Exploring aggregation genes in a P. aeruginosa chronic infection model

ABSTRACT

Bacterial aggregates are observed in both natural and artificial environments. In the context of disease, aggregates have been isolated from chronic and acute infections. Pseudomonas aeruginosa (Pa) aggregates contribute significantly to chronic infections, particularly in the lungs of people with cystic fibrosis (CF). Unlike the large biofilm structures observed in vitro, Pa in CF sputum forms smaller aggregates (~10–1,000 cells), and the mechanisms behind their formation remain underexplored. This study aims to identify genes essential and unique to Pa aggregate formation in a synthetic CF sputum media (SCFM2). We cultured Pa strain PAO1 in SCFM2 and LB, both with and without mucin, and used RNA sequencing (RNA-seq) to identify differentially expressed genes. The presence of mucin revealed 13 significantly differentially expressed (DE) genes, predominantly downregulated, with 40% encoding hypothetical proteins unique to aggregates. Using high-resolution microscopy, we assessed the ability of mutants to form aggregates. Notably, no mutant exhibited a completely planktonic phenotype. Instead, we identified multiple spatial phenotypes described as “normal,” “entropic,” or “impaired.” Entropic mutants displayed tightly packed, raft-like structures, while impaired mutants had loosely packed cells. Predictive modeling linked the prioritized genes to metabolic shifts, iron acquisition, surface modification, and quorum sensing. Co-culture experiments with wild-type PAO1 revealed further spatial heterogeneity and the ability to “rescue” some mutant phenotypes, suggesting cooperative interactions during growth. This study enhances our understanding of Pa aggregate biology, specifically the genes and pathways unique to aggregation in CF-like environments. Importantly, it provides insights for developing therapeutic strategies targeting aggregate-specific pathways.

IMPORTANCE

This study identifies genes essential for the formation of Pseudomonas aeruginosa (Pa) aggregates in cystic fibrosis (CF) sputum, filling a critical gap in understanding their specific biology. Using a synthetic CF sputum model (SCFM2) and RNA sequencing, 13 key genes were identified, whose disruption led to distinct spatial phenotypes observed through high-resolution microscopy. The addition of wild-type cells either rescued the mutant phenotype or increased spatial heterogeneity, suggesting cooperative interactions are involved in aggregate formation. This research advances our knowledge of Pa aggregate biology, particularly the unique genes and pathways involved in CF-like environments, offering valuable insights for developing targeted therapeutic strategies against aggregate-specific pathways.

INTRODUCTION

Bacterial aggregates are observed in both natural and artificial environments. In the context of disease, aggregates have been isolated from both chronic and acute infections and can be formed by bacteria, archaea, and fungi (1–4). Pseudomonas aeruginosa (Pa) is one such bacteria. As an opportunistic pathogen, Pa causes disease in those whose immune systems or barrier functions are compromised. This includes those with chronic and acute wounds, medical devices, and chronic infection in the lungs of people with the genetic disease cystic fibrosis (CF). Once chronic Pa colonization is established, a large proportion of the infecting bacteria grow within airway sputum as aggregates (~10–1,000 cells) (1). By contrast, in vitro Pa growth results in the formation of large structures containing millions of cells (5, 6). Previous studies of Pa cells in large, shaken flask cultures (macro-scale biofilm structures), have contributed significantly to our understanding of Pa growth, communication systems, and the mechanisms Pa utilizes to become tolerant to many antibiotics (1). However, growth in this context does not closely recapitulate that of actual infection—specifically the presence of aggregates. This divergence gives rise to a fundamental question in biofilm microbiology—how does spatially structured growth as aggregates influence infection?

While both biofilms and aggregates exhibit clinical tolerance to antimicrobial agents, it is probable that the underlying mechanisms contributing to this tolerance differ. As these phenotypes intersect, distinct differences between aggregates and biofilms have emerged. Notably, the exopolysaccharides Pel and Psl have been identified as crucial components for maintaining the integrity of Pa biofilms (7). However, even in the absence of these polysaccharides, Pa aggregates can still form in synthetic CF sputum media (SCMF2) (1). While the presence of Pel and Psl undeniably contributes to the tolerance of aggregates to therapeutic interventions like antibiotics and bacteriophages, the physical occurrence of aggregation appears to be closely linked to enhanced survival (1). Notably, the regulation of quorum sensing (QS) in Pa biofilms has predominantly been characterized as a binary on/off system for the coordination of group behaviors and the production of public goods (2). However, our previous research has revealed that in aggregates, the response to QS signals is considerably more diverse (8). For instance, aggregates in the path of a signal may not uniformly respond to it, and alterations in the expression of the signal receptor (LasR) can partially counteract this variability. This stark difference between biofilms and aggregates is noteworthy, especially when considering that Pa utilizes QS to regulate approximately 300 virulence genes. There are also now several examples from the microbiome that demonstrate how microbes can manipulate the spatial organization of their population or community, inferring changes in virulence. The range of functional outcomes that are mediated this way include regulation of QS, increased antibiotic tolerance, and cross-feeding of metabolites (8–11). This suggests that other virulent behaviors may be heterogeneously employed across individual aggregates during growth and that the ability to modulate pathogen position can modulate pathogen virulence. This could have significant consequences for the evolution of bacterial populations and ultimately how we should approach them therapeutically.

These observed functional differences between aggregates and biofilms highlight an urgent need to understand the mechanisms Pa utilizes to regulate aggregate formation to develop new therapeutic strategies. In this study, we identify a subset of genes that play an integral role in aggregate formation in a synthetic CF sputum media (SCFM2). Many of these genes encode hypothetical proteins. Using high-resolution microscopy, we have uncovered a spectrum of spatial phenotypes when aggregate genes are disrupted. Using available omics tools, we have been able to predict groups of functional pathways that contribute to such variations in the spatial structure of developing Pa aggregates. By mixing aggregate mutants with the WT PAO1, we found that the presence of fully functional cells incites further spatial heterogeneity, suggesting that multiple finely tuned biological systems are required for successful aggregation. These data present the use of SCFM2 as a tool to dissect the mechanisms Pa uses to form infection-relevant aggregates. More specifically, we demonstrate how we can utilize this knowledge to pair micron-scale spatial structure with the physiological response of Pa cells within aggregates.

