Waterlogging affects plant morphology and the expression of key genes in tef (Eragrostis tef)

Abstract Tef [Eragrostis tef (Zucc.) Trotter], an allotetraploid cereal that is a staple food to over 60 million people in the Horn of Africa, has a high nutritional content and is resistant to many biotic and abiotic stresses such as waterlogging and drought. Three tef genotypes, Alba, Tsedey, and Quncho, were subjected to waterlogging conditions and their growth, physiology, and change in transcript expression were measured with the goal of identifying targets for breeding cultivars with improved waterlogging tolerance. Root and shoot growth and dry weight were observed over 22 days. Stomatal conductance and chlorophyll and carotenoid contents were quantified. Microscopy was used to monitor changes in the stem cross sections. Illumina RNA sequencing was used to obtain the expression profiles of tef under flooding and control conditions and was verified using qPCR. Results indicated differences in growth between the three genotypes. Waterlogged Tsedey plants grew higher and had more root biomass than normally watered Tsedey plants. Quncho and Alba genotypes were more susceptible to the excess moisture stress. The effects of these changes were observed on the plant physiology. Among the three tested tef genotypes, Tsedey formed more aerenchyma than Alba and had accelerated growth under waterlogging. Tsedey and Quncho had constitutive aerenchyma. Genes affecting carbohydrate metabolism, cell growth, response to reactive oxygen species, transport, signaling, and stress responses were found to change under excess moisture stress. In general, these results show the presence of substantial anatomical and physiological differences among tef genotypes when waterlogged during the early growth stage.


