Keywords

The ecological success of the opportunistic pathogen Pseudomonas aeruginosa is based on its remarkable capability to adapt and survive in a plethora of diverse and challenging habitats (Goldberg 2000; Silby et al. 2011). The selection of beneficial mutations in the bacterial genome is driving long-term adaptation through the process of evolution and can also dramatically alter the genetic diversity of populations (Folkesson et al. 2012; Kordes et al. 2019b; Rossi et al. 2021). In addition to genetic diversity, phenotypic plasticity, which increases the fitness of the individual bacterium, allows bacterial adaptation to a changing environment. Phenotypic plasticity is defined as the ability of a single genotype to exhibit a range of different phenotypes in response to variation in the environment and is particularly suited to enact short-lived, energetically expensive, phenotype switching (Forsman 2015; Fusco and Minelli 2010; Kelly et al. 2012; Price et al. 2003; Schlichting and Pigliucci 1998). If however, the plastic response is close to the optimal phenotype, then achieving the optimal phenotype via phenotypic plasticity might be followed by the acquisition of mutations to fix the phenotype, in a process called genetic accommodation (Pigliucci et al. 2006; Price et al. 2003). Phenotypic plasticity therefore also promotes innovation and diversification, because phenotypic traits that are activated in a certain environment might eventually become fixed.

It has been demonstrated that during an infection, transcriptional regulators are hotspots for mutations. This can lead to a rewiring of the pathogen’s regulatory network and induce downstream transcriptional consequences that are beneficial in the prevailing host habitat (Marvig et al. 2015). Prominent examples are patho-adaptive mutations in lasR (encoding for a major quorum-sensing regulator that participates in the control of the production of virulence factors), mucA (causing a mucoid phenotype that protects against antibiotic treatment and immune defenses), and mexZ (leading to enhanced antibiotic resistance due to overexpression of efflux pumps) (Frimodt-Møller et al. 2018; Marvig et al. 2015; Winstanley et al. 2016; Yang et al. 2011).

1 P. aeruginosa, a Versatile Environmental Bacterium and Model Organism

P. aeruginosa is a prominent example of an extremely adaptable organism and opportunistic pathogen that serves as a well-studied model organism. Its remarkable metabolic versatility is reflected in its large genome of approx. 5.5–7 Mbp (Klockgether et al. 2011), consisting of a highly conserved core genome (Muthukumarasamy et al. 2020) and a diverse accessory genome (derived from various donor species) that varies greatly between individual strains (Pohl et al. 2014). The P. aeruginosa genome encodes an exceptionally large number of transcriptional regulators that build a complex regulatory network, thus providing the tools for flexible changes in bacterial behavior (Cases et al. 2003; Lee et al. 2006; Stover et al. 2000).

Key for bacterial survival under unfavorable conditions, are sigma factors that allow directional adaptation via the modulation of RNA polymerase promotor recognition and thus the transcriptional modulation of distinct sets of genes. Alternative sigma factors, in particular, are involved in the regulation of a large number of genes, the expression of which represents robust and effective responses to diverse external stressors. Regulatory networks of alternative sigma factors are organized in highly modular architectures, with function-specific crosstalk occurring between different sigma factor regulons, to orchestrate complex cellular processes such as motility and virulence (Binder et al. 2016; Schulz et al. 2015). In addition, two-component regulatory systems are involved in sensing and responding to environmental stimuli. They are typically composed of a membrane-bound histidine kinase as a receptor, which detects a specific external trigger and relays the signal via a phosphorylation cascade, and a corresponding response regulator, which receives the signal and in turn mediates a fine-tuned cellular response, e.g., by inducing expression changes in target genes (Krell et al. 2010). Two-component signaling plays a key role in both, biofilm formation (reviewed in Mikkelsen et al. 2011) as well as the expression of virulence-associated traits (reviewed in Balasubramanian et al. 2013). Similar to alternative sigma factors, multiple two-component signaling systems often regulate virulence phenotypes (Wang et al. 2021), highlighting again the complexity of the processes involved in the sensing of environmental signals and their translation into the production of appropriate phenotypes. Other multifaceted regulators that respond to a variety of (often unknown) environmental signals use the messenger bis-(3′-5′)-cyclic dimeric guanosine monophosphate (c-di-GMP) for intracellular signaling. c-di-GMP signaling plays a key role in the regulatory mechanisms that accompany the transition from a planktonic to a sessile biofilm state of growth. Thereby, high intracellular c-di-GMP levels induce biofilm traits, while reducing bacterial motility and virulence traits (Furukawa et al. 2006; Hengge 2009; Jenal and Malone 2006; Römling and Amikam 2006; Valentini and Filloux 2016).

Signal transduction occurs not only at the cellular level but also at the community level: Quorum sensing (QS) is a cell density-dependent regulatory mechanism that involves intercellular communication to coordinate and synchronize gene expression within a population (Papenfort and Bassler 2016). In P. aeruginosa, three interconnected, hierarchically organized QS-systems are known: the Las- and Rhl-systems (both depend on acylhomoserine lactones (AHLs) as autoinducers) (Latifi et al. 1995; Winson et al. 1995), as well as the Pseudomonas quinolone signal system (dependent on autoinducers of the 4-quinolone family) (Pesci et al. 1999). In total, up to 11% of all genes in P. aeruginosa are regulated by QS (Schuster and Peter Greenberg 2006; Whiteley et al. 1999), highlighting the importance of QS for a coordinated community behavior and phenotypic adaption at the population level (Williams and Cámara 2009). All in all, abundant, complex, and multi-layered regulatory mechanisms provide the basis for the remarkable adaptability of P. aeruginosa to diverse environmental conditions.

