Elucidating Drought Stress Tolerance in European Oaks Through Cross-Species Transcriptomics

The impact of climate change that comes with a dramatic increase of long periods of extreme summer drought associated with heat is a fundamental challenge for European forests. As a result, forests are expected to shift their distribution patterns toward north-east, which may lead to a dramatic loss in value of European forest land. Consequently, unraveling key processes that underlie drought stress tolerance is not only of great scientific but also of utmost economic importance for forests to withstand future heat and drought wave scenarios. To reveal drought stress-related molecular patterns we applied cross-species comparative transcriptomics of three major European oak species: the less tolerant deciduous pedunculate oak (Quercus robur), the deciduous but quite tolerant pubescent oak (Q. pubescens), and the very tolerant evergreen holm oak (Q. ilex). We found 415, 79, and 222 differentially expressed genes during drought stress in Q. robur, Q. pubescens, and Q. ilex, respectively, indicating species-specific response mechanisms. Further, by comparative orthologous gene family analysis, 517 orthologous genes could be characterized that may play an important role in drought stress adaptation on the genus level. New regulatory candidate pathways and genes in the context of drought stress response were identified, highlighting the importance of the antioxidant capacity, the mitochondrial respiration machinery, the lignification of the water transport system, and the suppression of drought-induced senescence – providing a valuable knowledge base that could be integrated in breeding programs in the face of climate change.

Several omics analyses (transcriptomic, proteomic, and metabolomic) have already been performed to identify drought-related patterns as well as their associated signaling and regulatory pathways in plants (Joshi et al. 2016), including various tree species (Harfouche et al. 2014). However, despite the high economic relevance of oaks as forestry species in Europe, molecular mechanisms of DS tolerance have explicitly been investigated so far only by a limited number of studies: in Quercus robur, transcriptomic and proteomic responses to long-term drought exposure elucidated the involvement of genes with protective functions, such as chaperones and dehydrins (Sunderlíková et al. 2009;Spieß et al. 2012) and highlighted the degradation of RuBisCO (Sergeant et al. 2011). Porth et al. (2005) assessed differences in DS responses in Q. robur and Q. petraea by analyzing a set of known osmotic-stressinduced transcripts. In addition, the high natural variation in DS related genes of Q. robur in contrast to Q. petraea provided first evidences of natural selection in the frame of drought adaptation processes (Homolka et al. 2012(Homolka et al. , 2013. This concept of local adaptation to changing environments is supported by a study of three European white oak species (Q. robur, Q. petraea and Q. pubescens) by Rellstab et al. (2016). In Q. suber, an activation of abscisic acid (ABA) responsive genes and ABA-dependent signaling was characterized in the root transcriptome in response to drought (Magalhães et al. 2015), and an increase of cytosine methylation was recognized in heat-treated leaves (Correia et al. 2013). A detailed analysis of the transcription factor QsMYB1 highlighted its role in the response to heat and drought stress in this species (Almeida et al. 2013). Finally, various studies were performed to elucidate the response to DS of Q. ilex, reporting the involvement of enzymes of carbohydrate and protein metabolism (Jorge et al. 2006), an increase in stress response proteins such as those involved in detoxifying reactive oxygen species (Echevarría-Zomeño et al. 2009), an increase of proline and g-Aminobutyric acid (GABA) in severely drought stressed seedlings (Rodríguez-Calcerrada et al. 2017), increased concentrations of sugars and polyphenolic compounds in leaves (Rivas-Ubach et al. 2014), and a down-regulation of glycolysis and stimulation of ATP synthesis in roots and seedlings subjected to water limitation (Valero-Galván et al. 2013;Simova-Stoilova et al. 2015). Despite this seemingly long list, a comprehensive species-spanning comparative analysis of DS response has not been performed so far.
Although oaks are generally considered to be DS tolerant, the level of tolerance differs considerably among oak species, reflecting their broad adaptation potential to environments with differing soil moisture regimes. In the current study we took advantage of the present DS tolerance gradient from the less tolerant pedunculate oak (Q. robur) via the quite tolerant pubescent oak (Q. pubescens) to the very tolerant holm oak (Q. ilex). In addition to their contrasting geographical ranges (pedunculate oak is a temperate-boreal species, holm oak a Mediterranean species and pubescent oak has an intermediate distribution), the three species are characterized by a different leaf habit, whereby pedunculate oak and pubescent oak are deciduous while holm oak is evergreen. All these differences were considered to unveil new regulatory candidate pathways and genes in the context of DS response by means of comparative transcriptomics, from the species up to the genus level.

Plant material and experimental site
For the purpose of comparative transcriptomics, three valuable European oak species with differing DS tolerance were chosen: Q. robur L. (pedunculate oak), a common deciduous broad leaved species showing a wide distribution range across Europe (Eaton et al. 2016), Q. pubescens Willd., the downy or pubescent oak, a more thermophilous (semi-) deciduous oak species which occupies almost all of Southern Europe, ranging into Western and Central Europe (Pasta et al. 2016), and the evergreen holm oak, Q. ilex L., a sclerophyllous species well adapted to drought occurring in the Central-Western Mediterranean basin . Three to five 9-year-old plants of Q. robur (provenance: Mitteldeutsches Tief-und Hügelland D-81705), Q. pubescens (provenance: Languedoc F-QPU701), and Q. ilex subsp. ilex (provenance: Languedoc F-QIL 702) were grown outdoors in two basins (6 Ã 4.5 Ã 2 m 3 each) for two years for root development from 2012 onwards in central Germany (50°10'08.9"N 8°37'48.6"E). The basins were filled with silty loam and covered with a semi cylindrical foil tunnel to allow irrigation control. Temperature, RH and solar radiation were captured by a weather station beside the basins. Monitoring of water tables in the basins was enabled by floaters within drainage pipes on the ground of the basins and kept at -1.8 m throughout the root development phase. Further details can be found in Früchtenicht et al. (2018b) and Kotrade et al. (2019).

