Towards the Identification of Antibiotic-Resistant Bacteria Causing Urinary Tract Infections Using Volatile Organic Compounds Analysis—A Pilot Study

Antibiotic resistance is an unprecedented threat to modern medicine. The analysis of volatile organic compounds (VOCs) from bacteria potentially offers a rapid way to determine antibiotic susceptibility in bacteria. This study aimed to find the optimal conditions to obtain the maximum number of VOCs detected which next allowed the assessment of differences in VOC profiles between susceptible and resistant isolates of Escherichia coli causing urinary tract infections. The analysis of VOCs in the headspace above the bacterial cultures allowed the distinguishing of resistant and susceptible bacteria based on the abundance of six VOCs with 85.7% overall accuracy. The results of this preliminary study are promising, and with development could lead to a practical, faster diagnostic method for use in routine microbiology.


Introduction
There is a need for more rapid determination of antibiotic resistance (ABR) in bacteria isolated from clinical samples. ABR is becoming more prevalent, as are urinary tract infections (UTIs) due to an aging population. Community-acquired UTIs have a prevalence of 0.7% in the United States corresponding to 7-8 million infections per year. This rises to 4% amongst healthcare-associated infections in the USA and 6% in Europe [1]. In healthcare-associated infections, the proportion of isolates resistant to specific antibiotics is rarely below 10% and in the case of ampicillin is above 40% [2]. Rapid susceptibility testing and appropriate treatment will decrease morbidity and healthcare costs, plus protect the efficacy of broad spectrum agents. their urine on or after admission. This preliminary study aimed to distinguish the most common and problematic ESBL variant seen in the UK, namely CTX-M-15, from susceptible strains of the same organism, in broth culture.
In order to establish whether ESBL strains could be distinguished from other strains of E. coli it was important to devise a methodology that was both standardised enough to observe any significant differences and simple enough to be performed in a routine laboratory. The simplest and most commonly performed method of analysing volatiles produced by microorganisms is analysis of the headspace above a vial of bacteria in planktonic culture. In an ideal situation, this would be performed directly from the vial into a suitable instrument such as a gas chromatograph-mass spectrometer (GC-MS). However, as such instruments are not currently available in routine microbiology laboratories a method for sampling remotely and transporting between sites was required. Solid-phase microextraction (SPME) fibres have gained popularity in recent years as sample extraction and preconcentration is achieved in a single step. However, when studying samples of a complex heterogeneous nature, a number of pitfalls must be avoided [17]. The precoated surface of the fibre is exposed to the headspace above liquid cultures of bacteria and over a period of time, VOCs are adsorbed onto the surface until an equilibrium is reached. The nature of this process, however, depends hugely on the composition of the surface, the time given to reach equilibrium and the temperature at which this occurs. During the equilibration, loss of compound can occur by either biodegradation or adsorption to other surfaces. Fibres are also coated with many different coatings and these all adsorb some compounds preferentially to others. Balasubramanian and Panigrahi provide an exhaustive review of these factors and others, and it can be seen that most studies that compare SPME techniques show greater variance between samples than with other techniques [18][19][20]. It was therefore very important for this study that readily available SPME fibres were used and incubation and storage conditions were carefully selected to optimise the competing goals of sample standardisation and repeatability of the method by routine laboratories.
The aim of this study was to find the optimal conditions to obtain the maximum number of VOCs detected in order to distinguish between susceptible and resistant strains of particular organisms and antimicrobials. The best outcome for any potential diagnostic method would be an ability to distinguish between strains of E. coli with different resistance profiles by a single sampling of vial headspace by SPME fibre. The first arm of this study therefore compared several replicates of two groups of four strains. One group were CTX-M-15 ESBL strains and the other susceptible strains of E. coli. Any discrimination of the two groups by VOC profile alone would be considered highly significant and important. However, if no such discrimination were to be seen, the next best scenario would be the ability to rapidly distinguish between the groups by monitoring the volatile profile over time until divergence was seen. Whilst not as significant as positive results in the first arm, these experiments would also demonstrate a potential diagnostic method.

