Stable isotope profiles reveal active production of VOCs from human-associated microbes

Volatile organic compounds (VOCs) measured from exhaled breath have great promise for the diagnosis of bacterial infections. However, determining human or microbial origin of VOCs detected in breath remains a great challenge. For example, the microbial fermentation product 2,3-butanedione was recently found in the breath of Cystic Fibrosis (CF) patients; parallel culture-independent metagenomic sequencing of the same samples revealed that Streptococcus and Rothia spp. have the genetic capacity to produce 2,3-butanedione. To investigate whether the genetic capacity found in metagenomes translates to bacterial production of a VOC of interest such as 2,3-butanedione, we fed stable isotopes to three bacterial strains isolated from patients: two gram-positive bacteria, Rothia mucilaginosa and Streptococcus salivarius, and a dominant opportunistic gram-negative pathogen, Pseudomonas aeruginosa. Culture headspaces were collected and analyzed using a gas chromatographic system to quantify the abundance of VOCs of interest; mass spectroscopy was used to determine whether the stable isotope label had been incorporated. Our results show that R. mucilaginosa and S. salivarius consumed D-Glucose-13C6 to produce labeled 2,3-butanedione. R. mucilaginosa and S. salivarius also produced labeled acetaldehyde and ethanol when grown with 2H2O. Additionally, we find that P. aeruginosa growth and dimethyl sulfide production are increased when exposed to lactic acid in culture. These results highlight the importance VOCs produced by P. aeruginosa, R. mucilaginosa, and S. salivarius as nutrients and signals in microbial communities, and as potential biomarkers in a CF infection.


Introduction
Breath metabolites have incredible potential to serve as non-invasive diagnostics of bacterial infection and other diseases, but the origins of the metabolites are often unknown. Microbes contribute significantly to the complex mix of molecules present in the human body. An average human likely contains >200 000 metabolites, and half or more of them are produced by or altered by the microbes that reside in and on humans (Wikoff et al 2009, Sharon et al 2014. Furthermore, most of the genes that distinguish one person from another are encoded by the microbial community that inhabits each person; human-associated microbes encode at least an order of magnitude more genes than the human genome (McFall-Ngai et al 2013). A very early attempt to profile breath and urine metabolites by Linus Pauling at Caltech in 1971 showed that starving the gut microbes led to more consistent breath and urine profiles across study participants (Pauling et al 1971). More recent studies have sampled breath VOCs to diagnose or detect bacterial infections in pneumonia (Fijten et al 2015) and tuberculosis (Phillips et al 2010).
Some of the metabolites produced by humanassociated microbes are volatile, have the capacity to travel and affect the physiology of neighboring microbial and human cells, and can be captured by breath sampling (Létoffé et al 2014, Audrain et al 2015. For example, 2,3-butanedione was detected in the breath of Cystic Fibrosis (CF) patients (Whiteson et al 2014). One challenge in identifying the cellular origin of volatile organic compounds (VOCs) is determining whether the metabolites are produced by microbial or human cells. Humans and bacteria share several fundamental pathways and have the ability to produce many of the same volatile molecules, such as isoprene and acetone, the two most abundant VOCs detected in breath (Fenske and Paulson 1999, Miekisch et al 2004, Thorn and Greenman 2012. In the case of 2,3-butanedione, a subset of microbes encode the enzymes from this fermentation pathway, however 2,3-butanedione is not known to be a product of human metabolism. Parallel metagenomic sequencing from CF patients with 2,3-butanedione in their breath showed Streptococcus and Rothia spp. contain the acetoin pathway genes leading to the production of 2,3-butanedione (Whiteson et al 2014). A similar companion study also produced in this special issue of the Journal of Breath Research shows the taxonomy of microbial genes responsible for the production of acetaldehyde and ethanol (Bos et al 2016). To determine whether 2,3-butanedione and other volatile molecules of interest such as acetaldehyde and ethanol are actively produced by bacterial isolates from CF patients, we employed the use of labeled stable isotopes fed to in vitro cultures to track bacterial metabolism.
