Comparison of Galleria mellonella, epithelial cell cytotoxicity, and mouse model of bacteremia to measure Pseudomonas aeruginosa virulence

ABSTRACT

Considerable effort has focused on identifying alternatives to mouse models in research studies. In the field of bacterial pathogenesis, Galleria mellonella and epithelial cell lines have been widely used for this purpose, but the concordance of these models with mice remains unclear. To begin to address this knowledge gap, we used 105 clinical isolates of Pseudomonas aeruginosa for which virulence had previously been determined in a mouse bacteremia model. A semistrong correlation was observed between G. mellonella median time to 50% mortality and mouse 50% pre-lethal dose (LD₅₀) values (Spearman’s rank correlation coefficient [ρ] = 0.75), whereas percent A549 epithelial-like cell lysis during co-culture showed a weak correlation to mouse LD₅₀ values (ρ=−0.47). Given the stronger correlation between G. mellonella and mouse virulence, we next examined whether G. mellonella could substitute for mice when asking questions about the virulence of large numbers of P. aeruginosa isolates. Results from mice indicated that isolates with resistance to more antibiotics were significantly less virulent, and the use of G. mellonella identified the same inverse correlation. Furthermore, both models found no evidence for the existence of hypervirulent clonal lineages. In particular, isolates belonging to sequence types defined as high-risk clones were not consistently more virulent than other isolates, despite the known association of high-risk clones with poor clinical outcomes. These findings suggest that G. mellonella can serve as an adequate substitute for mice when addressing specific population-based virulence questions, although conclusions should be confirmed in mice.

INTRODUCTION

Mouse infection models are arguably the gold standard for studying the virulence of bacteria such as Pseudomonas aeruginosa (1) and have many advantages, such as human-like innate and adaptive immune systems, genetically modifiable genomes, and an extensive body of comparative literature. However, mouse infection models are costly, highly regulated, and ethically controversial, which limits their utility in high-throughput assays. Galleria mellonella and mammalian cell culture models circumvent many of these disadvantages (2, 3). Both may be maintained at 37°C, the mammalian body temperature at which some P. aeruginosa virulence genes are upregulated (4). Cell co-culture assays have the added advantage of allowing for use of human cell lines, whereas G. mellonella harbor hemocytes that function similarly to mammalian neutrophils (2, 5), the immune cell primarily responsible for containing P. aeruginosa infections in mice and people (6–9). As a result, both G. mellonella and cell culture models have been widely used to quantify P. aeruginosa strain virulence (10–18), and results of these models have been shown to correlate with each other (19). However, the degree to which these models mirror P. aeruginosa infection outcomes in mice has been examined in only a few small studies (10–14, 18).

Most P. aeruginosa isolates fall into one of two genetically distinct clades: Group A and Group B (20, 21). These clades are largely distinguished by the presence of either the exoS or exoU gene, respectively, which encode type III secretion system (T3SS) effectors. The presence of the exoU gene and secretion of its corresponding effector protein are associated with more severe disease (15, 22–25). Within each of these large clades, subclades of multilocus sequence types (STs) exist, which are defined as isolates with the same alleles across seven housekeeping genes (26). Some STs have been designated as high-risk clones (HRCs) because of their global distribution, the disproportionate number of infections they cause, their high degree of antibiotic resistance, and their association with poor clinical outcomes (27–29). Even within a single P. aeruginosa ST, substantial isolate-to-isolate genomic variability exists, which drives phenotypic differences in virulence and antibiotic resistance (22, 23, 30–33). Some antibiotic resistance mutations confer fitness costs that reduce virulence (34–38), whereas others do not (38–40). Clarifying strain-to-strain differences in virulence and antibiotic resistance and the interplay between the two is crucial to a deeper understanding of the pathogenesis of P. aeruginosa infections.

In this study, we used 105 P. aeruginosa clinical isolates to examine how virulence measurements from the G. mellonella and mammalian cell co-culture models compare to those from mice. Results from G. mellonella showed a semistrong correlation with those from the mouse model, whereas cell cytotoxicity results correlated less well. We then used the G. mellonella and mouse results to examine the relationships between P. aeruginosa virulence and antibiotic resistance and whether hypervirulent clades of P. aeruginosa exist.

MATERIALS AND METHODS

P. aeruginosa isolates and culture conditions

A description of the P. aeruginosa isolates used in this study has been previously published (32, 41). Seventy-one isolates (“PABL”) were obtained at Northwestern Memorial Hospital in Chicago, IL, USA, from 1999 to 2003 from adults with P. aeruginosa bacteremia (42). Fifteen isolates (“PAC/S”) were obtained from the blood or peritoneal fluid of pediatric patients with Shanghai fever at Chang Gung Children’s Hospital in Taiwan from 2003 to 2008 (43). Nineteen isolates (“PASP”) were collected from patients with bacteremia in Spain between 2008 and 2009 (44). Bacteria were routinely cultured in lysogeny broth (LB) at 37°C with shaking at 250 rpm or on LB agar plates.

Whole-genome sequencing and phylogenetic analysis

The whole-genome sequence of each isolate has been previously published (32, 41, 45). Multilocus sequence typing (MLST) was performed using the P. aeruginosa PubMLST website (https://pubmlst.org/paeruginosa/) (46).

