A generic approach for the design of whole-genome oligoarrays, validated for genomotyping, deletion mapping and gene expression analysis on Staphylococcus aureus

Background DNA microarray technology is widely used to determine the expression levels of thousands of genes in a single experiment, for a broad range of organisms. Optimal design of immobilized nucleic acids has a direct impact on the reliability of microarray results. However, despite small genome size and complexity, prokaryotic organisms are not frequently studied to validate selected bioinformatics approaches. Relying on parameters shown to affect the hybridization of nucleic acids, we designed freely available software and validated experimentally its performance on the bacterial pathogen Staphylococcus aureus. Results We describe an efficient procedure for selecting 40–60 mer oligonucleotide probes combining optimal thermodynamic properties with high target specificity, suitable for genomic studies of microbial species. The algorithm for filtering probes from extensive oligonucleotides libraries fitting standard thermodynamic criteria includes positional information of predicted target-probe binding regions. This algorithm efficiently selected probes recognizing homologous gene targets across three different sequenced genomes of Staphylococcus aureus. BLAST analysis of the final selection of 5,427 probes yielded >97%, 93%, and 81% of Staphylococcus aureus genome coverage in strains N315, Mu50, and COL, respectively. A manufactured oligoarray including a subset of control Escherichia coli probes was validated for applications in the fields of comparative genomics and molecular epidemiology, mapping of deletion mutations and transcription profiling. Conclusion This generic chip-design process merging sequence information from several related genomes improves genome coverage even in conserved regions.


Background
Current hybridization technologies allow assaying thousands of nucleic acid sequences in a single reaction on a solid substrate. Such massively parallel systems offer unprecedented opportunities for basic research and diagnostic applications, including gene sequencing [1], detection of genetic polymorphisms [2], genome-composition analysis [3,4] and measurement of gene expression profiles in prokaryotes [5,6] or cancer cells [7]. Oligonucleotide probes (up to 70-mer) offer more flexibility than cDNA probes since they can be tailored according to optimal in silico physico-chemical and specificity properties, and applied to any sequence data.
Early available probe design software identified sets of probes sharing homogeneous thermodynamic properties for probe-target hybridization [8]. More elaborated software tools include cross-homology testing of probes against a reference database by BLAST (Basic Local Alignment Search Tool) [9,10] or prediction of secondary structures into the thermodynamically-based approach [11][12][13][14]. A frequent drawback of some of these algorithms is to yield an excessive number of unprocessed BLAST outputs that complicates final selection of the most specific probes. Furthermore, these approaches do not take into consideration probe interaction with microarray surface, in particular the impact of mismatches position between the target and probes, as shown by Hughes et al [15]. Designing reliable oligonucleotide probes with available software is quite difficult for bacterial genomes with low GC content [16], low complexity in sequence composition, or frequent conserved repeats leading to erroneous target identification by cross-hybridization.
The reported method (OliCheck) implements an algorithm for filtering oligonucleotide probes libraries sharing homogeneous thermodynamic properties by using positional information of predicted target-probe binding regions. An additional characteristic of OliCheck is to annotate probes recognizing highly conserved targets shared by different genomes. Staphylococcus aureus (S. aureus) was selected as a model organism for implementing and experimentally validating this approach. The choice of this clinically important pathogen for fundamental and applied genomic studies is prompted by the availability of several fully or partially sequenced strain genomes [16][17][18]. A set of feature elements was designed by OliCheck to yield an extensive S. aureus genome coverage. This S. aureus specific probe set together with control probes were used to manufacture an oligoarray that was extensively validated for comparative genomics, molecular epidemiology, mapping of deletion mutations, and transcription profiling applications. The specificity, signal-response linearity, and influence of hybridization temperatures for transcript profiling are also described.
Further genomic oligoarrays of several distinct microbial species have been successfully designed using this generic methodological approach.