This study focuses on several bacterial isolates from CF patients that are often found in the airways when the natural lung clearance mechanisms are impaired. Therefore, the molecules that are produced by these bacteria are potentially relevant for CF, chronic obstructive pulmonary disorder, and pneumonia. CF patients' lungs are chronically infected with opportunistic pathogens that significantly alter the health and physiology of the patient. Different types of infections can lead to varying patient outcomes (Hauser et al 2011, Ahlgren et al 2015). Among the most dominant bacteria are Pseudomonas aeruginosa, Burkholderia species, Staphylococcus aureus, Stenotrophomonas maltophilia, and Streptococcus pneumoniae (table 1). Rothia mucilaginosa has also been found to be a dominant strain in culture-independent studies of CF airway microbial communities (Lim et al 2013), and is highly abundant in CF sputum samples (Blainey et al 2012, Fodor et al 2012, Paganin et al 2015. Our goal was to track microbial production of VOCs in clinically relevant CF bacteria. Specifically, we were interested in P. aeruginosa, R. mucilaginosa, and Streptococcus salivarius. P. aeruginosa is a widespread opportunistic pathogen in CF which becomes more dominant through time in decades-long chronic CF infections. The recent detection of 2,3-butanedione in the breath of CF patients, along with the genes necessary for acetoin metabolism from metagenomic sequencing of the same samples (Whiteson et al 2014), motivated us to see if we could verify that CF isolates are indeed producing 2,3-butanedione. Analysis of culture-independent metagenomic sequence reads from CF sputum showed that R. mucilaginosa and a variety of Streptococcus spp. encode genes involved in the production of 2,3-butanedione (Whiteson et al 2014), which induces virulence in P. aeruginosa (Venkataraman et al 2011, Whiteson et al 2014 and has also been found to be present in the breath of CF patients (Whiteson et al 2014). R. mucilaginosa is an abundant and understudied bacteria common in all human oral-microbial communities, including CF sputum, and it is known to produce the toxic volatile acetaldehyde (table 1) (Moritani et al 2015). Streptococcus spp. are also common in all human oral microbial communities. A large fraction of the metagenomic reads that map to the 2,3-butanedione pathway-25% to 85%-originated from Streptococcus (Whiteson et al 2014). The sequences from CF sputum mapped equally well to several strains of Streptococcus spp., including S. parasanguinus, S. salivarius and S. pneumonia. We included a strain of S. salivarius cultured from a CF patient which is capable of producing 2,3butanedione based on genome sequence data.
To track VOCs produced by microbial metabolism, we used stable isotopes (D-Glucose-13 C 6 and 2 H 2 O), sampled headspaces of in vitro cultures, and analyzed these headspace samples using a gas chromatography multi column/detector system. Although metagenomic evidence suggests that Streptococcus spp. and R. mucilaginosa may produce 2,3-butanedione (Whiteson et al 2014), this study used stable isotopes to confirm its production by CF isolates. We also set up cultures of a clinical isolate of P. aeruginosa to study VOC production with and without 2,3-butanediol and other physiologically relevant products of S. salivarius and R. mucilaginosa. Future investigation for optimal P. aeruginosa growth conditions in supernatants from other bacteria is one of our goals in understanding how P. aeruginosa is affected by the neighboring CF community. The airways of people with CF have a lower pH than healthy airways, and immediately prior to a pulmonary exacerbation event, the pH drops even lower (Tate et al 2002, Ojoo et al 2005, Pezzulo et al 2012, Quinn et al 2015. To investigate the effect of low pH on our clinical isolates, our experimental conditions also involved the use of lactic acid or buffer to adjust the media pH. In summary, results show that levels of 2,3-butanedione, acetaldehyde, and dimethyl sulfide are increased when the pH is decreased and these signatures can be used to distinguish the metabolisms of P. aeruginosa, R. mucilaginosa, and S. salivarius.

