In this study, three strains of P. aeruginosa were used for biofilm cultivation, including type strain P. aeruginosa ATCC 10145 (DSM 50071), wild-type strain P. aeruginosa PAO1 originally isolated from a wound infection [26] and mucoid (alginate-overproducing) strain P. aeruginosa FRD1 (mucA22 mutation) isolated from the sputum of a CF patient [27].

Three agar media were employed for biofilm cultivation. As a non-selective complex growth medium, LB agar was used (35 g L-1 LB agar composed of (per L) 10 g tryptone, 5 g yeast extract, 5 g NaCl and 15 g agar, pH 7.0 $\pm$ 0.2) without KNO${}_3$ (LB). When preparing the agar media for anaerobic cultivation (LBN), potassium nitrate was also added to the 35 g-L 1 LB agar solution. The concentration of potassium nitrate in the LB agar medium was 10.11 g$\cdot$L${}^{-1}$ (final concentration 100 mM) (LBN). As a selective growth medium, Pseudomonas Isolation Agar (PIA) was used composed of (per L) 13.6 g agar, 1.4 g magnesium chloride, 20 g peptone, 10 g potassium sulfate and 0.025 g triclosan, pH 7 $\pm$ 0.2 (25 ${}^{\circ }$C) supplemented with 20 mL glycerol, without (PIA) and with the addition of KNO${}_3$ (final concentration 100 mM) (PIAN). After autoclaving, the agar media (20 mL) were filled into sterile glass Petri dishes (diameter, 90 mm) and allowed to solidify at room temperature.

For simulating growth conditions in the mucus of CF airways, Artificial Sputum Medium (ASM) was used as a solid agar medium based on the formulation of liquid ASM described by Sriramulu et al. [23] and Kirchner et al. [24]. ASM agar without (ASM) and with added KNO${}_3$ (ASMN) was prepared as follows. 0.8 g DNA, 1.0 g mucin and 1 g casamino acid were dissolved separately in 50 mL and 30 mL autoclaved ultrapure water, respectively, under constant stirring. The solutions were combined and 8 mL of a solution of 3 mg DTPA, 2.5 g NaCl and 1.1 g KCl in 30 mL water were added. For ASMN, KNO${}_3$ was added to a final concentration of 100 mM. After addition of sterile ultrapure water to a volume of 170 mL, the pH value of the solution was adjusted to 6.9 using 1 M Tris buffer and the total volume was adjusted to 200 mL. Finally, 3 g Bacto agar were added and suspended in the solution. ASM was sterilized by using a microwave autoclave (MediJet500, Phoenix Instrument GmbH, Garbsen, Germany). The procedure consisted of heating the medium to 135 ${}^{\circ }$C within 2 min 40 s, a holding time of 22 s, cooling to 100 ${}^{\circ }$C within 6 min. The medium was further cooled to 50 ${}^{\circ }$C in a water bath for 20 min. 1 mL sterile egg yolk emulsion was added to 199 mL of ASM and after mixing 20-mL portions of the agar medium were dispensed into sterile glass Petri dishes as described above.

Biofilms were grown as confluent bacterial lawns on the surface of the agar media as an in vitro biofilm model of P. aeruginosa described before [25]. In brief, single colonies of cultures grown overnight at 37 ${}^{\circ }$C were suspended in 0.14 M NaCl solution. Aliquots of 0.1 mL bacterial suspension were spread-plated on the agar media in glass Petri dishes (diameter, 100 mm) to a concentration of approximately 105 cells cm${}^{-2}$ on the medium surface. One PDMS film strip (0.35 mm $\times$ 55 mm) was placed in the centre of the glass lids. The inoculated agar plates without added KNO${}_3$ were incubated aerobically and agar plates with added KNO${}_3$ (final concentration 100 mM) were incubated anaerobically at 37 ${}^{\circ }$C for 48 h. Anaerobic cultivation was conducted in anaerobic jars with an oxygen-free atmosphere generated by Anaerocult® A (Merck KGaA, Darmstadt, Germany). All plates were incubated upside down [25].