Drought stress treatment and monitoring
One basin was chosen for the drought stress (DS) treatment, the other one as control (CO). In 2014, irrigation of the DS basin was stopped on DOY (day of year) 170 for 124 days while the CO basin was kept on a constant water table. The DS level in the plants was monitored by measuring predawn water potential (C PD ) with a scholander pressure bomb (Skye Instruments Ltd, UK) and the performance index (PIabs) with a Pocket PEA (Hansatech Instruments Ltd, UK). More detailed information on the experimental design, the environmental and physiological data were recently published in Früchtenicht et al. (2018b) and Kotrade et al. (2019).

Sampling and RNA extraction
Sampling was done on DOY 287 (Q. robur, Q. pubescens) and 288 (Q. ilex) in 2014 when C PD values showed significant differences between CO and DS groups in all species (for details see Kotrade et al. 2019). Leaves were taken from three trees per treatment and species, except for Q. robur, where only two trees of the CO group were sampled, since one individual perished. Per tree, three to five fully developed leaves of upper branches in southern orientation were sampled. For the evergreen species Q. ilex, only recent leaves developed in 2014 were collected. Leaves were immediately shock-frozen in liquid nitrogen and kept at -80°until RNA extraction. A modified CTAB-based RNA extraction protocol was used to extract RNA from leaves pooled for each tree (see Kotrade et al. 2019).

Library preparation and sequencing
Extracted total RNA was sent to VBCF NGS Unit (www.viennabiocenter.org/facilities) for library preparation and Illumina RNA sequencing. An overview of the samples sent for sequencing is given in Table S1. Shortly, after RNA quality control (Agilent 2100 Bioanalyzer), RNA library preparation of the 17 samples was done using the polyA enrichment strategy according to the manufacturer's protocol (NEB chemistry), pooled and sequenced together on three lanes (technical replicates). Sequencing of the first two lanes was done on a HiSeq 2500 as 150 bp PE reads in rapid mode, the third lane was sequenced as 125 PE reads.
Quality control and pre-processing of sequencing data All sequence reads were cleaned in order to guarantee high quality (HQ) data by removing remaining reads derived from the phiX genome (NCBI accession NC_001422) (Sanger et al. 1977), ambiguous reads, low quality regions (Q30), and short sequences (, 50 bp) from the datasets using bowtie2 (Langmead and Salzberg 2012), SAMtools (Li et al. 2009), and BEDtools (Quinlan and Hall 2010). Additionally, all HQ read pairs were tested whether their ends overlap using Flash with a minimal overlap of 10 bp and a maximal overlap of 250 bp (Mago c and Salzberg 2011).

De novo assembly
By combining all datasets (CO and DS) of each species, a de novo assembly was performed using Trinity (Grabherr et al. 2011;Haas et al. 2013) with default values and a minimal contig length of 100 bp to establish a species-specific leaf reference transcriptome. The resulting contigs were evaluated by mapping the pre-processed reads used for the assembly back to the assembled contig sequences using bowtie2 with default settings (Langmead and Salzberg 2012) and only those contigs consisting of at least five reads were retained.

Differential gene expression
Species-specific transcript abundance estimation was performed using RSEM (Li and Dewey 2011). RSEM performs abundance calculations on isoforms and genes. The gene expression level, that were used for further analysis is the sum of isoform expression levels. Additionally, RSEM outputs two files, one with the expected counts, that were used for DEG analysis with edgeR and one with TPM values (normalized by length) that were used comparing expression across the species (see below). CPM (counts per million mapped reads) values for each sample were calculated from expected counts using edgeR's cpm function. For annotation and differential gene expression only those genes were included that have a CPM greater than one in at least two samples. After filtering, further expression analysis was performed using RSEM expected counts and edgeR exactTest (Robinson et al. 2010;McCarthy et al. 2012) as well as p-value correction for multiple testing (FDR) with the Benjamini and Hochberg (BH) procedure. Significant genes were considered with |logFC| . 1 and a FDR , 0.05.

Functional annotation and classification
The de novo assembly filtered for low expressed genes were functionally annotated and classified as follows: Open reading frames (ORFs) were predicted using TransDecoder with a minimal length of 100 amino acids (Mago c and Salzberg 2011). Similarity searches were performed using BLASTX and BLASTP against the UniProt (Swiss-Prot) and against NCBI non-redundant Viridiplantae databases. Gene Ontology (GO) and KEGG pathways were assigned to the annotated BLASTX results using Trinotate (McCarthy et al. 2012). Similarities were considered significant with E-values # 1e-5. The presence of known functional protein domains was performed using HMMER (Eddy 1998) and the Pfam protein domain database.

GO enrichment analysis
Functional over-representation analysis of significant differentially expressed genes was performed using the R-package TopGO with the Fisher's exact test (Alexa et al. 2006). Gene Ontology (GO) categories are declared as significantly over-represented with a p-value level of , 0.01. GO enrichment visualization was performed using REVIGO (Supek et al. 2011).

Gene expression among species
The three species-specific assembled transcriptomes (including low expressed transcripts with CPM , 1) were compared to each other to identify orthologous genes in each of the other species transcriptomes using ORTHOfinder with default parameters (Emms and Kelly 2015). Only orthogroups that contain at least one gene of each species were considered for further analysis. Differential gene expression comparison among Quercus species was performed with exploratory data analysis methods. To compute the species-wise gene expression level for each orthologous group, normalized gene expression levels (TPM) of all genes were summed up per species. A pairwise comparison (Q. robur -Q. pubescens, Q. robur -Q. ilex and Q. pubescens -Q. ilex) has been performed with the workflow of the R package crossr (Mantegazza 2018) to find orthogroups that behave differently under DS condition between the species. Crossr fits a linear model and performs ANOVA (design formula: $species + treatment + species:treatment). The 200 orthologous groups with highest F-values of each pair from the ANOVA test were selected and used for further visualization.

