Introduction

The Cucurbitaceae family includes crops such as cucumbers, squashes, luffas, melons and watermelons, and is widely cultivated and distributed around the world. However, genomic resources for all cucurbits are still scarce, which makes this family less attractive for genetic study and molecular breeding. Cucumber (Cucumis sativus L., 2n = 2x = 14) with the lowest chromosome number in the genus Cucumis is one of the most cultivated vegetable crops in the world, so the extensive research on this species has always been beneficial for the agriculture industry (Koo et al. 2004). Biochemical, physiological and cytogenetic studies of this plant have been carried out for several decades (Bhaduri and Bose 1947; Burzyński et al. 2005; Chen et al. 1998; Janicka-Russak et al. 2008; Koo et al. 2002; Migocka and Klobus 2007; Ramachandran and Seshadri 1986), whereas the analysis of cucumber genes, gene transcription and expression has grown only slowly, since most of the genes remain unknown. Now that the entire cucumber genome has been sequenced (Huang et al. 2009) and the partial contigs have been already deposited in the NCBI database (http://www.ncbi.nlm.nih.gov), significant progress in research into the molecular biology of this species should be expected. However, no validation of suitable reference genes for cucumber gene expression analysis has been performed to date. An ideal reference gene, known as an internal control gene, should be expressed at a constant level across various conditions, such as developmental stages or tissue types, and its expression should be unaffected by experimental parameters (Schmittgen and Zakrajsek 2000; Thellin et al. 1999). Moreover, the reference gene and the target gene should have similar ranges of expression in the study samples (Cappeli et al. 2008). Common reference genes for normalization of qPCR data in various animal and plant tissues include β-actin (ACTB), tubulin (TUA), cyclophilin (CYP2), elongation factor (EF), RNA polymerase II (RP II), phospholipase A2 (PLA2), glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and 18S or 28S rRNAs (Czechowski et al. 2005; de Jonge et al. 2007; Murphy et al. 2003; Timmons et al. 2005; Zhaoguo et al. 2009). A number of new candidate reference genes for Arabidopsis have been recently proposed (Czechowski et al. 2005) and their expression stability has been already evaluated under some experimental conditions (Czechowski et al. 2005; Remans et al. 2008). They include AT5G15710 (F-box protein), AT2G28390 (SAND family protein) and AT5G08290 (mitosis protein YLS8) which proved to be highly stable during heavy metal stress.

A number of recently performed analyses have shown that in different experimental systems the expression of the commonly used reference genes 18S rRNA, ACTB, TUA or GAPDH often remains unstable (Barber et al. 2005; Bas et al. 2004; Dheda et al. 2004; Jemiolo and Trappe 2004; Nicot et al. 2005; Remans et al. 2008; Yperman et al. 2004; Zhaoguo et al. 2009). As a consequence, the normalization of expression data obtained by employing more advanced methods of gene expression study, such as real-time PCR or microarrays, results in the subsequent misinterpretation of the results (Schmittgen and Zakrajsek 2000; Tricarico et al. 2002). It is now assumed that the correct reference genes should be selected by evaluating data from expression studies with the available statistical algorithms, such as geNorm (Vandesompele et al. 2002), Bestkeeper (Pfaffl et al. 2004) or Normfinder (Andersen et al. 2004). Nevertheless, most of the studies validating reference genes in different experimental settings concern human and animal tissues. In plants, suitable internal controls for gene expression studies have been identified for rice (Ding et al. 2004; Jain et al. 2006; Kim et al. 2003), Arabidopsis thaliana (Czechowski et al. 2005), poplar (Brunner et al. 2004), potato (Nicot et al. 2005), grapevine (Reid et al. 2006), soybean (Jian et al. 2008; Libault et al. 2008), tomato (Exposito-Rodriguez et al. 2008), Orobanche ramose (González-Verdejo et al. 2008), coffee (Barsalobres-Cavallari et al. 2009) and peach (Tong et al. 2009).

