Ever since Bergh et al. (1989) published the numeric abundance of viruses in aquatic environments, viruses have been predicted to cause large fluctuations in global nutrient and biogeochemical cycles (Suttle, 2005; Danovaro et al., 2008; Falkowski et al., 2008). These global effects are primarily caused by microbial mortality following infection, which impacts microbial abundance and diversity (Suttle, 2005). As viruses are deprived of an own metabolism, they have to exploit their host cells’ resources to fuel viral replication during infection. To overcome possible bottlenecks in this process, many virus genomes have acquired host-derived genes with metabolic functions, which are hypothesized to reprogram the host’s metabolic processes to favor virus replication and are termed ‘auxiliary metabolic genes’ (AMGs) (Breitbart et al., 2007). The increased availability of metagenomic sequencing data over the past decade enabled identification of many AMGs in eukaryotic and prokaryotic viruses, with intervening roles in phosphate (Martiny et al., 2009) and nitrogen metabolism (Sullivan et al., 2010), photosynthesis (Lindell et al., 2005; Frank et al., 2013), the pentose phosphate pathway (Thompson et al., 2011), nucleic acid synthesis (Miller et al., 2003) and other metabolic pathways (Sharon et al., 2011). Currently, AMGs are known to be spread across nearly the entire central carbon metabolism (Hurwitz et al., 2013). In an attempt to organize the expanding list of AMGs, Hurwitz et al. (2015) proposed a subdivision into two classes. Class I contains annotations present in KEGG metabolic pathways, whereas Class II are all genes annotated only with a general metabolic function or absent from KEGG metabolic pathways (for example, transport functions).

Another feature of viral genomes is the high number of encoded open reading frames with no similarity to any known sequence (approximately 60% in marine viruses) (Paul and Sullivan, 2005), often referred to as the biological ‘dark matter’ (Rohwer and Youle, 2011). Especially early in infection, these small gene products of unknown function can influence the host metabolism through specific, often uncharacterized interactions, which are impossible to predict (Hatfull, 2008; Roucourt and Lavigne, 2009; Hargreaves et al., 2014). The identification of such viral–host interactions, and the viral genes involved, has proven to be labor-intensive research. However, recent advances in -omics techniques provide us with the tools to obtain novel insights into the host take-over process and to identify the affected host metabolic pathways (Van den Bossche et al., 2014; Sanchez et al., 2015).

In contrast to the increasing knowledge concerning the genetic potential of phages for host metabolism take-over, experimental data on the host metabolic changes caused by the phage infection and the influence of encoded AMGs on these changes are extremely scarce. Current advances in the field of metabolomics, triggered by the progress in mass spectrometry (Rabinowitz, 2007), allow analysis of multiple time points during infection with high metabolite coverage (Timischl et al., 2008). Ankrah et al. (2014) were the first to apply these approaches to investigate the phage infection of an environmentally relevant bacterium. Their results showed significant physiological differences between phage-infected and non-infected cells and raised the question if this response could be generalized to different phage–host systems. Furthermore, metabolomics has proven to be a valuable tool in functional genomics to characterize unknown proteins (Ehebauer et al., 2015). As enzyme inactivation leads to upstream accumulation of metabolites (Fendt et al., 2010), whereas protein activity modulation through post-translational modification or protein interaction also impacts metabolism (Oliveira et al., 2012; Schulz et al., 2014).

In this study, we selected the ubiquitous, Gram-negative bacterial pathogen, Pseudomonas aeruginosa, as a host system to investigate the phage–host metabolome interaction. This bacterium has an extreme metabolic versatility, allowing it to survive in diverse habitats ranging from coastal marine habitats over soil to even human tissue, making it an ideal model organism (Stover et al., 2000). Using flow injection analysis time-of-flight mass spectrometry (Fuhrer et al., 2011), we studied the metabolite dynamics in P. aeruginosa, during different stages of infection by diverse well-characterized lytic double-stranded DNA phages (Table 1, Ceyssens and Lavigne, 2010). The goal was to determine if phage infection results in a metabolic, ‘universal’ host response or if different phage species will yield distinguishable responses, a question that, to our knowledge, has never been addressed to date.