Physiological status of drought stressed plants
The three trees species were grown together in loamy soil in 2 m deep basins in a competition experiment setup, thus experiencing identical treatments of complete water withholding for four months (drought stress, DS) with no ground water in the basin or continuous well watering with maintaining the ground water level at -1.8 m (controls, CO), respectively. However, due to different morphological responses and properties, they in fact did not experience the same severity of drought stress (DS) during the experiment. After 4 months of withholding water, volumetric soil water content in the upper soil layers (10-20 cm depth) had decreased from 25-30 to below 10% in the DS basin. This resulted in significant decreases of predawn water potentials to -2.18 MPa (Q. robur, control [CO]: -0.38 MPa), -0.57 MPa (Q. pubescens, CO: -0.14 MPa) and -0.81 MPa (Q. ilex, CO: -0.13 MPa, as described in Früchtenicht et al. 2018b). Thus, while Q. robur (with the shallowest root system of the three species) was severely drought stressed, the other two species (with deeper rooting) experienced similar levels of moderate drought stress. This was also reflected in physiological parameters of the leaves: while the so-called performance index (PIabs) of the photosynthetic apparatus was significantly lower in DS Q. robur than in well-watered control trees; this was not the case in Q. pubescens and Q. ilex. PIabs is a chlorophyll fluorescence parameter summarizing various efficiencies of light collection and electron transport steps in the linear electron transport chain, which can be used as an indicator for e.g., DS (Früchtenicht et al. 2018a).
RNA sequencing output and pre-processing A total of 961 million raw reads were sequenced (258 mill. for Q. robur, 316 mill. for Q. pubescens, and 386 mill. raw reads for Q. ilex). The PhiX contamination was on average 0.28%. Due to high quality of the raw reads, on average 97.9% were retained after pre-processing, resulting in 256 mill. reads for Q. robur, 313 mill. reads for Q. pubescens, and 383 mill. reads for Q. ilex.

De novo assembly
After de novo assembly and further filtering of transcripts with less than five reads mapped, a total number of 335,571 transcripts for Q. robur, 420,306 for Q. pubescens and 100,158 transcripts for Q. ilex could be obtained. Best hit annotation against the whole UniProt Swiss-Prot databases showed that around 1% of the annotated transcripts were mouse (Mus musculus) contaminants in the Q. ilex assembly. Aligning the raw reads to the mouse coding sequences showed that the sample 23491 (see Table S1) from the Q. ilex control group contained a high number (3.64%) of mouse reads. To reduce false positive differentially expressed genes, transcripts that have more than 95% identical matches and an e-value # 1e-5 with a mouse gene and/or protein were removed for further analysis. One explanation could be, that the experiment was performed in the Scientific Garden of Goethe University with significant populations of various wild-living animals, and that colonization of the experimental setup with mice was observed. Especially the dense leaf cover of Q. ilex may have attracted the animals as a shelter during their activities and may have given rise to contamination of the leaves.
After filtering of low expressed genes (by keeping genes with a CPM . 1 in at least two samples), which is done to increase the sensitivity of the analysis of differential gene expression (cf Sha et al. 2015), the assemblies consisted of 68,377 transcripts for Q. robur, 70,721 for Q. pubescens, and 53,111 transcripts for Q. ilex, with a N50 value of 911 bp, 768 bp, and 1,069 bp, respectively (Table 1).

Annotation
Using UniProt Swiss-Prot database more than 60% of all genes in the three Quercus species could be annotated with an e-value #1e-5 while even 84-88% of the genes could be annotated when using NCBI nonredundant database filtered for Viridiplantae sequences. A summary of the annotation is given in Table 2. In more than 90% of these annotations, the recently published Quercus suber genome (CorkOak1.0, GCF_002906115.1) is assigned as best hit, followed by Juglans regia (GCA_001411555.1) in 1.6% of all annotations (Table S2). GO terms could be assigned to more than 62% and Pfam protein domains to more than 47% of all genes from the three species. Transdecoder could predict proteins in more than 68%.
The most frequent GO terms common to all three assembled leaf transcriptomes in the cellular component category are "nucleus", "integral component of membrane" and "plasma membrane"; in the molecular function category "ATP binding", "metal ion binding" and "RNA binding"; and in the biological process category "transcription, DNA-templated", "regulation of transcription, DNA-templated" and "defense response" (Table S3). The GO terms "endonuclease activity" in the molecular function and "RNA modification" in the biological function category have an overrepresentation of genes in Q. ilex compared to the other two species (4.3% and 3.3% of genes, respectively vs. around 2.5% and 2%, respectively), see Figure 1A-D and Table S3. Pfam functional domain analysis showed that pentatricopeptide repeat domains are more frequently annotated for the Q. ilex assembly ( Figure 2 and Table S4).

