A RT-qPCR assay based on SYBR Green detection was carried out to examine the stability of the expression of the 14 candidate genes (Table 1). The full sample set was included in each technical replicate to exclude any artefacts due to between-run variation. Each RT reaction was repeated once, and three independent technical replicates were performed for each experiment. The expression level of the candidate reference genes are presented as quantification cycle (Cq) values (Figure 1). The mean Cq values of the genes ranged from 17 to 32, with most lying between 20 and 25. CYP was the most highly expressed of the set, with a mean Cq of 19.6, and HDC the least (mean Cq of 32.7). EF1b showed the least variation (CV of 5.6%), while ACT2/7 (7.3%) and TUB4 (7.7%) were the most variable. The variation in Cq is illustrated as a scatter diagram in Additional file 2.

The variation in relative transcript quantity of the reference genes across all samples is shown as Figure 2. Here, transcript quantities are represented as percentages, relative to the aggregated reference transcript pool of each sample. The proportion of SKIP16, UKN2 and UKN1 transcript remained relatively constant across samples, while those of HDC and TUB4 were rather variable, especially with respect to developmental stage and tissue type. Although the expression level of UKN2 was fairly constant among almost all the samples, its expression was particularly low in the 2^nd triofoliolate at the stage when the 3^rd triofoliolate fully expanded. In contrast, the expression of HDC was particularly high in this tissue/developmental stage combination. TUA5 expression varied widely across developmental stages and tissue types, but was largely unaffected by photoperiodic treatment or cultivar. Thus, the transcript level of none of the reference genes was truly constant, rather it varied both temporally and spatially.

Melting curve analyses were performed following the RT-qPCR. The specificity of the amplicons was confirmed by the presence of a single peak (a representative trace is shown as Additional file 3). Electrophoretic separation of the amplicons produced a single fragment of the expected size in all cases, with no visible primer-dimer products. Five primer pairs were designed either to span an intron, or to target exon-exon junctions (Table 2), and used to compare amplicons derived from genomic DNA template with those from cDNA template. This comparison demonstrated that the cDNA template was free of contaminating gDNA. No amplification was detectable in the absence of template. Standard curves were generated using a ten-fold serial dilution of a cDNA pool, and these enjoyed a linear correlation coefficient (R2) of 0.994-0.999. Based on the slopes of these standard curves, the estimated PCR amplification efficiencies ranged from 94% to 106% (Table 2 and Additional file 4).

The expression stability of the set of candidate reference genes was examined by geNorm software, which calculates, for each gene, a measure of its expression stability (M) based on the average pairwise variation between all genes tested (Figure 3). Stepwise exclusion of the least stable gene allowed the genes to be ranked according to their M value (the lower the M value, the higher the gene's expression stability) 17, as depicted in Figure 3A. All the genes had an M value below the geNorm threshold of 1.5. Across all the samples, SKIP16 and UKN1 were the most stably expressed, and HDC the least. As a result, the latter was the first to be excluded from the analysis (Figure 3A). Among the various developmental stages, SKIP16 and UKN1 remained the most stable, and CYP the least stable. ACT11 and UKN1 were the most highly ranked across the set of tissues at the various developmental stages, while ACT2/7 was the least stable. In response to the short day (SD) and long day (LD) treatments, ACT11 and TUA5 were the most stable genes, and HDC the least; while in response to blue light (BL) and red light (RL) treatment, TIP41 and UKN2 were the most stable, and HDC the least.

To determine the optimal number of genes required for normalization, geNorm was used to calculate the pairwise variation (Vn/Vn+1) between sequential normalization factors (NF) (NFn and NFn+1) 17. As reported by Vandesompele et al (2002), a threshold value of 0.15 was adopted 17. In the SD/LD comparison, three genes was sufficient for normalization, since the V3/4 value was <<0.15 (Figure 3B). Differences in the expression stability of the candidate reference genes were less marked in the RL and BL photoperiodic treatment series, than in the other series (Figure 3). The V2/3 value for the RL/BL comparison was 0.091, so that TIP41 together with UKN2 would be sufficient for normalization purposes. Among the cultivars, the pair ACT11 and UKN2 produced a V2/3 value of 0.073. However, for the comparisons based on developmental stage and tissue type, four genes were necessary, since the V3/4 values lay above the threshold. When all the experimental samples were considered together, the V2/3 value was 0.196 and the V3/4 was 0.137, suggesting that the addition of a fourth gene did not improve the quality of the normalization (Figure 3B). Overall, the combination SKIP16, UKN1 and UKN2 was appropriate for all sets of samples.

