A genome-wide atlas of antibiotic susceptibility targets and pathways to tolerance – Nature.com

A genome-wide view of antibiotic sensitivity

To obtain a genome-wide view of the genetic determinants that can modulate antibiotic stress in S. pneumoniae, Tn-Seq was employed in the presence of 20 antibiotics (ABXs), representing 9 different ABX groups and four classes including cell wall synthesis inhibitors (CWSIs), DNA synthesis inhibitors (DSIs), 30S and 50S protein synthesis inhibitors (PSIs) and an RNA synthesis inhibitor (RSI) (Fig.1a). Six independent transposon libraries were generated and grown for approximately 8 generations in the absence and presence of an antibiotic at a concentration that reduces growth by approximately 3050% (Supplementary Data1). Tn-mutant frequencies are determined through Illumina sequencing from the beginning and end of the experiment with high reproducibility between libraries (R2=0.700.90; Supplementary Fig.1) which is consistent with previous Tn-Seq experiments6,15,16,18,39,40,41,42. Combined with the population expansion during the experiment each mutants fitness (WMT) is calculated to represent their environment-specific relative growth rate, which means that a mutant with for instance a fitness of 0.5 (WMT=0.5) grows twice as slow as the wild type (WT)6,18,39,43,44. Each genes antibiotic-specific fitness is statistically compared to baseline fitness without ABXs, and is represented as W (WABX WnoABX) and categorized as: (1) Neutral, W=0, a mutants relative growth is similar in the absence and presence of an ABX; (2) Negative, W<0, a mutants fitness is significantly lower and thus grows relatively slower in the presence of an ABX; (3) Positive, W>0, a mutants fitness is significantly higher and thus grows relatively faster in the presence of an ABX. All antibiotics trigger both positive and negative fitness effects (Fig.1b, Supplementary Data2), which are distributed across 22 different gene categories (Fig.1c). Importantly, enrichment analysis shows there are multiple expected patterns, for instance genes involved in DNA repair are enriched in the presence of fluoroquinolones; cell-wall, peptidoglycan, and cell-division genes are enriched in -lactams and glycopeptides; membrane integrity genes in lipopeptides; and transcription and translation in PSIs (Fig.1d). Additionally, throughout the manuscript, we validate a total of 49 predicted genotype phenotype interactions, which indicates the Tn-Seq data is of high quality and in line with previously shown accuracy6,15,16,18,39,40,41,42 (Fig.1e, Supplementary Data8).

a Project setup and overview. Tn-Seq is performed with S. pneumoniae TIGR4, which is exposed to 20 antibiotics at a concentration that reduces growth by 3050%. Genome-wide fitness is determined for each condition, suggesting a multitude of options exists to increase as well as decrease antibiotic sensitivity. A co-fitness network is constructed by adding Tn-Seq data from 17 additional conditions, which through a SAFE analysis highlights functional clusters, and connects known and unknown processes. The genome-wide atlas and network are used to develop an antibiotic-antibody combination strategy, and to map out the wide-ranging options that can lead to decreased antibiotic sensitivity in vitro and in vivo and that are associated with a higher rate of stop codons in clinical samples. b There are a large number of genetic options that can modulate antibiotic sensitivity; with significant increased (W<0.15) and decreased sensitivity (W>0.15) split over all antibiotics almost equally likely. c Additionally, increased and decreased antibiotic sensitivity are distributed across a wide variety of functional categories. d Enrichment analysis shows that some pathways/processes such as glycolysis are relatively often involved in modulating responses to antibiotics, while other processes are more specific. e Validated growth experiments (n3 independent experiments) performed throughout the project highlight the Tn-Seq data is of high quality. SEM are shown for each data point. Source data are provided as a Source Data file.