Design of a S. aureus oligoarray
The major genomic component of S. aureus is a 2.8 million base pairs (bp) circular chromosome with a low GC content (32.8%) which represents 2,595 ORFs in strain N315. The median length of strain N315 ORFs is 768 bp and the 25 th -75 th percentile ranges from 444-1,152 bp [16]. A probe design software (i.e. ArrayDesigner™) generated a comprehensive list of candidate 40-60 mer probes for each ORF of strain N315 according to the preset thermodynamic hybridization parameters (Fig. 1, step A). A similar process was performed with strain Mu50 and COL. This step yielded 417,776, 607,461 and 321,286 candidate probes for strains N315, Mu50 and COL, respectively.
Further selection of candidate oligonucleotides to sort out cross-reactive probes within the genome of each strain was achieved by OliCheck. This specificity filtering test is a mathematical transposition of Hughes et al observations [15] on the impact of mismatches position between the target and probe with respect to the microarray surface. The occurrence of mismatch(es) in the distal portion (solution end) of the probe leads to a strong decrease in signal intensity, as opposed to the proximal portion (surface end). The results of the specificity filtering test, when performed separately on each strain, yielded a list of 48,415, 48,510, and 34,303 probes for N315, Mu50 and COL, respectively ( Fig. 1, step B). In contrast to OliCheckfiltered probes that are expected to be devoid of crosshybridization, >85% of the probes selected by ArrayDe-signer™ alone displayed significant cross-hybridization with multiple ORFs.
For each strain, all accepted probes were further annotated by OliCheck against heterologous S. aureus genomes to identify probes common to the different genomes ( Fig. 1, step C).
To fulfill the manufacturing requirements of a S. aureus genome-wide oligoarray, a further probe selection was performed. This selection used a spreadsheet program to rank probes according to optimal strain coverage and thermodynamic criteria, for providing one or more nonoverlapping probes per target ORF (Fig. 1, step D).

In silico properties of the S. aureus oligoarray and manufacturing of StaphChip
The final set of 5,335 S. aureus OliCheck-filtered probes recognized 97.5, 93.0, and 81.0% of N315, Mu50, and COL ORFs, respectively. The low residual percentage of Schematic representation of StaphChip probe selection Figure 1 Schematic representation of StaphChip probe selection. All ORFs of N315 were loaded into ArrayDesigner™ (a) to select oligonucleotides according to their thermodynamical properties (Step A). The 417,776 N315-derived probes were filtered for target specificity using BLAST against N315 genome (Step B). Each probe should recognize a single target yielding a defined signal intensity threshold, i.e. outside the green box (b), otherwise it is rejected (c). During Step C, all accepted probes are aligned against heterologous S. aureus genomes (i.e. Mu50 or COL) to annotate probes common to the other genomes (d, e). Each probe should recognize a single target yielding a defined signal intensity threshold, i.e. inside the red box (d), with no other signals outside the green box; otherwise it is ignored (e). The process is repeated from Step A to C with the two other strain databases. Final selection by spreadsheet analysis (Step D) yielded a total number of 5,427 probes hybridizing with one or more S. aureus genomes.
Step C Step B Step D

5'427 Probes
Best probe selection Step C Step B Step D

5,427 Probes
Best probe selection ORFs (2.5% for strain N315 and 7.0% for Mu50) that escaped recognition by our final probe set mostly included (51/65 ORFs for N315) mobile genetic elements, located either on prophage or transposon elements. A likely explanation for exclusion of these targets by OliCheck is the occurrence of highly homologous (>98% nucleotide identity) sequence repeats found in other ORFs. Accordingly, 92 residual probes covering these ORFs had to be selected relying on step A ( Fig. 1) only.
To manufacture StaphChip, a total of 5,427 S. aureus probes were synthesized on the array together with a subset of control probes designed by OliCheck against E. coli K12 genome [19] (see methods).

Comparison of OliCheck with a currently used method
To validate the properties of OliCheck by comparison with an established method, we generated 60-mer oligonucleotides probes with homogenous thermodynamic criteria for N315 genome by the recently published Array-OligoSelector [13] tool. The final set of probes generated according to the ArrayOligoSelector procedure [13] (n = 2,592) was further tested for cross-homology by using OliCheck. A large percentage of these probes (83%) were sorted out by OliCheck because they failed specificity criteria defined by Hughes et al [15].

Comparison of in silico-predicted with experimentally detected hybridization signals
Among the total number of 5,427 S. aureus probes, 4,812 (89%) recognized targets common to all three strains (Fig.  2), thus affording StaphChip valuable properties for comparative genomic and transcriptomic studies on S. aureus. The finding of a significant number of probes (n = 401) commonly identified in N315 and Mu50 but not COL may be explained by their closely-related genomic contents [16].
The number of probes predicted to detect genomic elements from each S. aureus strain was compared to those experimentally recorded using StaphChip (Table 1). For each strain, >99.5% of the in silico-predicted hybridization signals were indeed detected by hybridization of genomic DNA on StaphChip probes. Most of the false-positive signals were recorded on probes that were generated using ArrayDesigner™ only (n = 92/104). In silico analysis determined that these false-positive signals did not originate from recognition of intergenic regions (data not shown). Thus, such false-positive frequency is not transferable to the whole array.