Materials and methods
Bacterial strains and growth conditions The bacterial strains chosen for this study were isolated from CF patients at the UCSD Adult CF Clinic: Pseudomonas aeruginosa PaFLR01, Rothia mucilaginosa RmFLR01, and Streptococcus salivarius SsFLR02. Strains were grown at 37°C for 24 h in Todd Hewitt (TH) broth (BD Bacto, Fisher Scientific). From the 24 h cultures, triplicate OD 500 0.05 cultures were started in TH broth with 30 mM D-Glucose-13 C 6 (Sigma-Aldrich and Cambridge Isotope Laboratory), unless otherwise noted. Some conditions involve the addition of 30% 2 H 2 O (Sigma-Aldrich) or 100 mM pH 6.5 potassium phosphate buffer (Fisher Scientific). We used OD 500 to measure cell densities, instead of OD 600 that is traditionally used, because P. aeruginosa strain PaFLR01 produces pyocyanin, a colorful phenazine that absorbs at 691 nm.

P. aeruginosa pyocyanin measurement
At the end of each experiment, P. aeruginosa cultures were centrifuged to pellet cells and supernatants were collected to read absorbance measurements at 691 nm with a spectrophotometer.
Headspace gas capture and gas chromatography A modified bioreactor and a gas chromatographic (GC) system were used for headspace gas collection and VOC analysis (Colman et al 2001, Umber et al 2013. In particular, replicates of three individual glass culture tubes were placed in the bulb of the bioreactor ( figure 1(a)) and, in order to replace any laboratory air, the reactor was flushed for 6 min at 1.5 l min −1 with ultra clean air collected at 10 000 feet at the Crooked Creek Research station in White Mountain (California). After flushing, the bioreactor was incubated at 37°C for 48 h statically and the headspace was collected and analyzed using a GC system as described in previous works (Colman et al 2001, Shin et al 2009, Umber et al 2013, Whiteson et al 2014. Briefly, headspace sample was cryogenically pre-concentrated in a stainless steel loop filled with glass beads. The sample was re-vaporized using The volume in each glass vial is 2 ml. Panel 1 is a simplified representation of our experiment controls. There are three media controls: Todd Hewitt (TH) broth with 30% 2 H 2 0, TH with 30 mM D-Glucose-13 C 6 , and TH with 30% 2 H 2 0 and 30 mM D-Glucose-13 C 6 . For the Pa control, we grew P. aeruginosa in TH with 30% 2 H 2 0. For the Rm and Ss controls, we grew each strain in TH with 30% 2 H 2 0 and 30 mM D-Glucose-13 C 6 . Panel II is an experiment set where we cultured P. aeruginosa with three metabolites: 2,3-butanediol, lactic acid, and pyruvate. When P. aeruginosa was cultured with 2,3-butanediol and lactic acid, TH and 30% 2 H 2 0 media was used. With pyruvate, TH and D-Glucose-13 C 6 was used as media. Panel III represents the supernatants from R. mucilaginosa and S. salivarius. We have supernatants from unbuffered and buffered pH 6.5 TH media.
hot water (at approximately 80°C) and split into five streams directed to a multi-column/detector GC system (all Hewlett Packard 6890) using UHP helium as carrier gas. The system consists of three Hewlett-Packard 6890 GC units. The first GC is equipped with two different column combinations output to an electron capture detector (ECD) and an FID; the second GC is output to an FID; and the third GC is equipped with two different column combinations and output to a quadrupole mass spectrometer detector working in selected ion monitoring (SIM) mode, and an ECD. The MSD is set to operate in SIM mode and scan mode simultaneously.
The different VOCs are expressed in part per trillion by volume. The average from the triplicate samples and a one-tailed student's t-test was used to identify VOCs that were significantly elevated compared to the media control.
Genome annotation P. aeruginosa and S. salivarius genomes were sequenced on the Illumina MiSeq platform at San Diego State University. R. mucilaginosa was sequenced on the Illumina MiSeq platform at microbesNG at the University of Birmingham. All sequenced genomes were uploaded onto RAST for annotation (Aziz et al 2008). Figure 1 displays the conditions of the experiment. All conditions were tested as biological triplicates in 2 ml volumes for 48 h, static. The media control included three conditions: TH broth with 2 H 2 O, TH broth with D-Glucose-13 C 6 , and TH with both 2 H 2 O and D-Glucose-13 C 6 . The P. aeruginosa control included two conditions: P. aeruginosa grown in TH with 2 H 2 O or TH with D-Glucose-13 C 6 and pyruvate. The R. mucilaginosa and S. salivarius controls included each strain grown in TH with 2 H 2 O and D-Glucose-13 C 6 (Panel I). P. aeruginosa was cultured in TH with the addition three metabolites: 20 mM 2,3-butanedione, 20 mM lactic acid, and 30 mM pyruvate (Panel II). We collected R. mucilaginosa and S. salivarius supernatants to perform cross-feeding experiments with P. aeruginosa, but P. aeruginosa did not grow. Therefore, we excluded that portion of the data set, but still kept the supernatants (Panel III) because the strains were cultured with a single labeled isotope, D-Glucose-13 C 6 .