For biofilm analysis, biomass was scraped from the agar surface using a sterile metal spatula, and suspended in 2 mL deionized water. The suspensions were vortexed for 2 min to disperse the biofilm cells. Serial dilutions of the suspensions were prepared in deionized water. For determination of total cell counts in the biofilm suspensions, bacteria were stained with the fluorochrome 4’,6-diamidino-2-phenylindole (DAPI). 1 ml of DAPI solution (25 $\mu$g mL${}^{-1}$ in 2% (vol/vol) formaldehyde) was added to 4 mL of cell suspension. After incubation at room temperature for 15 min in the dark the suspension was filtered through a black polycarbonate membrane filter (Millipore, 0.2 $\mu$m pore size). Cells on the membrane filters were manually counted under an epifluorescence microscope at 1000-fold magnification. 20 randomly selected fields of view were examined for each filter using a counting grid (100 $\mu$m $\times$ 100 $\mu$m). Results were expressed as cells cm${}^{-2}$ of the surface area of the biofilm-covered agar media.

Headspace sampling of the biofilm cultures was performed by TFME using commercial PDMS films (0.35 mm $\times$ 55 mm; thickness: 0.45 mm $\pm$ 10%; Goodfellow GmbH, Hamburg, Germany). For cleaning of the PDMS films LC-MS grade methanol from VWR (Leuven, Belgium) and 5% aqueous Decon 90 (VWR, Leuven, Belgium) solution was used. The cleaning process is described in [25].

The loaded sorption films were analysed by direct coupling of a thermodesorption system (TDS 3, Gerstel GmbH & Co. KG, Mühlheim a.d.R., Germany) incl. autosampler (TDS-A2, Gerstel GmbH & Co. KG, Mühlheim a.d.R., Germany) with a gas chromatograph (Agilent 6890 GC, Agilent Technologies Inc., Waldbronn, Germany) and a quadrupole mass spectrometer (Agilent 5975 MSD, Agilent Technologies Inc., Waldbronn, Germany). Adsorbed analytes were desorbed in the TDS at 200 ${}^{\circ }$C and transferred to the injector by a helium gas flow (ALPHAGAZ 1, 99.9%, Air Liquide, Düsseldorf, Germany) of 60 mL$\cdot$min${}^{-1}$, via a metallic transfer line with an inert surface (Prosteel, deactivated, Gerstel GmbH & Co. KG, Mühlheim a.d.R., Germany) at a temperature of 300 ${}^{\circ }$C. Cold injection system CIS 4 of Gerstel GmbH & CO. KG was used for introduction of the analytes. The injector temperature during thermal desorption is –10 ${}^{\circ }$C to trap the desorbed analytes in the liner filled with deactivated glass wool. The injector is heated to 270 ${}^{\circ }$C with a temperature-programming rate of 10 ${}^{\circ }$C$\cdot$s${}^{-1}$ to transfer the analytes from the liner to the GC column. The initial oven temperature is –10 ${}^{\circ }$C (hold time: 2 min), which causes the analytes to refocus at the column head. The GC column was heated to 325 ${}^{\circ }$C (hold time: 8 min) with a temperature programming rate of 10 ${}^{\circ }$C$\cdot$min${}^{-1}$. A DB-1 column (Agilent Technologies Inc., Waldbronn, Germany) with a length of 30 m, an inner diameter of 0.25 mm and a film thickness of 1 $\mu$m was used to separate the analytes. The separated analytes were transferred to the EI-qMS via a transfer capillary heated to 280 ${}^{\circ }$C. Ionization was performed by electron impact ionization at 70 eV. The qMS analysis was performed in scan mode with a scan range of 40–600 Da and a scanning rate of 2.24 scans$\cdot$s${}^{-1}$.

