Introduction

Thanks to advances in sequencing technology, it is possible to analyze the transcriptomes of many bacterial isolates with little effort, at a low cost, and within a short period of time. RNA sequencing data offer the potential for large-scale comparative studies, which aim at deciphering coordinated regulation of genes within an organism. Thus, new insights into complex bacterial adaptation strategies, for example during an infection process, can be achieved1.

The Gram-negative, opportunistic pathogen Pseudomonas aeruginosa is considered one of the most important pathogens involved in nosocomial infections2. The bacterium causes high morbidity and mortality, especially in patients with weakened immune systems and burn victims3. In cystic fibrosis (CF) patients, lifelong chronic P. aeruginosa infections lead to an over-activation of the immune system and to extensive tissue damage in the patients’ lungs4,5,6. In the course of these long-term infections, P. aeruginosa not only adapts by orchestrating the transcription of complex regulatory gene networks, but the bacteria also develop a number of pathoadaptive mutations that promote survival in the harsh environment of the human host7,8,9,10. The adapted phenotypes have fitness advantages over the originally infecting wild-type strain, e.g., by escaping immune recognition, saving energy by using public goods or developing resistance to antimicrobial agents4,7,11,12. Understanding the evolutionary trajectories that P. aeruginosa follows during long-term infections is an important task in order to optimize treatment and predict the clinical course of the infection13,14,15.

In this study, we analyzed the genomes and transcriptomes of a large number of clinical P. aeruginosa isolates previously recorded under both planktonic and biofilm growth conditions14,16,17. We found that complex gene expression patterns evolved independently in different clinical P. aeruginosa isolates and that a majority of the observed expression patterns could be attributed to a defective lasR gene, encoding the major quorum sensing (QS) regulator LasR18,19. Strikingly, whereas a great number of genes were differentially regulated in the lasR-proficient isolates upon shifting from planktonic to biofilm growth conditions, the lasR-defective isolates exhibited a transcriptional signature under planktonic culture conditions that already resembled a biofilm transcriptional profile. Our results demonstrate that the analysis of transcriptomic and genomic data from a large variety of clinical isolates provides new insights into how adaptive mutations drive gene expression programs and regulate phenotypic outcomes responsible for bacterial adaptation to altered and challenging host niche habitats.

Results

Distribution of single gene expression values across clinical P. aeruginosa isolates

We have previously transcriptionally profiled a collection of 414 clinical P. aeruginosa isolates following growth in lysogeny broth (LB) media until the early stationary phase (OD600 of 2)16,17,20. Here, we re-analyzed the data and recorded the distribution of individual gene expression values across all 414 clinical isolates. As exemplified in Fig. 1a, we identified genes that were expressed at high levels in most isolates, while other genes had overall low expression levels. In general, it appears that the higher the overall expression level of a particular gene, the higher the variation in expression level between the different clinical isolates (Fig. 1b). We also observed that some genes showed a much greater variation in their expression values, with some isolates having lower levels and others having higher levels. This at least bi-modal gene expression pattern is exemplified in Fig. 1a.

Fig. 1: Distribution of gene expression profiles across 414 clinical P. aeruginosa isolates.
figure 1

a The distribution of the expression values is shown as an example for the three genes cobK (blue, low expression), sahH (black, high expression), and mexY (red, bi-modal expression) across the 414 clinical isolates. Distribution was calculated using the density() function in R with nrpg values as input. b Mean gene expression values relative to the expression value range of all tested isolates. Expression values were normalized for gene length (RPKM). Blue line = linear regression; adjusted R2 = 0.7.

Full size image

We identified overall 136 genes with a bi-modal gene expression distribution by the use of an algorithm that distinguishes between uni- and at least bi-modality expressed genes (Supplementary Data 1, in some cases even a tri-modal distribution could be seen). We then classified the isolates into either a high- or a low-expression group with respect to each of the 136 at least bi-modal genes (distributions and cut-off visualization in Supplementary Fig. 7). The resulting binary dataset was clustered in a heat-map and we identified five gene clusters containing genes with either high or low expression patterns in the individual clinical isolates (Fig. 2). The largest cluster contained 76 genes (Fig. 2; gene cluster 2). High or low expression of those genes was identified as the main driver of the clustering of the individual clinical isolates into two major groups (Fig. 2). In this gene cluster 2, we found a significant enrichment of genes belonging to the KEGG pathway functional categories “phenazine biosynthesis”, “biofilm formation”, and “quorum sensing” (Supplementary Data 1).

Fig. 2: Distribution of bi-modal gene expression across 414 clinical P. aeruginosa isolates.
figure 2

Multimodal (R-package ‘multimode’, adjusted p-value ≤ 0.1) gene expression distributions were recorded, and the distribution of the 136 bi-modal genes (y-axis), which showed high expression (red) or a low expression (blue) across the 414 clinical isolates (x-axis), was recorded. The resulting binary matrix was used as a basis for a heat map (binary clustering of rows and columns). Clinical isolates harboring non-synonymous mutations in lasR are marked in green, insertions/deletions/premature stop codons (InDel/STOP) are marked in yellow. Clustering uncovered 5 groups of genes with a similar bi-modal expression pattern in the clinical isolates. They were colored on the y-axis for a better overview (Gene Cluster 1-5). Only bi-modally expressed genes that are assigned in both reference strains, PA14 and PAO1 were included.

