Our results (GEO GSE21868) showed that all probes present on the array were capable of hybridizing successfully (signal > background) when tested on the different samples. The sensitivity of the array was demonstrated by the wide signal dynamic range obtained (log2 values of 4 to 16). The experimental metrics report (NimbleScan v2.5) was used as recommended by Roche/Nimblegen to generate summary statistics that can be used to identify any potential problems during hybridization. Most of these metrics (interquartile density, signal range, uniformity mean, uniformity CV (coefficient of variation), number of empty features on the array, mean empty, the number of random control features present on the array, mean Random) assume probe randomization on the array surface. Thus, deviation from uniformity (i.e. outside the recommended value range) across the array could suggest potential artifacts during hybridization. Metric values (data not shown) for all 18 samples analyzed were within the recommended value range indicating that hybridization quality was satisfactory.

A crucial aspect of all microarray experiments is good system reproducibility enabling direct and reliable comparisons between different experiments. Precision describes how accurately the hybridization signal intensity can be reproduced and is usually reported as a correlation coefficient, standard deviation or average replicate error between duplicated experiments using the same RNA sample [19]. Raw expression data on the 18 (9 × 2) flax hybridization experiments were normalized and Pearson's correlation coefficients were calculated for the data sets of hybridization signal intensities. High correlation coefficients were obtained in all cases and the results obtained for the flax leaf sample are shown in Figure 2. These results (r ≥ 0.98; p-value < 2.2e^-16 ) indicate an almost perfect correlation between different technical experiments underlining the extremely high precision level for our flax oligo-array platform. Such reproducibility is similar to that obtained with Affymetrix GeneChips [20] and is probably also due to probe redundancy. Previous work has shown that the use of multiple independent oligonucleotides designed to hybridize to different regions of the same RNA improves signal-to-noise ratio and the dynamic range of detection, as well as minimizing cross-hybridization effects [21].

Accuracy describes how close to a true value a measurement lies. It can be estimated in experiments where a number of realistic targets are spiked at known concentrations into relevant RNA populations, or from comparisons with validation experiments, or by correlation of two different methods [19].

Gene-specific quantitative qRT-PCR was used for the cross-validation of platform performance and as an assessment of microarray accuracy. We selected 9 genes reflecting up-regulated (c20715, c2491 c602), down-regulated (c2533, c3323, c4370) or equally-expressed genes (c21991, c24118, c9380) in leaf (L) samples compared with stem (SIGC) samples. Selected genes also covered a wide range of signal intensity (5.3849-14.0473). Table 4 shows the expression levels of the same target genes detected by qRT-PCR and microarray analyses. Expression level was calculated as log2ratio of mean signal intensity across three technical replicates between L and SIGC samples. A statistically significant correlation (r = 0.9823, p = 2.376e^-06) between qRT-PCR and microarray results was obtained for all tested genes (Figure 3) indicating high concordance between these 2 data sets. The observation that the fold-change values in gene expression are lower for the microarray data as compared to the qRT-PCR data is most likely due to data compression resulting from limited dynamic range or signal saturation [22]. These data indicate that our oligo-array platform is able to accurately predict the direction of change of gene expression level (i.e. either up or down regulation) between subsets of interest.

The results of principal component analysis (PCA) of microarray data and representative K-means profiles of tissue-/stage-specific gene expressions are shown in Figure 4. Three major regions of dispersion are found: A) genes specifically expressed in roots (Figure 4, top, profile 1), B) genes expressed specifically in stems and leaves (Figure 4, left, profiles 2-4) and C) genes specifically expressed in seeds (Figure 4, right, profiles 5-7). Genes similarly expressed in all tissues are clustered in the middle of the figure.

J-Express generated a list of 1,357 specifically expressed genes of which 609 were root-specific genes, 599 leaf-specific genes, 79 shoot-specific genes and 70 seed-specific genes. Differential expression profiles of these genes were verified by a SAM test with a FDR less than 5%. Annotated genes in these different groups were then classified into different GO biological processes and the percentages of tissue-specific gene expressions were calculated for each process (Figure 5A).