The data obtained from the TD-GC-qMS analyses, (Agilent MSD Chemstation data format) were first converted into Agilent MassHunter data format using MassHunter GC/MS Translator software (Agilent Technologies Inc., Santa Clara, USA). Subsequently, a conversion into the “.abf” data format was carried out using the software Reifycs Analysis Base File Converter (Reifycs Inc., Tokyo, Japan). This enables feature analysis using MS-DIAL (version 4.18, RIKEN, Wako, Japan), which performs deconvolution and annotation of the data. The parameters used for deconvolution, feature analysis and annotation are listed in Supplementary Table 1 . The identification of the analytes based on the obtained EI-MS spectra was performed by using MS databases. The metabolomics specific GC-MS database MassBank of North America (MoNA) by Oliver Fiehn was used and analytes with a similarity score of $\ge$80% were putatively identified.

The statistical evaluation of the deconvoluted data was carried out with a standardized workflow. For this purpose, a peak intensity cut-off of 1000 counts and a similarity score cut-off of 70% was applied. Missing values were not refilled. To reduce the dimensionality of the data and to identify differe nces and similarities, non-supervised Principal Component Analysis (PCA) was performed using SIMCA software (Version: 16.0.2.10561; Satorius AG, Göttigen, Germany).

PCA is an algorithm that helps to reduce dimensionality while retaining most of the variance of the data that contains useful information and is frequently used in omics. Each principal component is identified by the direction and the amount of variance in the data and is a linear combination of the original variables. Each sample can then be represented by relatively few principal components instead of thousands of metabolites. For example, in Fig. 1a, the PC1 axis is explaining the variance between both samples groups which is the growth under aerobic or anaerobic conditions and on the PC2 axis, the variance within each of the sample groups is explained. For further information regarding the interpretation of a PCA, we recommend the publication of Ringnér [28]. Scores and loadings as well as the variance are calculated. Potentially interesting metabolites were further statistically analyzed by parametric significance test (t-test; two-sided distribution and heteroscedastic). The significance was evaluated using the p-value and classified into three significance levels, 0.001, 0.01 and 0.05.

Fig. 1.

Graphic representation of the evaluation for the differentiation of cultivation conditions ((aerobic (blue) vs. anaerobic (red) for the bacterial strains P. aeruginosa FRD1 and P. aeruginosa PAO1 using PCA and t-test. The biofilms were cultured aerobically on LB medium and anaerobically on LBN medium for 48 h at 37 ${}^{\circ }$C. The results of the PCA are shown as scores (a: PAO1; c: FRD1) and loadings plots (b: PAO1; d: FRD1). The peak intensities of selected VOCs shown in the bar charts correspond to the mean values from eight biological replicates. The standard deviation of the mean value is shown as an error bar. The substances (peak intensities are shown in the bar charts) (e): 1-tetradecene (PAO1), (f): 3-aminoquinoline (PAO1), (g): dimethyl disulfide (FRD1), (h): 2-aminoacetophenone (FRD1), (i): 2,5-hexadione (PAO1), (j): dimethyl disulfide (PAO1), (k): benzaldehyde (FRD1), (l): 2-hexanone (FRD1). The stars above symbolise the confidence level, where *** represents a p value of 0.001, **0.01 and * 0.05 (n = 8).

With the described materials and methods, both microbiological and analytical sample preparation is possible. The methods used for the characterization of the bacterial biofilms, for the sample preparation using polydimethylsiloxane (PDMS) sorbent films and for the analysis of the extracellular volatile metabolites are described.

