In vivo and in silico determination of essential genes of Campylobacter jejuni
© Metris et al; licensee BioMed Central Ltd. 2011
Received: 18 August 2011
Accepted: 1 November 2011
Published: 1 November 2011
In the United Kingdom, the thermophilic Campylobacter species C. jejuni and C. coli are the most frequent causes of food-borne gastroenteritis in humans. While campylobacteriosis is usually a relatively mild infection, it has a significant public health and economic impact, and possible complications include reactive arthritis and the autoimmune diseases Guillain-Barré syndrome. The rapid developments in "omics" technologies have resulted in the availability of diverse datasets allowing predictions of metabolism and physiology of pathogenic micro-organisms. When combined, these datasets may allow for the identification of potential weaknesses that can be used for development of new antimicrobials to reduce or eliminate C. jejuni and C. coli from the food chain.
A metabolic model of C. jejuni was constructed using the annotation of the NCTC 11168 genome sequence, a published model of the related bacterium Helicobacter pylori, and extensive literature mining. Using this model, we have used in silico Flux Balance Analysis (FBA) to determine key metabolic routes that are essential for generating energy and biomass, thus creating a list of genes potentially essential for growth under laboratory conditions. To complement this in silico approach, candidate essential genes have been determined using a whole genome transposon mutagenesis method. FBA and transposon mutagenesis (both this study and a published study) predict a similar number of essential genes (around 200). The analysis of the intersection between the three approaches highlights the shikimate pathway where genes are predicted to be essential by one or more method, and tend to be network hubs, based on a previously published Campylobacter protein-protein interaction network, and could therefore be targets for novel antimicrobial therapy.
We have constructed the first curated metabolic model for the food-borne pathogen Campylobacter jejuni and have presented the resulting metabolic insights. We have shown that the combination of in silico and in vivo approaches could point to non-redundant, indispensable genes associated with the well characterised shikimate pathway, and also genes of unknown function specific to C. jejuni, which are all potential novel Campylobacter intervention targets.
The rise of antibiotic resistance in pathogenic bacteria is a growing concern in the developed world necessitating knowledge-led approaches to identify new interventions and prevention strategies . One of the common sources of pathogenic bacteria is food, with the foodborne zoonotic pathogens Salmonella, Escherichia coli and Campylobacter being prime examples. Although it can be contended whether the use of antibiotics in the food industry contributes to antimicrobial resistance, it is clear that food-borne pathogens also increasingly acquire resistance to antimicrobial interventions. Multidrug resistance in Salmonella is well documented [2, 3]. For Listeria, antibiotic resistance has also been reported for strains isolated from food . In Campylobacter, resistance to ampicillin, erythromycin, tetracycline, and ciprofloxacin have all been reported [5–7].
In Europe, Campylobacter was the most frequent cause of food-borne illness in 2007, with over 200,000 laboratory-confirmed cases  although the total number of cases is thought to be approximately eightfold higher. Infection by Campylobacter is thought to be largely due to the consumption of contaminated poultry either through poor food preparation hygiene or under-cooking . While the symptoms associated with C. jejuni infection (diarrhoea, vomiting, and stomach pains) often only last between 2 to 5 days, sequelae of C. jejuni infection include more serious autoimmune diseases like Guillain-Barré syndrome, Miller-Fisher syndrome , and reactive arthritis . While human infection often does not require antibiotic intervention, the organism is endemic in poultry and farm animals, and it would be advantageous to have treatment options before entry in the food chain.
One approach for the identification of new antibiotic targets for a particular bacterial pathogen is to identify non-redundant cellular functions or metabolic pathways that are indispensible for growth and/or survival of that organism; for example, key metabolic enzymes or cell wall synthesis proteins. In the post-genomic era, genome analysis makes both bioinformatic predictions and targeted mutagenesis strategies feasible, due to the availability of large, curated datasets. However, genome annotation is often incomplete and incorrect, and metabolic redundancy (alternative pathways or catalytic activities) can confound such rational approaches. For instance, a comprehensive study in Salmonella of essential genes required during infection showed that many enzymes are not essential, partly due to metabolic redundancy . An alternative (experimental, high-throughput) strategy is the use of random approaches such as transposon mutagenesis to identify essential genes that would be needed to infect an animal model or to grow and proliferate [12–14].
In silico, essential genes of well characterised micro-organisms, such as E. coli, are predicted with high accuracy by Flux Balance Analysis (FBA) of the metabolic network . FBA consists of the computation of the fluxes going through the metabolic reactions when the cells are in a homeostatic state . The reactions are linked to the genes that encode the corresponding enzymes with Boolean relationships, and, a gene is predicted to be essential if in silico deletion results in negligible biomass . According to Feist et al., , about 90% of essential genes of E. coli can be predicted in a given environment. For micro-organisms other than the established model systems, the accuracy is lower (60-80%) [17, 18]. Nevertheless, it provides insights into cellular metabolism, which can be useful to identify potential new drug targets .
In this study we have constructed a genome scale metabolic model of the food-borne pathogen Campylobacter jejuni and discuss this organism's metabolism. Additionally, we have combined both in silico and in vivo approaches to make predictions about essential genes. A published FBA model of the closely related organism Helicobacter pylori provided the basis for a Campylobacter FBA model. The essential genes predicted from FBA of the reconstructed genome scale model of C. jejuni were compared to new experimentally generated transposon mutagenesis data, and a recently published independent whole genome transposon mutagenesis study . Whilst the overlap between the three methods is comparatively small, the combination of these methods has drawn focus to the shikimate pathway, a known target pathway for new antimicrobial interventions in other bacterial pathogens .
Metabolic network reconstruction
Number of reactions in the C. jejuni model (this paper) and in the H. pylori model .
Type of reaction
C. jejuni model
H. pylori model
The reactions of the central metabolism have been mainly drawn from the literature since contrarily to H. pylori, C. jejuni is predicted to contain a complete TCA cycle with some enzymes characteristic for anaerobes . C. jejuni does not metabolise glucose  and the genome annotation suggests that the Embden-Meyerhof pathway only functions in gluconeogenesis , so overall the space of solution of fluxes is different from H. pylori. The respiratory chain of C. jejuni is more complex than that of H. pylori, and C. jejuni can use sulphite as an electron donor .