Full size image

As many of the genes of these functional categories are under the control of the major QS regulator LasR, we evaluated whether mutations of the lasR gene could explain the distinct expression patterns of these at least bi-modally distributed genes. For this purpose, we analyzed previously published whole-genome sequencing data of the isolates16 and evaluated the lasR allele status in the 414 clinical isolates. In addition to gene-inactivating mutations due to frameshifts (insertion or deletions, summarized as InDels) or pre-mature stop codons (STOP), there were also non-silent mutations that led to changes in the amino acid sequence of LasR. As depicted in Supplementary Fig. 1, amino acid exchanges were found throughout the entire protein sequence. Nevertheless, there were marked hotspots in the N-terminal and C-terminal region, which are involved in dimerization and DNA binding, respectively21,22, indicating that they might result in the functional inactivation of LasR23.

There was a clear correlation between the expression status of bi-modal genes and the allele status of the transcriptional regulator LasR (Fig. 2; yellow and green squares). This indicates that variations in the expression pattern of a majority of the bi-modally expressed genes across the clinical isolates were associated with the presence/absence of a functional lasR gene and underscores the dominant role of LasR in shaping the transcriptional profile in our collection of clinical isolates. We did not identify large groups of clinical isolates that exhibited a characteristic expression pattern of genes belonging to the two other major clusters (Fig. 2; gene clusters 1 and 3). This suggests that there does not seem to be another major regulatory gene (in addition to lasR) that influences a comparably large fraction of the bi-modally expressed genes. Nevertheless, future work should concentrate on the identification of additional genes, which influence the expression of smaller groups of bi-modal genes.

Identification of the core lasR regulon

Since LasR shapes the transcriptional profile of a large fraction of our clinical isolates, we aimed for the identification of genes that were differentially regulated in the group of clinical isolates that expressed a wild-type lasR allele versus the group of isolates harboring a non-functional lasR. We concentrated on 28 randomly selected clinical isolates with lasR alleles exhibiting neither synonymous nor non-synonymous sequence variations in the lasR gene (as compared to the PA14 or PAO1 lasR gene) and 21 clinical isolates with lasR inactivating (InDels/stop) mutant alleles (hereinafter described as lasRWT and lasR* isolates, respectively). Both, planktonic and biofilm transcriptomes, were available for those 49 clinical isolates. We then calculated the differential gene expression between the lasRWT and lasR* group. In total, 722 genes were differentially regulated (corrected p-value ≤ 0.05 and log2-fold change |log2FC | ≥ 1). Of those, 412 genes were up- and 310 genes were downregulated in lasR* isolates compared to lasRWT isolates (Supplementary Data 2).

As expected, inactivation of lasR in the lasR* isolates led to a reduced expression of genes encoding for important virulence factors, such as phenazines (phzA-G), the protease LasA (lasA), the elastase LasB (lasB) or rhamnolipids (rhlAB). Accordingly, we identified genes belonging to the KEGG24 pathways of “phenazine biosynthesis”, “quorum sensing”, and “biofilm formation” as being significantly enriched (adjusted p-value ≤ 0.05, hypergeometric test) in the group of genes that were differentially expressed between the clinical lasRWT and lasR* isolates. We next phenotypically characterized a random selection of lasRWT (n = 10) and lasR* isolates (n = 8). In agreement with the finding that LasR governs the production of a range of virulence factors, we found a significantly reduced virulence of the lasR* isolates in a Galleria mellonella infection model compared to the isolates with functional LasR (Supplementary Fig. 2a). In addition, the lasR* isolates showed a significantly reduced protease and elastase production following growth in planktonic cultures (Supplementary Fig. 2b, c).

In order to further elucidate on the LasR regulon, we recorded the transcriptional profile of the reference strain UCBPP-PA14 (PA14 WT25) as compared to a clean lasR deletion mutant (PA14 ∆lasR26) under the same growth conditions as applied for the clinical isolates (Supplementary Data 3). Of the genes that were significantly downregulated in the lasR* versus lasRWT isolates, 90% were also significantly downregulated in PA14 ∆lasR vs. PA14 WT (Fig. 3a).

Fig. 3: The LasR regulon.
figure 3

Venn diagrams showing the overlap between genes that are downregulated (a) or upregulated (b) in lasR* as compared to lasRWT isolates (21 lasR* vs. 28 lasRWT isolates; blue) and the genes that are regulated respectively in PA14 ∆lasR as compared to PA14 WT (3 biological replicates; red) (log2FC > 1, p value < 0.05). Number of genes that were commonly (log2FC > 1) downregulated (c) or upregulated (d) in 1, 2, 3,…21 clinical lasR* isolates compared to all analyzed lasRWT isolates recorded (yellow bars). The proportion of genes that was identified to be also regulated in the constructed PA14 ∆lasR (compared to PA14 WT) is marked in red.

Full size image

In order to identify strain-specific as well as common transcriptional signatures associated with lasR deficiency, we calculated log2FCs of each of the 21 lasR* isolates against the group of lasRWT isolates. Totally, 1202 individual genes were down-regulated (log2FC ≤ −1) in at least 10 lasR* isolates and 74 genes were downregulated in all 21 lasR* isolates as compared to the group of lasRWT isolates (Supplementary Data 6). Of note, 72 of those 74 genes that were commonly downregulated across all 21 clinical lasR* isolates, were also identified as being part of the lasR regulon in the PA14 WT strain background (Fig. 3).