In general, research on the extracellular volatile metabolome of P. aeruginosa can be divided into five areas, (1) identification and differentiation of bacterial strains, (2) analysis of metabolic changes due to external factors, (3) identification of unknown metabolites, (4) identification of metabolic profiles using genome structure [16], and (5) effector function of volatiles in intermicrobial interactions such as the recently reported growth inhibition of the fungus Aspergillus fumigatus, relevant in CF airway infections, by P. aeruginosa volatiles [29]. Bos et al. [14] have already criticized the lack of comparability of published metabolome studies on different bacterial pathogens. Especially metabolome studies in the context of a disease, such as CF, should be performed in vitro under conditions like those in the human body. This necessity is due to the variability of the metabolome of bacteria to different external influences, such as nutrient deficiency and oxygen concentration. Furthermore, P. aeruginosa is a bacterium that is particularly well adapted to different environmental conditions and is capable of changing the phenotype , e.g., development from a non-mucoid to a mucoid variant during chronic CF airway infections [3]. In the following the influence of the phenotype and growth conditions on the metabolome of P. aeruginosa is shown.

An in vitro agar plate biofilm model was used throughout the study. The biofilms of P. aeruginosa were grown as confluent bacterial lawns directly on the surface of three different solid agar media, unsupplemented (LB, PIA, ASM) and supplemented with 100 mM KNO${}_3$ (LBN, PIAN, ASMN) for cultivation under aerobic and anaerobic conditions, respectively. This type of unsaturated biofilm model has proven useful for studying the morphology and composition of aerobic Pseudomonas biofilms such as those of P. aeruginosa, including the PAO1 and FRD1 strains used in the present study [30, 31, 32]. Recently, this model was used as the basis for a modification of the system suitable for analysing volatile organic compounds in the headspace of aerobically grown P. aeruginosa biofilm cultures [25]. In addition to the LB and PIA agar media conventionally used for P. aeruginosa cultivation a modified formulation of ASM was included that was originally developed to resemble in its composition (amino acids, mucins, DNA, among other substances) more closely conditions in CF lungs [23, 24]. So far, ASM has been described exclusively for the liquid cultivation of P. aeruginosa in the context of CF disease. For the cultivation of P. aeruginosa biofilms in our in vitro biofilm model the preparation of solid ASM was necessary and was accomplished by addition of agar to the liquid medium to obtain the final ASM as an agar medium as described above.

In the present study, monitoring of aerobically and anaerobically biofilm growth at 37 ${}^{\circ }$C for seven days had shown that the stationary growth phase with maximal cell numbers of biofilms were reached after 48 h of incubation. Taken together, the mean cell density values of all strains (P. aeruginosa ATCC 10145, PAO1 and FRD1) in the agar-grown biofilms varied between 4.6 $\times$ 10${}^9$ to 7.6 $\times$ 10${}^9$ cells/cm${}^2$ under aerobic conditions and 1.3 $\times$ 10${}^9$ to 6.1 $\times$ 10${}^9$ cells/cm${}^2$ under anaerobic conditionson LB/LBN, between 0.7 $\times$ 10${}^9$ to 2.4 $\times$ 10${}^9$ cells/cm${}^2$ and 3.3 $\times$ 10${}^9$ to 4.0 $\times$ 10${}^9$ cells/cm${}^2$ under aerobic and anaerobic conditions, respectively, on PIA/PIAN, and between 0.8 $\times$ 10${}^9$ to 2.2 $\times$ 10${}^9$ cells/cm${}^2$ and 0.4 $\times$ 10${}^9$ to 3.6 $\times$ 10${}^9$cells/cm${}^2$ under aerobic and anaerobic conditions, respectively, on ASM/ASMN (mean values of two to four independent experiments, exception for strain PAO1 biofilms on ASM/ASMN with data from only one growth experiment). These bacterial densities indicated the development of multilayer biofilms [33] on the surface of all agar media under both aerobic and anaerobic growth conditions.

Under anaerobic conditions no growth was observed on unsupplemented agar media, while addition of KNO${}_3$ (100 mM) to the agar media allowed bacterial growth, indicating that anoxic growth of P. aeruginosa biofilms via anerobic respiration by nitrate reduction (denitrification) occurred in the biofilm model as has been shown previously in anaerobic zones of mucus from CF airways [5] and in growth experiments in vitro under controlled laboratory conditions for clinical isolates from CF patients and laboratory strains like non-mucoid strain PAO1 and mucoid strain FRD1 used in the current study [34, 35].

