In order to pinpoint the best reference genes in Brachiaria, known reference genes from other species were used to BLAST search against a Brachiaria brizantha EST (expressed sequence tag) library constructed from ovaries of apomictic plants in megasporogenesis and megagametogenesis. This library was validated by sequencing and annotating 2,000 clones, and out of these sequences, eight genes were chosen due to their high similarity to sequences related to commonly used reference genes, including polyubiquitin, ubiquitin-conjugating enzymes, elongation factor-1 alpha, glyceraldehyde-3-phosphate dehydrogenase and tubulin. Specific primers were designed and tested for amplification efficiency, including two sets of primers for an ubiquitin-conjugating enzyme to use as an internal control (Table 1).

In order to find the best reference genes for relative quantification, a high quality starting material is needed. For that, total RNA was extracted from all tissue samples using the same extraction protocol 14 for the different Brachiaria organs. All samples were treated with DNAse to avoid misinterpretation of qRT-PCR results by genomic DNA contamination in cDNA samples. RNA quality analysis and quantitation were performed by agarose gel analysis and a Nano-Drop ND-1000 spectrophotometer (NanoDrop Technologies) measurement, respectively. This procedure was crucial to guaranteeing the same amount of starting material and equivalent efficiency of cDNA synthesis from total RNA samples.

Based on DNA analysis by agarose gel electrophoresis and the dissociation curves (additional file 1), one single PCR product with the expected size was amplified for each of the nine sets of primers selected for this analysis (not shown). After the PCR reaction, the entire raw fluorescence data generated in Opticon3 was used for the primer amplification efficiency calculation and Ct determination with the miner algorithm 26. This algorithm accounts for each PCR exponential curve, making it is possible to have accurate values for the quantification of qRT-PCR. The amplification efficiency using this program can vary between 50% and 150%, and for the nine tested primer pairs it varied from 0.87 ± 0.012 (87%) to 1.01 ± 0.009 (101%), which are expected amplification efficiencies between compared genes 26.

The median Ct data in the 22 samples are shown in Figure 1, and the Ct variations among the samples for the different primers are shown in Figure 2. The range and distribution of the Ct values allow for a visualization of the least variable genes among the samples. This provides an indication of the most stable genes, which were ubiquitin-conjugating enzymes (BbrizUBCE1 and BbrizUBCE2) and elongation factor-1 alpha (BbrizEF1), and showed the narrowest Ct range and therefore the least deviation from the Ct median by the different samples (Figure 1). The Ct values of the candidate reference genes in all samples were within 13.99 and 33.22 cycles, showing a high range of variation between them. BbrizEF1 showed the highest expression levels, whereas BbrizSUCOA showed the lowest within the chosen set of samples. Depending on the expression level of the genes tested, it is suitable to choose a reference gene with similar expression levels of the tested genes.

