A bacterial pan-genome makes gene essentiality strain-dependent and evolvable

Abstract

Many bacterial species are represented by a pan-genome, whose genetic repertoire far outstrips that of any single bacterial genome. Here we investigate how a bacterial pan-genome might influence gene essentiality and whether essential genes that are initially critical for the survival of an organism can evolve to become non-essential. By using Transposon insertion sequencing (Tn-seq), whole-genome sequencing and RNA-seq on a set of 36 clinical Streptococcus pneumoniae strains representative of >68% of the species' pan-genome, we identify a species-wide 'essentialome' that can be subdivided into universal, core strain-specific and accessory essential genes. By employing 'forced-evolution experiments', we show that specific genetic changes allow bacteria to bypass essentiality. Moreover, by untangling several genetic mechanisms, we show that gene essentiality can be highly influenced by and/or be dependent on: (1) the composition of the accessory genome, (2) the accumulation of toxic intermediates, (3) functional redundancy, (4) efficient recycling of critical metabolites and (5) pathway rewiring. While this functional characterization underscores the evolvability potential of many essential genes, we also show that genes with differential essentiality remain important antimicrobial drug target candidates, as their inactivation almost always has a severe fitness cost in vivo.

Fig. 1. Pan-genome coverage and essential genes of Streptococcus pneumoniae PG collection.

a, Number of gene clusters belonging to the core and accessory genomes for the 208 strains study group determined by three different methods (PanX, BF-Clust and PPanGGOLiN). Orange represents how much of the accessory genome (>68%) is present in the PG collection. b, Number of genes called essential, non-essential, or uncertain across 21 strains, with mutant libraries ranked by saturation measured as the percentage of the total unique TA sites in the genome occupied by the ‘mariner’ transposon. c, Left: three representative genes showing the criteria used to classify genes as ‘universal’, ‘strain-dependent’ and ‘accessory essentials’. Each dot represents the essentiality score obtained for a gene in a specific strain. A score >0.9916 categorizes a gene as essential and <0.0424 as non-essential. Right: the number of genes assigned to each class. d, Distribution of fitness effects of non-essential and strain-dependent essentials in vitro in rich medium (SDMM) and in vivo in a mouse infection model (lungs). Black lines inside each plot represent the median, and dotted lines the first and third quartiles. P values were obtained from Tukey’s corrected one-way analysis of variance (ANOVA). e, Validation of gene essentiality. Transformation efficiency (TE) ratio is the ratio between the number of colony forming units (c.f.u.) obtained after transforming 1,000 ng of a PCR product targeting an essential gene and the c.f.u. obtained after transforming 1,000 ng of a PCR product that inserts into a neutral genomic region. TE ratios are shown for five universal and six strain-dependent essential genes in four different strain backgrounds. Experiments were performed in triplicate for each strain. Each dot in the panel represents a single strain’s experimental replicate. Black dots indicate strains where the targeted gene is non-essential, and red dots where the gene is essential. Squares indicate transformation experiments with successful knockout recovery. Source data

Fig. 2. Δcmk/SP_1603 alters polar membrane lipid composition.