Screens such as Tn-Seq are geared toward highlighting the genetic regions and/or genes that are important under a specific screening condition. With increasing conditions, genes acquire profiles that reflect their involvement/importance in those conditions, where genes with similar profiles indicate having similar and/or shared tasks. Such profiles can thereby help fill gaps in pathways, and/or identify genes and gene clusters with similar roles. By building a correlation matrix based on each genes ABX fitness-profile patterns emerge along a similarity range; from genes with highly similar to contrasting profiles. Moreover, to increase statistical power (i.e., more conditions increases the ability to identify more and stronger associations) the ABX dataset was supplemented with previously collected Tn-Seq data from 17 additional non-antibiotic conditions18 (Supplementary Data3). This results in a 15191519 gene matrix where positive correlations between genes come from shared phenotypes (i.e., similar profiles), while negative correlations come from opposing phenotypic responses under the same condition (i.e., contrasting profiles; Supplementary Data4). By repeatedly hiding random parts of the data the stability and strength of each correlation is calculated and represented in a stability score (Supplementary Data5). The correlation matrix and stability score are turned into a network, where each node is a gene, and each edge is a correlation coefficient above a threshold (>0.75), which combined with the stability score indicates the strength of the relationship between two genes. (Fig.2a; Supplementary Data6). Spatial Analysis of Functional Enrichment (SAFE)45,46 is used to define local neighborhoods within the network, i.e., areas enriched for a specific attribute (e.g., a pathway or functional category), which identifies multiple clusters that represent specific pathways and processes including purine metabolism, cell-wall metabolism, cell division and DNA repair (Fig.2b; Supplementary Data7). Moreover, the network contains gene clusters of high connectivity identifying highly related genes including those within the same operon such as the ami-operon, an oligopeptide transporter, the dlt-operon which decorates wall and lipoteichoic acids with d-alanine, and the pst-operon a phosphate transporter (Fig.2c, IIII). Besides identifying known relationships, the network also uncovers interaction clusters between genes with known and unknown interactions and function. Several such clusters are highlighted in Fig.2c (IVVIII), including genes involved in purine metabolism (further explored below), threonine metabolism, and in secretion of serine-rich repeat proteins (SRRPs), which are important for biofilm formation and virulence47. Importantly, the identification of biologically relevant relationships among (clusters of) genes indicates the data is rich in known and new information.

a A 15191519 gene correlation matrix based on Tn-Seq data from 37 conditions generates a network with genes as nodes, and edges as interactions with a stability score and thresholded correlation >0.75. The network contains one large connected component and multiple smaller components placed underneath; b A SAFE analysis identifies at least 11 clusters within the network that represent specific pathways and processes; c The network contains highly connected clusters of smaller groups of genes for instance those within the same operon such as cluster: I. the ami-operon (unknown transport); II. the dlt-operon; and III. the pst-operon (phosphate transport). Several additional clusters are highlighted containing annotated and unannotated genes, connected through known and unknown interactions including cluster: IV. containing genes involved in purine metabolism and a putative deoxyribose transporter (boxed 1.); V. containing genes involved in threonine metabolism and several genes located as neighbors to SP_2066/thrC with unclear functions (boxed 2), including a regulator (SP_2062) and a transporter (SP_2065); VI. containing genes involved in secretion of serine-rich repeat proteins. Source data are provided as a Source Data file.

Two hundred and twenty-four genes with a known annotation are present in the data that have at least one significant phenotype in response to an antibiotic, which can be split over 21 functional groups according to a pathway or process they belong to (Fig.3a). Each group is characterized by having multiple instances of decreased fitness, indicating genes that upon disruption increase sensitivity to one or more antibiotics (negative phenotype). Additionally, each group, except for cell division, also has multiple instances that increase fitness, which is suggestive of genes that upon disruption decrease antibiotic sensitivity (Fig.3a; positive phenotype). Moreover, each antibiotic group triggers both negative and positive effects (Fig.3b). Where possible, the 21 functional groups are organized according to a pathway or process they belong to and each gene is combined with its antibiotic susceptibility profile. This results in an antibiotic susceptibility atlas, which shows on a fine-grained scale, how inhibiting a pathway or process can seemingly simultaneously lead to increased and decreased drug susceptibility in an antibiotic-specific manner (Fig.3c and Supplementary Figs.2 and 3). For instance, in the glycolysis group, knocking out any of the three genes involved in forming the phosphotransferase (PTS)-system (SP_0282-SP_0284) that imports glucose to generate glucose-6-phosphate (G-6P), has a negative effect on fitness in the presence of 30S and 50S PSIs as well as Synercid (a synergistic combination of two PSIs), while it increases fitness in the presence of all CWSIs (-lactams, glycopeptides, and daptomycin) and fluoroquinolones. Also, knocking out SP_0668 (gki, glucokinase), an enzyme that converts -D-Glucose into G-6P, has a positive effect on fitness in all CWSIs and a negative effect in 30S PSIs. In contrast, inhibiting SP_1498 (pgm, phosphoglucomutase), the major interconversion enzyme of G-6P and G-1P, has a negative effect on fitness with all antibiotics (Fig.3c). Additional detailed examples are highlighted in Fig.3c, for instance for pyruvate metabolism, where inhibiting lactate, or acetaldehyde production increases sensitivity to -lactams and glycopeptides and decreases sensitivity to 30S PSIs, inhibiting formate production decreases sensitivity to co-trimoxazole and 30S PSIs, and inhibiting acetyl-phosphate production decreases sensitivity to -lactams, glycopeptides, and co-trimoxazole. Within aspartate metabolism a range of changes can be triggered from increased sensitivity to -lactams, and glycopeptides, to decreased sensitivity to most other antibiotics. Moreover, the four genes involved in the production of threonine from L-aspartate trigger decreased sensitivity to fluoroquinolones and 30S and 50S PSIs. In the shikimate pathway inhibiting the production of chorismate from phosphoenolpyruvate (PEP) and erythrose-5-phosphate leads to increased sensitivity to -lactams, co-trimoxazole, and Synercid. Cell division is the only process that upon interference, only generates increased sensitivity, specifically to CWSIs and co-trimoxazole. Interfering with peptidoglycan synthesis also mostly leads to increased sensitivity to CWSIs, as well as to 30S PSIs, while changes to genes that are involved in anchoring proteins to the cell wall (SP_1218 [srtA], SP_1833) can decrease sensitivity to CWSIs. Lastly, interfering with protein turnover, for instance through the protease complex ClpCP (SP_2194, SP_0746) and the regulator CtsR (SP_2195), which are generally assumed to be fundamental for responding to stress48,49, leads to decreased CWSI sensitivity and increased sensitivity to 30S and 50S PSIs (Fig.3c and Supplementary Fig.2). Moreover, FtsH (SP_0013), important for clean-up of misfolded proteins from the cell wall, increases sensitivity to 30S PSIs and Synercid, indicating how important protein turnover is especially for surviving 30S PSIs, which can trigger the production of faulty proteins. Most importantly, these data show that, as expected, hundreds of options exist where disruption of a pathway or process leads to increased sensitivity to specific antibiotics. Remarkably, there seem to be almost as many options that can lead to decreased antibiotic sensitivity.