Use of StaphChip for deletion mapping and genomotyping applications
To evaluate the accuracy of StaphChip for deletion mapping, the Cy-3 labelled DNA from the SA113ica deletion mutant [20] was competitively hybridized with the Cy-5 labelled DNA from its isogenic parent. Figure 3 maps the ica-related signals which are missing in the ica mutant, as opposed to the unique positive signal generated by the tetracycline resistance marker used for the construction and selection of strain SA113ica. Fluorescent intensities of all other signals except ica-related genes were equivalent for both strains.
The potential of StaphChip for epidemiological typing was analysed by comparative genomic hybridization (CGH) of S. aureus strains from various epidemiological origins. Genomic DNA from each individual S. aureus strain labelled with Cy3 was co-hybridized with equivalent amounts of Cy5-labelled control genomic DNA (pooled from N315, Mu50 and COL) and analyzed by two-way hierarchical clustering (Fig. 4).
The recently sequenced community-acquired MRSA strain MW2 [17], not included in StaphChip probe design, revealed important differences with strains N315, Mu50, and COL, as shown by a major yellow region on Figure 4. The set of probes yielding negative signals with MW2 DNA (Fig. 4, cluster a) corresponds to ORFs coding for In silico specificity of selected probes  Mapping of a deleted gene region by StaphChip Figure 3 Mapping of a deleted gene region by StaphChip. Cy-5 labelled DNA of parental strain SA113 was co-hybridized with Cy-3 labelled DNA from its isogenic ica deletion mutant. Colored bars indicate the position of each probe used to map icarelated and adjacent ORFs. Background signals (green) were recorded from probes recognizing the ica-region known to be deleted in strain SA113ica (arrows), as opposed to positive red signals recorded in the wild-type strain. The tetracycline resistance marker used for the construction and selection of strain SA113ica is recorded in the green channel only. Dye swap experiments provided similar results (not shown). Data are raw signal intensities; background level is indicated by a dotted line. ica deleted strain Wild type strain antibiotic and heavy metal resistance determinants, bacterial adhesins and DNA modification enzymes lacking in this epidemic strain [17]. In contrast, extensively conserved genomic regions (Fig. 4, cluster b) are composed of house-keeping genes contributing to cell-wall, DNA synthesis, as well as essential metabolic enzymes.

Ica R Ica
Further analysis of strain SA113 and SA113ica compared to the other strains, showed no hybridization signals (Fig.  4, cluster c) for a number of genes (e.g. exotoxins and antibiotic resistance determinants) known to be missing in pathogenicity islands I-II-III of the NCTC8325 family [21].

Application of StaphChip for expression profiling studies
Control experiments were performed to study the doseresponse of labelled cDNA and influence of hybridization temperature on the linearity and intensities of fluorescent signals. Two portions of N315 cDNA were labelled during reverse-transcription, with either Cy-3 or Cy-5 and competitively hybridized on StaphChip. Characteristics of fluorescent signals obtained on N315 specific probes were compared to those recorded on control E. coli oligonucleotide probes, at either 55 or 60°C. The median level of fluorescence intensities were approximately 5-10 fold higher for S. aureus probes compared to E. coli probes.
On S. aureus capture elements, log-transformed signal intensities recorded with equivalent input amounts (5 or 10 µg) of Cy-3 and Cy-5 cDNA were highly correlated (R > 0.95). Linear regression of Cy-3 versus Cy-5 scatter plots yielded slopes from 0.94 to 1.02 at 55 or 60°C (Table 2). When 5 µg of Cy-3 was competitively hybridized with 10 µg of Cy-5 labelled cDNA, slopes from 0.48 to 0.53 were recorded at 55 or 60°C, as expected (Table 2). Since logtransformed signal intensities remained highly correlated (R>0.85), this assessed the robustness of the recorded signals. Furthermore, the median intensity of fluorescent signals was marginally altered by temperature changes (not shown).
In contrast, signal intensities from E. coli control probes were highly disturbed by temperature changes or by altering the ratios of fluorescently-labelled S. aureus cDNA (Table 2). Furthermore, the median intensity of fluorescent signals decreased by >2 fold at 60 compared to 55°C (not shown). Dye-swap experiments yielded equivalent results (not shown).