Selecting of the Optimal Microbiological Conditions
In order for any measurement of VOCs to be comparable between samples, the point at which the organisms are on their growth profile must be well described and understood. Broth cultures of E. coli were sampled over time using the SPME fibres to be used in the main study. For the optimisation of the incubation time, a laboratory DH5α strain of E. coli was used as an example organism. Results of the initial experiments on the evolution of the VOCs' profile over culture time can be seen in Figures 1-3. Figure 1 shows the number of different compounds detected at the different time points, Figure 2 shows the optical density and cell density of this strain at these time points, while Figure 3 shows the evolution of VOCs with time. The number of VOCs increased with the increasing incubation time.          The effect of different culture media on the VOC profiles produced by a wide variety of organisms has been well studied [21][22][23]. It is obvious that the substrate available to microorganisms greatly influences the VOCs they are able to produce, but other factors such as pH, salt concentration and temperature also have large effects. Depending on the manner in which the organisms are growing and multiplying, these initial conditions can change considerably as time progresses. In ex vivo samples, microbes are growing in competition with the host environment and although relatively steady states are common there are many confounding sources of VOCs. These include those produced by inflammatory host response [24], exogenous factors [25] and commensal organisms [26]. In vitro studies avoid these confounding factors, but introduce problems of their own. Most of these studies rely on headspace analysis of bacteria in planktonic culture. In the simplest case of a low cell count of a single strain of a microorganism inoculated into a small volume of liquid media, the organism will first pass though the lag phase. This is a distinct growth phase where the organism is acclimatising to the conditions into which it has been introduced and preparing for the next stage [27]. Next, the organism enters the log, or exponential growth phase. At first, the easiest to metabolise energy sources will be consumed [28] but as these substrates are depleted, hydrolytic pathways are induced [29] and catabolite repression is lifted [30]. It is during this phase that the number and quantity of VOCs produced increases. In fact, it has been shown that E. coli grown in batch culture in Luria−Bertani (LB) medium passes through four distinct phases in the log phase and each is likely to produce a different VOC profile [31]. It is likely that it is during this phase that the optimum time for volatile sampling is reached. As nutrients run out completely, the organism enters the stationary phase.
Next, nine VOCs representing different chemical groups and the compound characteristics for E. coli [32,33] were selected and the variation in the abundance of these compounds over time was followed ( Figure 4). A time course was performed to enable selection of a suitable time point for a diagnostic test. For this reason, the optimum sampling point was chosen to be at an optical density of approximately 1.6 at 550 nm. The representative chromatogram of the optimised conditions is shown in Figure 5, and the list of the tentatively identified VOCs is presented in Table 1.
Antibiotics 2020, 9, x FOR PEER REVIEW 5 of 13 The effect of different culture media on the VOC profiles produced by a wide variety of organisms has been well studied [21][22][23]. It is obvious that the substrate available to microorganisms greatly influences the VOCs they are able to produce, but other factors such as pH, salt concentration and temperature also have large effects. Depending on the manner in which the organisms are growing and multiplying, these initial conditions can change considerably as time progresses. In ex vivo samples, microbes are growing in competition with the host environment and although relatively steady states are common there are many confounding sources of VOCs. These include those produced by inflammatory host response [24], exogenous factors [25] and commensal organisms [26]. In vitro studies avoid these confounding factors, but introduce problems of their own. Most of these studies rely on headspace analysis of bacteria in planktonic culture. In the simplest case of a low cell count of a single strain of a microorganism inoculated into a small volume of liquid media, the organism will first pass though the lag phase. This is a distinct growth phase where the organism is acclimatising to the conditions into which it has been introduced and preparing for the next stage [27]. Next, the organism enters the log, or exponential growth phase. At first, the easiest to metabolise energy sources will be consumed [28] but as these substrates are depleted, hydrolytic pathways are induced [29] and catabolite repression is lifted [30]. It is during this phase that the number and quantity of VOCs produced increases. In fact, it has been shown that E. coli grown in batch culture in Luria−Bertani (LB) medium passes through four distinct phases in the log phase and each is likely to produce a different VOC profile [31]. It is likely that it is during this phase that the optimum time for volatile sampling is reached. As nutrients run out completely, the organism enters the stationary phase.
Next, nine VOCs representing different chemical groups and the compound characteristics for E. coli [32,33] were selected and the variation in the abundance of these compounds over time was followed ( Figure 4). A time course was performed to enable selection of a suitable time point for a diagnostic test. For this reason, the optimum sampling point was chosen to be at an optical density of approximately 1.6 at 550 nm. The representative chromatogram of the optimised conditions is shown in Figure 5, and the list of the tentatively identified VOCs is presented in Table 1.