Experimental design
CFC-11 is consistent across all samples Trichlorofluoromethane (CFC-11) is a VOC we do not expect to see produced by our strains and is constant across all samples (supplementary figure 2). Therefore, CFC-11 is a good negative control for the GC system.
Through fermentation, pyruvate can be converted to acetaldehyde by pyruvate decarboxylase (Pdc, E.C. 4.1. 1.1; figure 2). An example of an in vitro study of milk fermentation demonstrated through use of labeled isotopes that Streptococcus thermophilus was able to convert 13 C-glucose to acetaldehyde (Ott et al 2000). Surprisingly, while the CF isolates of R. mucilaginosa and S. salivarius produced acetaldehyde and ethanol, the cultures grown with D-Glucose-13 C 6 did not lead to production of 13 C-labeled acetaldehyde or ethanol (shaded areas of figures 3(b) and (c)). When R. mucilaginosa and S. salivarius are cultured in media with 2 H 2 O, both acetaldehyde and ethanol are labeled (unshaded areas of figures 3(b) and (c)). P. aeruginosa cultures did not produce any of the fermentation products shown in figure 3 in levels higher than the media controls. Labeled acetaldehyde was also detected in the media control when 2 H 2 O was present. We expected deuterium incorporation into acetaldehyde to depend on bacterial enzymes. However, acetaldehyde is a weak acid, and in the presence of base in the media the deuterium exchange with the alpha carbons of acetaldehyde could occur.
Acetaldehyde production may rely on acetate, acetyl-CoA, pyruvate, L-threonine, ethanol, or 2-deoxyribose-5-phosphate (Ott et al 2000). Our results suggest that acetaldehyde and ethanol production bypass glucose. Although both R. mucilaginosa and S. salivarius have the capacity to convert glucose to pyruvate, RAST annotations of the two strains suggest they do not have the genetic capability to convert pyruvate to acetaldehyde by pyruvate decarboxylase (E. C. 4.1.1.1; figure 2). Because the acetaldehyde and ethanol we detected in R. mucilaginosa and S. salivarius cultures are labeled in the presence of 2 H 2 O but not D-Glucose-13 C 6 , they may share a common pathway that involves the hydrolysis of 2 H 2 O.
Based on genome annotations (figure 2), P. aeruginosa, R. mucilaginosa, and S. salivarius are genetically capable of converting ethanol to acetaldehyde by alcohol dehydrogenase (E. C. 1.1.1.1) (figure 2). S. salivarius is also genetically capable of converting acetyl-CoA  Rm sup=spent medium from Rothia mucilaginosa after 48 h of growth in TH broth supplemented with D-Glucose-13 C 6 ; Ss=S. salivarius; Ss sup=spent medium from Streptococcus salivarius after 24 h of growth in TH broth supplemented with D-Glucose-13 C 6 . Error bars represent standard deviation from the mean. Shaded box represents conditions where only a single labeled isotope was used -D-Glucose-13 C 6 . Unshaded areas display samples that were grown in the presence of 30% 2 H 2 O and D-Glucose-13 C 6 . to acetaldehyde by acetaldehyde dehydrogenase (E.C. 1.2.1.10) (figure 2). P. aeruginosa has the genetic capability to convert acetate or acetyl-CoA to acetaldehyde through enzymes aldehyde dehydrogenase (E.C. 1.2.1.3) and acetaldehyde dehydrogenase (E.C. 1.2.1.10), respectively (figure 2). However, the culturing conditions did not lead P. aeruginosa to produce significant levels of acetaldehyde compared to R. mucilaginosa or S. salivarius, and acetaldehyde and ethanol were not labeled in the presence of 2 H 2 O in the P. aeruginosa cultures.