Table 1 Main characteristics of the phages used in this study

Materials and methods

P. aeruginosa strain and phage isolates

P. aeruginosa strain PAO1 (genotype 0002) was chosen as bacterial host system in view of its extensive genetic and phenotypic characterization. First isolated in 1955 in Melbourne, Australia from an infected wound (Holloway, 1955), it was the first fully sequenced P. aeruginosa strain (Stover et al., 2000). Six different lytic phages were selected to represent a diverse set of Pseudomonas bacteriophages (Table 1). Phage stocks were grown on their original P. aeruginosa propagation strain, that is, PAO1 (LUZ19, PEV2, 14-1, YuA and phiKZ) or Li010 (LUZ24) using standard soft agar overlay followed by PEG8000 precipitation, as previously described (Ceyssens et al., 2008), and stored in phage buffer (10 mM Tris-HCl, pH 7.5, 10 mM MgSO4, 150 mM NaCl) at 4 °C.

Construction of the virus population network

To illustrate the genome-based relationship between the selected phages, the predicted protein sequences representing the genomes of LUZ24, LUZ19, 14-1, PEV2, phiKZ and YuA, as well as their potentially relevant bacteriophages were clustered using the ACLAME database, version 0.4 ( with the database of ‘Viruses’ and E-value cutoff of 0.001 and all-to-all BLASTp searches with a cutoff E-value of 10−4 (Leplae et al., 2004). The degree of similarity between phages was calculated as the minus logarithmic score by multiplying hypergeometric similarity P-value by the total number of pairwise comparisons. Afterward, a protein-sharing network was generated with the Cytoscape software platform, version 3.1.1, using an edge-weighted spring embedded model (Shannon et al., 2003).

Medium and infection parameters characterization

All growth experiments were performed in minimal medium Pseudomonas (30 mM Na2HPO4, 14 mM KH2PO4, 20 mM (NH4)2SO4, 20 mM glucose, 1 mM MgSO4, 4 μM FeSO4). Infection curves were determined by following the OD600nm of a bacterial culture (OD600nm=0.3) every 5 min during infection until cell lyses was observed. Different multiplicity of infection (MOI) were used for all phages to ensure synchronized bacterial cell lysis and the optimal MOI was chosen (data not shown). Owing to differences in adsorption efficiency, the selected MOI varies between phages (Supplementary Figure S1). As the excess of phages is removed during the metabolome extraction, these differences will not impact the comparison between phages. During sampling, infection curves were made to assure infection of the entire culture (Supplementary Figure S1).

To check the efficiency of infection, cell counts of the infected culture were done before and 5 min after infection for all sampling and the obtained cell counts were compared. The efficiency of infection was >95% for phiKZ and 14-1, >90% for LUZ19 and YuA, >75% for PEV2 and >65% for LUZ24. Although there are differences in the efficiency of infection, they do not affect the comparison of metabolic alterations induced by the phages. As increasing the MOI did not affect the efficiency of infection, nor did it alter the infection curves, which show a synchronous infection (Supplementary Figure S1).

Metabolite extraction and analysis

Bacteria, grown in 75 ml of minimal medium Pseudomonas, were infected with phages at OD600=0.3 (1.25 × 108 CFU ml–1) and sampled for metabolite profiling by fast filtration (Link et al., 2013) at different time points during phage infection (Supplementary Figure S1). At each time point, the biomass quantity was measured by following the OD600 and the sampling volumes were adjusted to a volume corresponding to a biomass of a 1 ml culture at OD600=1.0 (approximately 4 × 108 CFU). The samples were vacuum filtered through a 0.45 μm pore size nitrocellulose filter (Millipore, Billerica, MA, USA) and washed with 2 ml ammoniumcarbonate solution (75 mM, pH 7.0). The filters containing the captured cells were incubated in 3 ml ethanol (60%) at 78 °C for 2 min, and subsequently snap frozen in liquid nitrogen and stored at –80 °C until further processing. Metabolite extracts were dried at 30 °C under vacuum and re-suspended in 100 μl water. The samples were profiled using negative mode flow injection-time-of-flight mass spectrometry and detected ions were annotated as previously reported (Fuhrer et al., 2011). For each phage infection, four biological replicates were sampled and two technical repeats were made of each independent biological sample.