Mouse infection model

A mouse model of bacteremia using tail vein injection of bacteria had been previously used to estimate the virulence of each P. aeruginosa isolate (32, 41). Briefly, each isolate was tested at a minimum of two doses, with 3–5 mice per dose (≥9 mice per isolate). Female 6- to 9-week-old BALB/c mice were used for all experiments. The bacterial CFU required to cause pre-lethal illness in 50% of mice (LD₅₀) following tail vein injection was calculated. Mice meeting pre-defined criteria (25) for pre-lethal illness were euthanized and scored as dead. For most of the isolates, LD₅₀ values were obtained and reformatted from Table S3 of Pincus et al. (32), distributed under the terms of the Creative Commons Attribution 4.0 International license (https://creativecommons.org/licenses/by/4.0/). For PABL019, the mouse LD₅₀ was obtained from Allen et al. (41).

G. mellonella infection model

G. mellonella were infected as previously described (10). Briefly, a series of doses was prepared for each isolate by centrifugation of bacterial cultures, washing of the pellet with phosphate-buffered saline (PBS), and suspension and dilution with PBS. Bacterial inocula were enumerated by plating onto lysogeny broth (LB) agar. An autoinjector was used to inject each dose into the last proleg of 10 larvae weighing between 200 and 350 mg. Larvae were incubated at 37°C and monitored until 100% mortality was obtained; the absence of movement of the larvae following shaking was used to determine mortality. For each defined bacterial dose that was injected into G. mellonella larvae, a 50% lethal time (LT₅₀) was calculated, at which half the infected larvae had died (Fig. S1A). This process was then repeated with a range of doses. The natural logarithms of these doses were plotted against the corresponding LT₅₀ values, and a linear regression was performed to establish the ln(CFU) vs LT₅₀ relationship (Fig. S1B). The time to death of 50% of injected larvae (LT₅₀) was determined with a custom R Script (https://github.com/ChrisAxline/Galleria_Code). This process was repeated for all isolates (Data S1). Each set of experiments included a PBS-only negative control. For each isolate, a minimum of two biological replicates were performed on different days by different researchers, spanning 5–28 doses tested per isolate. To compare isolate virulence, regression equations were used to predict the LT₅₀ values of each isolate at a dose of 3,000 CFU, the approximate median dose across all G. mellonella infections. This approach circumvents the need to experimentally prepare inocula of 3,000 CFU with high accuracy, which can be challenging for such a small dose and with a large collection of diverse isolates. Comparisons were also made at calculated doses of 300 and 30,000 CFU.

Cytotoxicity assay

A549 cells (ATCC) were grown at 37°C in Dulbecco’s Modified Eagle Medium (Corning) supplemented with 10% fetal bovine serum (Corning) under 5% CO₂ conditions. Cytotoxicity was quantified by measuring lactate dehydrogenase (LDH) release at 3 hours post-infection using the CyQUANT LDH Cytotoxicity Assay (Invitrogen) per the manufacturer’s instructions. Briefly, A549 cells were grown as a monolayer in 96-well plates for 22 hours, after which the culture medium was replaced with RPMI 1640 medium without phenol red (Gibco). Overnight bacterial cultures were grown in 5 mL of LB medium, and 500 μL of the overnight culture was pelleted, washed with PBS, and resuspended in 1 mL of PBS. A PA01 OD₆₀₀-CFU curve was used as a reference to further dilute the bacterial suspension to achieve an MOI of ~20. The plate was spun down at 500 × g for 5 min and then incubated for 3 hours at 37°C under 5% CO₂ conditions. After the incubation period, the supernatant was collected to perform the LDH assay. The percentage of cells lysed was calculated relative to the positive control obtained following the addition of 1% Triton X-100 to uninfected cells. Uninfected cells alone were used as a negative control and to normalize for background LDH release. For each isolate, three biological replicates were performed, each with three technical replicates. Results for PABL053, PABL056, and PABL063 were excluded from the final analyses due to an unusually high degree of variability between replicates.

Immunoblot analysis

For detection of in vitro secreted ExoU and ExoS proteins, subcultures were performed in LB medium supplemented with 20 mM MgCl₂ and 5 mM EGTA and were incubated at 37°C for 3 hours with shaking at 250 rpm. Proteins secreted into the culture supernatant were passed through a 0.2 μm filter and precipitated from solution by overnight incubation with 2 mL of 100% trichloroacetic acid (TCA) at 4°C with slow shaking. The precipitate was collected by centrifugation at 8,671 × g for 45 min at 4°C and resuspended and washed three times with 1 mL of ice-cold 100% acetone. The washed protein pellets were dried at 60°C and solubilized in Laemmli sample buffer (Bio-Rad). Samples were heated at 80°C for 10 min and then separated on a 4%–15% precast polyacrylamide gel (Bio-Rad). Proteins were transferred to a nitrocellulose membrane (Bio-Rad) by electrotransfer and blocked in 5% milk powder (Boston BioProducts) in 1× Tris-buffered saline (Bio-Rad) with 0.5% Tween-20 (National Diagnostics). Blots were incubated at room temperature for 1 hour with rabbit polyclonal antisera against ExoU or ExoS/T (25). Bound antibodies were detected using IRDye 680LT goat anti-rabbit secondary antibody (LI-COR) and visualized with the Odyssey M imaging system (LI-COR). Immunoblot results for the PABL set of isolates were previously published (41).

