Pilot Study To Evaluate Microarray Hybridization as a Tool for Salmonella enterica Serovar Typhimurium Strain Differentiation
http://www.100md.com
微生物临床杂志 2005年第8期
Institute of Microbiology, ETH Zürich, Wolfgang-Pauli-Str. 10, 8093 Zurich, Switzerland
Robert Koch Institut, 38855 Wernigerode, Germany
MicroDiscovery GmbH, Marienburger Strasse 1, 10405 Berlin, Germany
ABSTRACT
In developed countries, Salmonella enterica subspecies 1 serovars Enteritidis and Typhimurium range among the most common causes of bacterial food-borne infections. The surveillance and typing of epidemic Salmonella strains are important tools in epidemiology. Usually, Salmonella enterica subspecies 1 serovars are differentiated by serotyping for diagnostic purposes. Further differentiation is done by phage typing as well as molecular typing techniques. Here we have designed and evaluated a prototype DNA microarray as a tool for serovar Typhimurium strain differentiation. It harbors 83 serovar Typhimurium probes obtained by differential subtractive hybridization and from the public database. The microarray yielded reproducible hybridization patterns in repeated hybridizations with chromosomal DNA of the same strain and could differentiate five serovar Typhimurium reference strains (DT204, DT104, DT208, DT36, and LT2). Furthermore, the microarray identified two distinct groups among 13 serovar Typhimurium DT104 strains. This correlated with observations from pulsed-field gel electrophoresis analysis. Twenty-three further serovar Typhimurium strains were analyzed to explore future directions for optimization of the simple 83-probe DNA microarray. The data presented here demonstrate that DNA microarrays harboring small numbers of selected probes are promising tools for serovar Typhimurium strain typing.
INTRODUCTION
Salmonella spp. are pathogenic enterobacteria which can cause food-borne diseases ranging from a mild gastroenteritis to systemic infections (typhoid fever). In developed countries, infections with Salmonella enterica subspecies 1 serovars Enteritidis and Typhimurium range among the most common causes of bacterial food-borne infections.
For diagnostic purposes, Salmonella enterica subspecies 1 serovars are differentiated by serotyping (28), which allows reliable identification of organisms that cause infections associated with higher risks (i.e., serovar Typhi). For epidemiological surveys, detection of outbreaks, and uncovering of the chains of transmission, serotyping is insufficient because serovar Enteritidis and serovar Typhimurium often account for >80% of all human salmonelloses. For further differentiation, serotyping is complemented by phage typing as well as molecular typing techniques.
During the past decades serovar Typhimurium strains have been differentiated by phage typing (1). The Anderson phage typing scheme is based on the empirical observation that certain phages can cause lytic infections in one serovar Typhimurium strain but not another. A set of 34 bacteriophages allows more than 200 patterns of resistance or sensitivity, the "phage types" (definitive phage types [DTs]), to be distinguished. Phage typing has proven very useful for distinguishing many serovar Typhimurium strains. It has been invaluable for surveillance of the emergence and spread of epidemic serovar Typhimurium strains (i.e., DT204c and DT104) during the past decades (23, 40, 42-44, 49).
In spite of its success, phage typing has several drawbacks. (i) Phage typing has remained an empirical method, and there is only circumstantial evidence for the molecular mechanisms determining resistance or susceptibility to each Anderson typing phage. (ii) The power of resolution is limited. Several distinct serovar Typhimurium strains can belong to the same phage type. (iii) Phage typing relies on "unique" historical stocks which will not last forever. Replenishment of phage stocks will require propagation in Salmonella strains. Recent research has demonstrated that this can lead to recombination and mixed phage populations. Thus, the resulting phage preparations can show altered lysis patterns (36) and, consequently, can affect the consistency of the Anderson phage typing scheme.
Several molecular typing methods have been devised to differentiate serovar Typhimurium strains. This includes pulsed-field gel electrophoresis (PFGE), ribotyping, random amplification of polymorphic DNA, DNA fingerprinting, amplified length polymorphism analysis, enterobacterial repetitive intragenic consensus sequence typing, determination of outer membrane protein and multilocus enzyme patterns, and IS200 typing. Currently, these methods or combinations thereof are being employed successfully for outbreak investigations and identification of lines of transmission. In the past few years, the DNA microarray technology has been developed and is a potentially powerful tool for strain identification. DNA microarrays harboring probes for each open reading frame (ORF) of serovar Typhimurium strain LT2 (GenBank accession no. NC_003197) and additional probes derived from the serovar Typhi CT18 genome (GenBank accession no. NC_003198) and other Salmonella sequences from public databases have been designed (2, 15, 30). Hybridization of chromosomal DNA from standard Salmonella enterica laboratory collections allowed discrimination of a total of >100 strains of different S. enterica subspecies, various subspecies 1 serovars, and even different strains of the same serovar (29). However, so far no systematic studies have evaluated the suitability of DNA microarrays for the differentiation of serovar Typhimurium strains, e.g., for purposes of epidemiology.
Here, we describe a pilot study in which we assessed DNA microarrays as a tool for serovar Typhimurium strain differentiation. A DNA microarray was designed by using strain-specific probes obtained by differential subtractive hybridization and probes derived from public databases. Repeated hybridization experiments with a set of reference serovar Typhimurium strains demonstrated a high reproducibility of strain differentiation by this technique. Furthermore, we analyzed clinical serovar Typhimurium isolates of phage type DT104 and additional phage types. We found that the DNA microarray is useful for serovar Typhimurium strain differentiation. Implications for future applications and strategies for further optimization of the microarray are discussed.
MATERIALS AND METHODS
Bacterial strains. Serovar Typhimurium strain SL1344 has been described previously (39), strain DT104 (99-00971) was obtained from A. Hensel (Institut für Tierhygiene, Leipzig, Germany), and Vibrio cholerae El Tor Nent 720-95 was obtained from the Institut für Veterinre Bakteriologie, Bern, Switzerland.
Serovar Typhimurium isolates and serovar Typhi DNA (SARC2) were from the Robert Koch Institut, Wernigerode, Germany (Table 1). Chromosomal DNAs from Escherichia coli strain EDL933 and Yersinia enterocolitica strain 8081 were kindly provided by A. Rakin and S. Schubert (Max von Pettenkofer-Institut, Munich, Germany).
DNA manipulation. Chromosomal DNA of serovar Typhimurium strains DT104 (99-00971), DT204 (STM1690/75), DT208 (201IE888), DT36 (Anderson), and SARA2 (LT2) was prepared by standard methods (35). Chromosomal DNA was isolated from the strains studied by DNA hybridization, according to the protocol of Ausubel (3).
PFGE was performed as described previously (31).
SH and subcloning of the DNA fragments. Subtractive hybridization (SH) was performed by using the Clontech PCR-Select bacterial genome subtraction kit (PT3170-1; Clontech, Palo Alto, Calif.). Briefly, chromosomal serovar Typhimurium DNA (tester DNA) was digested with RsaI (four-base cutter, GTAC), purified, and divided into two aliquots. One aliquot was ligated to adapter 1, and the other was ligated to adapter 2R.
In a third tube, portions of each aliquot were combined (negative control for subtractive hybridization). Two rounds of hybridization with RsaI-digested DNA from another serovar Typhimurium strain (driver DNA) were performed at a temperature (48°C) lower than that recommended by the manufacturer (63°C) in order to increase the fraction of DNA fragments with low or no similarity between the driver and the tester DNAs. Then, a primary PCR and a secondary PCR with primers that anneal to adapters 1 and 2R were performed, and the PCR products were cloned into pCR2.1-Topo (Topo TA cloning kit; Invitrogen, Germany). Plasmids from lac-negative clones growing in Luria-Bertani medium (100 μg/ml ampicillin, 50 μg/ml kanamycin, isopropyl--D-thiogalactopyranoside, 5-bromo-4-chloro-3-indolyl--D-galactopyranoside) were analyzed by PCR (with primers T7 [5'-GTAATACGACTCACTATAGGGC-3'] and M13 reverse [5'-GGAAACAGCTATGACCATG-3']; annealing at 52°C and elongation for 1.5 min at 72°C for 35 cycles) and sequenced (AGOWA, Berlin, Germany). As probes for the microarray, we selected 500- to 800-nucleotide (nt)-long inserts (named A2 to G4 and 2SH4 to 3SH93; Table 2) showing <75% identity (blastn;www.ncbi.nlm.nih.gov/BLAST) over >50% of their entire sequence to serovar Typhimurium LT2, E. coli K-12, and Shigella genome sequences (http://www.ncbi.nlm.nih.gov/sutils/genom_table.cgi). Inserts with high similarities to mobile genetic elements (transposons, insertion sequence elements, phages) or pathogenicity islands were also included.
Microarray design and hybridization. Probes were generated by PCR (Table 2) and were analyzed by Scienion (Berlin, Germany) with a 2100 Bioanalyzer (Agilent Technologies) (data not shown). Eighty-three PCR products yielding a single band (Table 2) of the expected size were used for array design. Each probe was spotted onto two locations of the "amino slides" by using the Scienion processing system and cross-linked (300-mJ UV light; Stratalinker; Stratagene, The Netherlands). Quality control was performed by Scienion by SYBR Green staining of DNA and complex hybridization (with Cy3 and Cy5).
From each strain to be analyzed, 4 μg of chromosomal DNA was denatured, hybridized to random decamer primers, and labeled by primer extension by using Klenow polymerase and a deoxynucleoside triphosphate mix containing Cy5-dCTP. The labeled DNA was purified (MiniElute; QIAGEN), and the labeling efficiency was verified by gel electrophoresis. Array hybridization was performed in a formamide-based buffer (Sci-Hyb; Scienion) at 42°C for 20 h, followed by washing steps with increasing stringency (Sci-Wash; Scienion). The fluorescence signals (635-nm excitation) obtained with each probe were measured with an Axon 4100 scanner (16-bit; intensity range, 0 to 65,535), and the raw data were corrected for the background and processed with GenePix Pro V5.0 software. This was performed by Scienion.
Data processing. The data were transformed to log scale and normalized by quantile-based methods (37, 50). A two-component normal mixture model (24), p(x; m1, m2, s1, s2, w) = (1 – w) · n(x; m1, s1) + w · n2(x; m2, s2), was fitted to the normalized data by a maximum-likelihood method. The discriminant function (12) w · [n2(x)/p(x)] was used to represent the propensity of the sequence for being present (value close to 1) or absent (value close to 0) from the DNA of that strain. Discriminant values for all probes and all samples (strains) were calculated and stored in a signal probability matrix, with each column representing one array measurement and with each row representing the results obtained with one specific probe on the chip. For representation of the probability matrix, a coloring scheme was used: blue indicates state 1 (off), yellow indicates state 2 (on), and dark shading indicates indecisive measurements (p value of about 0.5).
Strain differentiation was studied as follows: pairwise correlation coefficients (16) were calculated from the signal probability values from each pair of hybridization experiments. The pairwise correlation coefficients were stored in a sample correlation matrix. For visualization of the results, the sample correlation matrix was displayed as a color scale ranging from green (weakly correlated) to red (strongly correlated) (adapted from the "sma" package [13]).
Dendrograms were calculated based on the signal probability matrix by using Euclidean distance and the average linkage method (12, 38). All calculations were implemented by using the R-statistics package (22).
