A single, partial cDNA fragment was cloned for each of these reference genes with the exception of β-TUB, for which two partial cDNA fragments were cloned (supplemental material). The cDNA fragments of the eight common reference genes EsGAPDH, EsUBQ, EsACT, EsEF, Esβ-TUB1, Esβ-TUB2, Esα-TUB, and Es18S ranged from 428 bp to 633 bp. BLASTN revealed that these amplified fragments had 85%–99% identity with similar sequences from other plant species, and six out of eight of these references genes had the maximum identity with a legume. For example, the EsGAPDH gene sequence had 91% identity with the Medicago truncatula GAPDH mRNA over a span of 574 bp (GenBank NoXM_003601780). All cDNA sequences were deposited to the GenBank database under accession numbers JN866814 to JN866821.

The primer sequence and amplicon characteristics of eight candidate reference genes for qRT-PCR are described in Table 1. The primer efficiency for each primer pair was greater than 90%. The specificity of each primer pair was further confirmed by gel electrophoresis analysis of the qRT-PCR amplification products (each primer pair amplified only a single product of the predicted size), and melting curve analysis (a single peak was obtained) (Figure 1).

Amplification of the EsACT gene in each of the 20 cDNA samples produced a single ACT-specific band with a predicted molecular size (approximately 300 bp) (data not shown), which confirmed that the RNA samples extracted from E. songoricum are appropriate for mRNA analysis. The expression of the eight reference genes from each RNA sample is shown in Figure 2. The median Cq value ranged from 10.53 to 28.78. Esα-TUB had the highest Cq value, which indicated relatively low expression. In contrast, Es18S had the highest expression level, and the threshold fluorescence was reached after approximately 12 cycles. Most Cq values were between 15 and 25 cycles, and the average Cq value of the samples (excluding Es18S) was 22 cycles.

GeNorm software was used to rank the eight reference genes based on the experimentally determined expression stability value M. Figure 3 shows the average expression values for the eight candidates under different conditions. According to the geNorm manual, the gene with the lowest M value is considered the most stable, while the highest M value indicates the least stable expression. All eight genes examined showed M values lower than the cut-off recommended by geNorm (M < 1.5). Across all the test samples, EsACT and Esβ-TUB1 were considered to be the most stable genes, while Es18S and Esα-TUB were shown to be the least stable genes (Figure 3A). Across the different tissues and developmental stages, similar result was obtained that EsACT and Esβ-TUB2 were the most stably expressed genes, while Es18S and Esα-TUB were the least stable (Figure 3B). For stress-treated samples, EsEF and Esα-TUB were the most stable reference genes, while EsGAPDH and Esβ-TUB2 were the least stable (Figure 3C). When combining adult plant material from the field with seedling material grown under controlled conditions, EsACT and Esβ-TUB2 genes demonstrated the best expression stability (Figure 3D).

The stability of the reference genes in a smaller experimental design was investigated. In root, stem and leaf samples of two-week-old seedlings grown under controlled conditions, the EsEF and Esβ-TUB2 genes were the most stable (Figure 3E). However, for field-grown adult samples (adult roots, assimilating branches and flowers), EsGAPDH and EsUBQ were the most stable (Figure 3F). For the different germination stages, EsEF and EsACT were the most stable, with an M value less than 0.1 (Figure 3G). Es18S ranked poorly across all tested conditions, while EsACT always showed a stable expression cross all the samples.

To determine the optimal number of reference genes required for accurate normalization, we calculated the pairwise variation (V_n/n+1) using geNorm. GeNorm uses 0.15 as the cut-off value, below which the inclusion of additional reference genes is not necessary. However, it has been suggested that this cut-off value is too strict [15]. In our analysis (Figure 4), for all the samples together, seven genes are required for proper normalization (V_7/8 = 0.163). For samples from the different tissues and developmental stages, the V_3/4 was 0.168, which is slightly greater than the recommended cutoff value (Figure 4). Under the stress conditions, the V_2/3 value is below 0.15 (0.147), which indicated that only the most two stable reference genes EsEF and Esα-TUB are needed to reliably normalize the expression data (Figure 4). Considering all the tested tissue samples, V_4/5 was 0.186, which is still higher than the cutoff, but when we analyzed smaller datasets, the V value improved in the seedling tissues (root, stem and leaf) and adult plant samples (branch, flower and root). The stability M value rank was slightly changed in seedling tissues and adult plant samples, but the pairwise variation (V_n/n+1) was much lower compared to the whole tissue samples. Only the combinations of the EsEF and Esβ-TUB2 (V_2/3 = 0.087), EsGAPDH and EsUBQ (V_2/3 = 0.113) genes were sufficient for reliable normalization of seedling tissues and adult plant tissues, respectively (Figure 4). Similar results were also found for the different stages of seed germination, only the most two stable genes (EsEF and EsACT) is enough for an accurate normalization result with the V_2/3 value less than 0.05 (Figure 4). The most stable reference genes and the optimal gene combinations were not identical for the different experimental conditions. The results obtained by geNorm are summarized in Table 2.