Our results clearly indicate that there is a core gene set, which is expressed in a lasR-dependent manner across many different P. aeruginosa isolates thus underscoring the value of a transcriptional profiling approach of multiple clinical isolates.

The complementary analysis of lasR-dependent gene expression in clinical isolates as well as in a targeted engineered reference strain allowed us to define the lasR core regulon, which consists of overall 138 genes (Table 1). Only those genes were included, which were downregulated in at least 90% (19 out of 21) of the analyzed single clinical lasR* isolates compared to the lasRWT isolates (n = 28), and which were also significantly downregulated in PA14 ∆lasR as compared to its PA14 wild-type.

Table 1 The lasR core regulon.
Full size table

Upregulation of gene expression in clinical isolates exhibiting a non-functional LasR

It is well established that LasR is a transcriptional activator that governs a large QS regulon27. However, we also found that lasR mutations in the clinical isolates led to an increased expression of certain genes when compared to lasRWT isolates (Fig. 3b). Table 2 lists a selection of 123 genes (organized in operon structures) of distinct functional categories that were expressed at higher levels in the absence of a functional lasR (full gene list in Supplementary Data 2). The majority of those genes (n = 57) are associated with the degradation of complex carbon sources, such as aromatic compounds or lipids. Another 21 genes are involved in various mechanisms of iron acquisition (such as hasRDE and pvdGEFN), and the kdpABC operon is associated with sensing low extracellular potassium levels. In addition, 12 genes were upregulated in the lasR* isolates that encoded for cup fimbriae or the biosynthesis of the pel exopolysaccharide. Moreover, the spermidine transport system encoding potABCD exhibited increased expression levels in lasR* isolates compared to lasRWT isolates. Furthermore, acquisition of extracellular phosphate either via the phn transport system, the (extracellular) degradation of phosphatidylcholine (glpQ and glpD), or the type 2 secretion system (T2SS, hxc) seems to be pronounced in the lasR* isolates. Interestingly, also multiple resistance genes including the multi-drug efflux pump mexCD-oprJ were significantly upregulated in lasR* isolates.

Table 2 Selection of upregulated genes in lasR* isolates under planktonic growth conditions as compared to lasRWT isolates. For a complete list of regulated genes see Supplementary Data 2.
Full size table

In contrast to the common set of genes that were consistently downregulated in the majority of the lasR* isolates (Fig. 3c), there were by far fewer genes that were commonly upregulated across the different lasR* isolates. As depicted in Fig. 3d, only one gene (PA14_33800) of the overall 3797 genes that reached the log2FC threshold in at least one clinical isolate was upregulated in a maximum of 17 clinical lasR* isolates, and only 18 genes were upregulated in at least 10 of the 21 clinical lasR* isolates. This indicates that an upregulation of genes in a lasR-deficient background is to a much higher degree dependent on the individual characteristics of a strain. In line with this, the overlap between the genes that were upregulated in both, PA14 ∆lasR and the lasR* clinical isolates, was much smaller as compared to the genes that were commonly downregulated (Fig. 3b).

LasR becomes less important under biofilm growth conditions

In addition to the transcriptional profiles of the lasR* (n = 21) and lasRWT (n = 28) isolates under planktonic growth conditions, we also recorded their transcriptomes under biofilm growth conditions14. While the planktonic gene expression profiles of the two groups of clinical isolates were clearly separated in a multi-dimensional scaling plot (MDS), their transcriptomes did not cluster separately when the isolates were grown under biofilm growth conditions (Fig. 4).

Fig. 4: Multidimensional scaling plots based on the transcriptomes of 21 clinical lasR* and 28 lasRWT P. aeruginosa isolates under planktonic and biofilm growth conditions.
figure 4

For transcriptional profiling, bacteria were grown in rich LB media until an OD600 of 2 at 37 °C and 180 rpm for planktonic growth (a), or statically in biofilms for 48 h (b). Reads were normalized for the individual library size and genes with coverage below 1 count per million were excluded from the analysis. Ellipses display the 95% confidence intervals.

Full size image

In accordance, only 53 genes (25 upregulated and 28 downregulated, Supplementary Data 4) showed a differential expression in the lasR* isolates (compared to lasRWT) under biofilm growth conditions (as compared to 722 genes under planktonic conditions). We also recorded the transcriptional profiles of PA14 WT and PA14 ∆lasR under biofilm growth conditions (Supplementary Data 5). Again, we detected a reduction (by approx. 20%) in the overall number of differentially regulated genes between PA14 WT and PA14 ∆lasR under biofilm growth conditions (780 and 978 genes were differentially regulated under biofilm and planktonic growth conditions, respectively, Supplementary Fig. 3).

Our finding that the differential gene expression in clinical isolates harboring a non-functional lasR gene, can be alleviated under biofilm growth conditions indicates that the activity of LasR becomes less important under biofilm growth conditions. We also found that genes of the pyocyanin biosynthesis gene cluster (phzG, phzF) were expressed at even higher levels in the lasR* isolates under biofilm growth conditions (Supplementary Data 4). This demonstrates that despite the finding that they belong to the core lasR regulon under planktonic growth conditions, they are not governed by LasR under biofilm growth conditions. Instead, there seem to be alternative ways to induce the phenazine biosynthesis genes under biofilm growth conditions that are even more efficient. Our finding is in accordance with previous reports on a strong pyocyanin production of late stationary phase-grown lasR mutants28.