2 P. aeruginosa: An Opportunistic Pathogen of High Clinical Relevance

High adaptability makes P. aeruginosa one of the most successful opportunistic pathogens and explains its dominant role as a pathogen of acute, as well as chronic infections across a wide variety of infection sites in the human host (Bodey et al. 1983). To establish and advance an acute infection, P. aeruginosa must quickly adapt to the host environment. This short-term adaptation typically requires a sophisticated and fine-tuned interplay of regulators that ensure survival and drive pathogenicity, e.g., via the facilitation of bacterial spread in damaged host tissues via the expression of genes encoding toxins and proteases. If not recognized and eliminated by the host immune system or eradicated by antibiotic treatment during early colonization, P. aeruginosa populations can cause chronic infections that exist for long enough to accumulate patho-adaptive mutations in the genome. Long-term adapted strains isolated from chronically infected CF lungs were shown to exhibit a large phenotypic variability (Clark et al. 2015; Workentine et al. 2013). Nevertheless, P. aeruginosa is commonly observed to become less invasive (e.g., via loss of the flagellum) and more persistent (Mahenthiralingam et al. 1994; Smith et al. 2006). Chronic persistent infections are characterized by the presence of aggregated, often surface-associated multicellular bacterial biofilm communities (Bjarnsholt 2013; Costerton 1999; James et al. 2008). Consequently, evolved phenotypic variants representing subpopulations with enhanced biofilm-forming capacities are repeatedly isolated from chronically infected CF lungs, such as alginate-overproducing mucoid strains (Boucher et al. 1997; Govan et al. 1992; Martin et al. 1993) or highly adherent small colony variants (SCV) (Häussler 2004; Häussler et al. 1999, 2003; Malone 2015). The biofilm lifestyle is considered as an adaptation to hostile environments and provides bacteria with many competitive advantages as compared to individual, free-floating planktonic cells. Biofilm-associated bacteria exhibit increased resistance toward various environmental stressors, disinfectants, antibiotic treatment, and the host immune response (Bridier et al. 2011; Ciofu and Tolker-Nielsen 2019; Hall-Stoodley et al. 2004; Hall and Mah 2017; Lebeaux et al. 2014). Thereby, biofilm-induced tolerance greatly varies among different clinical isolates and depending on the antibiotic was found to be increased up to 16,000-fold as compared to planktonic cells (Thöming and Häussler 2022). Chronic, biofilm-associated infections are furthermore characterized by high bacterial loads and increasing deterioration of lung function of, e.g., cystic fibrosis patients, thus significantly contributing to morbidity and mortality in these patients (Emerson et al. 2002; Govan and Deretic 1996; Mayer-Hamblett et al. 2014; Nixon et al. 2001; Wagner and Iglewski 2008). According to the Center for Disease Control (CDC) and the National Institutes of Health (NIH), biofilms are involved in 65–80% of all bacterial infections (Joo and Otto 2012) and therefore represent a huge financial burden for health care systems.

3 Old and New Antibacterial Strategies

The opportunistic pathogen P. aeruginosa can cause a multitude of difficult-to-treat infections. Successful treatment is complicated by (1) the frequent acquisition of multi-drug resistant phenotypes and (2) very limited treatment options in chronic biofilm-associated infections. To meet clinical demands, there is a need to expand the search for and the development of novel antimicrobial agents, as well as to invest in advanced diagnostics for earlier and more targeted antibacterial treatment and the implementation of effective infection control measures.

Conventional antimicrobial susceptibility testing (AST) is based on culture-dependent methods and is thus not only labor-intensive but also very time-consuming. The early implementation of more effective antibiotic treatments is crucial, particularly in critically ill patients. Especially in the light of increasing (multi-drug) resistance (WHO 2021), there is a need for earlier and reliable testing results on antimicrobial resistance profiles of infecting pathogens. Approaches to develop rapid culture-independent molecular diagnostics are promising to meet this need in future. The second major problem is diagnosing bacterial recalcitrance in chronic infections, which are mainly caused by biofilm-associated bacteria. Resistance profiles for these infections cannot be deduced from minimal inhibitory concentrations (MIC) of planktonic bacteria, as determined by conventional AST, and several studies suggest the implementation of standardized biofilm-specific diagnostics, in order to improve patient outcomes (Keays et al. 2009; Moskowitz et al. 2004; Müsken et al. 2017; Stewart 2015).