a The number of phenotypes scored for each pathway/process. Genes with at least one significant phenotype are split over 21 groups according to a pathway or process they belong to, which highlights how modulation of most pathways can lead to increased (negative phenotype) and decreased (positive phenotype) antibiotic sensitivity. b The number of phenotypes scored for each antibiotic group. While sensitivity to each antibiotic (group) can be increased by knocking out genes in the genome (negative phenotype), sensitivity can be decreased (positive phenotype) almost as often for most ABXs, except for Synercid, and to a lesser extent rifampicin, where most effects are negative. c Detailed view of 7 out of 21 groups/processes highlighting how modulation of specific targets within each process leads to changes in antibiotic sensitivity. Each group is indicated with a number that is the same as in a. Where possible, genes are ordered according to their place in a process/pathway, and gene numbers (SP_) are combined with gene names and annotation. Each indicated gene is combined with an antibiotic sensitivity bar indicating whether disruption leads to increased (red/negative fitness) or decreased (green/positive fitness) sensitivity to a specific or group of antibiotics. When phenotypic responses are the same, multiple genes are indicated with a single bar (e.g. SP0282/SP0283/SP0284 in glycolysis, or SP0413/SP1013/SP1361/SP1360 in Aspartate metabolism). Gene numbers in blue have no effect on growth in the absence of antibiotics when knocked out, while gene numbers in purple have a significant growth defect in the absence of ABXs (see for detailed fitness in the absence and presence of antibiotics Supplementary Data2). Essential genes are not indicated and genes with an asterisk have a partial or tentative annotation that has not been resolved. All 21 groups are listed in Supplementary Figs2 and 3. Source data are provided as a Source Data file.

By identifying targets that (re)sensitize bacteria against existing antibiotics, genome-wide antibiotic susceptibility data have the potential to guide the development of new antimicrobial strategies. One such strategy could be a combined therapeutic antibody-antibiotic approach; the antibody would target a gene product that is important for sensitivity to one or more antibiotics and ideally the product would be easily accessible for the antibody at the bacterial cell surface. To find suitable candidate targets, Tn-Seq data were filtered for gene products that, based on a known function or localization prediction, are likely to be present in the cell wall or membrane, and that when disrupted, increase sensitivity to one or more antibiotics. Moreover, it would likely be ideal if the gene is also important for survival in vivo. A strong candidate is SP_1505, which in the interaction network is most tightly linked to cell wall metabolism and cell-division genes (Fig.4a). After we previously hypothesized that it may play a role in cell wall integrity14, it was recently named cozEb, with a likely role in organizing peptidoglycan synthesis during cell division50, which fits its interaction profile (Fig.4a). Importantly, the antibiotic Tn-Seq data suggest that disruption creates increased sensitivity to vancomycin and rifampicin, while the product is critical in the presence of daptomycin, which was confirmed through individual growth curves (Fig.4b). The protein has eight predicted membrane-spanning domains (Fig.4c), and in vivo Tn-Seq predicts it is important for survival in both the nasopharynx and lungs (Fig.4a, Supplementary Data2). The gene was cloned into an expression plasmid generating an ~30kD product (Fig.4c), which was used to raise rabbit anti-CozEb antibodies, which were confirmed to be specific for the cozEb gene product (Fig.4c). Potential antibody in vitro activity was determined through a bacterial survival assay in the absence and presence of antibodies and either vancomycin or daptomycin. Incubating bacteria with antibodies or daptomycin has no significant effect on bacterial survival, while vancomycin alone at the concentration used slightly reduces the number of surviving bacteria. Moreover, combining the antibody with either vancomycin or daptomycin further reduces the number of surviving bacteria in vitro compared to any agent individually (Fig.4d). To assess whether the antibody-antibiotic approach works in vivo, mice were intranasally challenged with a bacterial inoculum either containing WT or cozEb. Two additional sets of mice were challenged with WT and 8h post infection they were either treated with daptomycin and control IgG antibody or with daptomycin and CozEb-specific antibody. Mice were sacrificed 24h post infection, and bacteria in the lungs were enumerated. As predicted by the in vivo Tn-Seq data the cozEb knockout has a significantly lower fitness in the lungs highlighted by an up to 2.5-log lower bacterial load compared to WT. Importantly, while the WT survives equally well in the presence of the low daptomycin concentration and the control IgG antibody, in the presence of daptomycin and the CozEb-targeting antibody, its survival in the lungs is significantly reduced and resembles that of the cozEb knockout (Fig.4e). This shows that by combining antibiotic and in vivo Tn-Seq with gene annotation information, a gene product can be selected that is central and critical to cell-wall synthesis and cell-division processes. Importantly, due to its presence in the membrane, it is directly targetable with an antibody, thereby sensitizing the bacterium to an antibiotic concentration it is normally not sensitive to.