| INTRODUCTION
Climate change is one of the most challenging problems facing global food security. Along with changes in temperature, the frequency and severity of both extreme droughts and extreme precipitation events are expected to increase (Hartmann, Tank, & Rusticucci, 2013) causing billions of dollars of crop losses (Bailey-Serres, Lee, & Brinton, 2012). Understanding of the genetic basis of the plant response to abiotic constraints is critical for the development of cultivars that are resilient to stress and to decrease the gap between yield potential and actual yield under stressed conditions (Setter & Waters, 2003).
Tef [Eragrostis tef (Zucc.) Trotter], an allotetraploid cereal crop, is the major crop in the Horn of Africa, especially in Ethiopia where it is annually cultivated on over three million hectares of land and is staple food to over 60 million people (CSA 2015). Of the approximately 350 members of the Eragrostis genus, tef is the only one cultivated for use as a human food although several species are used as livestock fodder or forage grasses (Ketema, 1993).
Tef grows in wide agroecological conditions ranging from semiarid areas with low rainfall to areas with high rainfall. It is also a dominant crop in poorly drained soils commonly known as vertisols which cover about 10% of the total area of Ethiopia, much of which are in high-rainfall areas (Mekonen, Tesfaye, & Bayu, 2013).
The combination of poorly drained soil and high rainfall is conducive to waterlogging and is responsible for both low yields and underutilization of this land for farming (Asamenew, Beyene, Negatu, & Ayele, 1993).
Waterlogging is a soil condition in which excess water causes inadequate gas exchange between the soil and the atmosphere. The effects on the plant are manifold. Oxygen diffuses approximately 10,000 times slower in water than in air, and the flow of oxygen into waterlogged soil is around 320,000 times less (Watanabe, Nishiuchi, Kulichikhin, & Nakazono, 2013). Therefore, the amount of oxygen available to the roots decreases (Jackson & Colmer, 2005;Lee et al., 2011), and photosynthesis and respiration are limited. Waterlogging can decrease a cell's resistance to pathogens (Hsu et al., 2013). Additionally, phytotoxic compounds such as sulfides and the reduced forms of minerals (e.g., Mn 2+ and Fe 2+ ) accumulate in waterlogged soil (Laanbroek, 1990;Nishiuchi, Yamauchi, Takahashi, Kotula, & Nakazono, 2012), creating oxidative stress (Fukao, Yeung, & Bailey-Serres, 2011).
Another morphological adaptation to flooding is the formation of barriers to radial oxygen loss (ROL) in the basal part of the root to minimize oxygen loss to the environment and keep the oxygen moving toward the root apex (Abiko et al., 2012;Colmer, Cox, & Voesenek, 2006;Nishiuchi et al., 2012;Watanabe et al., 2013). These barriers may also impede the movement of soil toxins and gases into the root (Armstrong, 1979;Colmer, 2003;Watanabe et al., 2013).
Both suberin (Watanabe et al., 2013) and lignin (Kotula, Ranathunge, Schreiber, & Steudle, 2009) may be components of this apoplastic barrier. A hybrid between wheat (Triticum aestevum) and a wild waterlogging-tolerant barley with constitutive aerenchyma and an inducible ROL barrier was found to be more waterlogging tolerant than the parental wheat line (Malik, Islam, & Colmer, 2011). Similarly, aerenchyma formation combined with an inducible barrier to ROL contributed to the waterlogging tolerance of a newly discovered teosinte (Zea nicaraguensis) compared to cultivated maize (Zea mays) (Abiko et al., 2012).
The production of reactive oxygen species (ROS) is a consequence of aerobic metabolism and may be a part of the signaling that stimulates the growth of adventitious roots (Steffens & Rasmussen, 2016). Detoxification of ROS is mediated by both enzymatic (superoxide dismutase, ascorbate peroxidase, etc.) and non-enzymatic (carotenoids, ascorbic acid, glutathione) antioxidants (Lekshmy, Jha, & Sairam, 2015). In crop plants, upregulation of antioxidants has been found to be correlated with abiotic stress tolerance (Rivera-Contreras et al., 2016;Ushimaru et al., 1997;Zhang et al., 2015) and has even been suggested to be the basis for waterlogging tolerance (Thirunavukkarasu et al., 2013).
Two strategies employed by rice to combat flood are (i) escape, rapid growth of the shoots that allows the plant to reach the surface as quickly as possible, and (ii) quiescence, entering a state of inactivity until the flooding conditions have passed (Nishiuchi et al., 2012).
Both strategies are mediated by ethylene. Escape is a useful response in the case of deep and long-lasting flooding, while quiescence is employed for shallower floods of short duration. Submergence-tolerant rice cultivars employing the quiescence strategy restrict the consumption of carbohydrates, retain chlorophyll to maintain limited photosynthesis, and express an ethylene-responsive factor (ERF) gene, Submergence 1A (SUB1A), which confers submergence tolerance by dampening ethylene production and GA responsiveness (Fukao, Xu, Ronald, & Bailey-Serres, 2006;Xu et al., 2006).
The physiological response of plants subjected to waterlogging stress involves a switch from aerobic respiration to fermentative metabolism, and consequently, genes that are involved in starch breakdown such as the glycolytic and fermentative pathways are affected (Parent, Capelli, Berger, Cr evecoeur, & Dat, 2008). Tolerant plants can also photosynthesize and respire while flooded (Caudle & Maricle, 2012). Thus, physiological measurements such as stomatal conductance and leaf chlorophyll content are valuable probes of the plant response.
The Tef Improvement Project was established to increase tef productivity using modern plant breeding methods and has undertaken genome and transcriptome sequencing for this purpose . Here, we report a study on the response to waterlogging of three tef genotypes-a landrace, Alba, and two improved varieties Tsedey and Quncho. Quncho, a cross between Dukem (maternal) and Magna (paternal), is a popular variety as it combines the high yield of Dukem with the high seed quality of Magna. Tsedey, an improved variety on the market since 1984, can grow in a wide range of soil, water, and air conditions. Here, growth and biomass have been measured for tef subjected to early (at 4 days) waterlogging. In addition, we combined an expression study using RNA-Seq and qPCR with physiological studies and microscopy to understand the response of tef to waterlogging at the tillering stage. The results show that there are substantial differences in growth, waterlogging tolerance, anatomy, and physiology between the three genotypes studied in the early and tillering stages of waterlogging, diversity that is necessary for breeding the trait into new germplasm and identification of the responsible genes using association mapping. The most tolerant genotype, Tsedey, has greater growth of both the shoot and the root, more constitutive aerenchyma, and aerenchyma production. Genes affecting carbohydrate metabolism, cell growth, response to reactive oxygen species, transport, signaling, and stress responses were found to change under stress in the Tsedey improved variety, consistent with the physiological and anatomical changes observed.

| MATERIALS AND METHODS
Experiments designed to monitor root and shoot growth, aerenchyma size, physiology, and gene expression are outlined in Table 1.

| Tef genotypes used
Three Eragrostis tef genotypes used in these experiments were as follows: (i) Tsedey (DZ-Cr-37), an improved variety whose genome and transcriptome have been sequenced and which is thought to be waterlogging resistant, (ii) Alba, a landrace obtained from the US Department of Agriculture (USDA), and (iii) Quncho (DZ-Cr-387 RIL-355), a popular high-yielding and white-seeded variety. Tsedey and Quncho were obtained from the Ethiopian Institute of Agricultural Research (EIAR).
2.2 | Growing conditions for Experiment 1 (Growth 1): plant height Three seeds of tef genotypes (Alba, Tsedey, and Quncho) were sown in pots and grown under long-day conditions (16-h light at 22°C and 65% relative humidity; 8-h dark at 18°C and 65% relative humidity).
For the waterlogging treatment, the pots were put in a plastic tray in which the water level was maintained at 1 cm below the surface of the soil. The control plants were put in an identical tray and watered from below every 3 days followed by drainage of the tray.
The waterlogging treatment was applied to 18 (9 controls, 9 waterlogged) 4-day-old plants. Day 0 marked the first day of treatment.
Plant height (from the base of the root to the end of the longest leaf) was recorded for the next 24 days, namely at 0, 1, 3, 4, 7, 8, 14, 17, and 22 days after waterlogging. 2.3 | Growing conditions for Experiment 2 (Growth 2): Early waterlogging time series of plant height, root length, and root/shoot dry weight Experiment 2 is a repetition of Experiment 1 using 66 seedlings of Tsedey and Quncho sown in pots and grown under long-day conditions as indicated above. For the waterlogging treatment, the pots were put in a plastic tray in which the water level was maintained at 1 cm below the surface of the soil. The control plants were put in an identical tray and watered from below every 3 days followed by drainage of the tray. Four days after germination, treatment started and the plants were grown for the next 24 days under control or waterlogging conditions. At 2,4,7,9,11,16, and 24 days after treatment, four plants were harvested, one root was reserved for cross-section, the remaining roots washed under tap water, the separated roots and shoots were dried at 60°C overnight, and the root and shoot dry weight and length were measured.

