Towards measuring growth rates of pathogens during infections by D2O-labeling lipidomics

RATIONALE Microbial growth rate is an important physiological parameter that is challenging to measure in situ, partly because microbes grow slowly in many environments. Recently, it has been demonstrated that generation times of S. aureus in cystic fibrosis (CF) infections can be determined by D2O-labeling of actively synthesized fatty acids. To improve species specificity and allow growth rate monitoring for a greater range of pathogens during the treatment of infections, it is desirable to accurately quantify trace incorporation of deuterium into phospholipids. METHODS Lipid extracts of D2O-treated E. coli cultures were measured on LC-ESI-MS instruments equipped with TOF and Orbitrap mass analyzers, and used for comparison with the analysis of fatty acids by isotope-ratio GC-MS. We then develop an approach to enable tracking of lipid labeling, by following the transition from stationary into exponential growth in pure cultures. Lastly, we apply D2O-labeling lipidomics to clinical samples from CF patients with chronic lung infections. RESULTS Lipidomics facilitates deuterium quantification in lipids at levels that are useful for many labeling applications (>0.03 at% D). In the E. coli cultures, labeling dynamics of phospholipids depend largely on their acyl chains and between phospholipids we notice differences that are not obvious from absolute concentrations alone. For example, cyclopropyl-containing lipids reflect the regulation of cyclopropane fatty acid synthase, which is predominantly expressed at the beginning of stationary phase. The deuterium incorporation into a lipid that is specific for S. aureus in CF sputum, indicates an average generation time of the pathogen on the order of one cell doubling per day. CONCLUSIONS This study demonstrates how trace level measurement of stable isotopes in intact lipids can be used to quantify lipid metabolism in pure cultures and provides guidelines that enable growth rate measurements in microbiome samples after incubation with a low percentage of D2O.


INTRODUCTION
Bacteria continually react to diverse stimuli, such as the availability of nutrients and electron acceptors, exposure to antimicrobial drugs or attack by the immune system. However, measuring microbial metabolites and growth rates within a complex environment still poses many technical challenges. Two recent advances in microbial ecology are beginning to make measuring average growth rates in environmental samples possible. The first advance is based on metagenomic DNA sequencing and takes advantage of the observation that growing cells yield more sequencing reads at genomic regions near the origin of replication [1,2] . This method is applicable to any microbial species in a microbiome as long as its assembled genome has a high sequence coverage. The second advance uses isotopic labeling to determine the biosynthesis rates of microbial lipid metabolites by mass spectrometry [3,4] . Stableisotope probing has a larger dynamic range than sequencing and can be used to quantify the slow growth rates that microbes have under environmental conditions. A limitation of stableisotope probing, however, is the identification of metabolites that are diagnostic for a specific microorganism. It is therefore desirable to combine isotopic labeling with a method such as lipidomics, which can detect a large number of microbial metabolites.
Lipids have been used for decades in ecology as markers of microbial metabolism, where they reveal information about viable biomass, nutritional status or changes of the microbial community structure [5,6] . Also, lipids can still be analyzed long after nucleic acids and peptides are degraded [7] . In order to estimate the growth rates of microbes, the active production of strain and genusspecific lipid metabolites can be measured with stable isotope labeling [3] . With advances in softionization mass spectrometry we can now attempt to combine isotope quantification and lipidomics for the study of microbes in situ .
Softionization mass spectrometry detects thousands of lipids in environmental extracts and would in principle be wellsuited to quantify the biosynthesis of lipid biomarkers by itself [8][9][10][11] .
However, extraction yields and ionization efficiency vary widely between samples [12][13][14][15][16] . This Ratios of isotopes can be measured with high accuracy by mass spectrometry, partly because ratiometric readouts vary less than absolute ion intensities [17] . For lipid biosynthesis, the incorporation of an isotope tracer such as 13 Clabeled substrates hence can provide a robust way to quantify anabolic activity and lipid turnover [18,19] . In microbiome samples bacteria differ widely in their ability to take up carbon sources and gases such as CO 2 , depending on their genetic capabilities and metabolic states. D 2 O is a nondiscriminating tracer of de novo lipid biosynthesis and thus often better suited for microbiome studies. D 2 Olabeling has recently been used to estimate in situ growth rates of Staphylococcus aureus in chronically infected lungs. After labeling of expectorated sputum with D 2 O, the deuterium enrichment of anteiso fatty acids was quantified using GC pyrolysis isotoperatio MS (GC/P/IRMS) in order to estimate the growth rate of the pathogen [3] . Previous studies have been used to study lipid biosynthesis with D 2 O in vivo [20][21][22][23][24][25] .
