of the GCN [42]. A total of 145 genes (Supplementary Table S3) were included to perform pairwise correlation analysis of gene expression values (Supplementary Table S3). The genes with significant correlations (p value ≤ 0.05) and absolute Pearson’s correlation (|r|) values ≥ 0.6 were used to develop the GCN. The final network included 140 genes and had a high clustering coefficient (0.619) and a characteristic path length of 1.977 (Supplementary Table S4).

To determine the relationships among the selected genes and to identify the hub genes in the early response to drought stress at 4 d in the shoot (Figure 4a), the topological features of the network were analyzed (Supplementary Table S4). We first looked at the genes displaying the highest node degree, which is the number of neighbors to which a node directly connects and characterizes the hub genes in a network [43]. The top five probes in the list belonged to Cluster 6 and represented two genes: four probes (Mtr.12290.1.S1_at, Mtr.12290.1.S1_s_at, Mtr.12291.1.S1_s_at, and Mtr.42984.1.S1_at), with degrees of 77, 78, 78, and 79 respectively, all represented one proline dehydrogenase 1 (namely MtPDH_1,2,3,4 in the GCN) gene (Mt v4 ID: Medtr7g020820; Mt v5 ID: MtrunA17_Chr7g0221391), encoding a key enzyme for proline degradation, of which its suppression induces proline accumulation [29]; one probe (Mtr.41252.1.S1_at) represented a gene predicted to encode for a BLH TF (Mt v4 ID: Medtr1g023050; Mt v5 ID: MtrunA17_Chr1g0155681) (degree = 76) that we named MtBLH6_2 in the GCN. MtKNOX3 and MtKNOX5 also showed a high node degree (51 and 59, respectively). They resulted highly co-expressed with four BELL/BLH genes (namely MtBLH1_BLH5, MtBLH1, MtBLH6_2, and MtBELL1_2 in the GCN), and anticorrelated with the PDH gene. MtKNOX3 and MtKNOX5 shared most of the connections (44). The subnetwork formed by the MtKNOX3 and MtKNOX5 genes stayed at the core of the GCN and is shown in green in Figure 4a. Probes, gene IDs, and the relative Pearson’s correlation of the genes shared by MtKNOX3 and MtKNOX5 are shown in Table S5. The connections shared by all the three MtKNOX3-like genes in Cluster 3, MtKNOX3, MtKNOX5, and MtKNOX10, were 26 and included the negative correlation with the MtPDH gene. KNOX2 were reported to act as transcriptional repressors [24], therefore possible MtKNOX3-like target genes would likely place in the anticorrelated Cluster 6. In this regard, the MtPDH gene, whose probes are highly anticorrelated with MtKNOX3, MtKNOX5, and MtKNOX10, may constitute a valuable downstream target that would link the MtKNOX3-like genes to drought response. The expression of the most abundant MtKNOX3-like and BLH genes identified in Cluster 3 were also analyzed in the transcriptomic data of Zhang et al. [29] in the roots sampled at the same time points of the shoot analysis in comparison to the MtPDH gene (Figure 5). The KNOX/BLH genes had a similar expression in the roots, with a transient increase at day 4 of drought stress. This increase corresponded to a decrease in the expression of the MtPDH gene, as observed in the shoot. Therefore, the opposite expression pathways of the MtKNOX3-like and MtPDH genes were conserved in both the shoot and roots of the plants under drought stress.

2.5. Gene Expression Analysis to Validate Clustering and GCN Analysis Predictions

The MtPDH gene acts as a main hub in the GCN constructed with Clusters 3 and 6 that represents the genes whose expression either increases or decreases specifically after 4 days of drought stress. To assess whether MtKNOX3-like TFs may directly or indirectly regulate the expression of MtPHD, we analyzed the expression of the MtPDH transcript in the leaves of the control and RNAi T2 plants sampled at early stages in our water withdrawing experiment. At time 0, RNAi lines already displayed a 1.4-fold significant increase in MtPDH expression with respect to the control lines. At 4 d of drought stress, MtPDH transcript abundance decreased in the control plants, but was instead still higher in RNAi plants (an example of a representative RNAi line is shown in Figure 6). This difference was only observed at 4 days because at 8 days of drought, RNAi plants also displayed a reduction in MtPDH expression with respect to the watered plants, similarly to the control plants. This confirmed previous transcriptomic data from Zhang et al. [29], as well as confirmed the predicted role of MtKNOX3-like genes in contributing to rapidly repress MtPDH gene expression at very early stages of drought stress response.