RESULTS
Probes for microarray design. There is considerable sequence diversity between different Salmonella subspecies and serovars (30). Much of this diversity is attributable to the loss or acquisition of DNA fragments by horizontal gene transfer (14). Even between different serovar Typhimurium strains, some DNA fragments do differ: variable sequences include those of antibiotic resistance genes, plasmids, and phages (17, 19, 26, 29, 47). Detection of these DNA fragments should allow serovar Typhimurium strain differentiation at high resolution with fairly simple microarrays composed of <200 probes. To test this hypothesis, we have designed a DNA microarray that includes probes from (i) Salmonella DNA fragments retrieved by SH experiments and (ii) sequences derived from sequences in a public database (National Center for Biotechnology Information [http://www.ncbi.nlm.nih.gov/]).
DNA microarray probes obtained by SH. SH allows the enrichment or retrieval of sequences present in one bacterial genome (designated the "tester") but absent from another (designated the "driver") (see Materials and Methods). These sequences would provide ideal probes for DNA microarray design.
The DNA of isolates of three serovar Typhimurium strains (DT204 = STM1690/75, DT104 = 99-00971, and DT208 = 201 IE 888; Table 1) known to differ significantly from each other (W. Rabsch, R. Prager, and H. Tschpe, unpublished observations) were chosen as testers. DT204 had caused a major epidemic (23, 42, 44) in the 1970s and 1980s and carries the SopE prophage, which encodes the type III effector gene sopE, in its chromosome (26, 27). Serovar Typhimurium DT104 (99-00971) is a representative of the antibiotic multiresistant phage type DT104 strains which emerged in several European countries as the predominant serovar Typhimurium phage type in human infections (31, 43, 49). Outbreaks have also been reported in the United States (5, 6, 34). Serovar Typhimurium DT208 (201IE888) is a representative of phage type DT208, which caused an outbreak in the former Soviet Union but which is also identified in current epidemiological surveys (11, 41).
We used DNA from the serovar Typhimurium strains mentioned above and from DT36 (Anderson), a strain highly susceptible to all Anderson typing phages. Tester-driver combinations were DT204 (STM1690/75)-DT104 (99-00971), DT208 (201IE888)-DT204 (STM1690/75), DT104 (99-00971)-DT36 (Anderson), DT204 (STM1690/75)-DT36 (Anderson), and DT208 (201IE888)-DT36 (Anderson). The hybridization steps were performed at 48°C to ensure enrichment of the tester DNA fragments with no significant similarity to the driver genome. DNA fragments were sequenced, and we selected 43 plasmids harboring 500- to 1,600-nt DNA inserts with no or limited similarity to known Salmonella or enterobacterial sequences for probe design (see Materials and Methods). Inserts with high similarities to mobile genetic elements (transposons, insertion sequence elements, phages) or pathogenicity islands were also included. Upon PCR amplification, 38 of these plasmids yielded unique products and were selected for DNA microarray design (Table 2; see Table S1 in the supplemental material). Nine of these probes originated from DT104 (99-00971), 24 originated from DT204 (STM1690/75), and 5 originated from DT208 (201IE888) (Table 2).
DNA microarray probes derived from prophages. The repertoire of prophages residing in the genome is highly diverse between different serovar Typhimurium strains (8, 17, 26). Unique sequences (immunity region, morons) which are present in one prophage but not in most others should be well suited for strain differentiation. PCR products corresponding to 21 unique prophage sequences from serovar Typhimurium strains, serovar Typhi, and serovar Paratyphi A (33) were selected for DNA microarray design (Table 2; see Table S2 in the supplemental material). Three probes were derived from the phage-encoded Shiga-like toxin A subunit (E. coli EDL 933; O157:H7) and cholera toxin A and B subunits (Vibrio cholerae El Tor Nent 720-95). It was unclear whether one of these toxins might also be present in some serovar Typhimurium strains.
DNA microarray probes from TTSSs. Type III secretion systems (TTSSs), their known effector genes, and chaperones are prerequisites for Salmonella virulence (18). The TTSSs and most of the effector genes are highly conserved between all Salmonella strains (25, 30). We designed probes based on sopE2, sopB, sipA, slrP, invB, invC, and prgH as positive controls (Table 2; see Table S2 in the supplemental material).
Effector genes sopE (probe D2 from SH) and avrA are more variable (20, 21, 26, 32) and were also included.
Additional probes for DNA microarray design. Additional probes were based on serovar Typhimurium SL1344 and DT204 (3341/78) DNA fragments retrieved in an earlier study (Table 2, probes pM48, K58, K51, and K9) (K. Ehrbar and W.-D. Hardt, unpublished data) and fimC (serovar Typhimurium LT2). Furthermore, open reading frames STM2240 (serovar Typhimurium LT2) and STY0312, STY1360, and STY1364 (serovar Typhi), which do not show similarity to other sequences in the public database, were selected for probe design (Table 2; see Table S2 in the supplemental material).
In an attempt to control the stringency of microarray hybridization probes E. coli frd (GenBank accession no.DH5_FRB; 90% identity to serovar Typhimurium LT2), E. coli kefc (GenBank accession no. DH5_KEFC; 83% identity to serovar Typhimurium LT2), E. coli agp (GenBank accession no. DH5_AGP; 80% identity to serovar Typhimurium LT2), and Yersinia enterocolitica thrc (GenBank accession no. Y_ENT_THRC; 76% identity to serovar Typhimurium LT2) (Table 2) were designed. The green fluorescent protein (gfp) (10) served as a negative control.
Microarray design. For microarray design, a total of 83 probes (250 to 1,500 nt long) were used, including 38 probes derived from subtractive hybridizations, 4 from earlier genetic screens, and 41 from the sequences in a public database (National Center for Biotechnology Information). Each probe was spotted onto two different locations of the amino slides. Probes E10, pM816, D6, and G3 each were present in four different spots on the microarray. A total of 100 microarrays were produced in two separate batches. Quality control (see Material and Methods) and hybridization experiments with identical DNA samples did not detect differences in the performance of the two different batches (data not shown).
Hybridization with five serovar Typhimurium reference strains. We performed repeated DNA microarray hybridization experiments to analyze the reproducibility and explore the quality of strain differentiation obtained by this method. For this purpose serovar Typhimurium DT36 (Anderson), DT204 (STM1690/75), DT104 (99-00971), DT208 (201IE888), and LT2 were chosen. The chromosomal DNA of one strain was isolated in two independent preparations, and each batch was hybridized with three DNA microarrays (hybridization data not shown). Each of the six hybridizations involved a separate labeling reaction. In this way we could determine whether technical problems associated with poorly reproducible steps of the hybridization procedure (see Materials and Methods) might interfere with DNA microarray-based strain typing.
Hybridization signals were transformed to log scale and normalized (see Materials and Methods). The signals obtained from pairs of identical probes on each array were very similar and showed typical correlation coefficients greater than 0.99 (data not shown). For all further calculations, the two signals were represented by their mean value.
Next, we used the hybridization signals to calculate the probability that a specific sequence is present in or absent from the strain. For this purpose, a two-component normal mixture model was fitted to the hybridization data by a maximum-likelihood method (see Materials and Methods). This yielded a "discriminant" function (see Fig. S1 in the supplemental material) which can use the hybridization signals to calculate for a probe the probability of a match (value close to 1) or no match (value close to 0) with the strain analyzed. The discriminant function was used to transform the 83 hybridization signals into a list of probabilities ("signal probability matrix") for the presence or absence of each probe (data not shown). To study strain differentiation, the signal probability list between each pair of DNA microarray hybridizations was compared. The resulting correlation coefficients were stored in a sample correlation matrix (see Materials and Methods). In Fig. 1, the sample correlation matrix for the analysis of the five representative serovar Typhimurium strains has been displayed in color code: red indicates strongly correlated, and green indicates weakly correlated.
The red diagonal is attributable to the strong correlation when each hybridization experiment is compared with itself. Besides this, the correlation matrix shows a chessboard pattern of strong and weak correlations. All hybridizations with DNA from the same strain yielded very similar results (large red areas on the chessboard). No similarity between the hybridization results from different serovar Typhimurium strains was observed. Therefore, the hybridization data are reproducible, and the prototype DNA microarray allows reliable differentiation of serovar Typhimurium DT36 (Anderson), DT204 (STM1690/75), DT104 (99-00971), DT208 (201IE888), and LT2.
Slight variations were observed within the group of six DNA microarray hybridizations performed with serovar Typhimurium DT36 (Anderson), as indicated by a shift to dark red in the correlation matrix. The same observation was made for serovar Typhimurium LT2. The reasons for this variation are unclear. However, strain differentiation was still clearly feasible.
The weakest correlation was observed between DT204 and the other strains, in particular, strains DT104 (99-00971) and DT208 (201IE888). This is at least partially attributable to the large number of DT204 (STM1690/75)-specific probes (24 in total; Table 2) present on the DNA microarray. In addition, DT204 (STM1690/75) failed to hybridize with many probes specific for the other strains (data not shown).
The hybridization patterns were also compared in a dendrogram. For this purpose, the signal probability matrix was analyzed by using Euclidean distance and the average linkage method (see Materials and Methods). The topology of the resulting dendrogram (Fig. 2) supported the results from the correlation analysis (Fig. 1). The branches from the six hybridizations of each serovar Typhimurium strain clustered closely together. The branch representing DT204 (STM1690/75) diverges early, while the branches for DT104 (99-00971) and DT208 (201IE888) cluster closely together.
It should be noted that this dendrogram does not provide information on the phylogenetic relationship between the strains. However, it is helpful for the recognition of groups of similar hybridization patterns. Throughout the rest of this study we have employed the dendrogram calculation in order to group the strains before plotting the color-coded pairwise correlation matrices (see below).
Taken together the results from this first series of hybridization experiments provided evidence for the reproducibility of the data and the feasibility of DNA microarray-based differentiation of serovar Typhimurium strains.
Microarray hybridization with various DT104 typed strains. The data presented above showed that the simple prototype DNA microarray could differentiate serovar Typhimurium strains which belong to different phage types. Next, we were interested in finding out whether different serovar Typhimurium isolates of the same phage type would yield similar hybridization results. To analyze this, we have chosen 13 serovar Typhimurium DT104 strains (Tables 3 and 4, strains 02-00822, 02-02682, 02-03881, 03-06256, 97-01178, 97-01193, 97-01864, 97-02103, 97-01174, 02-00429, 02-03122, 96-01186, and 97-05652). Three independent DNA microarray hybridizations (I, II, III) were performed with each strain, as described above. The DT104 (99-00971) reference strain was also hybridized once more.
The hybridization data were analyzed as described above. For better comparison, the results for the five reference strains shown in Fig. 1 and 2 have been included in this analysis (color coded in blue, grey, pink, marine, and yellow). The dendrogram (Fig. 3) revealed that all branches representing DT104 strains were clearly separated from the branches of reference strains serovar Typhimurium DT36 (Anderson), DT204 (STM1690/75), DT208 (201IE888), and LT2.
Interestingly, the 13 DT104 strains formed two clearly separated branches: subgroup 1 strains (02-00822, 02-02682, 02-03881, 03-06256, 97-01178, 97-01193, 97-01864, 97-02103, and 97-01174) were very similar to reference strain DT104 (99-00971). Subgroup 2 strains (02-00429, 02-3122, 96-01186, and 97-05652) formed a separate branch less similar to the DT104 reference strain.
The correlation analysis confirmed the finding of two DT104 subgroups (Fig. 4).
To explore these results in more detail, the 13 serovar Typhimurium DT104 strains were analyzed by PFGE (Fig. 5). All strains of subgroup 2 had a very similar PFGE pattern (Fig. 5, lanes 2 to 5). This pattern differed significantly (more than five bands) from that of the typical DT104 strains associated with the current epidemic (Fig. 5, lanes 11 to 15) (31) and from the additional subgroup 1 strains (Fig. 5, lanes 6 to 8 and 10).