| Growing conditions for Experiment 3: tef root visualization using rhizotrons
In experiment 3, seeds of Alba, Quncho, and Tsedey were grown in Plexiglass rhizotrons with dimensions of 20 cm wide, 30 cm high, and 1.5 cm deep. After germination, all plantlets except the one with the most visible root were culled. Water was available from the bottom of the rhizotron or could leak in at the seals on the sides. For the waterlogging treatment, the water level was maintained at 1 cm below the CANNAROZZI ET AL. were soaked in 2.5% bleach solution for 3-4 min and then repeatedly rinsed while agitating. They were then planted on wet paper towels in Petri dishes and then transferred to the rhizotrons after 4 days. Treatment was started one week after transplantation into the rhizotron.
The waterlogging conditions were the same as for Experiment 3.
2.6 | Growing conditions for Experiment 5 (Physiology) and Experiment 8 (RNA-Seq) Seedlings of the improved tef variety Tsedey were grown under long-day conditions (16 h of light at 24°C and 8 h of dark at 18°C, 65% relative humidity) in pots. The seedlings were grown for 19 days and then exposed for 9 days to two conditions (waterlogging and normal watering). Waterlogging was achieved by maintaining the water level at 1 cm below the soil surface. After 9 days of treatment, the stomatal conductance, chlorophyll a and b, and carotenoid content were measured, the plants were harvested, and RNA was extracted from leaf tissue (described below).
2.7 | Growing conditions for cross sections at the tillering stage Experiment 6 and Experiment 7 (qPCR) Alba, Tsedey and Quncho varieties were sown in 7.5-cm-diameter pots and grown under long-day conditions in the growth room as indicated above. Waterlogging treatment (water level maintained at 1 cm below the soil surface) was applied for 9 days to 19-day-old plants. After 9 days of growth under control or treatment conditions, the plants were harvested, the wet weight of the root and shoot was measured, the number of adventitious roots per plant was counted, plant material was collected for qPCR, and one or two adventitious roots were taken for cross sections. This procedure was performed in duplicate, both times with 10 or 11 plants (observation 1 and observation 2). For the cross sections, sometimes more than one cross section from the same plant and root location were taken  Figures S7, S8, S9, S10, S11, and S12 Tables 3, 4, 5, S4, and S5 which were considered as technical replicates and hence were not counted as independent observations. At the middle of the root, it was often difficult to get an unblemished cross section. The filenames for the cross sections are named for the plant (T: for Tsedey, Q: for Quncho, or A: for Alba), the number of the plants (1-17), the location on the root (oben (root base), mitte (middle), or unten (tip)), the replicate (1-5), and the magnification.

| Physiological measurements
Stomatal conductance of the flag leaf for the adaxial side was determined using an AP4 diffusion porometer (Delta T, Cambridge Life Sciences, Cambridge, UK) at the end of the water stress period.
Chlorophyll a and b, as well as carotenoids which comprise carotenes and xanthophylls, were extracted using 95% ethanol and measured with UV-Vis spectroscopy as previously described (Lichtenthaler, 1987). The amount of these pigments was normalized by fresh weight. For the physiological measurements, 10 measurements from 10 different plants were used.

| Statistical analysis for the physiology and growth measurements
Data are presented as means AE standard deviation. Statistical tests were made using R version 3.0.1 using the built-in function wilcox.test. First, an ANOVA test was used to determine whether significant differences existed. If so, outliers were removed using a modified Thompson tau test. Nonparametric tests (Mann-Whitney U-test or Kruskal-Wallis H tests) with a p-value of ≤.05 were used to determine statistical significance between treatment means.
2.10 | RNA extraction for RNA-Seq RNA was extracted from samples of leaf tissue using the TRIzol kit (ThermoFisher Scientific) according to the supplier's protocol. The quality and quantity of RNA were quantified using ND-1000 spectrophotometer for which the average 260/280 ratio was 2.0, indicating good-quality RNA.