Metabolite data analysis

A genome-scale metabolic model of P. aeruginosa (Oberhardt et al., 2008) and KEGG database were used to compile a metabolite reference list, according to which ions were assigned to metabolites applying a mass tolerance of 1 mDa and an intensity cutoff of 1500 counts. For each ion, only the metabolites with the highest annotation score were used, and for each metabolite, the best annotation was kept. Metabolite annotation and statistical analysis was performed using Matlab R2013b (Mathworks, Natick, MA, USA). A detailed description of these statistical analyses can be found in the Supplemental Materials and methods.

Results and Discussion

The selected Pseudomonas phages are highly diverse representatives within a broad virus population network

To investigate the occurrence of phage-specific effects on host metabolism, we aimed at selecting phages with possibly diverse infection phenotypes. To illustrate this, we constructed a virus population network, which includes the known (lytic) Pseudomonas phages, based on protein sequence similarities using the ACLAME database. The selected phages: LUZ19, LUZ24, PEV2, 14-1, phiKZ and YuA represent different branches within this network (Figure 1). Phages PEV2 and phiKZ are restricted to two components corresponding to the N4- and phiKZ-related groups, respectively; separated from the largest connected component. These individual components, attributable to their distinct shared core proteins, support the taxonomic descriptions as evolutionary distinct branches. In addition, phages LUZ24, 14-1 and LUZ19 fall into interconnected regions comprising the 7 LUZ24-like phages, 13 PB1-like phages and 3 phiKMV-like phages within the Autographivirinae clade, respectively. Each region contains one or more phage groups, implying their close evolutionary relationships. YuA shares proteins with multiple groups, with M6 as its closest relative. Collectively, as these phages belong to individual genera, the six studied phages provide an important basis for the understanding of the diversity of Pseudomonas phages in various viral lineages. Table 1 further illustrates that these phages differ strongly in genome size, GC-content and the duration of their infection cycle.

Figure 1
figure 1

Virus population network reveals distinct clusters within the Pseudomonas phage universe. The network consists of 498 phages belonging to the Myoviridae, Siphoviridae, Podoviridae, or uncharacterized and other phages, and 6953 relationships between them. The nodes in the network are linked by edges indicating the significant relationships between viruses, in terms of shared gene (protein) contents. The viruses with a stringency similarity score of 1, which is estimated using the hypergeometric formula, are shown. The network was visualized with an edge-weighted spring embedded layout using Cytoscape software platform (version 3.1.1), which places the viral genomes sharing more protein families closer in the display. Selected phages are shown in circle and named in bold. Orange=Myoviridae, teal=Siphoviridae, purple=Podoviridae, gray=uncharacterized.

Phage infection disrupts the metabolic steady state in exponentially growing cells

To investigate the dynamic response of the host metabolism to infection with different phages, we performed a time-course experiment, infecting P. aeruginosa with each of the six selected phages and sampling the metabolic content at different stages of infection. To ensure synchronized infection and sample uniformity during the experiment, the MOI was adjusted and killing curves were made for individual phages to confirm synchronized infection (Supplementary Figure S1, Materials and methods). Samples were analyzed with untargeted flow injection analysis time-of-flight mass spectrometry, which allowed high coverage measurements of the metabolome. A total of 6006 ions were detected, of which 918 could be assigned to P. aeruginosa metabolites. After stringent filtering of ion adducts, 375 unique ions corresponding to 518 metabolites, including mass isomers, were retained.

To compare the metabolite dynamics, metabolite fold changes were determined for each time point relative to time point zero of each infection. Without infection, metabolite levels remained stable, indeed only a small fraction of metabolites (3.2%) showed a (slight) increase after 45 min (log2(fold change)<0.5) (Figure 2a). Compared with the metabolic steady-state of the non-infected samples, phage infection was found to induce significant alterations (P-value 0.05 and abs(log2(fold change) >0.5) in up to 24.5% of the detected metabolites, as the sum of increased (19.7%) and decreased (4.8%) metabolites (PEV2, 25 min after infection, Figure 2a). This is consistent with the results observed for the lytic roseophage (Φ2047B) infection in Sulfitobacter sp. 2047, where approximately 25% of the 82 measured metabolites were altered after a single infection cycle (60–120 min; Ankrah et al., 2014), confirming that host physiology is significantly altered upon phage infection in different host systems.