We used the geNorm application for selecting the best reference gene for Brachiaria brizantha 17. GeNorm calculates a gene stability value (M) and a normalization factor (NF) based on the geometric mean of the expression values of the set of the control genes tested. The lower the M value, the more stably expressed the gene is. Also, the program enables the exclusion of the most unstable gene to recalculate the M value. Out of the eight genes used for analysis, only BbrizTUB showed an M value higher than the cutoff established by geNorm (M < 1.5), and two of them (ubiquitin-conjugating enzyme and elongation factor-1 alpha) showed the lowest M values, numbers well-suited for reference genes 2728. The primers used for the ubiquitin-conjugating enzyme (BbrizUBCE1 and BbrizUBCE2) and elongation factor-1 alpha (BbrizEF1) had M values of 0.47 and 0.79, respectively, when all of the genes for the calculation were considered (Figure 3a). However, after exclusion of the least stable gene, BbrizTUB, there was a decrease in the M value of all the other genes and also a change in the M values of the unstable genes, BbrizGDP and BbrizUBI (Figure 3b). To check if the primer design might interfere in the stability value of gene expression and to have an internal control of geNorm, a second pair of primers was used for the ubiquitin-conjugating enzyme that amplified a more external region of the gene. The M value for both sets of primers was the same, showing that the amplification region had no influence on the expression stability. Although the difference in the M value of the BbrizEF1 gene was higher when compared with BbrizUBCE, it is recommended its inclusion as a reference gene in qRT-PCR because of its high expression values, which is important whenever the tested genes are highly expressed. To support these data, Bestkeeper 19, another Excel-based tool based on pairwise correlation, was performed and showed similar results concerning the best reference genes for Brachiaria (data not shown). Having at least two reference genes is suggested for a more accurate qRT-PCR analysis. A previous report concerning the reference genes for Oryza sativa also showed that ubiquitin 5 and EF1 were the most stable genes for the tissues analyzed 21. In addition, a recent work on Vitis vinifera identified elongation factor-1 as one of the most stable genes for pre- and post-anthesis flowers, berries, leaves and roots 22. In species that show both apomictic and sexual development, differential expression screening in immature spikelets have been held in order to find genes related to apomixis development. For Eragrostis curvula it has been shown that, depending on the ploidy level and reproductive development of the plant, genes that are usually constitutive, such as ubiquitin and elongation factor can vary in expression level 25. In addition, for Paspalum notatum, another apomictic plant, among other genes, polyubiquitin and ribossomal protein showed different expression levels depending on the ploidy and reproductive development when comparing immature spikelets of apomictic vs sexual plants 24. Considering that in these two species the reproductive mode and ploidy level influence in expression levels of commonly used reference genes and if they will be used for apomixis molecular studies, stability in ovaries of sexual and apomictic plants is probably a relevant point to be considered. Therefore, the M value for only spikelets and ovary tissues in the four developmental stages in both apomictic and sexual B. brizantha was calculated (additional file 2). BbrizUBCE and BbrizTUB were again the more stable and the least stable genes respectively, while there was a slight difference in the order of stability of the other genes. The second and third best reference genes were BbrizE1F4A and BbrizEF1 with a 0.08 difference in M value.

The geNorm application considers two different factors in order to analyze gene expression stability: the average expression stability (M) and the pairwise variation (V). The pairwise variation (V_n/n+1) is calculated based on normalization factor values after the stepwise addition of a least stable reference gene (NF_n and NF_n+1) and indicates the minimum number of reference genes necessary for an accurate normalization. When analyzing all eight of the genes with a pairwise variation, there was not a significant difference in the V numbers, but there was an increase in the instability with the addition of BbrizSUCOA (V5/6) and BbrizTUB (V8/9, Figure 4), relatively unstable genes as shown by the Ct distribution in Figures 1 and 2. The optimal cutoff V number according to Vandersompele (2002) should be around 0.15, but other works using this application have shown a higher V number for the studied species 2930, depending on the amount of genes and type of samples tested. In B. brizantha, the addition of a fourth gene did not have a significant contribution to stability, comparing all tissues. Considering these values, we suggest that only two reference genes, BbrizUBCE and BbrizEF1, should be used for qRT-PCR experiments of root, leaf and reproductive tissues of the studied accessions. Besides, analysis of ovaries alone should be performed preferentially with BbrizUBCE, BbrizE1F4 and BbrizE1F.

Two accessions of Brachiaria brizantha from Embrapa were used in this work: BRA 002747 (B105), a sexual diploid (2n = 2x = 18), and BRA 00591 (B030), a facultative apomictic tetraploid (2n = 4x = 36) named B. brizantha (A. Rich) Stapf cv. Marandu, which were both cultivated in the field at the Embrapa Genetic Resources and Biotechnology.

For analysis of the most stable genes during male and female gametophyte development, ovaries and anthers of both accessions were dissected from flowers at four different stages of development before anthesis, as previously described by Rodrigues et al. (2003). For each RNA extraction experiment, around 1000 ovaries and 50 anthers of each of the four stages were isolated. Stages I and II are related to sporogenesis, and stages III and IV are related to gametogenesis 96. In addition, whole spikelet, leaf and root tissues were isolated from both B. brizantha accessions for RNA extraction.