Antibiotic susceptibility

Minimum inhibitory concentrations (MICs) of antibiotics were determined using microbroth dilution assays as previously described (47). Each isolate was tested against six agents: tobramycin (aminoglycoside), piperacillin/tazobactam (penicillin/β-lactamase inhibitor), aztreonam (monobactam), cefepime (cephalosporin), ciprofloxacin (fluoroquinolone), and meropenem (carbapenem). Antibiotic resistance was determined as non-susceptible (i.e., MICs in the intermediate or resistant range) or susceptible according to the Clinical and Laboratory Standards Institute 2024 breakpoints. Isolates were ranked from 0 (susceptible to all 6 antibiotics) to 6 (non-susceptible to all 6 antibiotics). All MIC tests were performed in duplicate; if the duplicates disagreed, a third replicate was performed.

Phylogenetic and cluster analysis

Core genome phylogenetic analysis was performed by aligning sequencing reads to the PA01 reference sequence using the BWA MEM algorithm in BWA v0.7.18 (48). Consensus sequences were generated with bcftools v1.20 and filtered to core positions present in 95% of isolates (49). The phylogenetic tree was generated using IQ-TREE 2 with the ModelFinder algorithm (50). The tree was midpoint-rooted and visualized with superimposed metadata using ggtree v3.16.3 in R v4.5.0 (51, 52). Cluster generation for hypervirulence analysis was performed using TreeCluster v1.0.4 with the Max Clade method and threshold 0.1 (53). Clusters were overlaid onto an unrooted version of the core genome phylogenetic tree with ggtree.

Statistical analyses

All statistical analyses were performed using R (v4.5.1) or GraphPad Prism (v10.2.3). The statistical tests used are indicated in the corresponding figure legends. P ≤ 0.05 was considered significant. Spearman’s rank correlation coefficient (ρ) strength was interpreted using previously published references: weak (|ρ| < 0.5), moderate (0.5 ≤ |ρ| < 0.75), semistrong (0.75 ≤ |ρ| < 0.85), and strong (|ρ| ≥0.85) (54).

RESULTS

Characterization of a collection of 105 P. aeruginosa clinical isolates

To compare the three models of infection, we utilized a collection of previously sequenced P. aeruginosa clinical isolates cultured from patients in the United States, Spain, and Taiwan (Fig. 1). These isolates were genetically diverse, consisting of 59 different STs (Data S1). Both the Group A and Group B clades of the P. aeruginosa population structure were represented, with 64 isolates containing only the exoS gene (exoS⁺) and 38 containing only the exoU gene (exoU⁺). PABL031 contained both effector genes (exoS⁺/exoU⁺), while PABL071 and PABL043 contained neither (exoS⁻/exoU⁻). Immunoblot analyses indicated all but 20 isolates secreted detectable amounts of the corresponding T3SS effector protein under inducing in vitro growth conditions. Of these non-secreting isolates (ExoS⁻/ExoU⁻), 14 were exoS⁺, 4 were exoU⁺, and 2 were exoS⁻/exoU⁻ (Data S1). These findings show that the collection of P. aeruginosa isolates is phylogenetically diverse and represents a variety of type III effector genotypes and secretion phenotypes.

Fig 1.

Phylogenetic and phenotypic diversity of P. aeruginosa clinical isolates. A maximum-likelihood core genome phylogenetic tree is shown, with a concentric ring indicating isolate geographic source. The presence of the exoS or exoU genes is indicated by the font color of the isolate name, and in vitro secretion of T3SS effectors ExoS and ExoU is indicated by tips with filled circles (secretion-positive) or open circles (secretion-negative). PABL043 (a PA7-like strain) is excluded from this tree because it is genetically highly divergent from the other isolates.

Virulence in a mouse model of bacteremia

All isolates had been previously assessed for virulence in mice by measuring the 50% pre-lethal dose (LD₅₀) in a tail vein injection bacteremia model (32, 41). These results are reproduced in Fig. 2. The isolates varied markedly in their virulence, with LD₅₀ values ranging over more than three logs (1,000-fold), with a median LD₅₀ value of 7.4 × 10⁶ (10^6.9) CFU. PABL012, an ExoS⁺ isolate, was the most virulent, with an LD₅₀ value of 6.3 × 10⁵ (10^5.8) CFU (Data S1). PABL019, an ExoS⁻ isolate that was obtained from the bloodstream of a patient with cystic fibrosis (CF), had the highest LD₅₀ value of 1.3 × 10⁹ (10^9.1) CFU. Consistent with prior studies (15, 32, 41, 55–58), these findings demonstrate substantial differences in the capacity of individual P. aeruginosa isolates to cause severe infections. These isolates serve as a reference for measuring the performance of the G. mellonella and cell co-culture models of infection.

Fig 2.

Virulence of P. aeruginosa clinical isolates in mice. BALB/c mice were previously challenged intravenously by tail vein injection with varying doses of the 105 P. aeruginosa clinical isolates and monitored for the development of pre-lethal infections. Results were used to calculate LD₅₀ values for each isolate. T3SS effector genotype for each isolate is indicated by bar color. Bar fill indicates detection of in vitro effector secretion (solid; either ExoS⁺ or ExoU⁺) or lack thereof (open, ExoS⁻/ExoU⁻). For convenience, the LD₅₀ values for all isolates except PABL019 were obtained and reformatted from Table S3 of Pincus et al. (32), distributed under the terms of the Creative Commons Attribution 4.0 International license (https://creativecommons.org/licenses/by/4.0/). For PABL019, the mouse LD₅₀ value was obtained from Allen et al. (41).