RESULTS

A unique subset of genes is critical for P. aeruginosa aggregate formation

We have previously shown that Pa growth as aggregates occurs in distinct phases, where single planktonic cells form aggregates that undergo dispersal and form new aggregates (3). Although this can be observed in vitro, we still do not understand the mechanisms required specifically for aggregate formation. The goal of this study was to determine whether a distinct subset of genes is required for Pa aggregate formation in a synthetic CF sputum media (SCFM2) compared to larger biofilm models. While SCFM2 replicates both physical and nutritional aspects of CF sputum, mucin is the only known required component for aggregation in this chronic infection model. Pa cultured in the absence of mucin is unable to form aggregates (1). We leveraged this mucin dependency to identify genes important for aggregate formation.

PAO1 was cultured in both SCFM2 and LB in the presence and absence of mucin for 8 hours. RNA sequencing (RNA-seq) revealed several genes that were significantly differentially expressed (DE) in the presence of mucin (Fig. 1). Of these DE genes, 40% were identified as hypothetical proteins, that is, no known function. Only two genes were determined to be significantly upregulated in SCFM2 with mucin: PA1530 and PA1531. Referenced as hypothetical proteins in Pa, orthologs in other Pseudomonas spp. also have no known function. Of the 17 significant DE genes, 15 genes were downregulated in the presence of mucin. We found that five of these genes are non-coding RNAs (ncRNAs) with known associations with Pa biofilm regulation: phrS, crcZ, rsmY, rsmZ, and P30 (12–15). Within known Pa quorum sensing (QS) systems, only pqsC expression met our significance cutoffs. We found that the pqsABCDE operon is consistently downregulated at 8 hours in the presence of mucin (SCFM2 and LB), suggesting PQS signaling is less important once aggregates have formed (>4 hours).

Fig 1.

Differentially expressed genes in aggregates. (a) Volcano plot of differentially expressed genes in SCFM + mucin compared to SCFM-mucin at 8 hours of growth. Upregulated genes are shown in red, downregulated genes are shown in blue, and non-significant genes are shown in gray. 50 most significantly differentially expressed genes are labeled. (b) Differentially expressed genes in SCFM + mucin compared to LB + mucin at 8 hours. (c) Heatmap of genes of interest and associated genes in their operon/pathway at 8 hours. Heatmap also includes several classical biofilm-associated genes such as polysaccharides and QS genes that notably do not show differential expression in aggregates in this study.

Next, we compared Pa cultured in SCFM2 with LB broth containing mucin. This allows us to identify genes that are specific to the nutritional environment of SCFM2 (Fig. 1). This comparison revealed several hundred significant DE genes between the two growth environments, approximately 13% of the total PAO1 genome. More than half of the DE genes were upregulated (52%) in SCFM2 when compared to LB. This is a significant contrast to the number of upregulated genes (PA1530 and PA1531) attributed to mucin comparing SCFM2 alone. Like our comparison of SCFM2+/mucin, a large percentage of DE genes were hypothetical proteins. Many pathway mapping programs (KEGG or GOseq) tend to exclude hypothetical proteins from analysis, making these tools inappropriate for our data. Therefore, we chose to focus on the 25 most upregulated and 25 most downregulated genes from this analysis (Table S1). Of this group, 40% of genes were identified as hypothetical proteins. Notable downregulated genes specific to growth as aggregates in SCFM2 include the RNA-binding regulator rsmA, O-chain length regulator wzz, and virulence regulator amiE (15). We also observed an upregulation of several genes in the pch, fptABCX, and glc operons, encoding for the siderophore pyochelin, a pyochelin receptor, and a glycolate oxidase, respectively (16).

To further explore where our RNA-seq data overlaps with other Pa growth environments we compared significantly dysregulated genes in SCFM2 aggregates with those identified in data sets from PAO1 grown on a biofilm disk, within a flow cell and in ex vivo CF sputum (Fig. S1) (4, 5). We identified >500 unique genes to SCFM2 and 420 genes similarly dysregulated by PAO1 grown in CF sputum, including PA0621, PA0643, PA0644, PA0646, PA0985, PA0998, PA1530, PA1531, PA4782, PA5102, and PA3406. These genes notably do not include those most commonly associated with the development of larger biofilms, such as the polysaccharide encoding genes pel, psl, or their related QS regulated las and rhl operons (Fig. 1c) (7–17) (6–9). We also found that Pa growth as aggregates results in the downregulation of ncRNAs and associated repressors that impede biofilm formation (13, 14, 18). These data suggest two things: (1) there are biological and metabolic processes that are specific to Pa growth as aggregates (Fig. 1c) and (2) that aggregate formation requires coordination across multiple biological pathways, including several that are still poorly defined.

Aggregation mutants display a range of spatial phenotypes

Our initial goal was to validate our RNA-seq data and identify which genes, when disrupted, result in a non-aggregating, planktonic phenotype. To do this, we cultured single transposon mutants of the 50 most significantly dysregulated genes in SCFM2 to assess how disruption of the gene impacted the ability of Pa to form aggregates (Fig. 2). Cultures were imaged with confocal laser scanning microscopy (CLSM) over 15 hours and image analysis software was used to create 3D digital renderings of individual aggregates. Objects were filtered by size based on the dimensions of a single Pa cell; objects with volumes between 1–5 μm³ were identified as single cells, and objects with volumes > 5 µm³ were identified as aggregates. Objects with volumes less than 1 µm³ were excluded from this analysis (as previously described (10)). Mutants were compared to WT aggregates quantitatively, characterized by the total biomass volume (μm³), and the average size of individual aggregates (μm³) over time as well as qualitatively, that is, the presence or absence of aggregation (Fig. 2; Fig. S2).