a cozEb/SP_1505 is tightly clustered with cell division and cell wall metabolism genes, it is predicted to increase sensitivity to glycopeptides and the lipopeptide daptomycin, and has a decreased fitness in the mouse lung and nasopharynx. b Reduced relative growth of cozEb validates its increased sensitivity to daptomycin and vancomycin. c CozEb has 8 transmembrane domains, which generates a ~30Kd product (BSA is shown as a control). The cloned protein was used to raise antibodies, which proofed to be specific for a product in the WT membrane, but does not bind anything in cozEb, indicating the antibodies are specific for the membrane protein CozEb. d Incubation of WT for 2h with vancomycin (Vanco) or daptomycin (Dapto) and in the presence of CozEb antibody, slightly but significantly decreases bacterial survival. Mean values SEM are shown from n3 independent experiments. e An in vivo lung infection with WT or cozEb confirms the mutant is less fit in vivo. Challenging the WT with daptomycin and IgG does not affect bacterial survival. In contrast, challenging with daptomycin and CozEb-specific antibodies, significantly reduces the recovered CFUs 24h post infection. Mean valuesSEM are shown from n10 mice/experiment. Significance is measured through a one-way ANOVA with Dunnett correction for multiple testing: *p=0.03, **p=0.001, ***p<0.001. Source data are provided as a Source Data file.

The example above illustrates how negative fitness indicates increased antibiotic sensitivity reflected by reduced relative growth, which can guide the development of (re)sensitizing approaches. In contrast, the occurrences of increased fitness in the dataset indicate that a large number of options exist that could lead to reduced antibiotic sensitivity (Fig.3). With increased fitness to 3 out of 4 antibiotic classes, the ami-operon is among genes with the greatest number of positive fitness effects. The operon forms a tight cluster in the interaction network (Figs.3 and 5a) and it is annotated as an oligopeptide transporter with no clear function. Two separate knockouts for SP_1888 (amiE) and SP_1890 (amiC) confirm that increased fitness results in decreased drug sensitivity in the form of increased relative growth in the presence of ciprofloxacin, vancomycin and gentamicin, and increased sensitivity (i.e., decreased relative growth) to Synercid (Fig.5b). There is limited evidence that the ami-transporter may have (some) affinity for at least two different peptides (P1 and P2)51,52,53. These have been theorized to possibly function as signaling molecules and under certain circumstances may be generated by the bacterium itself51,52,53. Both peptides were synthesized and while neither peptide affects growth of the WT or knockout mutants in the absence of antibiotics (Supplementary Fig.4), the WT grows slightly better in the presence of gentamicin and peptide P2, but not P1 (Fig.5b). This shows that some peptides may, at least partially, inhibit or occupy the ami-transporter, and thereby trigger decreased antibiotic sensitivity, in a similar manner as a knockout does. Besides peptides, the ami-transporter may be (non-selectively) transporting antibiotics into the cell, which could explain its effect on antibiotic sensitivity. To explore this, bacteria were exposed to ciprofloxacin or kanamycin and the internalized antibiotic concentration was determined through mass spectrometry for WT and both ami knockout mutants. In both mutants the amount of internalized ciprofloxacin was significantly lower (~1.7 in amiE, and ~2.3 in amiC), while the kanamycin concentration was found to be significantly lower in amiC (~2; Fig.5c). This shows that a functional ami-transporter increases the concentration of fluoroquinolones and 30S PSIs, suggestively by transporting them into the cell, and thereby, due to a higher internal concentration, enhancing the antibiotics inhibitory effects on growth. There are multiple examples that transporters can contribute to tolerance54,55, which we recently showed is also the case for the ade transporter in Acinetobacter baumannii, which contributes to fluoroquinolone tolerance7. However, those examples are mostly based on efflux pumps that actively decrease the antibiotic concentration in the cell through upregulation of such pumps. In contrast, with respect to the ami-operon it would be the reverse, i.e., inhibition instead of upregulation would lead to tolerance. To explore this possible effect on tolerance, the WT and amiE were exposed to either 10xMIC of gentamicin or vancomycin over a period of 24h. Approximately 1% of the WT population survives 4h exposure to gentamicin, while none of the population survives exposure past 8h. The amiE population displays a slower decline in survival with 1% of the population surviving the first 8h (tolerant cells)25. At ~10h the decline ceases and the remaining population (~0.01%) survives at least up to 24h, which is representative of a persister fraction25. In contrast, the WT and amiE mutant populations decline at similar rates when exposed to vancomycin, showing that inhibition of the ami-transporter can lead to tolerance and persistence in an antibiotic-specific manner while MICs of gentamycin and vancomycin for WT and amiE are similar (Supplementary Data1). Importantly, these data show that increased fitness indeed leads to decreased ABX sensitivity, which can translate into at least two phenotypes: increased relative growth and increased survival (i.e., tolerance).