Virulence factor expression becomes independent of lasR but not of rhlR under biofilm growth conditions

Previous work has shown that RhlR is able to regulate the expression of lasR-dependent genes in late stationary growth conditions so that LasR becomes dispensable29. Here, we aimed at evaluating whether also under biofilm growth conditions, LasR becomes dispensable due to the activation of RhlR in a LasR independent manner. Therefore, we searched our clinical isolates for variants harboring inactivating mutations in both lasR as well as rhlR. We identified 22 isolates with no mutation in the two genes (WT), 15 isolates with an inactivating mutation (InDel/Stop) exclusively in lasR (lasR*), and 7 isolates with InDel/stop mutations in rhlR and an InDel/stop and/or non-synonymous mutations in lasR (lasR**/rhlR*). For all of the isolates both, planktonic and biofilm transcriptomes, were available. As expected, the isolates harboring a rhlR inactivating mutation exhibited a significantly reduced expression of rhlI, the autoinducer synthase, which is under the direct control of RhlR (Supplementary Fig. 4). We then analyzed the expression of genes that clearly belonged to the LasR core regulon (138 genes that were commonly positively regulated by LasR under planktonic conditions; Table 1) under biofilm conditions (Fig. 5). We found that the expression of rhlAB, phzA-G, and hcnABC was independent of a functional lasR gene under biofilm growth conditions. In the lasR* isolates, expression levels even exceeded those in the lasRWT isolates. However, their expression was clearly dependent on a functional rhlR gene under biofilm growth conditions as the expression levels of these genes in the lasR**/rhlR* double mutants did not reach those of the wild-type. In accordance, the production of elastase was abolished in the majority of the lasR* and lasR**/rhlR* isolates when grown planktonically (shaking incubation for 24 h), whereas elastase production under biofilm conditions (static growth for 48 h) was at a similar level in lasR* but not in the lasR**/rhlR* strains as compared to lasR wild-type isolates (Supplementary Fig. 2c, d).

Fig. 5: Expression of lasR regulon genes in clinical lasRWT, lasR* and lasR**/rhlR* isolates under biofilm growth conditions.
figure 5

Expression values (log10(nrpg + 1)) of a rhlAB, b phzABCDEFG, c hcnABC, d aprIAFEDX, e pqsABCDE in lasRWT (n = 22), lasR* (n = 15), and lasR**/rhlR* (n = 7) clinical isolates under biofilm growth conditions are depicted. Asterisk: p-value < 0.05, Wilcoxon rank sum test or Welch’s t-test. Boxplot elements are: center line—median; box limits—upper and lower quartiles; whiskers—1.5× interquartile range; points—gene expression values (log10(nrpg + 1)). Reads are shown as log10 normalized million reads per gene.

Full size image

Our results thus demonstrate that RhlR acts downstream of LasR to impact the expression of many genes and that RhlR becomes activated under biofilm growth conditions independent of LasR to induce expression of genes that belong to the LasR core regulon under planktonic conditions. Supplementary Fig. 5 depicts the comparison of the expression of all LasR core regulon genes (n = 138) among the lasRWT, lasR* and lasR**/rhlR* isolates under planktonic and biofilm growth conditions. The data confirm that RhlR becomes activated under biofilm growth conditions independent of LasR to induce expression of the LasR-regulon genes.

Of note, the expression of the alkaline protease biosynthesis genes (aprIAFEDX) was dependent on lasR even under biofilm conditions and the additional inactivation of rhlR did not reduce gene expression further (Fig. 5d). In accordance, both, lasR* and lasR**/rhlR* mutant isolates showed in skim milk agar plate-grown colony biofilms a significantly reduced proteolytic activity as compared to lasRWT strains (Supplementary Fig. 2b). Furthermore, the expression of the Pseudomonas Quinolone Signal (PQS) biosynthetic gene cluster (pqsABCDE) seemed to be independent of both lasR and rhlR under biofilm growth conditions (Fig. 5e).

We also analyzed the genes that were expressed at higher levels in the absence of lasR under planktonic growth conditions (Fig. 6). Again, we found that upregulation of genes in the lasR mutant under planktonic growth conditions (Fig. 6a–c) seemed to be dependent on a functional rhlR gene as the cupA2-5, phnCDEHIJK and pvdGEFN gene expression levels were back to the wild-type level in the double lasR**/rhlR* mutant isolates. Under biofilm growth conditions (Fig. 6d–f) these genes were not upregulated in the lasR-deficient isolates. Their expression level was even below that of the wild-type under biofilm conditions in the lasR-deficient isolates but back to normal in the lasR**/rhlR* clinical isolates. Again, as depicted in Supplementary Fig. 6, the comparison of the expression of all (n = 412) genes, which have been identified to be upregulated in the lasR-deficient isolates, among the lasRWT, lasR* and lasR**/rhlR* isolates confirm that RhlR acts downstream of LasR to induce the expression of genes in the absence of lasR under planktonic conditions. However, this lasR-mediated inhibition of gene expression is lost under biofilm conditions.

