The extended probe set (63 oligonucleotides targeting gyrB gene of 24 most common food‐ and water‐borne pathogens and indicator organisms at the species and/or genus level) was validated using pure cultures of the reference strains. For microarray validation each strain was hybridized separately onto the microarray. All labelling reactions were performed using a mixture containing all reverse complement (RC) and competitive (CO) oligonucleotides (Appendices S1 and S2), therefore eliminating all predicted false positive results. This validation revealed only one false positive signal remaining on the array (Cam_2221 probe giving signal with Enterococcus faecalis target; weighted mismatch value 2.4). Design and experimental validation of the additional CO oligonucleotides targeting this probe–target pair are planned before further application of the microarray. The summary of validation data is presented in Fig. 1.

Interesting findings were made regarding previously developed V. cholerae‐specific probes (Vch_1776, Vch_1795 and Vch_1839; Kostićet al., 2007). Hybridization with DNA of the closely related V. fluvialis Sas27 and V. mimicus Sas23 isolates yielded a positive signal with three and two of these probes respectively. Subsequent comparative phylogenetic analysis of gyrB sequences of these species and Vch probes revealed that these probes were in fact also perfect‐match probes for V. fluvialis and V. mimicus species. Detailed analysis of the gyrB alignment of these two strains revealed that there was not a single position left where a discriminating probe could be designed following previously described criteria (3′ terminal cytosine residue).

In order to ascertain absolute sensitivity (lowest number of detectable cells) of the newly developed microarray an overnight culture of Salmonella spp. was serially diluted and plated onto TSA‐Y agar [from 4.7 × 10⁸ colony‐forming units (cfu) ml⁻¹ down to 0.47 cfu ml⁻¹]. Our results showed that the absolute sensitivity of the microarray is approximately 10⁴ cfu. In addition, relative sensitivity (lowest abundance of targeted pathogen in non‐targeted background) of the microarray was tested by artificial contamination of water samples. The latter were prepared and analysed in parallel using microarray and fluorescence in situ hybridization (FISH) analysis. Detailed description of the spiked samples and an overview of the results are summarized in Table 1. Samples A to C harbouring only Salmonella spp. were used as a positive control and for initial method optimization. Relative sensitivity of the microarray detection was demonstrated by the analysis of the spiked samples F to H. These samples harboured low amounts of targeted bacteria (Escherichia coli) spiked in a high amount of non‐targeted bacteria (Burkolderia phytofirmans strain PsJN). As predicted, and confirmed by FISH, the relative abundance of E. coli in these samples was in the range of 0.04–0.4% (Table 1– shaded values). Microarray analysis of these samples exhibited clearly detectable signals for all four E. coli‐specific probes (Fig. 2A).

Reliability of parallel detection was shown with spiked samples D and E. All four pathogens that were spiked in the sample were successfully detected by the microarray (Fig. 2B). Quantification of relative abundance of each bacterium was done via FISH by performing four independent hybridizations for each sample always using a strain‐specific probe labelled with Cy3 and EUB338 probe mix labelled with fluorescein (Table S4). In general, all strain‐specific probes exhibited significantly weaker signals than the universal EUB338 probe mix. Staphylococcus aureus was not detectable using a specific probe (Sau), probably due to the too strong fixation (4% paraformaldehyde (PFA) solution, fixing agent for Gram‐negative cells, was used on all filters; Gram‐positive S. aureus would require ethanol as fixing agent). However, summarized relative abundances of the other three pathogens (established by using Daime software) corresponded to 90% of the total biovolume (represented by EUB338 stained cells).