| Transcriptome library construction and sequencing for RNA-Seq
The RNA extracted from plants grown under waterlogging and normal watering conditions was sent to Fasteris (Geneva, Switzerland, www.fasteris.com) for further quality testing and sequencing using Illumina HiSeq2000 with the goal of analysis for differential expression.
Two biologic replicates were made leading to two libraries for control (GNY1 and GNY10) and two from the waterlogged plants (GNY3 and GNY12). The GNY10 and GNY12 libraries were prepared using the AccuPrime polymerase (Invitrogen, Carlsbad, CA) and following the protocol for high GC content. The GNY1a and GNY3a samples were prepared with the TruSeq SBS version 5 kit and the data analysis pipeline consisting of HiSeq Control Software version 1.1.37.8, RTA 1.7.48, and CASAVA 1.7. GNY1b, GNY3b, GNY10, and GNY12 also used the TruSeq SBS version 5 kit and flow cell version 3 with software: HiSeq Control Software version 1.4.8, RTA 1.12.4.2, and CASAVA 1.8.2. The six cDNA libraries were sequenced to generate a total of 205 million single-end reads. Before assembly, the reads were trimmed such that the Phred quality scores were above 28. In addition, all primer and adaptor sequences detected by FastQC were removed.

| Quantification of gene expression levels and differential expression experiments
The reads from each condition were mapped onto the 14,057 scaffolds of size 1000 bp or greater obtained from the recently sequenced tef genome (Cannarozzi et al., 2014) using STAR 2.3.0 with the default parameters. These were converted to BAM format with SAMtools (Li et al., 2009).
A count table was obtained using the HTSeq-count program with options stranded=no, type=gene, and attribute=ID (Anders, Pyl, & Huber, 2015) and using the Maker gene predictions provided by the Tef Genome Project (Et_genome_1.0.fasta.gz) (Cannarozzi et al., 2014). HTSeq tabulates the percentage of uniquely mapping reads, reads that map to no feature in the predicted transcriptome, ambiguous reads (map to more than one gene simultaneously), and reads that do not map uniquely to one location. Only uniquely mapping reads were used for counting. The newCountDataSetFromHTSeq-Count function of the DESeq package was used to generate the count table used as input into DESeq, a Bioconductor package that estimates the variance-mean dependence of high-throughput sequence data and tests for differential expression using the negative binomial model (Anders & Huber, 2010).

| Background set
The transcriptome used in these experiments was from the file Et_all.maker.transcripts.shortids.fasta.gz provided by the Tef Improvement Project. This file contains genes predicted in the genome by Maker and described in (Cannarozzi et al., 2014). To define a background set for the enrichment analysis, the reads were mapped onto the set of 42,052 genes predicted from the tef genomic scaffolds, creating a background set consisting of all 34,761 genes detected in the experiment. The background set was annotated using Blast2GO with the default parameters (E-value 1.0 e-03, blastp, non-redundant database at NCBI). Differentially regulated genes were defined as those found by DESeq to have an adjusted pvalue of less than 0.05 and a fold change of 2 or more. Assignment and clustering of GO terms produced biologic process, molecular function, and cellular component of these differentially regulated terms. Enrichment analysis using Fisher's exact test at a = 0.05 was used to determine genes differentially expressed.
Sequences were also placed into MapMan functional categories based on the sequence similarity using the Web interface to Mercator (http://mapman.gabipd.org/web/guest/mercator) (Lohse et al., 2014) using annotated sequences from the following: the Arabidopsis Information Resource (TAIR) (Berardini et al., 2015), SwissProt/ UniProt plant proteins (Bateman et al., 2017), TIGR5 rice proteins (Kawahara et al., 2013) and the COG database (Tatusov et al., 2003) and BLAST cutoff 80. Multiple bin assignments were allowed. Enrichment analyses of various pathways using a Wilcoxon rank-sum test with Benjamini-Hochberg corrections for multiple testing were conducted using a local installation of the MapMan software.
2.14 | Verification of expression levels using qPCR Plant material was collected as described above and was then ground in liquid nitrogen. Total RNA was extracted using the Total RNA Isolation System (Promega, USA) and treated with the DNAfree kit from Life Technologies, USA. RNA was quantified using a NanoDrop ND-1000 spectrophotometer (Thermo Scientific, USA).
For calculating the relative gene expression, the 2-DDCq method was used (Livak & Schmittgen, 2001). CYP and PP2A were used as reference genes. Each biologic replicate was sampled three times, and the measured Cq values were averaged.