Reproducibility of StaphChip for expression profiling
To evaluate the reproducibility of fluorescent signals (Fig.  5), eight independent hybridization experiments were performed under identical conditions using Cy-3 labelled Comparative genome hybridization using clustering analysis Figure 4 Comparative genome hybridization using clustering analysis. Genomic DNA of each individual S. aureus strain was labelled with Cy3 and co-hybridized with equivalent amounts of Cy5-labelled genomic DNA pooled from N315, Mu50 and COL. Background-subtracted data were expressed as Log 10 ratios and analyzed by two-way clustering using GeneSpring 6.1. Probes yielding positive and negative signals are shown in blue and yellow, respectively. The significance of black bars (a, b, and c) is indicated in the text. Note that the figure resolution does not allow visualising single probe differences but only clusters of probes.

N315
Mu50 N315 cDNA, derived from overnight cultures. Maximal relative errors [(average-min)/average] of fluorescent probe signal intensities were more widely scattered on E. coli (Fig. 5B) compared to S. aureus (Fig. 5A) capture elements, thus reflecting the instability of poorly specific interactions with S. aureus targets and E. coli probes. The 90 th percentile of maximal relative errors from S. aureus probes represented <25% of average signal intensities. In contrast, the same percentile of maximal relative errors recorded from E. coli probes represented >100% of average signal intensities. Furthermore, the 50 th percentile on E. coli probes was superior to the 90 th percentile of maximal relative errors on S. aureus probes.
Finally, we compared signal intensities generated from individual gene transcripts covered by two or more adjacent but non-overlapping S. aureus probes. Figure 6 shows average fluorescence intensities and their maximal relative errors (A) and the cumulative distribution of maximal relative errors (B) For nearly 90% of the gene transcripts (n = 2,269), maximal relative errors of ORF-related signals were <60%. This yields evidence that multiple probes recognizing the same transcripts provide reproducible signals and that StaphChip provides reliable and robust determinations of genome-wide transcripts. Figure 7 shows the fold changes estimated by quantitative RT-PCR for 18 transcripts found to be either up regulated, down regulated or unchanged on the StaphChip. A good fitting was observed between both methods, with a correlation coefficient of 0.91 obtained from linear fitting. Of note, the quantitative changes recorded with RT-qPCR tended to be higher than those quantified with the microarray. This finding is supported by previous observations that the dynamic range of RT qPCR is higher than that of microarray [22,23], that only reach 2-3 orders of magnitude. These data validate the use of our method for quantitative gene-expression analysis.

Discussion
Several recent studies have shown the usefulness of microarrays for genomic and global transcriptomic studies of microbial pathogens [23][24][25]. Each step of microarray experiments needs to be optimized and validated, from array design and manufacture to data collection and analysis. Among critical technical parameters that need to be controlled are microarray surface chemistry, probe sequence, probe deposition process, and hybridization conditions. Accuracy of microarray-generated data can be improved by using multiple replicates, dye swaps and statistical data analysis. Compared to cDNA microarrays, oligoarrays provide a flexible design and are considered more reliable in terms of sensitivity and specificity [26][27][28][29]. Reported drawbacks of cDNA or PCR probes include: i) unpredicted secondary structures, ii) uncontrolled cross-hybridization occurring on repeats or partially homologous regions of PCR probes, and iii) varying amounts and purity of spotted products [27,30]. Recent software development allows genome-wide selection of sub-sequences that uniquely identify genes. Ideally, these approaches should amplify fragments of constant length, thus minimizing the differences in PCR amplification efficiency as well as in hybridization kinetics [31]. However, the extent of cross-hybridization has rarely been evaluated and reported, and thus may lead to severe errors in higher level data analysis, such as clustering [32], genome composition analysis and genotyping for molecular epidemiology [4].
Most oligoarray applications dedicated to prokaryotes were developed by companies using proprietary algorithms [33,34] whose detailed properties are rarely available. Furthermore, the lack of published validation data prevents adequate comparison of those short-probe oligoarrays with investigator-designed oligonucleotide arrays.
To date, several strategies for oligoarray design have been described, but their experimental validation is often  [35] or absent [36]. A drawback of these approaches is to apply thermodynamics laws on probe/ target interaction as determined in solution [37], but ignoring effects related to oligonucleotide immobilization on a microarray surface [15,38].
To address this issue, OliCheck considered the influence of predicted probe/target binding with respect to its position along the immobilized probe, as demonstrated by Hughes et al [15]. This tool allowed selecting probes from large oligonucleotide libraries in order to provide multiple genome coverage, suitable for epidemiological and transcriptomic studies.
As a specific application, OliCheck was used to design StaphChip, an oligoarray dedicated to genomic studies on S. aureus, a clinically important pathogen with a low GC content and numerous sequence repeats. The 5,427 feature elements were selected to ideally cover all ORFs of three S. aureus genomes with two non-overlapping probes, as validated under experimental conditions. A particular achievement of this strategy is to yield crossannotations between the designed probes and homologous ORF regions conserved across several genomes. Any new genome sequence can be screened by OliCheck to identify probes that can specifically detect homologous ORFs. Cross-annotations on the recently released S. aureus MW2 genome [17] yielded 78% gene coverage. It should be mentioned that the recently published S. aureus COL genome [39] composition and annotation have changed significantly since the early release by TIGR in 2003.