The lack of new antibacterial drugs to combat severe P. aeruginosa infections has been highlighted by the World Health Organization (WHO), who classified multidrug-resistant P. aeruginosa as a critical threat, with the highest priority for research and development of new antibiotics (WHO 2017). In addition, there is a need to overcome the complete lack of biofilm-specific antimicrobials on the market. Researchers worldwide are working on the discovery of new antimicrobial compounds, as well as substances with an anti-biofilm potential. The current state of development for novel therapeutic options to combat P. aeruginosa infections is reviewed in (Yaeger et al. 2021). Drug development involves, to a large extent, classical approaches for drug discovery, such as the application of high-throughput screenings of large compound libraries (including up to 106 small molecules), followed by multiple confirmation and validation steps of potential hits. Multiple screens have discovered potential novel lead compounds that showed antimicrobial activity against multi-drug resistant P. aeruginosa (Li et al. 2020; Luther et al. 2019) or anti-biofilm activity (Andersen et al. 2021; Li et al. 2020; Sambanthamoorthy et al. 2012). However, classical drug development is a very time-consuming and costly process and the overall probability of success is low. To bypass this obstacle, an alternative strategy makes use of existing drugs: approved substances are screened in order to identify novel uses outside their original scope, a strategy called drug repurposing or drug repositioning (reviewed in Ashburn and Thor 2004; Pushpakom et al. 2019). For example, QS-inhibitors with virulence-attenuating potential (Baldelli et al. 2020; D’Angelo et al. 2018; Ho Sui et al. 2012; Imperi et al. 2013a, 2013b) and a c-di-GMP inhibitor with anti-biofilm activity (Lieberman et al. 2014) have been identified in libraries of FDA-approved drugs. Another promising example is promethazine, an FDA-approved drug, which harbors synergistic activity when combined with standard antibiotics. Promethazine interferes with the bioenergetics of biofilm-grown bacteria, thereby sensitizing generally tolerant bacterial cells in the biofilm state (Donnert et al. 2020).

In addition to large-scale screens, the era of big data and artificial intelligence offers new innovative opportunities in drug discovery (reviewed in Gupta et al. 2021). In silico analyses no longer depend on labor-, time-, and cost-intensive screens in the laboratory. It was demonstrated that genome mining approaches can aid the identification of novel natural products with antimicrobial activity, based on similarities with known bioactive compounds of known structure and function (Cao et al. 2019; Mohimani et al. 2014). In addition, virtual screening approaches have also been applied to predict the anti-biofilm potential of existing drugs. Binding capacities of drug candidates to the solved protein structure of targets of interest were calculated by molecular docking. This approach has been used to identify inhibitors and antagonists of quorum sensing in independent studies (Mellini et al. 2019; Sadiq et al. 2020; Soukarieh et al. 2021).

A major limitation of these in silico approaches is the need for detailed a priori knowledge of structural features of either the drug (to identify analogs), or the target (to identify interfering substances). Recently, sophisticated machine learning approaches paved the way to overcome this bottleneck. Advanced deep learning algorithms were successfully applied to predict antibacterial activities of small molecules: an artificial neural network was first trained with experimental data on bacterial growth inhibition, of a relatively small sample set of 2000 molecules. In the second step, the built model was applied to a gigantic chemical library comprising >107 Mio molecules in order to identify potential lead compounds (Stokes et al. 2020). This machine learning approach led to the identification of halicin, a molecule that exhibits broad-spectrum bactericidal activity. It is particularly noteworthy that halicin was discovered by the algorithm even though its structure is quite divergent from conventional antibiotics. A second example that demonstrates the power of machine learning approaches is the recent development of the artificial intelligence (AI)-based program AlphaFold, the first computational approach to successfully predict protein structures based solely on their amino acid sequence (Jumper et al. 2021). The integration of physical and biological knowledge on structural features of proteins, together with multi-sequence alignments into a deep learning algorithm, enabled accurate 3D structure predictions with near experimental accuracy, even in cases where no similar protein structure is known. Systematic mapping of the pathogen’s genomic mutational space, can reveal the deleterious effects of target inhibition and thus identify critical structural motifs as potential new targets for new antibacterial strategies, including the development of novel drugs as well as vaccines.

These very striking examples impressively demonstrate the potential of AI-based computational approaches and the use of big data to facilitate the identification of novel active compounds. However, knowledge-based approaches, which aim at uncovering novel antibiotic- or anti-biofilm targets, have also made significant advancements over the last decade. A deep understanding of bacterial regulatory processes that are critical for the establishment and maintenance of an infection promises to identify so far unexplored possibilities of intervention. Particularly the decreasing cost of next-generation sequencing (NGS), but also advances in mass spectrometry technologies, have progressed the genomic era into the discipline of functional genomics. In genome-wide approaches, functional genomics uses high-throughput methods to uncover dynamic gene/protein functions and interactions. Vast global datasets such as those generated by the so-called -omics technologies (genomics, transcriptomics, proteomics, metabolomics, and phenomics) have been used to profile bacterial isolates. The challenge today is to link these dynamic, large-scale functional datasets and thus to explain phenotypic behavior in response to an intervention.

In addition to the discovery of promising new drug targets for antibacterial/antiviral intervention, it will be important to determine to what extent they are conserved, e.g., within all isolates in a given pathogen species, and whether the targets are expressed under infection-relevant conditions in the human host. We are only at the beginning of a comprehensive understanding of the regulatory mechanisms that respond to prevailing environmental conditions, involved in the expression of phenotypic traits. In the following section, we review the recent developments in transcriptional profiling and explain what can be learned from the application of RNA-seq in the context of P. aeruginosa infections. We discuss how large transcriptome data sets obtained from diverse clinical isolates can be used to gain an expanded understanding of bacterial adaptation during the infection process in the human host.