In order to demonstrate the applicability of this microarray system for the detection and identification of food‐borne pathogens, an extensive validation was performed using artificially and naturally contaminated food samples. Microarray results (summarized in Table 2) showed a high degree of reproducibility between replicate spike sets as well as good correlation to the results obtained by conventional microbiological analysis (data not shown). Salmonella spp. and Listeria monocytogenes were detectable at the level of 5 cfu per 25 g food after specific enrichment (RVS and Fraser respectively). The detection limit for Yersinia enterocolitica was found to be > 10 cfu per 25 g food (inoculum of 10 cfu per 25 g food could not be detected; inoculum of 30 cfu per 25 g food was clearly detectable). The presence of Campylobacter spp. in the spiked samples was confirmed by three probes (Cam_1556, Cam_2027 and Cam_2221). Ten cfu per 25 g food gave unambiguous positive signals on the microarray; lower amounts were not tested. Furthermore, Campylobacter jejuni‐specific probe Cje_2000 enabled unambiguous differentiation of C. jejuni and C. coli on the microarray.

In order to further test the application potential of the microarray for food analysis and to demonstrate the advantages of its multiplexing feature, a minced meat (pork and beef mix) matrix was spiked in parallel with S. Typhimurium DSM554, L. monocytogenes SLCC 2755 and Y. enterocolitica NCTC10460. Minced meat is a very complex and challenging matrix containing a native bacterial flora comprising, among others, Proteus sp., Aeromonas sp., Citrobacter sp. and Y. enterocolitica (as determined by microarray analysis of non‐spiked samples after pre‐enrichment in buffered peptone water; data not shown). Spiked meat samples were submitted to universal pre‐enrichment (BPW for 18 h at 37°C) followed by selective enrichment (RVS, Fraser and ITC broths for Salmonella enterica, L. monocytogenes and Y. enterocolitica respectively; for 48 h at 37°C or 25°C). Salmonella enterica was the only spiked microorganism that could be detected already after buffered peptone water pre‐enrichment. This was true for both microarray and classical microbiological (selective agar plates) analysis. Accordingly, Salmonella spp. specific signals could also be obtained from RVS enrichment. Listeria monocytogenes could only be detected after selective enrichment in Fraser broth. Correspondingly to the results obtained in the single spiking experiments, the detection limit for both Salmonella spp. and L. monocytogenes was 1–10 cfu per 25 g food. Even though Y. enterocolitica was present in the matrix as part of the native matrix flora it was suppressed below detectable levels during the course of biological enrichment. Both microbiological and microarray results for Yersinia spp. were negative.

The previously established gyrB sequence database (Kostićet al., 2007) was extended by downloading new gyrB sequences from the NCBI database (http://www.ncbi.nlm.nih.gov) as well as by sequencing gyrB genes of strains used for microarray validation. Probes were designed using the ARB software package (Ludwig et al., 2004) following already established and validated probe design criteria: (i) 3′ terminal cytosine residue, (ii) placement of the diagnostic mismatch(es) as close to the 3′ end of the probe as possible, (iii) similar melting temperature (targeted 60°C), and (iv) probe length between 17 and 28 nucleotides (Kostićet al., 2007). CalcOligo 2.03 (Stralis‐Pavese et al., 2004) was used to create an Excel table indicating predicted melting temperatures (based on the nearest neighbour model and SantaLucia (SantaLucia et al., 1996) parameters), length and GC content of the probes and the number of weighted mismatches between each probe–target pair. Nearest neighbour Tm values were calculated with concentration settings of 250 nmol for oligonucleotide and 50 mmol for Na⁺. Factors for weighing mismatches in CalcOligo were as follows: positions: 5′ 1st 0.3; 5′ 2nd 0.6; 5′ 3rd 0.8; 3′ 1st 4.0; 3′ 2nd 2.0; 3′ 3rd 1.2; all other positions 1.0. Base pairs: AC 1.2; TC 1.2; GU 0.7; TG 0.4; all other mismatched base pairs 1.0 (where first and second nucleotides refer to probe and target sequence respectively). Probe–target pairs with weighted mismatches values up to 0.5 were expected to yield positive hybridization under the conditions applied. Furthermore, based on the results obtained in the preceding study (Kostićet al., 2007), additional CO oligonucleotides were designed in order to suppress potential false positive signals. Probe Msi_294 targeting pmoA gene of Methylosinus trichosporium OB3b was used as control oligonucleotide (Stralis‐Pavese et al., 2004). A complete list of the RC oligonucleotide probes and competitive oligonucleotides used in this study can be found in Tables S1 and S2.