2.6. Mt KNAT3/4/5 like-RNAi Plants Accumulate Less Proline during Drought Stress

At the transcriptional level, the knock down of MtKNOX3-like genes seems to delay downregulation of the proline dehydrogenase gene MtPDH at early stages of drought stress. To assess whether RNAi lines were affected in proline accumulation upon water withdrawal, proline content was measured and compared in the leaves of either the control or Mt KNAT3/4/5-like RNAi lines at 4, 8, and 12 days of drought stress (Figure 7).

Since the beginning of the drought stress, a slight increase in proline was observed after 4 and 8 days in both CTR and RNAi plants. From 8 d to 12 d of drought stress, a significant proline increase was observed. The proline levels in CTR plants showed a 7-fold increase, whereas the increase in RNAi plants was only of 3.5-fold. The overall reduction in proline accumulation observed in the RNAi plants may relate to a delay in downregulating the MtPDH gene at early stages of drought stress due to the MtKNOX3-like genes’ knock down and may account for the reduced tolerance to drought stress.

2.7. KNOX/BLH Binding Site Motifs in the MtPDH Regulatory Sequences

We searched the MtPDH gene, 3000 bp upstream nucleotide sequences, and intronic regions for possible KNOX/BLH binding sites. Since no binding site specificity information is available for KNOX2 TFs of the KNAT3/4/5 clade, we used the binding sites’ degenerated sequences that were previously described for KNOX1 and BLH proteins [44]. We searched for three regulatory motifs that we named KNOX/BLH_bs1 (KGACM), KNOX/BLH_bs2 (TGAYTGA), and KNOX/BLH_bs3 (TGATKKGA) (Supplementary Table S8). The results are shown in Figure 8 and Supplementary Figure S2. In total, 23 KNOX/BLH binding motifs were found in the MtPDH gene, of which 14 in the promoter and 9 in the intronic regions. In particular, the third intron (187 bp length) of the MtPDH gene was strikingly enriched in KNOX/BLH binding sites (six binding sites) and may constitute a main regulatory element that could be tested in further analysis to validate MtKNOX3-like TFs and the MtPDH regulatory relationship.

3. Discussion

We investigated the possible role of MtKNOX3-like TFs that are members of the TALE superfamily of homeobox proteins in drought stress response. MtKNOX3-like genes belong to the KNAT3/4/5-like clade of KNOX2, which are highly co-expressed during plant development in both Arabidopsis and M. truncatula, thus pointing to functional redundancy. Arabidopsis single mutants for KNAT3/4/5 genes do not display any phenotype [24]. In our previous work on the role of KNAT3/4/5-like genes in M. truncatula nodule symbiosis, we observed no phenotype when only MtKNOX3 was knocked down [28]. Moreover, we observed that downregulating MtKNOX3 would result in an overexpression of MtKNOX5 and an overexpression of MtKNOX3 associated with a reduced expression of MtKNOX9 and MtKNOX10, thus suggesting strong transcriptional feedback regulation amongst the KNAT3/4/5-like genes [28]. Therefore, we used RNAi lines in which four MtKNOX3-like genes have been knocked down. One of the RNAi constructs we used did not allow the recovery of viable plants, whereas a second construct allowed us to obtain plants producing seeds that were less tolerant to drought stress. Biochemical and physiological parameters, analyzed under well-watered or drought conditions, pointed to a reduced tolerance of RNAi plants to stress compared to the control plants.

A role for MtKNAT3/4/5-like genes in root symbiotic nodule formation was recently shown [28]. Being recognized as developmental regulators, their possible role in the response to abiotic stress was neglected and poorly investigated so far. One study in A. thaliana linked KNAT3 and BLH1 TFs to ABA response. BLH1 was induced by ABA and the encoded protein physically interacted with KNAT3 to modulate ABA sensitivity during seed germination and early seedling development [25]. Arabidopsis KNAT3 and BLH1 were also found amongst the genes upregulated during drought stress response in a microarray experiment [26]. Whereas no further study reported possible roles of the AtKNAT3/4/5 subfamily in stress response, several BLH TFs were instead related to different type of stresses in cotton, rice, and soybean [46].