3.1.2 Investigation of the Extracellular Metabolome from Two P. aeruginosa Strains with Mucoid and Non-mucoid Phenotypes

In addition to the cultivation atmosphere, the phenotype of the bacteria may have a significant effect on the metabolome. In cystic fibrosis, the mucoid phenotype is dominant in chronic infections, whereas in initial infections, especially in younger CF patients, non-mucoid strains in the lungs are dominant [3]. A statistical and comparative study of the non-mucoidstrain P. aeruginosa PAO1 (acute infection isolate) and mucoid strain P. aeruginosa FRD1 (chronic infection isolate) was performed to get more information whether the phenotype has an influence on the extracellular volatile metabolome of P. aeruginosa. Cultivation in the in vitro biofilm model was performed under anaerobic conditions using the complex LB medium supplemented with potassium nitrate as described by Koehler et al. [25]. Eight biological replicates were investigated per bacterial strain.

The comparison of the two phenotypes using a PCA is shown in Fig. 2a. A distinction between the two phenotypes is not possible by means of two main components, since the scatter of results in the scores plot is large (0.2–20, blue dots) for the mucoid strain P. aeruginosa FRD1. In contrast, the scatter of results for the strain P. aeruginosa PAO1 is small (2–6, red dots), but the results for the mucoid strain are partly within the range for the non-mucoid strain P. aeruginosa PAO1. Furthermore, the substance classes are not grouped according to the two phenotypes (Fig. 2b). Therefore, differentiation of the two phenotypes cannot be done at the substance class level. However, the influence of the phenotype, for example the formation of alginate (mucoid phenotype) on the molecular level cannot be excluded. For this purpose, the peak intensities of the identified metabolites were compared with each other and their significance was determined using the t-test. Using the example of six metabolites, the difference in mean peak intensities is shown in Fig. 2, with the significance of the difference in peak intensity varying for the selected metabolites.

Fig. 2.

Comparison of the two strains P. aeruginosa FRD1 (blue) and PAO1 (red) with mucoid and non-mucoid phenotype, respectively, by PCA, statistical evaluation by t-test. Cultivation in the in vitro biofilm model was performed under anaerobic conditions using LB medium supplemented with potassium nitrate for 48 h at 37 ${}^{\circ }$C. Figures a and b: results of the PCA as scores (a) and loadings plot (b). Figures c to h: difference in peak intensity between the two phenotypes considered, using the example of (c): octanol, (d): benzaldehyde, (e): propylbenzene, (f): dimethyl disulfide, (g): 2-undecanone, (h): 4-methylquinoline. The mean values of the peak intensities of the biological replicates produced and their standard deviation as error bars are shown in form of bar charts. The stars above symbolise the confidence level, where *** represents a p value of 0.001, ** 0.01 and * 0.05.

The peak intensities of certain metabolites were significantly increased in the mucoid phenotype compared to the non-mucoid phenotype (see Fig. 2a–e). However, the opposite trend could be observed for certain metabolites (see Fig. 2f–h). For example, a 4-fold higher peak intensity was observed for benzaldehyde in mucoid strains. The example of the metabolite dimethyl disulphide shows that a significant reduction of peak intensity can also occur in the mucoid phenotype compared to the non-mucoid phenotype. The peak intensity of the mucoid phenotype is reduced by a factor of 13. In total, a significant difference in peak intensity was found for 28 metabolites. The statistical comparison of the mVOCs between the two phenotypes revealed three metabolites whose difference in peak intensity was a confidence level of p = 0.001. Eleven metabolites had a confidence level of p = 0.01 and fourteen metabolites had a confidence level of p = 0.05. A list of these metabolites is shown in Supplementary Table 4. Previously, secondary electrospray ionization-mass spectrometry (SESI-MS) was applied to produce fingerprints of the P. aeruginosa strains PAO1 and FRD1 directly from the breath of infected mice [37]. It was demonstrated that SESI-MS was capable of differentiating infected versus uninfected mice as well as infections caused by strain PAO1 versus FRD1. In addition, comparison of the in vivo volatiles with in vitro volatiles from liquid laboratory cultures showed that only 25–34% of peaks were shared between the in vivo and in vitro fingerprints [37].