4 Transcriptional Profiling Approaches for a Comprehensive Understanding of P. aeruginosa Adaptation

Bacteria adapt to their environment through the expression of specialized phenotypic traits that confer a selective advantage under the prevailing conditions. Complex regulation of gene expression ensures that only those genes that are required, or advantageous, under the given conditions are switched on and translated, ensuring appropriate phenotypic traits. Studying specific environment-driven transcriptional responses is particularly useful for the identification of regulons that can be assigned to stimulus-specific master regulators.

In order to identify environmentally induced gene expression changes in the pathogen P. aeruginosa, early transcriptional studies used simple binary comparisons that typically included the cultivation of a laboratory reference strain under two (or a few more) experimental conditions in vitro. The analysis of gene expression changes in P. aeruginosa upon the switch from a motile to a sessile lifestyle is a prominent example of such an in vitro approach (Dötsch et al. 2012). However, an early microarray-based meta-analysis indicated that the identified biofilm-specific gene expression signatures varied substantially from study to study, depending on the design and practical implementation of the biofilm assay used (e.g., in terms of media composition, the applied biofilm model, static or dynamic cultivation conditions, etc.) (Patell et al. 2010). It became clear that the environment had a substantial impact on the transcriptome. In this respect, it was interesting to see that the variability of the gene expression profile was higher in the P. aeruginosa type strain PA14 cultivated under different infection-relevant environmental conditions, as compared to different strain backgrounds cultivated under the same environmental condition (Dötsch et al. 2015). The P. aeruginosa transcriptional response as a function of the environment can be accessed if P. aeruginosa is cultivated under as many experimental conditions as possible. The goal of this approach is to induce the expression of each individual gene under at least one condition, in order to obtain a more complete picture of transcriptional regulation. With the increasing availability of P. aeruginosa RNA-seq data (recent developments and applications of NGS are reviewed in Valli et al. 2020), the identification of the full extent of the reaction norm seems possible. Thereby, the reaction norm describes the range of expressed phenotypes (here: gene expression) of an isogenic organism in response to different environmental conditions (Schlichting and Pigliucci 1998). In addition, new bioinformatic tools have been developed to integrate large data sets and use them to explain bacterial behavior. For example, it has been demonstrated that the unsupervised machine-learned ADAGE model (Tan et al. 2016; https://adage.greenelab.com/) captures gene expression patterns and identifies regulons that can be linked to different biological processes or experimental conditions (Clay et al. 2020; Doing et al. 2020; Harty et al. 2019; Koehorst et al. 2016). Thus, with the continuous progression and the rapid increase in the availability of transcriptional profiles, the focus of transcriptome analysis shifted from descriptive studies on gene-level variations to comprehensive studies that analyze integrated regulatory networks of co-regulated genes or functional pathways on a global scale (Rajput et al. 2021; Sastry et al. 2019).

4.1 RNA-seq: Progress and Advances in Mimicking Infections in Laboratory Experiments

The niche-specific environment in the human host has a significant influence on phenotypic adaptation of the pathogen during an infection. Therefore, in order to more closely resemble the host habitat, in vitro experiments for transcriptome analyses have been undertaken under conditions mimicking the infection environment. For example: by cultivating bacteria in artificial sputum media reflecting the nutrient composition in CF lungs (Fung et al. 2010; Palmer et al. 2007; Tata et al. 2016), or by adding host factors to the cultivation medium, such as exudates from burn wounds (Gonzalez et al. 2018), mucin glycans, components of native human mucus (Wheeler et al. 2019), expectorated sputum from the human respiratory tract (Cattoir et al. 2012), or by direct cultivation of bacteria in blood from trauma patients (Elmassry et al. 2019).

The presence of commensals and other co-existing pathogens at the infected site, and in particular their secreted metabolites, also have an impact on P. aeruginosa during infection. Complex reciprocal interactions and crosstalk during co-infection in mixed-species models have been observed and a common theme is the activation of virulence mechanisms. For example, S. aureus peptidoglycan was shown to induce the production of extracellular virulence factors in P. aeruginosa (Korgaonkar et al. 2013), and phenazine biosynthesis genes were upregulated in P. aeruginosa, following the induction by ethanol produced and secreted by C. albicans (Doing et al. 2020). In vitro experiments also indicated that the presence of host cells per se induced physiological changes in infecting bacteria (Chugani and Greenberg 2007; Felgner et al. 2020; Frisk et al. 2004; Gellatly et al. 2012), and vice versa (Jenner and Young 2005). As a further advancement in transcriptional profiling approaches, animal models were developed that more closely mimic the in vivo conditions in the human host (reviewed, e.g., in Lorenz et al. 2016). For example, the gene expression status in P. aeruginosa was analyzed in an acute mouse pneumonia model (Pan et al. 2020), a mouse tumor model that resembles the chronic infection status (Bielecki et al. 2013), an acute murine burn wound infection model, and in a chronic murine surgical wound infection model (Turner et al. 2014). To mimic human physiology of the infected body site more closely, ex vivo porcine lung models (Harrington et al. 2022; Harrison and Diggle 2016) and in vivo (burn) wound models in pigs have been developed (Davis et al. 2008). During acute infections, a common theme was the upregulation of several regulators involved in bacterial virulence, including the type III secretion system that allows the bacteria to directly inject toxins into host cells (Pan et al. 2020). Moreover, genes involved in chemotaxis and flagellum-mediated motility were shown to be required for bacterial fitness in acute infections, but dispensable in chronic wound models (Turner et al. 2014).