Oligonucleotides for immobilization were custom‐synthesized (VBC Genomics, Vienna, Austria) with a 5′ primary amino group, followed by a C6 spacer and five thymidine residues preceding the probe sequence. A 384‐well flat bottom plate (Ritter, Schwabmünchen, Germany) was prepared with 30 µl of 50 µM oligonucleotide solution in 50% DMSO spotting buffer. Microarrays were spotted with an OmniGrid spotter (1 TeleChem SMP3 pin) at 45% relative humidity and 21°C onto silylated slides containing aldehyde groups (CEL Associates, Pearland, TX, USA). Arrays were spotted in triplicates to allow a statistical correction of errors. Slide processing was carried out as described before (Stralis‐Pavese et al., 2004). Processed slides were stored desiccated at room temperature in the dark.

Strains used for the validation of the new probe set are listed in Table S3. Strains were obtained from the University of Sassari and the German Collection of Microorganisms and Cell Cultures (DSMZ GmbH, Braunschweig, Germany). DSMZ strains were cultivated according to the DSMZ (http://www.dsmz.de) users instructions. Total genomic DNA was isolated from the cultures that were homogenized using 0.1 mm glass beads (BioSpec Products, Bartlesville, OK, USA) and a bead mill homogenizer (Retsch, Haan, Germany) followed by standard organic extraction with phenol and chloroform (Sambrook and Russel, 2001). Nucleic acids were precipitated by the addition of 0.7 volumes of 2‐propanol and 0.1 volumes of 3M sodium acetate (30 min at −20°C), pelleted by centrifugation, rinsed with 70% ethanol, air‐dried and finally dissolved in 50 µl TE buffer containing RNaseA (1 µg µl⁻¹).

The gyrB gene was amplified using a mixture of universal primers UP1, UP1G, UP2r and UP2Ar (Yamamoto and Harayama, 1995). PCR reactions were performed in 100 µl aliquots using the FailSafe PCR PreMix E (Epicentre, Madison, WI, USA) and 4 U Taq DNA polymerase (Invitrogen, Carlsbad, CA, USA) with 50–100 ng DNA as template. Amplification conditions were 95°C for 5 min, followed by 35 cycles of: 1 min at 95°C, 1 min at 58°C, 2 min at 72°C, followed by a final elongation step of 10 min at 72°C (Kostićet al., 2007). PCR products were subsequently purified using a commercial PCR purification kit (QIAGEN, Hilden, Germany) according to manufacturer's instructions and eluted in 25 µl dH₂O. DNA concentration was measured using the NanoDrop spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA) and adjusted to an end concentration of 50 ng µl⁻¹.

gyrB PCR products were cloned using the TOPO TA Cloning Kit for Sequencing (Invitrogen, Carlsbad, CA, USA) and sequenced using M13 primers as described before (Kostićet al., 2007). Sequences were analysed using Sequencher v 4.5 (Gene Codes Corporation, Ann Arbor, MI, USA) and subjected to preliminary nucleotide – nucleotide blast analysis within the NCBI database (http://www.ncbi.nlm.nih.gov/blast). Accession numbers of obtained gyrB sequences are listed in Table S3. Subsequently, sequences were imported into the gyrB database in ARB where phylogenetic analysis was performed.

During each labelling and hybridization experiment a pmoA PCR product from Methylosinus trichosporium OB3b was included as an internal positive control and subsequently used for normalization of the results. Genomic DNA extraction and pmoA PCR amplification were performed as described before (Bodrossy et al., 2003). Amplicons were purified using a commercial PCR purification kit (QIAGEN, Hilden, Germany) according to manufacturer's instructions. DNA concentration was measured using the NanoDrop spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA) and the concentration was adjusted to 50 ng µl⁻¹.