In conclusion, the DNA microarray allowed the reliable and reproducible identification of different isolates of the same phage type. Moreover, it could differentiate two groups of DT104 strains. This was unexpected, because the DNA microarray did not specifically include probes designed for this purpose. However, this observation suggested that more sophisticated subdifferentiation of serovar Typhimurium DT104 strains might be feasible by including additional probes on the next microarray generation.
Microarray hybridization with various phage-typed strains. So far, our study had focused on the differentiation of strains representing five different phage types. However, the number of serovar Typhimurium strains in nature is much higher. Even the Anderson phage typing scheme is able to differentiate >200 different types of serovar Typhimurium strains. In order to explore the potential and the limitations of the current prototype microarray, we have studied strains representing various phage types. Here, each strain was hybridized only once. Therefore, the data can provide general insights, but they are not statistically significant. Strains belonging to phage types DT204 (00-02379, 00-07509, 00-09603, 00-10022, and 99-07382), DT193 (02-01414, 02-04395, 02-04515), DT120 (03-01606, 03-01663, and 03-01860), DT92 (03-00009, 03-0008), DT12 (03-6841, 03-07846), DT17 (03-01862), DT177 (03-05510), DT186 (03-5115) DTu302 (03-6382), DT7 (03-6600), DT9 (03-01779), DT170 (04-0093), DT2 (04-0110), DT175 (305-70), DT68 (11635), and DT40 (6772/96) were analyzed as described above. For comparison, data for the five reference strains (Fig. 1 and 2) have been included (color coded in blue, grey, pink, marine, and yellow). As expected, all five reference strains were located in distinct branches of the dendrogram (Fig. 6). The vast majority of the other serovar Typhimurium strains clustered in three distinct branches of the dendrogram (termed X, Y, and Z; Fig. 6). These branches were clearly distinct from those of the five reference serovar Typhimurium strains. The hybridization patterns of two strains, DT193 (02-01414) and DT177 (03-05510), differed from those of all of the other strains analyzed. This indicated that the prototype DNA microarray can differentiate at least 11 different types of serovar Typhimurium strains, the five reference strains (Fig. 1); DT104 subgroup 2 (Fig. 3 and 4); clusters X, Y, and Z; and the two strains DT193 (02-01414) and DT177 (03-05510).
The five additional DT204 strains (00-02379, 00-07509, 00-09603, 00-10022, and 99-07382) clustered closely together (Fig. 6, cluster X) and were located close to the DT204 (STM1690/75) reference strain. However, they were more closely related to each other than to the DT204 reference strain. This suggested that the DNA microarray can subdifferentiate at least some DT204 strains. Further experiments are required to substantiate this hypothesis.
Interestingly, the DT68 (11635) strain, which carries the SopE prophage (26), was also located in the same branch as the DT204 strains. Similarly, 03-6382 (Dtu302) clustered closely with the DT104 (99-00971) reference strain. Further studies are needed to determine whether strains DT68 (11635) and DT204 or strains Dtu302 (03-6382) and DT104 may have a recent common ancestry.
Cluster Y harbored both DT92 isolates and both DT12 isolates. The three DT120 isolates were located in cluster Z (Fig. 6). Thus, strains of these three phage types might be recognized specifically by the prototype DNA microarray and differentiated from the five reference serovar Typhimurium strains. Further experiments are needed to substantiate this hypothesis.
The three DT193 isolates (02-01414, 02-04395, and 02-04515) were located in separate branches of the dendrogram, implying that this DT group includes heterogeneous serovar Typhimurium strains which can be differentiated by the DNA microarray. Overall, clusters Y and Z harbored quite diverse groups of strains, as judged by phage typing. It is likely that the prototype DNA microarray is simply insufficient to differentiate between these strains. In this case, additional probes would be required in order to improve strain differentiation by future generations of a serovar Typhimurium typing microarray.
Evaluation of the probes chosen for microarray design. For the future design of improved serovar Typhimurium typing arrays, it would be useful to analyze in more detail the performance of each probe on the current DNA microarray. For this purpose, the signal probability matrix (see Materials and Methods) for each probe in each hybridization experiment has been evaluated (see Fig. S2 in the supplemental material). Twelve DNA probes matched every serovar Typhimurium strain analyzed. This included most probes derived from TTSSs as well as probes E6 (transposase), A7 (prophage ST64T), A8 (transposase), A6 (SPI3), and B4 (100% identity to LT2) obtained by subtractive hybridization. Twelve DNA probes did not match any of the serovar Typhimurium strains analyzed. This included the negative control (GFP); prophage-encoded genes for CTXA, CTXB, and EDL-STXA; the serovar Paratyphi B sequences Paratyphi-Contig 8 and Paratyphi-Contig 35; the serovar Typhi sequences Typhi-STY1360, Typhi-STY0312, and Typhi-STY4629; Y. enterocolitica THRC; DH5-AGP; and DH5-KEFC. Some of these probes might be omitted in future. However, a few "positive and negative controls" should be retained. It might be of interest to retain probes GFP (truly negative control), Y_ENT_THRC (76% identity to LT2), DH5_AGP (80% identity to LT2), and DH5_KEFC (83% identity to LT2). In contrast to these "absent" probes, control probe DH5_FRB (90% identity) yielded matches in almost all hybridization experiments. Thus, these probes can provide a useful control for the general stringency of DNA microarray hybridization. In addition, these observations indicate that DNA fragments less than 83% identical between two serovar Typhimurium strains can be employed on future DNA microarrays.
Several probes derived from the serovar Typhimurium LT2 (10 probes) and serovar Typhi CT18 (6 probes) genome sequences contributed to strain differentiation. This pertained especially to probes derived from unique prophage sequences. Thus, additional completed Salmonella genomes will provide promising sources for the design of additional probes.
The vast majority (30 of 38) of probes derived by subtractive hybridization clearly contributed to strain differentiation. Due to the possibility to experimentally enrich DNA fragments that differ between two strains of choice, subtractive hybridization will provide a powerful tool by which additional probes for improved DNA microarray-based strain differentiation can be obtained.
DISCUSSION
Fine differentiation of serovar Typhimurium strains is of great importance for epidemiology and the identification of chains of transmission. Phage typing and several molecular typing techniques are routinely used for this purpose. Potentially, microarrays offer a powerful alternative for phage typing. In fact, chip-based strain typing has been demonstrated for Listeria spp. and Campylobacter spp. (45, 46), and a DNA chip representing each serovar Typhimurium LT2 open reading frame was found to differentiate strains from Salmonella reference collections (29). Here, we report on the first pilot study to use simple DNA microarrays for serovar Typhimurium strain differentiation. The prototype DNA microarray harbored 83 probes. It yielded reproducible hybridization patterns in repeated hybridizations with the same strain and could differentiate five reference strains (Fig. 1 and 2), two subgroups of serovar Typhimurium DT104 (Fig. 3 to 5), two types of DT204 strains, as well as three clusters of serovar Typhimurium wild-type isolates (Fig. 6). This demonstrated that DNA microarrays harboring small numbers of selected probes are promising tools for serovar Typhimurium strain differentiation.
Our study included merely 44 serovar Typhimurium strains. This included the most prevalent serovar Typhimurium strains from Germany, but it is just a small fraction of the existing serovar Typhimurium strains with distinct genetic makeups (>>200). Thus, it is conceivable that the prototype DNA microarray can differentiate even more serovar Typhimurium strains than is evident from this study. More hybridization experiments would be required to address this point.
Future generations of the DNA microarray should include additional probes that will allow the accurate discrimination between those strains not differentiated by the current DNA microarray (i.e., subgroup 1 DT104 strains [Fig. 5, lanes 6 to 8 and 10] and strains of clusters Y and Z [Fig. 6]) or strains which are going to emerge in the future.
How many strain-specific probes are required Optimal strain differentiation was achieved for the reference strain DT204 (STM1690/75; Fig. 1). This strain yielded strong correlations in repeated hybridizations but was clearly distinct from other DT204 strains (Fig. 6) and yielded only a weak correlation with all other reference strains (Fig. 1). The DNA microarray harbored 31 probes that yielded positive hybridization signals with DT204 (STM1690/75) but not with one or more of the other strains. Twenty-one of these "useful" probes were derived from DT204 (STM1690/75) by subtractive hybridization. This number of probes represents an estimate for the upper limit, since all differences are expected to show up in this experimental setup. Sensible strain differentiation can probably be achieved with fewer strain-specific probes. Reference strain DT104 (99-00971) provides an example: only six of the nine probes derived from DT104 (99-00971) by subtractive hybridization yielded positive hybridization signals with DT104 (99-00971) but not with one or more of the other strains. Still, the prototype microarray could differentiate DT104 (99-00971) from all other reference strains and could even distinguish between two different subgroups of DT104 strains (Fig. 4).
How can the prototype DNA microarray be improved Clearly, additional DNA probes that are present in some serovar Typhimurium strains but absent from others are needed. As discussed above, DNA fragments of several hundred base pairs <83% identical between different strains will be useful for strain typing. It should be noted that (in contrast to many other DNA array studies) it has proven useful to focus on DNA similarity and not on open reading frames. Thus, many probes obtained by subtractive hybridization represent partial open reading frames or noncoding DNA.
In theory, optimal efficiency would be achieved if each of the probes would recognize a different subset of serovar Typhimurium strains. More precisely, an optimal (orthogonal) set of probes would show a vanishing correlation for each pair of probes or samples. Currently, there are two main strategies for acquisition of additional probes: ongoing sequencing projects which include several serovar Typhimurium strains and strains from other S. enterica subspecies 1 serovars will provide rich sources for probe generation. Alternatively, subtractive hybridization can be used. This method has provided many useful probes in this study. It is fairly inexpensive and allows the enrichment of DNA sequences present in one strain but absent from another. Therefore, it is ideally suited for the identification of probes which distinguish between strains which are not clearly distinguished by the prototype DNA microarray (i.e., different DT104 strains of subgroup 1 [Fig. 4 and 5] and different strains of clusters Y and Z [Fig. 6]).
In conclusion, this pilot study has demonstrated that DNA microarrays are useful tools for serovar Typhimurium strain differentiation. Most likely, strains from other S. enterica subspecies 1 serovars can be differentiated in a similar manner. If one considers that the technology for chip production, hybridization, and analysis has become widely available in recent times, DNA microarray strain typing holds many promises for the fast and efficient detection of chains of transmission and for epidemiology in general. However, it should be noted that today's high costs for DNA microarray production and the labeling must drop significantly before this technology will be widely applied in diagnostic and reference laboratories.
ACKNOWLEDGMENTS
We thank Alexander Rakin, Sren Schubert, and Martin Braun for discussion, strains, and chromosomal DNA and Holger Eickhoff and Sandra Prietz (Scienion) for excellent service in microarray production and hybridization.
This work was funded in part by a grant from the Kompetenznetzwerk "Genomforschung an pathogenen Bakterien" Verbund 3 (BMBF, Germany) to W.-D. Hardt.
Supplemental material for this article may be found at http://jcm.asm.org.
REFERENCES
Anderson, E. S., L. R. Ward, M. J. Saxe, and J. D. de Sa. 1977. Bacteriophage-typing designations of Salmonela typhimurium. J. Hyg. (London) 78:297-300.
Andrews-Polymenis, H. L., W. Rabsch, S. Porwollik, M. McClelland, C. Rosetti, L. G. Adams, and A. J. Baumler.2004 . Host restriction of Salmonella enterica serotype Typhimurium pigeon isolates does not correlate with loss of discrete genes. J. Bacteriol. 186:2619-2628.