The biosynthesis of folate results in the production of glycolaldehyde, which can cause cell damage by electrophilic attack of negatively-charged molecules . In the H. pylori and E. coli models, glycolaldehyde is converted to glycolate by a glycolaldehyde dehydrogenase and glycolate is degraded further through glyoxylate metabolism [16, 18]. Alternatively glycolate can diffuse out of the cell in the case of E. coli. Since a glyoxylate oxidase (Cj1213c) has been annotated  and since glyoxylate has been shown to be degraded via a malate synthase in H. pylori, we assumed a malate synthase in our model despite the genome lacking the corresponding annotation as in H. pylori[30, 31].
Amino acid metabolism
The reactions for the amino acid metabolism have been mainly drawn from the genome annotation. Contrarily to H. pylori, C. jejuni seems to have the capacity to synthesize all the amino acids and vitamins it requires. This was shown experimentally with BIOLOG phenotype microarrays, where respiration was detected on growth medium containing salts and a carbon source only . From the genome sequence, complete pathways for the synthesis of isoleucine, leucine and valine, are present in C. jejuni, whereas these pathways are incomplete or absent in H. pylori. In addition to the amino acids required by H. pylori, a complete pathway to synthesize histidine was found in C. jejuni (cj0317, cj1315c, cj1597-99, cj1600-01, cj1603-04) while orthologs of these genes are absent in the H. pylori genome. The gene for the last step of methionine synthesis is predicted to be present in C. jejuni (cj1201), so no demand reaction was included in the model to artificially consume S-adenosylmethione. In common with H. pylori, only one gene of the methionine salvage pathway was found to be present in C. jejuni (cj0117). In the iIT341 GSM/GPR H. pylori model, the pathway was nevertheless included to ensure the recycling of 5-methylthioadenosine, a by-product of spermidine biosynthesis, to methionine, based on the assumption that the pathways vary from one micro-organism to another. However it has recently been suggested that the last steps of spermidine biosynthesis in C. jejuni differ from the pathway proposed for H. pylori eliminating the necessity for recycling 5-methylthioadenosine, so the methionine salvage pathway was not included in this model.
The reactions for the nucleotide metabolism have been almost exclusively derived from the genome annotation. The pathway for the synthesis of IMP is more similar to that of E. coli, rather than that of H. pylori. Only a few genes have been annotated at the level of nucleotide inter-conversions such as cj0293 which has been predicted to encode for a nucleotidase . However most of the products of the reactions catalysed by this enzyme are not reutilised in the metabolic network, they are dead-ends. So either cj0293 is incorrectly annotated, or genes encoding enzymes to utilize the products of the nucleotidase have not been annotated.
Vitamin and cofactor metabolism
In the H. pylori model iIT341 GSM/GPR, it was assumed that pimelate diffuses into the cell and that the first step of the synthesis of biotin is catalysed by pimelyl-CoA synthetase although no locus was found for such a gene . Having no better alternative, we kept the assumption of the H. pylori model. C. jejuni does not contain ubiquinone, but uses menaquinone 6 and a methyl-substituted menaquinone . It has been shown that for Streptomyces coelicolor, H. pylori, C. jejuni and Thermus thermophilus, the pathway for menaquinone synthesis diverges from the one from E. coli with futalosine as an intermediate . However the pathway is not completely elucidated so in this model, the equations of the pathway of E. coli were kept as in the model iIT341 GSM/GPR. As for the ubiquinone, the same three gene orthologs of the genes present in H. pylori have been predicted to be present in C. jejuni. Thiamine is essential for the growth of some strains of Campylobacter but not for the strain NCTC11168 . According to the genome annotation, the pathway for its synthesis is complete.
Cell wall metabolism
There is little evidence about the composition of the cell wall of C. jejuni in the literature, and the genome annotation suggests that the pathways are neither those of E. coli nor those of H. pylori. For instance, only 2 genes are annotated as part of the fucose biosynthesis pathway in C. jejuni (cj1407c and cj1428c), while 6 such genes are annotated in H. pylori. However in the absence of better data, the pathways for the cell wall metabolism were copied from the model iIT341 GSM/GPR. It has been shown that the fatty acid composition of C. jejuni changes with the environmental conditions and/or the growth rate [37, 38] but the predominant fatty acid have been reported to be the saturated hexadecanoic acid, the unsaturated octodec-1l-enoic acid and to a lesser extent cyclopropane in C19 and tetradecanoic acid which increased with stress [37–40]. These are the same fatty acids that are included in the model iIT341 GSM/GPR.
It is not clear from the genome sequence nor from the literature how C. jejuni assimilates sulphur, and hence we have used the assumptions used for H. pylori. Transport reactions were mainly drawn from circumstantial evidence. Three "sink reactions" were added to the model because the pathways for the degradation of the corresponding products are unknown, and the ones used are the same as the ones introduced in E. coli and H. pylori.
Exploration of the predicted metabolism of C. jejuni
To test the model, the production of biomass from different carbon sources was simulated and the results are shown in the table in the Additional file 2: substrate utilisation. Out of the 19 substrates tested, three are not included in the model (bromosuccinate, methyl pyruvate, α-hydroxybutyrate) and for four of them, it is not clear from the literature whether they can be used as a sole carbon source or not. According to the model, glutamate, citrate, α-ketoglutarate, aspartate, asparagine, L-lactate, L and D-malate, succinate, fumarate, pyruvate and serine can be metabolised in agreement with the data of the literature. The model also allowed the metabolism of proline and L-glutamine, which did not produce significant respiration with the BIOLOG experiments , however they were metabolised once aspartate and serine were depleted, alongside other chemicals in the defined media . It is not clear whether formate can be used as an electron donor only  or a carbon source as well [42, 43], but cannot be used as a sole carbon source according to the microarray experiments  and is not predicted to be metabolised in the model. Cysteine has been shown to be metabolised [37, 42] but was not predicted to be sufficient as a sole carbon source. Finally, no regulatory constraint was considered in any of the simulations, the only constraints used were the rate of consumption of the carbon source and oxygen source.
In silico prediction and experimental identification of essential genes in C. Jejuni
The metabolic model in combination with FBA was used to predict the metabolic genes that are essential for the production of biomass in rich medium (see Materials and Methods). This generated a list of 176 predicted essential genes (see Additional file 3: predictions of essential genes).