Environmental samples, however, provide particular challenges. For example, microbes often grow slowly in situ and one can expect low rates of deuterium incorporation into lipids [3] . Whether trace levels of incorporation can reliably be detected remains to be studied before deuterium incorporation can be used to determine lipid biosynthesis rates in situ by lipidomics.
In this study we apply stable isotope probing with D 2 O and measure deuterium incorporation by MSbased lipidomics. This approach can be used to obtain labeling rates for individual intact lipids in environmental samples, where microorganisms typically grow slowly. We begin by characterising several technical aspects that have to do with quantifying low levels of deuterium labeling. We then refine our application of D 2 Olabeling lipidomics by tracking of lipid labeling in E. coli during the stationarytolog phase transition. This reveals lipids that have distinct labeling dynamics that are not obvious from measuring absolute analyte concentrations alone.
Lastly, we test D 2 Olabeling lipidomics in a clinical context with the aim to measure the growth of 105 110 115 S. aureus in cystic fibrosis lung infections. In sum this study establishes principles for how growth rates of microbes in situ can be estimated by stable isotope probing lipidomics.

RESULTS
When cells grow after addition of heavy water, newly synthesized biomass will contain more D.
This also means that each lipid pool will be a mixture of molecules that vary more in their D abundance. The introduced heterogeneity causes broadening of chromatographic peaks, which could skew the isotope ratio observed by LCMS as ionization efficiency varies over time [26,27] . In order to evaluate how this affects isotope quantification by lipidomics, we grew an E. coli culture in F WATER ) and measured lipids after chromatographic separation using an ESITOF mass spectrometer [28] .
E. coli has a comparatively simple lipid composition and its lipid metabolism has been studied for decades [29] . The bacterium therefore provides a solid model system to develop and test methods for stable isotope labeling lipidomics. E. coli lipid extracts contains mainly phosphatidylethanolamines (PE), phosphatidylglycerol (PG) and cardiolipins (CL) [30] . When fully labeled in 4 % D 2 O, the molecular ions from PE and PG lipids extend over a range of 8 m/z . As expected, labeling causes a shift of retention time ( Figure 1A). Strongly deuterated molecules elute earlier than lighter ones ( Figure 1B). The maximum shift in retention time is about half of the chromatographic peak width, which indicates that all molecular species have overlapping elution profiles ( Figure 1C). We expect that this degree of shifting typically does not alter isotope ratios, as long as a moderate amount of labeling is used and the mass spectrum is integrated over a sufficiently large retention time window.
The quantification of D in intact lipids is complicated by 13  Values lower than 4 % are expected because hydrogen atoms from the unlabeled carbon source are incorporated into the lipids and biosynthetic enzymes favor 1 H over D due to kinetic isotope effects.
Note that this calculation assumes that N and Obound hydrogen atoms equilibrate fully with water during extraction and chromatography [31] . Additionally, the natural level of D, which is about 0.015 %, is neglected for the purpose of this study.
In order to evaluate the utility of D 2 Olabeling lipidomics for estimating bacterial growth rates, we grew E. coli cultures in glucose minimal medium ranging from 0.0156 % (natural abundance) to 4 % 2 F WATER and quantified the glycerophospholipids PE and PG. 2 F LIPID increases linearly (R 2 > 0.99) with 2 F WATER ( Figure 3A and B). Analyzing the same samples on a Q Exactive Plus Orbitrap operated at R=35,000, yields nearly identical slopes. Assuming a detection limit of 0.03 % 2 F LIPID for D 2 Olabeling lipidomics, we suggest that incubating cells for 15 minutes with 5 % 2 F WATER is a useful range to quantify lipid biosynthesis from microbes growing at one doubling per day ( Figure 3C). These boundary conditions indicate that D 2 Olabeling lipidomics can be developed further into a method to estimate microbial growth rates in environmental samples [3] . An important consideration for microcosm incubations is that two separate populations of molecules cooccur after labeling, a pool that contains low natural D abundance and a new pool that is enriched in D. High labeling strength would create molecules that occur further away from the monoisotopic mass in the spectrum and become difficult to quantify.