Ausubel, F. M. 1997. Current protocols in molecular biology. John Wiley & Sons, Inc., New York, N.Y.
Beltran, P., S. A. Plock, N. H. Smith, T. S. Whittam, D. C. Old, and R. K. Selander.1991 . Reference collection of strains of the Salmonella typhimurium complex from natural populations.J. Gen Microbiol. 137:601-606.
Besser, T. E., C. C. Gay, J. M. Gay, D. D. Hancock, D. Rice, L. C. Pritchett, and E. D. Erickson. 1997. Salmonellosis associated with S. typhimurium DT104 in the USA. Vet. Rec. 140:75.
Besser, T. E., M. Goldoft, L. C. Pritchett, R. Khakhria, D. D. Hancock, D. H. Rice, J. M. Gay, W. Johnson, and C. C. Gay. 2000. Multiresistant Salmonella Typhimurium DT104 infections of humans and domestic animals in the Pacific Northwest of the United States. Epidemiol. Infect. 124:193-200.
Boyd, E. F., F. S. Wang, T. S. Whittam, and R. K. Selander. 1996. Molecular genetic relationships of the Salmonellae. Appl. Environ. Microbiol. 62:804-808.
Brussow, H., C. Canchaya, and W. D. Hardt. 2004. Phages and the evolution of bacterial pathogens: from genomic rearrangements to lysogenic conversion. Microbiol. Mol. Biol. Rev. 68:560-602.
Carniel, E., I. Guilvout, and M. Prentice. 1996. Characterization of a large chromosomal "high-pathogenicity island" in biotype 1B Yersinia enterocolitica.J. Bacteriol. 178:6743-6751.
Cormack, B. P., R. H. Valdivia, and S. Falkow.1996 . FACS-optimized mutants of the green fluorescent protein (GFP). Gene 173:33-38.
Davies, R. H., R. Dalziel, J. C. Gibbens, J. W. Wilesmith, J. M. Ryan, S. J. Evans, C. Byrne, G. A. Paiba, S. J. Pascoe, and C. J. Teale. 2004. National survey for Salmonella in pigs, cattle and sheep at slaughter in Great Britain (1999-2000).J. Appl. Microbiol. 96:750-760.
Duda, R. O., and P. E. Hart. 1973. Pattern classification and scene analysis. John Wiley & Sons, Inc., New York, N.Y.
Dudoit, S., Y. H. Yang, M. J. Callow, and T. P. Speed 2000, posting date. Statistical methods for identifying differentially expressed genes in replicated cDNA microarray experiments. Technical report 578. [Online.]. http://www.stat.berkeley.edu/users/terry/zarray/Html/matt.html.
Edwards, R. A., G. J. Olsen, and S. R. Maloy.2002 . Comparative genomics of closely related salmonellae.Trends Microbiol. 10:94-99.
Eriksson, S., S. Lucchini, A. Thompson, M. Rhen, and J. C. Hinton.2003 . Unravelling the biology of macrophage infection by gene expression profiling of intracellular Salmonella enterica.Mol. Microbiol. 47:103-118.
Feller, W. 1950. An introduction to probability theory and its applications. John Wiley & Sons, Inc., New York, N.Y.
Figueroa-Bossi, N., S. Uzzau, D. Maloriol, and L. Bossi. 2001. Variable assortment of prophages provides a transferable repertoire of pathogenic determinants in Salmonella. Mol. Microbiol. 39:260-272.
Galan, J. E. 2001. Salmonella interactions with host cells: type III secretion at work. Annu. Rev. Cell Dev. Biol. 17:53-86.
Guerra, B., E. Junker, A. Miko, R. Helmuth, and M. C. Mendoza.2004 . Characterization and localization of drug resistance determinants in multidrug-resistant, integron-carrying Salmonella enterica serotype Typhimurium strains. Microb. Drug Resist. 10:83-91.
Hardt, W. D., and J. E. Galan. 1997. A secreted Salmonella protein with homology to an avirulence determinant of plant pathogenic bacteria. Proc. Natl. Acad. Sci. USA 94:9887-9892.
Hardt, W. D., H. Urlaub, and J. E. Galan.1998 . A substrate of the centisome 63 type III protein secretion system of Salmonella typhimurium is encoded by a cryptic bacteriophage. Proc. Natl. Acad. Sci. USA 95:2574-2579.
Ihaka, R., and R. Gentleman. 1996. R: a language for data analysis and graphics. J. Comput. Graph. Stat. 5:299-314.
Kuhn, H., W. Rabsch, H. Tschape, and E. Tietze. 1982. Characterization and epidemiology of a Salmonella typhimurium epidemic strain. Z. Arztl. Fortbild (Jena) 76:607-610. (In German.)
McLachlan, G. J., and D. Peel. 2000. Finite mixture models. Wiley series in probability and statistics, John Wiley and Sons, Inc., New York, N.Y.
Mirold, S., K. Ehrbar, A. Weissmüller, R. Prager, H. Tschpe, H. Rüssmann, and W. D. Hardt. 2001. Salmonella host cell invasion emerged by acquisition of a mosaic of separate genetic elements, including Salmonella pathogenicity island 1 (SPI1), SPI5, and sopE2. J. Bacteriol. 183:2348-2358.
Mirold, S., W. Rabsch, M. Rohde, S. Stender, H. Tschape, H. Russmann, E. Igwe, and W. D. Hardt. 1999. Isolation of a temperate bacteriophage encoding the type III effector protein SopE from an epidemic Salmonella typhimurium strain. Proc. Natl. Acad. Sci. USA 96:9845-9850.
Pelludat, C., S. Mirold, and W. D. Hardt. 2003. The SopEPhi phage integrates into the ssrA gene of Salmonella enterica serovar Typhimurium A36 and is closely related to the Fels-2 prophage. J. Bacteriol. 185:5182-5191.
Popoff, M. Y., J. Bockemuhl, and A. McWhorter-Murlin.1992 . Supplement 1991 (no. 35) to the Kauffmann-White scheme. Res. Microbiol. 143:807-811.
Porwollik, S., E. F. Boyd, C. Choy, P. Cheng, L. Florea, E. Proctor, and M. McClelland. 2004. Characterization of Salmonella enterica subspecies I genovars by use of microarrays. J. Bacteriol. 186:5883-5898.
Porwollik, S., R. M. Wong, and M. McClelland. 2002. Evolutionary genomics of Salmonella: gene acquisitions revealed by microarray analysis. Proc. Natl. Acad. Sci. USA 99:8956-8961.
Prager, R., A. Liesegang, W. Rabsch, B. Gericke, W. Thiel, W. Voigt, R. Helmuth, L. Ward, and H. Tschape. 1999. Clonal relationship of Salmonella enterica serovar typhimurium phage type DT104 in Germany and Austria. Zentbl. Bakteriol. Parasitenkd. Infektkrankh. Hyg. Abt. 1 Orig. 289:399-414.
Prager, R., S. Mirold, E. Tietze, U. Strutz, B. Knuppel, W. Rabsch, W. D. Hardt, and H. Tschape. 2000. Prevalence and polymorphism of genes encoding translocated effector proteins among clinical isolates of Salmonella enterica. Int. J. Med. Microbiol. 290:605-617.
Prager, R., W. Rabsch, W. Streckel, W. Voigt, E. Tietze, and H. Tschape.2003 . Molecular properties of Salmonella enterica serotype Paratyphi B distinguish between its systemic and its enteric pathovars. J. Clin. Microbiol. 41:4270-4278.
Rabatsky-Ehr, T., J. Whichard, S. Rossiter, B. Holland, K. Stamey, M. L. Headrick, T. J. Barrett, and F. J. Angulo.2004 . Multidrug-resistant strains of Salmonella enterica Typhimurium, United States, 1997-1998. Emerg. Infect. Dis. 10:795-801.
Sambrook, J., E. F. Fritsch, and T. Maniatis. 1989.Molecular cloning: a laboratory manual , 2nd ed. Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y.
Schmieger, H. 1999. Molecular survey of the Salmonella phage typing system of Anderson. J. Bacteriol. 181:1630-1635.
Schuchhardt, J., D. Beule, A. Malik, E. Wolski, H. Eickhoff, H. Lehrach, and H. Herzel. 2000. Normalization strategies for cDNA microarrays. Nucleic Acids Res. 28:E47.
Steinhausen, D., and K. Langer. 1977. Clusteranalyse: Einführung in Methoden und Verfahren der automatischen Klassifikation: mit zahlreichen Algorithmen, FORTRAN-Programmen, Anwendungsbeispielen und einer Kurzdarstellung der multivarianten statistischen. Verfahren, 1. Erste Editierte Auflage de Gruyter, Berlin, Germany.
Stocker, B. A., S. K. Hoiseth, and B. P. Smith. 1983. Aromatic-dependent Salmonella spp. as live vaccine in mice and calves. Dev. Biol. Stand. 53:47-54.
Threlfall, E. J., J. A. Frost, L. R. Ward, and B. Rowe. 1994. Epidemic in cattle and humans of Salmonella typhimurium DT 104 with chromosomally integrated multiple drug resistance. Vet. Rec. 134:577.
Threlfall, E. J., C. J. Teale, R. H. Davies, L. R. Ward, J. A. Skinner, A. Graham, C. Cassar, and K. Speed. 2003. A comparison of antimicrobial susceptibilities in nontyphoidal salmonellas from humans and food animals in England and Wales in 2000. Microb. Drug Resist. 9:183-189.
Threlfall, E. J., L. R. Ward, and B. Rowe.1978 . Epidemic spread of a chloramphenicol-resistant strain of Salmonella typhimurium phage type 204 in bovine animals in Britain. Vet. Rec. 103:438-440.
Threlfall, E. J., L. R. Ward, and B. Rowe.1997 . Increasing incidence of resistance to trimethoprim and ciprofloxacin in epidemic Salmonella typhimurium DT104 in England and Wales. Euro. Surveill. 2:81-84.
Threlfall, E. J., L. R. Ward, and B. Rowe.1978 . Spread of multiresistant strains of Salmonella typhimurium phage types 204 and 193 in Britain. Br. Med. J. 2:997.
Volokhov, D., V. Chizhikov, K. Chumakov, and A. Rasooly. 2003. Microarray-based identification of thermophilic Campylobacter jejuni, C. coli, C. lari, and C. upsaliensis. J. Clin. Microbiol. 41:4071-4080.
Volokhov, D., A. Rasooly, K. Chumakov, and V. Chizhikov. 2002. Identification of Listeria species by microarray-based assay.J. Clin. Microbiol. 40:4720-4728.
Walker, R. A., E. Lindsay, M. J. Woodward, L. R. Ward, and E. J. Threlfall. 2001. Variation in clonality and antibiotic-resistance genes among multiresistant Salmonella enterica serotype typhimurium phage-type U302 (MR U302) from humans, animals, and foods. Microb. Drug Resist. 7:13-21.
Wells, J. G., B. R. Davis, I. K. Wachsmuth, L. W. Riley, R. S. Remis, R. Sokolow, and G. K. Morris. 1983. Laboratory investigation of hemorrhagic colitis outbreaks associated with a rare Escherichia coli serotype. J. Clin. Microbiol. 18:512-520.
Wybot, I., C. Wildemauwe, C. Godard, S. Bertrand, and J. M. Collard. 2004. Antimicrobial drug resistance in nontyphoid human Salmonella in Belgium: trends for the period 2000-2002. Acta Clin. Belg. 59:152-160.