Comparison of VOC Profiles of Resistant and Susceptible Bacteria
In order to verify the applicability of this method for distinguishing resistant and susceptible bacteria, the VOC profiles were detected at the optimised time point. The comparison study was performed based on the true replicates of organisms from different primary cultures on different days. In total there were 84 samples comprising 45 susceptible and 39 resistant isolates, allowing for multiple isolates of the 8 strains. Across all 84 samples a total of 85 VOCs (

Comparison of VOC Profiles of Resistant and Susceptible Bacteria
In order to verify the applicability of this method for distinguishing resistant and susceptible bacteria, the VOC profiles were detected at the optimised time point. The comparison study was performed based on the true replicates of organisms from different primary cultures on different days. In total there were 84 samples comprising 45 susceptible and 39 resistant isolates, allowing for multiple isolates of the 8 strains. Across all 84 samples a total of 85 VOCs (Table 2) were detected although some occurred very infrequently. Some 30 VOCs were detected in the headspace above the LB medium, used for bacterial culture, and this information was included in further analysis. The VOCs detected in the headspace above the bacterial cultures were characteristic for the E. coli species, with a predominant VOC being indole [32,33]. Bacteria produce a wide and diverse range of VOCs. These compounds may be produced as byproducts of metabolism, but they can also act as signalling molecules for communication between bacteria, or between bacteria and host. Although these interactions are not fully understood, they are thought to play an important role in antibiotic resistance, as described in a review by Bos et al. [34]. The statistical analysis showed that five volatiles (retention times (RT) expressed in min: 16.68, 18.5, 19.2, 19.47, 20.40) were positively associated with the samples from resistant strains with p < 0.05 in all five cases (p = 0.039, p = 0.011, p = 0.024, p = 0.030, p = 0.003, respectively). Two volatiles (RT: 24.36, 25.07) were positively associated with the samples from sensitive strains (p = 0.005, p = 0.045, respectively). In a multivariate analysis, the six volatiles with RT (mins): 16.68, 18.5, 19.47, 20.40, 24.36, 25.07 independently and jointly contributed to overall discrimination between susceptible and resistant strains with 85.7% overall accuracy, with almost the same accuracy under leave-one-out cross-validation. A total of 91.1% of samples from sensitive strains and 79.5% of samples from resistant strains were correctly classified and discriminant scores are summarised in Figures 6 and 7. The microbiological methods used here involved true replicates of organisms from different primary cultures on different days, leading to a relatively large variance between samples. The fact that a significant discrimination between the groups (AUROC 0.912) could be seen using only six volatiles is extremely encouraging.  Unfortunately, exact identification of the VOCs discriminating the susceptible and resistant bacteria was not possible, except for the tentative identification of butanoic acid and 2-dodecanone. The mass spectra of the other four VOCs are presented in the Figure S1. The abundance of butanoic acid was higher in the resistant bacteria, while the abundance of 2-dodecanone was higher in the resistant strains. Butanoic acid is the product of bacterial fermentation [35]. Interestingly, 2dodecanone was detected in E. coli strains after the incubation with biosilver nanoparticles, having bacteriostatic and bactericidal properties.  Unfortunately, exact identification of the VOCs discriminating the susceptible and resistant bacteria was not possible, except for the tentative identification of butanoic acid and 2-dodecanone. The mass spectra of the other four VOCs are presented in the Figure S1. The abundance of butanoic acid was higher in the resistant bacteria, while the abundance of 2-dodecanone was higher in the resistant strains. Butanoic acid is the product of bacterial fermentation [35]. Interestingly, 2dodecanone was detected in E. coli strains after the incubation with biosilver nanoparticles, having bacteriostatic and bactericidal properties. Unfortunately, exact identification of the VOCs discriminating the susceptible and resistant bacteria was not possible, except for the tentative identification of butanoic acid and 2-dodecanone. The mass spectra of the other four VOCs are presented in the Figure S1. The abundance of butanoic acid was higher in the resistant bacteria, while the abundance of 2-dodecanone was higher in the resistant strains. Butanoic acid is the product of bacterial fermentation [35]. Interestingly, 2-dodecanone was detected in E. coli strains after the incubation with biosilver nanoparticles, having bacteriostatic and bactericidal properties.
The results of this preliminary study are promising, and with development could lead to a practical diagnostic method for use in routine microbiology to distinguish between strains that do or do not carry ESBL. Currently, organisms are incubated for 24 h in a primary broth before transfer to a second standardised broth. However, infected urine typically contains greater than 10 5 cfu/mL of bacterial cells, this being two log fold below the optimum sampling point seen. If the aim is to rapidly screen for ESBL strains present in patient urine, then it is conceivable that this could be possible by enriching with broth and incubating the samples directly. In order to do this, it must be accepted that better characterisation of the discriminating volatiles is necessary.
The method of VOC detection by GC-MS has successfully identified important volatiles by retention time, but full identification has not been possible other than a chemical group (mostly alkanes). If further identification can be achieved, tailoring of the culture or enrichment media can be performed, promoting production of the VOCs of interest, thus increasing the sensitivity of the technique. It is also possible that simpler and cheaper methods of detecting the volatiles could be employed, such as those based around electrochemical sensing.