Summary of the genome wide transposon mutagenesis for C. jejuni
Number of inserts measured
Number of inserts in genes
Percentage of inserts within genes
Number of genes with 1 or more inserts
Number of genes with no insert
List of essential "metabolic genes" according to the different techniques
cj0024, cj0026c, cj0066c, cj0075c, cj0127c, cj0146c, cj0172c, cj0187c, cj0194, cj0196c, cj0197c, cj0205, cj0227, cj0237, cj0240c, cj0273, cj0274, cj0286c, cj0288c, cj0296c, cj0297c, cj0298c, cj0306c, cj0307, cj0321, cj0326, cj0332c, cj0360, cj0384c, cj0405, cj0432c, cj0433c, cj0434, cj0435, cj0437, cj0443, cj0453, cj0490, cj0541, cj0542, cj0559, cj0576, cj0580c, cj0585, cj0589, cj0638c, cj0647, cj0699c, cj0716, cj0764c, cj0766c, cj0767c, cj0795c, cj0798c, cj0806, cj0813, cj0821, cj0822, cj0847, cj0853c, cj0858c, cj0862c, cj0891c, cj0905c, cj0918c, cj0932c, cj0947c, cj0949c, cj0955c, cj0992c, cj0995c, cj1039, cj1044c, cj1046c, cj1048c, cj1067, cj1080c, cj1081c, cj1088c, cj1096c, cj1104, cj1114c, cj1133, cj1149c, cj1150c, cj1151c, cj1152c, cj1177c, cj1183c, cj1196c, cj1198, cj1202, cj1213c, cj1238, cj1243, cj1248, cj1364c, cj1398, cj1400c, cj1401c, cj1402c, cj1404, cj1407c, cj1424c, cj1428c, cj1476c, cj1498c, cj1515c, cj1529c, cj1530, cj1605c, cj1634c, cj1641, cj1645, cj1672c, cj1685c
FBA & transposon mutagenesis of this study
cj0027, cj0116, cj0117, cj0324, cj0387, cj0394c, cj0442, cj0514, cj0516, cj0581, cj0597, cj0639c, cj0641, cj0686, cj0840c, cj0894c, cj0927, cj1008c, cj1045c, cj1054c, cj1058c, cj1082c, cj1131c, cj1250, cj1274c, cj1288c, cj1346c, cj1347c, cj1366c, cj1403c, cj1443c, cj1536c, cj1607, cj1610, cj1652c
FBA & transposon mutagenesis of Stahl and Stintzi
cj0132, cj0147c, cj0308c, cj0356c, cj0451, cj0503c, cj0722c, cj0845c, cj0861c, cj0925, cj1201, cj1239, cj1290c, cj1291c, cj1531, cj1535c
FBA & the 2 transposon mutagenesis methods
cj0231c, cj0545, cj0707, cj0810, cj0855, cj0895c, cj1644, cj1676
One important caveat of the essential gene predictions using FBA is that some pathways were copied from H. pylori. In addition many genes annotated to be part of the cell wall were not taken into account in the model. However many genes involved in fatty acid metabolism and all the genes involved in fatty acid elongation are predicted to be essential by FBA: cj0328c (FabH), cj1303 (FabH2), cj0442 (FabF), cj0435 (FabG), cj0273 (FabZ), and cj1400c (FabI). FabD (cj0116) and FabF are also predicted essential from the transposon study. Fatty acid biosynthesis in Campylobacter is likely to contribute to the biosynthesis of the cell envelope, so it is perhaps not surprising that this is a key pathway.
Comparison of the essential gene predictions with a published study
In this study we have combined bioinformatic approaches to construct and validate a genome-wide model of metabolism of Campylobacter jejuni, the first such model of this important food-borne pathogen. Flux Balance Analysis has been used to predict those proteins that, if removed from the model, result in loss of biomass production. To complement this in silico predictive approach, we have used random transposon mutagensis coupled to gene-specific PCR to identify those genes that contain one or more transposon insert (dispensable genes for growth under laboratory conditions) and those genes that do not contain a transposon insert, the putative essential genes.
In silico determination of essential genes
Although the reconstruction of the metabolic network of C. jejuni is based on limited biochemical data, it was possible to formulate a hypothesis on the metabolism of this pathogen. The reconstruction pointed out the main areas of uncertainty: the cell wall metabolism and nucleotide pathways. It was also found that the pathway for sulphur assimilation is not obvious from the genome annotation.
A malate synthase activity was an assumption in our model, and this activity has been demonstrated in H. pylori. Based on the annotation, the genome does not encode a malate synthase, and extensive BLAST searching using both the malate synthase A (aceB) and malate synthase G (glcB) sequences did not reveal any match in the H. pylori or C. jejuni genomes or any genome from the epsilon sub-division of Proteobacteria. A new class of malate synthase enzymes may be present in H. pylori and possibly in this clade of life, that does not have sequence homology to known characterised malate synthase enzymes from other bacteria, although this requires further biochemical evidence. The reconstruction of the model was mainly based on conventional genome annotation employing BLAST searches. More sophisticated annotation methods have been proposed to address functional divergence amongst proteins that share sequence similarity [50, 51]. For instance, we compared the EC numbers obtained with the EFICAz  and PRIAM  tools to the EC numbers of the reactions linked to a unique gene in our model (287 reactions) and found discrepancies for 30 and 27 reactions respectively. Based on the original genome annotation, some enzymes in our model could catalyse diverse reactions, while the more sophisticated annotation tools suggested more metabolic specificity. An example is cj0324, originally annotated as a ubiquinone/menaquinone methyltransferase (EC 2.1.1.-) , the PRIAM tool  suggests specifically demethylmenaquinone methyltransferase activity (EC 22.214.171.124), which is more likely as menaquinone and a methyl substituted menaquinone have been isolated in C. jejuni rather than ubiquinone . We also checked our metabolic model against the 'expert community' subsystem annotation presented in The SEED , which returned discrepancies for 20 of the reactions considered above. For some genes, the precise annotation depends on the method used: returning to our glycolate to glyoxylate interconversion hypothesis, Cj1213c is a putative glycolate oxidase subunit D (EC 126.96.36.199) [23, 55], or an alkylglycerone-phosphate synthase (EC 188.8.131.52) , or a D-lactate dehydrogenase (EC 184.108.40.206)  making the degradation of glycolate into glyoxylate an uncertain assumption.
An inherent limitation of the FBA method is the suitability of the objective function . For instance, by optimising the biomass, FBA does not take into account the microaerophilic and capnophilic properties of C. jejuni. These may indeed constitute additional constraints like a maximum concentration of oxygen-sensitive enzymes neglected in these simulations except as modelled by the limiting uptake rate of oxygen. Alternatively, trade-off functions may be more appropriate objective function than the optimisation of the biomass with these kind of micro-organisms .