Virulence in a G. mellonella infection model

We next quantified the virulence of each isolate in G. mellonella. Unlike other bacterial species, many P. aeruginosa strains kill G. mellonella at extremely low inocula, where LD₅₀ estimates become difficult to obtain (10). We therefore used time to death as a measure of virulence, as we have previously shown this approach provides a consistent and reliable measure of G. mellonella virulence (10). Isolate S7 was the most virulent, with an LT₅₀ value of 10.3 hours (Fig. 3). PABL019 was the least virulent isolate, with an LT₅₀ value of 35.6 hours. The median LT₅₀ value for all isolates was 12.6 hours. These results demonstrate substantial variability in the virulence of P. aeruginosa isolates in the G. mellonella model.

Fig 3.

Virulence of P. aeruginosa clinical isolates in G. mellonella. G. mellonella larvae were infected with each P. aeruginosa isolate, and virulence was quantified by extrapolating an LT₅₀ value for a dose of 3,000 CFU. T3SS effector genotype for each isolate is indicated by bar color. Bar fill indicates detection of in vitro effector secretion (solid; either ExoS⁺ or ExoU⁺) or lack thereof (open, ExoS⁻/ExoU⁻).

Virulence in a cell co-culture system

Next, each isolate was tested for its ability to lyse human cells. We chose A549 cells, which are a cell line of pulmonary epithelial-like cells, because the lungs are a common site of P. aeruginosa infections (59) and epithelial cells have been used extensively to characterize P. aeruginosa cytotoxicity (15, 60–62). Bacteria were co-incubated with A549 cells, and cytotoxicity was quantified after 3 hours, a time point commonly used to measure P. aeruginosa cytotoxicity (15, 63). Most of the isolates caused little or no cytotoxicity, but 32 (30%) lysed greater than 5% of the A549 cells (Fig. 4; Data S1). Among these 32 isolates, PABL072 showed the highest cytotoxicity (52%), and the average cytotoxicity for this subset of isolates was 33%. Thus, only a minority of isolates caused appreciable cytotoxicity under these conditions.

Fig 4.

Virulence of P. aeruginosa clinical isolates in an epithelial-like cell lysis model. A549 cells were co-incubated with each isolate for 3 hours, and LDH was measured as a proxy for cell lysis. Detergent-induced lysis served as the positive control for complete lysis. T3SS effector genotype for each isolate is indicated by bar color. Bar fill indicates detection of in vitro effector secretion (solid; either ExoS⁺ or ExoU⁺) or lack thereof (open, ExoS⁻/ExoU⁻). Bars indicate the mean of three biological replicates; error bars indicate standard deviations.

Comparison of G. mellonella and cell co-culture models to the mouse model

We compared the virulence measures of the 105 P. aeruginosa isolates in the G. mellonella and cell co-culture models to those of the mouse bacteremia model. To avoid the assumption of a linear relationship between virulence measurements across models, we used Spearman’s rank correlation coefficient (ρ) to quantify monotonic associations. Extrapolated LT₅₀ values for 3,000 CFU in G. mellonella correlated in the “semistrong” range when compared to mouse LD₅₀ values (ρ = 0.75, P < 0.0001, Fig. 5A). Similar correlations were observed using LT₅₀ values for 300 CFU (ρ = 0.75, P < 0.0001) and 30,000 CFU (ρ = 0.64, P < 0.0001) (Fig. S2). In contrast, comparison of cell cytotoxicity to mouse LD₅₀ values yielded a correlation coefficient in the “weak” range (ρ = −0.47, P < 0.0001, Fig. 5B). This was largely driven by the absence of appreciable cytotoxicity associated with two-thirds of the isolates, preventing this model from discriminating virulence differences among these isolates. Thus, G. mellonella was superior to the cell co-incubation model in measuring P. aeruginosa virulence when a mouse model of bacteremia was used as the gold standard.

Fig 5.

Concordance of G. mellonella or cell co-culture cytotoxicity with mouse virulence measurements. (A) The LT₅₀ values at an extrapolated dose of 3,000 CFU in G. mellonella are plotted against corresponding LD₅₀ values from the murine bacteremia model, yielding a Spearman’s rank correlation coefficient (ρ) of 0.75 (P < 0.0001). (B) A549 cytotoxicity values are plotted against corresponding LD₅₀ values from the murine bacteremia model, yielding a ρ of −0.47 (P < 0.0001).

One potential difficulty in comparing the three model systems is that their readouts differ in range. While mouse LD₅₀ values vary markedly, necessitating the use of a log scale, virulence values in G. mellonella and cell culture systems vary less and are suitable for a linear scale. To better facilitate comparisons across these ranges, isolates were ranked from highest (rank 1) to lowest (rank 105) virulence for each model. A heatmap was then generated to visually compare the relative virulence of isolates across models (Fig. 6). Isolates with a high mouse virulence rank (1–35) were significantly more likely to have a high G. mellonella virulence rank (Fisher’s exact test, P = 1.12 × 10⁻⁷), and isolates with a low virulence rank in mice (70–105) were also more likely to have a low virulence rank G. mellonella (P = 5.42 × 10⁻⁹). In contrast, isolates within the intermediate mouse virulence rank (36–69) were not significantly more likely to have an intermediate G. mellonella virulence rank (P = 0.12). For example, moderately virulent isolates in mice such as PABL078, PABL030, and PABL073 (ranks 49, 52, and 53, respectively) were among the most virulent in G. mellonella (ranks 2, 6, and 9, respectively). In contrast, cell co-culture rankings correlated less well with those of the mouse model. Although isolates with a high mouse virulence rank were significantly more likely to have a low cell survival rank (i.e., high cytotoxicity; P = 2.04 × 10⁻⁵), there were some highly virulent isolates (e.g., PABL012, S10, PABL049) that exhibited very little cytotoxicity. Isolates with low mouse virulence ranks did not show a significant association with high cell survival ranks (P = 0.051). These observations suggest that G. mellonella adequately identifies isolates that are at the extremes of virulence in mice, but that cell co-culture cytotoxicity measurements do so less well.