a The ami-operon forms a tight cluster, and upon knockout is predicted to decrease sensitivity to most antibiotics, and increase sensitivity to Synercid. b Growth curves of individual knockout mutants of amiE and amiC validate changes in antibiotic sensitivity; i.e., they show that positive fitness translates into decreased ABX sensitivity and increased relative growth, while negative fitness translates into increased ABX sensitivity and decreased relative growth. Additionally, growth curves suggest the transporter phenotypically responds to peptide P2. Mean valuesSEM are shown from n3 independent experiments. c Intracellular antibiotic accumulation analysis shows that the WT strain with an intact transporter reaches a higher intracellular antibiotic concentration, suggesting the transporter is involved in importing antibiotics, explaining why a knockout or occupation with a peptide such as P2, can lead to decreased antibiotic sensitivity. Mean valuesSEM are shown from n3 independent experiments. d Besides that modulation of the transporter leads to positive fitness, which translates into decreased ABX sensitivity and increased relative growth in the presence of gentamicin or vancomycin, it also leads to increased survival (i.e., tolerance) to gentamicin, but not vancomycin. Mean valuesSEM are shown from n=4 independent experiments. Significance is measured through a one-way ANOVA with Dunnett correction for multiple testing: *p=0.05, **p=0.01, ***p=0.001. Source data are provided as a Source Data file.

Among the 21 functional groups, purine metabolism has some of the largest number of positive fitness effects, mostly with -lactams and glycopeptides (Figs.3a and 6a). Moreover, two regulators (SP_1821/1979) associated with this pathway decrease sensitivity to -lactams and/or glycopeptides and two neighboring genes with unknown function have either the same (SP_0830), or the opposite effect (SP_1446) on antibiotic sensitivity as their defined neighbor, suggesting they may be involved in the same process as their neighbor (Fig.6a). Furthermore, the global interaction network positively links an ABC transporter (SP_0845-0848, Figs.2c, 6a) with multiple genes in this pathway due to their similar profiles. This operon is annotated as a putative deoxyribose transporter, and to verify whether an interaction exists with purine metabolism, single and double knockouts were created between SP_0846 (the transporters ATP binding protein) and SP_0829/deoB. Their profiles suggest they do not affect growth in the absence of ABXs and have increased sensitivity to Synercid, which was confirmed with individual growth curves (Fig.6b). However, when both knockouts are in the same background, their increased sensitivity to Synercid is masked. Thus, as indicated by the network, these results show that the ABC transporter indeed has a genetic interaction with purine metabolism/salvage, but plays an unknown role. Importantly, this confirms that the global network includes valuable interactions that can be explored to uncover functional relationships.