Overall, these results are in general agreement with the known physiological processes of the different organs/tissues thereby suggesting that our platform is capable of generating biologically-useful gene-expression data. For example, leaves are well known to be the plant's main photosynthetic organs and leaf structure is closely associated with its photosynthetic function and chloroplast biogenesis [23-31]. The observation that the GO process 'photosynthesis' is entirely composed of genes specifically expressed in the leaves (Figure 5A), as are the GO processes 'cellular component biogenesis', and 'protein modification process' is in agreement with the biological structure and role of this organ. In root tissues the GO processes 'cell growth' and 'response to stress' were the most represented functions in agreement with the biological activities of this organ [32,33]. As in other plant species, the flax stem is characterized by the differentiation of different specialized tissues [18,34] and the majority stem-specific gene expression in the development GO process is in keeping with the biological activity of this organ. Finally, cell-cycle and cellular homeostasis GO processes were associated with embryo-specific gene expressions characteristic of embryogenesis [35].

Similar analyses (Figure 5B) of the relationship between different tissue-specific genes and metabolisms also suggest that our platform is capable of generating biologically-relevant information. Of interest are the relatively high percentages of genes associated with lipid and carbohydrate metabolisms in seeds since direct quantitative measurements [36] have shown that sucrose and lipids represent the largest metabolite pools in flax seeds. Similarly, secondary metabolism represents the major metabolism in flax stems and is most likely associated with the lignification of xylem secondary cell walls [34].

We compared the expression of four genes known to be involved in the process of secondary cell wall biosynthesis and wood formation characteristic of inner stem tissues [37]. Expression levels of cellulose synthase (c2533), phenylalanine ammonia-lyase (c247), transcinnamate 4-hydroxylase (c11079), and caffeoyl-CoA 3-O-methyltransferase (c15771) were calculated as log2ratio of the signal intensity in each tissue vs. the average of signal intensity across all analyzed tissues (Figure 6). As expected, the great majority of these genes were more highly expressed in inner tissues (at both developmental stages examined). These observations confirm that the flax oligo-array platform is able to accurately discriminate gene expression profiles in different tissues.

In order to verify that our oligo-array platform could be used in future studies to provide biologically-relevant information about cell wall formation and fibre development in flax, we compared expression profiles between the heavily lignified inner stem tissues and the outer stem tissues enriched in weakly-lignified bast fibres at two different developmental stages.

The results (Additional files 1 and 2) showed that in vegetative stage samples, 203 genes were over-expressed in inner stem tissues as compared to outer stem tissues (cut-off: log2ratio (SIV/SOV) > 2), and 229 genes were over-expressed in outer stem tissues as compared to inner stem tissues. Examination of the 203 genes over-expressed in the inner stem tissues showed that 56 genes (28%) were potentially associated with cell wall formation and xylem identity. Of these differentially-expressed cell wall genes, 50% (28 genes) were associated with lignification and included both monolignol biosynthetic genes and laccases involved in the oxidative polymerisation of this phenolic polymer. Such an observation is in agreement with the fact that flax inner stem tissues, mainly composed of xylem, are much more heavily lignified than the outer tissues containing cellulose-rich bast fibes [34]. Other cell-wall related genes coded for proteins involved in polysaccharide cell wall polymer synthesis and remodelling. In addition, 3 genes (C1109, C3396, C5556) coding for fasciclin-like arabinogalactan proteins (FLAs) were also over-expressed in inner tissues as compared to outer tissues. Interestingly, FLA genes are highly expressed in flax outer stem tissues, as well as in poplar tension wood where they have been hypothesized to be involved in the formation of cellulose-rich gelatinous fibres (g-fibres) [17,37,38]. Our observation (see below) that 3 other FLA genes (C2947, C3576, C5237) are more highly expressed in flax outer stem tissues when compared to inner stem tissues could suggest that these proteins play a role in secondary cell wall formation in both flax inner and outer stem tissues.

Similar analyses (Additional files 1 and 2) of the 229 genes over-expressed in flax outer stem tissues when compared to inner stem tissues revealed that only 16 genes (7%) were associated with cell wall formation. In contrast, 25 genes were potentially associated with lipid and wax metabolism, 8 genes were related to photosynthesis and 8 genes were stress-related. These differences reflect the different physiological status of flax inner and outer stem tissues. For example, the high percentage of 'cell-wall-related' transcripts in inner stem tissues is associated with secondary xylem formation while the relative abundance of photosynthesis-related transcripts reflects the fact that outer stem tissues are green. Nevertheless, it is interesting to note that 6 genes (C1423, C6325, C353, C1410, C8886, C5052) over-expressed in outer stem tissues corresponded to Lipid Transfer Proteins (LTPs). LTPs have been previously associated with diverse aspects of cell wall development and formation and have also been shown to be more highly expressed in flax outer stem tissues [17,39]. Such an observation could suggest that such proteins might be associated with fibre maturation in flax. As indicated above, 3 FLAs (C2947, C3576, C5237) were also over-expressed in outer stem tissues as compared to inner stem tissues.