We re-analyzed the transcriptomic data of drought stress in M. truncatula with a novel targeted knowledge-based approach using clustering and GCN analysis to predict possible targets of MtKNOX3-like TFs. The clustering gene expression data allows to identify the groups of genes that may act in the same pathway/response and may share gene function.

Our targeted cluster analysis identified 10 clusters characterized by different expression patterns. It has been observed that the genes whose expression patterns are strongly anticorrelated can also be functionally related [47]. This is the case for transcriptional repressors and their target genes that may cluster into separate anticorrelated groups. Since KNOX2 TFs are predicted to be transcriptional repressors, we focused on the cluster where most of the aMtKNOX3-like genes are placed (Cluster 3), and on the anticorrelated one (Cluster 6). The genes in Clusters 3 and 6 were characterized by a transient increase or decrease in expression, respectively, at 4 d of drought stress. Three out of the four MtKNOX3-like genes targeted by our RNAi constructs, i.e., MtKNOX3, MtKNOX5, and MtKNOX10, were placed in Cluster 3 and resulted highly anticorrelated to a PDH gene (MtPDH) placed in Cluster 6. This negative regulatory relationship between the three MtKNOX3-like genes and MtPDH was confirmed in our transgenic RNAi lines, which displayed a constitutive higher expression of MtPDH in leaves and a delay in MtPDH downregulation at early stages of water deprivation with respect to control plants.

It was clearly shown that M. truncatula tightly regulates proline production and accumulation during drought stress progression by the upregulation of several genes encoding P5CS, a key enzyme in proline synthesis, and concomitantly repressing genes coding for proline degrading enzymes such as PDH [29]. P5CS genes were progressively induced during drought stress, whereas PDH genes were specifically downregulated at day 4. In our analysis, the expression pattern of P5CS genes corresponded to Cluster 1 (Supplementary Figure S1), whereas the PDH genes were placed in Cluster 6. Zhang et al. [29] also showed that proline metabolite steadily accumulated in both shoots and roots from 4 d onwards. However, root proline content under drought stress was several folds higher than in shoots, despite the higher expression levels detected for P5CS in the shoots. All these findings suggested that proline accumulation in different organs and tissues may depend on the equilibrium amongst catabolism, biosynthesis, and transport. Proline metabolism is strongly connected to many cellular processes and is considered a regulatory hub of signaling pathways [48]. For this reason, a complex regulation of the expression of the genes involved in proline metabolism is expected.

Our cluster analysis successfully identified MtPDH as an important hub of the regulatory network that acts specifically, and transiently, at 4 days after water depletion. This regulatory network involves gene encoding of three MtKNOX3-like TFs and their several possible BELL/BLH interactors as potential repressors of PDH genes and may represent an important early switch in the response to water scarcity. We identified several KNOX/BLH binding sites in the regulatory regions of the MtPDH gene. The KNOX/BLH binding sites were found in the 3000 bp upstream of the PDH gene and intronic regions. Strikingly, the third intron of the gene was enriched in KNOX/BLH motifs. These findings suggest that the regulatory relationship between MtKNOX3-like TFs and MtPDH may be direct, although further analysis would be needed to confirm this hypothesis. No direct negative regulators of PDH have been identified so far. Several bZIP TFs are known to induce PDH expression [48], whereas only one mechanism to negatively regulate PDH expression has been demonstrated [49]. Veerabagu et al. showed that the Arabidopsis type-B response regulator 18 (ARR18) physically interacts with bZIP63 and negatively interferes with the positive transcriptional activity of bZIP63 on the PDH1 promoter.

As mentioned above, two proline biosynthetic genes (P5CS) (Mt v4 ID: Medtr4g020110, Mt v5 ID: MtrunA17_Chr4g0008951—Mt v4 ID: Medtr7g063650; Mt v5 ID: MtrunA17_Chr4g0008951), represented in the microarray by three probes (Mtr.33511.1.S1_at, Mtr.42902.1.S1_s_at, and Mtr.11510.1.S1_at, respectively) were placed instead in Cluster 1 (Supplementary Figure S1 and Table S6). This cluster is characterized by the genes that increased as drought stress progressed, with a peak expression at severe drought stress. They all had a high cluster score (> 0.9) and may also represent important hubs in the regulatory network formed by the anticorrelated Cluster 1 and Cluster 7 (Supplementary Figure S1 and Table S6). However, these two clusters are not related to most MtKNOX3-like genes and their BELL/BLH partners, and there are no indications that they could be part of the KNOX2/BLH regulatory pathway.