Decreasing the pH of TH broth to pH 6.5 by adding 100 mM potassium phosphate (figure 3) led to an increase in growth and an even greater increase in the production of 13 C labeled 2,3-butanedione and unlabeled acetaldehyde by R. mucilaginosa (figures 3(a) and (b); OD 600 of unbuffered cultures reached 0.73-0.76 while the OD 600 of the pH 6.5 culture reached 2.48-2.55). S. salivarius, on the other hand, showed higher 2,3-butanedione production in unbuffered media compared to buffered pH 6.5 media. To verify the presence and measure the concentrations of 2,3butandione in our samples, Voges-Proskauer assays were performed on supernatants (supplementary figure 1). The results were consistent with our gas chromatography findings, in that we detected high levels of 2,3-butanedione in the R. mucilaginosa supernatant ( figure 3(a) and supplementary figure 1).
Some limitations of studying pH dependent microbial physiology include the additional nutrients provided by the buffer (i.e., perhaps the phosphate buffer provides phosphate which boosts growth); buffered media may also eradicate micro-gradients relevant in the CF lung. For example, R. mucilaginosa produced more 2,3-butanedione in buffered pH 6.5 media while S. salivarius produced more when grown in unbuffered media. S. salivarius cultured in unbuffered media may include micro-gradients that are even more acidic than the buffered pH 6.5 media. Early dairy literature demonstrated that as pH becomes more acidic, Streptococci produce a higher yield of 2,3butanedione (diacetyl) (Michaelian et al 1938). Without the restriction of a buffered environment, S. salivarius has the ability to further decrease the pH of its micro-environment. Therefore, the local pH may be much lower in an unbuffered condition than the buffered pH 6.5 media, driving the physiology of a subpopulation of cells. While buffers are useful to control the pH, especially for bulk measurements of VOCs like those presented here, future studies using pH sensitive dyes and imaging may allow observation of the local spatial structure of interactions.
P. aeruginosa physiology is affected by exposure to microbial metabolites P. aeruginosa is a dominant opportunistic pathogen in CF and for this reason, to understand the polymicrobial community dynamics, it is essential to investigate its role and interactions with other bacteria and their metabolites. When P. aeruginosa was exposed to 2,3butanediol, lactic acid, and pyruvate in TH broth, not only did we detect production of dimethyl sulfide (DMS; figure 4(a)), we also detected production of pyocyanin in the culture ( figure 4(b)). DMS is a VOC that is well studied in the marine food web and is important in environmental sciences (Yoch 2002, Carrión et al 2015; DMS has also been detected in the breath of CF patients (Kamboures et al 2005), and P. aeruginosa has been shown to produce DMS and stimulate the growth of Aspergilles (Briard et al 2016). The stable isotope labeled substrates ( 2 H 2 O and D-Glucose-13 C 6 ) did not lead to label incorporation into DMS, potentially because glucose is not a preferred carbon source for P. aeruginosa, and DMS production may not depend on 2 H 2 O exchange.
Pyocyanin is an active redox phenazine produced by P. aeruginosa and found in the lungs of CF patients (Mentel et al 2009, Hunter et al 2012, Glasser et al 2014. Interestingly, DMS, pyocyanin production, and growth are higher when P. aeruginosa is cultured with 20 mM lactic acid (figures 4(a) and (b)). Lactic acid both lowers pH and is a preferred carbon source for P. aeruginosa, which could lead to better growth and therefore increased production of DMS, pyocyanin and other metabolic products. In our conditions, neither R. mucilaginosa nor S. salivarius produce DMS. As such, we find that among our strains, DMS production is unique to P. aeruginosa.

Production of acetone and isoprene
Acetone and isoprene are the two most abundant VOCs detected in breath and can be produced by either human or bacterial cells. In our study, we find that acetone was detected as a labeled volatile in each of the media controls and experimental conditions where 2 H 2 O was present ( figure 5(a)). Therefore, the production of acetone cannot be attributed to any of the strains because of its presence in the media controls. The use of 2 H 2 O is not a reliable method for identifying metabolic production of acetone and potentially other VOCs including butanal, 2-methylpropanal, 3-methylbutanal, 2-methylbutanal, and 2-butanone (data not shown) because the deuterium from the heavy water may be exchanging with hydrogens in the media.