The FBA method has the potential of being condition specific to determine the essential genes. In this study, they were determined in laboratory conditions. However as more data become available on the conditions in the chicken gut, the model has the potential of being used in situations relevant to the food industry.
Flux balance analysis methods have a good track record of predicting essential genes , however, they only focus on metabolism-related genes. In this study, the FBA model only contained reactions linked to 388 genes, 24% of the total genome. In vitro transposition has the advantage of targeting the whole genome. We describe the construction of two transposon mutant libraries in C. jejuni NCTC11168 and the mapping of a total of 9550 inserts in the genome, this represents a coverage of 5.94× using the method of Stahl and Stintzi . Data from the combined transposon mutagenesis libraries predicted 233 essential genes (14% of the genome total). While FBA only uses a subset of genes from the genome, the number of predicted essential genes was similar at 175 (11% of the genome). These numbers are similar to the published C. jejuni study (194 genes, 12% of the genome) although the overlap between the published study and the data presented in this study is comparatively small (only 8 genes predicted by all three methods). Compared to other published essential gene prediction studies, the number of predicted essential genes is between the lower quartile and median with respect to total number of essential genes and percentage of the genome predicted to be essential. However, reviewing all published microbial essential gene predictions, we noted there was no correlation between number of predicted essential genes and genome size (data not shown). Clearly the relationship between genome size, complexity of niche, and indispensable genes is complex, plus a number of caveats should always be considered when interpreting this sort of data: any gene containing at least one insert can be said to be non-essential under the growth conditions described. However, the inverse logic is not true. The absence of an insert in a gene does not necessarily mean that the gene is required for growth and hence essential. Transposon insertion may not have occurred for a variety of reasons: chance, sequence bias of the transposase or transposon depletion during the reaction. Although no detailed studies of the sequence preference of either transposases used here have been carried out and it is generally assumed they are essentially random, we used two different transposases in an attempt to minimise any effect of sequence bias. An over-representation of small hypothetical proteins (and accordingly, small genes lacking an insert) may have resulted from the random nature of the transposon insertion: i.e. the smaller the gene, the smaller the chance of transposon insertion; however, no gene size bias was observed when comparing genes with insert with genes encompassing the entire genome.
The genome of C. jejuni NCTC 11168 totals 1,641,481 bases and thus our libraries represent insertions in only ~0.006% of the possible positions. It is possible that the number of inserts identified is an under representation of the actual total since inserts close together would generate very similar sized PCR products that may fail to be discriminated on the agarose gels. Additionally, the abundance of individual mutants in the isolated pooled genomic DNA may also affect whether a band is visible. Since the genomic DNA used was isolated from pooled colonies, it is possible that any mutant resulting in reduced growth and hence colony size would be under represented in the pooled material and as a consequence, would not be detected. The whole genome in vitro transposition presented in this study should be seen as a high-throughput method, as opposed to a high precision method. A number of caveats should not be ignored: the library unlikely represents all possible insertion points and some regions may be naturally more resistant to accepting an insert. Detection is constrained by the primer library, which in this case was optimized for microarray probe generation. The PCR and agarose gel-based approach also suffers from more common technical drawbacks such as smaller PCR products are more favourably amplified that longer ones and accuracy of sizing gel fragments is not infallible. Additionally, polar effects due to operon structure, may result in the null recovery of some mutants, if the transcription and translation of upstream genes is perturbed by a transposon insert, as was shown for the C. jejuni fur gene which in itself is not essential 
Functional vs. topological determination of essential genes
Another source of information which covers a high percentage of the genome is the Protein-Protein Interaction (PPI) network of C. jejuni, obtained by yeast two-hybrid methods which covers 80% of the proteome . Each protein is a node and if they interact, they are linked by an edge. Essential genes have been linked to the topological properties of the PPI network, as it has been shown that essential genes are more likely to be hubs of the PPI network than by chance [59, 60]. We have investigated whether there is a correlation between the degrees of the nodes the PPI network of C. jejuni and essential genes determined using FBA and transposon mutagenesis. No correlation was observed, contrarily to what was predicted by Parrish et al.. They based their analysis on putative essential proteins which were orthologs of Escherichia coli and Bacillus subtillis essential proteins. However, it has been shown that these bacteria do not share many essential genes, especially B. subtillis. More recent analyses of binary PPI networks suggest that the relationship between hubs and essential proteins is more complex, with hubs being correlated to genetic pleiotrophy; that is hubs are proteins that have many phenotypes when the gene encoding that protein is deleted . The interpretation of PPI networks remains ambiguous and models to explain the universal structure of PPI networks have been proposed to be related to evolutionary principles such as duplication and mutation of a few ancestors  or to the potential of proteins to bind together because of their physical properties, such as binding affinity and folding .
The shikimate pathway
Whilst the genome-wide comparison of gene essentiality with PPI hubs has not been fruitful in this study, the shikimate pathway in particular exhibits a large number of proteins with high degrees (see Figure 4). Since the interpretation of physical PPI network remains ambiguous, the high degrees of this pathway could be interpreted in multiple ways. Nonetheless, the combination of essential gene prediction methods has drawn focus to this particular pathway as a potential target for intervention, which should be investigated further using conventional genetic tools. The shikimate pathway has been the subject of antimicrobial research in previous studies [21, 64, 65]. As reported by other groups, the shikimate pathway is present in bacteria, plants, and fungi, but absent in humans, making it the target for novel antimicrobials and herbicides . More specifically, Zucko et al. show that a complete shikimate pathway is present in 76% of 442 bacterial genomes studied, although largely incomplete in Archaea . Two E. coli studies also identify essential genes from this pathway: aroH, aroK were predicted essential by Gerdes et al. and aroB, aroD, aroE, aroC, and pheA plus the entire trpABCDE operon were predicted essential by Joyce et al.. The aroD gene is also predicted to be essential in the refined H. pylori metabolic model .
It is noteworthy that without any prior expectations of pathway targets, the methods presented in this work point towards a known target pathway for novel antimicrobial interventions. However, the ultimate validation of our approach requires further laboratory investigation that is beyond the scope of this paper.
We have presented the first curated metabolic model of the important pathogen Campylobacter jejuni and discussed insights into the organism's metabolism. Flux Balance Analysis used in combination with a transposon mutagenesis library has been used to make predictions about essential genes, and these predictions have been further informed with reference to other published studies, such as the PPI dataset. This analysis has provided the basis for further laboratory investigations and suggests a re-evaluation of a previously scrutinized pathway, which may turn out to be the Achilles heel of this food-borne pathogen.