For the SSELO a modified protocol of Rudi and colleagues (2003) was applied. After purification, PCR products were treated with shrimp alkaline phosphatase (SAP) (Roche Diagnostics, Penzbeg, Germany) in order to dephosphorylate remaining nucleotides. Samples consisting of 20 µl purified PCR product (50 ng µl⁻¹), 2 µl Thermo Sequenase reaction buffer (Amersham Biosciences, Piscataway, NJ, USA) and 4 µl SAP (1 U µl⁻¹) were incubated at 37°C for 30 min, followed by 10 min at 95°C to inactivate enzyme activity. The SAP‐treated gyrB PCR products were then used as a template for the cyclic labelling reaction. The control pmoA PCR product was further diluted to an end concentration of 5 ng µl⁻¹. For the SSELO approach a set of RC oligonucleotides (see Table S1), lacking the 3′ terminal cytosine residue, was custom synthesized (VBC Genomics, Vienna, Austria). Furthermore, a set of CO oligonucleotides (Table S2) was synthesized with a 3′ phosphate modification. Lyophilized oligonucleotides were dissolved to an end concentration of 100 pmol µl⁻¹ and stored at −20°C. For labelling reaction a mix of oligonucleotides, containing each RC and CO oligonucleotide at an end concentration of 1 pmol µl⁻¹, was prepared.

The cyclic labelling was performed in 10 µl aliquots consisting of 1× Thermo Sequenase reaction buffer, 10 ng SAP‐treated control pmoA PCR product, 1 pmol of each reverse complement oligonucleotide, 10 pmol Tamra‐ddCTP (PerkinElmer Life and Analytical Sciences, Boston, MA, USA), 10 pmol of each ddATP, ddTTP, ddGTP (Roche Diagnostics GmbH, Penzbeg, Germany), 3 U Taq DNA polymerase (Invitrogen, Carlsbad, CA, USA) and 100 ng of SAP‐treated gyrB PCR product. Reaction conditions were 25 cycles of 30 s at 95°C followed by 75 s at 60°C, carried out in a thermocycler. After cyclic labelling, samples were used directly for hybridization, without further purification.

Hybridization was carried out as described before (Stralis‐Pavese et al., 2004). Labelled targets (10 µl) were mixed with hybridization buffer (pre‐warmed to 65°C). Final concentration of the hybridization buffer was: 6× SSC, 1× Denhardt's reagent (Sigma, St Louis, MO, USA), 0.1% SDS. Hybridization was performed for 2 h at 55°C. After hybridization, slides were washed in a 2× SSC, 0.1% SDS wash solution for 5 min, followed by two wash cycles for 5 min in 0.2× SSC, and a final wash for 5 min in 0.1× SSC, all at room temperature. Slides were dried with an oil‐free air gun and scanned immediately.

Microarrays were scanned at 10 µm resolution using a GenePix 4000B laser scanner (Axon Instruments, Foster City, CA, USA). PhotoMultiplier Tube (PMT) gain was adjusted to scan the spots below the saturation level. Scanned images were saved as multilayer tiff images and analysed with the GenePix Pro 6.0 software (Axon Instruments, Foster City, CA, USA). Microsoft Excel was used for statistical analysis and presentation of the results. Microarray hybridization results were normalized to the signal obtained from the internal control oligonucleotide (Msi_294) and expressed as percentage, 100% equalling the signal of the control probe. Non‐specific signals reached in some cases a value of up to 8% (of the control signal, Msi_294). Therefore, a cut‐off value of 10% was chosen to allow for unambiguous differentiation of positive signals.