In summary, the investigation of the influence of the mucoid phenotype of the bacterium P. aeruginosa on its metabolism showed that at the level of substance classes as well as two main components, no distinction was possible in this study. However, differences could be detected at the molecular level. It was shown that, in addition to the cultivation conditions (aerobic and anaerobic), the mucoid phenotype of the bacterial strain must be taken into account in metabolome studies.

3.1.3 Influence of the Culture Media on the Extracellular Volatile Metabolome of P. aeruginosa

In addition to the influencing variables already studied and discussed, the culture medium could have an impact on the extracellular volatile metabolome of P. aeruginosa. In most metabolome studies published on P. aeruginosa the full medium LB-Lennox and bacteria selective growth media are used [14]. Such selective culture media contain only the substances that the respective bacterium needs to grow. Additionally, these culture media contain chemicals that suppress the cultivation of other bacteria. The Pseudomonas isolation (PIAN) agar is such a P. aeruginosa specific medium. In order to investigate the influence of the culture medium on the metabolome of P. aeruginosa in the context of CF disease two clinical strains P. aeruginosa PAO1 and P. aeruginosa FRD1 were cultivated under anaerobic conditions on both culture media (LBN and PIAN medium). Potassium nitrate was added to both culture media to ensure anaerobic cultivation. Five biological replicates of the strain P. aeruginosa FRD1 and three biological replicates of the strain P. aeruginosa PAO1 were grown on PIAN medium. The detected extracellular volatile metabolites were compared with those of eight biological replicates of the two strains cultivated on LBN medium. To exclude the influence of the phenotype described in chapter 3.1.2, the comparison of the extracellular volatile metabolites was performed exclusively between cultures with the same phenotype, e.g., FRD1 on LBN vs. FRD1 on PIAN.

The scores and loading plots obtained by performing a PCA are presented in Fig. 3a–d. From these results a significant differentiation of the cultivation on both culture media is possible using two main components. To differentiate between the two culture media, only the first principal component is necessary, which explains 84% of the variance for strain P. aeruginosa PAO1 and 60% for strain P. aeruginosa FRD1. The second principal component only describes the scatter within the group. For PAO1 this principal component describes 91% variance and for FRD1 79%. Based on the high values, the qualitative observation of the differentiation between the two culture media by means of PCA can be confirmed quantitatively. Considering the corresponding loadings plots, the detected extracellular volatile metabolites can be assigned to the nutrient medium.

Fig. 3.

Comparison of two different culture media (LBN: blue; PIAN: red) in a cultivation of the two strains P. aeruginosa PAO1 (a, b) (n = 5) as well as FRD1 (e, f) (n = 3) under anaerobic conditions in an in vitro biofilm model by PCA. Cultivation in the in vitro biofilm model was carried out under anaerobic conditions in LB medium supplemented with potassium nitrate and in the Pseudomonas isolation agar supplemented with potassium nitrate. The bacteria were cultivated for 48 hours at 37 ${}^{\circ }$C. The mean values of the signal intensities and the corresponding standard deviations are shown as bar graphs. Significance analysis of the difference in signal intensities was performed using a t-test (two-sided distribution, heteroscedastic). Metabolites shown are: c = 1-hexadecanol, d = pentadecane, g = 2-nonanone, h = 1-octanol (all ***; p $\ge$ 0.001).

Furthermore, the variance between the biological replicates in the LBN medium for the cultivation of P. aeruginosa FRD1 (see Fig. 3e) as well as for P. aeruginosa PAO1 (see Fig. 3a) is small. A higher variance for both bacterial strains as well as a significantly larger number of extracellular volatile metabolites is achieved by cultivation on PIAN medium. A t-test was used to investigate the differentiation of both cultivation media at the molecular level. Fig. 3c–f show two substances for each of the two bacterial strains, whose mean peak intensity is significantly higher when cultivated on PIAN medium (confidence level: p = 0.001).