Fig. 6: Expression of genes negatively affected by the presence of lasR in clinical lasRWT, lasR*, and lasR**/rhlR* isolates.
figure 6

Expression values are shown for planktonic (a–c) and biofilm growth conditions (d–f). Expression values (log10(nrpg + 1)) of (a, d) cupA2-5; (b, e) phnCDEHIJK; (c, f) pvdGEFN in lasRWT (n = 22), lasR* (n = 15), and lasR**/rhlR* (n = 7) clinical isolates are depicted. Asterisk: p-value < 0.05, Wilcoxon rank sum test. Boxplot elements are: center line—median; box limits—upper and lower quartiles; whiskers—1.5× interquartile range; points—gene expression values (log10(nrpg + 1)). Reads are shown as log10 normalized million reads per gene.

Full size image

rhlR and phoB dependent PA14 gene expression profiles under planktonic and biofilm growth conditions

We next evaluated whether rhlR acts downstream of lasR also in the PA14 type strain to induce the expression of genes in the absence of lasR under biofilm growth conditions. We, therefore, generated clean ∆rhlR and ∆lasR/∆rhlR PA14 deletion mutants using a CRISPR/Cas9-based approach and recorded their transcriptional profiles under planktonic and biofilm growth conditions. Again, we found that the expression of the lasR core regulon genes was dependent on the presence of lasR under planktonic, but not under biofilm conditions, while the lack of rhlR reduced the expression of the lasR regulon genes under both, planktonic and biofilm conditions (Fig. 7a). Since it has recently been demonstrated that LasR becomes dispensable under low phosphate growth conditions30, we also generated a deletion in the gene encoding the phosphate starvation response regulator PhoB as well as a ∆lasR/∆phoB mutant. However, the lack of phoB did not have any influence on the expression of the PA14 lasR core regulon genes (Fig. 7).

Fig. 7: Expression of genes that are affected by the presence of lasR in the clinical lasR* isolates in PA14 and its isogenic ∆lasR, ∆rhlR, and ∆phoB single and double mutants.
figure 7

a Gene expression values (log10(nrpg + 1)) of the lasR core regulon genes (n = 136) and b of the lasR repressed (n = 412) genes in the ∆lasR, ∆rhlR, ∆phoB, ∆phoB/∆lasR, and ∆lasR/∆rhlR deletion mutants under planktonic (left) and biofilm (right) growth conditions. Asterisk: p-value < 0.05, Wilcoxon rank-sum test. Boxplot elements are: center line—median; box limits—upper and lower quartiles; whiskers—1.5x interquartile range; points—gene expression values (log10(nrpg + 1)).

Full size image

We also had a closer look at genes (n = 412) that were expressed at a higher level in the clinical lasR* vs. lasRWT isolates. Expression of those genes was demonstrated to be less consistent across the clinical isolates (Fig. 3d). This indicates that the lack of lasR might not directly control their expression but rather that the clinical lasR-defective isolates tend to produce a certain, possibly compensatory, gene expression profile that can be repeatedly found in the clinical lasR mutants. If this is true, one would not necessarily expect to see this profile of gene expression in the clean PA14 lasR deletion mutant. Indeed, expression of the 412 genes that were upregulated in the clinical lasR* was not differentially regulated in either of the ∆lasR, ∆rhlR, and ∆phoB single and double mutants under planktonic or biofilm growth conditions (Fig. 7b).

Discussion

In this study, we took advantage of the extensive transcriptome as well as whole-genome sequencing data on a large collection (>400) of clinical P. aeruginosa isolates14,16. Those strains have been isolated from different infected body sites and they were recovered from both, acute and chronic infections. In order to enable the successful growth of the highly diverse clinical isolates, we recorded the transcriptomes following growth under rich media (LB) conditions. Although growth e.g. in a synthetic sputum medium might have reflected the conditions within the human host more accurately, the non-selective cultivation conditions accommodated the growth of all clinical isolates. Despite the fact that the transcriptional profiles were recorded under rich medium growth conditions, the recorded transcriptional patterns are nevertheless the result of genomic variations that have evolved during the infection process in the human host.

The extensive data allowed us to cluster the clinical isolates according to a high vs. low expression status of a large subset of expressed genes. Although we did not account for absolute differences in gene expression levels, we were able to group isolates according to the complex high/low gene expression profiles that developed repeatedly in different P. aeruginosa strains. Our analysis identified mutations within lasR as major contributors to evolved gene expression patterns and therefore as main drivers of adaption towards the habitat of the human host. Indeed, inactivating mutations within lasR are well-known pathoadaptive mutations, frequently found in P. aeruginosa isolates from the respiratory tract of CF patients23,31,32,33. LasR-deficient isolates have been associated with advanced disease and major changes in the expression of QS-regulated genes9,11,34. However, especially the apparent reduced ability to produce virulence factors raises the question of the fitness advantage of such an evolutionary path26,35,36.

The analysis of engineered lasR deletion mutants in P. aeruginosa type strains has previously uncovered the LasR regulon27,37. This regulon comprises many virulence-associated genes, which are positively regulated by LasR activity. In this study, we identified genes that are commonly found to be differentially regulated in lasR-deficient (lasR*) versus lasR-proficient (lasRWT) clinical isolates. This core regulon included many genes, which were also found to be LasR-dependent in the PA14 type strain. However, we also found additional pathoadaptive traits in the lasR* isolates. Recording of transcriptional profiles of clinical lasR* isolates informed on important co-selected or possibly compensatory adaptations that might explain the success of the lasR mutations in the clinical context. Most remarkably, we found that the gene expression profile of the lasR* isolates did not differ to the same extent to those of lasRWT clinical isolates when cultured under biofilm growth conditions. Indeed, it seems that the lasR* isolates exhibit a transcriptional profile typical for growth in a biofilm, already under planktonic culture conditions. The lasR* isolates were found to regulate 1.52-fold less genes compared to lasRWT isolates (25% and 38% of the whole genome respectively, Supplementary Data 7) when switching to the biofilm growth mode. Genes encoding for the cup fimbriae38 and the exopolysaccharide Pel39 were already expressed under planktonic growth conditions in the lasR* isolates. Furthermore, despite the fact that the transcriptional profiles have been recorded under rich medium conditions, genes important for metabolizing complex carbon sources such as phosphatidylcholine or amino acids and for the assimilation of phosphate as well as iron uptake were highly expressed in the lasR* isolates. Amino acids and polyamines are major carbon sources in the CF lung40,41, while phosphate and iron are important for the establishment of an infection42,43,44.