Isoprene levels were higher in both P. aeruginosa and R. mucilaginosa cultures, and it was labeled when P. aeruginosa and R. mucilaginosa were cultured in media with 2 H 2 O ( figure 5(b)). Labeled isoprene was not detected in the media controls or in P. aeruginosa cultures lacking 2 H 2 O, suggesting that label incorporation was due to bacterial metabolism. It is not clear why isoprene levels were sometimes lower in R. mucilaginosa and S. salivarius cultures. P. aeruginosa, R. mucilaginosa, and S. salivarius share genes in the isoprenoid biosynthesis pathway, farnesyl diphosphate synthase (E.C. 2.5.1.10) and geranylgeranyl pyrophosphate synthetase (E.C. 2.5.1.29). P. aeruginosa also has several other genes involved in the isoprenoid pathway, but neither P. aeruginosa nor R. mucilaginosa have isoprene synthase (IspS, E.C. 4.2.3.27), the gene required to convert dimethylallyl diphosphate (DMAPP) to isoprene. There may be other mechanisms or pathways that allow P. aeruginosa and R. mucilaginosa to produce isoprene. Other bacteria that have been shown to produce isoprene are Bacillus subtilis (Julsing et al 2007, Xue andAhring 2011) and Escherichia coli (Zhao et al 2011, Zurbriggen et al 2012. Although isoprene was detected as labeled in R. mucilaginosa control cultures, it was not detected in the supernatants. Further investigation is needed to understand the role of microbial metabolism in isoprene production.

Conclusions
Analysis of VOCs in breath often leads to the detection of molecules that could be of human or microbial origin. Our microbial culture experiments using stable isotope labeled substrates demonstrate that microbial metabolism is responsible for the production of several VOCs found in breath, including 2,3-butanedione, acetaldehyde, ethanol, and isoprene. GC analysis and Voges-Proskauer test results have confirmed that clinical isolates R. mucilaginosa and S. salivarius produce 2,3-butanedione, which has previously been supported by metagenomics data (Whiteson et al 2014). Acetaldehyde is an important molecule for human health because it is a toxic carcinogenic molecule that is produced by members of the oral microbial community. Although it can be derived from pyruvate, acetaldehyde production can start from other substrates, including acetyl-CoA, acetate, 2-deoxyribose-5-phosphate, ethanol, and L-threonine (Ott et al 2000). We found that our P. aeruginosa strain produced DMS, which is more abundant when P. aeruginosa is cultured in TH supplemented with lactic acid. Finally, we found that labeled isoprene was produced by P. aeruginosa and R. mucilaginosa. Isoprene is one of the most abundant VOCs detected in breath, but when compared to other studies that measure concentrations of isoprene, our strains produced much less than what has been detected in breath Media=average values of three media controls; Pa=P. aeruginosa; Rm=R. mucilaginosa; Rm sup=spent medium from R. mucilaginosa after 48 h of growth in TH broth supplemented with 30 mM D-Glucose-13 C 6 ; Ss=S. salivarius; Ss sup=spent medium from S. salivarius after 24 h of growth in TH broth supplemented with 30 mM D-Glucose-13 C 6 . Error bars represent standard deviation from the mean. Shaded box represents conditions where 30 mM D-Glucose-13 C 6 was used. Unshaded areas display samples that were grown in the presence of 30% 2 H 2 O or 30% 2 H 2 O and 30 mM D-Glucose-13 C 6 . (King et al 2009, Khan et al 2013. This approach is promising for teasing apart the interactions that affect breath VOC production in human-associated microbial communities. For example, future studies that involve feeding stable isotope labeled substrates to one type of bacteria, and cross-feeding the metabolic products to another bacteria, could help us understand metabolic interdependencies and their VOC signatures in polymicrobial infections. As non-invasive diagnoses are being developed, identifying the contributing bacteria of an infection in CF is ideal for employing more specific and timely treatments. Lab results may take several days while results from breath samples can be instantaneous. Studies focused on the microbial origin of volatile molecules are critical for using breath data for earlier diagnosis and targeted treatments of infections.