The aim of this research was to evaluate the stability of twelve cucumber genes for the purpose of normalization in studying cucumber target gene expression. These included commonly used cucumber reference genes, such as actin (ACT), elongation factor (EF), cyclophilin (CYP), ubiquitin (UBQ) and tubulin (TUA), and the newly identified candidates for reference genes that encode clathrin adaptor complex subunit (CACS), F-box protein, tonoplast intrinsic protein (TIP41), mitosis protein (YSL8), protein phosphatase 2 (PDF2), helicase (HEL) and protein homolog of At4g33380. The last seven genes were identified by screening of the cucumber genome using the known Arabidopsis homologs that have been shown to be stably expressed in different experimental conditions (Czechowski et al. 2005). The stability and suitability of cucumber genes as reliable internal controls were assessed by three different statistical applets: geNorm (Vandesompele et al. 2002), NormFinder (Andersen et al. 2004) and Bestkeeper (Pfaffl et al. 2004). Material for the analysis was collected from a wide range of experimental series including different tissue types/state of development/treatments/time-course of treatments. Therefore, the results obtained in the study could be useful in many further expression analyses of cucumber target genes.

Materials and methods

Plant material

Plants were grown under a 16-h photoperiod (180 mmol m−2 s−1) at 24°C during the day and 22°C during the night for 6–7 days or 2 weeks on standard nutrient solution (Migocka and Klobus 2007), pH 5.0 (with metals) or 6.0 (with other supplements). All growth solutions were filter-sterilized. For the assays including short-term metal treatment, 6 day-old plants were transferred on nutrient solutions supplemented with 20 μM metals (Cd, Pb, Cu, Zn, Mn, Ni) for 4, 12 or 36 h (Experiment 1). The solutions containing metals were exchanged every 12 h to ensure a steady metal concentration was accessible to plant roots. In the experiments with long-term metal stress, the germinated seeds were planted at once on nutrient solution containing metals and grown for 2 weeks (Experiment 2). During that time the solutions with metals were exchanged twice a week. In this experimental system, Pb was applied at 20 μM concentration while the level of other metals was decreased to 10 μM (Cd, Mn, Zn, Ni) or 5 μM (Cu), as those metals in higher concentrations considerably affected plant growth. For the assays including short-term treatment of plants with plant growth regulators or the agents inducing salt, osmotic or oxidative stress (Experiment 3), 7 day-old seedlings were transferred on the nutrient solutions containing NaCl (50 or 100 mM), mannitol (100 or 200 mM), PEG (5 or 10%), GSH, GSSH, DTT or H2O2 (all to the final 5 mM concentration), ascorbic acid (10 mM), ABA, IAA, 2,4-d, kinetin, salicylic acid or GA3 (all to the final 5 μM concentration). For each treatment, four root samples, four leaf samples or four shoot samples of 100 mg from four different plants were taken for RNA extraction and immediately frozen in liquid nitrogen before storage at −80°C.

Total RNA isolation and cDNA synthesis

Frozen cucumber tissues were disrupted under frozen conditions using mortar and pestle. Approximately 50 mg of the ground powdered tissues were resuspended in 500 μl of cold Tri-Reagent (Sigma). The following total RNA extraction was performed according to the manufacturer’s protocol. The RNA pellet was washed with ice-cold 75% ethanol, air-dried and dissolved in 30 μl of DEPC-water. To avoid any DNA contamination, samples were treated with RNAse-free DNAse I (Fermentas) at 37°C for 30 min. The reaction was stopped by addition of 2.5 mM EDTA to the samples immediately before 10 min incubation at 65°C. RNA concentration and purity was determined before and after DNA digestion using a NanoDrop spectrophotometer ND-1000 (Thermo Scientific) and the 260/280-nm ratio showed expected values between 1.8 and 2.0. The RNA integrity of randomly chosen samples was also confirmed by electrophoresis in denaturing formaldehyde agarose gel. Before cDNA synthesis, RNA samples were also screened for genomic contamination by PCR using the primer pairs designed for real-time PCR analysis according to the following scheme: 2 min at 94°C, followed by 40 cycles of 30 s at 94°C, 30 s at 55°C and 1 min at 72°C, with final extension for 10 min at 72°C. Finally, 1 μg of total RNA was used for cDNA synthesis with 2.5 μM oligonucleotide dT primers using the First Strand Synthesis Kit (Applied Biosystem) according to the manufacturer’s instructions.