In plants, the variation in proline anabolism and catabolism gene expression directly relates to the high proline accumulation [48,50,51]. No proline accumulation is detected under well-watered conditions, thus implying that proline accumulation is specifically induced by low water potential. Increasing the abundance of specialized metabolites and associated primary metabolites, such as ascorbates and proline, are considered as a marker for enhanced biotic and abiotic stress tolerance [8]. Indeed, transgenic RNAi lines displayed a delay in the accumulation of proline with respect to the controls, confirming a possible role of MtKNOX3-like TFs in rapidly switching off the proline degradation genes to speed proline accumulation in response to stress and to protect the cell membranes from drought-induced osmotic stress.

During drought stress, both osmotic and oxidative stresses are induced [7]. Proline and sugars act as osmolytes and scavengers of reactive oxygen species (ROS). They keep osmotic balance and cell turgor, stabilize membranes, and consequently prevent electrolyte leakage. They also bring ROS concentrations to normal ranges, thereby preventing oxidative bursts in plants [7,8].

MtKNOX3-like RNAi lines, besides being less tolerant to drought, rapidly underwent dark-induced senescence and were less tolerant to oxidative stress when subjected to oxidative paraquat treatment. These findings indicate that MtKNOX3-like genes may have a role in responding to other stress, involving oxidative bursts, through the modification of proline catabolism.

Overall, our approach, that integrated in planta functional studies and molecular and physiological phenotyping with a targeted knowledge-based bioinformatic analysis, was successful in identifying a novel KNOX/PDH regulatory module acting at early stages of drought stress perception. In our hands, partitioning the complex transcriptomics data based on deep knowledge of gene function and relative pathways allows to analyze subsets of data and reduce noise in the bioinformatic analysis, thus prompting to explore novel genetic relationships in complex biological networks.

The novel role in drought tolerance identified for KNOX2 TFs, until now known as players in developmental processes, opens new intriguing questions about genetic networks acting in the trade-off between abiotic stress response and growth.

4. Materials and Methods

4.1. Plant Material, Growth Conditions, and Stress Treatment

Medicago truncatula genotype R108 seeds were provided by INRA Montpellier, SGAP Laboratory, Mauguio, France. Seeds were scarified with sandpaper, sterilized for 15 min in bleach (12% [v/v] sodium hypochloride, Dexal, Eurospin IT), rinsed 5–6 times in sterile water, and vernalized in the dark at 4 °C in water-containing Petri dishes. After 2 days, the seedlings were transferred to 20 °C in the dark for germination. After 24 h, the germinated seedlings were transferred to soil (Cactaceae soil S.Q. 12, Vigor Plant Italia) and grown in a pot at 23–26 °C under long-days conditions (16 h light/8 h dark at 100 μmol/m²/s) and 40% relative humidity. For standard growing conditions, the plants were watered from the soil top every third day by alternating between water and a nutritive solution (3 mL/L Bayfolan Universale N/P/K, Bayer garden).

M. truncatula germinating seeds were selected and sown in squared plastic pots (9 × 9 × 10 cm with 0.6 L in volume), filled with pre-autoclaved soil (Cactaceae soil S.Q. 12, Vigor Plant Italia), and grown at 23–26 °C under long-days conditions (16 h light/8 h dark). The photon flux density at soil level was 100 μmol/m²/s and supplied mainly with cool light. The plants were watered from the soil top every second day, alternating between water and a nutritive solution (3 mL/L Bayfolan Universale N/P/K, Bayer garden) until 24 days after planting. At that time, most of the plants showed an emergence of the seventh leaf and an expanded sixth leaf. Half of the plants were put under drought-stress imposition by withholding watering, while the remaining half were kept with regular watering. For RNA isolation and metabolite/proline/sugar analysis, pools of the third, fourth and fifth fully expanded leaves were sampled in biological triplicates for both the control and drought-stressed plants at all time points. The leaf samples were frozen in liquid nitrogen and stored at −80 °C prior to RNA isolation and metabolite analysis. Soil humidity was measured daily via the Flower Care Smart Sensor (http://www.huahuacaocao.com/product (accessed on 29 June 2023)) and soil water loss was assessed via daily weighting watered and drought-stressed reference plant pots.