Species-specific differential gene expression between CO and DS
Considering 24,287 Q. robur genes, 415 genes were found significantly differentially expressed, with 132 higher and 283 lower expressed in the CO group than in the DS group. In Q. pubescens, where 27,983 genes were analyzed, 79 genes were found to be significant differentially expressed, with 48 genes significantly higher and 31 significantly lower expressed in the CO group compared to the DS group. The differential gene expression analysis of the 26,211 Q. ilex genes resulted in 222 significantly different expressed genes after water deprivation. 112 were significant higher and 110 were significant lower expressed in the CO group than in the DS group. A summary statistic for all species of the differential gene expression analysis and GO enrichment analysis is given in Table 3 and Figure 3, including MA and volcano plots, heatmaps and clusters.
GO enrichment analysis of the Q. robur differentially expressed gene dataset showed that 84 GO terms are significantly enriched. Most enriched GO terms in the category biological processes are "defense response", "response to biotic stimulus" and "response to stress" (Table  S5). Visualization ( Figure 4A) of enriched GO terms shows further an enrichment of GO terms connected with phenylpropanoid and monoterpenoid metabolic processes (e.g., ECOD) as well as ion transport (e.g., YSL1 and OCT7) ( Table S6).
The 40 top upregulated genes (FDR , 2.38E-03) include those encoding for the NRX2 and DOX1, proteins that protect the plant against oxidative stress. Further, a BURP domain protein RD22like, a known dehydration-responsive protein, and a L-ascorbate n oxidase-like was listed (Table S6). Further, we found the gene LEA known to play a major role in drought and other abiotic stresses tolerance in plants, and WAT1, an auxin transporter. Among the entire list of 132 upregulated genes, we additionally found two genes belonging to the NAC domain-containing proteins (e.g., RD26), two related to sugar transport/metabolism (SPS4 and MSSP2). Among the 40 most downregulated genes (FDR , 4.54E-09), three cucumisinlike were listed, four genes that are annotated as LRR-RLKs (leucinerich repeat receptor-like serine/threonine-protein kinases) and many other receptor-like proteins. Additionally, GLR2.8, the receptorlike protein EIX2, the beta-amyrin 28-oxidase, and two genes annotated as glutathione S-transferase U7-like were listed (Table S6). Further genes also found to be downregulated in the entire dataset of 283 genes were several sugar transporters (Table S6). Besides a high enrichment of "defense response" GO terms in Q. pubescens, the GO enrichment analysis showed an enrichment of amino acid activation and especially (lysyl)-tRNA aminoacylation (Table S5 and Figure 4B). Concordantly to the GO enrichment analysis, five lysine-tRNA ligases were found significantly downregulated. Additionally, two genes annotated as F-box protein SKIP23, GLR3.3 and GLR3.6, the receptor-like protein EIX1, and cytochrome P450 87A3 were listed. Further downregulated were cellulose synthase-like protein E6, TERC, and linamarin synthase 2, which is involved in the biosynthesis of linamarin, a toxic cyanogenic glucoside (Table S7).
The significantly upregulated genes included genes involved in hormonal pathways, such as ERF4 and CYP74A1 / AOS1, involved in jasmonic acid biosynthesis, further two genes involved in sugar transport/metabolism (GPT2 and STP13), additionally, two genes of the phenylpropanoid pathway, PAL and 4CL, as well as PAT24 and the bHLH92 transcription factor, which is known to respond to osmotic stress. Further, two heat shock proteins were found, LOX3.1, BAT1, which is involved in GABA transport, and the E3 ubiquitin-protein ligase PUB23, that negatively regulates water stress response in Arabidopsis thaliana (Table S7).
In Q. ilex, the most enriched GO terms of the category biological processes were "defense response", "DNA integration" and "response to stress" (Table S5). Notable are also an enrichment of "DNA methylation on cytosine" (CMT1 and DRD1) and processes involved in "cellular respiration" (XYLT) ( Figure 4C and Table S8). In the list of the 40 most upregulated genes (FDR , 1.07E-03), the majority of genes is related to disease resistance including two that were annotated as LRK10, BDA1 and NPR4, both Ankyrin repeat-containing proteins that are regulators of plant immunity, and CDR1 and SUMM2. Further, three genes belong to the LRR-RLKs, additional three genes belong to SRKs, and COMT, involved in lignin biosynthesis, AOS3, involved in jasmonic acid biosynthesis, and OBE, genes that are required for the maintenance and/or establishment of both the shoot and root meristems, were included. Further, two genes homologous to cytochrome P450 genes: CYP75B1 and cytochrome P450 71A1. Considering the entire list of 112 genes, also a dehydration responsive protein, WAK5, MLP43 or EXPA1 were listed (Table S8).
Among many hypothetical and uncharacterized proteins in the 40 most downregulated genes (FDR , 4.66E-03), we found two genes annotated as CRK22, a putative disease resistance protein, and a LRR-RLK. Additionally, CUL1, a member of the SCF-complex, involved in jasmonate signaling, and MIK2, involved in cell wall integrity sensing, were found. Notable in the further list is also ZFWD1, DREB3, and two genes annotated as WAK1 and GAT1 (Table S8).

Comparison of orthologous genes between species
To compare gene expression levels of genes that are present in all three species, orthologous groups of genes were identified across the species resulting in 14,005 orthologous groups that contain at least one gene of each species. As more genes of one species can be related to one orthogroup, in total 25,597 genes of Q. robur, 23,747 of Q. pubescens, and 22,616 of Q. ilex were assigned to orthogroups.

Differences between oak species
The selection of the 200 orthologous genes with highest F-values (pairwise ANOVA test included in crossr workflow) of each comparison (Q. robur -Q. pubescens, Q. robur -Q. ilex and Q. pubescens -Q. ilex) led to 517 orthologous genes that show different expression in at least one species after DS (Table S9). Further visual examination of the expression patterns of these 517 orthologous genes could group 161 genes in categories that show a) a species-specific reaction, b) a different reaction pattern between deciduous and evergreen species, or c) a resemblance of the expression pattern to the gradually increasing DS tolerance of the three species, Q. robur , Q. pubescens , Q. ilex.
a. Species-specific reaction to drought stress: Orthogroups that fall in this category only show a different expression pattern between CO and DS in one of the species while in the other two species an idle expression was seen. In total, 72 orthogroups were found, 56 specific for Q. robur, four for Q. pubescens, and 12 for Q. ilex. Selected orthogroups are depicted in Figure 5. Regarding Q. robur, 32 genes were up-and 24 were downregulated, among them, homologs of the E3 ubiquitin protein ligase DRIP2 and SGR, the first a master-regulator of DS tolerance and the latter involved in chlorophyll a degradation. Further, the sugar transporter GFT1. 14, MSSP2, WRKY53 and VEP1 were found in the species specific DEG analysis of Q. robur. DS led to the upregulation of four orthogroups specific for Q. pubescens genes, including choline monooxygenase (CMO) homolog, a homolog of the f-box protein CPR30, involved in the defense response, and a homolog of the heat stress transcription factor A-4a. Quercus ilex reacted to DS with the upregulation of 10 genes and with the downregulation of two genes, among them a NADK3 homolog, modulating abscisic acid responses, two genes homologous to cytochrome P450 genes: the flavonoid 39-monooxygenase (CYP75B1), and a CYP71B5 homolog, the transcription factor bHLH30, and a homolog of a AMP deaminase. b. Reaction pattern of deciduous compared to evergreen species: Here, expression patterns of the orthologous genes are contrary for the deciduous species (Q. robur and Q. pubescens) compared to the evergreen species (Q. ilex) after DS. Among the 37 orthologous genes that might play an important role for the adaptive potential to DS in evergreen Q. ilex were EDR1, AOC, 4CLL7, FREE1, the GABA transporter 1 (GAT1), a 15.4 kD class V heat shock protein homolog (HSP15.4) and NADK3. The expression pattern of selected orthologous genes is shown in Figure 6.
n  c. Response along the DS-tolerance gradient: To encounter genes that are connected to the gradual DS tolerance which is present in the three species, we compared the not so DS tolerant Q. robur to the semi DS tolerant Q. pubescens with the DS tolerant Q. ilex to find patterns resembling this gradient as well as genes that follow the gradient of the measured level of water potential. The expression pattern of selected orthologous genes can be found in Figure 7. Among the 17 orthologous genes that show a gradually increasing gene expression with increasing DS tolerance, starting from a downregulation in Q. robur, a rather stable expression in Q. pubescens, and an upregulation in Q. ilex, a dnaJ protein ERDJ2-like homolog, a folylpolyglutamate synthase (FPGS), two homologs of TWINKLE, which is known to be involved in organelle DNA replication, a LRK10 homolog, and a 28 kD heat-and acid-stable phosphoprotein-like gene were observed. Vice versa, among the 11 genes that were characterized by an upregulation of gene expression in Q. robur and a downregulation in Q. ilex (with a stable expression in Q. pubescens), we found the indole-3-acetic acidinduced protein ARG7, the TCP9 TF, MTERF5, involved in ABArelated stress response, and a Zinc finger CCCH domain-containing protein 63 homolog. Furthermore, 21 genes showed downregulation in Q. robur and an upregulation in Q. pubescens (and a stable expression in Q. ilex), e.g., AGO5, the guanine nucleotidebinding protein-like NSN1, homologs to VQ motif-containing protein 22 (VQ22, JAV1) and to the WRKY transcription factor 33 / WRKY24-like, whereby VQ proteins are known to interact with WRKY transcription factors (Chi et al. 2013), a homolog to heavy metal-associated isoprenylated plant protein, shown to be involved also in transcriptional responses to cold and drought. And further 13 genes showed an upregulation in both, Q. pubescens and Q. ilex, and a downregulation in Q. robur. Among them, the cinnamoyl-CoA reductase 1, involved in lignin biosynthesis, PAO1, ACOX1, involved in jasmonate biosynthesis, plant L-ascorbate oxidase (AO), a homolog of a known stress-responsive GST gene, and a scopoletin glucosyltransferase-like homolog (TOGT1).