To test for H/D exchange we compare UPLCESITOF with GC/P/IRMS, which quantifies nearnatural isotopic composition of fatty acids [35] . Albeit the two methods are distinct in many ways, they should yield a similar linear relationship between 2 F WATER and 2 F LIPID [36,37] . For lipidomics, we determine an average slope for intact lipids produced by E. coli in glucose minimal medium of 0.577±0.003 (Figure S1; 2 F WATER between 0.125 and 4 %). Slightly higher slopes have been reported previously for E. coli fatty acids using GC/P/IRMS (0.65±0.04 for C16:0, 0.60±0.02 for C16:1 and 0.63±0.03 for C18:1) [37] . Growth on acetate raises 2 F in E. coli fatty acids analyzed by GC/P/IRMS, and it does so also for intact lipids measured by LCMS ( Figure S1). Overall, we obtain comparable slopes by lipidomics and GC/P/IRMS. hydrogens, which is in line with prior assessments of Cbound hydrogen exchange [38] . Together these tests imply that for lipids labeled well above natural Dabundance, no relevant artifacts of  Figure 5B and C). The formation of CFAs in E. coli is a postsynthetic modification of the unsaturated phospholipids that occurs predominantly as cultures enter the stationary phase. CFA synthase has an unusual regulation that involves enzyme instability as well as transcription of the cfa gene from two distinct promoters [39,40] . This means that, although CFA synthase is synthesized at basal levels throughout the growth curve, a transient spike in activity occurs during the logtostationary phase transition. In agreement with this regulation, CFAs largely dilute out during stationarytolog phase transition. Using D 2 Olabeling lipidomics we detect small levels of production of CFA lipids as well as D incorporation, which shows that CFA lipids were actively made during stationary phase exit ( Figure 5D).
Untargeted labeling reveals striking differences between phospholipids. Here we describe the uL scenario in detail. Some D 2 Olabeling patterns fit an exponential growth model ( Figure 6).
Other lipids, in particular CFAcontaining lipids, were inconsistent with simple exponential de novo production. For them the growth model needs to be extended. We include a parameter that accounts for lipid biomass in the inoculum that is inactive, i.e. not exponentially reproduced during the stationarytolog phase transition (see Methods for details).
The isotopic labeling patterns of E. coli phospholipids are dominated by their two fatty acyl chains, as they contain most of the Cbound hydrogens. A major trend we notice is that lipids that contain unsaturated fatty acids label rapidly, while fully saturated lipids incorporate label more steadily ( Figure 6; also Figure S3 and S4). In E. coli , unsaturated fatty acids are made during de novo fatty acid biosynthesis and not generated by modification of saturated fatty acids or phospholipids [29] . The faster labeling of unsaturated lipids we observe thus likely reflects that the unlabeled inoculum contained little unsaturated phospholipids, because most got converted into CFA during stationary phase. A second common trend is that most CFAcontaining lipids show slow initial increase of 2 F LIPID and often do not reach full saturation levels. This reflects that CFA lipids are only produced in small quantities after inoculation and hence a large proportion of is stimulated so that it gains label rapidly and to high levels ( Figure 6). This lipid occurs only in trace amounts in stationary phase, as does its precursor PG(14:0/16:1). Therefore, the material produced during outgrowth of the cultures is highly labeled and dominates the PG(14:0/16:0(Cp)) pool. Overall, these tests indicates that D 2 O addition allows a readout of how much of the material has been newly synthesized even for minority components, whose absolute concentrations can be challenging to quantify in complex lipid extracts. that anteiso fatty acids are produced also by other bacterial species. In the context of CF sputum, Prevotella melaninogenica and Stenotrophomonas maltophilia are relevant sources of anteiso C15:0 and anteiso C17:0 fatty acids in some CF patients [3,41,42] . Certain phospholipids, specifically those that contain anteiso fatty acyls, may therefore be more specific markers of S. aureus in CF infections and could be used to assess activity of the pathogen by lipidomics. To evaluate this hypothesis, we analyzed samples that had been collected and characterised as part of a longitudinal study of CF patients undergoing pulmonary exacerbations [42] .