In order to validate the above‐described experimental set‐up and to confirm the detection threshold, spiking experiments were performed. Cultures of E. coli DSM 5313, S. enterica DSM 17058, P. aeruginosa DSM 50071, S. aureus ATCC 6538 and B. phytofirmans strain PsJN DSM 10436 were grown in 10% tryptic soy broth (Merck KGaA, Darmstadt, Germany) overnight at 37°C with agitation. Overnight cultures were characterized by measuring the optical density (OD) at 600 nm. Additionally, the cfu per ml were determined by plating serial dilutions on 10% Tryptic Soy Agar plates (Merck KGaA, Darmstadt, Germany). After incubation of the TSA‐Y agar overnight at 37°C the colonies were counted. For each sample 2 l of sterile 0.9% NaCl solution was prepared and spiked with a defined volume of one or more cultures. Overview of the spiked samples is presented in Table 1.

For DNA extraction, microorganisms were harvested from 1 l aliquots by filtration through 0.22 µm pore size GS filters (Millipore Corporation, Billerica, MA, USA). Following filtration filters were thoroughly rinsed with 10 ml 0.9% NaCl solution and cells were pelleted by centrifugation. Pellets were resuspended in 800 µl supernatant and suspension was homogenized using 0.1 mm glass beads (BioSpec Products) and bead mill homogenizer (Retsch). Genomic DNA was extracted following standard protocol for organic extraction with phenol and chloroform (Sambrook and Russel, 2001).

For the FISH analysis 500 ml aliquots of the spiked water samples were filtered through 0.22 µm pore size polycarbonate filters (Millipore Corporation). After filtration, all fixation and washing steps were performed as described before (Loy et al., 2005) and filters were stored at −20°C before in situ hybridization. Probes used for FISH analysis were selected using probeBase (Loy et al., 2003) and are listed in Table S4. Labelled probes were purchased from Thermo Electron GmbH (Ulm, Germany). Lyophilized oligonucleotides were dissolved to an end concentration of 300 ng µl⁻¹ and stored at −20°C. For in situ hybridization working solutions of 30 ng µl⁻¹ were prepared (with exception of EUB338 probe mix that was used at 50 ng µl⁻¹ concentration). In situ hybridization on filters was performed following a previously published protocol (Glöckner et al., 1996). Polycarbonate filters (47 mm filter diameter, 35 mm diameter of effective filter area) containing paraformaldehyde‐fixed bacteria were cut in eight sections. Each filter section was hybridized for 90 min with 30 µl of hybridization solution containing 3 µl of both Cy3‐labelled specific probe and fluorescein‐labelled EUB338 mix in an equilibrated chamber at 46°C. Following in situ hybridization filters were stringently washed for 15 min at 48°C, rinsed in ice‐cold double‐distilled water and dried on Whatman 3M paper. Subsequently, filters were stained with 100 µg ml⁻¹ DAPI (4′,6′‐diamidino‐2‐phenylindole) solution for 5 min on ice, rinsed in ice‐cold double‐distilled water, followed by ice‐cold 50% ethanol and dried on Whatman 3M paper. After drying filters were mounted on glass slides and embedded with Citiflour AFI (Citiflour Ltd, Canterury, UK). Hybridized filters were analysed using an epifluorescence microscope (Carl Zeiss, Göttingen, Germany) equipped with a mercury lamp and appropriate fluorescence filter sets (Loy et al., 2005). For calculating total cell numbers probe‐ and DAPI‐stained cells were counted at a magnification 1000× in 10 randomly chosen fields. Assessment of relative abundance, related to biovolume, was performed by analysing 20 randomly taken images using the Daime software (Daims et al., 2005).