Reaction upon DS on the genus level
Using the strict criteria for individual DEG analysis (FDR , 0.05 and |logFC| . 1), no orthologous groups were found that contain significantly differentially expressed genes for all species after DS. However, considering a p-value , 0.1, five orthologous groups were found to be upregulated, while nine were found to be downregulated similarly across all three Quercus species (Table S10). Among the upregulated genes, a homolog to the glucose-6-phosphate/phosphate-translocator (GPT), DSP1 and G3Pp1 were listed (Figure 8). Among the downregulated orthologous genes, CRK10, CNGC1, and BAM4 were found ( Figure 8).

DISCUSSION
This study describes the species-specific DS response on the transcriptional level of three European oak species with respect to various biological processes, such as hormonal pathways, osmoprotection, photosynthesis and cell wall remodeling. Through the comparison of orthologous genes present in all three species, we were able to point toward differences between deciduous and evergreen species as well as highlight genes whose transcriptional behavior correlates to the DS-tolerance gradient, from the least to the most tolerant species. When drought occurs during vegetative growth, plants react in a flexible way and reprogram growth (for reviews see Claeys and Inzé 2013;Pierik and Testerink 2014). In this respect, an important feature is a drought avoidance strategy including the rearrangement of root architecture and increasing the root/shoot ratios, thus ameliorating the negative effect of a naturally drying soil on plant water supply (for reviews Malamy 2005;Brunner et al. 2015). Root/shoot ratios also depend on latitudinal origin and season, more northern provenances show a higher root/shoot ratio at the end of the growing season (Cannell and Willett 1976). In the current study, the root depth was measured in autumn: Q. pubescens showed the deepest root system (-165 cm), followed by Q. ilex (-150 cm) and Q. robur (-100 cm) (Früchtenicht et al. 2018b); results that correlate with the different severity of the DS experienced by the leaves, but not with origin or genotype of the plant material. Under the severe stress experienced by the Q. robur leaves, stomatal closure over long periods is unavoidable, resulting in CO 2 deprivation and Mehler reaction with subsequent ROS stress. According to the C PD values above -0.8 MPa, Q. pubescens only experienced very mild DS (Früchtenicht et al. 2018b), nevertheless, we found a decent response to DS on the transcriptional level.
Species-specific response to DS As a general feature, Q. robur revealed much more changes in gene expression (415 differentially expressed genes between CO and DS) than the two Mediterranean (and more drought tolerant) species, in accordance with its much lower water potential during water withholding despite the identical abiotic conditions, to which the three species were exposed. Apparently, while Q. pubescens and Q. ilex were in a state of drought avoidance/drought resistance reactions, Q. robur already suffered severely from the treatment, resulting in more fundamental changes in metabolism and the onset of catabolic processes.
Q. robur: As expected, the analysis of GO terms ("biological processes") showed a significant enrichment of "defense response"-related GO terms (biotic and abiotic) in all three species, however, Q. robur further showed an enrichment of GO terms related to "secondary metabolism" (e.g., phenylpropanoid, monoterpenoid), an indication that the antioxidant machinery was induced. This is also supported by the expression of ROS scavengers, NRX2 and DOX1, known to reduce oxidation (De León et al. 2002;Kneeshaw et al. 2017). Also, sucrose was produced (SPS4), likely functioning as ROS scavenger, too (Solís-Guzmán et al. 2017). Cell integrity/stability and cell wall composition both play important roles in DS response and tolerance mechanisms (Tenhaken 2014;Feng et al. 2016). Through the activation of ABA/auxin-mediated gene expression (RD22, WAT1) and further genes related to cell wall remodeling (AUG8, PDCB3), Q. robur most likely induced its adaptation to drought by strengthening the secondary cell wall of fibers (Yamaguchi-Shinozaki and Shinozaki 1993;Ranocha et al. 2013;Shi et al. 2014;Phillips and Ludidi 2017) and by closing plasmodesmata (Sager and Lee 2014). Furthermore, under DS, the photosynthetic apparatus is extremely stressed due to stomatal closure and subsequent Samples (control and drought stress) are clustered along the horizontal axis, differentially expressed genes along the vertical axis. Genes and samples are hierarchically clustered based on the Euclidean distance of gene abundance values and complete linkage clustering. Red and green color intensity indicate gene upregulation and downregulation, respectively. In (C) the expression patterns for each gene (grey) and the mean expression profile within that cluster (blue) is shown. CO 2 depletion (Früchtenicht et al. 2018a). In our dataset, DRIP2 was upregulated suggesting an inactivation of DREB2A-mediated gene expression in the long-term response to DS in Q. robur (Qin et al. 2008), which further hints toward a higher susceptibility to DS (Sakuma et al. 2006) in this species in comparison to Q. pubescens and Q. ilex. Further, catabolism in form of chlorophyll a breakdown (SGR) was induced (Sakuraba et al. 2014;Sato et al. 2018), supporting previous studies, where significantly lowered chlorophyll content during DS were reported (Früchtenicht et al. 2018a). But, repair mechanisms (DEGP1) of PSII damages already started (Sun et al. 2010;Kley et al. 2011;Zienkiewicz et al. 2013). Effects of DS on genes involved in carbohydrate metabolism were found in all three oak species. Sugars are not only the product of photosynthesis, and thus subject to drastic concentration changes under DS, but may also play a role as osmotica in oaks  (Épron and Dreyer 1996). In Q. robur, sugar synthases and transporters were activated (SPS4, MSSP2, GPT2) but also downregulated (GFT1, SWEET7), likely contributing to osmotic adjustment to maintain metabolic activity in source leaves (Büttner and Sauer 2000;Rautengarten et al. 2016;Solís-Guzmán et al. 