For the detection of the metabolites 2-nonanone and 1-octanol, a higher signal intensity was observed when using PIAN agar (Cultivation of P. aeruginosa FRD1, with a statistical significance of 99.9%) (Fig. 3g,3h). Similarly, higher signal intensities were detected for 1-hexadecanol and pentadecane (cultivation of P. aeruginosa PAO1 on PIAN agar) with the same statistical significance (Fig. 3c, 3d). Overall, significance analysis of culturing P. aeruginosa FRD1 on LBN and PIAN agar revealed 102 metabolites with a significant difference in signal intensity. 21 metabolites could be assigned to the significance level p = 0.001, 33 metabolites to the significance level p = 0.01, and 48 metabolites to the significance level p = 0.05.

The analysis for the bacterial strain P. aeruginosa PAO1 revealed 123 metabolites with a significant difference in signal intensity. 25 metabolites could be assigned to the significance level p = 0.001, 54 metabolites to the significance level p = 0.01 and 44 metabolites to the significance level p = 0.05. An overview of the determined metabolites of the two investigated bacterial strains is given in the supplement (P. aeruginosa FRD1: Supplementary Table 5; P. aeruginosa PAO1: Supplementary Table 6).

First, this analysis shows that the culture medium has a significant influence on the extracellular volatile metabolome, due to the high number of metabolites with significantly different signal intensities. On the other hand, due to the similar number of metabolites with significant difference in signal intensity, it was shown that the influence of the culture medium is independent of the mucoid phenotype of the bacterium. Analogous to the results of the investigation of the influence of the mucoid phenotype, a difference can be detected on the molecular level.

The extracellular volatile metabolome of the two clinical strains P. aeruginosa FRD1 as well as PAO1 and the reference strain P. aeruginosa ATTC 10145 was investigated. The in vitro metabolome investigations of the bacterial strains were performed under CF-like conditions using the solid ASM described above. The mVOCs of three biological replicates of each bacterial strain were identified after deconvolution and feature analysis by the GC-EI-MS specific metabolomics database MoNA. Following data processing and peak identification, the statistical evaluation of the metabolites was carried out performing a PCA following the identification of specific metabolites by evaluating the resulting score and loading plots (see Fig. 4).

Fig. 4.

Comparison of the available P. aeruginosa strains (Fig. A: blue = ATCC 10145; red = PAO1; black = FRD1) by means of their extracellular volatile metabolites. The three bacterial strains were cultivated under anaerobic conditions and on ASMN agar using the in vitro biofilm model. Comparison was performed using unsupervised PCA. Fig. 4A shows the scores and the loadings plot. In the loadings plot, the metabolites are labeled, which are shown as bar graphs in Fig. 4B (a–l). a = benzophenone, b = tetradecane, c = toluene, d = 2-ethylhexanal, e = 3-methoxypropylamine, f = trans-2-dodecenyl acetate, g = 1-hexadecanol, h = pentadecane, i = methyl benzoate, j = 2-octanone, k = isoprene, l = 1-decanol. * = 0.05; ** = 0.01; *** = 0.001.