Reproducibility of fluorescent signals in replicate experiments
The properties of StaphChip design were confirmed by comparative genome hybridization and global gene expression studies. Work in progress assesses the reliability of StaphChip for monitoring S. aureus transcription changes during biofilm formation, endocytic stage, or expression of antibiotic resistance. Another achievement (unpublished data) was the design of oligonucleotide probes for the genomes of Neisseria meningitidis A and B, E. coli K12, Erwinia carotovora and E. chrysanthemi; having GC contents ranging from 32.8 to 52%. Furthermore, Oli-Check design expands oligoarray use for the study of hostpathogen interactions by potentially preventing crosshybridization between bacterial probes and contaminating host nucleic acids.

Conclusion
In summary, this work describes a validated approach to select optimal oligoarray capture elements for S. aureus expression analysis and comparative genome hybridization studies. This generic approach will enable researchers to develop customized oligoarrays for genomic studies of any sequenced microorganism.
Reproducibility of fluorescent signals recorded from multiple non-overlapping capture elements for common transcripts

Design of specific oligonucleotide probes
Step A An initial set of candidate oligonucleotide probes was generated by ArrayDesigner™ 1.17 (Premier Biosoft Intl) using the following probe parameters: (i) 40-60 bp probe length, (ii) 65 ± 10°C target Tm, (iii) <-5.0 kcal/mol for hairpins, (iv) <-8.0 kcal/mol for self-dimers, and (v) dinucleotide repeats shorter than 5 bp (Supplementary material provides OliCheck input format description, for using other probe design software or probe list  (Fig. 1a).
Further selection of specific oligonucleotide probes was performed by the design of an original program called OliCheck. This approach is derived from the experimental findings of Hughes et al. that microarray hybridization signals are mostly influenced by mismatches in the solutionend (distal part) rather than surface-end (proximal part) portion of the oligonucleotide probe [15]. OliCheck queries the locally available S. aureus genome databases by performing a BLAST (BLASTN for Windows, version 2.2.2) [9] analysis for each probe.
Step B A first probe selection is performed by aligning each probe against its own genome, e.g. N315 candidate probes against N315 genome (Fig. 1b, and 1c).
Each BLAST output is analyzed to extract alignment information. The best scored target-probe alignment (perfect match) is considered as the target of the probe, therefore exempted from further testing. To avoid cross-hybridization, all other alignments are tested using the specificity filtering test. For this test, any 60-mer probe that would align to the target with >11 matched nucleotides by its distal third (solution-end), or >31 matched nucleotides by its proximal two-thirds (surface-end) is rejected. An additional condition is the exclusion of irrelevant target-probe pairs with >18 consecutive nucleotide matches.
Step C Further probe selection is achieved by aligning each candidate probe from each genome (e.g. N315) against the genomes of the other S. aureus strains (e.g. Mu50 and COL) (Fig. 1d, and 1e). This process allows annotating probes detecting homologous targets in the other genomes.
Each BLAST output is analyzed to extract alignment information. The best scored target-probe alignment is tested to predict a high hybridization signal. This efficiency test requires the absence of any mismatch in the distal half and <20 mismatches in the proximal half of the probe. If those requirements are not fulfilled, the probe is rejected; otherwise the alignment is considered appropriate for detecting a homologous target and the process is continued. To avoid cross-hybridization with targets from other genomes, further alignments obtained with that probe are checked by the specificity filtering test, as defined in step B. Each probe fulfilling these requirements is annotated as detecting a unique homologous sequence target.
Step D Using a spreadsheet program, the probes were sorted by their ORF target names and their best thermodynamic criteria. Probes showing the best combination of thermodynamic criteria and strain coverage were selected for microarray manufacturing (Fig. 1f)