a Key steps in purine metabolism with the same color coding as in Fig.3. SP_1097 is listed as well, for which we found no change in ABX sensitivity, which is denoted with np for no phenotype. The putative deoxyribose transporter (SP_0845-0848), a high-connectivity cluster in Fig.2, is also shown. b Single knockouts for deoB/SP_0829 and SP_0846, as well as a double knockout show that mutants and WT grow equally well in the absence of antibiotics. In the presence of Synercid, as predicted and indicated by their ABX sensitivity bar, the single knockouts display a higher sensitivity to the drug then the WT. The double mutant suppresses the increased Synercid sensitivity phenotype of the single mutants, indicating that the positive interaction that is found in the co-fitness network leads to a positive genetic interaction between these genes. c Single and double knockouts of SP_1097 and SP_1645/relA grow just as well as WT in the absence of antibiotics. As predicted SP_1097 is equally sensitive to cefepime as the WT, while relA has decreased sensitivity as indicated by its ABX sensitivity bar in a. Additionally, the double knockout has decreased sensitivity to cefepime, indicating the dominant phenotype of relA. d The phenotype of SP_0831 was validated in growth as well, showing no change in growth in the absence of ABX, and decreased sensitivity (i.e., increased relative growth) in the presence of cefepime (FEP). e The alarmone (p)ppGpp is below the limit of detection (b.l.d.) in the absence of stress, upon induction with mupirocin it is synthesized in equal amounts in WT, SP_0831 and SP_1097, while it cannot be synthesized when relA is absent. f Synthesis of di- and trinucleotides is significantly affected in the different mutants upon mupirocin exposure. Mean valuesSEM are shown from n3 independent experiments. Significance is measured through a paired t-test with an FDR adjusted p value for multiple comparisons: *p<0.05, **p<0.01, ***p<0.001, ns not significant. Source data are provided as a Source Data file.

Furthermore, within purine metabolism the alarmone (p)ppGpp is synthesized from GTP and/or GDP. Like other bacterial species, S. pneumoniae likely responds to (some) ABXs via induction of the stringent response pathway56, in which relA (SP_1645) is the key player with both synthetase and hydrolase activity57. Additionally, SP_1097 is annotated as a GTP diphosphokinase and may be involved in the synthesis of pppGpp from GTP (Fig.6a). Our data suggests, and we confirmed for the -lactam cefepime (Fig.6c), that when synthesis of the alarmone is inhibited by deletion of relA, similar to many other interactions in purine metabolism, this leads to reduced -lactam and glycopeptide sensitivity manifested by increased relative growth (Fig.6c). Moreover, while SP_1097, as predicted, does not change ABX sensitivity (Supplementary Data2, Fig.6), a double knockout of relA-SP_1097 seems to further decrease sensitivity to cefepime by further increasing relative growth (Figs.6c and7a). Additionally, besides a change in growth, the single relA and double knockout (relA-SP_1097), also increases tolerance to cefepime by ~1000-fold at 24h (Fig.7b), without changing the MIC (Supplementary Data1). To understand how relA and SP_1097 affect purine metabolism, we used LC/MS to measure (p)ppGpp, ADP, ATP, GDP, and GTP. Additionally, we included SP_0831 a purine nucleoside phosphorylase involved in nucleotide salvage, which has the same ABX profile as relA (Fig.6a, d), but should not directly affect (p)ppGpp synthesis. While (p)ppGpp is below the limit of detection during normal growth in any of the strains, as expected relA and the double mutant relA-SP_1097 are unable to synthesize the alarmone when exposed to mupirocin, a strong activator of the stringent response (Fig.6e, Supplementary Data9). In contrast, WT, SP_0831, and SP_1097 synthesize (p)ppGpp upon mupirocin exposure to a similar extent (Fig.6e). Concerning the di- and trinucleotides in the pathway, upon mupirocin exposure GTP and GDP are significantly reduced in WT, SP_0831, and SP_1097, likely because they are used for (p)ppGpp synthesis (Fig.6f, Supplementary Data9). In contrast, while ATP and ADP again remain constant for the relA mutants, ATP and ADP synthesis are significantly increased upon mupirocin exposure, especially for WT and SP_1097. This suggests that during activation of the stringent response, synthesis from IMP is directed toward AMP, and not necessarily GMP, at least not enough to replenish GTP and GDP. Additionally, upon mupirocin exposure, ATP only minimally increases for SP_0831, while it increases over twofold for WT and SP_1097 (Fig.6f). It has been shown for bacteria including Escherichia coli and Staphylococcus aureus that a decreased ATP concentration can decrease sensitivity to ABXs such as ciprofloxacin58. Additionally, in S. aureus (p)ppGpp overexpression has been associated with decreased sensitivity to linezolid59. Our data suggest that (p)ppGpp and ATP synthesis may be intrinsically linked, i.e., at least in S. pneumoniae the inability to produce the alarmone also results in lowered ATP synthesis, which is associated with a lowered ABX sensitivity to -lactams and glycopeptides. However, SP_0831 shows that even if (p)ppGpp can be synthesized, modulation of purine metabolism, for instance through the salvage pathway, can result in decreased ATP synthesis, and can lead to lowered ABX sensitivity (i.e., increased relative growth). Importantly, in many bacterial species, alarmone production is generally assumed to be triggered in response to different types of stress and has been shown to affect a large variety of processes including nucleotide synthesis, lipid metabolism, and translation. (p)ppGpp is thereby a ubiquitous stress-signaling molecule that enables bacteria to generate a response that is geared toward overcoming the encountered stress. However, contradictory results between species indicate a possible non-uniformity across bacteria, leaving much to be learned about how the alarmone and the processes it can control fit into the entire organismal (response) network56. Our data suggest that the inability (i.e., due to mutations) to generate the alarmone in S. pneumoniae in response to -lactams and glycopeptides is linked to reduced ATP, which under specific circumstances may be an optimal response, as it results in decreased ABX sensitivity translating into increased relative growth and tolerance, and thereby a higher probability to survive the insult (Figs.6c and 7a, b).