Selection of cucumber reference genes and primer design

Twelve genes were selected for investigation to identify the most stable reference genes for qPCR expression analyses of target cucumber genes. These included the five genes already identified and commonly used as internal controls (ACT, CYP, TUA, UBQ and EF) as well as the newly identified homologs of Arabidopsis thaliana stable reference genes: clathrin adaptor complex subunit (CACS; At5g46630), expressed protein At33380, TIP41(At54g34270), helicase (HEL; At1g58050), phospholipase 2 (PDF2; Atg13320), mitosis protein (YSL8; At5g08290) and F-box (At5g15710). The cucumber whole-genome shotgun reads were queried with Arabidopsis nucleotide sequences using BLASTn to select cucumber homologs. The selected cucumber sequences were then input into the programs identifying the full cDNA sequences (comprising the 5′ and 3′ ends): GeneMark (http://exon.gatech.edu/eukhmm.cgi) and FGENESH (http://www.softberry.ru/berry.phtml). The primer pairs for qPCR amplification of the newly identified expressed sequence tags as well as for the already identified cucumber reference genes were designed using Lightcycler Probe Design Software 2 (Roche). The primers were further employed in a standard RT-PCR reaction using the Titan One Step RT-PCR System (Roche) to check for size specificity of the amplicon size. The amplicons of seven newly identified reference genes were sequenced to confirm specificity of the PCR products.

Quantitative real-time PCR and data analysis

Real time PCR was performed in 20-μl capillaries (Roche) with a LightCycler 2.0 (Roche). Reaction mixtures contained 1 μl of cDNA, 5 μl of RealTime 2× PCRMaster Mix SYBR® (A&A Biotechnology) and 1 μl of each of 10 μM primers in a total volume of 10 μl. Primer sequences for the reference genes are listed in Table 1. The following amplification conditions were applied: 30 s at 95°C; 45 cycles of 10 s at 95°C, 10 s at 55°C and 15 s at 72°C, with final melting for 15 s at 65°C. The general quality assessment of the PCR results was based on the amplification and melting curve profile of the samples in relation to the assay controls (non-template controls). Melting curve analysis was performed to confirm the specific amplicon and to identify putative unspecific PCR products (e.g. primer dimers, reaction mix contamination). Successive dilutions of the sample with the lowest crossing point (Cp) were used as a standard curve. Amplification efficiency was around 2. For each RNA extraction, measurements of gene expression were obtained in triplicate, and the mean of these values was used for further analysis. The data obtained were converted into correct input files, according to the requirements of the software, and analyzed using three different VBA (Visual Basic Applet) applets: geNorm (version 3.4), NormFinder (version 0.953) and BestKeeper (version 1.0).