Total RNA was extracted from each pool sample with TRIZOL (1/10 w/v) (Invitrogen™) with a modified method from the manufacturer's instructions. Samples were ground with a drill (AD-18 S Bionic Drill set) holding an RNAse-free polystyrene pistil. After extraction, the RNA sample was dissolved in 15–20 μl of 0.1% diethyl pyrocarbonate (DEPC)-treated water. RNA was treated with DNAse using on-column Qiagen DNAse Treatment (RNeasy MicroKit, Qiagen, Valencia, CA, USA) according to the manufacturer's instructions. The RNA concentration and A₂₆₀/A₂₈₀ ratios were determined before and after DNAse I treatment using a Nano-Drop ND-1000 spectrophotometer (NanoDrop Technologies), and 1.1% agarose gel electrophoresis was conducted to visualize the integrity of the RNA. Only the RNA samples with A₂₆₀/A₂₈₀ ratios between 1.9 and 2.1 and A₂₆₀/A₂₃₀ ratios greater than 2.0 were used for the analysis.

First strand cDNAs were synthesized from 1.5 μg of total RNA with OligodT and Superscript II enzyme (Invitrogen™). The first strand synthesis system was used according to the manufacturer's instructions.

Several described plant housekeeping genes were selected for the analysis. Genes already described as good reference genes for other plant species were used to BLAST search against B. brizantha EST (expressed sequence tag) ovaries libraries, and the list of selected sequences is shown in Table 1. Primers were designed within 800 bp of the polyadenylation site, since the primers came from an EST library constructed using the OligodT priming strategy. Primer 3.0 software was used for primer design. Amplicon lengths varied from 100 to 200 bp, with melting temperatures (Tm) varying between 59 – 60°C and primer lengths between 20–23 bp. The primers were screened for hairpins, dimmer formation, and target specificity by BLASTN http://www.ncbi.nlm.nih.gov/BLAST against the nr databank. Primer pairs were tested for specificity by RT-PCR and also by qRT-PCR, followed by a dissociation curve and agarose gel electrophoresis.

PCR reactions were performed in 96-well plates with the Chromo4 Real-Time PCR Detector System (BioRad^®) using SYBR^® Green to detect dsDNA synthesis. Reactions were done in 20 μL volumes containing PCR Buffer (Invitrogen™), 1.5 mM MgCl₂, 0.1 mM dNTPs, 0.25 U Taq Platinum (Invitrogen™), 0.1× SYBR Green (Amersham™), 200 nM of each primer and 10 μl sscDNA (corresponding to 5 ng of total RNA). Aliquots from the same sscDNA sample were used with all primer sets in two separate experiments. Two biological replicates for each of the 20 samples were used for real-time PCR analysis, and three technical replicates were analyzed for each biological replicate.

Reactions were run in a BioRad qRT-PCR machine using the following cycling parameters: 94°C for 5 min, 40 cycles of 94°C for 15 s, 60°C for 10 s, 72°C for 15 s and 60°C for 35 s. No-template controls (NTC) were included for each primer pair, and each PCR reaction was performed in triplicate. Dissociation curves for each amplicon and agarose gel were then analyzed to verify the specificity of each amplification reaction; the dissociation curve was obtained by heating the amplicon from 40°C to 100°C and reading at each °C.

The calculation of primer amplification efficiency and Ct determination were done using the miner algorithm 26. Raw fluorescence data generated with the Opticon 3 software (BioRad) was used for these calculations. After running the miner algorithm, Ct values were transferred as a Microsoft Excel file (Microsoft, Redmond, WA) for further gene expression stability analysis.

For analysis of gene expression stability and rank, geNorm v. 3.4 software was used. The Microsoft Excel file (Microsoft, Redmond, WA) with the raw expression Ct values for each tested gene in the 22 different samples generated with the miner algorithm was first analyzed with the qBase software version 1.3.4 http://medgen.ugent.be/qbase/ and then transferred into the expression stability program geNorm, version 3.4 http://medgen.ugent.be/~jvdesomp/genorm/, as described by Vandesompele et al. (2002). This application defines the most stable genes by calculating the mean pairwise variation between a particular gene and all the others used in one experiment and determines an M value. The highest M value corresponds to the least stable expression in a set of samples. As a result, the normalization factor (NF) is defined, by considering the M value of the most stable genes. This information allows for the establishment of the minimum number of reference genes required for an accurate calculation of the relative expression of a target gene. This ideal number is given by the inclusion of a certain number of genes in the NF calculation until there is no significant contribution to an additional gene. The raw data from the two biological replicas was used for gene stability analysis and both showed similar results.