Quantitative RT-PCR has become a powerful tool for analysis of gene expression because of its high throughput, sensitivity, and accuracy . However, the use of one or more stably expressed reference genes to normalise the variation introduced by RNA sample quality, RNA input quantity, and RT enzymatic efficiency is essential to achieving reliable results [13, 17]. To obtain a solid basis for normalisation of gene expression data, it is advisable to validate the expression stability of candidate reference genes under the conditions studied, rather than using reference genes published elsewhere . Validation of reference genes has been simplified with the design of statistical algorithms, such as geNorm and NormFinder, which not only test the expression stability of reference genes, but can also determine the number of reference genes required to provide accurate normalisation [17, 22].
This study describes the validation of candidate reference genes for normalisation of gene expression in perennial ryegrass. The most comprehensive dataset contains 422 field-grown leaf and stubble samples collected at different growth stages following a range of defoliation treatments and representing spatial and temporal replicates. Using geNorm, eEF1A (s) and eEF1A (m) were identified as the two most stable genes across the wide range of samples tested, followed by TBP-1 and YT521-B. Use of all four of these genes is recommended for normalisation, based on a suggested pairwise variation threshold of 0.15 . However, this is not an absolute rule and depends on the data. In the current study, based on V ≤ 0.15, two reference genes are sufficient for normalisation of qRT-PCR data from the callus, inflorescence and root samples, the etiolated seedlings of different cultivars, the laboratory-grown samples to evaluate water stress, and the field-grown samples harvested at the peak of each season.
One of the factors that may have made it more difficult to achieve V ≤ 0.15 in the current study is the large number of samples and treatments tested. Datasets containing smaller numbers of samples and treatments tended to require fewer reference genes for accurate normalisation  than larger datasets (> 50 samples; 2-4 treatments). As far as we are aware, the maximum number of qRT-PCR samples analysed hitherto using geNorm was 91 samples . Moreover, even other studies that analysed smaller sample numbers (< 50) were not able to obtain V values smaller than 0.21 when they tested eight or ten candidate reference genes in human breast cancer and osteoarthritic cartilage samples [30, 31]. Perennial ryegrass, like the human species, is an outcrossing, heterogenous species; hence within a sample of the ryegrass population there can be considerable genetic diversity between plants [32, 33]. Thus, more reference genes may be required to stabilise the variability in gene expression. The larger threshold for variability in reference gene expression may have also been a result of the fact that the plants were field-grown, and thus exposed to constantly changing environmental conditions.
In slight contrast to geNorm, the alternative algorithm, NormFinder, ranked eEF1A (s), YT521-B, eEF1A (m), and TBP-1 as the four most stably expressed genes in this dataset, with eEF1A (s) and YT521-B providing the best combination of two genes for normalisation of gene expression data. Although the reduction in the stability value when using the single, most stably expressed gene (0.039) compared with the two most stably expressed (0.030) is not large, if small differences in gene expression are to be detected then this increase in the accuracy of normalisation is still desirable.
Some studies that have utilised both geNorm and NormFinder have reported minor changes in gene stability ranking [26, 34–36], while others have observed relatively substantial changes [i.e. up to 15 places between the two methods; [37–39]. GeNorm and NormFinder rely on different mathematical approaches to calculate stability. GeNorm selects two genes with a low intra-group variation and approximately the same non-vanishing inter-group variation. In comparison, NormFinder selects the two best genes with minimal combined inter- and intra-group expression variation , which can have a notable effect on the subsequent gene stability ranking . Therefore, the fact that the ranking of candidate reference genes by NormFinder is not always identical to that defined by geNorm is not surprising. Some studies have chosen to base their reference gene selection on geNorm results because of its ability to identify the appropriate number of reference genes for accurate normalisation [37, 41]. Others have chosen NormFinder, because it examines the stability of each single reference gene independently, and not in relation to the other genes as geNorm does; which is important considering our limited knowledge regarding gene co-regulation .