a Relative growth rates (i.e., fitness) of 16 knockout mutants involved in 7 processes measured in the presence of 7 antibiotics, validate that decreased ABX sensitivity (i.e., increased relative growth) can be achieved by modulating a wide variety of processes. Mean valuesSEM are shown from n3 independent experiments. b Significantly increased survival during exposure to 510xMIC of an ABX over a 24h period is observed for 9 out of 12 knockouts. Significance is measured with an ANOVA with Dunnett correction for multiple comparisons: **p<0.01, ***p<0.001. Mean valuesSEM are shown from n=4 independent experiments. c Tn-Seq data with a positive fitness in the presence of at least one antibiotic (y-axis) is plotted against in vivo Tn-Seq data (x-axis). Note that only in vivo data is shown that is predicted to have no more than a small fitness defect, no fitness defect or an increased predicted in vivo fitness, either during nasopharynx colonization or lung infection. Circled and indicated with arrows are SP_0829 in red and SP_1396 in black. d In vitro growth curves validate decreased sensitivity (i.e., increased relative growth) to cefepime (SP_0829) and meropenem (SP_1396). Mean valuesSEM are shown from n=3 independent experiments. e Mice were challenged with WT and MT in a 1:1 ratio of which half received ABX 16h post infection (p.i.), and all were sacrificed 24h p.i. Displayed are the MTs competitive index (CI) in the nasopharynx and lung, and in the presence and absence of cefepime (SP_0829) or meropenem (SP_1396). In all instances, the addition of ABX significantly increases the CIof the mutant. Significance is measured with a MannWhitney test **p<0.01, ***p<0.001. Mean valuesSEM are shown from n7 mice/experiment. Source data are provided as a Source Data file.

For instance, in the glycolysisTo further confirm that antibiotic sensitivity can be decreased by inhibiting a variety of processes, knockouts (KOs) were generated for fourteen mutants from 8 different processes. Moreover, an additional goal was to determine what increased fitness (i.e., decreased ABX sensitivity) would look like phenotypically, and thus whether it would translate into increased relative growth and/or tolerance. Of the 14 mutants with a Tn-Seq predicted increased fitness, 13 display an increased ability to grow in the presence of an ABX compared to the WT. Moreover, eight mutants, which inhibit several different processes including different metabolic pathways, transport, and transcription and translation, displayed tolerance, while retaining a similar MIC, and thereby have an increased ability to survive high-level exposure to an ABX (510xMIC) for at least 24h (Fig.7a, b, Supplementary Data1,8). Note that we validated 49 single KO genotype phenotype associations in this study, with an equal distribution across the entire spectrum of ABX sensitivity (Fig.1e, Supplementary Data8). This highlights that our approach uncovered a detailed genome-wide ABX sensitivity atlas composed of a multitude of genes, pathways and processes that when modulated can increase and/or decrease ABX sensitivity. The validation experiments highlight that the resulting fitness accurately predicts the relative growth rate of a mutant, which we have previously shown for hundreds of other negative fitness phenotypes6,14,15,16,18,39,40,41,42,44,60,61,62. Moreover, it turns out that in the majority of cases, increased fitness not only results in increased relative growth in the presence of an antibiotic, but also tolerance. Thereby, the part of the atlas that depicts decreased ABX sensitivity (i.e., increased fitness) includes a genome-wide tolerome, composed of a wide variety of pathways and processes that when modulated trigger tolerance in vitro in an ABX dependent manner.