In a process of genetic assimilation45, induced traits can lose their environmental sensitivity and thereby become robust to the environment (i.e. they no longer require the environmental signal for their expression). It seems that the switch to the biofilm mode of growth, which is expected to be triggered by the environmental conditions in the human host, might have become fixed in the lasR* population45,46.

The results of our study also showed that the genes of the lasR regulon do not depend on the presence of lasR, but on the presence of rhlR under biofilm growth conditions. An exception was the alkaline protease gene cluster (apr) and the PQS synthesis operon, whose expression remained dependent on the presence of a functional lasR even under biofilm growth conditions. It has been described previously that LasR becomes dispensable under conditions where RhlR is activated by different means, in order to drive genes belonging to the LasR regulon. A decoupling of the QS networks has been described under diverse environmental conditions such as low phosphate30, late stationary phase28 and now biofilm growth conditions. Interestingly, the phosphate-starvation regulator PhoB was previously shown to interact either directly or indirectly via RhlR and PqsR on the expression of the otherwise LasR-dependent pyocyanin synthesis gene clusters47. However, our results indicate that the phosphate starvation response regulator PhoB does not play a role in the activation of lasR/rhlR dependent genes under biofilm growth conditions, despite the fact that many genes that are activated in the clinical lasR mutants are involved in the acquisition of phosphate.

In conclusion, we demonstrate in this study that the analysis of large sample sizes of clinical isolates provides an interesting alternative to in vitro long-term experimental evolution experiments. Our data illustrate the power of profiling large sample sizes of clinical isolates to discover novel evolutionary trajectories in pathogen populations on their way of adaptation to the conditions encountered during an infection process within the human host. An integrated analysis of large-scale genomic and transcriptomic data together with clinical metadata, including detailed knowledge about the isolate-specific history (e.g. duration of infection and infection site), would further extend our knowledge and contribute to a more comprehensive understanding of adaptive evolution during infection. Profiling the consequences of pathoadaptive mutations promises to identify predictable pathways to successful long-term persistence within a host and thus highlights opportunities for developing novel strategies to combat therapy-refractory chronic infections.

Methods

Bacterial strains and growth conditions

Bacteria were grown in standard rich medium culture conditions (LB) with constant shaking (180 rpm) at 37 °C unless otherwise stated. All strains and plasmids used in this study are listed in Table 3. If necessary, LB medium was supplemented with gentamicin (Gm; 15 µg/ml for E. coli or 50 µg/ml for P. aeruginosa), or streptomycin (Sm; 50 µg/ml for E. coli or 500 µg/ml for P. aeruginosa) to retain plasmids.

Table 3 Strains and plasmids used in this study.
Full size table

Generation of scarless deletion mutants using a novel CRISPR/Cas9-recombineering method

We developed a targeted CRISPR/Cas9-recombineering system in order to generate clean deletion mutants in P. aeruginosa. For this endeavor, we resorted to the pS448·CsR plasmid developed for the accelerated genome engineering in the closely related bacterium species P. putida48. This vector controls the constitutive expression of a target-specific small guide RNA (sgRNA) and the inducible production of the Cas9 endonuclease upon addition of 3-methylbenzoate (3-mBz). In Pseudomonas spp., it has been previously shown that using synthetic, linear DNA (either ssDNA or dsDNA) as a mutagenic template for genomic manipulations is futile unless an efficient DNA recombinase is also expressed49,50. Thus, we constructed a second vector expressing the gene encoding for the Ssr recombinase from P. putida DOT‐T1E51. The ssr coding sequence was excised from pSEVA658-ssr as an AvrII/BamHI restriction fragment and cloned anew into the same restriction sites of the isopropyl β-D-1-thiogalactopyranoside (IPTG)-inducible vector pSEVA624, resulting in plasmid pSH624-ssr.

For cloning each gene-specific spacer into pS448·CsR, we followed the indications by Wirth and collaborators48. Briefly, 100 pmol of each primer pair constituting the gene-specific spacer sequence were mixed and phosphorylated for 30 min at 37 °C in a 50-μl mixture with the T4 Polynucleotide Kinase (PNK, New England Biolabs). Next, the primers were denatured for 5 min at 95 °C and annealed to each other by allowing the mixture to slowly cool down to room temperature for at least 2 h. Both complementary oligonucleotides were designed to generate the protruding ends of the Eco31I restriction site (BsaI isoschizomer) when hybridizing (see Supplementary Table 1 for the details). Hence, it was possible to clone the dsDNA spacer fragment into the Eco31I restricted vector pS448·CsR (Table 3) to generate each gene-specific sgRNA-expressing vector. All oligonucleotides used for spacer construction, PCR amplification, as well as the 100 bp recombineering oligos for introducing the deletions are listed in Supplementary Table 1.