Table 1 Cucumber reference genes and their primer sequences
Full size table

Results

Expression profiles of reference genes

The raw expression data analyses obtained from the qPCR assay revealed differences between the twelve reference genes (Electronic Supplementary Material Table). Based on the Cp value, the traditional reference genes encoding EF, cyclophilin, actin and tubulin were expressed at the highest level, reaching threshold fluorescence after 13–19, 15–20, 16–21 or 17–21 cycles, respectively. The transcript of CACS was slightly less abundant with the Cp in the range of 18–22 cycles. In addition, it was observed that the Cp values for GW881873 and TIP41 were not significantly higher (20–23 cycles), whereas in the case of F-box, YSL8 and HEL, these values reached 20–24 cycles. The genes with the least abundant transcripts included UBQ and PDF2 with Cp values in the range of 22–26 cycles. The expression of each gene varied across all RNA samples studied, with higher expression variations observed for TUA, ACT, EF and CYP (5 cycles) and lower expression variations for the remaining genes (below 5 cycles). When samples of one experiment were considered, the expression variations were lower for all genes except ACT. The discrepancy in the expression of the twelve genes in different cucumber tissue samples implied that a selection of reliable reference genes to normalize gene expression under various experimental conditions should be the prerequisite. In this experiment, the relative expression values were used for each cDNA sample as inputs for the geNorm and the NormFinder algorithms. In addition, amplification efficiencies and Cp values obtained during expression measurements of reference genes were used for the BestKeeper applet.

geNorm analysis

For each reference gene, the geNorm algorithm calculated the gene expression stability measure M as the average pairwise variation V for that gene with the other tested reference genes. The lower value of M reflected more stable gene expression and was the basis for ranking the genes by expression stability. Vandesompele et al. (2002) recommend the use of at least three reference genes for robust normalization of expression stability. To evaluate the impact of differential treatment/organ combinations on the expression stability of the reference genes, the geNorm input was separated to consider only short-term metal treatment (Experiment 1), long-term metal treatment (Experiment 2) or short-term treatment of plants with plant growth regulators or agents inducing salt, osmotic or oxidative stress (Experiment 3). According to the results obtained for Experiment 1 (Fig. 1, top), the average expression stability values M for the traditional genes were significantly (TUA, UBQ, CYP) or slightly (EF, ACT) higher than for the novel cucumber reference genes identified for this assay. Thus, the expression of the novel reference genes was significantly more stable under short-term metal stress, with the three genes TIP41, GW881873 and CACS ranked in the top positions. In addition, CACS and GW881873 together with F-box also proved to be the most stable genes in Experiment 2 (Fig. 1, middle); however, the other novel reference genes, such as HEL, YSL8 and PDF, were ranked among the five least stable genes during the 2 weeks-long heavy metal stress. In this experiment, the traditional reference gene, EF, was also stably expressed (fourth position according to geNorm), while TUA displayed the lowest stability of all the tested genes. Additionally, CACS was again highly ranked in Experiment 3, though EF and TIP41 were the most stable genes under different abiotic stresses and the growth regulators (Fig. 1, bottom). Another traditional reference gene, UBQ, also remained highly stable in this experiment, since it was ranked by the applet in the fourth position. On the contrary, YSL8, TUA and ACT were among the three least stable reference genes, demonstrating the highest M values. Based on the results obtained for the three different experimental series, it may be concluded that genes encoding CACS and TUA displayed the most and the least stable expression, respectively (Fig. 1). Considering the whole geNorm curves, the M values of the reference genes were the most diverse in the experiment, including the samples from roots, shoots and leaves of plants treated with metals for 2 weeks. Hence, the expression of some reference genes seems to be less stable when comparing different tissues and organs or during the long-term stresses than during the short-term changes in the environment. It has been recently shown that the use of the geometric mean of several internal control genes as a normalization factor is more accurate than the use of a single reference gene. However, the inclusion of unstable genes might in turn negatively influence the normalization factor. The optimal number of control genes for normalization was determined with geNorm by means of the pairwise variations (V n/n+1) between the sequential normalization factors (NF n and NFn+1) after successive inclusion of less stable reference genes (Fig. 2). According to Vandesompele et al. (2002), a variation of <0.15 indicates no significant contribution of an additional control gene to the normalization factor, which means that the optimal number of reference genes is achieved and the inclusion of an additional reference gene is not required. In all three experiments, V 2/3 was already below this cut-off, implying that the employment of the two best reference genes is likely to result in reliable gene expression normalization of target genes (Fig. 2).