Fig 6.

Ranked comparison of the virulence of the P. aeruginosa isolates measured by the three infection models. Mouse LD₅₀ values (“Mouse”), G. mellonella LT₅₀ values at 3,000 CFU (“G. mellonella”), and A549 percent cell survival (derived from cytotoxicity measurements; “% Cell Survival”) are shown in a heatmap format. For each metric, isolates are ranked from 1 (most virulent by that metric) to 105 (least virulent), and tiles are colored by virulence rank (see scale). In general, when multiple isolates shared identical virulence values within the same infection model, isolates were assigned consecutive ranks in an arbitrary order. However, for “% Cell Survival” data, where many isolates exhibited 100% cell survival, all isolates with 100% survival were ranked “105.” Isolates are ordered based upon their virulence rank in mice. The adjacent annotation bar (“T3SS”) indicates each isolate’s T3SS effector genotype (by color) and secretion phenotype (by crosshatch pattern). In the text, an isolate that secretes neither ExoS nor ExoU is referred to as ExoS⁻/ExoU⁻.

The usefulness of G. mellonella to address population-based questions of P. aeruginosa virulence

Since G. mellonella infection outcomes had a semistrong correlation with those in mice, we next asked whether this insect model could be used to answer questions about P. aeruginosa pathogenesis. We used the “gold-standard” mouse virulence values to answer the question and then determined whether the G. mellonella model provided the same answer.

We first asked whether antibiotic-resistant isolates were less virulent than antibiotic-susceptible isolates. The MICs of six antibiotics, each representing a major class of antipseudomonal antibiotics, were determined for all isolates (Data S1). Isolates were then ranked from 0 to 6 based on the number of tested antibiotics to which they exhibited non-susceptibility. The most commonly observed non-susceptibility was to aztreonam (48%) and the least was meropenem (18%). As a group, isolates with higher ranks of antibiotic resistance exhibited significantly higher mouse LD₅₀ values (lower virulence) (Kendall’s rank correlation, τ = 0.20, P = 0.007) (Fig. 7A), confirming previous reports (34–38). This relationship between antibiotic non-susceptibility and virulence was also detected by the G. mellonella model (τ = 0.30, P = 4.9 × 10⁻⁵) (Fig. 7B). These findings suggest that the G. mellonella model is adequate to uncover relationships between virulence and antibiotic resistance.

Fig 7.

Relationship between antibiotic resistance and virulence of P. aeruginosa isolates in mice and G. mellonella. Each isolate was assigned an antibiotic resistance rank (0–6) based on the number of six tested antibiotics to which it was non-susceptible (i.e., intermediately susceptible or resistant). (A) Comparison of antibiotic resistance ranks to mouse LD₅₀ values. A modest positive association between antibiotic resistance levels and mouse LD₅₀ values was observed (Kendall’s rank correlation, τ = 0.20, P = 0.007). Kendall’s tau (τ) indicates the strength and direction of association between two variables (e.g., τ = 1 indicates a perfect positive association, and τ = 0 indicates no association). (B) Comparison of antibiotic resistance ranks to G. mellonella LT₅₀ values. A similar degree of association (τ = 0.30, P = 4.9 × 10⁻⁵) was noted. Each symbol represents an individual P. aeruginosa isolate. Colors represent T3SS genotypes. Filled circles represent isolates that secreted either ExoS or ExoU, and open circles represent isolates that secreted neither.

We next asked whether some lineages of P. aeruginosa were characterized by unusually high levels of virulence, similar to what has been observed with Staphylococcus aureus (64), Listeria monocytogenes (65), and Klebsiella pneumoniae (66). We first focused on HRCs due to their role in causing disproportionate numbers of infections. Currently, consensus criteria for defining lineages as HRCs have not yet emerged, and different definitions have been used (27, 28, 67). For this study, we defined P. aeruginosa HRCs as STs reported in the literature as “HRCs,” found on at least two continents, reported to be highly resistant to antibiotics, and more frequently cultured from patients than other P. aeruginosa lineages. To address the last criterion, we determined the relative abundance of each ST among P. aeruginosa genomes deposited in NCBI (Fig. S3). Using these criteria, we identified 10 STs that represented the most common HRCs present in our collection. A total of 32 isolates (30.5%) in our collection belonged to these 10 HRC STs. The virulence of these HRCs did not differ significantly from non-HRCs in either the mouse or G. mellonella infection model (Fig. 8A and B). These findings remained consistent when the analysis was restricted to the five most common HRCs in the NCBI repository (Fig. 8C and D). Interestingly, our analysis demonstrated that many of the lineages defined as HRCs in the literature contain a relatively low proportion of antibiotic-resistant isolates (Fig. 8). Eighteen of the 32 HRC isolates (56.2%) had a resistance rank of 0 or 1, while only 9 of 32 (28.1%) had a rank of 4–6. Together, these findings suggest that HRCs are no more or less virulent than non-HRC P. aeruginosa lineages, and that the G. mellonella model mimics the mouse model in demonstrating this relationship.

Fig 8.