Reconstruction of the metabolic network for FBA
The reconstruction of the metabolic network is based on the genome sequence of C. jejuni NCTC 11168  and a recently curated and updated annotation . Where C. jejuni had H. pylori orthologs (they share about 2/3 of their genome), the reactions were taken from the H. pylori model iIT341 GSM/GPR  with the same assumptions. The reactions were also checked against on-line databases [69, 70] and also the literature on C. jejuni. In particular, the reactions for the central metabolism and respiration were drawn from a recent review . The conventions for the names of chemicals and reactions were kept as close as possible to the H. pylori model ilT341 GSM/GPR. The reactions added to the model were elementary and charge balanced based on a neutral intracellular pH. Where possible, the reactions were associated to genes that encode the proteins which catalyse them, with Boolean relationships. This means that for reactions catalysed by isozymes or different proteins, the "and", "or" Boolean operations between the genes was used .
Validation test for the model
To check that the model allowed the bacteria to metabolise the expected substrates, it was tested against BIOLOG microplates where the respiration of C. jejuni fed on different carbon source was measured , and, other literature data [37, 42, 43, 72]. Due to the scarcity of data in the literature, the model could only be tested with oxygen for respiration.
Prediction of essential genes by FBA
FBA consists of the computation of the possible fluxes, ν, going through the reactions of the metabolic network at steady-state. The system of equations is defined by the stoichiometric matrix, S, containing the stoichiometric coefficients of the metabolic reactions, with m being the number of metabolites, and n the number of fluxes . At steady-state (i.e. during balanced growth when the biomass composition is assumed to be constant), S.ν = 0. There are more fluxes than metabolites (n > m), so the system is underdetermined. The fluxes are bound by thermodynamic feasibility so the space of solutions is a convex space . It has been shown with some organisms, notably with E. coli, that the flux through the biomass is optimised for the uptake of nutrients during balanced growth, so the biomass equation can be used as an objective function to reduce the space of solutions. With less well characterised micro-organisms like C. jejuni, the biomass composition is not known quantitatively. The biomass composition was assumed to be the same as for H. pylori except for (a) vitamin B6, which was added to the equation as the genome is predicted to encode the entire biosynthetic pathway bar one gene and (b) thiamine, for which the active form was assumed to be thiamine diphosphate rather than thiamine.
Exploration of the space of solution: Even after appointing an objective function, there may be more than one solution to the optimisation problem. These solutions are referred to as silent phenotypes as the growth rate is the same but the internal organisation of the fluxes is different . Due to the design of the algorithm, the solution returned by simplex linear programming generally minimizes the number of fluxes.
To estimate whether a gene is essential or not, the ratio, Gr, of the biomass flux when the gene is absent to the biomass flux when the gene is present was calculated in a given environment . A gene was considered essential if Gr <= 10-9 (arbitrary value).
Medium composition: Minimal medium was used to validate the model. Its substrates were derived from the BIOLOG medium experiments http://www.biolog.com. The experiments to screen for essential genes were carried out in Brucella medium which is a rich medium containing pancreatic digest of casein, peptic digest of animal tissues, dextrose, yeast extract, sodium chloride and sodium bisulfite . Since the composition of this medium is unknown, the medium was assumed to be similar to yeast extract , the composition used for simulations is indicated in the table in the Additional file 5: medium composition. The chemicals are allowed to enter or leave the system through exchange reactions. The exchange fluxes of the carbon source were fixed to a maximum of 20 mmol/g dry weight of biomass/h, which is close to a maximum uptake rate for E. coli and 5 mmol/g dry weight of biomass/h for oxygen, which is about a fourth of the maximum fluxes measured in air for E. coli as C. jejuni is microearophilic. The other nutrients present in the medium were assumed to be non-limiting with an arbitrary uptake higher boundary of 1,000 mmol/g dry weight of biomass/h.
Two in vitro transposition libraries were constructed using mariner transposase, essentially as described by Gaskin and van Vliet  and Tn7 transposase from New England Biolabs with C. jejuni NCTC11168 genomic DNA. These were introduced into C. jejuni NCTC 11168 cells by natural transformation  and plated onto Blood Agar Base no.2 (Oxoid) plates supplemented with 5% v/v defribinated horse blood and kanamycin 50 μg/ml. After ~48 hours incubation at 42°C under microaerophilic conditions (5% oxygen, 10% carbon dioxide, 85% nitrogen) colonies were pooled and genomic DNA extracted using QIAgen genomic-tips (QIAgen).
Mapping of transposon insertions
Genomic DNA from the pooled colonies was used as template in polymerase chain reactions using a transposon specific primer and individual gene specific primers. Briefly, 50 μl reactions were set up using 100 ng genomic DNA, 10-50 pmol of each primer and 25 μl of HotStarTaq mix (QIAgen). Cycling conditions were 95°C for 15 min, followed by 30 cycles of 95°C 30 sec, 50°C 30 sec, 72°C 90 sec and a final 72°C 15 min extension step. Aliquots from each reaction were run on 0.8% agarose gels, which were stained with ethidium bromide. Gel images were captured using a GeneDoc system (Anachem). The sizes of observed bands were calculated using Labimage (Kapelan Bio-Imaging GmbH) and this data was inputted into Excel (Microsoft). Based on the transposon study, 47 insertional inactivation mutants were created using conventional methods to validate the Tn7/Mariner findings.
The work was supported by the Institute Strategic Programme Grant of the BBSRC to the IFR.