The expression of the EsDREB2 gene in response to drought stress was assessed using the one or two most stable reference gene(s) (EF and α-TUB) for normalization, which was validated by geNorm as described above and shown in Figures 3,4. The geometric average of the most stable reference genes in the stress treatment (EF and α-TUB) was used as an internal control. The expression pattern of EsDREB2 did significantly change with the reference gene(s) used for normalization. With either EsEF used alone for normalization or EsEF and Esα-TUB combined together as reference genes, EsDREB2 was found to be up-regulated by drought stress. The transcript gradually increased and peaked at 6 h and then began to decline at 12 h (Figure 5B,C). However, normalizing with Esα-TUB alone revealed that the expression level of EsDREB2 was decreased by drought stress (though it slightly increased at 6 h) (Figure 5A).

Eremosparton songoricum (Litv) Vass. seeds were collected from the Gurbantunggut Desert, within the Xinjiang Uygur autonomous region of the P. R. China (88°24′67′′E, 45°58′11′′N, 667m asl), and were soaked in 98% (v/v) sulfuric acid for 10-15 min to break the physical dormancy of seeds followed by washing with ddH₂O. Seeds were placed on moist filter paper in petri-plates, and seedlings were grown at 25 °C with cycles of 12 h light and 12 h darkness (100 μmol m⁻²·s⁻¹) and 60% relative humidity prior to stress treatments.

We exposed E. songoricum to nine different stress conditions to evaluate the stability of the tested reference genes. The stress treatments included common stresses experienced by this species in its natural environment (e.g., salinity, drought, grazing) and more common stresses that alter gene expression in plants in general (e.g., free radicals or oxidative potential, heavy metal exposure, and ABA). For NaCl, PEG, ABA, oxidation and metal treatments, 2-week-old seedlings were placed on filter paper saturated with 8 mL of the following solutions: 250 mM NaCl, 20% (w/v) polyethylene glycol (PEG6000), 100 μM ABA, 50 mM H₂O₂, 0.5 mM CuSO₄ and 0.5 mM ZnSO₄, respectively. For cold and heat treatments, 2-week-old seedlings were incubated at 4 °C and 40 °C, respectively. For UV radiation treatment, 2-week-old seedlings were exposed to 0.5 w/m² UV-B irradiation. For mechanical wounding, leaves were cut six times per leaf with a razor blade [8]. Two-week-old seedlings grown under identical conditions and handled in a similar manner served as the control (e.g., control plants were placed onto water-saturated filter paper). For all stress treatments described above, tissue samples (e.g., leaves, stems and roots) were harvested after 4 hours. For comparison with laboratory-grown E. songoricum materials, field-grown adult roots, assimilating branches and flowers were collected from the same location as the seed source described above (the Gurbantunggut Desert). Because seed germination is one of the most important stages for stress resistance [52], additional seeds were placed onto moist filter paper placed in petri-plates (as described above), and whole seeds were collected on days 1, 3 and 7 after germination. After harvest, all samples were flash frozen in liquid nitrogen and stored at −80 °C prior to RNA extraction.

Total RNA extraction was performed using RNAiso™ plus reagent (Takara, Japan) following the manufacturer’s instructions with little modification. Genomic DNA contamination was eliminated using RNase-free DNaseI (Takara, Japan). RNA concentration, purity, and integrity were determined using a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific, USA) and visually assessed via gel electrophoresis (1.2% agarose). Only RNA samples with a 260/280 ratio between 1.9 and 2.1 and 260/230 ratio higher than 2.0 were used for subsequent analyses. First strand cDNA was synthesized from 1 μg total RNA, l μL oligo-dT, 1 μL random hexamers and 4 μL 5× Primerscript Buffer (Takara, Japan). The RT-PCR reaction was carried out at 37 °C for 30 min on a C1000™ Thermal cycler (Bio-Rad, USA) in a final volume of 20 μL, and inactivation of the enzyme was achieved at 85 °C for 5 min. Prior to qRT-PCR, cDNA samples were tested at templates by RT-PCR for the ACT gene (data not shown). Finally, all cDNA was stored at −20 °C.