When the target gene EF-Tu was quantified using the four reference genes recommended by geNorm, the two suggested by NormFinder or the least stable gene there were some differences in the calculated transcript abundance. The geNorm and NormFinder strategies differed in their estimation of EF-Tu at the 1- and 3-leaf stages of regrowth. In an ideal situation, normalisation using the genes defined by geNorm or NormFinder would have produced exactly the same result, indicating that there was no additional benefit in using four reference genes as opposed to two. While this wasn't the case, it is difficult to say whether the mathematical approach used by geNorm or NormFinder is superior. Therefore, the fact that the trend in transcript abundance throughout regrowth remained the same for both strategies suggests that either approach could be used for normalisation. Both approaches demonstrate the same up- or downregulation of the target gene, it is just the magnitude of the effect that differs, which could be taken into consideration when interpreting results. The trend in transcript abundance produced from normalisation using the least stable gene differed to the other two strategies. This highlights the importance of validating reference gene stability to ensure that low precision or misleading results do not occur [11, 12, 14].
Some of the reference genes used in the current study have been used previously as reference genes in other species. In plant studies, eEF1A is probably the most commonly used reference gene [42–44], although its expression profile has not always been validated before use. Using geNorm and/or NormFinder, eEF1A has been ranked as one of the top two reference genes in rice, perennial ryegrass, potato, and Lolium temulentum [23–25, 45], and third best in Arabidopsis (Arabidopsis thaliana L.), Brachypodium distachyon, and Ectocarpus siliculosus [11, 26, 27]. However, while eEF1A was proven to be a good reference gene in the current, and other published studies, there is also evidence that its expression profile is not as consistent as that of other reference genes that have been tested [18, 19, 39, 40]. One possible reason for this is that eEF1A is encoded by several genes in a multigene family, members of which have been shown to exhibit differential expression [46, 47]. In the current study the gene expression of two individual eEF1A genes in perennial ryegrass was analysed. The moderately expressed eEF1A (s) had a more stable expression profile than the more highly expressed eEF1A (h). This difference in the expression stability may explain why contrasting results exist for many tested reference genes [18, 24, 46], and highlights the importance of validating expression stability of candidate reference genes before use for normalisation.
Brunner et al.  suggested that simultaneous amplification of two or more members of a reference gene family by a single primer pair designed in the conserved coding region could result in more stable gene expression than a single gene amplification. This assumes that genes within the same family will have balanced expression, which is not always the case, as demonstrated with eEF1A here and by Reid et al. , with ubiquitin by Jain et al. , and with actin by Jian et al. .
The reference gene YT521-B, although identified as the most unstable gene in half of the datasets, was selected by NormFinder, along with eEF1A (s), as providing the best two-gene combination for normalisation of gene expression data in the most comprehensive dataset. To our knowledge YT521-B has not been tested before for use as a reference gene. In Arabidopsis, Bläsing et al.  illuminated five-week old seedlings for 4 hours in the presence of ambient or < 50 ppm carbon dioxide (close to compensation point where the rates of photosynthesis and respiration are balanced). They identified an YT521-B-like family protein (At5 g61020; orthologous to YT521-B from the current study) as a carbon fixation-responsive gene, which perhaps makes it slightly surprising that it was considered to be a stably expressed gene in the field-grown samples collected following a range of defoliation treatments. However, in eukaryotic cells, YT521-B are said to be ubiquitously expressed nucleolar proteins  playing vital roles in the assembly of genes into transcription centres and allowing efficient gene expression regulation . YT521-B appears to be a good reference gene candidate; however it is vital that it is validated due to its instability under some experimental conditions.
To the best of our knowledge, no other study has attempted to analyse gene expression in field-grown perennial monocotyledons, let alone perennial ryegrass. To successfully commercialise perennial plants specific for pastoral/turf/biofuel use, an understanding of gene expression in these plants during their regrowth cycle is necessary before we can harness the power of biotechnology for various industries. The prelude to this would be to validate a set of reference genes in order to harmonise the data from various experiments that are expected to follow suit. We believe that we have achieved this by the validation of several suitable reference genes for normalisation of target genes involving not only perennial ryegrass plants raised in controlled conditions, but also in the field.