Yang, Y. H., S. Dudoit, P. Luu, D. M. Lin, V. Peng, J. Ngai, and T. P. Speed. 2002. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucleic Acids Res. 30:e15.(C. Pelludat, R. Prager, H)
Robert Koch Institut, 38855 Wernigerode, Germany
MicroDiscovery GmbH, Marienburger Strasse 1, 10405 Berlin, Germany
ABSTRACT
In developed countries, Salmonella enterica subspecies 1 serovars Enteritidis and Typhimurium range among the most common causes of bacterial food-borne infections. The surveillance and typing of epidemic Salmonella strains are important tools in epidemiology. Usually, Salmonella enterica subspecies 1 serovars are differentiated by serotyping for diagnostic purposes. Further differentiation is done by phage typing as well as molecular typing techniques. Here we have designed and evaluated a prototype DNA microarray as a tool for serovar Typhimurium strain differentiation. It harbors 83 serovar Typhimurium probes obtained by differential subtractive hybridization and from the public database. The microarray yielded reproducible hybridization patterns in repeated hybridizations with chromosomal DNA of the same strain and could differentiate five serovar Typhimurium reference strains (DT204, DT104, DT208, DT36, and LT2). Furthermore, the microarray identified two distinct groups among 13 serovar Typhimurium DT104 strains. This correlated with observations from pulsed-field gel electrophoresis analysis. Twenty-three further serovar Typhimurium strains were analyzed to explore future directions for optimization of the simple 83-probe DNA microarray. The data presented here demonstrate that DNA microarrays harboring small numbers of selected probes are promising tools for serovar Typhimurium strain typing.
INTRODUCTION
Salmonella spp. are pathogenic enterobacteria which can cause food-borne diseases ranging from a mild gastroenteritis to systemic infections (typhoid fever). In developed countries, infections with Salmonella enterica subspecies 1 serovars Enteritidis and Typhimurium range among the most common causes of bacterial food-borne infections.
For diagnostic purposes, Salmonella enterica subspecies 1 serovars are differentiated by serotyping (28), which allows reliable identification of organisms that cause infections associated with higher risks (i.e., serovar Typhi). For epidemiological surveys, detection of outbreaks, and uncovering of the chains of transmission, serotyping is insufficient because serovar Enteritidis and serovar Typhimurium often account for >80% of all human salmonelloses. For further differentiation, serotyping is complemented by phage typing as well as molecular typing techniques.
During the past decades serovar Typhimurium strains have been differentiated by phage typing (1). The Anderson phage typing scheme is based on the empirical observation that certain phages can cause lytic infections in one serovar Typhimurium strain but not another. A set of 34 bacteriophages allows more than 200 patterns of resistance or sensitivity, the "phage types" (definitive phage types [DTs]), to be distinguished. Phage typing has proven very useful for distinguishing many serovar Typhimurium strains. It has been invaluable for surveillance of the emergence and spread of epidemic serovar Typhimurium strains (i.e., DT204c and DT104) during the past decades (23, 40, 42-44, 49).
In spite of its success, phage typing has several drawbacks. (i) Phage typing has remained an empirical method, and there is only circumstantial evidence for the molecular mechanisms determining resistance or susceptibility to each Anderson typing phage. (ii) The power of resolution is limited. Several distinct serovar Typhimurium strains can belong to the same phage type. (iii) Phage typing relies on "unique" historical stocks which will not last forever. Replenishment of phage stocks will require propagation in Salmonella strains. Recent research has demonstrated that this can lead to recombination and mixed phage populations. Thus, the resulting phage preparations can show altered lysis patterns (36) and, consequently, can affect the consistency of the Anderson phage typing scheme.
Several molecular typing methods have been devised to differentiate serovar Typhimurium strains. This includes pulsed-field gel electrophoresis (PFGE), ribotyping, random amplification of polymorphic DNA, DNA fingerprinting, amplified length polymorphism analysis, enterobacterial repetitive intragenic consensus sequence typing, determination of outer membrane protein and multilocus enzyme patterns, and IS200 typing. Currently, these methods or combinations thereof are being employed successfully for outbreak investigations and identification of lines of transmission. In the past few years, the DNA microarray technology has been developed and is a potentially powerful tool for strain identification. DNA microarrays harboring probes for each open reading frame (ORF) of serovar Typhimurium strain LT2 (GenBank accession no. NC_003197) and additional probes derived from the serovar Typhi CT18 genome (GenBank accession no. NC_003198) and other Salmonella sequences from public databases have been designed (2, 15, 30). Hybridization of chromosomal DNA from standard Salmonella enterica laboratory collections allowed discrimination of a total of >100 strains of different S. enterica subspecies, various subspecies 1 serovars, and even different strains of the same serovar (29). However, so far no systematic studies have evaluated the suitability of DNA microarrays for the differentiation of serovar Typhimurium strains, e.g., for purposes of epidemiology.
Here, we describe a pilot study in which we assessed DNA microarrays as a tool for serovar Typhimurium strain differentiation. A DNA microarray was designed by using strain-specific probes obtained by differential subtractive hybridization and probes derived from public databases. Repeated hybridization experiments with a set of reference serovar Typhimurium strains demonstrated a high reproducibility of strain differentiation by this technique. Furthermore, we analyzed clinical serovar Typhimurium isolates of phage type DT104 and additional phage types. We found that the DNA microarray is useful for serovar Typhimurium strain differentiation. Implications for future applications and strategies for further optimization of the microarray are discussed.
MATERIALS AND METHODS
Bacterial strains. Serovar Typhimurium strain SL1344 has been described previously (39), strain DT104 (99-00971) was obtained from A. Hensel (Institut für Tierhygiene, Leipzig, Germany), and Vibrio cholerae El Tor Nent 720-95 was obtained from the Institut für Veterinre Bakteriologie, Bern, Switzerland.
Serovar Typhimurium isolates and serovar Typhi DNA (SARC2) were from the Robert Koch Institut, Wernigerode, Germany (Table 1). Chromosomal DNAs from Escherichia coli strain EDL933 and Yersinia enterocolitica strain 8081 were kindly provided by A. Rakin and S. Schubert (Max von Pettenkofer-Institut, Munich, Germany).
DNA manipulation. Chromosomal DNA of serovar Typhimurium strains DT104 (99-00971), DT204 (STM1690/75), DT208 (201IE888), DT36 (Anderson), and SARA2 (LT2) was prepared by standard methods (35). Chromosomal DNA was isolated from the strains studied by DNA hybridization, according to the protocol of Ausubel (3).
PFGE was performed as described previously (31).
SH and subcloning of the DNA fragments. Subtractive hybridization (SH) was performed by using the Clontech PCR-Select bacterial genome subtraction kit (PT3170-1; Clontech, Palo Alto, Calif.). Briefly, chromosomal serovar Typhimurium DNA (tester DNA) was digested with RsaI (four-base cutter, GTAC), purified, and divided into two aliquots. One aliquot was ligated to adapter 1, and the other was ligated to adapter 2R.
In a third tube, portions of each aliquot were combined (negative control for subtractive hybridization). Two rounds of hybridization with RsaI-digested DNA from another serovar Typhimurium strain (driver DNA) were performed at a temperature (48°C) lower than that recommended by the manufacturer (63°C) in order to increase the fraction of DNA fragments with low or no similarity between the driver and the tester DNAs. Then, a primary PCR and a secondary PCR with primers that anneal to adapters 1 and 2R were performed, and the PCR products were cloned into pCR2.1-Topo (Topo TA cloning kit; Invitrogen, Germany). Plasmids from lac-negative clones growing in Luria-Bertani medium (100 μg/ml ampicillin, 50 μg/ml kanamycin, isopropyl--D-thiogalactopyranoside, 5-bromo-4-chloro-3-indolyl--D-galactopyranoside) were analyzed by PCR (with primers T7 [5'-GTAATACGACTCACTATAGGGC-3'] and M13 reverse [5'-GGAAACAGCTATGACCATG-3']; annealing at 52°C and elongation for 1.5 min at 72°C for 35 cycles) and sequenced (AGOWA, Berlin, Germany). As probes for the microarray, we selected 500- to 800-nucleotide (nt)-long inserts (named A2 to G4 and 2SH4 to 3SH93; Table 2) showing <75% identity (blastn;www.ncbi.nlm.nih.gov/BLAST) over >50% of their entire sequence to serovar Typhimurium LT2, E. coli K-12, and Shigella genome sequences (http://www.ncbi.nlm.nih.gov/sutils/genom_table.cgi). Inserts with high similarities to mobile genetic elements (transposons, insertion sequence elements, phages) or pathogenicity islands were also included.
Microarray design and hybridization. Probes were generated by PCR (Table 2) and were analyzed by Scienion (Berlin, Germany) with a 2100 Bioanalyzer (Agilent Technologies) (data not shown). Eighty-three PCR products yielding a single band (Table 2) of the expected size were used for array design. Each probe was spotted onto two locations of the "amino slides" by using the Scienion processing system and cross-linked (300-mJ UV light; Stratalinker; Stratagene, The Netherlands). Quality control was performed by Scienion by SYBR Green staining of DNA and complex hybridization (with Cy3 and Cy5).
From each strain to be analyzed, 4 μg of chromosomal DNA was denatured, hybridized to random decamer primers, and labeled by primer extension by using Klenow polymerase and a deoxynucleoside triphosphate mix containing Cy5-dCTP. The labeled DNA was purified (MiniElute; QIAGEN), and the labeling efficiency was verified by gel electrophoresis. Array hybridization was performed in a formamide-based buffer (Sci-Hyb; Scienion) at 42°C for 20 h, followed by washing steps with increasing stringency (Sci-Wash; Scienion). The fluorescence signals (635-nm excitation) obtained with each probe were measured with an Axon 4100 scanner (16-bit; intensity range, 0 to 65,535), and the raw data were corrected for the background and processed with GenePix Pro V5.0 software. This was performed by Scienion.
Data processing. The data were transformed to log scale and normalized by quantile-based methods (37, 50). A two-component normal mixture model (24), p(x; m1, m2, s1, s2, w) = (1 – w) · n(x; m1, s1) + w · n2(x; m2, s2), was fitted to the normalized data by a maximum-likelihood method. The discriminant function (12) w · [n2(x)/p(x)] was used to represent the propensity of the sequence for being present (value close to 1) or absent (value close to 0) from the DNA of that strain. Discriminant values for all probes and all samples (strains) were calculated and stored in a signal probability matrix, with each column representing one array measurement and with each row representing the results obtained with one specific probe on the chip. For representation of the probability matrix, a coloring scheme was used: blue indicates state 1 (off), yellow indicates state 2 (on), and dark shading indicates indecisive measurements (p value of about 0.5).
Strain differentiation was studied as follows: pairwise correlation coefficients (16) were calculated from the signal probability values from each pair of hybridization experiments. The pairwise correlation coefficients were stored in a sample correlation matrix. For visualization of the results, the sample correlation matrix was displayed as a color scale ranging from green (weakly correlated) to red (strongly correlated) (adapted from the "sma" package [13]).
Dendrograms were calculated based on the signal probability matrix by using Euclidean distance and the average linkage method (12, 38). All calculations were implemented by using the R-statistics package (22).
RESULTS
Probes for microarray design. There is considerable sequence diversity between different Salmonella subspecies and serovars (30). Much of this diversity is attributable to the loss or acquisition of DNA fragments by horizontal gene transfer (14). Even between different serovar Typhimurium strains, some DNA fragments do differ: variable sequences include those of antibiotic resistance genes, plasmids, and phages (17, 19, 26, 29, 47). Detection of these DNA fragments should allow serovar Typhimurium strain differentiation at high resolution with fairly simple microarrays composed of <200 probes. To test this hypothesis, we have designed a DNA microarray that includes probes from (i) Salmonella DNA fragments retrieved by SH experiments and (ii) sequences derived from sequences in a public database (National Center for Biotechnology Information [http://www.ncbi.nlm.nih.gov/]).