Obviously, the selection regime in vivo is far more complex and stricter than in a test tube, which raises the question whether many of the options that decrease ABX sensitivity in vitro, including those that increase tolerance, would be available in vivo as well. To explore this, all the Tn-Seq data with a positive fitness in the presence of at least one antibiotic was combined with in vivo Tn-Seq data and filtered for those genes with no or only a small fitness defect predicted in vivo during nasopharynx colonization or lung infection (Fig.7c, Supplementary Data2). Two genes were selected that we had confirmed for decreased ABX sensitivity in vitro: (1) SP_0829/deoB synthesizes Ribose-1P and is involved in purine metabolism (Fig.6a). deoB has no effect on in vitro growth in the absence of ABX (Fig.7a, d), as predicted it grows better in the presence of cefepime (Fig.7a, d), but it does not affect survival/tolerance (Fig.7b); (2) SP_1396/pstA is the ATP binding protein of a phosphate ABC transporter (Supplementary Fig.3). pstA has no effect on in vitro growth (Fig.7a, d), it has a higher relative growth rate in the presence of meropenem (Fig.7a, d), and it also increases survival/tolerance (Fig.7b). Both mutants were mixed with WT in a 1:1 ratio and used in an in vivo mouse infection competition model as we have done previously18. Of the infected mice, half were administered antibiotics at 16h post infection, and were sacrificed 6h later to determine the strains competitive index (CI) (Fig.7e). Importantly, while both mutants may have a slight disadvantage compared to the WT when colonizing the lung or nasopharynx, their CI increases significantly in the presence of ABXs, leading to increased survival compared to the WT (Fig.7e, Supplementary Data10). Combining antibiotic- with in vivo Tn-Seq highlights the ability to predict the existence of a wide array of possible alterations of specific genes, pathways and processes that can have a beneficial effect in vivo in the presence of antibiotics. Such changes could thereby contribute to escape from antibiotic pressure and even create a path toward the emergence of antibiotic resistance.

There is likely significant overlap in the selective pressures a bacterial pathogen would experience in a mouse infection model compared to the human host. This raises the possibility that those gene disruptions that are predicted by Tn-Seq to lead to decreased antibiotic sensitivity and that simultaneously have no more than a minimal defect in vivo, could also have an advantage in the human host in the presence of ABXs and thereby contribute to ABX escape and/or the emergence of resistance. A premature stop codon most closely reflects the effect a transposon insertion has on a gene; i.e., it disables a gene. We thereby hypothesized that stop codons in certain gene sets predicted by Tn-Seq could be enriched for in antibiotic-resistant clinical isolates. To test this hypothesis 4 gene sets were compiled consisting of those that upon disruption: (1) decrease antibiotic sensitivity in at least 1 antibiotic and have no strong defect in vivo; (2) decrease antibiotic sensitivity in at least 1 antibiotic and have a defect in vivo; (3) have little to no effect on antibiotic sensitivity and in vivo; (4) have no effect or increase antibiotic sensitivity and have a defect in vivo (Fig.8a, b; Supplementary Fig.5). Thousands of strains were selected from the PATRIC63,64 database that could be split into a group of co-trimoxazole (SXT) resistant and a group of -lactam resistant strains, and each group was matched with an equal number of sensitive strains from the database. In all strains in the SXT and -lactam groups, irrespective of resistant or sensitive status, the number of stop codons in gene sets 1 and 3 are highest, which reflects the Tn-Seq predicted in vivo effects, i.e., while gene sets 1 and 3 contain mostly genes with potentially neutral effects, gene sets 2 and 4 contain many genes that are suggested to have a defect in vivo when disabled (e.g. with a stop codon) (Fig.8c). Moreover, SXT resistant isolates in gene set 1 more often contain a stop codon compared to sensitive strains, and in -lactam resistant isolates this is true for gene sets 13 (Fig.8d). While these are not ideal comparisons, for instance the entire ABX profile is not clear for many strains, different changes than premature stops could have ABX/in vivo modulating effects, strains could have experienced different ABX and/or in vivo selective pressures, and genetic changes can be strain-background dependent, it shows that genetic changes that can affect ABX and/or in vivo sensitivity, which are predictable with Tn-Seq, readily occur in clinical samples. This in turn underscores that ongoing infections may consist of variants that enable different paths to adjusting to, or overcoming a challenging host/ABX environment.

a Based on in vivo and ABX Tn-Seq data, four gene sets consisting of 34 genes each were compiled with specific fitness profiles in the presence of antibiotics and in vivo. Shown are the in vivo effects for nasopharynx, while lung data are depicted in Supplementary Fig.5. W represents the fitness difference of a gene in a specific condition (e.g., an antibiotic, in vivo) minus its fitness in vitro in rich medium. Dashed lines indicate significance cut-offs, grayed-out dots indicate genes with no significant change in fitness in the presence of antibiotics, colors represent antibiotics and are the same as in Fig.1. b Detailed distributions for each gene set highlight whether effects in the presence of antibiotics, in the nasopharynx and lungs increase (+), do not affect (0) or decrease () relative fitness. Gene set rationales are described in the text. c The total number of stop codons in each gene set for 2296 co-trimoxazole and 1166 -lactam resistant and sensitive strains. d The number of sensitive and resistant strains with at least one stop codon in a gene in each gene set. Significance is measured through a Fishers exact test: **p<0.01, ***p<0.001, ****p<0.0001. Source data are provided as a Source Data file.

Here is the original post:
A genome-wide atlas of antibiotic susceptibility targets and pathways to tolerance - Nature.com