- Amini S, Tavazoie S: Antibiotics and the post-genome revolution. Curr Opin Microbiol. 2011, 14: 513-518.PubMedView ArticleGoogle Scholar
- Parry CM, Threlfall EJ: Antimicrobial resistance in typhoidal and nontyphoidal salmonellae. Curr Opin Infect Dis. 2008, 21: 531-538.PubMedView ArticleGoogle Scholar
- Rowe B, Ward LR, Threlfall EJ: Multidrug-resistant Salmonella typhi: a worldwide epidemic. Clin Infect Dis. 1997, 24 (Suppl 1): S106-109.PubMedView ArticleGoogle Scholar
- Lo Nostro A, Pesavento G, Ducci B, Nieri D, Comodo N: Prevalence and antibiotic susceptibility of Listeria spp. isolated from raw meat and retail foods. Food Control. 2010, 21: 708-713.View ArticleGoogle Scholar
- Chatre P, Haenni M, Meunier D, Botrel MA, Calavas D, Madec JY: Prevalence and antimicrobial resistance of Campylobacter jejuni and Campylobacter coli isolated from cattle between 2002 and 2006 in France. J Food Prot. 2010, 73: 825-831.PubMedGoogle Scholar
- Deckert A, Valdivieso-Garcia A, Reid-Smith R, Tamblyn S, Seliske P, Irwin R, Dewey C, Boerlin P, McEwen SA: Prevalence and antimicrobial resistance in Campylobacter spp. isolated from retail chicken in two health units in Ontario. J Food Prot. 2010, 73: 1317-1324.PubMedGoogle Scholar
- Gormley FJ, Strachan NJ, Reay K, MacKenzie FM, Ogden ID, Dallas JF, Forbes KJ: Antimicrobial resistance profiles of Campylobacter from humans, retail chicken meat, and cattle feces. Foodborne Pathog Dis. 2010, 7: 1129-1131.PubMedView ArticleGoogle Scholar
- Westrell T, Ciampa N, Boelaert F, Helwigh B, Korsgaard H, Chríel M, Ammon A, Mäkelä P: Zoonotic infections in Europe in 2007: a summary of the EFSA-ECDC annual report. Euro Surveill. 2009, 14: pii = 19100Google Scholar
- Jacobs BC, van Belkum A, Endtz HP: Guillain-Barre' Syndrome and Campylobacter Infection. Campylobacter. Edited by: Nachamkin I, Szymanski CM, Blaser MJ. 2008, Washington, DC: American Society for Microbiology, 245-261. ThirdView ArticleGoogle Scholar
- Hannu T, Mattila L, Rautelin H, Pelkonen P, Lahdenne P, Siitonen A, Leirisalo-Repo M: Campylobacter-triggered reactive arthritis: a population-based study. Rheumatology (Oxford). 2002, 41: 312-318.View ArticleGoogle Scholar
- Becker D, Selbach M, Rollenhagen C, Ballmaier M, Meyer TF, Mann M, Bumann D: Robust Salmonella metabolism limits possibilities for new antimicrobials. Nature. 2006, 440: 303-307.PubMedView ArticleGoogle Scholar
- Akerley BJ, Lampe DJ: Analysis of gene function in bacterial pathogens by GAMBIT. Methods Enzymol. 2002, 358: 100-108.PubMedView ArticleGoogle Scholar
- Autret N, Charbit A: Lessons from signature-tagged mutagenesis on the infectious mechanisms of pathogenic bacteria. FEMS Microbiol Rev. 2005, 29: 703-717.PubMedView ArticleGoogle Scholar
- Mecsas J: Use of signature-tagged mutagenesis in pathogenesis studies. Curr Opin Microbiol. 2002, 5: 33-37.PubMedView ArticleGoogle Scholar
- Kauffman KJ, Prakash P, Edwards JS: Advances in flux balance analysis. Curr Opin Biotech. 2003, 14: 491-496.PubMedView ArticleGoogle Scholar
- Feist AM, Henry CS, Reed JL, Krummenacker M, Joyce AR, Karp PD, Broadbelt LJ, Hatzimanikatis V, Palsson BO: A genome-scale metabolic reconstruction for Escherichia coli K-12 MG1655 that accounts for 1260 ORFs and thermodynamic information. Mol Syst Biol. 2007, 3: 121-PubMed CentralPubMedView ArticleGoogle Scholar
- Nogales J, Palsson BO, Thiele I: A genome-scale metabolic reconstruction of Pseudomonas putida KT2440: iJN746 as a cell factory. BMC Syst Biol. 2008, 2: 79-PubMed CentralPubMedView ArticleGoogle Scholar
- Thiele I, Vo TD, Price ND, Palsson BO: Expanded metabolic reconstruction of Helicobacter pylori (iIT341 GSM/GPR): an in silico genome-scale characterization of single- and double-deletion mutants. J Bacteriol. 2005, 187: 5818-5830.PubMed CentralPubMedView ArticleGoogle Scholar
- Lee DS, Burd H, Liu J, Almaas E, Wiest O, Barabasi AL, Oltvai ZN, Kapatral V: Comparative genome-scale metabolic reconstruction and flux balance analysis of multiple Staphylococcus aureus genomes identify novel antimicrobial drug targets. J Bacteriol. 2009, 191: 4015-4024.PubMed CentralPubMedView ArticleGoogle Scholar
- Stahl M, Stintzi A: Identification of essential genes in C. jejuni genome highlights 5 hyper-variable plasticity regions. Funct Integr Genomics. 2011, 11: 241-257.PubMedView ArticleGoogle Scholar
- Ducati RG, Basso LA, Santos DS: Mycobacterial shikimate pathway enzymes as targets for drug design. Curr Drug Targets. 2007, 8: 423-435.PubMedView ArticleGoogle Scholar
- Gundogdu O, Bentley SD, Holden MT, Parkhill J, Dorrell N, Wren BW: Re-annotation and re-analysis of the Campylobacter jejuni NCTC11168 genome sequence. BMC Genomics. 2007, 8: 162-PubMed CentralPubMedView ArticleGoogle Scholar
- Parkhill J, Wren BW, Mungall K, Ketley JM, Churcher C, Basham D, Chillingworth T, Davies RM, Feltwell T, Holroyd S, et al: The genome sequence of the food-borne pathogen Campylobacter jejuni reveals hypervariable sequences. Nature. 2000, 403: 665-668.PubMedView ArticleGoogle Scholar
- Kelly DJ: Complexity and Versality in the Physiology and Metabolism of Campylobacter. Campylobacter. Edited by: Nachamkin I, Szymanski CM, Blaser MJ. 2008, Washington, DC: American Society for Microbiology, 41-61. ThirdView ArticleGoogle Scholar
- Velayudhan J, Kelly DJ: Analysis of gluconeogenic and anaplerotic enzymes in Campylobacter jejuni: an essential role for phosphoenolpyruvate carboxykinase. Microbiology. 2002, 148: 685-694.PubMedView ArticleGoogle Scholar