DNA microarray probes obtained by SH. SH allows the enrichment or retrieval of sequences present in one bacterial genome (designated the "tester") but absent from another (designated the "driver") (see Materials and Methods). These sequences would provide ideal probes for DNA microarray design.
The DNA of isolates of three serovar Typhimurium strains (DT204 = STM1690/75, DT104 = 99-00971, and DT208 = 201 IE 888; Table 1) known to differ significantly from each other (W. Rabsch, R. Prager, and H. Tschpe, unpublished observations) were chosen as testers. DT204 had caused a major epidemic (23, 42, 44) in the 1970s and 1980s and carries the SopE prophage, which encodes the type III effector gene sopE, in its chromosome (26, 27). Serovar Typhimurium DT104 (99-00971) is a representative of the antibiotic multiresistant phage type DT104 strains which emerged in several European countries as the predominant serovar Typhimurium phage type in human infections (31, 43, 49). Outbreaks have also been reported in the United States (5, 6, 34). Serovar Typhimurium DT208 (201IE888) is a representative of phage type DT208, which caused an outbreak in the former Soviet Union but which is also identified in current epidemiological surveys (11, 41).
We used DNA from the serovar Typhimurium strains mentioned above and from DT36 (Anderson), a strain highly susceptible to all Anderson typing phages. Tester-driver combinations were DT204 (STM1690/75)-DT104 (99-00971), DT208 (201IE888)-DT204 (STM1690/75), DT104 (99-00971)-DT36 (Anderson), DT204 (STM1690/75)-DT36 (Anderson), and DT208 (201IE888)-DT36 (Anderson). The hybridization steps were performed at 48°C to ensure enrichment of the tester DNA fragments with no significant similarity to the driver genome. DNA fragments were sequenced, and we selected 43 plasmids harboring 500- to 1,600-nt DNA inserts with no or limited similarity to known Salmonella or enterobacterial sequences for probe design (see Materials and Methods). Inserts with high similarities to mobile genetic elements (transposons, insertion sequence elements, phages) or pathogenicity islands were also included. Upon PCR amplification, 38 of these plasmids yielded unique products and were selected for DNA microarray design (Table 2; see Table S1 in the supplemental material). Nine of these probes originated from DT104 (99-00971), 24 originated from DT204 (STM1690/75), and 5 originated from DT208 (201IE888) (Table 2).
DNA microarray probes derived from prophages. The repertoire of prophages residing in the genome is highly diverse between different serovar Typhimurium strains (8, 17, 26). Unique sequences (immunity region, morons) which are present in one prophage but not in most others should be well suited for strain differentiation. PCR products corresponding to 21 unique prophage sequences from serovar Typhimurium strains, serovar Typhi, and serovar Paratyphi A (33) were selected for DNA microarray design (Table 2; see Table S2 in the supplemental material). Three probes were derived from the phage-encoded Shiga-like toxin A subunit (E. coli EDL 933; O157:H7) and cholera toxin A and B subunits (Vibrio cholerae El Tor Nent 720-95). It was unclear whether one of these toxins might also be present in some serovar Typhimurium strains.
DNA microarray probes from TTSSs. Type III secretion systems (TTSSs), their known effector genes, and chaperones are prerequisites for Salmonella virulence (18). The TTSSs and most of the effector genes are highly conserved between all Salmonella strains (25, 30). We designed probes based on sopE2, sopB, sipA, slrP, invB, invC, and prgH as positive controls (Table 2; see Table S2 in the supplemental material).
Effector genes sopE (probe D2 from SH) and avrA are more variable (20, 21, 26, 32) and were also included.
Additional probes for DNA microarray design. Additional probes were based on serovar Typhimurium SL1344 and DT204 (3341/78) DNA fragments retrieved in an earlier study (Table 2, probes pM48, K58, K51, and K9) (K. Ehrbar and W.-D. Hardt, unpublished data) and fimC (serovar Typhimurium LT2). Furthermore, open reading frames STM2240 (serovar Typhimurium LT2) and STY0312, STY1360, and STY1364 (serovar Typhi), which do not show similarity to other sequences in the public database, were selected for probe design (Table 2; see Table S2 in the supplemental material).
In an attempt to control the stringency of microarray hybridization probes E. coli frd (GenBank accession no.DH5_FRB; 90% identity to serovar Typhimurium LT2), E. coli kefc (GenBank accession no. DH5_KEFC; 83% identity to serovar Typhimurium LT2), E. coli agp (GenBank accession no. DH5_AGP; 80% identity to serovar Typhimurium LT2), and Yersinia enterocolitica thrc (GenBank accession no. Y_ENT_THRC; 76% identity to serovar Typhimurium LT2) (Table 2) were designed. The green fluorescent protein (gfp) (10) served as a negative control.
Microarray design. For microarray design, a total of 83 probes (250 to 1,500 nt long) were used, including 38 probes derived from subtractive hybridizations, 4 from earlier genetic screens, and 41 from the sequences in a public database (National Center for Biotechnology Information). Each probe was spotted onto two different locations of the amino slides. Probes E10, pM816, D6, and G3 each were present in four different spots on the microarray. A total of 100 microarrays were produced in two separate batches. Quality control (see Material and Methods) and hybridization experiments with identical DNA samples did not detect differences in the performance of the two different batches (data not shown).
Hybridization with five serovar Typhimurium reference strains. We performed repeated DNA microarray hybridization experiments to analyze the reproducibility and explore the quality of strain differentiation obtained by this method. For this purpose serovar Typhimurium DT36 (Anderson), DT204 (STM1690/75), DT104 (99-00971), DT208 (201IE888), and LT2 were chosen. The chromosomal DNA of one strain was isolated in two independent preparations, and each batch was hybridized with three DNA microarrays (hybridization data not shown). Each of the six hybridizations involved a separate labeling reaction. In this way we could determine whether technical problems associated with poorly reproducible steps of the hybridization procedure (see Materials and Methods) might interfere with DNA microarray-based strain typing.
Hybridization signals were transformed to log scale and normalized (see Materials and Methods). The signals obtained from pairs of identical probes on each array were very similar and showed typical correlation coefficients greater than 0.99 (data not shown). For all further calculations, the two signals were represented by their mean value.
Next, we used the hybridization signals to calculate the probability that a specific sequence is present in or absent from the strain. For this purpose, a two-component normal mixture model was fitted to the hybridization data by a maximum-likelihood method (see Materials and Methods). This yielded a "discriminant" function (see Fig. S1 in the supplemental material) which can use the hybridization signals to calculate for a probe the probability of a match (value close to 1) or no match (value close to 0) with the strain analyzed. The discriminant function was used to transform the 83 hybridization signals into a list of probabilities ("signal probability matrix") for the presence or absence of each probe (data not shown). To study strain differentiation, the signal probability list between each pair of DNA microarray hybridizations was compared. The resulting correlation coefficients were stored in a sample correlation matrix (see Materials and Methods). In Fig. 1, the sample correlation matrix for the analysis of the five representative serovar Typhimurium strains has been displayed in color code: red indicates strongly correlated, and green indicates weakly correlated.
The red diagonal is attributable to the strong correlation when each hybridization experiment is compared with itself. Besides this, the correlation matrix shows a chessboard pattern of strong and weak correlations. All hybridizations with DNA from the same strain yielded very similar results (large red areas on the chessboard). No similarity between the hybridization results from different serovar Typhimurium strains was observed. Therefore, the hybridization data are reproducible, and the prototype DNA microarray allows reliable differentiation of serovar Typhimurium DT36 (Anderson), DT204 (STM1690/75), DT104 (99-00971), DT208 (201IE888), and LT2.
Slight variations were observed within the group of six DNA microarray hybridizations performed with serovar Typhimurium DT36 (Anderson), as indicated by a shift to dark red in the correlation matrix. The same observation was made for serovar Typhimurium LT2. The reasons for this variation are unclear. However, strain differentiation was still clearly feasible.
The weakest correlation was observed between DT204 and the other strains, in particular, strains DT104 (99-00971) and DT208 (201IE888). This is at least partially attributable to the large number of DT204 (STM1690/75)-specific probes (24 in total; Table 2) present on the DNA microarray. In addition, DT204 (STM1690/75) failed to hybridize with many probes specific for the other strains (data not shown).
The hybridization patterns were also compared in a dendrogram. For this purpose, the signal probability matrix was analyzed by using Euclidean distance and the average linkage method (see Materials and Methods). The topology of the resulting dendrogram (Fig. 2) supported the results from the correlation analysis (Fig. 1). The branches from the six hybridizations of each serovar Typhimurium strain clustered closely together. The branch representing DT204 (STM1690/75) diverges early, while the branches for DT104 (99-00971) and DT208 (201IE888) cluster closely together.
It should be noted that this dendrogram does not provide information on the phylogenetic relationship between the strains. However, it is helpful for the recognition of groups of similar hybridization patterns. Throughout the rest of this study we have employed the dendrogram calculation in order to group the strains before plotting the color-coded pairwise correlation matrices (see below).
Taken together the results from this first series of hybridization experiments provided evidence for the reproducibility of the data and the feasibility of DNA microarray-based differentiation of serovar Typhimurium strains.
Microarray hybridization with various DT104 typed strains. The data presented above showed that the simple prototype DNA microarray could differentiate serovar Typhimurium strains which belong to different phage types. Next, we were interested in finding out whether different serovar Typhimurium isolates of the same phage type would yield similar hybridization results. To analyze this, we have chosen 13 serovar Typhimurium DT104 strains (Tables 3 and 4, strains 02-00822, 02-02682, 02-03881, 03-06256, 97-01178, 97-01193, 97-01864, 97-02103, 97-01174, 02-00429, 02-03122, 96-01186, and 97-05652). Three independent DNA microarray hybridizations (I, II, III) were performed with each strain, as described above. The DT104 (99-00971) reference strain was also hybridized once more.
The hybridization data were analyzed as described above. For better comparison, the results for the five reference strains shown in Fig. 1 and 2 have been included in this analysis (color coded in blue, grey, pink, marine, and yellow). The dendrogram (Fig. 3) revealed that all branches representing DT104 strains were clearly separated from the branches of reference strains serovar Typhimurium DT36 (Anderson), DT204 (STM1690/75), DT208 (201IE888), and LT2.
Interestingly, the 13 DT104 strains formed two clearly separated branches: subgroup 1 strains (02-00822, 02-02682, 02-03881, 03-06256, 97-01178, 97-01193, 97-01864, 97-02103, and 97-01174) were very similar to reference strain DT104 (99-00971). Subgroup 2 strains (02-00429, 02-3122, 96-01186, and 97-05652) formed a separate branch less similar to the DT104 reference strain.
The correlation analysis confirmed the finding of two DT104 subgroups (Fig. 4).
To explore these results in more detail, the 13 serovar Typhimurium DT104 strains were analyzed by PFGE (Fig. 5). All strains of subgroup 2 had a very similar PFGE pattern (Fig. 5, lanes 2 to 5). This pattern differed significantly (more than five bands) from that of the typical DT104 strains associated with the current epidemic (Fig. 5, lanes 11 to 15) (31) and from the additional subgroup 1 strains (Fig. 5, lanes 6 to 8 and 10).