Fig. 1
figure 1

Ranking of the candidate reference genes according to geNorm analysis. The M value reflecting gene expression stability was calculated separately for short-term metal treatment (Experiment 1) and long-term metal treatment (Experiment 2) as well as for treatment concerning salt, osmotic and oxidative stresses and plant growth regulators (Experiment 3)

Full size image
Fig. 2
figure 2

Determination of the optimal number of control genes for normalization. A variation <0.15 indicates no significant contribution of an additional control gene to the normalization factor, and hence the use of two best reference genes should result in reliable gene expression normalization in all three experiments

Full size image

NormFinder analysis

Similar to geNorm, the NormFinder program is another Visual Basic application tool for Microsoft Excel used for the determination of expression stability of reference genes. This program allows the estimation not only of the overall expression variation of the candidate genes but also of the variation between sample subgroups of one sample, e.g. roots, shoots and leaves. The results of the NormFinder analysis are presented in Table 2. Samples from three experiments were input separately and analyzed separately as one group or as three subgroups of one group. This separation resulted from a previous observation that the separation and definition of sample subgroups could have a significant effect on NormFinder’s output (Tong et al. 2009).

Table 2 Ranking of candidate genes and their expression stability values calculated by NormFinder
Full size table

The group of samples from short-term metal stress was divided into three subgroups according to the length of metal treatment: 4, 12 and 36 h. The seven subgroups of the samples separated according to the metal treatment (six metals plus control) were also analyzed but the output was not significantly affected (data not shown). The set of samples from long-term metal stress was divided into three subgroups based on the tissue type (roots, shoots and leaves). In addition, analysis of this set was performed including seven subgroups separated according to the type of metal. However, this separation did not influence the output, and therefore the data were not included in Table 2. Three subgroups in the third set of experiments were separated as follows: osmotic and salt stress (1), oxidative stress (2) and growth regulator treatment (3).

The NormFinder outputs with different sample subgroups and without any subgroups demonstrated common features. In the experiments concerning heavy metal stress, the best reference genes assigned by the applet included CACS ranked in the first position and EF, TIP41, GW881873 and F-box, ranked among the top positions in different order depending on the time-course of the stress. On the contrary, TUA displayed the least stable expression under heavy metal stresses. More striking differences between experiments 1 and 2 were related to the middle positions of the ranking: CYP and UBQ were expressed the least stably during short-term metal stress while their stability during permanent metal stress was ranked higher. Similar to the geNorm analyses, the least stable genes in Experiment 1 included mostly traditional reference genes: TUA, CYP and UBQ. However, NormFinder assigned EF as the most stable reference gene during temporary metal treatment, whereas geNorm ranked it in one of the last positions (Fig. 1, top). In Experiment 2, the results obtained from both geNorm and NormFinder calculations were consistent, indicating CACS, F-box and GW881873 as the best internal controls and TUA, HEL and YSL8 as the least stable genes. The NormFinder analyses of the results from Experiment 3 were also in accordance with the geNorm output. As was shown by the two applets, under the short-term courses of growth regulator treatment and abiotic stresses other than heavy metal pollution, the most reliable genes included EF, TIP41, F-box, CACS and UBQ, whereas the expressions of YSL8, TUA and ACT were the least stable. As for the geNorm results, the NormFinder output indicated that the novel cucumber reference genes might be suitable candidates for reliable internal controls useful for normalization assays. According to geNorm and NormFinder calculations, CACS should be the most reliable gene for the study of the expression of cucumber target genes in different tissues and under heavy metal stress, whereas EF showed remarkable stability of expression level under a broad range of various temporary treatments of cucumber plants.