| Preparation and microscopy of cross sections of adventitious roots
The adventitious roots were cut into three sections of equal length, referred as "base" (closest to the shoot), "middle," and "tip" (closest to the root tip or apex). These were embedded in 5% agarose and sliced into 100-to 200-lm slices using a Vibratome. No staining or dyes were used. The cross sections were visualized in a Zeiss Axioskop 2 microscope with an AxioCam color (412-312) camera using 59 to 209 magnification. The microscope software was AxioVision Release 4.8.2 SP2 (06-2012), and ImageJ (Duarte et al., 2016) was also used in the image processing. The expansin sequences from (Li et al., 2003) were downloaded from the National Center for Biotechnology Information (NCBI) Web site (Coordinators, 2017). The accession numbers are as follows:

| Quantification of aerenchyma
AF247163.  At each time point, shoot dry weight, root dry weight, root length and shoot length, and the number of leaves were measured ( Figure 1 for all days, Table 2 for day 24, and all raw data in Tables S1 (Quncho) and S2 (Tsedey)). As in Experiment 1, Tsedey grew more robustly than Quncho in the waterlogged environment.

| Shoot growth
In the Tsedey cultivar, the shoot length, measured from the base The summary of growth statistics at 24 days after waterlogging is shown in

| Root growth
The root structure of tef plants contains many fine root hairs. As the plants were grown in soil, it was necessary to wash the root ball before drying and measurement. To remove all soil without damaging, the fine root structure was difficult and added uncertainty to the measurements of root weight. In terms of root length, the roots of Tsedey subjected to early waterlogging were significantly longer at 9 and 11 days after stress, while the roots of Quncho subjected to early waterlogging were significantly shorter at 11 days after treatment ( Figure 1). Similarly, Quncho roots had significantly less dry weight at 24 days after waterlogging where they were reduced by 56% compared to that of Tsedey which had increased in weight by 48% ( For comparison, the experiment was repeated with Alba, Tsedey, wheat, and maize (Experiment 4). In Tsedey, the control roots were longer, but the waterlogged plants had thicker roots and more adventitious roots than normally watered plants (Fig. S2) In wheat, after one day of waterlogging, the root systems of the control and waterlogging plants were similar (Fig. S3). However, after 3 days of waterlogging, the control plants had distinctly more roots than the waterlogged plants, and this phenomenon became more pronounced with time. Both control and waterlogged plants grew adventitious roots, but they were more pronounced in the control CANNAROZZI ET AL. 3.4 | Physiological response of tef to waterlogging at the tillering stage Nineteen-day-old seedlings of improved tef variety Tsedey (DZ-Cr-37) were exposed to either normal watering or waterlogging for 9 days. Physiological measurements of stomatal conductance, as well as carotenoid, chlorophyll a, and chlorophyll b, were made before the plants were harvested and the tissue sent for sequencing (Experiment 5). Only stomatal conductance was altered significantly under waterlogging conditions (Figure 2). The difference was large, however, as stomatal conductance was roughly three times higher under waterlogging conditions than controls.

| Aerenchyma formation in roots under
waterlogging at the tillering stage  Four-day-old plants were grown in soil either with normal watering or with water maintained at 1 cm below the soil surface. All values are means (n = 3 or 4) AE SD. Significance differences at p ≤ .05 are denoted with a * (Mann-Whitney). In addition, the number of adventitious roots was counted for replicates of this experiment. For all three genotypes, the number of adventitious roots increased with waterlogging. The variance was high, however, from 109% in Alba to 336% in Tsedey (Table S3).
The roots of one replicate from each day of Experiment 2 (waterlogging of 4-day-old plants for 22 days) were used to visualize the formation of aerenchyma of Quncho and Tsedey with samples taken at 2, 4, 7, 9, 16, and 24 days after waterlogging (Fig. S6). Quncho was the earliest to show aerenchyma formation which appeared only 4 days after waterlogging, while in Tsedey they were first observed 7 days after stress.
3.6 | RNA-Seq study of differential expression at the tillering stage  sequence reads were generated with Illumina sequencing (Table 3).
These reads were filtered for adaptor and vector sequences and trimmed such that the Phred score was greater than 30 across the sequence.
3.6.1 | Annotation of the background set using

Blast2GOand Mercator
Although the Tef Genome Project provided annotations, only a small percentage of sequences had gene ontology (GO) annotations, which are helpful for enrichment analysis. Therefore, the background genes were reannotated with descriptions and GO terms, InterPro protein signatures (Finn et al., 2017), and KEGG pathways (Kanehisa, Furu- GO terms were assigned via Blast2GO using the GOSlim plant annotations resulting in 27,338 proteins annotated with GO terms in the background set (Fig. S8). For the biologic process, the GO terms "metabolic process," "cellular process," "single-organism process," "biologic regulation," and "response to stimulus" were the five most represented categories. The molecular function category was dominated by "catalytic activity" and "binding" with "transporter activity" also being well represented. The cellular components most often found were "cell," "organelle," and "membrane".
In addition, the background set was annotated with pathways using the KEGG annotation system in Blast2GO. Prediction of the pathways containing the background proteins resulted in 127 pathways containing 8,768 proteins (25.2% of the genes detected in the experiment were classified). The pathways including the most transcripts were "purine metabolism," "starch and sucrose metabolism," "phenylpropanoid biosynthesis," "phenylalanine metabolism," "pyrimidine metabolism," and "glycolysis and glucogenesis." 3.6.2 | Annotation of the background set using