In conclusion, the DNA microarray allowed the reliable and reproducible identification of different isolates of the same phage type. Moreover, it could differentiate two groups of DT104 strains. This was unexpected, because the DNA microarray did not specifically include probes designed for this purpose. However, this observation suggested that more sophisticated subdifferentiation of serovar Typhimurium DT104 strains might be feasible by including additional probes on the next microarray generation.
Microarray hybridization with various phage-typed strains. So far, our study had focused on the differentiation of strains representing five different phage types. However, the number of serovar Typhimurium strains in nature is much higher. Even the Anderson phage typing scheme is able to differentiate >200 different types of serovar Typhimurium strains. In order to explore the potential and the limitations of the current prototype microarray, we have studied strains representing various phage types. Here, each strain was hybridized only once. Therefore, the data can provide general insights, but they are not statistically significant. Strains belonging to phage types DT204 (00-02379, 00-07509, 00-09603, 00-10022, and 99-07382), DT193 (02-01414, 02-04395, 02-04515), DT120 (03-01606, 03-01663, and 03-01860), DT92 (03-00009, 03-0008), DT12 (03-6841, 03-07846), DT17 (03-01862), DT177 (03-05510), DT186 (03-5115) DTu302 (03-6382), DT7 (03-6600), DT9 (03-01779), DT170 (04-0093), DT2 (04-0110), DT175 (305-70), DT68 (11635), and DT40 (6772/96) were analyzed as described above. For comparison, data for the five reference strains (Fig. 1 and 2) have been included (color coded in blue, grey, pink, marine, and yellow). As expected, all five reference strains were located in distinct branches of the dendrogram (Fig. 6). The vast majority of the other serovar Typhimurium strains clustered in three distinct branches of the dendrogram (termed X, Y, and Z; Fig. 6). These branches were clearly distinct from those of the five reference serovar Typhimurium strains. The hybridization patterns of two strains, DT193 (02-01414) and DT177 (03-05510), differed from those of all of the other strains analyzed. This indicated that the prototype DNA microarray can differentiate at least 11 different types of serovar Typhimurium strains, the five reference strains (Fig. 1); DT104 subgroup 2 (Fig. 3 and 4); clusters X, Y, and Z; and the two strains DT193 (02-01414) and DT177 (03-05510).
The five additional DT204 strains (00-02379, 00-07509, 00-09603, 00-10022, and 99-07382) clustered closely together (Fig. 6, cluster X) and were located close to the DT204 (STM1690/75) reference strain. However, they were more closely related to each other than to the DT204 reference strain. This suggested that the DNA microarray can subdifferentiate at least some DT204 strains. Further experiments are required to substantiate this hypothesis.
Interestingly, the DT68 (11635) strain, which carries the SopE prophage (26), was also located in the same branch as the DT204 strains. Similarly, 03-6382 (Dtu302) clustered closely with the DT104 (99-00971) reference strain. Further studies are needed to determine whether strains DT68 (11635) and DT204 or strains Dtu302 (03-6382) and DT104 may have a recent common ancestry.
Cluster Y harbored both DT92 isolates and both DT12 isolates. The three DT120 isolates were located in cluster Z (Fig. 6). Thus, strains of these three phage types might be recognized specifically by the prototype DNA microarray and differentiated from the five reference serovar Typhimurium strains. Further experiments are needed to substantiate this hypothesis.
The three DT193 isolates (02-01414, 02-04395, and 02-04515) were located in separate branches of the dendrogram, implying that this DT group includes heterogeneous serovar Typhimurium strains which can be differentiated by the DNA microarray. Overall, clusters Y and Z harbored quite diverse groups of strains, as judged by phage typing. It is likely that the prototype DNA microarray is simply insufficient to differentiate between these strains. In this case, additional probes would be required in order to improve strain differentiation by future generations of a serovar Typhimurium typing microarray.
Evaluation of the probes chosen for microarray design. For the future design of improved serovar Typhimurium typing arrays, it would be useful to analyze in more detail the performance of each probe on the current DNA microarray. For this purpose, the signal probability matrix (see Materials and Methods) for each probe in each hybridization experiment has been evaluated (see Fig. S2 in the supplemental material). Twelve DNA probes matched every serovar Typhimurium strain analyzed. This included most probes derived from TTSSs as well as probes E6 (transposase), A7 (prophage ST64T), A8 (transposase), A6 (SPI3), and B4 (100% identity to LT2) obtained by subtractive hybridization. Twelve DNA probes did not match any of the serovar Typhimurium strains analyzed. This included the negative control (GFP); prophage-encoded genes for CTXA, CTXB, and EDL-STXA; the serovar Paratyphi B sequences Paratyphi-Contig 8 and Paratyphi-Contig 35; the serovar Typhi sequences Typhi-STY1360, Typhi-STY0312, and Typhi-STY4629; Y. enterocolitica THRC; DH5-AGP; and DH5-KEFC. Some of these probes might be omitted in future. However, a few "positive and negative controls" should be retained. It might be of interest to retain probes GFP (truly negative control), Y_ENT_THRC (76% identity to LT2), DH5_AGP (80% identity to LT2), and DH5_KEFC (83% identity to LT2). In contrast to these "absent" probes, control probe DH5_FRB (90% identity) yielded matches in almost all hybridization experiments. Thus, these probes can provide a useful control for the general stringency of DNA microarray hybridization. In addition, these observations indicate that DNA fragments less than 83% identical between two serovar Typhimurium strains can be employed on future DNA microarrays.
Several probes derived from the serovar Typhimurium LT2 (10 probes) and serovar Typhi CT18 (6 probes) genome sequences contributed to strain differentiation. This pertained especially to probes derived from unique prophage sequences. Thus, additional completed Salmonella genomes will provide promising sources for the design of additional probes.
The vast majority (30 of 38) of probes derived by subtractive hybridization clearly contributed to strain differentiation. Due to the possibility to experimentally enrich DNA fragments that differ between two strains of choice, subtractive hybridization will provide a powerful tool by which additional probes for improved DNA microarray-based strain differentiation can be obtained.
DISCUSSION
Fine differentiation of serovar Typhimurium strains is of great importance for epidemiology and the identification of chains of transmission. Phage typing and several molecular typing techniques are routinely used for this purpose. Potentially, microarrays offer a powerful alternative for phage typing. In fact, chip-based strain typing has been demonstrated for Listeria spp. and Campylobacter spp. (45, 46), and a DNA chip representing each serovar Typhimurium LT2 open reading frame was found to differentiate strains from Salmonella reference collections (29). Here, we report on the first pilot study to use simple DNA microarrays for serovar Typhimurium strain differentiation. The prototype DNA microarray harbored 83 probes. It yielded reproducible hybridization patterns in repeated hybridizations with the same strain and could differentiate five reference strains (Fig. 1 and 2), two subgroups of serovar Typhimurium DT104 (Fig. 3 to 5), two types of DT204 strains, as well as three clusters of serovar Typhimurium wild-type isolates (Fig. 6). This demonstrated that DNA microarrays harboring small numbers of selected probes are promising tools for serovar Typhimurium strain differentiation.
Our study included merely 44 serovar Typhimurium strains. This included the most prevalent serovar Typhimurium strains from Germany, but it is just a small fraction of the existing serovar Typhimurium strains with distinct genetic makeups (>>200). Thus, it is conceivable that the prototype DNA microarray can differentiate even more serovar Typhimurium strains than is evident from this study. More hybridization experiments would be required to address this point.
Future generations of the DNA microarray should include additional probes that will allow the accurate discrimination between those strains not differentiated by the current DNA microarray (i.e., subgroup 1 DT104 strains [Fig. 5, lanes 6 to 8 and 10] and strains of clusters Y and Z [Fig. 6]) or strains which are going to emerge in the future.
How many strain-specific probes are required Optimal strain differentiation was achieved for the reference strain DT204 (STM1690/75; Fig. 1). This strain yielded strong correlations in repeated hybridizations but was clearly distinct from other DT204 strains (Fig. 6) and yielded only a weak correlation with all other reference strains (Fig. 1). The DNA microarray harbored 31 probes that yielded positive hybridization signals with DT204 (STM1690/75) but not with one or more of the other strains. Twenty-one of these "useful" probes were derived from DT204 (STM1690/75) by subtractive hybridization. This number of probes represents an estimate for the upper limit, since all differences are expected to show up in this experimental setup. Sensible strain differentiation can probably be achieved with fewer strain-specific probes. Reference strain DT104 (99-00971) provides an example: only six of the nine probes derived from DT104 (99-00971) by subtractive hybridization yielded positive hybridization signals with DT104 (99-00971) but not with one or more of the other strains. Still, the prototype microarray could differentiate DT104 (99-00971) from all other reference strains and could even distinguish between two different subgroups of DT104 strains (Fig. 4).
How can the prototype DNA microarray be improved Clearly, additional DNA probes that are present in some serovar Typhimurium strains but absent from others are needed. As discussed above, DNA fragments of several hundred base pairs <83% identical between different strains will be useful for strain typing. It should be noted that (in contrast to many other DNA array studies) it has proven useful to focus on DNA similarity and not on open reading frames. Thus, many probes obtained by subtractive hybridization represent partial open reading frames or noncoding DNA.
In theory, optimal efficiency would be achieved if each of the probes would recognize a different subset of serovar Typhimurium strains. More precisely, an optimal (orthogonal) set of probes would show a vanishing correlation for each pair of probes or samples. Currently, there are two main strategies for acquisition of additional probes: ongoing sequencing projects which include several serovar Typhimurium strains and strains from other S. enterica subspecies 1 serovars will provide rich sources for probe generation. Alternatively, subtractive hybridization can be used. This method has provided many useful probes in this study. It is fairly inexpensive and allows the enrichment of DNA sequences present in one strain but absent from another. Therefore, it is ideally suited for the identification of probes which distinguish between strains which are not clearly distinguished by the prototype DNA microarray (i.e., different DT104 strains of subgroup 1 [Fig. 4 and 5] and different strains of clusters Y and Z [Fig. 6]).
In conclusion, this pilot study has demonstrated that DNA microarrays are useful tools for serovar Typhimurium strain differentiation. Most likely, strains from other S. enterica subspecies 1 serovars can be differentiated in a similar manner. If one considers that the technology for chip production, hybridization, and analysis has become widely available in recent times, DNA microarray strain typing holds many promises for the fast and efficient detection of chains of transmission and for epidemiology in general. However, it should be noted that today's high costs for DNA microarray production and the labeling must drop significantly before this technology will be widely applied in diagnostic and reference laboratories.
ACKNOWLEDGMENTS
We thank Alexander Rakin, Sren Schubert, and Martin Braun for discussion, strains, and chromosomal DNA and Holger Eickhoff and Sandra Prietz (Scienion) for excellent service in microarray production and hybridization.
This work was funded in part by a grant from the Kompetenznetzwerk "Genomforschung an pathogenen Bakterien" Verbund 3 (BMBF, Germany) to W.-D. Hardt.
Supplemental material for this article may be found at http://jcm.asm.org.
REFERENCES
Anderson, E. S., L. R. Ward, M. J. Saxe, and J. D. de Sa. 1977. Bacteriophage-typing designations of Salmonela typhimurium. J. Hyg. (London) 78:297-300.
Andrews-Polymenis, H. L., W. Rabsch, S. Porwollik, M. McClelland, C. Rosetti, L. G. Adams, and A. J. Baumler.2004 . Host restriction of Salmonella enterica serotype Typhimurium pigeon isolates does not correlate with loss of discrete genes. J. Bacteriol. 186:2619-2628.
Ausubel, F. M. 1997. Current protocols in molecular biology. John Wiley & Sons, Inc., New York, N.Y.