Fig 2.

Transposon mutant aggregate phenotypes. (a) Examples of aggregation, with WT shown in green and two representative mutants shown in red. Transposon mutant PA2111 is an example of the entropic phenotype, with long chains of stacked cells oriented along the coverslip. Transposon mutant PA0998 is an example of an impaired phenotype, with slow growth and loosely packed cells that become most apparent at later timepoints. Scale bar is 10 µm (b) Average aggregate volume of transposon mutants over time. Mutants are grouped by phenotype. Mutant aggregates differ significantly from WT but vary over time. Three biological replicates ± SEM, significance calculated using Kruskal-Wallis (P value < 0.0001) with multiple comparisons test (alpha 0.05).

These experiments identified 13 genes of interest, where Pa growth occurred as unique aggregation-deficit phenotypes when compared to WT PAO1 aggregation. We defined these aggregate phenotypes as “normal,” “entropic,” or “impaired” (Fig. 2a). It is important to note that the disruption of no gene resulted in a completely “non-aggregating” planktonic phenotype. Entropic aggregates are the result of Pa cells tightly packed side by side in long “raft-like” structures. In comparison, impaired aggregates consist of loosely packed Pa cells with a number of planktonic cells during growth. A unique spatial arrangement, observation of the entropic phenotype in other studies of Pa is limited. Initially described, “depletion” or “entropic” was used to describe Pa growth in relation to the physical forces or entropy present after the addition of polymers to media (19). A second example describes the significance of Pa’s O-antigen in mediating this spatial structure (19, 20).

We observed the entropic phenotype in cultures of Tn-mutants PA0435, PA0646, PA2111, PA2207, PA4782, and PA5102. Here, cells pack tightly side by side, forming chains or stacks that elongate as the population matures. These structures form channels along the coverslip, overlaid by dense populations of disorganized planktonic cells and aggregates in the middle and upper layers of the Z-stack (Fig. 2a). Entropic mutants were seen to maintain large planktonic populations, likely contributing to their accelerated growth rate compared to the WT. PA5102, PA2207, and PA0435 exhibited a loss of chain structures and the corresponding population decline at 14 hours (Fig. S1). Although we observe a general increase in growth rate compared to the WT, among entropic mutants this is variable. For example, PA0646 and PA2111 display a similar doubling time as the WT (80 minutes), while PA5102 and PA2207 exhibit robust, rapid growth that surpasses WT by 4 hours (94–108 minutes). PA0435 and PA4782 have an initial growth delay in SCFM2, entering the exponential phase at 5 and 10 hours, respectively. In addition, both mutants form aggregates that are not significantly different from WT by volume (Fig. 2b; Fig. S1). PA5102 and PA2207 produce aggregates that are significantly larger (by volume) than WT, as well as maintaining their distinct morphology. However, PA5102 WT-like aggregation was able to be restored after complementation (Fig. S2).

The impaired phenotype (characterized by loosely packed Pa cells) exhibits a similar growth rate to the WT PAO1, although significantly slower than that of entropic mutants (Fig. 2a). This group includes the transposon mutants PA0998 (pqsC), PA1864, PA3009, PA3160 (wzz1), and PA3406 (hasD). All mutants except for PA1864 exhibit impaired or delayed growth, entering exponential growth around 9 hours as compared to 4 hours in WT cultures (Fig. S1). The impaired mutant PA0998 (pqsC) demonstrates a significant growth deficit when compared to WT, revealing a previously unreported dependency on PQS quorum sensing for normal growth and wild-type aggregate formation in a CF lung-like environment. PA0998 (pqsC) and PA3160 (wzz1) produce aggregates that are significantly larger than the WT at 4 hours, but transition to a primarily planktonic population of cells between 4 hours and the start of exponential growth at 9 hours (Fig. 2b). The characteristic “impaired” architecture becomes most apparent between 10 and 15 hours of growth. PA1530 and PA1531 exhibit a “mixed” phenotype in which they exhibit impaired spatial structure during the first 10–12 hours of growth, after this time point, we observed the formation of entropic stacks. Complementation of PA0998, PA1864, and PA3406 resulted in a reversion to WT-like aggregate formation (Fig. S2).

Predictive modeling reveals functional pathways specific to Pa aggregates

Next, we wanted to better understand the impact of disrupting aggregation genes on the physiology of Pa. Of the 13 prioritized genes (Table 1), 10 code for a hypothetical protein. We assigned predicted functions and pathways to each gene using our predictive modeling pipeline (Fig. S2). This pipeline uses sequential and structural homology to assign gene functions and co-expression patterns to predict protein participation within known pathways. Interestingly, only three genes have known or suggested links to previously reported Pa biofilm functions: PA3406 (hasD) a component of the HasAD hemophore, PA0998 (pqsC) a component of the PQS quorum sensing (QS) pathway, and hypothetical protein PA1864 which has homology to a transcriptional repressor of FimWX-mediated surface adhesion (Table 1) (21–24). PA1864 and hasD were upregulated in aggregate conditions 28 and 39-fold, respectively, while pqsC was downregulated −7-fold.

TABLE 1.