Mercator
Mercator, a Web application that assigns DNA or protein sequences to one of 35 MapMan bins (Lohse et al., 2014), was also used for annotation of the background set. MapMan categorization provides an overview of metabolism and cellular process and is tailored for functional annotation of plants (Thimm et al., 2004). Many of these categories deal with metabolic pathways and enzyme functions, providing an overview of the metabolic networks involved in the response. The results of the mapping of the background set revealed the categories: transport (11.7%), cell wall (10.0%), protein (8.3%), and cell (8.3%) to be the most represented functional categories (Fig. S9).

| Differential expression
Differential expression analysis was performed to determine the change in transcript expression from the waterlogging treatment to the control (normal watering) (Fig. S7). First, the reads were mapped to the genome using STAR aligner (Dobin et al., 2013). A count table was generated for each predicted gene from the tef genome for each sequencing dataset using HTSeq (Anders et al., 2015), which was given the genome and the predicted locations of the genes provided by the tef genome sequencing (Cannarozzi et al., 2014). Only the roughly 70% of the reads that map uniquely to one location in the genome were used for counting (Table 3)  and cyclophilin/peptidyl-prolyl Isomerase (CYP), have been identified as the most stable for use as reference genes in qPCR of abiotic stress response genes in sorghum (Reddy et al., 2016). All measurements were taken using both reference genes. Comparisons between the RNA-Seq expression measurements in Tsedey and the qPCR expression in Tsedey, Alba, and Quncho are shown in Figure 5 and Table S6. All four genes were regulated in the same direction with similar magnitudes. The use of the two reference genes resulted in similar behavior except in the Alba genotype for the aquaporin gene and for granule-bound starch synthase ( Figure 5).

| DISCUSSION
Eight experiments were carried out to probe the morphology of roots, growth of shoots and roots, and gene expression. The experiments were carried out either on 4-day-old seedlings with parameter measurement every 2-4 days for up to 24 days (early waterlogging) or on plants at the tillering stage (19-day-old plants which were then waterlogged for 9 days).

| Growth
The Tsedey genotype grew faster under early waterlogging conditions, while the growth of Quncho was suppressed (Figure 1 and S1). Measurements of Tsedey seedling growth over time showed that  Table 2). The largest difference between Tsedey and Quncho tef genotypes was seen in the roots, where Tsedey consistently grew a larger root mass when waterlogged. The root length decreased for both genotypes, but the number of adventitious roots increased for both, resulting in a denser root mass. The root dry weight was significantly increased in Tsedey by 148%, while in Quncho it was reduced to 48%. A part of the difference may be attributed to the fact that thicker roots are less likely to be damaged or removed when washing the root ball. In a similar study, the dry weight of the flooding-tolerant Zea nicaraguensis decreased by 54%, while in the flooding-susceptible Zea mays, the dry weight was reduced by 60% (Abiko et al., 2012). A global increase in biomass under waterlogging conditions is often associated with waterlogging adaptation (Naidoo & Naidoo, 2015) as is the ability to produce adventitious roots (Steffens, Wang, & Sauter, 2006). The number of leaves was not affected, but the leaves of Quncho started browning after 16 days, a sign of waterlogging stress.

| Aerenchyma formation at the tillering stage
Cross sections of three tef genotypes (Alba, Quncho, and Tsedey) Zea mays roots formed a negligible amount of aerenchyma in drained soil which increased to 15% by area under waterlogging. The roots of Z. nicaraguensis, a waterlogging-tolerant species, had 22% aerenchyma by area, which increased to 29% under waterlogging (Abiko et al., 2012). Aerenchyma are induced by waterlogging but are found constitutively in wetland or waterlogging-adapted species (Abiko et al., 2012;Justin & Armstrong, 1987). A QTL for aerenchyma formation has been identified in barley, and the newly developed markers explain 44% of the phenotypic variance (Zhang et al., 2016), while a recently discovered allele in wild barley is said to account for 76.8% of the phenotypic variance (Zhang et al., 2017).