We selected the following seven classic reference genes spanning a range of biological functions as reference gene candidates in E. songoricum: 18S ribosomal RNA (18S), elongation factor 1 alpha (EF), cytoskeletal structural protein (ACT), ubiquitin protein (UBQ), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), and the cytoskeletal structural proteins α-tubulin (α-TUB) and β-tubulin (β-TUB). The primers for ACT, 18S, β-TUB and the gene of interest EsDREB2 were designed with Primer Premier 5.0 with the mRNA sequences of ACT (EU529707.1), 18S (AF293760.1), β-TUB (U12286.1) and EsDREB2 of E. songoricum (HQ687367). The primers for EF, UBQ, GAPDH and a-TUB were based on the literatures [43,53–55]. The information on the primers for the reference genes is provided in supplemental material (Table S1).

PCR was performed using TaKaRa Taq™ (TaKaRa, Japan), and the reaction conditions were as follows: 30 cycles with denaturation at 94 °C for 30 s, annealing at 55 °C for 30 s, and extension at 72 °C for 45 s. An initial denaturation step of 3 min at 95 °C and a final elongation step at 72 °C for 7 min were performed. The final amplification products were checked on a 1% (w/v) agarose gel in 1× TAE buffer. Amplicons were purified using the EZ Spin Column DNA Gel Extraction Kit (BIO BASIC INC, Canada) and cloned into the pMD™19-T vector according to the manufacturer’s instructions. Positive colonies containing recombinant plasmids were then sent to the Beijing Genomics Institute (Beijing, China) for DNA sequencing. The resulting DNA and deduced polypeptide sequences were used to query the appropriate databases using the BLAST algorithm [56,57] of the National Center for Biotechnology Information (NCBI).

Based on the gene sequences obtained from homology cloning, eight gene-specific primer sets were designed for qRT-PCR (Table 1). Specifically, primers were designed with Primer Premier 5.0 using the following as criterion: amplicon length from 100 to 300 bp and a Tm of 58 ± 1 °C. Amplification efficiency (E) was evaluated using a standard curve generated by qRT-PCR using a ten-fold dilution series over at least four dilution points that were measured in triplicate. Primer specificity was confirmed using melting-curve analysis after qRT-PCR and gel electrophoresis analysis of the amplicon. For a more detailed explanation of all qRT-PCR experiments, see the MIQE guidelines proposed by Bustin [58].

All cDNA samples were diluted 5-fold with RNase-free water before being used as templates in quantity analysis. Real-time PCR reactions using SYBR Premix Ex Taq™ (Takara, Japan) were performed using the CFX96 Real-Time PCR Detection System (Bio-Rad, USA) with 96-well plates. The reaction mixture consisted of 2 μL 1:5 diluted cDNA samples, 0.4 μL each of the forward and reverse primers (10 μM), 10 μL real-time master mix and 7.2 μL PCR-grade water in a final volume of 20 μL. Two biological replicates for all of the samples and three technical replicates of each biological replicate with a no-template control (NTC) were also used. The RT-PCR protocol was as follows: 30 s initial denaturation at 95 °C, 40 cycles of 94 °C for 5 s and 60 °C for 30 s. To verify the specificity of each primer, a melting-curve analysis was included (65 °C to 95 °C with fluorescence measured every 0.5 °C). Data were analyzed using Bio-Rad CFX Manager software (version 1.6) (Bio-Rad, USA).

GeNorm v. 3.5 (a publicly available software tool) was used to rank and analyze the stability of expression of the reference genes in each of the samples [15]. According to the geNorm manual, the Cq values were imported into Microsoft Excel and transformed into relative quantities using the formula Q = (1 + E) ^{(min Cq − sample Cq)} and then imported into geNorm (version 3.5). The expression stability value M and pairwise variation value V for each reference gene with all other genes were automatically analyzed and ranked according to their expression stability. Then, the optimal number of reference genes for normalization was determined.

EsDREB2 was cloned from our lab (GenBank No HQ687367) (data not shown). To assess the validity of the procedure for the selection of reference genes detailed above, we evaluated the relative expression level of EsDREB2. The expression of the EsDREB2 gene was evaluated across five samples: 0 h (control), 2 h, 6 h, 12 h, and 48 h after drought treatment. The expression of EsDREB2 was analyzed by Bio-Rad CFX Manager software (version 1.6) (Bio-Rad, USA) and is shown as the mean ± SE of three replicates for each sample.