Convergent Within-Host Adaptation of Pseudomonas aeruginosa through the Transcriptional Regulatory Network

Abstract

Bacteria adapt to their host by mutating specific genes and by reprogramming their gene expression. Different strains of a bacterial species often mutate the same genes during infection, demonstrating convergent genetic adaptation. However, there is limited evidence for convergent adaptation at the transcriptional level. To this end, we utilize genomic data of 114 Pseudomonas aeruginosa strains, derived from patients with chronic pulmonary infection, and the P. aeruginosa transcriptional regulatory network. Relying on loss-of-function mutations in genes encoding transcriptional regulators and predicting their effects through the network, we demonstrate predicted expression changes of the same genes in different strains through different paths in the network, implying convergent transcriptional adaptation. Furthermore, through the transcription lens we associate yet-unknown processes, such as ethanol oxidation and glycine betaine catabolism, with P. aeruginosa host adaptation. We also find that known adaptive phenotypes, including antibiotic resistance, which were identified before as achieved by specific mutations, are achieved also through transcriptional changes. Our study has revealed novel interplay between the genetic and transcriptional levels in host adaptation, demonstrating the versatility of the adaptive arsenal of bacterial pathogens and their ability to adapt to the host conditions in a myriad of ways. IMPORTANCE Pseudomonas aeruginosa causes significant morbidity and mortality. The pathogen's remarkable ability to establish chronic infections greatly depends on its adaptation to the host environment. Here, we use the transcriptional regulatory network to predict expression changes during adaptation. We expand the processes and functions known to be involved in host adaptation. We show that the pathogen modulates the activity of genes during adaptation, including genes implicated in antibiotic resistance, both directly via genomic mutations and indirectly via mutations in transcriptional regulators. Furthermore, we detect a subgroup of genes whose predicted changes in expression are associated with mucoid strains, a major adaptive phenotype in chronic infections. We propose that these genes constitute the transcriptional arm of the mucoid adaptive strategy. Identification of different adaptive strategies utilized by pathogens during chronic infection has major promise in the treatment of persistent infections and opens the door to personalized tailored antibiotic treatment in the future.

Keywords: Pseudomonas aeruginosa; antibiotic resistance; infectious disease; microbial genetics; transcription regulation network; transcriptional adaptation; within-host adaptation.

FIG 1

Different genetic events can converge at the transcriptional level. Gene X is positively regulated by TF2 and TF4, TF2 is positively regulated by TF1, and TF4 is positively regulated by TF3. Three different LOF events occurred in the three different strains. TF2 underwent LOF in strain 1, TF4 underwent LOF in strain 2, and TF1 underwent LOF in strain 3. Despite these differences, all three events are expected to lead to some reduction in the expression of gene X, demonstrating transcriptional convergence. Note that all analyses in this study were performed at the level of progenitor-progeny isolate pairs of the strains.

FIG 2

Mean regulatory outcome scores (mROS) of genes across the transcriptional regulatory network of P. aeruginosa. The transcriptional regulatory network of P. aeruginosa (25) is shown, with nodes colored by the mROS values of the corresponding genes. Red, positive mROS; blue, negative mROS; arrows, positive regulation; blunt-end arrows, negative regulation.

FIG 3

Distribution of mean regulatory outcome scores (mROS values) and heterogeneity scores (HS values). (A and B) mROS (A) and HS (B) values of all genes in the transcriptional regulatory network. Notable genes mentioned in the main text are denoted. (C) Plot of mROS versus HS values of all genes. Genes with mROS values of >0.15 and HS values of >2.5 are marked in red, and those with mROS values of <−0.15 and HS values of >2.5 are marked in blue.

FIG 4

Heterogeneity score (HS) associates novel processes and genes with host adaptation. (A) mROS and HS values of the components of the ethanol oxidation system. The ErbSR TCS regulates most components involved in ethanol oxidation, including the EraSR TCS, which regulates exaA. For clarity, additional regulators are not shown. All genes involved in this system have low mROS values (blue) and high HS values (orange). Wide arrows show genes in their corresponding genomic locations; all genes reside in the same genomic region except mqoA, which is in a different locus. Arrows pointing to the right correspond to genes on the forward strand, and arrows pointing to the left correspond to genes on the reverse strand. Same-operon genes are in the same color. Green dashed arrows connect genes to their products. Black arrows correspond to positive regulation by transcription factors. (B) LOF events of TFs/SFs leading to predicted increased expression of the pyeR-pyeM-xenB operon in the different strains. Arrows, positive regulation; blunt-end arrows, negative regulation; green node, pyeR-pyeM-xenB; blue nodes, nodes of genes encoding TFs/SFs whose LOFs lead to predicted increased expression of pyeR-pyeM-xenB; gray nodes, all other nodes. The size of a node is proportional to the number of progenitor-progeny pairs in which it underwent LOF.

FIG 5

Mean regulatory outcome scores (mROS) and heterogeneity scores (HS) of selected KEGG pathways. Selected KEGG pathways with the highest average mROS values (A), the lowest average mROS values (B), and the highest average HS values (C). Green dots indicate the mean value for the genes included in the pathway, and gray dots indicate the mean for all genes not included in the pathway. Pathways passing the statistical test for the mROS or HS values are written in red. *, P value < 0.1; **, P value < 0.01; ***, P value < 0.001; ****, P value <0.0001. All P values are corrected for the testing of multiple hypotheses. TCA, tricarboxylic acid.

FIG 6

The concordance score distinguishes a subset of strains associated with mucoid adaptation. (A) Histogram of the concordance scores of progenitor-progeny pairs for cluster 14. (B) Box plots of the concordance scores for cluster 14 of progenitor-progeny pairs with and without mucA LOF mutations.

FIG 7

TRN propagation algorithm. An example of the computation of the regulatory outcome score (ROS) by the TRN propagation algorithm. Gene C (node C) is positively regulated by TF B (node B) and negatively regulated by TF D (node D). TF B is positively regulated by TF A (node A). In the progenitor-progeny pair in the example, TFs A and D were determined to undergo loss-of-function (LOF). In the blue box, the effect of the LOF of TF A on gene C is demonstrated. TF A is two edges away from gene C, and the product of the weights of all the edges between them is 1 × 1 = 1. The final score of the impact of TF A on gene C is therefore −1 × (product of weights)/number of edges = −0.5. In the red box, the effect of the LOF of TF D on gene C is demonstrated. TF D is one edge away from gene C, and the product of the weights of all the edges between them is −1. The final score of TF D on gene C is therefore −1 × (product of weights)/number of edges = 1. In the green box, the final ROS of gene C is demonstrated. It is defined as the sign of the sum of the scores of gene C corresponding to all the TFs predicted to affect its expression in the progenitor-progeny pair, in this case the sign of −0.5 + 1, which is +1. The expression of gene C is therefore predicted to increase in the progenitor-progeny pair due to the summed effects of all LOF events.