A lipid that is appears wellsuited to monitor the growth of S. aureus is PG( a C15:0/ a C17:0). This compound was detected in lipid extracts of S. aureus and its structure assigned based on the m/z of the molecular ion in positive and negative ionization mode as well as MS/MS fragmentation spectra. Subsequently, signals from this lipid were also detected in expectorated sputum from several CF patients with S. aureus infection (Figure 7). Two patient whose lung infections did not contain S. aureus showed no signal corresponding to PG( a C15:0/ a C17:0). Based on these observations PG( a C15:0/ a C17:0) in CF sputum appears to be an specific marker for S. aureus in CF sputum.
Microbial lipid metabolites make up only a minute fraction of the total lipid content of CF sputum. The high sensitivity of ESI mass spectrometry allows detection of trace components, however, we had to use concentrated lipid extracts to yield sufficiently high signal intensities for the target analyte. In order to minimize contamination of the mass spectrometer, we only collected MS data at retention times that are needed to detect abundant phospholipids of S. aureus (58 min.).
Labeling of CF sputum with 4% D 2 O for 1 hour resulted in 0.054±0.04 at% D enrichment of PG( a C15:0/ a C17:0). This value can be used to estimate that the average generation time of S. aureus was approximately one cell doubling per day. This estimation is based on a previously established procedure that takes into account diffusion of the label, cell maintenance and other factors [3] . For comparison, the generation time estimated by the D/H ratio of anteiso C15:0 fatty acid in this sample was 3.3 days [42] . The slower estimate based on GC isotoperatio MS could for example be caused by contributions of anteiso C15:0 from other sources or variability in the production rates of anteiso C15:0 containing phospholipids in S. aureus . In summary, these initial tests indicate that it is possible to measure the activity of microbial pathogens in situ by D 2 Olabeling lipidomics. Its main benefits are that LCMS has increased species specificity, requires smaller sample amounts, it is faster than alternative MS methods [43] . Furthermore, D 2 Olipidomics can be performed on instrumentation that is available in many biomedical laboratories.

CONCLUSIONS
The combination of D 2 Olabeling and lipidomics allows a robust isotope ratio measurement, which reveals dynamic aspects of biosynthesis not accessible from absolute concentrations alone. The technology is also sufficiently sensitive to be adapted for environmental samples. ( Figure S5). M1/M0 could be a simple readout of D incorporation in environmental samples. When we assume an excess of unlabeled over labeled lipid, as it is the case for many environmental incubations, we anticipate an optimal labeling strength that causes the greatest change of the M1/M0 ratio. This is achieved when the ∆MW of the newlymade lipid is about +1.5 Da. Overall, a concentration of 23 % 2 F WATER seems most suited for environmental microcosm incubations. The optimal value will depend on the complexity of the lipid sample, i.e. whether D incorporation can be assessed from isotopologue distributions or M1/M0 ratio. Another consideration is that the fraction of D that enters the lipid varies with microbial metabolism [37,44] . We estimate that the combination of D 2 Olabeling and lipidomics as used here can roughly quantify growth rates greater than one cell doubling per day after labeling for 15 minutes. Currently, differences in the lipid composition between microbes are already used to identify strains by chemotaxonomy [45,46] . By combining largescale lipid detection with the quantification of isotopic labeling, new applications might become possible. These include identifying microbial adaptations to drugs, determining instantaneous microbial growth rates and forecasting composition of microbial community composition after exposure to a stressor.