Table of genes of interest, prediction function, pathway, and aggregate phenotype^a

Gene	Known function	Predicted function	Predicted pathway	Phenotype
Fold change
PA1531	Hypothetical	ABC transport-substrate-binding protein	GLU/ARG/ORN import, synthesis	Mixed
2307.34
PA1530	Hypothetical	ABC transport-periplasmic binding protein	Periplasm nutrient scavenging	Mixed
120.63
PA3406	hasD	Heme acquisition via T1SS	HasAP heme acquisition	Impaired
39.02
PA0435	Hypothetical	Sodium:proline symporter, PepSY TM domain	Proline import	Entropic
34.91
PA1864	Hypothetical	ycdC-like negative transcriptional regulator	FimW-FimX mediated surface adherence	Impaired
27.89
PA5102	Hypothetical	Fatty acid desaturase	Glycine betaine catabolism	Entropic
27.09
PA2207	Hypothetical	Tripartite tricarboxylate transporter TctA	Tripartite tricarboxylate transport	Entropic
25.3
PA0646	Hypothetical	F-type pyocin	Mucin-specific downregulation of pyocins	Entropic
−4.25
PA0998	pqsC	QS signal response	PQS quorum sensing system	Impaired
−7.25
PA3009	Hypothetical	ABC transport–ATP-binding protein	Unknown	Impaired
−8.7
PA4782	Hypothetical	Secreted signal lipoprotein	Unknown	Entropic
−9.31
PA3160	Wzz1	O-antigen chain length regulator	O-antigen regulation	Impaired
−9.75
PA2111	Hypothetical	Putative allophanate hydrolase subunit 1	Urea cycle/glutamyl cycle	Entropic
−9.78

^a Genes of interest identified from CLSM screening of transposon mutants. Gene function and predicted pathways for hypothetical proteins are informed from our prediction pipeline.

Using our pipeline, we are now able to propose a model of the genes and pathways critical for aggregate formation (Fig. 3; Fig. S3). We were able to group them by function, specifically by metabolism, iron acquisition, interaction and competition, surface modification, attachment, and QS-related communication.

Fig 3.

Proposed pathways of genes of interest. Hypothetical gene functions and pathways were predicted using the computational pipeline. Genes have been grouped by predicted function. Upregulated functions are shown in green, downregulated functions are shown in red. Gene names within function indicate associated genes that were present in our data set. The figure was created with Biorender.

Metabolic shifts

Within the 13 genes, we identified an upregulation of multiple predicted amino acid transporters. PA1531 is predicted to be the substrate binding protein (SBP) of an ABC transporter and has high sequential and structural similarity to SBPs specific to glutamate, arginine, lysine, and ornithine. PA1531 is the most upregulated gene across all aggregate conditions, reaching a > 2,300-fold change in expression by 8 hours. PA1530 is predicted to be a periplasmic binding protein (PBP), which couples with specific ABC transporters in the periplasm to enhance nutrient uptake (25). PA1530 is also highly upregulated in aggregates at a 121-fold change. Our data strongly suggest co-expression and interaction between PA1530 and PA1531, indicating a critical role for amino acid transport and consumption during aggregate formation (Table 1). PA0435, a predicted proline/sodium symporter, is upregulated >30-fold. This protein also contains a pepSY domain, which is common in membrane proteins and has a proposed role in regulating local peptidase activity (26). PA2207 is upregulated 25-fold and is annotated in other Pa strains as tctA, the symport protein of the larger tripartite tricarboxylic transporter (TTT) assembly. TTTs are not well characterized but have been shown to use ion gradients to bring citrate into the cell—although their substrate specificity is thought to be low (27, 28). By contrast, we observed a >8-fold downregulation of PA3009, which is predicted to be the ATP binding component of an ABC transporter.

Analysis of PA5102 and PA2111 provides support that distinct metabolic shifts occur as cells transition from a planktonic to an aggregate growth style. Upregulated ~27-fold at 8 hours, PA5102, using our functional pipeline, we predict is a fatty acid desaturase that acts on exogenously sourced saturated fatty acids (SFAs) (Fig. 3). This is further supported by the structural similarity between PA5102 and pseudomonas fatty acid desaturase DesB—known to desaturate exogenous SFAs. PA5102 is less similar to DesA, which functions on membrane-bound SFAs (29)(Fig. S6). STRING analysis predicts an association between PA5102 and the glycine betaine and carnitine catabolism pathways, which begins with host-derived SFAs such as phosphatidylcholine or sphingomyelin (29). This pathway results in the production of amino acids such as glycine is further supported by the downregulation of betI in aggregate samples at 4 hours, followed by the upregulation of both PA5102 and dhcB (Fig. 3). PA2111 is predicted to be an allophanate hydrolase that is downregulated −9-fold in aggregates. Allophanate hydrolases are involved in the conversion of allophanate and H₂O into ammonia and CO₂, ultimately leading to the formation of carbamoyl phosphate, a major input into the urea cycle (30). This downregulation coupled with similar growth of PA2111 to the WT and a lack of other DE urea cycle-associated genes suggests that the urea cycle may not be utilized as heavily in aggregates as it is in planktonic cells (Fig. S1).

Iron acquisition

PA3406 (hasD) is the inner membrane component of the HasAD hemophore and is upregulated in aggregates 39-fold at 8 hours, accompanied by upregulation of the remaining HasAP heme acquisition and pch/fptABCX pyochelin operons (Fig. 1 and 3). Interrupting the HasAP secretion complex (and therefore, heme acquisition capacity) in a HasD transposon mutant leads to a growth delay and a loosely packed, impaired biofilm phenotype with a large planktonic population (Fig. 2).

Interaction and competition

PA4782 (−9-fold change) is predicted to be a secreted lipoprotein with a signal peptide, although our pipeline was unable to assign a specific function or pathway due to a lack of homology to proteins in other organisms. PA0646 is an F-type pyocin that is downregulated −4-fold and is part of a larger trend of mucin-specific pyocin downregulation (31) (Table S2). Pyocins are secreted particles used by Pa to compete against other organisms and even other strains of Pa. We saw an interesting pattern of pyocin downregulation that is consistent and exclusive to mucin-containing samples and includes a mixture of R and F pyocins, with one S pyocin—pyoS5 (Table S2) (31). Transposon mutants of both PA4782 and PA0646 form entropic aggregates (Table 1).