Generally, food samples, including meat, cheese and eggs, were purchased at the local retail level. Food samples were transported cooled at 4°C to the laboratory and immediately artificially contaminated. Twenty‐five grams of each sample was inoculated with different spike levels: (i) 1–10 cfu (low level), (ii) 10–100 cfu (medium level), (iii) 100–1000 cfu (high level) and (iv) 0 cfu (negative control level) of selected food borne pathogen and enriched according to the ISO standard method (ISO 11290‐1, 1996, ISO 6579, 2002, ISO 10273, 2003, ISO 10272‐1, 2006). In detail, the spike strains included S. enterica ssp. enterica serovar Typhimurium DSM 544, S. enterica ssp. enterica serovar Enterica DSM 9898, L. monocytogenes: SLCC 2755 serovar 3a, NCTC 5105 serovar 1/2b and NCTC 7973 serovar 1/2a, C. jejuni DSM 4688, Campylobacter coli DSM 4689 and additionally Y. enterocolitica NCTC 10460. The spike strains stored at −80°C as cryobeads were cultivated overnight in brain heart infusion (BHI) broth (Merck KgA, Darmstadt, Germany) for 18 ± 2 h at the optimal growth temperature (25°C for Y. enterocolitica, 37°C for Salmonella spp. and Listeria monocytogenes, 41,5°C for Campylobacter spp.). To obtain cells in a late‐logarithmic growth phase 100 µl of the overnight cultures were re‐inoculated into 10 ml BHI broth and grown for additional 6–8 h at the corresponding temperature. Campylobacter sp. strains were incubated at 42°C for 44–48 h under microaerophilic conditions and then immediately used for spiking. All cultures were adequately diluted (in 10‐fold steps; up to the level of 10⁻⁷ or 10⁻⁸) in order to obtain the required amount of bacterial cells for spiking. The exact amount used was subsequently quantified by plate counting. All experiments were performed in triplicates resulting in nine sample sets. An overview of the spiked samples is presented in Table 2.

Precisely, 25 g food samples (chicken meat, egg) spiked with different spike levels of Salmonella spp. were preenriched in 225 ml buffered peptone water (BPW) (Merck KGaA, Darmstadt, Germany) for 18 h at 37°C according to ISO 6579. Subsequently, RVS broth (Oxoid Ltd, Cambridge, UK) was inoculated with 100µl BWP and incubated at 42°C for 24 h. Quantification using plate count method was performed on XLD agar (Oxoid Ltd, Cambridge, UK). Twenty‐five gram of each food sample (blue veined cheese, paté) spiked with Listeria monocytogenes was incubated in 225 ml of Half Fraser broth (Merck KGaA, Darmstadt, Germany) for 24 h at 30°C, followed by secondary enrichment in Fraser broth (Merck KGaA, Darmstadt, Germany) for 48 h at 37°C. Bacterial growth was determined using selective PALCAM (Solabia Biokar Diagnostics, Pantin Cedex, France) and OCLA agar (Oxoid Ltd, Cambridge, UK). Pork meat samples spiked with Y. enterocolitica were initially enriched in ITC broth (Merck KGaA, Darmstadt, Germany) for 48 h at 25°C. Quantification was carried out on CIN agar after alkali treatment (Oxoid Ltd, Cambridge, UK). Selective enrichment of chicken meat samples spiked with Campylobacter spp. was performed using Bolton broth (Oxoid Ltd, Cambridge, UK). Samples were incubated in an oxygen‐reduced atmosphere (anaerobic jar with 5% oxygen and 10% carbon dioxide atmosphere) for 48 h at 42°C. Quantification on selective agar plates, DNA extraction and following microarray analysis were performed from each enrichment broth. Genomic DNA was isolated from each enrichment step following the Nucleospin tissue kit (Clontech‐Takara Bio Europe, Saint‐Germain‐en‐Laye, France) user manual instructions and eluted in 100 µl buffer BE.

In order to further test the application potential of the microarray for food analysis and to demonstrate the advantages of its multiplexing feature, a minced meat (25 g of pork and beef mix) matrix was spiked in parallel with different spike levels: (i) 1–10 cfu (low level), (ii) 10–100 cfu (medium level), and (iii) 0 cfu (negative control level) of S. Typhimurium DSM554, L. monocytogenes SLCC 2755 and Y. enterocolitica NCTC10460.

Considering the requirements of the routine food diagnostics in terms of easy handling, all tests with food samples were done without quantifying and adjusting the amount of template DNA. PCR amplification was performed in 100 µl aliquots with 5 µl DNA sample as a template. Subsequently, 20 µl of the purified PCR products was used for the SAP treatment. Other conditions were the same as described above.