Relationship between HRC status and virulence in mice and G. mellonella. Comparison of (A) mouse LD₅₀ values and (B) G. mellonella LT₅₀ values of isolates belonging to 10 HRC STs (235, 111, 274, 253, 244, 395, 179, 308, 175, and 463) to those of the remaining isolates. No statistical differences between the two groups were noted (P = 0.90 and P = 0.46, respectively). Comparison of (C) mouse LD₅₀ values and (D) G. mellonella LT₅₀ values of isolates belonging to five HRC STs (235, 111, 274, 253, and 244) to those of the remaining isolates. Again, no statistical differences were noted (P = 0.31 and P = 0.43, respectively). Each symbol represents an individual isolate. The antibiotic resistance rank (0–6) of each isolate is indicated by gray to red hue. Mann–Whitney U test, ns = not significant.

We next examined whether any lineages of P. aeruginosa are associated with unusually high levels of virulence, independent of HRC status. We grouped isolates based on core genome similarity using a phylogeny-based clustering tool (Fig. S4) (53). Seven clusters of genetically related isolates were generated, with five isolates not clustering into any group (Fig. 9). Mouse LD₅₀ values differed significantly across clusters (Kruskal–Wallis test, P = 0.006). Cluster 7, comprised of all exoU⁺ isolates and PABL071 (exoS⁻/exoU⁻), had the lowest median LD₅₀ value (highest virulence), but no cluster was significantly more or less virulent by pairwise comparisons. Similarly, the clusters did not significantly differ in their virulence in G. mellonella. This analysis did not identify hypervirulent lineages of P. aeruginosa, suggesting that either such lineages do not exist or that analysis of larger isolate collections allowing for more granular groupings will be necessary to identify such lineages. In addition, the mouse and G. mellonella models yielded similar results when applied to this question.

Fig 9.

Comparison of the virulence of P. aeruginosa isolates in distinct genetic clusters. (A) Comparison of the mouse LD₅₀ values of isolates belonging to seven defined phylogenetic clusters. Median mouse LD₅₀ values differed significantly across clusters (Kruskal–Wallis test, H = 17.92, P = 0.006), but no significant pairwise differences were noted (Bonferroni-adjusted Dunn’s post hoc multiple comparisons, P > 0.05). (B) Comparison of G. mellonella LT₅₀ values of isolates belonging to the seven clusters. No significant differences were found between the clusters (H = 4.69, P = 0.58). Each symbol represents an individual isolate. Bars represent medians, boxes represent the interquartile range (25th–75th percentiles), and brackets represent minimum and maximum values.

Differential roles of the T3SS in the models

To explore the mechanistic basis for the differential performance of the mouse and G. mellonella models, we focused on the T3SS, as it is a major pathogenic determinant of P. aeruginosa and has been shown to impact virulence in both models (25, 68).

In mice, ExoU⁺ and ExoS⁺ isolates were generally more virulent than ExoS⁻/ExoU⁻ isolates (P < 0.0001), with 11 of the 12 least virulent isolates being ExoS⁻/ExoU⁻ (Fig. 2). As a group, ExoU⁺ isolates were more virulent than ExoS⁺ isolates (P = 0.041, Fig. 10A), with 18 of the 25 most virulent isolates being ExoU⁺ (Fig. 2). The overall population also displayed a moderate phylogenetic signal associated with virulence (Pagel’s λ = 0.60, P = 4.79 × 10⁻⁵, Fig. S5A), which may be partly driven by the phylogenetic clustering of exoU⁺ B isolates. However, several ExoS⁺ isolates exceeded ExoU⁺ isolates in virulence, suggesting that additional factors beyond these T3SS effector proteins contribute to overall virulence. In fact, the most virulent isolate overall was PABL012, an ExoS⁺ isolate (Fig. 2). Some isolates (e.g., PABL104, PABL056, PABL092) had detectable in vitro effector secretion but exhibited low virulence in both infection models (Fig. 6). This suggests that in vitro effector secretion may not fully reflect in vivo secretion or that differences in the presence, type, and expression levels of other virulence factors (e.g., adhesins, siderophores, and proteases) may influence overall virulence. These results are in consistent with previously observed general relationships between T3SS phenotypes and disease severity in mice and in patients (15, 22, 23, 25) but indicate that exceptions are relatively common.

Fig 10.

Comparison of virulence of P. aeruginosa isolates in mice and G. mellonella based on ExoS and ExoU secretion status. (A) Mouse virulence data for each T3SS phenotype. ExoS⁺ isolates were less virulent than ExoU⁺ isolates (P = 0.041). (B) G. mellonella virulence data for each T3SS phenotype. No significant difference in virulence between ExoS⁺ and ExoU⁺ isolates was observed (P = >0.99). Each symbol represents an individual P. aeruginosa isolate. Kruskal–Wallis test; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns = not significant.

Results using G. mellonella were similar to those in mice but with a few notable differences. Both ExoS⁺ and ExoU⁺ isolates were more virulent than ExoS^–/ExoU^– isolates (P = 0.0034 and P = 0.0006, respectively), but there was substantial overlap between all three groups (Fig. 10B). For example, the seventh most virulent isolate in G. mellonella was PASP398, an ExoS^–/ExoU^– isolate (Fig. 3). In contrast to the mouse model, ExoU⁺ isolates, as a group, were not significantly more virulent than ExoS⁺ isolates (Fig. 10B); 6 of the 10 most virulent isolates were ExoS⁺ (Fig. 3). Additionally, the population displayed no evidence of any phylogenetic signal associated with virulence (Pagel’s λ = 0.008, P = 0.87, Fig. S5B), suggesting virulence is not lineage specific in this model. One explanation for these results is that individual T3SS effector proteins have differential impacts on virulence in G. mellonella compared to mice, which may partially explain why the correlation between virulence measures in these two models was not higher.