4.2. Plasmid Construction and Plant Transformation

Constructs for RNA interference (RNAi) of RNAi constructs targeting multiple M. truncatula KNAT3/4/5-like genes were previously described [28]. For stable plant transformation, two RNAi constructs targeting multiple KNOX2 genes, one based on the KNOXII domain of the MEINOX (131 bp) (Mt KNAT3/4/5-like RNAi) and one as part of the homeodomain (HOX) (134 bp) (Mt KNAT3/4/5-like hox RNAi) [28], were introduced into the Agrobacterium tumefaciens strain EHA105 [52] via electroporation.

M. truncatula Agrobacterium-mediated genetic transformation and plant regeneration were done according to Ratet and Trinh (EMBO practical course on the New Plant model System Medicago truncatula. 19 November–1 December 2001, Gif-sur-Yvette, France) [33] using healthy, young expanded leaves from 4 to 6-week-old plants, with some modifications. An additional step was introduced to minimize Agrobacterium overgrowth after the 2 days of leaf explant cocultivation. The plant material was washed 4–5 times with sterile water containing 400 mg/L of Augmentin, blotted dry on sterile filter paper and transferred to a SH3a solid medium. This simple additional step prevented Agrobacterium overgrowth and a good number of viable calli were obtained, most of which gave rise to somatic embryos within 3 weeks. Emerging embryos were transferred to a fresh medium containing 40 mg/L of kanamycin for selection.

4.3. Plant Genotyping and Selection of Transgenic Plants

M. truncatula’s positive primary transformants (T0) and their T1 and T2 progeny harboring the RNAi constructs were identified via a direct PCR amplification from cotyledon or through young leaf tissues using the Phire Plant Direct PCR Kit (Thermo Scientific, Waltham, MA, USA). A small piece of plant material was cut using a sterile pipette tip, placed in 20 μL of dilution buffer and crushed by pressing it briefly against the Eppendorf tube wall. For DNA amplification, 0.5 uL of the supernatant was used as template for a 20 μL PCR reaction containing Phire Plant PCR Buffer (including dNTPs and MgCl₂), 0.5 μM of each primer, and 0.4 μL of the Phire Hot Start II DNA Polymerase. Amplifications were carried out in a thermal cycler (Eppendorf Mastercycler Personal, Hamburg, Germania) with cycling conditions as follows: 98 °C for 5 min; 40 cycles at 95 °C for 5 s, 55–65 °C for 5 s, and 72 °C for 20 s; and final extension at 72 °C for 1 min. The primers used are shown in Supplementary Table S7.

4.4. Physiological and Biochemical Measurements

Water loss was measured as previously described [53]. Ten fresh trifoliate expanded leaves from well-watered 5-week-old M. truncatula plants were detached from three T1 plants for each T0 original transformed plant and immediately weighed. The leaves were then placed in ventilated petri dishes and incubated in the growth chamber at 23 °C under continuous light. Fresh weights of the leaves were measured at time 0 and at designated time intervals (16, 20, 24, and 40 h). Water loss was calculated on the basis of the initial weight of the plants. All plant materials were sampled at 9:00, 3 h after the beginning of the day cycle.

To evaluate the oxidative stress tolerance, the excised leaf discs obtained from CTR and RNAi M. truncatula 5-week-old plants were subjected to methyl viologen (paraquat). Paraquat is the common name of the herbicide methyl viologen (MV; N,-N′-dimethyl-4,-4′-bipyridinium dichloride), which acts in the production of reactive oxygen species (ROS) via a light dependent mechanism. The ion leakage out of the leaf discs, due to the destruction of membrane lipids, was measured as an increase in the conductance of the floating solution, as described by Kasajima et al. (2017) [54].

For inducing the senescence of the detached leaves in the darkness, two to three leaves from 5-week-old M. truncatula plants were excised and placed in Petri dishes, wrapped with double-layer aluminum foil, and then kept at 22 °C. Biological triplicates were used. The total chlorophyll content was measured in the detached leaves at the initial experimental stage (0 d) and after 6 and 10 days, as previously described [37].

Proline accumulation was measured on pools of the third, fourth, and fifth fully expanded leaves, sampled in biological triplicates for both the control and drought-stressed plants at 0, 4, 8, and 12 days, following Lee et al.’s (2018) procedure [55].