a, Schematic representation of Cmk’s function in CMP recycling. Molecules in bold represent the substrate and product of the enzyme’s reaction (CMP, CDP), and the ones directly involved in CMP production. cmk deletion affects the pathways where CDP_diacylglycerol (red and bold) is an intermediate metabolite. b, Growth of wild type and (complemented) SP_1603 knockouts in SDMM. c, Genes involved in the synthesis of the phospholipids (red tone molecules) PtdGro and CL, and the glycolipids (violet tone molecules) Glu-DAG and Gal-Glu-DAG. Molecule colours relate with bar colours in d. d, Membrane polar lipid composition of three strains and their SP_1603 knockouts. Data show existing differences in lipid composition between strains, highlighting how there is no single perfect way of building a lipid membrane. Bar graph represents the mean read counts (n = 2) of the different ¹⁴C-labelled polar lipids in the membrane of the WT strains PG04, PG27 and TIGR4 (filled bars) and SP_1603 knockouts (open bars). Red bars, PtdGro counts; pink, CL; dark violet, Glu-DAG; and light violet, Gal-Glu-DAG. e, Growth phenotypes of the different strains grown in the presence of daptomycin. For the WT strains, increasing concentrations of daptomycin caused an extended lag phase, diminution of the growth rate and lower maximum ODs. In contrast, daptomycin only mildly affected growth of the SP_1603 knockouts. (i) WT growth curves in the presence of 30 μg ml⁻¹ daptomycin. (ii) Growth rate measured as hr⁻¹ (left) and maximum OD (right) of the different strains growing in increasing concentrations of daptomycin. (iii) Growth curves of the different strains at 20 (TIGR4), 30 (PG04) and 40 (PG27) μg ml⁻¹ daptomycin. (iv) Growth rate (left) and maximum OD (right) of the curves depicted in iii. P values were obtained from an ordinary one-way ANOVA (uncorrected Fisher’s least significant difference (LSD) test). Data shown in b and e are the results from a single experiment with n = 3 biologically independent samples per condition. Independent repetition of each experiment in triplicate showed identical results. Error bars represent s.d. Source data

Fig. 3. Capsule sugar transferase characterization.

a, Schematic of capsule biosynthesis in S. pneumoniae. Genes inside the red box are accessory essentials, likely because the essential molecule UP gets ‘trapped’ if these downstream capsule genes are non-functional. b, Genetic interactions identified by Tn-seq using libraries constructed in the background of four different strains with a sugar transferase knockout. Each data point represents a gene; red dots, mostly in capsule metabolism, depict a suppressive interaction with the sugar transferase. These suppressor interactions are possible because locking the attachment of the initial UDP-sugar residue to UP prevents lethal dead-end lipid carrier trapping. c, Growth of WT strains PG06 and TIGR4 and their sugar transferase knockouts in rich medium at two choline concentrations. Autolysis in S. pneumoniae relies on the amount of LytA protein (SP_1937) bound to the choline moiety of wall teichoic acids. Accordingly, the sugar transferase knockout-observed early autolysis relates to the amount of choline in the medium. Indicated are time differences in the onset of autolysis between the two conditions for WT (black) and knockouts (red). Shown are data from n = 3 biologically independent samples; independent repetition (3) showed similar results; error bars represent s.d. d, The sugar transferase has a synthetic lethal interaction with genes involved directly and indirectly in cell wall and teichoic acid biosynthesis. One of these interactions occurs with the response regulator CiaR (SP_0798), which upregulates teichoic acid biosynthesis genes. The requirement of CiaR suggests that teichoic acid production needs to be upregulated in the context of the sugar transferase knockout. The horizontal bars of squares indicate the type of interaction identified between the sugar transferase and the indicated gene for each strain (white squares indicate no identified interaction for a specific strain). Metabolites in red (for example, UDP-Glucose) are known to accumulate in the context of acapsular S. pneumoniae strains. LTA, lipoteichoic acid; WTA, wall teichoic acid. Source data

Fig. 4. Redundancy in magnesium transport is strain-dependent.

a, SWISS-MODEL structures of core strain-dependent essential CorA (SP_0185) (Qmean = −2.27) and the second non-essential transporter SP_1751 (Qmean = -−2.82). Uniform red colour of the structure of SP_0185 indicates that the protein sequence of the transporter is fully conserved among S. pneumoniae strains. Non-essential transporter coded by SP_1751 conserved amino acids are represented in grey. Colours other than grey indicate residues with allelic variation across the PG collection. b, Genetic interactions with SP_0185 in strain PG04. Each dot represents a gene interacting with SP_0185, its type of interaction and functional category. Note that 5 out of 22 genes with a positive interaction code for magnesium-binding proteins, suggesting that preventing the use of Mg²⁺ for non-essential reactions can reduce the impact of ΔSP_0185. c, Growth of WT PG04 and ΔSP_0185 in SDMM with different MgCl₂ concentrations and hexammine cobalt. Low Mg²⁺ significantly affects growth of ΔSP_0185 compared with WT PG04, while addition of hexammine cobalt abolishes growth of the knockout. Data shown are from n = 3 biologically independent samples, while independent repetition (3) showed similar results. Error bars represent s.d. d, Doubling time versus maximum OD observed for different strains growing in rich medium with MgCl₂ at 0.15 mM and in the absence/presence of hexammine cobalt. Growth of WT strains in which SP_0185 is essential is similar to that of PG04-ΔSP_0185 (black dots). White and red dots represent WT strains where SP_0185 is non-essential or essential, respectively. Dots numbered 1 to 4 correspond to the numbered curves in c. e, Schematic showing the most common genomic configuration of SP_0185 and neighbouring uvrA, while some strains have a prophage in between. f, Transcript abundances of the two corA genes. Each dot represents a different WT strain, colours indicate SP_0185 essentiality and arrows indicate prophage presence. R and P values were derived from a least-squares linear regression. Source data