2017). Regarding control mechanisms on the molecular level, many transcription factors (TFs) are described to be drought-inducible. Confirming earlier results on NAC TF activation after drought in Q. robur (Spieß et al. 2012), NAC domain-containing proteins were found to be upregulated also in our study after DS, likely leading to drought-induced senescence, fiber development and further abiotic stress responses (Podzimska-Sroka et al. 2015;Wu et al. 2016). Interestingly, both members of the DET1-HY5 pathway were activated, suggesting DS-induced leaf curling in Q. robur (cf Abrams 1990). A well-known family of proteins related to heat stress and other abiotic stresses are heat shock proteins (HSPs) (De Maio 1999), among them one group called the Late embryogenesis abundant (LEA) proteins. In this context, Spieß et al. (2012) already described the upregulation of a LEA by DS in Q. robur, which could be confirmed in our study, indicating that this species increased its tolerance to dehydration through chaperone activity (Hong-Bo et al. 2005).
Q. pubescens: Interestingly, in Q. pubescens much less genes (n = 79) were significantly differentially expressed after DS in comparison to Q. robur (415) and Q. ilex (222). The reason could be that the predawn water potential (see Table 1 in Kotrade et al. 2019) showed the less negative water potential (-0.47 CPD) and the least difference in water potential between CO and DS trees on the DOYs of sampling (287 and 288), indicating that the DS perceived on leaf cellular level was lower in Q. pubescens as compared to Q. robur and Q. ilex in the chosen, competitive experimental setup. Concerning its water potential, Q. pubescens only reached a level of mild DS. Nevertheless, it reacted to water withholding with the induction of jasmonic acid (JA) biosynthesis through LOX3.1 and AOS1, a process, that might play also an important role in the formation of antioxidants (Hause et al. 2002). Besides JA, also GABA plays an important role in DS tolerance (Shelp et al. 1999;Li et al. 2016Li et al. , 2017Mekonnen et al. 2016). The upregulation of the GABA permease BAT1 in Q. pubescens could hint toward the onset of catabolic processes by reducing the protein synthesis machinery, activating protein degradation, and subsequently export of amino acids from leaves (Dündar and Bush 2009). This can be supported by the GO analysis, that detected an enrichment of amino acid activation and especially (lysyl)-tRNA aminoacylation after DS treatment, resulting from significant downregulation of genes annotated for lysine-tRNA ligase, e.g., lysyl-tRNA synthetase (KRS) (Wu et al. 2007). The important role of ROS scavenging, sugar allocation and regulation of osmotic state in Q. pubescens after the drought treatment was further reflected in the differential expression of four genes related to these processes: GPT2, STP13, CMO, and bHLH92 (Yamada et al. 2011). Further, enhanced root growth could mediate DS tolerance in Q. pubescens by the upregulation of the F-box protein CPR30, a gene that may interact with multiple ASK proteins, which are known to be involved in regulating stomatal opening and root growth (Li et al. 2012). Like Q. robur, Q. pubescens activated heat shock proteins (ERD8/HSP80/HSP90.2, HSP83/HSP90.1) and heat stress transcription factors (HSFA4b) to maintain correct folding and protection of protein degradation (Clément et al. 2011;Pérez-Salamó et al. 2014). A possible adaptational mechanism to drought stress in Q. pubescens could be the strengthening of the cell wall through the incorporation of lignin by the activity of PAL, 4CL, and COMT (Moura et al. 2010;Pandey et al. 2010;Byeon et al. 2014;Le Gall et al. 2015;Wang et al. 2017), while the depletion of thylakoid membranes in chloroplasts through TERC may lead to a decrease in photosynthesis, most likely a starting point of catabolic processes (drought-induced senescence) (Batra et al. 2014;Schneider et al. 2014).
Q. ilex: Next to a significant enrichment of "defense response"-related GO terms (biotic and abiotic), Q. ilex showed additionally an enrichment in GO terms of "cytosine methylation" and "cellular respiration". Interestingly, in evergreen cork oak (Q. suber), an increase of cytosine methylation has been reported in heat-treated leaves (Correia et al. 2013), suggesting that Q.ilex may use the same mechanism of DNA structure reorganization to deal with DS. Quercus ilex reacted upon drought similarly as Q. pubescens by the activation of JA biosynthesis and signaling through AOS3 and CUL1 (Hause et al. 2002;Nagels Durand et al. 2016) and by increasing ABA-mediated drought tolerance (MLP43; Wang et al. 2016), most likely through ABA-responsive gene expression (bHLH30 with Arabidopsis homolog bHLH32; 94). As in the other two species, oxidative stress and the production of ROS is scavenged, in this case through the production of NADP(H) via the activity of NADK3 (Mittler 2002;Li et al. 2018) and through the production of flavonoids via CYP75B1 (Narusaka et al. 2004;Fini et al. 2011). It is likely that DS also impaired the flowering time through deactivation of COL14, which is supported by a study showing a 30% decrease in flower production of Q. ilex by drought   (Ogaya and Peñuelas 2007). One of the major responses to DS in Q. ilex involved genes related to cell wall remodeling conferring drought tolerance, like COMT, CSLE6, MIK2, ARAD1, WAK1, WAK5 and ExPA1 (Harholt et al. 2006;Moura et al. 2010;Kohorn and Kohorn 2012;Le Gall et al. 2015;Feng et al. 2016;Van der Does et al. 2017). Further, many LRR-RLKs, SRKs and CRKs responded to DS (e.g., LRK10), indicating their important regulatory function in this species (Marshall et al. 2012;Lim et al. 2015;Lu et al. 2016). Interestingly, the Pfam functional domain analysis showed that PPR (pentatricopeptide repeat) domains are more frequently annotated for the Q. ilex assembly than for the other two species (Figure 2). PPR proteins are involved in all aspects of organelle RNA metabolism (Yagi et al. 2014) and a recent study (Xing et al. 2018) showed that 154 PPRs in Populus were induced by biotic and abiotic treatment. However, roles of PPR genes in higher plant species remain largely unknown (Xing et al. 2018).