Several isolates exhibited pronounced discordance in virulence between the mouse and G. mellonella models (Fig. 6). We assessed whether ExoU and ExoS secretion phenotypes contribute to these model-dependent differences by focusing on isolates with the greatest divergence in virulence ranks between models. We identified the 10 isolates for which the virulence rank in G. mellonella most exceeded the virulence rank in mice (i.e., were relatively more virulent in G. mellonella than in mice) and the 10 isolates for which the virulence rank in mice most exceeded that in G. mellonella (i.e., were relatively more virulent in mice than in G. mellonella) (Fig. 6; Table S1). Both sets of isolates spanned a large range of virulence values in each model (Fig. S6). However, isolates with a greater relative virulence rank in G. mellonella were statistically more likely to be from Group A (predominantly exoS⁺ isolates) than were those with a greater relative virulence rank in mice (P = 0.033, Table 1; Fig. S7). This finding, together with the observation that ExoS⁺ isolates had higher virulence relative to ExoU⁺ isolates in G. mellonella than in mice (Fig. 10), suggests that G. mellonella are more affected by ExoS or another Group A-associated factor than mice.

TABLE 1.

Phylogenetic clade distribution among isolates with the greatest relative virulence rank difference between the mouse and G. mellonella models^b

	Number of isolates with the greatest relative virulence ranka
Phylogenetic clade	G. mellonella	Mouse
Group A	9	5
Group B	0	5

^a Top 10 isolates for which the virulence rank in G. mellonella exceeded that in mice, and the top10 isolates for which the virulence rank in mice exceeded that in G. mellonella. PABL043 (a PA7-like isolate) was excluded from this analysis because it belongs to neither Group A nor Group B.

^b Fisher’s exact test, P = 0.033.

We next examined whether in vitro type III secretion might explain why the cell co-culture results differed from the mouse results. This difference was largely driven by the fact that the ExoS⁺ and ExoS^–/ExoU^– isolates all displayed <5% cytotoxicity despite showing a broad range of mouse LD₅₀ values (Fig. 4; Data S1). In contrast, all ExoU⁺ isolates caused appreciable cytotoxicity, consistent with previous reports demonstrating ExoU is the predominant cytotoxin of P. aeruginosa (14, 69). Substantial differences were apparent in the proportion of cells lysed by ExoU⁺ isolates during infection, suggesting that factors in addition to the secretion of ExoU are involved in modulating the cytotoxicity of these isolates, potentially including differences in exoU expression or effector delivery. However, even when only ExoU⁺ isolates were included in the analysis, cytotoxicity correlated poorly with mouse LD₅₀ values (ρ = −0.30, P = 0.095), indicating that cytotoxicity alone only modestly predicts virulence in the murine model (Fig. S8).

DISCUSSION

To test the utility of using G. mellonella and cell co-culture models as alternatives to a mouse model of bacteremia, we quantified the virulence of P. aeruginosa. We found that G. mellonella produced results comparable to those of the mouse model, but cell lysis of A549 epithelial-like cells was a less reliable measure. We then showed that G. mellonella and mice yielded similar answers to P. aeruginosa population-based questions regarding virulence. These findings suggest that G. mellonella provides a high-throughput, cost-effective, and scalable substitute for the mouse bacteremia model in screening large numbers of P. aeruginosa isolates for virulence. However, final conclusions should be confirmed in mice.

Our results support previous reports showing a correlation between P. aeruginosa virulence measured by G. mellonella and murine models (10–12) and further quantify the extent of this correlation. Several biological differences may explain why the correlation was not higher. Mice and G. mellonella are markedly different in anatomy, physiology, and immunology. In addition, the method of inoculation (parenchymal injection in G. mellonella vs intravenous injection in mice) differed, as likely did dissemination (70, 71). Additionally, P. aeruginosa virulence factors may be differentially active in each host, either from variations in their targets or in host defense systems that detect or neutralize their activities. In this regard, a closer examination of our results provides several clues to these molecular differences. First, ExoS⁺ isolates as a group were as virulent as ExoU⁺ isolates in G. mellonella but less virulent in mice (Fig. 10). For mice, this observation is consistent with previous reports demonstrating that in a mouse pneumonia model, deletion of exoU attenuated virulence to a greater degree than deletion of exoS (25). It is possible that ExoS or another Group A-associated factor is more active in G. mellonella than in mice or that ExoU or a Group B-associated factor is less active. Notably, the role of ExoS in G. mellonella killing has not been evaluated, but an exoU-deficient mutant of P. aeruginosa was not attenuated in killing G. mellonella (68). Second, ExoS^–/ExoU^– isolates were generally less virulent in mice, but some of these isolates exhibited high virulence in G. mellonella, such as PASP398, PABL042, and PABL011 (Fig. 6). This finding suggests that unidentified factors other than ExoS and ExoU contribute relatively more to outcomes in G. mellonella than in mice. Defining which P. aeruginosa virulence determinants differentially impact illness severity in mice and G. mellonella is an exciting new avenue of research that may better define the conditions under which G. mellonella can be used with confidence as a substitute for mice.