Figure 2
figure 2

(a) Percentage of altered Pseudomonas metabolites during the course of infection of P. aeruginosa with six different phage clades (P-value 0.05 and log2(fold change) −0.5 or 0.5). Fold changes were determined by comparison of metabolite levels before infection against the mentioned time point during infection. (b) Percentage of metabolites changed in multiple phage infections, metabolites were considered significantly changed when P-value 0.05 and log2(fold change) −0.5 or 0.5.

A ‘universal’ metabolic host response to Pseudomonas phage infection does not occur

To identify potentially common metabolic responses to phage infection, changes in all annotated metabolites were compared among the different phages. In total, 56% of the metabolites were significantly altered by at least one phage (P-value <0.05 and abs(log2(fold change) >0.5). However, only 2.4% of the metabolites were changed upon infection with each of the six phages, indicating the absence of a general response to phage infection (Figure 2b). In contrast, 17.30% of the metabolites were found to be significantly increased/decreased by a single phage, whereas staying stable in all others (Figure 2b,Supplementary Table S1).

The small percentage of changing metabolites (2.4%, nine metabolites) shared between all phages are involved in nucleotide and nucleotide sugar metabolism (Supplementary Table S2). Nucleotide metabolism produces the required nucleotides for viral genome replication, which was shown to be crucial during T7 infection (Qimron et al., 2006). UDP-glucose and UDP-galactose are key intermediates for the production of oligo- and polysaccharides (Bosco et al., 2009; Ebrecht et al., 2015). For example, the nucleotide sugars (UDP-glucose, UDP-alpha-D-ManNAc3NAcA and UDP-alpha-D-GlcNAc3NAcA) are precursors for lipopolysaccharide biosynthesis (Westman et al., 2009).

Hierarchical clustering of the metabolites and phages based on the metabolite fold changes relative to the uninfected state confirms the absence of a ‘universal’ host response (Figure 3). For each phage, the time point responses form a closely related cluster branch, and the metabolite groups that attribute to the branch separation are shown on the top of the figure and will be discussed in later paragraphs.

Figure 3
figure 3

Hierarchical clustering based on correlations between mean metabolite fold change vectors reveals phage-specific changes in host metabolism. Color bar indicates the fold change of different metabolites (columns) for the different samples (rows). On the left is the hierarchical tree showing the relation between the different samples. On the right is the different time points and infecting phage are indicated for each row of metabolite fold changes. On the top is the corresponding pathways are shown for clusters of metabolites. AA, amino acids; nucl, nucleotide; pant, Panthotenate; PPP, pentose phosphate pathway.

A comparative pathway analysis between phages reveals a central role for pyrimidine metabolism and accumulation of nucleotide/amino sugars

To assess whether specific metabolic pathways are targeted during phage infection, metabolite set enrichment analysis was performed to identify pathways enriched for accumulated or depleted metabolites. All significantly affected pathways have been summarized in Supplementary Figure S2. Aside from a gradual change over time in pathways involved in nucleotide and amino-acid metabolism (for example, PEV2), only limited temporal variation was found for the different phages. To facilitate visual comparison between phages, the enrichment analyses over the different time points were combined (Figure 4).

Figure 4
figure 4

Metabolic pathway enrichment for different phages. The results for the enrichment analysis are shown for all six phages (columns) and for all respectably significantly decreased and increased pathways (rows). The color scale indicates the cutoff for the P-values. Note, pathways are only shown if at least one time point shows an enrichment P-value <0.05.

Most of the enriched pathways can be linked to specific requirements during phage replication. First, most negatively impacted pathways are related to amino-acid metabolism, with a decrease in the metabolism of 14/20, 11/20, 10/20, 4/20 and 3/20 amino acids (and precursors) during the infection of YuA, 14-1, PEV2, LUZ19 and LUZ24, respectively (Figure 4). This decrease can be explained by either a decreased production or an increased incorporation of these amino acids in proteins, not compensated by increased production. An increased consumption of amino acids during phage particle formation seems the most logical hypothesis. However, not all phages deplete amino acids to an equal extent, whereas the giant phage phiKZ does not even exhibit this decrease. This observation indicates different levels of dependency on the host cell resources, as discussed later.