BestKeeper analysis

Based on the results obtained in the geNorm and NormFinder analyses, we excluded the two most unstable genes in each experiment, namely TUA and CYP, TUA and HEL or YSL8 and TUA in experiments 1, 2 and 3, respectively. The expression of the ten remaining reference genes was further analyzed using the Bestkeeper applet. The third Excel-based spreadsheet software application designed to test gene expression stability estimates inter-gene relationships of possible reference gene pairs by the analysis of the numerous pairwise correlations of raw Cp values of each gene. Moreover, since the previously presented geNorm and NormFinder software was restricted to reference gene expression analyses, the BestKeeper software could analyze the expression levels of the target genes together with the reference genes (Pfaffl et al. 2004). For each pairwise correlation, the Pearson correlation coefficient (r) and the probability (P) value were calculated. Four different target genes were used for the evaluation of reference gene stability, using this applet. Similar to the previous analyses, the samples were analyzed as three separate inputs according to the category of experimental conditions.

The results are presented in Table 3. Reference genes used for the test demonstrated significant correlation with each other and strong correlation with the BestKeeper index (r > 0.9 in the vast majority of cases). The strongest inter-gene correlation was observed for CACS and TIP41 in the short-term metal stress, F-box and CACS in the long-term metal stress and TIP41 and EF during short-term treatment of plants with oxidative, osmotic or salt stress or growth regulators (r > 0.93). The highest Pearson correlation coefficient indicates that these pairs of genes showed the most similar expression patterns in the conditions of our experiments. CACS demonstrated the strongest correlation with the BestKeeper index in Experiment 1 (r = 0.98), while F-box (r = 0.99) and TIP41 (r = 0.983) were the most reliable reference genes in experiments 2 and 3, respectively. However, the differences in the Pearson correlation coefficients between the first and the second reference gene in the three rankings were minute. The Bestkeeper selection of the best reference genes was in agreement with the geNorm and NormFinder analyses. According to the calculations based on the raw Cp values in relation to Bestkeeper index, YSL8 (r = 0.765), PDF (r = 0.873) and ACT (r = 0.884) were the least stable genes in experiments 1, 2 and 3, respectively. All three applets ranked PDF and ACT poorly in experiments 2 and 3, respectively, whereas YSL8 held a higher position in the geNorm and NormFinder rankings than in the Bestkeeper calculations. Nevertheless, the three different statistics produced highly comparable results in selecting the most reliable cucumber reference gene for normalization of target gene expression in different tissues and under heavy metal stress (CACS) or during the exposure of plants to other environmental stresses or growth regulators (EF).

Table 3 The expression stability values for cucumber reference genes calculated by the BestKeeper applet
Full size table

Discussion

The invention of quantitative PCR (qPCR) has significantly improved the detection and quantification of expression profiles of selected genes in distinct biological samples. The main advantages of this technique are high sensitivity, high specificity and broad quantification range (Bustin et al. 2005; Bustin and Nolan 2004; Gahon et al. 2004; Ginzinger 2002). However, in recent years it has been well proven that accurate data normalization is an absolute requirement for correct measurement of gene expression using the qPCR method (Andersen et al. 2004). Until recently, it has often been assumed that the best normalization was achieved using an internal reference gene, since this takes into account variation in the amount of starting material between samples caused, e.g., by sample-to-sample variations, variations in RNA integrity, RT efficiency differences and cDNA sample loading variations. Commonly used housekeeping genes, e.g. GAPDH, albumin, actins, tubulins, cyclophilin, micro-globulins, 18S rRNA or 28S rRNA seemed to be suitable, since they are present in all nucleated cell types, being necessary for basic cell survival. Hence, their expression has been considered to be stable in various tissues. Nevertheless, numerous treatments and studies have already shown that the expression of these genes might also be regulated and vary under experimental conditions (Thellin et al. 1999). Therefore, normalization with multiple reference genes is becoming the standard and the reports that identify such genes in plant research have been increasing in number now that algorithms measuring gene expression stability are available (Vandesompele et al. 2002; Andersen et al. 2004; Pfaffl et al. 2004). The present study reports the first validation of housekeeping genes in cucumber allowing the identification of the most suitable reference gene(s) for normalization of qPCR data. The detailed analysis included a broad spectrum of samples obtained from cucumber plants: the experiments considered the analyses of gene expression in different plant tissues (roots, shoots, leaves), under different external factors (stresses or growth regulators) and under different time-courses for the presence of external factors (metals).