Differentiation between the investigated strains of P. aeruginosa is possible by the first two principal components, as can be seen from the scores plot in Fig. 4A. Cumulatively, these two principal components explain a variance of 98 %. Furthermore, the first principal component is suitable to explain the difference between the strains P. aeruginosa ATCC 10145 and PAO1 and the strain P. aeruginosa FRD1. For explaining the difference between P. aeruginosa ATCC 10145 and P. aeruginosa PAO1 the second principal component can be applied. The variance between the replicates of one strain, except for the replicates of P. aeruginosa PAO1, is low compared to cultivation on the other culture media used (see Fig. 4a). In the replicates of the strain P. aeruginosa PAO1, only one outlier can be detected. This outlier could be caused on the one hand by irregularities in the analysis or on the other hand by deviation in the sample preparation. The biological variance must also be considered in the discussion, but since only technical replicates were analysed, this can be assumed to be low. Despite this outlier, a clear differentiation between the bacterial strains with the first two principal components as well as an identification of bacterial strain-selective metabolites using the loadings plot (see Fig. 4A,4b) is possible. Based on the scores and loadings plot, it can be assumed that the difference between the bacterial strains P. aeruginosa ATCC 10145 and FRD1 as well as between P. aeruginosa PAO1 and FRD1 is significantly larger than between P. aeruginosa ATCC 10145 and PAO1. This can be explained by the differentiation of the mentioned strains already in the first principal component as well as the number of identified features, shown in the loadings plot (see Fig. 4A,4b). An evaluation of this observation can be carried out with the help of a significance analysis. For this purpose, the variance of the peak intensity of the identified features between all strains was tested using a t-test (two-sided distribution, heteroskedastic). By means of the significance analysis, a total of 70 metabolites could be identified that show a significant difference (at least 95%) in peak intensities between the strains P. aeruginosa FRD1 and ATCC 10145 (see Supplementary Table 7). Analogous to the qualitative observation from the scores and loadings plot, the number of metabolites with a significant difference in peak intensity decreases significantly when comparing the bacterial strains P. aeruginosa FRD1 and PAO1 (36 metabolites) and P. aeruginosa ATCC 10145 and PAO1 (28 metabolites). In addition to the decrease in total metabolites, the proportion of metabolites with the greatest significance (p = 0.001; 99.9%) in the total number also decreases. Both this quantitative and qualitative comparison using scores and loadings plot reveals that P. aeruginosa ATCC 10145 and PAO1 are more similar to each other in their extracellular volatile metabolome than P. aeruginosa FRD1. This can be attributed to the different phenotype. Furthermore, it can be observed that the difference of P. aeruginosa FRD1 to P. aeruginosa ATCC 10145 is higher than to P. aeruginosa PAO1. This could be explained by different origins of the strains (P. aeruginosa ATCC 10145 as a neotype strain of P. aeruginosa and P. aeruginosa PAO1 as originally a clinical isolate). P. aeruginosa PAO1 was originally isolated from a burn wound of a patient and was thus exposed to environmental influences. These environmental influences may have led to mutations that made the P. aeruginosa PAO1 strain resistant to such influences, resulting in the presence of different genes despite the same mucoid phenotype, thereby diverging the phenotypes between the two bacterial strains. Although Kavanagh, who first used P. aeruginosa ATCC 10145 in 1947 [38], did not provide any information on the date of origin or the origin of the bacterial strain [38]. Due to this lack of information regarding the original isolation of P. aeruginosa ATCC 10145, the explanation is based exclusively on theoretical considerations.

On the molecular level, for example, some molecules of the substance class of methyl ketones (e.g., 2-octanone, 2-hexanone, etc.) could be identified (see Supplementary Table 7). Forney et al. [39, 40] have described the formation of methyl ketones from n-alkanes in bacteria in two publications. Furthermore, Thijsse et al. [41] found that P. aeruginosa strains accumulate methyl ketones, such as 2-hexanone. This was supported by the research results of Robinson et al. [42]. 2-Hexanone and other methyl ketones have already been identified as volatile metabolites in several studies of the metabolome of P. aeruginosa [14, 18, 43, 44, 45]. Methyl ketones, such as 2-nonanone [17], have also been detected in CF sputum. However, as Bos et al. [14] report, methyl ketones can also be produced by other bacteria such as S. aureus. Since S. aureus is also a CF-relevant pathogen, the influence of other CF-relevant pathogens must also be taken into account when analysing the extracellular volatile metabolome of P. aeruginosa in the context of respiratory infections in CF disease.