Second, multiple pathways involved in nucleotide metabolism have been influenced. The pyrimidine pathway, which produces the required pyrimidine nucleotides for DNA replication, is enriched for increased metabolites for all analyzed phages, except for 14-1 (Figure 4). Using a Venn diagram, the overlap in changed metabolites between phages was visualized (Supplementary Figure S3). Most of the final pyrimidine nucleotides are increased in all infections, whereas the intermediates only increase for phiKZ and even decrease for PEV2, LUZ24 and YuA (Supplementary Figure S3). Interestingly, the purine metabolism was only enriched for increased metabolites during infection by YuA. This hints at a difference in the need/consumption of pyrimidine versus purine nucleotides during phage infection. Metabolite levels of glutathione metabolism are decreased by both LUZ19 and YuA (Supplementary Figure S2). Glutathione is essential for oxidative stress protection and increased activity of ribonucleotide reductases in nucleotide metabolism (Masip et al., 2006). These enzymes catalyze the formation of deoxyribonucleotides from ribonucleotides, an essential step for phage DNA replication.

Metabolites of panthotenate and coenzyme A (CoA) metabolism are decreased in all phages, except phiKZ (Figure 4). CoA has an important role in many host pathways that supply essential components and energy to the cell (Leonardi et al., 2005). The enrichment of the pantothenate and CoA pathway among decreased metabolites would slow down the host metabolism. This may be explained by the phage-induced host metabolism take-over or shutdown. However, although the metabolite levels of pantoate and aspartate levels are indeed decreased, some intermediate metabolites and CoA levels are even increased by certain phages, for example, PEV2, LUZ24 and YuA (Supplementary Figure S4). In these cases, the observed decrease in CoA metabolism could also be caused by increased CoA consumption.

The third pathway affected for all phages, except YuA, is the amino sugar and nucleotide sugar metabolism (Supplementary Figure S5). Nucleotide sugars are the building blocks for peptidoglycan and lipopolysaccharides, required for cell growth and division. Multiple examples exist of phage proteins that inhibit cell division or cell wall biosynthesis, such as lysis protein E of phiX174 (Bernhardt et al., 2001) or the kil proteins in phages Mu (Waggoner et al., 1989), P22 (Semerjian et al., 1989), lambda (Sergueev et al., 2001) and Sf6 (Casjens et al., 2004). Homologs of these proteins share only low levels of sequence similarity (Sau et al., 2008), and have not yet been identified in these Pseudomonas phages. However, the increased levels in the nucleotide sugars appears to be indicative of the presence of phage proteins inhibiting cell division or cell wall biosynthesis. This is further illustrated by the enrichment of peptidoglycan and lipopolysaccharide biosynthesis during PEV2, LUZ24 and phiKZ infection. In fact, Phikzlikevirus members encode a tubulin-like protein (Kraemer et al., 2012), which positions the phage DNA in the center of the cell and induces an arrest in host cell division during infection (Ceyssens et al., 2014). Other cell division inhibiting early proteins were also identified in LUZ7, an N4likevirus like PEV2 (Wagemans et al., 2014).

Most of the remaining pathways enriched for increased metabolites are involved in host carbohydrate metabolism, which produces the required building blocks for biomass production, however, the affected carbon sources vary between phages (Figure 4). Interestingly, none of the central carbon metabolism pathways (for example, TCA cycle, glycolysis) were enriched. To conclude, this pathway enrichment analysis reveals a primarily phage-specific impact on host physiology, a central need for pyrimidine nucleotides and a widespread accumulation of nucleotide sugars during phage infection.

Host stress metabolites reveal phage-specific impact on different stress responses

In response to different stresses, P. aeruginosa produces secondary messenger molecules, which enable cell-to-cell communication and coordinated responses (Jimenez et al., 2012). To assess if phage infection would trigger these host stress responses, the detected metabolites were analyzed. Although many of them were not detected (for example, AHLs, c-di-GMP and cAMP), both ppGpp and Pseudomonas quinolone signal (PQS) could be analyzed.