Fig. 5. Farnesyl-PP synthase (SP_1205) essentiality is easily bypassed.

a, Synthesis pathway of UP. The enzyme encoded by SP_1205 catalyses the condensation between isopentenyl diphosphate (IPP) and geranyl diphosphate (GPP) into farnesyl diphosphate (FPP), which is then converted to UPP by enzyme SP_0261, after which UPP is dephosphorylated by SP_0457 into its active form UP. b, Disaccharide metabolism and bacitracin resistance related genes are involved in bypass mechanisms of SP_1205. Suppressor mutations of SP_1205 essentiality were identified with WGS (boxes with white background) and synthetic lethality interactions (boxes with pink backgrounds) identified with Tn-seq using libraries constructed in a PG06-SP_1205 knockout. The identification of three functionally related suppressive SNPs (SP_1176 A17P, SP_1884 I34T and SP_1885 H182N) suggests that transport of the disaccharide trehalose (or a similar compound) is a central hub for bypassing SP_1205. Gene SP_1205 has a synthetic lethal interaction with two enzymes (SP_1883 and SP_1724) that hydrolyse the disaccharides trehalose and sucrose to produce glucose-6-phosphate (which feeds into glycolysis), the pentose-phosphate pathway and capsule biosynthesis (left panel). One of the genes with synthetic lethality with SP_1205 is the UPP phosphatase BacA (SP_0457). Interestingly, a TIGR4-derived SP_1205 knockout presented two compensatory mutations associated with the cell wall: one mutation is located in murE (SP_1530 A430S), which is involved in peptidoglycan synthesis, and the second mutation is located in a non-characterized membrane protein coding gene (SP_0454 H581N), which forms a complex with BacA (right panel). Source data

Fig. 6. S. pneumoniae strains can evolve different mechanisms to overcome the essentiality of formate tetrahydrofolate reductase (SP_1229).

a, Role of the product of the strain-dependent essential gene SP_1229 in the one-carbon pool by folate pathway and suppressor mutations in PG16 bypassing SP_1229 essentiality (boxes with white background). Dashed arrows indicate multiple metabolic reactions (not-shown) that connect the depicted pathways. Two of these mutations, present in two independent backgrounds, create premature stop codons in SP_0918 (Q136*) and SP_1378 (Q319*), which connect to the one-carbon pool by folate pathway through the S-adenosyl-methionine cycle. SP_0918, when deleted in strain D39, diminishes the production of spermidine, while SP_1378 codes for an S-adenosyl-l-methionine (SAM)-dependent RNA-methyltransferase. Three additional PG16-SP_1229 knockouts contain suppressor mutations in genes involved in guanosine ribonucleotide metabolism and (p)ppGpp biosynthesis (SP_0012-T104A and SP_1645-S310F, SP_1645-A378T and SP_1738-Y35C, respectively). b, Growth of WT strains PG04, TIGR4 and PG16, and their SP_1229 knockouts in rich medium in the presence/absence of the tetrahydrofolate (THF) synthesis inhibitory drug SXT. SP_1229 is non-essential in PG04 and TIGR4, and essential in PG16. Data are from n = 3 biologically independent samples per condition, independent repetition of each experiment (3) showed similar results. Error bars represent s.d. c, Identification of genetic interactions through Tn-seq using libraries constructed in two backgrounds where SP_1229 is non-essential (PG04 and TIGR4) and two knockouts derived from PG16 with different suppressor mutations (SP_0918 Q136* and SP_1378 Q319*). Source data