Comparison (Additional files 1 and 2) of inner and outer stem tissue expression profiles (cell-wall related genes) at the green capsule stage revealed a very similar differential expression pattern to that observed for the vegetative stage suggesting that little modification/evolution of inner and outer stem transcriptomes occurs between these two developmental stages. This was confirmed by the observation that only 3 cell-wall related genes (C25634, C2733, C11945) were up-regulated in inner stem tissues at the vegetative stage as compared to the green capsule stage, and no cell-wall related genes were up-regulated in inner stem tissues at the green capsule stage as compared to vegetative inner stem tissues. In outer stem tissues, 8 cell wall genes (C31544, C37539, C6312, C59350, C6820, C44241, C5939, C51183) were more highly expressed at the vegetative stage as compared to the green capsule stage. Two of these genes (C44241, C5939) code for xyloglucan endotransglycosylase/hydroalses (XTHs), one gene (C6820) codes for a secondary cell wall associated glycosyltransferase and one gene for a FLA. Since these 4 genes are implicated in cell wall remodelling and assembly events, such an observation could suggest that cell wall formation in outer stem tissues is more active at the vegetative stage as compared to the green capsule stage. No cell wall genes were more highly expressed at the green capsule stage as compared to the vegetative stage.

The vegetative stage (7-8 weeks) corresponds to the fast-growth stage associated with the increase in plant height and fibre lengthening above the snap point, as well as fibre cell wall thickening below the snap point [40]. At the green capsule stage (10-11 weeks) fibres have stopped elongating and further fibre development is restricted to continued fibre cell wall thickening [40,41]. Our results would suggest that continued fibre thickening (as well as continued secondary xylem formation in flax inner stem tissues) is associated with continued expression of cell wall related genes.

Altogether these analyses confirm that our oligo-array platform represents a powerful tool for investigating cell wall development and fibre formation in flax stems. Our arrays should also prove extremely useful for investigating other interesting biological processes in flax such as oil and seed formation, disease resistance and tolerance to abiotic stress.

We used our platform to see whether the flax microarrays were able to detect differentially-expressed genes between 2 contrasting flax genotypes - Drakkar and Belinka. Drakkar produces better quality fibres than the variety Belinka, as well as being more resistant to the fungal pathogen Fusarium http://www.lin-itl.com[42]. RNA was isolated at the green capsule stage from outer stem tissues of field-grown flax plants and 3 biological and 2 technical repetitions were used for each variety. Following hybridization, the Log2ratio was calculated as the intensity signal of each gene in the Drakkar genotype vs. Belinka. The Pearson's correlation was > 0.99 for technical repetitions and > 0.97 for biological repetitions for both Belinka and Drakkar. A total of 428 up-regulated genes and 367 down-regulated genes were found. Only 269 genes were annotated or associated with a known biological function. (Additional files 3 and 4).

Since the 2 genotypes show differences in fibre quality and pathogen resistance, we focused our attention on those differentially-expressed genes associated with cell wall biosynthesis and response to biotic stimuli. Fibre quality in flax is associated with both the structure of the cellulose-rich secondary cell wall and the architecture (length, diameter) of individual fibres [16,34,43]. The observation (Figure 7A) that 10 cell-wall related genes were up-regulated in Drakkar, as compared to Belinka is therefore extremely interesting. These genes include 1 gene (cellulose synthase catalytic subunit - c1532) involved in cellulose biosynthesis, 3 genes (xylose synthase - c59577; secondary cell wall-related glycosyltransferase family 47 - c9188; secondary wall-associated glycosyltransferase family 8 D - c7526) involved in secondary cell wall hemicellulose synthesis, 3 genes (phenylalanine amonnia lyase - c59528; caffeic acid o-methyltransferase - c629; coniferyl alcohol 9-O-methyltransferase (c26370), involved in phenylpropanoid/lignin biosynthesis, 1 gene (xyloglucan endo-transglycosylase - c5939) involved in cell wall expansion, 1 gene (cell wall apoplastic invertase - c7453) involved in sucrose partitioning [44], and 1 gene (fasciclin-like AGP 2 - c51183) that has previously been associated with both flax fibre formation [45] and G-fibre formation in tree reaction wood [46,47]. Only 1 cell wall related gene (xyloglucan endotransglucosylase hydrolase - c33660) associated with wall remodeling events was up-regulated in Belinka as compared with Drakkar. The up-regulation of these genes could suggest that cell wall biosynthesis is more dynamic in the Drakkar variety and is possibly related to the different fibre quality characteristic of this variety. However, further functional validation is obviously necessary before confirming such hypotheses.