4.2 Ex Vivo Transcriptomics: Analyzing Gene Expression During an Infection Process

With continuous optimization of sample preparation and remarkable advances in NGS technology, researchers finally succeeded in analyzing bacterial mRNA extracted directly from an infected human tissue and organ. Transcriptional profiling of ex vivo samples can be considered the gold standard when analyzing phenotypic adaptation of pathogenic bacteria to the human infection site. Early ex vivo approaches focused on samples with rather low complexity, such as E. coli from urinary tract infections (Bielecki et al. 2014; Hagan et al. 2010), or S. aureus from prosthetic joints (Xu et al. 2016). Major breakthroughs in P. aeruginosa infection research were recently achieved by the analysis of ex vivo transcriptional profiles from chronic wound infections and chronic CF lung infections. Intact bacterial mRNA could be obtained from CF sputum (Cornforth et al. 2018; Rossi et al. 2018), from soft tissue infections such as chronic wound infections and burn wounds (Cornforth et al. 2018), and directly from an explanted chronically infected lung from a CF patient (Kordes et al. 2019b). Despite the challenges of ex vivo RNA-sequencing, and although approx. 95% of the obtained total reads represented human transcripts, all three approaches obtained sufficient numbers of bacterial reads to allow the identification of a robust P. aeruginosa ex vivo transcriptome. In agreement with previous findings that show a greater impact of variances in environmental conditions than of the genetic background on the transcriptional response (Dötsch et al. 2015), different bacterial strains shared highly similar ex vivo transcriptional signatures despite underlying genetic diversity (Kordes et al. 2019b; Rossi et al. 2018). This indicates that there is a common in vivo transcriptional profile tied to the environment of chronically infected CF lungs. Ex vivo transcriptomics of P. aeruginosa in the respiratory tract of CF patients revealed, for example, bacterial adaptation to oxidative stress and microaerophilic conditions. Moreover, genes encoding for various iron acquisition and uptake systems, as well as genes involved in antimicrobial resistance, were expressed at a higher level in vivo. In vivo transcriptional signatures also reflected a physiological state of bacteria characterized by low (energy) metabolism (e.g., downregulation of NADH dehydrogenases, fatty acid metabolism, and amino acid biosynthesis). Furthermore, (sub-)populations of bacteria appeared to be in a non-motile state, indicated by low ex vivo expression levels of motility-related genes (Cornforth et al. 2018; Kordes et al. 2019b; Rossi et al. 2018).

4.3 Dual-seq: Understanding the Crosstalk and Interplay Between Host and Pathogen During Infection

Until recently, the dynamics of reciprocal changes in gene expression programs of both host and pathogen during an infection have remained largely unexplored. However, more recent technical advances enabled the development of dual RNA-seq approaches that allow the simultaneous analysis of changes in gene expression profiles of the host and the pathogen. Dual RNA-seq approaches have been applied to P. aeruginosa in, e.g., a zebrafish model (Kumar et al. 2018) and an acute pneumonia model in mice (Damron et al. 2016). Both studies provided novel insights into the host-specific immune response triggered by P. aeruginosa and identified the battle for iron as crucial during infections. Furthermore, the analysis of dynamic host–pathogen interactions uncovered a crosstalk between P. aeruginosa and murine macrophages. Upon host cell contact, P. aeruginosa produces an exo-product (spermidine) that triggers phagocytic uptake of the bacteria (Felgner et al. 2020). In addition, a dual RNA-seq approach on P. aeruginosa cells internalized within bladder epithelial cells, uncovered rapid bacterial adaptation to the prevailing stress conditions, indicating that intracellular survival seems to play an important role in bacterial long-term survival during chronic and recurrent urinary tract infections (Penaranda et al. 2021).