Bacterial Cultures and Sampling
To first establish the protocol for VOC sampling, 250 mL of Luria−Bertani (LB) broth was inoculated with a laboratory strain of E. coli and incubated overnight at 37 • C. This was then subcultured into a bank of headspace vials containing 3 mL of LB at a ratio of 20 µl/mL. These were reincubated at 37 • C with shaking. At 30 min intervals, SPME fibres were inserted into the headspace and transferred to a 55 • C water bath for one hour, stopping further growth and allowing adsorption of volatiles onto the fibre. Optical density of duplicate broths at each time point was measured at 550 nm and dilutions performed to keep the measurement in the optimum range for the spectrophotometer. Plate counts were performed on LB agar to give cell counts at these densities. After incubation, the fibres were analysed by GC-MS as described in 4.2. These results were used to gauge the optimum optical density at which to sample E. coli volatiles by SPME.
Following this, a selection of E. coli isolates were obtained from the University of Bristol, consisting of four wild type CTX-M ESBL positive urinary isolates from the community [36] and four ESBL negative laboratory strains. Single colonies from fresh overnight cultures of each organism on LB agar at 37 • C were used to inoculate 250 mL of LB broth. These were then incubated overnight at 37 • C with shaking. 200 µL of each primary broth was then transferred to 2 × 10 mL of LB broth in identical headspace vials. Turbidity of one of the vials was monitored by optical density at 550 nm until the required growth density was achieved. At this point the SPME fibre with Carboxen ® /Polydimethylsiloxane (CAR/PDSM) coating was inserted through the septum into the headspace and further incubated for 1 hour at 55 • C as before. The fibres were then retracted into their holders and the ends sealed before being placed in a transportable icebox at −15 • C. When all fibres had been exposed, the box was transported to the gas chromatography-mass spectrometer (GC-MS) laboratory. Four biological replicates of each strain were performed on different days from fresh primary agar culture each time.