- Kelly DJ: The physiology and metabolism of Campylobacter jejuni and Helicobacter pylori. Symp Ser Soc Appl Microbiol. 2001, 16S-24S.Google Scholar
- Myers JD, Kelly DJ: A sulphite respiration system in the chemoheterotrophic human pathogen Campylobacter jejuni. Microbiology. 2005, 151: 233-242.PubMedView ArticleGoogle Scholar
- Glomb MA, Monnier VM: Mechanism of protein modification by glyoxal and glycolaldehyde, reactive intermediates of the Maillard reaction. J Biol Chem. 1995, 270: 10017-10026.PubMedView ArticleGoogle Scholar
- Pitson SM, Mendz GL, Srinivasan S, Hazell SL: The tricarboxylic acid cycle of Helicobacter pylori. Eur J Biochem. 1999, 260: 258-267.PubMedView ArticleGoogle Scholar
- Boneca IG, de Reuse H, Epinat JC, Pupin M, Labigne A, Moszer I: A revised annotation and comparative analysis of Helicobacter pylori genomes. Nucleic Acids Res. 2003, 31: 1704-1714.PubMed CentralPubMedView ArticleGoogle Scholar
- Tomb JF, White O, Kerlavage AR, Clayton RA, Sutton GG, Fleischmann RD, Ketchum KA, Klenk HP, Gill S, Dougherty BA, et al: The complete genome sequence of the gastric pathogen Helicobacter pylori. Nature. 1997, 388: 539-547.PubMedView ArticleGoogle Scholar
- Line JE, Hiett KL, Guard-Bouldin J, Seal BS: Differential carbon source utilization by Campylobacter jejuni 11168 in response to growth temperature variation. J Microbiol Methods. 2010, 80: 198-202.PubMedView ArticleGoogle Scholar
- Lee J, Sperandio V, Frantz DE, Longgood J, Camilli A, Phillips MA, Michael AJ: An alternative polyamine biosynthetic pathway is widespread in bacteria and essential for biofilm formation in Vibrio cholerae. J Biol Chem. 2009, 284: 9899-9907.PubMed CentralPubMedView ArticleGoogle Scholar
- Carlone GM, Anet FA: Detection of menaquinone-6 and a novel methyl-substituted menaquinone-6 in Campylobacter jejuni and Campylobacter fetus subsp. fetus. J Gen Microbiol. 1983, 129: 3385-3393.PubMedGoogle Scholar
- Hiratsuka T, Itoh N, Seto H, Dairi T: Enzymatic properties of futalosine hydrolase, an enzyme essential to a newly identified menaquinone biosynthetic pathway. Biosci Biotechnol Biochem. 2009, 73: 1137-1141.PubMedView ArticleGoogle Scholar
- Dickgiesser N, Czylwik D: Chemically defined media for auxotyping of Campylobacter jejuni. Zentralbl Bakteriol Mikrobiol Hyg A. 1985, 260: 57-64.PubMedGoogle Scholar
- Leach S, Harvey P, Wali R: Changes with growth rate in the membrane lipid composition of and amino acid utilization by continuous cultures of Campylobacter jejuni. J Appl Microbiol. 1997, 82: 631-640.PubMedView ArticleGoogle Scholar
- Martinez-Rodriguez A, Mackey BM: Physiological changes in Campylobacter jejuni on entry into stationary phase. Int J Food Microbiol. 2005, 101: 1-8.PubMedView ArticleGoogle Scholar
- Lambert MA, Patton CM, Barrett TJ, Moss CW: Differentiation of Campylobacter and Campylobacter-Like Organisms by Cellular Fatty-Acid Composition. Journal of Clinical Microbiology. 1987, 25: 706-713.PubMed CentralPubMedGoogle Scholar
- Moss CW, Kai A, Lambert MA, Patton C: Isoprenoid quinone content and cellular fatty acid composition of Campylobacter species. J Clin Microbiol. 1984, 19: 772-776.PubMed CentralPubMedGoogle Scholar
- Hoffman PS, Goodman TG: Respiratory Physiology and Energy-Conservation Efficiency of Campylobacter jejuni. Journal of Bacteriology. 1982, 150: 319-326.PubMed CentralPubMedGoogle Scholar
- Mohammed KA, Miles RJ, Halablab MA: The pattern and kinetics of substrate metabolism of Campylobacter jejuni and Campylobacter coli. Lett Appl Microbiol. 2004, 39: 261-266.PubMedView ArticleGoogle Scholar
- Westfall HN, Rollins DM, Weiss E: Substrate utilization by Campylobacter jejuni and Campylobacter coli. Appl Environ Microbiol. 1986, 52: 700-705.PubMed CentralPubMedGoogle Scholar
- Holmes K, Tavender TJ, Winzer K, Wells JM, Hardie KR: AI-2 does not function as a quorum sensing molecule in Campylobacter jejuni during exponential growth in vitro. BMC Microbiol. 2009, 9: 214-PubMed CentralPubMedView ArticleGoogle Scholar
- Palyada K, Sun YQ, Flint A, Butcher J, Naikare H, Stintzi A: Characterization of the oxidative stress stimulon and PerR regulon of Campylobacter jejuni. BMC Genomics. 2009, 10: 481-PubMed CentralPubMedView ArticleGoogle Scholar
- Reid AN, Pandey R, Palyada K, Naikare H, Stintzi A: Identification of Campylobacter jejuni genes involved in the response to acidic pH and stomach transit. Appl Environ Microbiol. 2008, 74: 1583-1597.PubMed CentralPubMedView ArticleGoogle Scholar
- Wright JA, Grant AJ, Hurd D, Harrison M, Guccione EJ, Kelly DJ, Maskell DJ: Metabolite and transcriptome analysis of Campylobacter jejuni in vitro growth reveals a stationary-phase physiological switch. Microbiology. 2009, 155: 80-94.PubMedView ArticleGoogle Scholar
- Parrish JR, Yu J, Liu G, Hines JA, Chan JE, Mangiola BA, Zhang H, Pacifico S, Fotouhi F, DiRita VJ, et al: A proteome-wide protein interaction map for Campylobacter jejuni. Genome Biol. 2007, 8: R130-PubMed CentralPubMedView ArticleGoogle Scholar
- Crossley RA, Gaskin DJ, Holmes K, Mulholland F, Wells JM, Kelly DJ, van Vliet AH, Walton NJ: Riboflavin biosynthesis is associated with assimilatory ferric reduction and iron acquisition by Campylobacter jejuni. Appl Environ Microbiol. 2007, 73: 7819-7825.PubMed CentralPubMedView ArticleGoogle Scholar
- Arakaki AK, Tian W, Skolnick J: High precision multi-genome scale reannotation of enzyme function by EFICAz. BMC Genomics. 2006, 7: 315-PubMed CentralPubMedView ArticleGoogle Scholar