Extended Data Fig. 1. Streptococcus pneumoniaePan-Genome collection phylogenetic diversity and pan-genome coverage.

a) Maximum-likelihood phylogenetic tree based on core genes SNPs of the 208 strains in the pan-genome study group. This group consists of the 36-strain PG-collection, 23 reference genomes, 111 strains from several worldwide surveillances’ programs (30 from Massachusetts, US, 30 from Maela, Thailand, 30 from Southampton, UK, and 21 from Malawi), and 38 isolates obtained before 1974. Labels from the strains selected for our PG collection are in bold and have a larger font size. Black outer strip color symbol and black tree branches also indicate the PG collection strains. Panel shows that PG-collection scatters evenly through the phylogenetic tree. b) Number of BF-clusters not-represented in the 36 strain PG-collection. More than half (55%) of the non-represented genes are present in less than 5 isolates. c) Pan-genome content saturation. Source data

Extended Data Fig. 2. Characteristics of the different essentialome classes.

a) Gene transcript abundance obtained by RNAseq collected from early exponential phase cultures of the 36 PG-collection strains growing in rich THY medium. Graph shows genes grouped by their essentialome classification. Black lines inside each plot represent the median, and dotted lines the first and third quartiles. p-values are obtained from a Tukey’s corrected one-way ANOVA. Results show that core universal, core strain-dependent and accessory essential genes tend to be more highly expressed compared to non-essential genes; b) Sequence diversity of S. pneumoniae genes split into the different essentialome classes and measured using the diameters of the clusters obtained by BF-Clust. Black lines inside each plot represent the median, and dotted lines the first and third quartiles. p-values are obtained from a Kruskal-Wallis test comparison. Strain-dependent versus accessory and non-essential classes presented the same p-value (<0.001). Results show that core universal and core strain-dependent essential genes are less genetically diverse than accessory and non-essential genes. The higher diversity in accessory essentials is at least partially caused by many of them being involved in ‘addiction-like’ systems, for example, systems consisting of two or more genes in which one of the genes is only required if the other/s are present (for example phage repressors, antitoxins from toxin-antitoxin systems, and methylases from restriction-modification systems). These genes may thus tolerate more mutations than a core genome gene involved in central processes, such as DNA replication; c) Functional categories enriched in Universal, Core Strain-dependent, and Accessory essentials. Each chart represents a COG category and the fraction of genes belonging to each essentialome class. Color coding is the same as previous panels. Asterisks indicate enrichments of essentialome class within COG category (adjusted p-value< 0.05). Analysis shows that universal essential genes are enriched in 5 central cellular processes categories, while core strain-dependent and accessory essentials are enriched for only two of these categories. The most enriched functional category of accessory essentials (cell wall, membrane and envelope biogenesis) can be explained by the high diversity of essential capsule type-specific biosynthesis genes. Source data

Extended Data Fig. 3. Methodology for essentiality validation, and evolvability, and genetic interaction identification of essential genes.

a) Strategy deployed to obtain knockouts in different genes and representative results for each essentialome class. The amount of PCR knockout product used (1000 ng) was 20 times higher than routinely used. b) Knockout confirmation and identification of suppressor mutations. WGS analysis with breseq of clones selected from transformations revealed whether clones were true knockouts (targeted gene coverage is near zero, middle panel) or merodiploids (target gene coverage same as average, homology arms coverage twice the average, bottom panel). WGS and breseq were also used to identify SNPs in genes putatively involved in gene essentiality suppressor mechanisms. c) Genetic interactions between strain-dependent and non-essential genes. Tn-Seq experiments using libraries constructed in different knockouts identified four different types of interactions: 1) a positive interaction occurs when a double mutant (knockout, “Δ”, plus transposon, “::tn”) presents a growth advantage compared to the individual mutants; 2) a suppressive interaction occurs when an essential gene becomes non-essential in the context of the knockout; 3) in a negative interaction the double mutant presents a growth defect, and 4) in a synthetic lethal interaction the double mutant is non-viable. Tn-Seq fitness calculations were used to identify positive and negative interactions, and combined with the TRANSIT binomial method to identify suppressive interactions and synthetic lethality. Source data