4.4 Single-Cell RNA-seq: Analyzing the Heterogeneity in Gene Expression Within a Population

It is well known that expression levels of genes can vary between individual cells within a bacterial community, resulting in a phenotypically heterogeneous population (Ackermann 2015; Evans et al. 2020). Within a bacterial biofilm, heterogeneity can be further sustained due to differences in the microenvironment, characterized by nutrient and oxygen availability, as well as pH gradients, thus, leading to heterogeneity in metabolic states between bacterial biofilm-associated cells (Stewart and Franklin 2008). This adds another layer of complexity to the analysis of phenotypic adaptation and the study of underlying transcriptional responses of bacteria to biofilm growth. The commonly performed bulk transcriptional profiles of thousands of bacterial cells reflect the accumulated gene expression status of all cells present in the population but mask distinct transcriptional signatures or gene expression patterns of single cells within a heterogeneous population. Investigating differences in the physiology of individual cells within a biofilm population could be key for a more comprehensive understanding. Biofilm tolerance is one example for which physiological heterogeneity within a multicellular community has been suggested to be a contributing factor (Schiessl et al. 2019; Stewart 2015). For example, phenazine-mediated cellular redox-balancing was shown to activate bacterial metabolism in microaerobic biofilm regions, and induced biofilm tolerance to ciprofloxacin (Schiessl et al. 2019). Furthermore, previous research indicates that changes in cellular respiration and the intracellular pH drive bacterial tolerance phenotypes (Arce-Rodríguez et al. 2022; Donnert et al. 2020), which may result from local differences in, e.g., pH or osmolarity in the biofilm environment. In this context, it might be interesting to study spatial differences in the transcriptional response of bacteria. Microarray analyses following laser capture microdissection technology were applied in an early approach to investigate local transcriptomes in biofilms. Subpopulations with spatially distinct mRNA abundances of genes involved in general metabolic pathways were identified (Williamson et al. 2012). Recently, a targeted high-resolution transcriptome-imaging approach based on sequential fluorescence in situ hybridization (par-seqFISH) was introduced to record spatial expression patterns of 105 genes at a single-cell level (Dar et al. 2021). In addition, major advances have been achieved in untargeted, probe-independent technologies to study the transcriptional diversity of single cells. However, single-cell RNA-sequencing (scRNA-seq) is still challenging for bacteria due to the very low amounts of mRNA in each individual bacterial cell with approx. 10–100 fg per cell (Imdahl and Saliba 2020). Nevertheless, recent successes (Blattman et al. 2020; Imdahl et al. 2020; Kuchina et al. 2021) hold great promise that scRNA-seq will soon be available as a standard application and will allow the investigation of single-cell transcriptional landscapes. Although the small amounts of on average 0.4 mRNA copies per gene and cell will not allow the capture of complete transcriptomes, scRNA-seq will give valuable insights into the distribution of the most highly expressed genes in each bacterium across populations. It may even be possible to build a more comprehensive transcriptional profile at the single-cell level if single-cell transcriptional profiles were integrated with bulk population profiles to extrapolate the full transcriptome from the partial gene expression signatures.

5 Global Profiling Approaches and Functional Genomics to Study Bacterial Adaptation

An adaptive gene expression profile produced in a challenging environment might become fixed in bacterial isolates that prevail for a prolonged period of time under the respective condition. This can be achieved by the accumulation of patho-adaptive mutations, e.g., in global regulators, which then fix the expression of a downstream regulon in an ON or OFF state. The analysis of the phenotypic traits of naturally evolved clinical isolates can thus be of great value as those adapted phenotypes might reflect important adaptations to the environment from which they have been recovered. Due to advances in next-generation sequencing, large-scale genome and transcriptome approaches to investigate not only single but also large collections of clinical isolates are now feasible and affordable. For studying bacterial adaptation to the infection within the human host, different approaches are used. In longitudinal studies, sequential bacterial isolates recovered from individual patients are followed over a long period of time. Those studies are well suited for investigating the nature and dynamics of evolutionary adaptation and have been applied for studying long-term adaptation during chronic infections, such as in the lungs of CF patients (Gabrielaite et al. 2020; Klockgether et al. 2018; Marvig et al. 2015; Smith et al. 2006; Wiehlmann et al. 2007; Yang et al. 2011). Alternative approaches are based on the cross-sectional analysis of large collections of heterogeneous clinical isolates. The analysis of strains from diverse geographic origins, different patients, infection sites, and infection states (e.g., acute vs. chronic) can give new information on the overall diversity potential of a bacterial species and its possibilities for adaptation to a multitude of host niches. Adaptability and phenotypic diversity in a population of genetically heterogeneous clinical P. aeruginosa isolates are reflected, for example, in the expression of very diverse biofilm structures (Fig. 11.1).

Fig. 11.1
16 images of different infection sites and stages with pathogens. A measurement in micrometers is given on the lower left side of each image.

Clinical isolates of the opportunistic pathogen P. aeruginosa from diverse infection sites and stages (acute vs. chronic) produce a large array of structurally different in vitro biofilms. Distinct biofilm features of individual strains are formed with high reproducibility. Biofilms were grown in static microtiter plates for 48 h. Image stacks were acquired in the bottom of the wells using a confocal microscope (Leica SP8) following live/dead staining (Müsken et al. 2010; Thöming et al. 2020). Living cells are shown in green (Syto9); dead cells are shown in red (propidium iodide: PI). 3D reconstructions were generated with the Imaris Software. The scale bar represents 50 μm

Full size image

With the aim to investigate changes in transcriptional profiles, both under standard laboratory conditions and in vitro biofilm growth conditions, a large set of genetically diverse clinical P. aeruginosa isolates has been analyzed (Dötsch et al. 2015; Thöming et al. 2020). Interestingly, the overall variability of transcriptional profiles across very heterogeneous clinical P. aeruginosa isolates was relatively small (Dötsch et al. 2015; Thöming et al. 2020). Nevertheless, transcriptional profiles exhibited a higher diversity when recorded under biofilm-growth conditions as opposed to standard lab conditions (planktonic growth in rich medium) (Thöming et al. 2020). Growth within biofilms might be expected to mimic the growth state of some of the clinical isolates in the human host during a chronic infection more closely than planktonic growth in rich medium. Thus, some of the clinical isolates could be expected to have acquired adaptive genomic mutations, which have been selected during their in vivo biofilm growth. While those mutations do not necessarily impact the transcriptional profile under non-selective planktonic conditions, they are more likely to do so under biofilm growth condition and thus they might drive the diversity of the transcriptional profiles under biofilm growth (Thöming et al. 2020).