| Physiological response at the tillering stage
Measurements of the seedling growth over time show that the largest changes in growth occurred directly after the application of the stress, while in this experiment the tissue was collected 9 days after the application of the stress. For the physiological measurements, no significant difference was found in chlorophyll a, chlorophyll b, or carotenoid which may indicate that the plants were not stressed to a large extent or had already adapted to the stress F I G U R E 5 Validation of RNA-Seq expression measurements by qPCR in three tef genotypes. Aquaporin (a), bifunctional nuclease 2-like gene (b), cinnamoyl reductase-like gene (c), and granule-bound starch synthase (d) were regulated under waterlogging conditions using RNA-Seq. Expression measurements for these four transcripts were verified with qPCR, using either CYP or PP2A as reference genes. Three replicates were used for each genotype, condition, and transcript ( Figure 2). Stomatal conductance, however, increased threefold under flooding. Caudle and Maricle suggested that the characteristics of plants that correlated best with tolerance to flooding were photosynthesis rate, respiration upon flooding, and the ability to avoid oxygen shortage (Caudle & Maricle, 2012). Leaf-level stomatal conductance was also suggested to be an indicator of flooding tolerance in comparison with the responses of the flooding-sensitive Johnson grass (Sorghum halepense) and the flooding-tolerant common reed (Phragmites australis). In the sensitive Sorghum halepense, transpiration decreased in response to flooding, while in the tolerant Phragmites australis, the opposite occurred (Waring & Maricle, 2013). A study in tolerant versus sensitive wheat concluded that a significant reduction in grain and straw yield happened only if the waterlogging was prolonged for more than 20 days (Arduini, Orlandi, Pampana, & Masoni, 2016 (Zou et al., 2013), while in Jatropha 1968 mRNA transcripts had a significant change in abundance (Juntawong et al., 2014). Similarly, in cypress tree (Taxodium "Zhongshansa 406"), 2090 differentially expressed genes were found in roots, while only 394 were found in shoots (with 174 shared between the two groups). The RNA used in the current study was extracted from seedlings (not including the roots) 9 days after waterlogging at the tillering stage, which may explain the relatively small number of genes detected.
The differentially regulated genes are ordered by their MapMan annotation bins in Tables 4 and 5 and are discussed in this order.  (Jordan & Chollet, 1983) and it also acts as a chaperone during stress (Rokka, Zhang, & Aro, 2001). In tomato, RuBisCO activase has been found to decrease under flooding stress and is susceptible to degradation from reactive oxygen species (Ahsan et al., 2007).
Scaffold2486 contains two tandem genes annotated by Blast2GO as expansin genes, in which only one was identified as an expansin by MapMan. With only two exons, the ambiguous gene looks incomplete, supporting the annotation of MapMan which annotates it as a gene of unknown function (Table 4). Under flooding conditions, rapid elongation of internodes and petioles is often observed as plants use an escape strategy to try to reach an environment with a better oxygen supply. Expansins and XTHs are involved in breaking down the cell walls to allow rapid expansion and growth (Li et al., 2002;Rose, Braam, Fry, & Nishitani, 2002;Sampedro & Cosgrove, 2005;Tsuchiya, Satoh, & Iwai, 2015) and have been shown to contribute to rapid coleoptile growth in rice (Miro & Ismail, 2013). One b-expansin has been shown to induce the extension of maize coleoptiles (Cosgrove, Bedinger, & Durachko, 1997). Expansins have also been implicated in the formation of adventitious roots in the loblolly pine (Pinus taeda) (Hutchison, Singer, McInnis, Diaz-Salaz, & Greenwood, 1999) and deepwater rice (Cho & Kende, 1997). The action of XTHs involves breaking linkages in the xyloglucan-cellulose network and then reforming them in a new position, allowing expansion without impairing overall wall integrity. XTH transcripts have been shown to increase in maize roots after 12 h of hypoxia due to flooding (Saab & Sachs, 1996).
Lysigenous aerenchyma result from the death of certain cells in the root cortex and are found in crops such as wheat, rice, barley, and maize (Shiono, Takahashi, Colmer, & Nakazono, 2008). Ethylene accumulation has been implicated as the primary signal in the formation of lysigenous aerenchyma (Shiono et al., 2008). This triggers a signal transduction pathway involving phosphoinositides and Ca 2+ (Drew, He, & Morgan, 2000). Although no genes in this pathway have been detected to have differential expression under waterlogging, an XTH is upregulated (Table 4). An XTH has also been found to be associated with aerenchyma formation (Saab & Sachs, 1996).
In maize, a mechanism for cortical cell-specific programmed cell death has been proposed that includes the generation of ROS in combination with suppression of a metallothionein ROS scavenger.
Addition of the regulated tef expansin sequences to the phylogenetic tree of Li et al. (2003) shows that the regulated tef sequence is of type b1, the most closely related to the Li sequences EXPb1.10 which forms a clade with EXPb1.7, EXPb1.8, EXPb1.9, and EXPb1.11 ( Fig. S11). Expression of b-expansins has been correlated with internodal elongation in deepwater rice (Lee & Kende, 2001).
Peroxidases also affect cell wall extensibility. Correlations between the decrease in cell wall-bound peroxidase activity and coleoptile elongation have been observed (Ismail, Ella, Vergara, & Mackill, 2009;Lee & Lin, 1995). Peroxidase activity has been found to be significantly higher in flooding-sensitive genotypes of rice.
Indeed, in the current study, a class III peroxidase, Et_s190.0.32, is also downregulated.