Stability of expression was then re-analysed using the program NormFinder, which is based on a variance estimation approach 21, and ranks the genes according to their stability under a given set of experimental conditions. The ranking generated by this approach was slightly different from that determined by geNorm (Table 3). ACT11 and UKN1 were still ranked the highest for tissue samples, and ACT11 and UKN2 the highest for inter-cultivar comparisons. HDC, CYP and ACT2/7 ranked consistently poorly. Among developmental stages, EF1b and MTP emerged as the most stably expressed (ranked second and third by geNorm) (Figure 3). ACT11 and TUA5 were identified by both NormFinder and geNorm as being among the three most stable genes under SD and LD treatments. When evaluated across all the experimental samples, the same four genes were identified by both programs, although their rank order was slightly altered.

The expression pattern of GmFTL3, a soybean FLOWERING LOCUS T (FT) ortholog, was analysed using the selected reference genes (Figure 4). In A. thaliana, FT acts as a floral promoter and an integrator of various flowering pathways 4344454647. GmFTL3 has been proposed as a flowering promoter, since its ectopic over-expression in A. thaliana is associated with an extremely early flowering phenotype (unpublished data). Its pattern of expression was assessed at five distinct vegetative growth stages. When normalized using SKIP16, UKN1, MTP and EF1b as reference genes, transcript abundance gradually increased over time, peaking at the onset of flowering (the fourth trifoliolate leaf fully expanded) (Figure 4E). Similar expression patterns were generated when either three or two of the most stable genes (as identified by geNorm) were used for normalization (Figure 4C and 4D). When only one reference gene was employed, its expression was also rather similar to the above patterns (Figure 4A and 4B), but differences were evident in estimated transcript abundance, which was higher when normalized against SKIP16 than against UKN1, presumably because UKN1 transcript level was greater than that of SKIP16 (Figure 1). Normalization based on either of the less stable genes CYP or TUB4 produced a picture of GmFTL3 expression in which transcript level was constant during the vegetative growth stages (Figure 4F and 4G). Its relatively less abundant expression at the onset of flowering was a consequence of CYP and TUB4 up-regulation during this period. It suggested that not only the stability but also the abundance of a reference gene affected the normalized results.

The soybean cultivar Kennong18 (KN18) was used for most experiments. Plants were grown in a growth chamber under short day conditions (8 h light/16 h dark) at a temperature 25°C - 28°C. Seedling tissues were harvested before the expansion of the unifoliolate leaf. The root, hypocotyl, epicotyl, cotyledon, unifoliolate leaf and shoot apex (including the apical meristem and immature leaves) were sampled when the unifoliolate leaves had become fully expanded (about two weeks after sowing). A further sample of the root, along with the stem, unifoliolate leaves, various trifoliolate and lateral leaves, the petiole and the flowers were harvested when the fourth trifoliolate had become fully expanded (45 days after sowing, flowering onset). Pods and seeds were sampled at seven, 14 and 21 days after flowering, and at maturity. The aerial part of plants was also harvested respectively when the unifoliolate, first, second, third trifoliolate, and fourth trifoliolate were fully expanded (Additional file 5, indicated in yellow and green). To study the effect of altering the photoperiod, seedlings were exposed to either a long day (LD, 18 h light/6 h dark) or a short day (SD, 8 h light/16 h dark) regime. Fully expanded unifoliolate leaves were collected at 4 h intervals over 48 h, then the seedlings were transferred to either constant white light (LD) or constant darkness (SD), and the unifoliolate leaves re-sampled at 4 h intervals over a further 48 h (Additional file 5, indicated in grey). The effect of exposure to either red (RL) or blue (BL) light was monitored in etiolated seedlings subjected to red (Red-LED, 658 nm) or blue (Blue-LED, 436 nm) light in a growth chamber under LD conditions. The unifoliolate leaves were harvested at 4 h intervals over 48 h (Additional file 5, indicated in red and blue). Six further soybean cultivars were included: Heihe 27 (HH27), Zhonghuang 13 (ZH13), Jidou 12 (JD12), Tiefeng 31(TF31), Suinong 14 (SN14) and Fudou 1 (FD1). These seedlings were grown under either SD or LD conditions and the unifoliolate leaves were sampled 30 min before the lights were turned off (Additional file 5, indicated in purple). Totally, the experimental samples comprised 44 at various stages of development, 60 exposed to various photoperiod treatments, and 12 involving six different cultivars (Additional file 5). All samples were immediately frozen in liquid nitrogen and stored at -80°C until required.