Surface modification

Pa as aggregates also exhibit a pattern of downregulation of well-described extracellular components, which cause aggregation-deficient phenotypes. PA3160 (wzz1) is a regulator of O-antigen length on LPS and is downregulated 10-fold in aggregate conditions (Table 1). These findings are accompanied by a trend of downregulation in other O-antigen and LPS-modifying genes including hisH2, wapR, and phoP, suggesting a decrease in O-antigen and LPS modification in aggregates (20). More specifically, these data suggest that cells capable of forming aggregates must adopt a distinct outer membrane composition.

QS-related communication

Our data set only identified significant differential expression of one QS system, when Pa exists as aggregates—PQS. PA0998 (pqsC) is significantly downregulated −7-fold in aggregates at 8 hours. The remaining PQS operon is also downregulated, although at lesser levels (Fig. 1). Despite low expression levels at all timepoints, the transposon mutant of pqsC shows a severe growth deficiency in SCFM2. These data indicate that tight regulation of the PQS quorum sensing system is integral for aggregate formation. pqsABCD are needed for the synthesis of the PQS signal molecules (32). In the disruption of pqsC, Pa cannot form WT aggregates in a CF-like environment—timelapse microscopy reveals sparse populations composed mostly of planktonic cells (Fig. 2a). The PQS system has been documented as critical for several components of biofilm formation, including iron chelation and induction of the oxidative stress response (33, 34). Gene expression patterns in our data also point to the importance of these two functions in the formation and maintenance of aggregates.

Co-culture with the WT PAO1 results in further spatial heterogeneity

We co-cultured aggregate mutants (expressing GFP) with the WT PAO1 (expressing mCherry) in a 1:1 ratio in SCFM2. Using high-resolution microscopy, we observed developing populations to assess three things: (1) the ability to restore the WT aggregate phenotype, (2) whether any mutant can outcompete the WT, and (3) whether the addition of WT cells influences the spatial organization of mutants. We found that the growth of fully functional WT cells with aggregate mutants resulted in changes in observed spatial organization (Fig. 4) as well as in total biomass, average aggregate volume, and volume of planktonic cells (Fig. 5).

Fig 4.

WT and transposon mutant co-culture. Confocal laser scanning microscopy of WT PAO1 (green) and Tn-mutants (red) co-cultured in SCFM2 at a 1:1 ratio after 5 hours. Images are representative of each mix of which there were three biological replicates. The scale bar is 5 µm.

Fig 5.

Comparisons of WT (solid bars) and transposon mutant (striped bars) (a) total biomass (b) average aggregate volume, and total volume of planktonic cells after 15 hours growth in SCFM2. Mutants are grouped by phenotype. Three biological replicates ± SEM, significance of WT vs. mutants was calculated using ordinary two-way ANOVA (P value < 0.0001) with Fischer’s LSD multiple comparisons test (alpha 0.05).

We found that mutants PA4782, PA1864, PA0435, and 2207 restored “WT-like” aggregation by volume when mixed with the WT. However, strains did not maintain separate clonal populations. Instead, we observed extensive mixing of strains. By contrast, PA0646 resulted in WT aggregate phenotypes, though strains maintained separate clonal populations. Strains did not form mixed structures and underwent separate dispersal events (Fig. 4). This is captured in PA0646, with higher overall biomass and aggregate volume at 15 hours, shortly after a WT dispersal event (Fig. 5). PA4782 showed no significant difference in any measure compared to the WT. PA1864 and PA0435 exhibited significant differences in both aggregate volume and the number of planktonic cells, with mutant strains producing smaller aggregates than WT.

PA3160 and PA3009 mixtures also revealed strains maintaining distinct clonal populations but did not restore WT-sized aggregates. Although there were no significant differences between strains, these clonal populations formed extensive lawns of cells with distinct layers (Fig. 4). Finally, multiple combinations retained the original mutant phenotype. Specifically, PA1531, PA1530, PA 2111, PA5102, PA0998, and PA3406 mutants retained either an entropic or impaired aggregate structure. Interestingly, several mutants like PA3406 and PA5102 were able to incorporate WT cells into mixed entropic structures (Fig. 4).

From the opposite perspective, in many cases, WT populations exhibited significant variation in total biomass, aggregate size, and planktonic populations between co-cultures (Fig. S4). For example, when the PA1531 mutant is cultured alone, it forms significantly smaller aggregates and has a reduced total biomass and planktonic population when compared to the WT. Conversely, when mixed, WT PAO1 exhibits significantly larger phenotypes when compared to its growth in the presence of other mutants (Fig. S4). When we compared each complemented mutant with the WT PAO1, at both 4 (Fig. S2c) and 7 hours (Fig. S2d), there was no significant difference in the average aggregate size between the WT and any of the complemented mutants. This provides strong evidence that gene restoration reverts the impaired and entropic phenotypes to WT aggregate formation at these time points. These findings strongly suggest that aggregation mutations can be “rescued” in some cases. However, they also suggest that aggregation requires multiple biological processes, likely under precise temporal control.

DISCUSSION

We hypothesized that Pa aggregate formation requires a unique subset of genes when compared to much larger mm-scale biofilms. We have previously shown that the presence of mucin is a requirement for aggregation in our chronic Pa infection model of synthetic CF sputum (SCFM2) (1, 35, 36). By contrast, there is evidence that mucin promotes dispersal in Pa biofilms in an acute infection model, highlighting its potential importance in mediating the aggregate-planktonic lifestyle switch (37). Here, we leveraged this as a tool to control aggregation and study the differences between planktonic cells and aggregates of Pa in SCFM2.