First, ppGpp controls the bacterial growth rate by inhibiting the production of ribosomal RNA (Potrykus and Cashel, 2008; Potrykus et al., 2011). It also fine tunes the metabolism by transcriptional regulation of enzymes involved in amino-acid metabolism and binding of enzymes involved in nucleotide metabolism (Gallant et al., 1971; Hochstadt-Ozer and Cashel, 1972; Potrykus and Cashel, 2008). High levels of this molecule are linked to the stringent response, which occurs on exposure to amino-acid starvation or heat shock (Traxler et al., 2008; Hauryliuk et al., 2015). The level of this molecule increases during LUZ19, 14-1 and YuA infection, whereas it remains stable for phiKZ and even decreases for LUZ24 and PEV2 ( This clear distinction between the phages does not appear to be correlated to other observed metabolic differences, such as the decreased amino-acid levels in multiple phages (YuA, 14-1, PEV2, LUZ19 and LUZ24). This seems to indicate an active interference in the stringent response during phage infection.

The second stress metabolite, the 'PQS', triggers autolysis and DNA fragmentation, whereas simultaneously lowering the metabolic activity in undamaged bacteria and increasing the protection against oxidative stress (D’Argenio et al., 2002). It has a central selective role for the survival of the fittest (Häussler and Becker, 2008). The PQS metabolite level increases during YuA, 14-1 and LUZ24 infection (, which hints at a host response aimed to protect the fittest (uninfected) cells by slowing down metabolism and killing the susceptible (infected) cells. Surprisingly, PQS levels are not increased upon phiKZ, LUZ19 and PEV2 infection. Either the activation of the PQS system is not a general host response or these phages have evolved unknown mechanisms to block PQS synthesis.

Observed correlations between the phage-specific metabolic responses and predicted AMGs

To investigate what is causing the phage-specific impact on host physiology, all phage genomes were screened for potential AMGs using the bioinformatics tool BLASTn (Table 2) (Altschul et al., 1990). The number of potential AMGs identified was limited in all phages (around 3% of all genes) and most were Class I AMGs predicted to be involved in nucleotide metabolism. Their presence likely explains the common impact on pyrimidine metabolism in most phages, corresponding to the central need for nucleotides in viral genome replication. In fact, the increased levels of metabolites in this pathway, as well as the amino sugar and nucleotide sugar metabolism, were also hypothesized from a metagenomic screen for enriched viral auxiliary genes from the Global Ocean Survey (Enav et al., 2014). The authors hypothesized that viruses carry these metabolic genes to expand the nucleotide pool in two ways, by recycling the nucleotides of the cellular genome and transcriptome (Powell et al., 1992; Ueno and Yonesaki, 2004; Lavigne et al., 2013) and/or by directing the host metabolism to provide substrates for de novo synthesis (Thompson et al., 2011).

Table 2 Predicted AMGs present in phage genomes as well as their class

Indeed, the giant myovirus phiKZ redirects the host metabolism to provide substrates for de novo synthesis (Figure 5). The genome of this giant virus contains seven AMGs, which can be linked to the pyrimidine pathway (Table 2). Directly after infection, the intermediates of de novo pyrimidine synthesis show a clear increase (for example, N-carbamoyl-l-aspartate, (S)-dihydroorotate), followed by a slow increase in the nucleotide monophosphates (Figure 5). Considering the coding potential of this giant virus (280.3 kb) and its host-independent transcriptional regulation scheme (Ceyssens et al., 2014), it is tempting to rationalize a similar independence of available host resources at the metabolome level. Indeed, phiKZ appears to have no influence on the cell’s metabolites during early infection, but approximately 10% of all annotated metabolites gradually accumulate during late infection (Figure 2a).

Figure 5
figure 5

Both the strategy of recycling and de novo synthesis of pyrimidine nucleotides are present in the set of phages. Graphs show the fold changes of metabolites in the pyrimidine pathway at different stages of infection (black-control, orange-YuA and blue-phiKZ). YuA recycles nucleotides (orange shade), characterized by a rapid increase of nucleotides. PhiKZ showed an increased de novo synthesis (blue shade), characterized by an increase in early building blocks. The y axis shows the fold change, x axis shows the time points during infection (in minutes), annotated metabolite name and ion mass are shown above the graph. The colored squares indicate the enzymatic reactions encoded by YuA (orange) and phiKZ (blue). The metabolites mentioned in the text are also highlighted.