VOC Analysis
After the extraction and transport, the fibers were immediately introduced into a gas chromatography injector port with 0.75 mm ID splitless glass liner (PerkinElmer, Inc., Waltham, MA, USA), set to splitless mode and a temperature of 240 • C. The thermal desorption of VOCs was carried out for 10 min to avoid carryover (data not presented). Analysis of VOCs was performed using a Perkin Elmer GC-MS Clarus 500 instrument (PerkinElmer, Inc., Waltham, MA, USA). The separation of VOCs was undertaken using a Zebron-624 column; 60 m × 0.25 mm × 1.40 µm, (Phenomenex, Torrance, CA, USA). The carrier gas was 99.9995% pure helium (AirProducts, Crewe, UK) at a constant flow of 1.5 mL/min. The oven temperature program was set as follows: 40 • C hold for 2 min, and increase to 240 • C at a rate of 10 • C/min and then maintained at the final temperature for 8 min, giving a total run time of 30 min. Mass spectra were obtained by electron ionisation acquired in full scan mode with a scan range of m/z 35-450 used for data acquisition. The operating conditions for the mass spectrometer were as follows: electron ionisation mode at an energy of 70 eV; transfer line and ion source temperatures were 280 • C and 180 • C, respectively. Total ion chromatograms (TIC) were analysed with the Turbomass software (PerkinElmer, Inc., Waltham, MA, USA) and compounds were expressed as the area under the curve. For the identification, the peaks were deconvoluted using the free Automated Mass Spectral Deconvolution and Identification System (AMDIS) software by the National Institute of Standards and Technology (NIST, Gaithersburg, MD, USA) and tentatively identified by comparison of the mass spectra with the NIST/EPA/NIH Mass Spectral Library (version 2.2, 2014, Gaithersburg, MD, USA). Only the components with a match factor >80% were listed by name in the Tables. Moreover, retention indexes for the detected compounds were calculated according to d'Acampora Zellner and co-authors [37]

Data Analysis
Comparison of the VOC profiles of ESBL positive and negative strains was performed at each retention time using the Mann−Whitney test. The peak areas of VOCs which significantly differed between sensitive and ESBL-producing strains were used in a multivariate assessment using linear discriminant analysis. A square root variance stabilising transformation of peak areas was undertaken prior to multivariate assessment in order to produce a robust model not prone to the presence of outliers. Leave-one-out cross-validation was used in the assessment of predictive accuracy.

Conclusions
The concept of rapidly detecting antibiotic-resistant bacteria by analysing VOCs is ambitious, but our results and those of others suggest the approach is promising. Differences in ESBL-positive and -negative isolates were rapidly distinguished by SPME with GC-MS in under 2 h, which compares well with 24 h or more using routine methods. Statistically significant differences were found between the VOC profiles of CTX-M-15 positive and negative bacteria. The study of VOCs and linking specific VOCs to certain antibiotic strains may help elucidate a better understanding of bacteria−host interactions in addition to allowing specific detection at earlier time points compared to conventional methods.
The next stage of this study will be to expand the VOC profile testing of ESBL strains of E. coli such as those carrying other CTX-M variants or those producing AmpC enzymes. It is also important to test strains with non-beta-lactam resistance profiles. Following this, attempts can be made to accurately categorise novel strains in a fully blinded study. Furthermore, with more complete identification of VOCs of note, it may be possible to develop specific enrichment media to allow direct examination urine samples as well as the sensor-based devices dedicated to the individual VOCs that can be used in a medical, point-of-care practice.
Supplementary Materials: The following are available online at http://www.mdpi.com/2079-6382/9/11/797/s1, Figure S1: The mass spectra of unknown compounds used for the distinguishing of resistant and sensitive strains.