4.5. Quantitative RT-PCR Analysis of Gene Expression

For quantitative real-time PCR (qRT-PCR) gene expression analysis, sample homogenization was performed with a TissueLyser (Qiagen, Hielden, Germany) following the manufacturer’s procedures. Total RNAs were extracted from the frozen leaves of M. truncatula using the Trizol reagent (Sigma-Aldrich, Darmstadt, Germania, http://www.sigmaaldrich.com/ (accessed on 29 June 2023)) and treated with DNase I (Qiagen, Hielden, Germany). First-strand cDNA was synthesized from 1.5 μg of the total RNA using the Superscript III first-strand synthesis system (Invitrogen, Darmstadt, Germania). Primer design was performed using Primer3 software (available at http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi (accessed on 29 June 2023)). qRT-PCR reactions were performed using the Eco Real-Time PCR System (Illumina, San Diego, CA, USA) following the manufacturer’s instructions, using 50–70 ng of the template cDNA (Biotool easy mix with an Eva green fluorophore) and 300 nM of the final primer concentration. Cycling conditions were as follows: 95 °C for 10 min; and 40 cycles at 95 °C for 15 s, 60 °C for 15 s, and 72 °C for 15 s. Three technical replicates and two independent biological experiments were performed in all cases. A ratio with MtEF1-α reference genes was used for the different experiments (the primers used are listed in Supplementary Table S7). The ratio value of the experimental control condition was set up to 1 as a reference to determine the relative expression of the fold change.

4.6. Gene Expression Meta-Analyses and Microarray Pathway Annotation

The transcriptomic data of either shoot or root drought response series from the article of Zhang et al. [11], based on the Affymetrix Medicago GeneChip containing 50,900 probes, were retrieved from the M. truncatula Gene Expression Atlas (http://mtgea.noble.org/v2/ (accessed on 29 June 2023)). The genes encoding the TFs were annotated and classified by family based on the TF annotations in the PlantTFDB database [36]. Gene IDs were retrieved by the Expression Atlas Server (https://lipm-browsers.toulouse.inra.fr/pub/expressionAtlas/app/mtgeav3/ (accessed on 29 June 2023)) [56,57]. Specific pathways were annotated based on the key words analysis, current literature, and homology with the A. thaliana genes. Different modules were annotated: TF, CK, GA, ABA, oxidative stress, and proline metabolism. The analyses started from the microarray values of 2468 selected probes representing the pathways of interest, from which the expression data were extracted and used to create a starting sub-database (Supplementary Table S1).

4.7. Correlation Analysis and Gene Co-Expression Networks

The pipeline for the k-means cluster analysis was performed according to Testone et al. (2019) [37] with some modifications. Briefly, mean values of the expression data for the 2468 selected probes from the drought stress shoot samples [29] were log-transformed using log2(x + 1) for data normalization. The optimum number of clusters was determined based on the converging results of the sum of squared errors (SSE) estimate and the Calinsky criterion [37]. Data scaling, k-means clustering, and visualization were performed in R as previously described [37]. The cluster analysis data are available in Supplementary Table S6. Clusters 3 and 6 were selected for the GCN’s construction, and the expression data for the TALE genes were also added. From the selected clusters, only the genes with a cluster score ≥ 0.8 were used to perform pairwise correlation analysis of the expression values from the biological triplicates. The expression data were log-transformed using log2(x + 1) for normalization, and Pearson pairwise correlation analysis was conducted across the selected samples using the “corrplot” and “hclust” packages of R software (https://rdrr.io/cran/corrplot/man/corrplot-package.html (accessed on 29 June 2023)). Significant correlations (p-value ≤ 0.05) with a Pearson’s correlation coefficient (r) ≥ |0.6| were used to develop the GCNs. The Cytoscape software platform v. 3.5.1 [42] (Shannon et al., 2003) was used to visualize the networks, to determine the relationships among the selected genes, and to identify the hub genes in the orchestration of the response to water withdrawing at day 4.

4.8. KNOX/BLH Binding Sites Analysis

To identify the putative cis-acting elements of KNOX/BLH TFs in MtPDH gene sequences, the 3000 bp length located upstream of the gene and the intronic regions were retrieved from the genome of M. truncatula (Mt4.0 v1) using Phytozome 13 [58] (Supplementary Figure S2). Promoter and intronic sequences were then searched for specific motifs using RSAT Plants [59]. The motifs searched were the following: KNOX/BLH_bs1 (KGACM); KNOX/BLH_bs2 (TGAYTGA); and KNOX/BLH_bs3 (TGATKKGA).