Differential response between deciduous and evergreen species
Through the comparison of orthologous genes, we found 37 that show a contrasting expressional behavior between the deciduous (Q. robur, Q. pubescens) and the evergreen species (Q. ilex) after DS perception. In deciduous oaks, GAT1 (GABA transporter 1) was activated but repressed in the evergreen species, indicating that the turnover and likely the accumulation of GABA is one of the mechanisms for adaptation to drought in the deciduous species. Also, JA (jasmonic acid) and ethylene seemed to be synthesized preferably in deciduous oaks through the activation of AOC, 4CLL7, and ERD1, but repressed in evergreens. These genes were already shown to be related to drought but could also be further associated with leaf senescence (He et al. 2002;Tang et al. 2005;Moura et al. 2010;Li et al. 2017), a process, that might have been already initiated in the deciduous species after DS. In addition, the small heat shock protein HSP15.4 and S1P were activated in DS deciduous but repressed in DS evergreens, indicating that deciduous oaks prevented DS damage by increased chaperone activity (Al-Whaibi 2011) and by stimulating the endoplasmic reticulum (ER) stress signaling pathway, which can occur when unfolded or misfolded proteins accumulate in the ER (Liu et al. 2007). Vacuolar protein transport seemed to be enhanced in deciduous species through the upregulation of FREE1 (Gao et al. 2015). Furthermore, out of the 37 orthologous genes, 12 showed a differential pattern only in the evergreen species between the CO and DS cohort, among them, orthologous genes that are involved in ROS scavenging (NADK3, CYP71B1 and CYP71B5) as well as bHLH30, that might confer ABA-mediated drought tolerance (Kim and Kim 2006), and genes that have been discussed before in the species-specific response pattern.
Response along the DS-tolerance gradient Comparing Q. robur with the more drought tolerant Q. pubescens and Q. ilex, we found a higher number of genes involved in ROS scavenging in the DS cohorts of the more drought tolerant species. Maintaining a certain ROS level involves several proteins that are either involved in ROS production or ROS scavenging. FPGS2 for example, a regulator of cellular folate homeostasis (antioxidant activity), is upregulated in Q. ilex but downregulated in Q. robur. Similarly, increased carotenoid and ascorbic acid (AsA) contents improve tolerance to abiotic stresses through their antioxidant activity (Akram et al. 2017), exemplified in the expression pattern of PDS (downregulated in Q. robur and upregulated in Q. ilex), a gene that is involved in carotenoid biosynthesis  (Hugueney et al. 1992). L-ascorbate oxidase (AO) upregulation in the Mediterranean species and downregulation in Q. robur possibly is connected to the latter's lower DS tolerance (Fotopoulos et al. 2008). ROS scavenging as a predominant DS tolerance process is also reflected in the upregulation of GST and scopoletin glucosyltransferase (TOGT1) in Q. pubescens and Q. ilex, which are involved in ROS detoxification by glutathione (Xu et al. 2015) and scopolin (Zandalinas et al. 2016;Döll et al. 2018), respectively. Besides the reduction of ROS, increased DS-tolerance is possibly also mediated by the maintenance of the mitochondrial respiration machinery (Atkin and Macherel 2009), as can be seen through the increased expression of TWINKLE in Q. pubescens and Q. ilex, a gene involved in mitochondrial DNA replication (Diray-Arce et al. 2013) and by the repression of mTERF5, a gene involved in the regulation of mitochondrial gene expression (Jõers et al. 2013). Further, these species activated the biosynthesis of lignin through the expression of CCR1 (Srivastava et al. 2015), required for lignification of interfascicular fibers and the xylem (Vanholme et al. 2010). The inactivation of PAO1, observed in Q. robur, together with PAO5, has already been proven to confer tolerance to drought in A. thaliana (Sagor et al. 2016).
Regarding the hormonal cross-talk in response to DS, most of the genes that were upregulated in Q. robur but downregulated in the tolerant species were linked to hormonal pathways, e.g., TCP9 transcription factor, involved in the control of leaf development via the jasmonate signaling pathway (Danisman et al. 2012), the indole-3acetic acid-induced protein ARG7 and the RNA-binding protein ZFWD2, indicating that in the severely stressed Q. robur further phytohormone-related DS responses were activated. The DS tolerant species also activated the heat shock protein machinery. ACX1, a gene that is described to be induced by DS was upregulated in Q. pubescens and Q. ilex. It interacts directly with HSP90, one of the key chaperones involved in modulating a multitude of cellular processes under both physiological and stress conditions (Chong et al. 2015). ERDJ, which is a drought inducible co-chaperone component of the HSP70 system (Ohta and Takaiwa 2014), was upregulated in Q. ilex. The overexpression of a homologous gene in Nicotiana, NtDnaJ1, enhanced drought tolerance possibly through regulating expression of stressresponsive genes (Xia et al. 2014). The combined upregulation of VQ22 and WRKY33 in DS Q. pubescens is connected to the regulation of downstream gene expression. This transcriptional regulation could confer drought (and heat) tolerance in deciduous oaks since VQ22-WRKY33 binds to the promoter region of stress-responsive genes Jing and Lin 2015). Another putative drought toleranceinducing gene that was downregulated in Q. robur only is NSN1, which leads to growth retardation and premature senescence when silenced (Jeon et al. 2015).
Taken together, antioxidant activity seems to be one of the major characteristics for enhanced DS tolerance in oaks. In the more tolerant species, many genes involved in the production of antioxidants (vitamins, carotenoids) and in ROS scavenging were upregulated compared to the less tolerant species. Also, the maintenance of the mitochondrial respiration machinery, the lignification of the water transport system and the suppression of drought-induced senescence might confer a higher tolerance to DS in Q. pubescens and Q. ilex compared to Q. robur.
Common response to drought on the genus level Despite the fact, that Old World oaks (Group Ilex, incl. Q. ilex) were separated from the New World oaks (Group Quercus, incl. Q. robur and Q. pubescens) more than 50 million years ago in the early Eocene (Hubert et al. 2014), highlighting a large difference on sequence level between the two groups, and an estimated divergence of the European white oaks 5 million years ago (Hubert et al. 2014;Leroy et al. 2017), we found 14 orthologous groups, that showed the same expression pattern across the three Quercus species, among them the gene CRK10 of the CRR-RLK family, which was downregulated in response to DS in all three species. The role of RLKs in responses to DS has been discussed already for plants in general (Marshall et al. 2012) and for Q. ilex in this study (see above). As suggested by the diversity and size of the RLK family (with more than 600 and 1,100 family members in A. thaliana and rice, respectively; Shiu et al. 2004), a genus-or even a speciesspecific concerted response pattern of its members to DS can be expected. Tolerance to DS on the genus level might additionally be mediated by the downregulation of BAM4, that is involved in starch breakdown (Kossmann and Lloyd 2000), and its activity is known to be induced by several abiotic stresses (Kaplan and Guy 2004). A betaamylase 1 mutant in A. thaliana exhibited improved tolerance to DS through reduced starch breakdown in guard cells which led to decreased stomatal opening (Prasch et al. 2015). In many plant species, sugar transport to the cytoplasm plays a role in increasing DS tolerance (Yamada and Osakabe 2017). In Q. petraea, total soluble carbohydrates (sucrose, glucose, fructose) increased upon water shortage, thereby contributing to drought-induced osmoregulation (Épron and Dreyer 1996). In our analyzed oak species, we found an upregulation of GPT, a gene that is involved in the transport of glucose 6-phosphate across plastid membranes, and G3Pp1, which is involved in the glycerol-3phosphate shuttle. G3P is a well-known inducer of systemic immunity in plants (Chanda et al. 2011) and our results indicate its involvement in drought stress perception. Apparently, oaks respond to DS by activating hexose allocation to the cytosol for osmotic adjustment. In addition, the upregulation of DSP1 indicates an increased sensitivity toward drought stress in all three species (Liu et al. 2015). Interestingly, through the downregulation of CNGC1 after DS perception, one could speculate, that stress-induced calcium uptake into oaks is hindered and as an effect, root growth might be enhanced (Ma et al. 2006;Jha et al. 2016;Wilkins et al. 2016) likely with the attempt to reach a waterbearing soil level.