Several isolates in our collection warrant further investigation. The first is PABL019, which was the least virulent isolate in both G. mellonella and mice (rank #105, Fig. 5A and 6). Interestingly, this isolate was obtained from a blood culture of an individual with CF. CF-associated P. aeruginosa isolates have been reported to be attenuated in virulence in both models (12, 72), likely due to their adaptation during chronic infection of the airways (3, 55, 73, 74). Nearly all of the other tested isolates in our study were from non-CF infections, so we were unable to further examine the behavior of CF isolates in G. mellonella, but future studies using such isolates would further test the ability of G. mellonella to mirror mice in this context. The second isolate is PABL002, which demonstrated high virulence in mice (rank #24) but relatively modest virulence in G. mellonella (rank #71) (Table S1). This isolate may be useful in defining those pathogenic factors that are critical in mice but fail to influence virulence in G. mellonella. Conversely, PASP398 had low virulence in mice (rank #76) and relatively high virulence in G. mellonella (rank #7), suggesting the presence of factors that enhance pathogenicity in G. mellonella but not in mice. These and similar isolates may serve as useful tools to dissect model-specific virulence determinants.

In comparison to G. mellonella, the cell co-culture model performed less well in predicting virulence in mice (Fig. 5B). At first glance, this result is surprising, as ExoU⁺ isolates exhibit high levels of virulence in mice and are also highly cytotoxic against epithelial cell lines (Fig. 2) (13–15, 25). However, closer examination suggests three contributors to the weaker correlation between these models. First, although ExoU⁺ isolates exhibited notable cytotoxicity, the remaining isolates caused minimal cytotoxicity. As a result, this assay was unable to discriminate virulence differences within the groups of ExoS⁺ and ExoS^–/ExoU^– isolates, which comprised ~67% of the tested isolates. Second, while many ExoU⁺ isolates were indeed highly virulent in mice and associated with substantial cell lysis, many noncytotoxic ExoS⁺ isolates were also highly virulent in the mouse model, such as PABL012, S10, and PABL049 (Fig. 6). Third, even among ExoU⁺ isolates, the correlation between cytotoxicity and mouse LD₅₀ values was poor (ρ = −0.30, Fig. S8). P. aeruginosa harbors multifactorial mechanisms to disseminate in the host, evade the immune response, and alter host physiology; epithelial cells alone may not capture the complexity of these host–pathogen interactions (75). Despite the well-established cytotoxic effects of ExoU, its contribution to overall virulence may be modulated or supplemented by additional factors. Alternatively, cell lysis may be an in vitro phenomenon that does not reflect a more nuanced pathogenic role for ExoU in vivo. Finally, the conditions of our cell lysis assays, such as the MOI, time of co-incubation, and host cell type, may not reflect the in vivo conditions under which ExoU exerts its impact on disease progression (76).

The impact of antimicrobial resistance on P. aeruginosa virulence is somewhat controversial. Our findings with 105 P. aeruginosa isolates across two infection models show an inverse relationship between antibiotic resistance and virulence, which could indicate a tradeoff between these two properties. However, these associations do not necessarily imply that antimicrobial resistance determinants themselves cause reduced virulence. P. aeruginosa strains inhabiting hospitals and long-term care facilities are heavily exposed to antibiotics but also to other selective pressures intrinsic to survival in patients and the healthcare environment. Adaptations that facilitate persistence in these niches would be expected to co-occur with antibiotic resistance alleles and may ultimately be responsible for the inverse relationship we observed. For example, P. aeruginosa lineages cultured from chronically infected CF patients simultaneously accumulate disruptions of bacterial virulence attributes and mutations that enhance antibiotic resistance, presumably due to independent immune and antimicrobial selective pressures (74, 77).

Hypervirulent clades have been well described in several bacterial species (64–66), but whether such clades exist in P. aeruginosa remains unclear. We were unable to detect groups of genetically related isolates that were highly virulent in mice or G. mellonella. In particular, the virulence levels of defined HRCs, which have previously been linked to poor clinical outcomes in patients (27, 28, 34, 78), were not significantly different from non-HRCs in either model. This finding suggests that HRC-associated poor outcomes stem mainly from their resistance to antibiotic treatments or their propensity to infect patients with comorbid conditions. Interestingly, many isolates belonging to HRC STs in our collection were relatively antibiotic susceptible (Fig. 8). This finding supports previous reports suggesting that HRCs are more prone to acquire and maintain resistance determinants, but that many isolates within HRCs remain susceptible to antibiotics (28). Quantification of these rates and properties is necessary to formulate a more precise definition of HRCs.

Our study has several limitations. The clinical isolates we used predominantly originated from bloodstream infections and may not accurately reflect the virulence or clonal distribution of P. aeruginosa isolates from other infections. Additionally, sampling spanned only three continents; broader geographic coverage is necessary to demonstrate whether our results are generalizable. Unlike murine models, which use inbred mice, G. mellonella larvae may vary in genetic background and husbandry, introducing possible variability in LT₅₀ values. Cytotoxicity was assessed exclusively in A549 cells, a well-established cell line for cell lysis assays, but results may differ with other cell lines. We used a 3-hour readout for cytotoxicity, but longer times likely would have detected cytotoxicity among non-ExoU isolates. However, this likely would come at the cost of saturated cell lysis for the ExoU⁺ isolates. Finally, we recognize that definitive conclusions on the utility of alternative infection models such as G. mellonella and cell co-culture assays cannot be drawn from a single type of mouse model (bacteremia) and cell culture readout (cytotoxicity) with a single bacterial species (P. aeruginosa). However, we believe that this study will serve as an incremental step towards the accumulation of sufficient data to answer this important question.

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/iai.00335-26.

Data S1. iai.00335-26-s0001.xlsx.

Various data.

Supplemental material. iai.00335-26-s0002.docx.

Fig. S1 to S8; 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.