In silico comparison of our algorithm with ArrayOligoSelector
Three probes set of 60-mer probes with homogenous thermodynamic criteria (Tm = 60°C) were generated using default parameters for N315 genome by: i) ArrayDesigner ii) ArrayOligoSelector for all candidate probes, iii) Array-OligoSelector for the best candidate probes (one per ORF). The output lists generated by ArrayOligoSelector were reformatted to match OliCheck input file format. The list of probes generated by either software was further processed by OliCheck for cross-homology filtering. The sets of probes selected by each method were further compared for homology using BLAST. Alignment showing E-value <1E-20 were considered as homologous.

Microarray manufacturing
The StaphChip microarray was manufactured by in situ synthesis of 8,455 long oligonucleotide probes (Agilent). It consists of 5,427 S. aureus and 2,873 E. coli specific probes, together with A. thaliana control probes for spiked controls.

Preparation of the labelled nucleic acids
For comparative genome hybridization, each S. aureus strain was grown overnight in 2 ml Mueller-Hinton broth (MHB) and total DNA was extracted and purified using DNeasy columns (QIagen) following manufacturer's instructions. DNA purity and concentration were assayed by spectrophotometer. 2 µg DNA were labelled by the Klenow fragment of DNA polymerase I (BioPrime, Invitrogen) with Cyanine-3 or Cyanine-5 coupled dCTP (NEN) for 2 hours at 37°C, then stopped by the addition of 5 µl 0.5 M EDTA. Labelled DNA was purified on QiaQuick columns (Qiagen).
For gene expression analysis, total RNA was extracted from 2 ml exponential or overnight cultures using the Rneasy kit (Qiagen) as previously described [41]. Batches of 5 or 10 µg of total S. aureus RNA were spiked with increasing amounts of different Arabidopsis thaliana mRNAs (SpotReport, Strategene), used as external calibrators. The RNA mixture was labelled by Cy-3 dCTP or Cy-5 dCTP, using the SuperScript II (Invitrogen) following manufacturer instructions. Labelled cDNA was then purified onto QiaQuick columns.

Hybridization and scanning parameters
Unless specified, equivalent amounts of cDNA (or genomic DNA) labelled with Cyanine-3 or Cyanine-5, were diluted in 250 µl Agilent hybridization buffer, and hybridized at a temperature of 60°C for 17 hours in a dedicated hybridization oven (Robbins Scientific). For comparative genome hybridization, genomic DNA from each individual S. aureus strain was labelled with Cy3 and cohybridized with equivalent amounts of Cy5-labelled genomic DNA pooled from N315, Mu50 and COL [42]. Slides were washed, dried under Nitrogen flow and scanned (Agilent) using 100% PMT power for both wavelengths. Data were extracted and processed using Feature Extraction™ software (version 5.0, Agilent).
For gene expression analysis, saturated spots were excluded from subsequent analysis. Local backgroundsubtracted signals were corrected for unequal dye incorporation or unequal load of labelled product. The algorithm consisted of a rank consistency filter and a curve fit using the default LOWESS (locally weighted linear regression) method. Spots showing a reference signal lower than background plus two standard deviations were also excluded from subsequent analyses.
For comparative genome hybridization, local background-subtracted data were expressed as Log 10 ratios and analyzed by two-way clustering using GeneSpring 6.1 (SiliconGenetics).

Real time, quantitative PCR analysis
The expression of 18 genes involved in metabolic pathways was quantitatively assayed using by 1 step RT-qPCR using a SDS7700 (Applied Biosystems, Framingham, MA). Primers and probes were identified by scanning each gene sequence using the software Primer Express 2.0(Applied Biosystems). All identified sequences were further aligned on the whole genome of sequenced strains to ensure gene specificity and conservation of the target sequence between strains. Optimal concentration of primers and Taqman probes (labelled with FAM in 3' and coupled to TAMRA in 5' as quencher and purchased from Eurogentec, Seraing, Belgium), determined accordingly to manufacturer's instructions were 200 nM and 100 nM per reaction, respectively. Primers and probes were mixed in Platinum qRT-PCR Thermoscript kit (Invitrogen) with 0.4 ng of total purified RNA. Fold changes were calculated after normalization with the expression level of the 16s rRNA gene as previously described [41].