- Claudel-Renard C, Chevalet C, Faraut T, Kahn D: Enzyme-specific profiles for genome annotation: PRIAM. Nucleic Acids Res. 2003, 31: 6633-6639.PubMed CentralPubMedView ArticleGoogle Scholar
- EFICAz. [http://cssb2.biology.gatech.edu/EFICAz/data/bac/cje_eficaz_4ec.html]
- PRIAM. [http://www.priam.prabi.fr/REL_MAY11/index_may11.html]
- Overbeek R, Begley T, Butler RM, Choudhuri JV, Chuang HY, Cohoon M, de Crecy-Lagard V, Diaz N, Disz T, Edwards R, et al: The subsystems approach to genome annotation and its use in the project to annotate 1000 genomes. Nucleic Acids Res. 2005, 33: 5691-5702.PubMed CentralPubMedView ArticleGoogle Scholar
- The Seed. [http://seed-viewer.theseed.org/seedviewer.cgi]
- Schuster S, Pfeiffer T, Fell DA: Is maximization of molar yield in metabolic networks favoured by evolution?. J Theor Biol. 2008, 252: 497-504.PubMedView ArticleGoogle Scholar
- Papp B, Teusink B, Notebaart RA: A critical view of metabolic network adaptations. HFSP J. 2009, 3: 24-35.PubMed CentralPubMedView ArticleGoogle Scholar
- van Vliet AH, Wooldridge KG, Ketley JM: Iron-responsive gene regulation in a campylobacter jejuni fur mutant. J Bacteriol. 1998, 180: 5291-5298.PubMed CentralPubMedGoogle Scholar
- Jeong H, Tombor B, Albert R, Oltvai ZN, Barabasi AL: The large-scale organization of metabolic networks. Nature. 2000, 407: 651-654.PubMedView ArticleGoogle Scholar
- Yu H, Kim PM, Sprecher E, Trifonov V, Gerstein M: The importance of bottlenecks in protein networks: correlation with gene essentiality and expression dynamics. PLoS Comput Biol. 2007, 3: e59-PubMed CentralPubMedView ArticleGoogle Scholar
- Yu H, Braun P, Yildirim MA, Lemmens I, Venkatesan K, Sahalie J, Hirozane-Kishikawa T, Gebreab F, Li N, Simonis N, et al: High-quality binary protein interaction map of the yeast interactome network. Science. 2008, 322: 104-110.PubMed CentralPubMedView ArticleGoogle Scholar
- Solé R, Pastor-satorras R, Smith E, Kepler T: A model of large-scale proteome evolution. Advances in Complex Systems. 2002, 5: 43-54.View ArticleGoogle Scholar
- Deeds EJ, Ashenberg O, Shakhnovich EI: A simple physical model for scaling in protein-protein interaction networks. Proc Natl Acad Sci USA. 2006, 103: 311-316.PubMed CentralPubMedView ArticleGoogle Scholar
- Davies GM, Barrett-Bee KJ, Jude DA, Lehan M, Nichols WW, Pinder PE, Thain JL, Watkins WJ, Wilson RG: (6S)-6-fluoroshikimic acid, an antibacterial agent acting on the aromatic biosynthetic pathway. Antimicrob Agents Chemother. 1994, 38: 403-406.PubMed CentralPubMedView ArticleGoogle Scholar
- Vianna CP, de Azevedo WF: Identification of new potential Mycobacterium tuberculosis shikimate kinase inhibitors through molecular docking simulations. J Mol Model. 2011Google Scholar
- Zucko J, Dunlap WC, Shick JM, Cullum J, Cercelet F, Amin B, Hammen L, Lau T, Williams J, Hranueli D, Long PF: Global genome analysis of the shikimic acid pathway reveals greater gene loss in host-associated than in free-living bacteria. BMC Genomics. 2010, 11: 628-PubMed CentralPubMedView ArticleGoogle Scholar
- Gerdes SY, Scholle MD, Campbell JW, Balazsi G, Ravasz E, Daugherty MD, Somera AL, Kyrpides NC, Anderson I, Gelfand MS, et al: Experimental determination and system level analysis of essential genes in Escherichia coli MG1655. J Bacteriol. 2003, 185: 5673-5684.PubMed CentralPubMedView ArticleGoogle Scholar
- Joyce AR, Reed JL, White A, Edwards R, Osterman A, Baba T, Mori H, Lesely SA, Palsson BO, Agarwalla S: Experimental and computational assessment of conditionally essential genes in Escherichia coli. J Bacteriol. 2006, 188: 8259-8271.PubMed CentralPubMedView ArticleGoogle Scholar
- KEGG. [http://www.genome.jp/kegg-bin/show_organism?menu_type=pathway_maps&org=cje]
- BioCyc. [http://biocyc.org/CJEJ192222/organism-summary]
- Reed JL, Famili I, Thiele I, Palsson BO: Towards multidimensional genome annotation. Nat Rev Genet. 2006, 7: 130-141.PubMedView ArticleGoogle Scholar
- Guccione E, Leon-Kempis Mdel R, Pearson BM, Hitchin E, Mulholland F, van Diemen PM, Stevens MP, Kelly DJ: Amino acid-dependent growth of Campylobacter jejuni: key roles for aspartase (AspA) under microaerobic and oxygen-limited conditions and identification of AspB (Cj0762), essential for growth on glutamate. Mol Microbiol. 2008, 69: 77-93.PubMedView ArticleGoogle Scholar
- Palsson BO, (Ed.): Systems Biology. 2006, Cambridge University Press
- Becker SA, Feist AM, Mo ML, Hannum G, Palsson BO, Herrgard MJ: Quantitative prediction of cellular metabolism with constraint-based models: the COBRA Toolbox. Nat Protoc. 2007, 2: 727-738.PubMedView ArticleGoogle Scholar
- Zimbro MJ, Power DA, (Eds.): Difco & BBL Manual. 2003
- Raghunathan A, Reed J, Shin S, Palsson B, Daefler S: Constraint-based analysis of metabolic capacity of Salmonella typhimurium during host-pathogen interaction. BMC Syst Biol. 2009, 3: 38-PubMed CentralPubMedView ArticleGoogle Scholar
- Varma A, Palsson BO: Stoichiometric flux balance models quantitatively predict growth and metabolic by-product secretion in wild-type Escherichia coli W3110. Appl Environ Microbiol. 1994, 60: 3724-3731.PubMed CentralPubMedGoogle Scholar
- glpk. [http://www.gnu.org/software/glpk/glpk.html]
- Gaskin DJ, van Vliet AH: Random mutagenesis strategies for Campylobacter and Helicobacter species. Methods Mol Biol. 2010, 634: 37-52.PubMedView ArticleGoogle Scholar