| Changes in carbohydrate metabolism
The MapMan category primary metabolism includes the metabolism of amino acids, nitrogen metabolism, sugar and its derivatives, and energy. Plant roots subjected to flood stress switch from aerobic respiration to fermentative metabolism, requiring soluble sugar, which is produced from carbohydrates stored as starches. Starch breakdown is the result of the action of the hydrolytic enzymes alpha amylase, b-amylase, debranching enzyme, and alpha glucosidase (Guglielminetti, Yamaguchi, Perata, & Alpi, 1995). Several genes that function in the metabolism of sugar and its derivatives were upregulated under water stress. One is granule-bound starch synthase (Et_s66-0.13-1 and Et_s6672-0.28-1), an enzyme that converts ADP-glucose to amylose. This gene is more upregulated in Tsedey than in Quncho or Alba ( Figure 5). Another upregulated gene is Et_2217-0.39-1, bamylase, which is active in the conversion of starch to maltose. Amylase activity has a positive correlation with both shoot and root elongation and with plant survival when comparing tolerant and nontolerant rice genotypes (Ismail et al., 2009

| Signaling and transport
Aquaporins form a family of membrane proteins that allow control of water flow through a membrane either by changing the abundance of these proteins in the membrane or by changing the rate of flow through the pores. Aquaporin activity may be key in influencing water transport through waterlogged roots (Bramley, Turner, Tyerman, & Turner, 2007) and may be a critical part of determining a plant's isohydric characteristics, that is, how much they maintain the same water potential during variations in water availability. In the current work, two aquaporins of the TIP4 subtype, Et_s819-0.0 (TIP4-2) and Et_s4179-0.34 (TIP4-1), were downregulated. Downregulation of the PIP and TIP aquaporin subtypes under waterlogging stress has been found in Arabidopsis (Liu et al., 2005). In sorghum, waterlogging-tolerant genotypes showed differential regulation of PIP2-6, PIP2-7, TIP2-2, TIP4-4, and TIP5-1. However, SbTIP4-4 was downregulated in the sensitive genotypes (Kadam et al., 2017). TIP2 regulates the response to abiotic stresses (salt and drought) in bread wheat (Triticum aestivum) (Kayum et al., 2017) and tomato (Solanum lycopersicum) (Sade et al., 2009). In addition to the aquaporins, some proteins involved in metal and nitrate transport (Et_s7784-0.11, Et_s6198-0.9, Et_s594-1.9) were upregulated.

| Not found
Surprisingly, none of the stress-induced hormones in the ABA, GA, or JA pathways had detectable changes in expression after 9 days of waterlogging stress. Ethylene has been shown to play a role in both the induction of the GA pathway (Voesenek et al., 2003) and aerenchyma formation in maize and rice (Nishiuchi et al., 2012). Gene regulation in Arabidopsis shoots under flooding was also affected in a mutant with an ethylene signaling mutation. In addition, genes associated with ABA biosynthesis were upregulated in shoots and downregulated in roots, indicating a role for ethylene in the response. An ABA signaling mutation, abi4-1, affected the expression of several systemic responsive genes, suggesting that ABA biosynthesis may be part of the systemic response to flood response (Hsu, Chou, Peng, Chou, & Shih, 2011).

| CONCLUSION S
We report here on the morphology, growth, physiology, and differential gene expression in tef under conditions of long-term water stress starting at the tillering stage and early in development. Although tef has been reported to be waterlogging tolerant, this is the first quantification of changes to gene expression, morphology, and physiology under waterlogging. Three tef genotypes, two improved varieties and one landrace, were investigated and differences in waterlogging tolerance among the three genotypes were found. The improved variety Tsedey was found to have increased root and shoot biomass under waterlogging conditions when waterlogged in early development. It formed more adventitious roots and aerenchyma than the other two genotypes and is a promising candidate for further study. Differences in plant growth characteristics peaked at 9-11 days after the onset of stress for early waterlogging. Differential expression of 19-day-old seedlings waterlogged at the tillering stage shows changes in the cell wall, carbohydrate metabolism, and upregulation of genes involved in the response to ROS. In addition, genes affecting transport and lignification were affected. The identification of lines more tolerant to waterlogging conditions will aid the cultivation of cereal crops in poorly drained soils as tolerant waterlogging lines will be introgressed to elite tef genotypes, and the relevant genetic loci may be introduced to other cultivated crops. As this study shows that there is significant genetic diversity in tef genotypes with respect to waterlogging tolerance, a large-scale screening is underway to identify and quantify the performance of 500 tef genotypes under waterlogging stress.

ACKNOWLEDG MENTS
We thank Syngenta Foundation for Sustainable Agriculture and the University of Bern for financial and technical support. We also thank Christopher Ball and Jasmin Sekulovski for taking care of our plants.