Total RNA was extracted using the TRIzol reagent (Invitrogen, CA, USA) according to the manufacturer's instructions. Alternatively, total RNA from the petioles was isolated by the CTAB method 57. Only RNA preparations having an A260/A280 ratio of 1.8-2.0 and an A260/A230 ratio >2.0 were used for subsequent analysis. RNA integrity was verified by 2% agarose gel electrophoresis followed by SYBR Green staining. Before cDNA synthesis, the RNA was treated with RQ1 RNase-free DNase (Promega, Madison, WI, USA), according to the manufacturer's instructions, and first-strand cDNA synthesis was carried out using 4 μg RNA with the help of the RevertAid first strand cDNA synthesis kit (Fermentas, St. Leon-Roth, Germany) and oligo-dT primers, according to the manufacturer's protocol.

A set of 14 candidate reference genes was selected. This comprised seven conventionally used housekeeping genes; the soybean orthologs of the A. thaliana reference genes TIP41 (At4G34270), HDC (At1G58050) and UKN2 (At4G33380); and SKIP16 (At1G06110), MTP (At2G41790), PEPKR1 (At1G12580) and UKN1 (At3G13410), which were identified as potential reference genes via a soybean microarray gene expression analysis 37.

Primers were designed using Beacon Designer v7.0 (Premier Biosoft International, Palo Alto, California, USA) with melting temperatures 58-60°C, primer lengths 20-24 bp and amplicon lengths 60-134 bp. Experimental details are given in Table 2. Exon/intron boundaries were determined by aligning each cDNA sequence with its corresponding genomic sequence, downloaded from Phytozome http://www.phytozome.net/cgi-bin/gbrowse/soybean/. Five primer pairs were directed to locate on different exons or directly spanning exon-exon junction of each cDNA (Table 2). For each primer pair, reaction efficiency estimates were derived from a standard curve generated from a serial dilution of pooled cDNA. Mean quantification cycle (Cq) values of each ten-fold dilution were plotted against the logarithm of the cDNA dilution factor. An estimate of PCR efficiency was derived from the expression [10^(1/-S)-1] × 100%, where S represents the slope of the linear regression 58.

RT-qPCR was conducted using an ABI StepOne Detection System (Applied Biosystems, USA), based on SYBR Premix Ex Taq polymerase (TaKaRa, Toyoto, Japan). Each 15 μl reaction comprised 4 μl template, 7.5 μl 2× SYBR Premix, 0.3 μl (200 nM) of each primer and 0.3 μl ROX. The reactions were subjected to an initial denaturation step of 95°C/10s, followed by 40 cycles of 95°C/5s and 60°C/60s. A melting curve analysis was performed at the end of the PCR run over the range 60-95°C, increasing the temperature stepwise by 0.5°C every 10s. Baseline and quantification cycle (Cq) were automatically determined using the StepOne Software v2.0. Zero template controls were included for each primer pair, and each PCR reaction was carried out in triplicate.

Cq values were converted into relative quantities via the delta-Cq method using the sample with the lowest Cq as calibrator and incorporating the calculated amplification efficiencies for each primer pair (Table 2). The stability of reference gene expression was analysed with the geNorm (v3.5) and NormFinder (v0.953) software packages 1920. The former derives a stability measure (M), and via a stepwise exclusion of the least stable gene, creates a stability ranking. It also estimates the number of genes required to calculate a robust normalization factor (NF). NormFinder uses an ANOVA-based model to estimate intra- and inter-group variation, and combines these estimates to provide a direct measure of the variation in expression for each gene. All other statistical analyses were performed with SPSS (v13, SPSS Inc., Chicago, IL).

The stability of the reference gene set was validated using the 3,092 Genevestigator soybean genome microarray dataset, available at http://www.genevestigator.ethz.ch 55. The Meta-Profile Analysis tool was used to represent each reference gene's expression stability according to its UniGene IDs (see Table 1).