Polysaccharides may not be required for early Pa aggregate formation

Our previous studies have identified the ability of Pa to form aggregates in the absence of Pel, Psl, or alginate (1). It is important to note that previous experiments provided qualitative data, simply examining the ability of PAO1 lacking polysaccharides to form aggregates. Here, using RNA-seq to compare planktonic cells with aggregates, we did not observe significant differential expression of the genes pel, psl, or alg and their associated pathways. Neither did we see significant dysregulation within the associated QS pathways Las or Rhl. We did however observe downregulation of the ncRNAs (such as rsmA) and their targets in aggregates (Fig. 3). This suggests that regulation of polysaccharides in aggregates may occur post-transcriptionally (as shown previously for Psl in larger biofilms (18)). Further studies at a higher temporal resolution are a necessary next step to understand this further. Our findings point to shifts in several metabolic pathways required for the switch from planktonic cells to aggregates in SCFM2 instead of a primary requirement for exopolysaccharides. However, many of these pathways have not been previously linked to aggregates and are not well defined, therefore revealing novel insights into the metabolic needs of aggregates (Fig. 1c and 3).

Aggregation requires temporal regulation of specific metabolic genes

Each of the 13 genes prioritized in this study demonstrates changes in expression between time periods of 2, 4, and 8 hours. Genes that are upregulated in aggregates at 8 hours, such as PA5102 and PA2207, may be required to adopt the slower metabolism and doubling time of aggregated cells. Mutants of these genes form entropic aggregates with rapid growth rates and large populations of planktonic cells. Disrupting genes that are downregulated after initial aggregation occurs (>4 hours) also results in different mutant phenotypes, despite their perceived unimportance at this stage. This suggests a role in maintaining the aggregate once it has formed. Hypothetical proteins, PA2111, PA3009, and PA4782, are upregulated at earlier time points. Mutants of these genes are unable to form WT aggregates, indicating early expression of these genes is required for aggregate formation. PA2111 is predicted to be a subunit of an allophanate hydrolase, part of a complex that converts allophanate into CO₂ and ammonium, which eventually feeds into the urea cycle. Expression of this gene in SCFM2 cycles over time, peaking at 4 hours and decreasing to −10-fold downregulation at 8 hours (Fig. 1c; Table 1). This is evidence of increased nitrogen levels, possibly attributed to amino acid catabolism during initial aggregate growth at 4 hours, creating urea as a waste product. This expression pattern is specific to aggregate growth in SCFM2, where amino acids are the primary carbon source. In LB, expression of PA2111 peaks later at 8 hours (Fig. 1c). Studying gene expression patterns in this way provides new insight into aggregate metabolism and the impact of the microenvironment over time.

Amino acids are abundant in SCFM2 (and actual CF sputum) (14). Accordingly, several highly upregulated genes in aggregates (PA1530, PA1531, PA2207, and PA0435) form part of an amino acid importer (PA1530, PA1531, PA0435), or can feed directly into the TCA cycle (PA2207). PA1531 is consistently upregulated in SCFM2 + mucin samples, exhibiting >2,000-fold upregulation compared to planktonic samples (Fig. 1). Our computational pipeline predicts this hypothetical protein is the SBP of an ABC transporter, with specificity for arginine, lysine, ornithine, and glutamine, while PA0435 is a predicted proline/sodium symporter. We also see upregulation in SCFM + mucin of two probable amino acid aminotransferases (PA0870 and PA3139) that could facilitate this reaction (Fig. 1c). PA2207 (tctA) is the transmembrane symport protein of a tripartite tricarboxylic transporter (TTT), which are not well characterized but have been shown to facilitate citrate uptake in several organisms (Fig. 3). In parallel, PA5102 is a predicted fatty acid desaturase associated with carnitine/glycine betaine metabolism and is upregulated in aggregates at 8 hours. This is preceded by the downregulation of repressor betI at 4 hours. We do however observe a cognate upregulation of dhcABR, which aligns with STRING co-expression predictions and is required in the final steps of the overall conversion of eukaryotic SFAs to glycine (38)(Fig. 1c and 3). The dchAB genes have been linked to the metabolism of many small molecules, including other amino acids such as lysine (39). It is important to note that the use of porcine mucin in SCFM2 provides an abundant source of carnitine in this environment. Overall, the involvement of BetI indicates a role in choline metabolism and associated osmo-protection in Pa as aggregates. Previous studies have demonstrated the importance of choline metabolism and osmotic stress response during Pa infection (38, 40). Specifically providing a fitness advantage to Pa in murine lung infection models and Pa isolates from people with CF, respectively. This also highlights the different adaptive measures adopted by Pa as aggregates compared to planktonic cells. Interestingly, this pathway is particularly important in the context of infections, such as in the lungs of people with CF (41).

PA2207, PA0435, and PA5102 are entropic mutants with prolific replication rates (Fig. S1). Evidence that repression of these highly expressed genes results in robust growth tells us several things about aggregate metabolism and its links to aggregate formation. First, the difference in growth rate between entropic and impaired mutants suggests that these spatial phenotypes are metabolically diverse (Fig. S1). Second, that these mutations are not lethal suggests that Pa can compensate for their lack of function. Redundancy in the genome could explain this, where several homologs could be co-opted to perform some of these functions.

Some QS systems are redundant during aggregation

While the Las and Rhl QS systems are not significantly DE across our samples, our data suggest that tight regulation of the PQS QS system is integral for aggregate formation. PA0998 (pqsC) is significantly downregulated by more than sevenfold when compared to planktonic cells at 8 hours (Fig. 1). Despite low expression levels at all timepoints, the transposon mutant pqsC has a severe growth deficiency in SCFM2. PqsABCD is required for the synthesis of the PQS signal molecules (32). Here, we show that in the absence of PqsC, Pa cannot form WT aggregates in a CF-like environment, timelapse microscopy reveals sparse populations composed mostly of planktonic cells (Fig. 2). To our knowledge, this is the first evidence of explicit link between the ability of Pa to produce PQS and aggregate formation. The PQS system is well described as important for several components of biofilm formation, including iron chelation and the induction of oxidative stress response (33, 34). Gene expression patterns in our data also point to the importance of these two functions in the formation and maintenance of aggregates. We observed ~40-fold upregulation of PA3406 (hasD) in aggregates at 8 hours, accompanied by upregulation of the rest of the hasAP heme acquisition and pch pyochelin operons (Fig. 1). Interrupting the HasAP secretion complex (and therefore heme acquisition capacity) in hasD transposon mutants leads to a growth delay and a loosely packed, impaired biofilm phenotype with a large planktonic population (Fig. 2). Circumventing several steps in the TCA cycle, glyoxylate shunting allows for the conversion of isocitrate to malate and is often observed in Pa isolates from human infections. Here in SCFM2 aggregates, we see an upregulation of glyoxylate oxidase subunits glcDE (42) (Fig. 1 and 3).