Several approaches have been proposed to measure the stability of reference gene expression and it is assumed that a comparison of different algorithms allows better evaluation (Tong et al. 2009). Therefore, we used three statistical applets, geNorm, NormFinder and BestKeeper, to validate five cucumber reference genes in different experimental conditions. Since the number of potential candidate genes identified in cucumber has been limited to date, we have selected and identified seven novel reference genes based on the available data from the genome-wide study and testing of Arabidopsis thaliana reference genes (Czechowski et al. 2005). The novel cucumber reference genes included clathrin adaptor complex subunit (CACS), homolog of At4g33380, RNA helicase, PDF2-like protein, YSL8-like mitosis protein, TIP41-like protein and F-box protein. Their homologs in A. thaliana displayed a remarkably stable expression pattern, as calculated by geNorm software on a variety of samples including different developmental stages, organs, tissues, genotypes and various cultivation conditions (Czechowski et al. 2005). This report has clearly shown that among the 21 analyzed Arabidopsis reference genes, the majority of novel genes were expressed considerably more stably than the traditional and commonly used ACT, UBQ or EF. In our research, the novel cucumber reference genes were also mostly ranked higher than traditional genes, except EF which was assigned as the best internal control in Experiment 3.

Comparing the outputs obtained from three different applets, it may be concluded that CACS should be the most reliable reference gene for studies on the effect of temporary and permanent heavy metal stress on target gene expression in cucumber roots, shoots and leaves. This gene also ranked very high in the geNorm analyses of Arabidopsis samples from different developmental series and different nutrient-related stresses (Czechowski et al. 2005). Thus CACS may be considered as a reliable internal control for organ and tissue expression analyses as well as during heavy metal stress or deprivation and/or readdition of different nutrients (N, C, P, S). Other suitable novel genes considered as reliable genes during heavy metal stress included cucumber F-box and At4g33380 homolog, as well as TIP41-like protein. Their Arabidopsis homologs ranked high in the complex analyses of Czechowski et al. (2005); although TIP41 was ranked as the weakest candidate of ten reference genes tested by Remans et al. (2008) in samples from roots and leaves of plants treated with Cu and Cd. Due to problems with primers specific for the genes, Remans et al. (2008) could not verify the expression of Arabidopsis At4g33380 and CACS under Cd and Cu stress. Comparing the expression of some novel and traditional (ACT, EF, UBQ) reference genes, Remans et al. (2008) selected F-box, YSL8 and SAND family protein as the most stable genes when considering the effect of Cu and Cd treatment on Arabidopsis roots and leaves. In our analyses concerning heavy metal stress and different cucumber organs, the stability of YSL8 expression was lower than the stability of some other novel reference genes. This might be a result of the differences in the range of samples that were used for the expression assay. These samples included not only roots and leaves but also shoots collected from plants and the treatment included six (Cd, Pb, Cu, Zn, Mn, Ni) and not two different heavy metals.

Interestingly, only the novel reference genes were among the best five reference genes assigned by the software through the analyses of samples from Experiment 1, whereas the analyses of samples from Experiment 2 revealed that of the five traditional cucumber reference genes tested, EF and UBQ could be considered as stable internal controls. Arabidopsis EF1α and UBQ10 were also highly ranked in the Czechowski et al. (2005) analyses of 79 samples from different developmental series. Nutrient-related stresses significantly affected EF1α but not UBQ expression patterns (Czechowski et al. 2005). Thus, it is suggested that EF and UBQ should be used as internal controls in organ/tissue/developmental expression analyses rather than during various cultivation conditions.