A total of 14 genes involved in biotic stress response were up-regulated in Drakkar as compared with Belinka, while 9 genes were up-regulated in Belinka as compared with Drakkar (Figure 7B). Interestingly, 4 of the most up-regulated genes (c31122, c59522, c26764, c402) in Drakkar code for cysteine proteases known to play an important role in programmed cell death (PCD) associated with the hypersensitive response (HR) [48,49]. Two other highly up-regulated genes (c7149 and c48400) encode germin-like proteins potentially associated with ROS production and cell-wall cross-linking in defense [50]. One up-regulated gene (Nbi-D - c37826) has been previously shown to be involved in flax resistance to rust (Melampsora lini) [51]. In Belinka, the most highly up-regulated gene (c505) as compared to Drakkar correspond to a hypersensitive-induced response protein.

Overall, these results would suggest that our flax microarray platform is capable of detecting biologically-relevant differential gene expression between contrasted flax varieties. As such, our microarrays represent a powerful tool for identifying candidate genes potentially associated with quality-related polymorphism and therefore represent a valuable contribution to molecular-based plant improvement programs.

5 μg aliquots of total RNA were treated with DNaseI using a TURBO DNA-free Kit (Ambion) according to the manufacturer's instructions, then first-strand cDNA was synthesized using M-MuLV RNase H^- reverse transcriptase (Finnzymes) with 2.5 μg of random hexamers and 500 ng of oligo(dT)12 according to the manufacturer's instructions. The reaction was stopped by incubation at 70°C for 10 min, and the reaction mixture treated with RNaseH (BioLabs) according to the manufacturer's instructions before dilution with 600 μL of sterile de-ionized water.

Transcript levels were assessed by qRT-PCR, in assays with triplicate reaction mixtures (final volume, 20 μL) containing 5 μL of cDNA, 0.5 μM of both forward and reverse primers, and 1X DyNamo Capillary SYBR Green qRT-PCR mix (Finnzymes). qRT-PCR experiments used a balanced randomized block design, as recently advised [58]. A LightCycler (Roche) was used to acquire the CT values for each sample. The following standard protocol was applied for all amplifications: 10 min at 95°C, followed by 45 cycles of 10 s at 95°C, 15 s at 60°C, and 15 s at 72°C. A melting curve analysis was added to each PCR program and the size of PCR products was assessed by electrophoresis in agarose gels. The primer sequences used for all target genes are presented in Additional file 5.

Relative standard curves describing the PCR efficiencies (E) for each primer pair were generated for each amplicon according to Larionov et al. [59]. Normalization of qRT-PCR was performed using reference genes (R) according to Gutierrez et al. [60]. Twelve genes (Additional file 5) were chosen for their stability across the set of microarray experiments previously shown in this paper. Their expression was assessed by qRT-PCR, and they were ranked according to their stability of expression using geNorm software [61]. The_c3168 (lcl|genolin_c3168 628 nt) and c10916 (lcl|genolin_c10916 813 nt) genes were the most stably expressed ones among the 12 tested and, thus, were used to normalize the qRT-PCR data. The normalized expression patterns obtained using both reference genes were similar, so only the data normalized with c10916 (highly similar to a Dehydrodolichyl diphosphate synthase 6 of Arabidopsis thaliana) are shown in this article. Gene expression was calculated using CT and E values with the formula, E_T^(CT_tissueA^-CT_tissueB⁾/E_R(^(CT_tissueA^-CT_tissueB⁾, where (T) is the target gene and (R) the reference gene, (tissueA) is related to cDNA from the tissue showing the lowest expression and (tissueB) from the tissue showing the highest expression. All qRT-PCR results represent means as calculated from the three technical replicates [58].