Extended Data Fig. 4. SP_1603 knockout characterization.

a) Growth phenotype of wild type and derived SP_1603 knockouts in rich medium (SDMM) in the presence/absence of additional uracil, cytidine, or choline. Knockout phenotypes are potentially caused by a decreased replication rate provoked by a dCTP shortage. While addition of excess uracil and/or cytidine could potentially restore growth, it had no effect. Alternatively, CMP accumulation could be negatively affecting pathways or reactions in which CMP is a product. In S. pneumoniae, genes SP_1273 and SP_1274 generate CMP from CDP-choline and are involved in decorating teichoic acid with choline^,. While CMP accumulation could thus affect teichoic acid choline decoration, the addition of 10x more choline to the medium also did not compensate growth. Data are obtained from n = 3 biologically independent samples per condition, independent repetition (3) showed similar results. Error bars represent standard deviation. b) TLC run of C-14 acetate labeled lipids extracted from the different strains. Figure 4c summarizes this experiment. Source data

Extended Data Fig. 5. Genes and pathways with genetic interactions with capsule sugar transferase genes.

Selection of genetic interactions identified by Tn-Seq using libraries constructed in sugar transferase knockouts derived from four different strains. The order of the bars consisting of 4 squares indicate the strain, and the color the type of interaction that was identified. White square depicts no identified interaction. Metabolites in red bold indicates those that accumulate in acapsular S. pneumoniae strains. Gal-Glu-DAG: galactosyl-glucosyl-diacylglycerol, Glu-DAG: glucosyl-diacylglycerol, TA: teichoic acid, LTA: lipoteichoic acid, WTA: wall teichoic acid. Source data

Extended Data Fig. 6. Minodronate (MNDR) specifically targets S. pneumoniae farnesyl-PP synthase (SP_1205) and synergizes with vancomycin.

Growth of WT strains PG06 (in which SP_1205 is non-essential), TIGR4 and PG16 (in which SP_1205 is essential) and their derived SP_1205 knockouts in rich medium (SDMM), in the presence of non-inhibitory concentrations of vancomycin (VNC), in the presence or absence of the SP_1205 targeting drug MNDR, or both. Panels show the suppressor mutations that enable SP_1205 knockout in strains it is essential in. MNDR is a bisphosphonate drug currently in a phase-3 trial for human use, which targets human FPP synthase to suppress bone resorption and bone loss. However, there is evidence that bisphosphonates may also have antimicrobial potential. The drug inhibits growth of the wild-type strains but not the SP_1205 knockouts. Because UP is critical for the cell wall, inhibition of UP synthesis may thus synergize with cell-wall synthesis inhibitors, like vancomycin. The combination of MNDR with vancomycin at a subinhibitory concentration has a synergistic effect on wild-type strains while there is no effect in ΔSP_1205 strains. Results shown are from n = 3 biologically independent samples per condition, independent repetition of each experiment (3) showed similar results. Error bars represent standard deviation. Source data

Extended Data Fig. 7. S. pneumoniae requires SP_1205 when growing in the presence of sucrose or bacitracin.

a) Carbon source and SP_1205 requirement. Growth of WT strains PG06 (in which SP_1205 is non-essential) and TIGR4 (in which SP_1205 is essential) and their derived SP_1205 knockouts in rich medium (SDMM) with either glucose or sucrose as the carbon source, and with or without MNDR. b) Bacitracin sensitivity of SP_1205 knockouts. Growth of WT strains PG06 (in which SP_1205 is non-essential), TIGR4 and PG16 (in which SP_1205 is essential), and their derived SP_1205 knockouts in rich medium (SDMM) in the presence or absence of the UPP recycling inhibitory drug bacitracin (Bac). Data in a and b are the results from n = 3 biologically independent samples per condition, independent repetition of each experiment (3) showed similar results. Error bars represent standard deviation. Source data