Transcriptional profiling of clinical isolates under different growth conditions holds great promise to provide a more holistic picture of the environment-dependent expression of certain adaptive traits. Furthermore, the availability of large amounts of transcriptional profiles and the possibility for exploratory analysis of gene expression patterns coupled with new information on genomic sequence variation provides the unique opportunity to link transcriptional signatures to causative adaptive mutations, and thus to assign phenotypes to genotypes.

5.1 Gene Expression Patterns of Individual Genes Across a Collection of Isolates

Transcriptional profiles do not only reveal complex adaptation strategies of individual bacterial pathogens, e.g., to conditions prevailing in the human host, but profiles can also be used to study gene expression patterns of individual genes across a collection of isolates. A recent study re-analyzed the transcriptional profiles of >400 genetically diverse clinical P. aeruginosa, which have been isolated from a multitude of different infection sites and states (Jeske et al. 2022). The work revealed that although some genes were expressed at a high level and some at a lower level, the great majority of the genes were expressed at a comparable level across the isolates. Nevertheless, although all transcriptional profiles were recorded under the identical non-selective, rich medium condition, some genes exhibited a bi-modal expression pattern: i.e., high expression in one subpopulation and low expression levels in another subpopulation of clinical strains. Clustering of clinical isolates based on high/low expression of the bi-modally expressed genes revealed groups of genetically unrelated strains exhibiting similar expression patterns across those bi-modally expressed genes. This indicates that a certain gene expression status has evolved independently in several clinical isolates and thus might indicate an important adaptation to the conditions in the human host during an infection process. Of note, the convergent changes in gene expression of a large fraction of those bi-modally expressed genes could be attributed to mutations in the lasR gene. LasR is a key QS regulator and mutations in lasR are frequently found in P. aeruginosa strains isolated from chronic infections in CF patients (D’Argenio et al. 2007; Feltner et al. 2016; Hogardt and Heesemann 2013). Furthermore, lasR mutations have previously been shown to have pleiotropic effects, affecting more than one phenotypic trait in P. aeruginosa (Dötsch et al. 2015). Interestingly, although reduced expression of a common LasR-dependent regulon (including QS-regulated virulence factors, such as phenazines, rhamnolipids, LasA protease and LasB elastase (Jeske et al. 2022)) could be found across the clinical lasR-defective isolates during planktonic growth, this was not the case when the transcriptional profiles of the lasR mutants were recorded under biofilm growth conditions. Under those conditions, the lasR mutants exhibited transcriptional profiles that were comparable to those of the lasR wild-type isolates. Obviously under biofilm growth conditions, as was shown before for phosphate-limiting conditions (Soto-Aceves et al. 2021) and late stationary phase cultures (Cabeen 2014), LasR became dispensable, and the second homoserine lactone-dependent QS regulator, RhlR, took over. Frequently observed lasR mutations might not only be without functional consequences, when it comes to the expression of downstream-regulated genes during chronic biofilm-associated infections, but they might even confer fitness advantages to the pathogen (Jeske et al. 2022).

5.2 Identification of Core Phenotypic Traits by Applying Global Correlation Studies

Large-scale profiling approaches can be used to uncover commonly (or uniquely) expressed genes across bacterial strains when cultivated under identical environmental conditions. The common responses, which are shared by the entire population, uncover important adaptive traits and represent a universal response at the species level to the given environment. However, distinct responses, which are shared by only sub-groups of a population, might also be identified. Subgroup-specific gene expression programs might reveal parallel evolution of adaptive traits in independent strains. The challenge then is to link the distinct gene expression pattern of the bacterial sub-groups to associated phenotypes. Multi-layered integrative analysis of genomic, proteomic, metabolomic, and transcriptomic data combined with detailed phenotypic information on the clinical isolates has the potential to gain new knowledge on the importance of a variety of infection-relevant traits, e.g. virulence (Depke et al. 2020; Kordes et al. 2019a), antibiotic resistance (Khaledi et al. 2016, 2020; Seupt et al. 2021), and biofilm formation (Erdmann et al. 2019; Thöming et al. 2020).