CONCLUSION
A plant's answer to DS is manifold and depends on many environmental factors as well as on species-specific or even individual-specific mechanisms and preconditions. The analysis of orthologous genes that reacted the same in response to DS on the genus level revealed the induction of commonly known drought-responsive transcription factors and kinases, as well as genes involved in cell wall remodeling and lignification. Our data further indicate, that DS tolerance on the genus level could be mediated by a reduced starch breakdown, and further by an elevated sugar and glycerol-3-phosphate transport. At the species level, Q. robur reacted to the severe stress reflected in the very low water potential by activating ABA-mediated gene expression, the ROS scavenging machinery, and adjusting the cellular osmotic potential to maintain metabolic activity in the cell. However, catabolic reactions (chlorophyll breakdown) started, possibly indicating the onset of drought-induced senescence. Despite the late sampling date, the catabolic reactions are probably not caused by daylength-induced senescence processes since they were not detected in the CO group and the sampling date was chosen before major chlorophyll breakdown starts in Q. robur according to a previous study (Holland et al. 2014). Although Q. pubescens experienced a much lower level of DS on the cellular level compared to Q. robur (see Früchtenicht et al. 2018b), similar molecular mechanisms were activated, such as ROS scavenging for cell detoxification, and genes involved in root growth and the regulation of stomatal opening, but also catabolic processes leading to a decrease in photosynthesis. Differently than Q. robur, Q. pubescens activated the heat-shock protein/chaperone network as well as lignification. In the most tolerant species, Q. ilex, also the activation of the ABA-mediated response to DS (as Q. robur) was detected, but in addition predominantly genes related to ROS scavenging and cell wall remodeling were regulated, which might help allowing to maintain growth and vigor after DS perception.
Taken together, a higher antioxidant activity seems to be one of the major characteristics for enhanced DS tolerance in oaks. But also the maintenance of the mitochondrial respiration machinery, the lignification of the water transport system and the suppression of droughtinduced senescence might confer a higher tolerance to DS; a valuable knowledgebase of putative candidate genes, that could be integrated in breeding strategies for drought tolerant oaks.

ACKNOWLEDGMENTS
This work was funded by the research-funding program "LOEWE -Landes-Offensive zur Entwicklung Wissenschaftlich-Ökonomischer Exzellenz" of the Hessian Ministry for Science and the Arts.