As the other phages contain smaller genomes, they may lack the coding capacity to generate a functioning host metabolism and rely more on depletion of the host resources. This may explain the immediate increase in nucleotide monophosphates (for example, dTMP and dCMP) in the pyrimidine metabolism during the infection of siphovirus YuA. Most likely, this is the result of host genome degradation, because of unchanged levels of the above mentioned intermediates (Figure 5) and the presence of a predicted exonuclease (gp6) in the YuA genome (Table 2). In comparison with phiKZ, the smaller phage YuA (58.6 kb) uses a ‘leeching’ strategy, aimed at harvesting metabolites from the host. In fact, after infection an immediate negative effect on approximately 17% of all measured metabolites is observed, which results in a distinct metabolic effect for this phage (Figure 3) and a high number of negatively impacted pathways (Figure 4). However, the immediate degradation of the host genome may also explain the overall negative effect observed in YuA. This is a clear example of how the presence of a single AMG, in this case a nuclease, results in a distinct different metabolic impact.

A second example of the role AMGs might have, has been found when comparing the impact of LUZ19 and LUZ24. At first sight, these two phages show very similar percentages of metabolite increase and decrease during infection (Figure 2a), which may be explained by their relative close relationship within the population network (Figure 1). Both phages cause an increase in nucleotide levels after the first third of infection (Supplementary Figure S6). As they both encode an endonuclease, gp23 for LUZ19 and gp35 for LUZ24 (Table 2), the observed increase is likely caused by host genome degradation, which has already been experimentally shown for LUZ19 (but not LUZ24) at the start of viral genome replication (after the first third of infection) (Lavigne et al., 2013). However, the hierarchical clustering revealed more differences at the metabolite level between LUZ24 and LUZ19 (Figure 3). A comparison of the encoded AMG content revealed three AMGs that are present in LUZ24, but absent in LUZ19 (Table 2).

LUZ24 gp19 is a predicted l-glutamine-D-fructose-6-phosphate amidotransferase (also called glucosamine-6-phosphate synthase, glmS). This protein catalyzes the rate-limiting enzymatic step in de novo biosynthesis of UDP-GlcNAc, the activated form of N-acetylglucosamine (Milewski, 2002). This metabolite is required for the biosynthesis of peptidoglycan and lipopolysaccharide (Mirelman and Nuchamowitz, 1979; King et al., 2009). Late in LUZ24 infection, the level of UDP-GlcNAc (C00043) is indeed significantly increased (FC35 minutes=1), which is not observed for LUZ19. This indicates that LUZ24 gp19 is indeed a novel functional glucosamine-6-phosphate synthase. The presence of this AMG resulted in a lower enrichment of the amino sugar metabolism for increased metabolites in LUZ24, compared with LUZ19 (Supplementary Figure S2). Most likely, this is the result of a decreased accumulation of the precursors before this rate limiting enzymatic step. Although no stable lysogenic isolates could be obtained for this phage (Ceyssens et al., 2008), it is genetically homologous (71% sequence identity) to the temperate phage PaP3 (Tan et al., 2007). This AMG could be an artifact of its former temperate nature, where it would yield a fitness benefit to the host during lysogeny (Feiner et al., 2015). The two other AMGs, solely present in LUZ24, are involved in glutathione metabolism (Table 2). Metabolites of this pathway are only decreased during LUZ19 infection, whereas they remain stable during LUZ24 infection. This decrease is likely caused by oxidative stress because of phage progeny production, the AMGs in LUZ24 can compensate this drop and thus reduce oxidative stress (Figure 4).

Phage 14-1 encodes two predicted AMGs, one involved in pyrimidine metabolism and a homolog to the T4 polynucleotide kinase (PnkP) (Table 2). During infection, 14-1 exhibits both increased (7.2%) and decreased (7.7%) metabolites (Figure 2a). Despite the presence of this AMG, it is the only phage where the pyrimidine pathway is not enriched for increased metabolites (Figure 4). In T4, PnkP repairs host-induced cleavage of host tRNAs (for example, lysine) by ribotoxins to circumvent the host’s attempt to block T4 protein synthesis (Amitsur et al., 1987; Wang et al., 2015). Metabolites associated with the amino-acid metabolism are decreased early in the short 14-1 infection (<15 min), whereas towards the end of infection this negative effect is less pronounced (Supplementary Figure S2). These observations indicate the presence of the PnkP homolog may enable 14-1 to respond to a triggered host defense and complete its rapid infection cycle.