One of our most interesting findings was that instead of an expected binary phenotype of aggregation or no aggregation, we identified several intermediate spatial structures between planktonic growth and WT aggregates. This suggests that failure to meet any one of the required pathways has the potential to inhibit WT aggregation to varying degrees. Approaching the study of aggregate spatial structure as a continuum instead of binary provides a platform to study how changes in the microenvironment shape spatial structure. Due to the nature of Pa’s sociality, it is likely that spatial organization within aggregates can impact the ability to share public goods, the use of gradient-driven communication, and antibiotic tolerance. For example, the loose packing in impaired aggregates would likely require public goods and signal molecules to travel farther, possibly exceeding their effective distance (8). The tight stacking in entropic mutants leaves little, if any, space between cells, reducing the distance that public goods and signals must diffuse to be shared (Fig. 2 and 4). As a result, interactions that result from direct contact are more likely.

We decided to explore this further, by mixing fully functional WT cells with aggregate mutants. We asked if the spatial organization of mutants could be reverted to WT when cultured in SCFM2 (Fig. 4 and 5). Combinations of strains resulted in further shifts in spatial organization. For example, entropic aggregators PA4782 and PA2111 (labeled red) incorporated WT cells (shown in green) in two different ways. PA4782 and WT displayed a close-to-normal aggregate phenotype (as determined by average aggregate volume); however, aggregates were not clonal and contained a mixture of cells from both strains. PA2111 retained an entropic phenotype but was able to integrate WT cells into its distinct “raft-like” formation. Mutant PA1864 alone is characterized as an impaired aggregator; however, the addition of WT cells produced aggregates similar in volume to WT. These preliminary findings of the WT’s ability to influence the spatial organization of a mutant imply the occurrence of cooperative interactions during growth. Interestingly, during co-culture, entropic aggregates (PA4782 and PA2111) behave differently from each other. This raises many interesting questions about the combination of biological systems required for aggregation, specifically how the disruption of genes impacts physiology that relates to interactions with other cells. On a broader scale, the evolving accessibility of Pa cells within different formations to secreted factors and contact-dependent interactions may significantly influence the overall functionality of an aggregate (as seen previously (8)).

Conclusions

We provide evidence that amino acid uptake and metabolism, PQS, iron scavenging, oxidative stress response, and LPS modification play fundamental roles in the process of Pa aggregation. The absence of WT aggregation and the existence of multiple spatial phenotypes when these systems are disrupted supports this. Importantly, these mutations are not lethal and there is still much to be learned about aggregate metabolism, associated pathways, and their contribution to aggregation. The ability to temporally regulate aggregate genes is critical, where genes and operons need to be expressed at a particular time or growth phase to facilitate later functions (PA5102, PA1864) that are required for WT aggregation. This result is especially important when considering using aggregation genes as potential therapeutic targets. At large, although are results relate directly to a model of chronic infection, we believe these data enhance our understanding of aggregate biology. Lewin et al. developed a framework to assess model accuracy by comparing gene expression between models and natural environments. All 13 genes, except PA0646, PA3406, and PA3160, were similarly expressed in synthetic CF sputum (SCFM2) compared to natural CF sputum (43). This supports the further study of these genes in relation to aggregation, allowing future studies to explore their role in more complex and infection-like environments. In relation to other Pa strains, there is much scope for the exploration of how and if aggregation-related genes are conserved in different niches. Turner et al. used transposon sequencing (Tn-seq) to identify genes essential for PA14 virulence in murine infection models. Among the genes we have identified here, PA1864 and PA3009 were required for PA14 to establish infection in a murine chronic wound model. Pathway analysis predicted similar regulatory pathways for these genes, consistent with those identified in our mutants (35). This finding warrants further investigation into the contribution of these genes in other infection sites, specifically with a focus on clinical isolates. At a larger scale, unique requirements for aggregation could be used to inform better applications of existing therapeutics or the development of new ones against other aggregate-forming bacteria.

MATERIALS AND METHODS

Strains, media, and culture techniques

Pseudomonas aeruginosa PAO1 wild type::gfp (containing plasmid pMRP9-1 (17)) was cultured in standard lab media (LB) from frozen stock overnight at 37°C with shaking (200 rpm). Cells were back diluted 1:20 in fresh media and grown to log-phase ~3 hours, then washed with PBS (pH7.0) before inoculation. Pa cells were inoculated at approximately 10⁵ cells per mL and incubated without shaking at 37°C as previously described (1). Transposon mutants were obtained from the University of Washington and confirmed by Illumina sequencing (44). Strains were grown in LB (Gm50) and stocked at −80°C. We transformed Tn mutants were transformed with the fluorescent protein-expressing plasmid pMP7605::mCherry (45). SCFM2 + mucin (Porcine mucin, 250 mg) media was prepared as previously described (46). SCFM2 – mucin media was prepared identically with the exclusion of mucin. LB + mucin was prepared by adding mucin to sterile LB at the same concentration as SCFM2 + mucin media. LB –mucin consisted of sterile LB only.