To achieve the deletions, P. aeruginosa PA14 or its ∆lasR derivative were previously electroporated with the recombineering plasmid pSH624-ssr. The ssr-carrying strains were grown overnight at 37 °C in 10 ml of LB medium supplemented with Gm. Next, 5 ml of fresh LB medium containing 50 µg/ml Gm and 1 mM IPTG were added to the cultures for inducing the ssr gene. After 3 h incubation at 37 °C, electrocompetent cells were prepared from 6 ml suspension of the induced cells as indicated by Choi and colleagues52, with the only modification that cells were finally resuspended in 200 μl of 0.3 M sucrose. Aliquots of 100 μl of cells were electroporated with 250 ng of the pS448•CsR-derivative harboring the appropriate spacer and 100 pmol of the mutagenic, recombineering oligonucleotide specific for each gene deletion (Supplementary Table 1). Cells were recovered for 2 hours in 2 ml of LB medium containing 2 mM 3-mBz for the induction of the cas9 gene. After incubation, P. aeruginosa cell dilutions were plated onto LB-Agar containing Gm, Sm and 2 mM 3-mBz to select for plasmids and to counterselect the WT cells by means of CRISPR/Cas9 specific targeting of each spacer region. After 24 h of incubation at 37 °C, successfully generated deletion mutants were identified by colony PCR using primers flanking each gene (Supplementary Table 1). The efficiency of mutation was ~70% of the total colonies assayed. Altogether, this method allowed us to engineer P. aeruginosa mutants with clean deletions of rhlR and phoB within only 5 working days. Finally, the constructed mutant strains were cured of the plasmids by three consecutive passages in LB devoid of antibiotics and selected by sensitivity to the antibiotics and by PCR using primers specific for ssr and cas9 genes (Supplementary Table 1).

Transcriptional profiling of planktonic and biofilm-grown bacteria

We re-analyzed the transcriptional profiles of previously characterized clinical P. aeruginosa isolates collected across Europe16,20. Transcriptional profiles were recorded in LB, a universal, rich medium that allows the growth of a large variety of adapted clinical isolates. In the course of this study, we complemented the available dataset by recording transcriptomes of PA14 WT, as well as its deletion mutants (see Table 3) under identical planktonic16 and biofilm14 growth conditions. In brief, for planktonic growth 10 mL LB medium was inoculated with a starting optical density (OD600) of 0.05, and bacteria were cultivated at 37 °C and constant shaking (180 rpm) until they reached an OD600 of 2.02–2.28 (early stationary phase). For the harvest, 1 ml of bacterial suspension was mixed with an equal volume of RNAprotect (Qiagen), incubated for 10 min, and centrifuged at 8000 rpm for 10 min.

A randomly selected sub-group (n = 77) of our strain collection was chosen to be analyzed under biofilm growth conditions14. Biofilms were inoculated at a starting OD600 of 0.002 in LB and grown statically in half-area 96-well µClear microtiter plates (Greiner Bio-One) at 37 °C in a humid atmosphere. After 48 h, mature biofilms (10 wells per replicate) were resuspended and harvested in an equal volume of RNAlater (Qiagen). Three independent biological replicates were used for each strain and condition.

Total RNA from cell pellets was extracted using the RNeasy MiniKit (Qiagen) following an initial QIAshredder step according to the manufacturer’s instructions. DNA was removed by DNase I (Ambion) treatment, and the RNase inhibitor RNAsin (Promega) was added to protect the eluted RNA.

For cDNA library preparation53,54, total RNA was fragmented (150–350 bp) in Fast AP-Buffer (Thermo Scientific), DNA was digested by TURBO™ DNase (Invitrogen), and RNA fragments were phosphorylated by FastAP alkaline phosphatase (Thermo Scientific). Custom-made barcodes were ligated to the RNA using the T4 ligase (New England Biolabs) and fragments were subsequently purified and concentrated using the RNA Clean & Concentrator™-25 Kit (Zymo Research) following the manufacturer’s instructions. Ribosomal RNA was removed by using the RiboZero Bacteria kit (Illumina). cDNA libraries were synthesized using the SMARTScribe Reverse Transcriptase (Takara) followed by a PCR enrichment using the AccuPrime HiFi Taq polymerase (Invitrogen). Enzymatic reactions were carried out in the presence of SUPERase·In™ RNase Inhibitor (Invitrogen); RNACleanXP beads (Agencourt) were used for all RNA purification steps. Quality checks were performed before, during, and after cDNA library preparation with the RNA Nano Kit and an Agilent Bioanalyzer 2100 (Agilent Technologies). Libraries were sequenced on an Illumina HiSeq (single-end mode; 1 × 50 bp) or an Illumina NovaSeq 6000 (paired-end mode; 2 × 50 bp).

Transcriptome analysis

Data analysis was performed in the R statistical environment (version 3.6.3)55 Reads were mapped against the UCBPP-PA14 reference genome (NCBI Reference sequence: NC_008463.1) using stampy (version 1.0.23)56. Normalization was performed using the command calcNormFactors() of the R-Package edgeR (version 3.28.1)57 and normalized read counts were extracted using the cpm() command of the same package. Genes with less than 1 count per million (cpm) in at least 21 isolates were excluded from the analysis. p-value correction outside of the edgeR-based differential gene expression analyses was calculated by p.adjust() of the stats R-package (version 3.6.1) with the adjustment method ‘fdr’.