The last phage, PEV2 belongs to the same genus (N4likevirus) as roseophage Φ2047B, studied by Ankrah et al. (2014). As both phages share a general genome organization, similarity in the host response was expected. Indeed, after one infection cycle, an increase is observed for approximately 20% of the detected metabolites, whereas only a small fraction (4.8%) was significantly decreased (Figure 2a). However, early in infection, up to 10% of measured metabolites were decreased, something which was not observed for the roseophage (Figure 2a). This is due to a drop in amino-acid-related metabolites early in infection, which is restored over time. PEV2 encodes only two predicted AMGs, which fail to explain the observed effects on amino-acid metabolism (Supplementary Figure S2).

Although the identified and described AMGs already offer some explanations for the phage-specific metabolic effects, many other effects cannot be attributed to predicted AMGs at this time. Given the high number of encoded viral genes that lack any functional prediction in all of these phages, we can speculate on the significant importance of unknown, ‘non-enzymatic’ peptides in metabolism take-over. Such AMGs could technically be defined as Class II AMGs, although they lack any direct metabolic function. More and more of such peptides are being identified, for example, the interaction of an early protein of LUZ7 (gp30) with a protein involved in polyamine metabolism (PA4114) (Wagemans et al., 2014) or the early protein of phiKMV (gp9), which is hypothesized to impact the TCA cycle (Roucourt et al., 2009). These findings emphasize the need for more efforts to characterize the function of such small, unknown viral gene products.

An open access database enables straightforward interpretation and interactive analysis of the metabolomics data

Unfortunately, metabolomics data sets of phage–host systems are scarce and hardly accessible, despite the fact that the entire community could benefit from such large data sets. For this reason, an interactive database was set up providing open access to the data and a straightforward to use, visual interface to analyze and compare them ( The data interpretation is possible at different levels: comparative analysis between phages, phage-specific interpretation and pathway/metabolite-tailored visualization. These tools should enable researchers to look at specific metabolites/responses upon phage infection. Finally, this database is a first step toward an online platform capable of integrating proteomics, metabolomics and transcriptomics data sets, which, because of the technological advances in these fields, enables the study of phage infection at a systems level.


Implications for phage biology

In this work, we provide a detailed insight into the changing metabolic content of phage-infected P. aeruginosa cells, within a single infection cycle, and conclude that the phage-specific alterations of the host metabolism are more than a mere depletion of the host metabolites. A clear distinction can be made between ‘leeching’ phages, which take all existing resources for viral replication (for example, YuA), and phages that actively modulate the metabolism, by redirecting the cell’s metabolism to produce new resources, required for phage production (for example, phiKZ). The great diversity in genetic features of each individual phage is the driver for this metabolic diversity. We further show that specific metabolic pathways, targeted by the AMGs, are statistically enriched in the set of metabolites changing upon infection. Therefore, each specific set of AMGs will result in a unique metabolic phenotype after infection. From a more philosophical perspective, this altered metabolic state of the phage-infected cell is consistent with the definition of viral life (virocell concept) (Forterre, 2013).

Implications for global ecosystem and fluxes studies

The presence of viruses in microbial communities impacts nutrient fluxes. Recently, a model was presented which demonstrated the impact of virus infection on marine food webs and stressed the need to take the viral presence into account (Weitz et al., 2015). However, as the (current consensus is that the) virome contains an almost unlimited source of AMGs from which phages retain those beneficial for their needs (Sharon et al., 2011), not only the occurrence of phage infection but also the diversity of the phages present will impact the net metabolic effect of phage infections in complex bacterial populations. This is an additional element that should be considered when trying to understand such complex systems and the role phages have in them, such as in global nutrient fluxes.

Implications for biotechnology

Finally, this study provides a valuable resource to investigate the link between the metabolic effects of a phage and its encoded AMGs and the search for unknown ‘non-enzymatic’ AMGs. To catalyze the research, an interactive database ( was set up to allow fast visualization of the data depending on the research question. In future, this database could be expanded with data from single phage gene expression, transcriptome and protein interaction data to allow a thorough analysis of phage infection at the systems’ biology level. With the newly developed phenomics technique (Sanchez et al., 2015), we expect the discovery of phage-encoded AMGs to boom in the coming years. Such ‘metabolic modulators’ could provide key tools toward novel biotechnological applications in the advanced metabolic tailoring of bacteria.