Comparison of aggregate and planktonic cultures and RNA extraction

After a period of 2, 4, or 8 hours, cultures were pelleted and re-suspended in DNA/RNA Shield (Zymo Research). These time points were chosen to represent different stages of aggregation (as described in reference (1)). Samples were flash-frozen and stored at −80°C. The 0-hour timepoint was inoculated at a calculated OD₆₀₀ equivalent of 8 hours of growth for both LB and SCFM2 media, 0.53 and 1, respectively. This acted as a density control. After inoculation samples were grown statically at 37°C for 15 minutes and then harvested and frozen identically to other time points.

Cell pellets were thawed on ice and 100 µL of RNase/DNase-free water was added. Samples were transferred to bead beater tubes and 600 µL of Qiagen lysis buffer was added. Samples were bead beat at RT for 5 min, and then RNA was extracted using the Qiagen RNeasy Mini kit and eluted in molecular-grade water. RNA was stored at −80°C, and quantity and quality were measured using Qubit RNA High Sensitivity and RNA IQ kits (Thermofisher). Multiple technical replicates were preformed and pooled until there was at least 4 µg of total RNA for each biological replicate. To note, RNA was not recoverable from SCFM2 –mucin samples before 8 hours due to a lag phase in these conditions.

RNA sequencing (RNA-seq) and DE analysis

Illumina RNA sequencing was performed by Novogene. FastQC was used to generate quality reports of the raw RNAseq FastQ reads. The data were used in MultiQC to make one comprehensive quality control report. Qiagen CLC Workbench was used for differential expression analysis. FastQ files generated by Novogene were imported into CLC Workbench as paired-end reads with the following settings: minimum distance = 1, maximum distance = 1,000, Illumina options = remove failed reads, join reads from different lanes. The reference annotated gene track was created using reference genome PAO1_187 (NCBI) and reads were mapped using default mapping options. Reads were also mapped to annotated rRNA using default options; these maps were excluded from differential expression analysis. The remaining unmapped reads were run through the CLC Workbench Differential Gene Expression Analysis workflow. For DE comparisons between time points within a single condition, the following settings were used: analysis was run in Batch, define batch units using metadata column: “Condition,” test differential expression due to “Timepoint,” select “All group pairs for the Comparisons.” For DE comparisons between different conditions at one time point, the following options were used: analysis was run in Batch, define batch units using metadata column: “Timepoint,” test differential expression due to “Condition,” select “All group pairs for the Comparisons.” The resulting tables of differentially expressed genes were filtered by fold change ≥2 and False Discovery Rate (FDR) P-value ≦ 0.05. Relevant data sets can be found at NCBI with the BioProject ID PRJNA1168577.

Assessment of P. aeruginosa Tn-mutant aggregation

The initial determination of the ability of Pa to aggregate involved the top 50 significant upegulated or downregulated genes. Transposon mutants were obtained from the PAO1-ordered transposon mutant library (University of Washington (44)). Each of the 50 mutants was inoculated into 500 µL SCFM2 + mucin at 37°C. Cultures were observed using phase contrast microscopy at 4 and 8 hours for signs of aggregation. We identified 13 mutants with significantly different aggregation compared to the WT PAO1, prioritizing these mutants for further investigation. Each of the 13 mutants was transformed with the fluorescent protein-expressing plasmid pMP7605::mCherry (45). For mixed strain experiments, WT PAO1 and mutants were inoculated 1:1. The Pa mutants PA0998, PA1531, PA1864, PA3406, and PA5102 were complemented by electroporation with the pET21b + plasmid (GenScript) containing the relevant gene sequence. Transformants were selected on ampicillin-containing agar plates and cultured in SCFM2 as above. All images were acquired with the Zeiss LSM 880 confocal laser scanning microscope utilizing Zen image capture software. Bacterial cells were visualized via mCherry (excitation wavelength of 587 nm/emission wavelength of 610 nm) or GFP (excitation wavelength of 489 nm/emission wavelength of 508 nm) with a 63× oil immersion objective. SCFM2 images were acquired by producing 512- by 512-pixel (0.26- by 0.26-µm pixels) 8-bit z-stack images that were 60 µm from the base of the coverslip. The total volume of the 60 µm z-stack images were 1093.5 µm³. Control images of uninoculated SCFM2 were acquired using identical settings to determine the background fluorescence for image analysis. Microscopy files were exported in CZI format.

Image analysis

Image analysis. All imaging was performed with identical image capture settings. To determine the background fluorescence in SCFM2 as previously described (1, 8, 46). For aggregate studies in SCFM2, isosurfaces were produced for all remaining voxels after background subtraction with the surpass module in Imaris as previously described (1). All image data were exported as a CSV and imported into R. Here, detected aggregate isosurfaces were then ordered by volume. Objects that were >0.5 and <5.0 µm³ were categorized as dispersed biomass, and objects that were >5.0 µm³ were categorized as aggregated biomass. The total dispersed biomass was calculated as the sum of all the dispersed objects. Graphs were generated with GraphPad Prism 8.

Computational modeling

Since many pathway mapping programs struggle to accommodate hypothetical proteins, we used a predictive modeling pipeline to estimate the functions and pathways of these critically important genes (Fig. 2a). Protein sequences of identified hypothetical proteins were retrieved from the Pseudomonas Genome Database (47). Sequence homology to other proteins was determined using HMMER (48). Structural homology was queried using both Dali (https://ekhidna2.biocenter.helsinki.fi/dali/) and SWISS-Model (49). This structural homology information was used to predict protein structures using AlphaFold2 ColabFold v1.5.2 (50). Protein-protein interactions were mapped using STRING pathway analysis, which utilizes information from experiments, as well as large omics data sets (51). This information taken in conjunction with the predicted function allowed us to postulate mechanisms and pathways that these unannotated proteins are likely to be involved in (Table 1).

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/jb.00429-24.

Supplemental materials. jb.00429-24-s0001.pdf.

Fig. S1 to S5, Table S2, and figure legends.

Supplemental table. jb.00429-24-s0002.xlsx.

Table S1

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.