Multimodality of gene expression was assessed by testing the normalized expression values of all genes of the tested 414 clinical P. aeruginosa isolates using the command modetest() of the R-package multimode (version 1.4)58 in default settings. Multimodality was assigned when the calculated and corrected p-value was <0.1. Isolates were sorted into either high or low expressing group by determining the last minimum before the last clear maximum of the distribution of expression values (see Supplementary Fig. 7). The expression of each gene that met the definition of a multimodal gene distribution was evaluated in each isolate and isolates were sorted into one of two groups accordingly: high or low expression of the respective gene.

Differential gene expression analysis was performed using the R-Package edgeR using the commands glmQLFTest() and topTags() with a corrected p-value of 0.05 as a cut-off for a significant differential gene expression and a threshold of log2FC ≥ 1 for upregulation and log2FC ≤ −1 for downregulation.

KEGG24 pathway enrichment was performed using the KEGGREST (version 1.26.1)59 package in R with the “pau” pathways to map the differentially regulated genes. Enrichment was performed via the p.hyper() function of the stats package with a corrected p-value of 0.05 as a cut-off for a significant enrichment.

The regulon robustness, as well as the core regulon analysis, were performed by calculating a differential gene expression of each individual lasR* isolates against all lasRWT isolates (n = 28). The number of isolates with log2FC ≥ 1 or log2FC ≤ −1 for each gene was counted and summed over the number of tested lasR* isolates (n = 21). Finally, the lasR core regulon was determined as the intersection of genes that showed a log2FC above or below the respective threshold in at least 90% of the isolates (n = 19), and genes that are significantly regulated in the clean PA14 ∆lasR compared to PA14 WT.

MDS plots were calculated based on the normalized read counts created in the differential gene expression analysis described above and the PlotMDS() function in the edgeR package in R, taking all analyzed genes into account for the calculation of the dissimilarity matrix. The data was visualized using the ggplot2 R-package with the command stat_ellipse() for the 95% confidence interval ellipses.

Comparisons of gene cluster expression were generally performed by combining the complete transcriptional profiles of isolates of interest into a DGEList object as described above to subsequently perform the described normalization and low coverage gene exclusion. Expression values of the genes of interest were then extracted as library size-normalized counts per million (cpm() command in R; values described as “nrpg”) and a value of 1 was added to the all nrpg values prior to log10 transformation in order to prevent the creation of infinite values.

DNA sequencing and SNP calling

In order to identify mutations in the sequences of lasR (PA14_45960) and rhlR (PA14_19120), previously published whole-genome sequencing data of 414 clinical isolates16,17 were screened for sequence variations in the respective genes. Mapping was accomplished using stampy (version 1.0.23)56 and variant calling was performed using SAMtools (version 0.1.19)60 with UCBPP-PA14 (NCBI Reference sequence: NC_008463.1) or PAO1 (NCBI Assembly: GCA_000006765.1 ASM676v1) as a reference. The strain background of the clinical isolates was assessed based on the phylogenetic analysis documented in Khaledi et al.16. PA7-like isolates were excluded in this study.

Phenotypic characterization of clinical isolates

Phenotypic data for randomly selected clinical isolates representing a lasR wild-type allele status (lasRWT; n = 10), strains harboring a non-functional lasR (lasR*; n = 8) and isolates with inactivating mutations in both lasR and rhlR (lasR**/rhlR*; n = 4) was collected in previous studies14,17. In vivo virulence was determined 48 h post infection as % killing of Galleria mellonella larvae. For infection, bacterial strains were grown overnight and serially diluted in PBS. 20 µl of the bacterial suspension was injected into the last left proleg of the larvae with an MOI (multiplicity of infection) of 100 cfu (10 larvae per bacterial strain). Relative proteolytic activity was assessed by measuring clearing zones of spotted bacteria on cation-adjusted Müller-Hinton (MH) broth containing 10% (v/v) milk after 24 h of incubation. Overnight cultures were adjusted to an OD600 of 0.025 and 5 μl of the bacterial suspensions were spotted on top of the agar plates. An Elastin Congo Red (ECR) assay was applied to determine the elastolytic activity of secreted proteases. Bacteria were grown for 24 h in planktonic cultures (shaking at 180 rpm), or for 48 h in biofilms (static incubation) as described above. After harvest, 100 µl of bacteria-free culture broth was incubated with the substrate (900 μl ECR buffer containing 100 mM Tris [pH 7.5], 1 mM CaCl2, supplemented with 22.5 mg/ml ECR [Sigma-Aldrich]) for 3 h at 37 °C and 900 rpm. Elastase secretion was determined by measuring the absorbance of the supernatant at OD = 495 nm.

Statistical analysis

The shapiro.test() function of the stats package in R (v 3.6.1) was applied to test for normal distribution of the compared groups. If no normal distribution was observed (p-value < 0.05) in one of the groups, the wilcox.test() command with default settings from the same package was employed to test for statistical significance.

Overrepresentation analysis was calculated using the phyper() command of the stats package in R with the lower.tail option set to “FALSE” to test for enrichment.

Statistical analyses for phenotypic data were performed in GraphPad Prism (v 8.3.0) by using Kruskal-Wallis (comparison of three groups) and Mann–Whitney tests (pair-wise comparison).

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.