Beltran, P., S. A. Plock, N. H. Smith, T. S. Whittam, D. C. Old, and R. K. Selander.1991 . Reference collection of strains of the Salmonella typhimurium complex from natural populations.J. Gen Microbiol. 137:601-606.
Besser, T. E., C. C. Gay, J. M. Gay, D. D. Hancock, D. Rice, L. C. Pritchett, and E. D. Erickson. 1997. Salmonellosis associated with S. typhimurium DT104 in the USA. Vet. Rec. 140:75.
Besser, T. E., M. Goldoft, L. C. Pritchett, R. Khakhria, D. D. Hancock, D. H. Rice, J. M. Gay, W. Johnson, and C. C. Gay. 2000. Multiresistant Salmonella Typhimurium DT104 infections of humans and domestic animals in the Pacific Northwest of the United States. Epidemiol. Infect. 124:193-200.
Boyd, E. F., F. S. Wang, T. S. Whittam, and R. K. Selander. 1996. Molecular genetic relationships of the Salmonellae. Appl. Environ. Microbiol. 62:804-808.
Brussow, H., C. Canchaya, and W. D. Hardt. 2004. Phages and the evolution of bacterial pathogens: from genomic rearrangements to lysogenic conversion. Microbiol. Mol. Biol. Rev. 68:560-602.
Carniel, E., I. Guilvout, and M. Prentice. 1996. Characterization of a large chromosomal "high-pathogenicity island" in biotype 1B Yersinia enterocolitica.J. Bacteriol. 178:6743-6751.
Cormack, B. P., R. H. Valdivia, and S. Falkow.1996 . FACS-optimized mutants of the green fluorescent protein (GFP). Gene 173:33-38.
Davies, R. H., R. Dalziel, J. C. Gibbens, J. W. Wilesmith, J. M. Ryan, S. J. Evans, C. Byrne, G. A. Paiba, S. J. Pascoe, and C. J. Teale. 2004. National survey for Salmonella in pigs, cattle and sheep at slaughter in Great Britain (1999-2000).J. Appl. Microbiol. 96:750-760.
Duda, R. O., and P. E. Hart. 1973. Pattern classification and scene analysis. John Wiley & Sons, Inc., New York, N.Y.
Dudoit, S., Y. H. Yang, M. J. Callow, and T. P. Speed 2000, posting date. Statistical methods for identifying differentially expressed genes in replicated cDNA microarray experiments. Technical report 578. [Online.]. http://www.stat.berkeley.edu/users/terry/zarray/Html/matt.html.
Edwards, R. A., G. J. Olsen, and S. R. Maloy.2002 . Comparative genomics of closely related salmonellae.Trends Microbiol. 10:94-99.
Eriksson, S., S. Lucchini, A. Thompson, M. Rhen, and J. C. Hinton.2003 . Unravelling the biology of macrophage infection by gene expression profiling of intracellular Salmonella enterica.Mol. Microbiol. 47:103-118.
Feller, W. 1950. An introduction to probability theory and its applications. John Wiley & Sons, Inc., New York, N.Y.
Figueroa-Bossi, N., S. Uzzau, D. Maloriol, and L. Bossi. 2001. Variable assortment of prophages provides a transferable repertoire of pathogenic determinants in Salmonella. Mol. Microbiol. 39:260-272.
Galan, J. E. 2001. Salmonella interactions with host cells: type III secretion at work. Annu. Rev. Cell Dev. Biol. 17:53-86.
Guerra, B., E. Junker, A. Miko, R. Helmuth, and M. C. Mendoza.2004 . Characterization and localization of drug resistance determinants in multidrug-resistant, integron-carrying Salmonella enterica serotype Typhimurium strains. Microb. Drug Resist. 10:83-91.
Hardt, W. D., and J. E. Galan. 1997. A secreted Salmonella protein with homology to an avirulence determinant of plant pathogenic bacteria. Proc. Natl. Acad. Sci. USA 94:9887-9892.
Hardt, W. D., H. Urlaub, and J. E. Galan.1998 . A substrate of the centisome 63 type III protein secretion system of Salmonella typhimurium is encoded by a cryptic bacteriophage. Proc. Natl. Acad. Sci. USA 95:2574-2579.
Ihaka, R., and R. Gentleman. 1996. R: a language for data analysis and graphics. J. Comput. Graph. Stat. 5:299-314.
Kuhn, H., W. Rabsch, H. Tschape, and E. Tietze. 1982. Characterization and epidemiology of a Salmonella typhimurium epidemic strain. Z. Arztl. Fortbild (Jena) 76:607-610. (In German.)
McLachlan, G. J., and D. Peel. 2000. Finite mixture models. Wiley series in probability and statistics, John Wiley and Sons, Inc., New York, N.Y.
Mirold, S., K. Ehrbar, A. Weissmüller, R. Prager, H. Tschpe, H. Rüssmann, and W. D. Hardt. 2001. Salmonella host cell invasion emerged by acquisition of a mosaic of separate genetic elements, including Salmonella pathogenicity island 1 (SPI1), SPI5, and sopE2. J. Bacteriol. 183:2348-2358.
Mirold, S., W. Rabsch, M. Rohde, S. Stender, H. Tschape, H. Russmann, E. Igwe, and W. D. Hardt. 1999. Isolation of a temperate bacteriophage encoding the type III effector protein SopE from an epidemic Salmonella typhimurium strain. Proc. Natl. Acad. Sci. USA 96:9845-9850.
Pelludat, C., S. Mirold, and W. D. Hardt. 2003. The SopEPhi phage integrates into the ssrA gene of Salmonella enterica serovar Typhimurium A36 and is closely related to the Fels-2 prophage. J. Bacteriol. 185:5182-5191.
Popoff, M. Y., J. Bockemuhl, and A. McWhorter-Murlin.1992 . Supplement 1991 (no. 35) to the Kauffmann-White scheme. Res. Microbiol. 143:807-811.
Porwollik, S., E. F. Boyd, C. Choy, P. Cheng, L. Florea, E. Proctor, and M. McClelland. 2004. Characterization of Salmonella enterica subspecies I genovars by use of microarrays. J. Bacteriol. 186:5883-5898.
Porwollik, S., R. M. Wong, and M. McClelland. 2002. Evolutionary genomics of Salmonella: gene acquisitions revealed by microarray analysis. Proc. Natl. Acad. Sci. USA 99:8956-8961.
Prager, R., A. Liesegang, W. Rabsch, B. Gericke, W. Thiel, W. Voigt, R. Helmuth, L. Ward, and H. Tschape. 1999. Clonal relationship of Salmonella enterica serovar typhimurium phage type DT104 in Germany and Austria. Zentbl. Bakteriol. Parasitenkd. Infektkrankh. Hyg. Abt. 1 Orig. 289:399-414.
Prager, R., S. Mirold, E. Tietze, U. Strutz, B. Knuppel, W. Rabsch, W. D. Hardt, and H. Tschape. 2000. Prevalence and polymorphism of genes encoding translocated effector proteins among clinical isolates of Salmonella enterica. Int. J. Med. Microbiol. 290:605-617.
Prager, R., W. Rabsch, W. Streckel, W. Voigt, E. Tietze, and H. Tschape.2003 . Molecular properties of Salmonella enterica serotype Paratyphi B distinguish between its systemic and its enteric pathovars. J. Clin. Microbiol. 41:4270-4278.
Rabatsky-Ehr, T., J. Whichard, S. Rossiter, B. Holland, K. Stamey, M. L. Headrick, T. J. Barrett, and F. J. Angulo.2004 . Multidrug-resistant strains of Salmonella enterica Typhimurium, United States, 1997-1998. Emerg. Infect. Dis. 10:795-801.
Sambrook, J., E. F. Fritsch, and T. Maniatis. 1989.Molecular cloning: a laboratory manual , 2nd ed. Cold Spring Harbor Laboratory Press, Cold Spring Harbor, N.Y.
Schmieger, H. 1999. Molecular survey of the Salmonella phage typing system of Anderson. J. Bacteriol. 181:1630-1635.
Schuchhardt, J., D. Beule, A. Malik, E. Wolski, H. Eickhoff, H. Lehrach, and H. Herzel. 2000. Normalization strategies for cDNA microarrays. Nucleic Acids Res. 28:E47.
Steinhausen, D., and K. Langer. 1977. Clusteranalyse: Einführung in Methoden und Verfahren der automatischen Klassifikation: mit zahlreichen Algorithmen, FORTRAN-Programmen, Anwendungsbeispielen und einer Kurzdarstellung der multivarianten statistischen. Verfahren, 1. Erste Editierte Auflage de Gruyter, Berlin, Germany.
Stocker, B. A., S. K. Hoiseth, and B. P. Smith. 1983. Aromatic-dependent Salmonella spp. as live vaccine in mice and calves. Dev. Biol. Stand. 53:47-54.
Threlfall, E. J., J. A. Frost, L. R. Ward, and B. Rowe. 1994. Epidemic in cattle and humans of Salmonella typhimurium DT 104 with chromosomally integrated multiple drug resistance. Vet. Rec. 134:577.
Threlfall, E. J., C. J. Teale, R. H. Davies, L. R. Ward, J. A. Skinner, A. Graham, C. Cassar, and K. Speed. 2003. A comparison of antimicrobial susceptibilities in nontyphoidal salmonellas from humans and food animals in England and Wales in 2000. Microb. Drug Resist. 9:183-189.
Threlfall, E. J., L. R. Ward, and B. Rowe.1978 . Epidemic spread of a chloramphenicol-resistant strain of Salmonella typhimurium phage type 204 in bovine animals in Britain. Vet. Rec. 103:438-440.
Threlfall, E. J., L. R. Ward, and B. Rowe.1997 . Increasing incidence of resistance to trimethoprim and ciprofloxacin in epidemic Salmonella typhimurium DT104 in England and Wales. Euro. Surveill. 2:81-84.
Threlfall, E. J., L. R. Ward, and B. Rowe.1978 . Spread of multiresistant strains of Salmonella typhimurium phage types 204 and 193 in Britain. Br. Med. J. 2:997.
Volokhov, D., V. Chizhikov, K. Chumakov, and A. Rasooly. 2003. Microarray-based identification of thermophilic Campylobacter jejuni, C. coli, C. lari, and C. upsaliensis. J. Clin. Microbiol. 41:4071-4080.
Volokhov, D., A. Rasooly, K. Chumakov, and V. Chizhikov. 2002. Identification of Listeria species by microarray-based assay.J. Clin. Microbiol. 40:4720-4728.
Walker, R. A., E. Lindsay, M. J. Woodward, L. R. Ward, and E. J. Threlfall. 2001. Variation in clonality and antibiotic-resistance genes among multiresistant Salmonella enterica serotype typhimurium phage-type U302 (MR U302) from humans, animals, and foods. Microb. Drug Resist. 7:13-21.
Wells, J. G., B. R. Davis, I. K. Wachsmuth, L. W. Riley, R. S. Remis, R. Sokolow, and G. K. Morris. 1983. Laboratory investigation of hemorrhagic colitis outbreaks associated with a rare Escherichia coli serotype. J. Clin. Microbiol. 18:512-520.
Wybot, I., C. Wildemauwe, C. Godard, S. Bertrand, and J. M. Collard. 2004. Antimicrobial drug resistance in nontyphoid human Salmonella in Belgium: trends for the period 2000-2002. Acta Clin. Belg. 59:152-160.
Yang, Y. H., S. Dudoit, P. Luu, D. M. Lin, V. Peng, J. Ngai, and T. P. Speed. 2002. Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucleic Acids Res. 30:e15.(C. Pelludat, R. Prager, H)