The analyses of reference gene stability in the third experiment also revealed similarities in geNorm, NormFinder and BestKeeper outputs: all algorithms proved that EF and TIP41 were the most reliable reference genes for the normalization of gene expression results obtained from studies using cucumber roots treated with various stresses (oxidative, osmotic or salt stress) and growth regulators. According to geNorm and NormFinder, TUA and YSL8 were the least stable reference genes in Experiment 3, so they were excluded from further Bestkeeper analysis. On the other hand, Bestkeeper selected GW881873 and ACT as the least reliable internal controls and ACT was also ranked poorly by geNorm and NormFinder.

Hence, the completion of the three analyses has shown that although all three softwares produced slightly different rankings of reference gene stability, the set of most reliable genes was similar in their outputs. Rankings for the three diverse experiments were different, particularly in the middle positions. The inconsistency between geNorm, NormFinder and BestKeeper results has already been observed and is generally expected since all three applets are based on distinct statistical algorithms (Cruz et al. 2009). It has been recently proposed that the geometric mean of several internal control genes as a normalization factor is more accurate than the use of a single reference gene. The rankings presented in this study provide a basis for the further selection of genes that could be combined together to normalize qPCR analysis of cucumber gene expression. The results also confirm that the study of reference gene expression stability should be frequently performed before the normalization of target gene expression data. In general, the analyses revealed that novel cucumber reference genes CACS, TIP41, F-box protein, GW881873 and traditional EF were among the best reference genes for accurate data normalization in the conditions of our experiments. In contrast, the traditional reference genes (UBQ, CYP, TUA and ACT) were ranked lower in the stability rankings. Previous data regarding traditional reference genes as internal controls are inconsistent. It was previously proved that UBQ10 expression was unstable in different tissues at various developmental stages in rice, soybean and peach (Coker and Davies 2003; Jain et al. 2006; Tong et al. 2009). However, the gene displayed high stability in Arabidopsis throughout the development and under a wide range of environmental conditions including biotic and abiotic stress as well as growth regulators (Czechowski et al. 2005). UBQ was also assigned as a reliable internal control in studies on tomato subjected to nitrogen, cold and light stress (Løvdal and Lillo 2009). Some recently published reports have also proved that ACT might not fulfill the requirements of reference gene definition (Gutierrez et al. 2008; Nicot et al. 2005; Tong et al. 2009). This might be of great importance, since this gene is one of the most frequently used reference genes in studies on gene expression in many organisms, including plants. Among the tested cucumber reference genes, both CYP and TBA were ranked poorly as reliable internal controls when considering overall analyses. It had been previously proved that the expression of CYP varied in different tissue samples of peach, in tomato subjected to different biotic and abiotic stress, and in grapevine throughout berry development (Nicot et al. 2005; Reid et al. 2006; Tong et al. 2009). CYP mRNA synthesis was also shown to be variable at different developmental stages of maize and stimulated in maize and bean by mercuric chloride treatment and other abiotic stresses, including heat-shock, wounding, salt stress and low temperature (Marivet et al. 1992, 1995). In addition to this, different drugs significantly induced CYP transcription in human tissues (Pérez et al. 2003). Therefore, particular caution should be taken when CYP is considered as a reliable reference gene. A similar conclusion has been proposed for TUA, which displayed highly variable expression patterns in different peach tissues samples (Tong et al. 2009). On the other hand, the expression of TUA and UBQ remained quite stable in various poplar tissues collected at different developmental stages, and at different times of the year (Brunner et al. 2004).

Taken together, the data obtained during previous studies and our study confirmed the need for the evaluation of reference genes under different experimental conditions, proving that a reference gene that is stable in one experiment may not be a reliable internal control in other conditions. Our results also confirm the data from previous studies indicating that the novel reference genes should be selected and evaluated alongside the traditional reference genes, as they are very often more stable in various experimental conditions than the commonly used internal controls. This first evaluation of the expression stability of cucumber reference genes in a wide range of experimental samples should be useful in further molecular research in this species. Furthermore, the novel cucumber reference genes that displayed a remarkably stable expression pattern in our experimental samples can be successfully used for the accurate quantification of cucumber target gene expression using real-time PCR assay.