5.3 Identification of Group-Specific Traits That Have Evolved via Parallel Evolution

A prominent example of the successful association of distinct sub-group phenotypes with characteristic transcriptional profiles has been recently demonstrated for biofilm formation. Thöming et al. (2020) profiled the biofilm structures of >400 clinical P. aeruginosa isolates by the use of a high-throughput confocal microscopy screen (Müsken et al. 2010). Despite a broad diversity (represented by microscopic images of a sub-selection of clinical isolates, Fig. 11.1), three major characteristic biofilm structures were repeatedly produced in genetically independent strains. The analysis of biofilm-induced gene expression changes uncovered a cluster-specific convergence of differentially regulated genes during biofilm growth as compared to the planktonic growth state. Thereby, the clinical isolates that formed dense, unstructured biofilms showed an upregulation of numerous genes encoding for matrix components (e.g., alg, pel, or psl genes (Mann and Wozniak 2012)) and attachment (cupA-C gene clusters involved in the assembly of cell surface fimbriae (Vallet et al. 2001)). In contrast, clinical isolates that formed fine filamentous biofilms exhibited a biofilm-specific upregulation of phenazine genes encoding for the redox-active electron shuttle pyocyanin (Wang et al. 2010), while a third group, which was characterized by the formation of small microcolony-like biofilm aggregates, showed a combined regulation of both, the phenazine genes and genes involved in matrix production. The data provide an example of how the combined analysis of (biofilm) phenotypes and transcriptional profiles can uncover different strategies that the opportunistic pathogen P. aeruginosa applies in order to build a biofilm. Interestingly, an extended analysis of additional infection-relevant phenotypes revealed further virulence characteristics that could be assigned to the isolates in the individual biofilm clusters. Clinical isolates that produced structurally related in vitro biofilms exhibited also specific pathogenicity and motility behavior. They even exhibited distinct metabolic profiles as determined in an untargeted metabolomics approach (Depke et al. 2020). All in all, this shows that different sub-groups of clinical isolates repeatedly emerge, possibly in different environments of the human host. However, at the same time, the studies demonstrate that despite the complexity of phenotypes and the underlying genomic variations, the evolution of phenotypes is not random and very similar complex phenotypes characterized by common biofilm/virulence/motility/metabolite expression patterns evolve during human infections again and again.

5.4 Persistence of Transcriptional Responses as Memory Responses

C-di-GMP can affect the transcription of a multitude of target genes and is known to be involved in the motile to sessile lifestyle switch (Hengge 2009). However, many clinical isolates of P. aeruginosa exhibit increased levels of c-di-GMP even under non-inducing planktonic conditions. In this respect, it is worth noting that elevated c-di-GMP levels in clinical isolates are not necessarily the result of adaptive genetic mutations. Recent evidence suggests that individual transcriptional responses can persist as memory responses for several generations, even after the removal of the external stimulus (Casadesús and D’Ari 2002; Kordes et al. 2019a; Lambert and Kussell 2014; Lee et al. 2018; Norman et al. 2013; Ronin et al. 2017; Wolf et al. 2008). A multigenerational memory of surface-exposed P. aeruginosa cells observed during early biofilm formation has been linked to the Pil-Chp system (Lee et al. 2018), which involves type IV pili and intracellular cyclic AMP signaling (Luo et al. 2015). Furthermore, it was shown that the acute virulence phenotype of P. aeruginosa in an in vivo invertebrate G. mellonella infection model was associated with low c-di-GMP levels, which were induced by the invertebrate host conditions and memorized by the bacterial population when cultured outside the host for several generations (Kordes et al. 2019a). Vice versa, persisting high levels of c-di-GMP despite growth under non-c-di-GMP inducing conditions, has been demonstrated in a clinical small colony variant (SCV) P. aeruginosa isolate (Koska et al., in preparation). In this SCV, alginate expression levels were linked to elevated c-di-GMP levels and only gradually switched back to wild-type levels after prolonged incubation under non-inducing conditions. Much remains to be learned about the memory state of the expression of distinct bacterial traits, in particular on the molecular mechanisms of how a bacterial behavior is passed on to descendants over multiple generations that have never experienced the phenotype-inducing conditions.

6 Conclusion and Outlook

Pseudomonas aeruginosa is a highly adaptable opportunistic pathogen that can cause a broad range of difficult-to-treat infections. Treatment options, especially for persistent biofilm-associated infections, are very limited and there is an urgent need to identify new opportunities for clinical intervention. Transcriptome analyses have contributed significantly to a better understanding of how bacteria adjust gene expression to survive in challenging environments. Reports on the in vivo transcriptome of P. aeruginosa from an explanted CF lung represent breakthroughs in the study of transcriptional adaptation to the host habitat during chronic infections. In addition, functional genomics studies that integrate transcriptomes with other -omics data promise an advanced multi-layered understanding of bacterial adaptation during infection. The integration of the complex data, including phenotypic data, offers the unique opportunity not only to identify the transcriptional consequences of acquired patho-adaptive mutations but also to decipher how a particular gene expression profile is translated into better-adapted phenotypes. Furthermore, the profiling of diverse and large sets of clinical isolates promises to infer evolutionary pathways that define, e.g., virulence, biofilm formation, and antibiotic resistance phenotypes. In this context, it appears that the bacteria follow different parallel evolutionary paths depending on their location in the human host and the type of infection (acute or chronic) as a result of selection pressure in the different hostile host environments.

In the future, a more holistic understanding of the opportunistic pathogen P. aeruginosa will be achieved by integrating more and more detailed data sets. This will include phenotypic data that allow increasingly accurate stratification of different phenotypic traits. This is an important prerequisite for more robust correlations between expressed phenotypes and the underlying regulatory mechanisms. The integration of large-scale, multi-layered -omics datasets holds great potential for identifying key pathways essential for specific adaptive traits during host infection. This knowledge contributes to the identification of new targets for novel therapy options as well as novel diagnostic markers as predictors for the severity of the disease.