Recording isotope labeling dynamics of lipids can help to rationalize microbial lipid function and metabolism. These efforts will benefit from related lines of research in environmental microbiology 320 325 330 and in human physiology that measure the synthesis and turnover of lipids with isotope labeling lipidomics, mass isotopomer distribution analysis or biomarkers analysis [8,21,47] .

Deuteriumenriched growth medium
M9 minimal medium was prepared with 3.8 μM thiamine pyrophosphate and glucose (22.2 mM) or sodium acetate (15 mM) [48] . All media were sterilized by filtration (0. These were in turn calibrated against the VSMOW, GISP, and SLAP international standards [49] . More enriched samples were measured against working standards made inhouse, ranging from 0.050 % to 0.150 % 2 F WATER . The presence of doublysubstituted species (DOD) was not taken into consideration due to fast equilibration of water molecules below 0.0150 %. Samples beyond this scale were no longer in the linear response range of the instrument, and we calculated 2 F WATER based on the gravimetric preparation of the medium. Their 2 F values were confirmed by water isotope analysis after dilution with natural abundance water of known isotopic composition.

E. coli cultures
Escherichia coli K12 (FRAG1) was streaked on LB agar plates for single colonies and used to inoculate 6 mL precultures of M9 minimal medium with glucose as carbon source [50] . Absorbance was recorded at 562 nm on a plate reader (Synergy 4, BioTek).

Liquid chromatography mass spectrometry (LCMS)
For LCMS analysis lipids were extracted based on the procedure by Matyash et al. [51] . Cell pellets LCMS data were collected on an Acquity IClass UPLC coupled to a Xevo G2S TOF mass spectrometer (Waters). Intact polar lipids were separated on an Acquity UPLC CSH C18 column (2.1 mm × 100 mm, 1.7 μm; Waters) at 55°C following a protocol established by Waters Corporation and adapted in our laboratory [28] . Samples were run in three randomized instrument replicates (injection volume 5 μL). LCTOFMS E data was collected in positive and negative mode using electrospray ionization (ESI) with a desolvation temperature of 550°C and source temperature of 120 °C.
Lipids were identified by the mass to charge ratio ( m/z ) of their molecular ion, their fragmentation products in positive and negative mode, and comparison to representative standards.
Note that that the assignment of sn 1 and sn 2 fatty acyl positions is tentative [30] . Lipids are named based on LIPIDMAPS classification [52] . Table S1 provides a summary of ions and retention times used for quantification. The monoisotopic intensity alone yields incorrect values for absolute lipid concentrations in labeling experiments, where some of the signal has shifted to higher masses. So we quantified the absolute concentrations of labeled lipids by comparing the sum of all intensities of its isotopologues with those of an internal standard ( Figure S6). Peaks in extracted ion chromatograms were integrated using the software MAVEN [53] . Subsequent analysis was done in R [54] . Models of isotopic distribution patterns were calculated using the R package Isopat (Martin Loos, EAWAG, Switzerland).
In order to calculate the generation time we fit our data to an exponential growth model. We samples is published elsewhere [42] . Samples used for LCMS were from Patients 1 (2nd hospitalization, day 5; method development), Patient 2 (day 2; growth rates). S. aureus negative controls: Patient 5 (day 1), Patient 7 (day 18 and 19).
Lipid extracts from S. aureus (grown in LB medium) were prepared and analyzed by LCMS as for E. coli . For sputum, 10 mg of lyophilized material was extracted by the same method. The lipid extract was dissolved in 100 µL methanol and 0.2 µL were injected for routine lipidomic profiling. Detection of anteiso containing phospholipids in sputum was performed by injecting 5 µL lipid extract and restricting flow into the ESI source to a retention time window of 58 min.

Calculation of deuterium content in lipids ( 2 F LIPID )
The mass spectra of a labeled and unlabeled lipid were compared to determine the fractional abundance of D ( 2 F LIPID ). We calculated the molecular weights of the two isotopologue distributions and dividing their difference by the number of Cbound hydrogen atoms. Note that we did not calculate the molecular weights using accurate masses. At 30,000 mass resolving power each signal M1, M2, etc contains isotopologues which differ slightly in mass. Measured masses have additional experimental uncertainties. For simplicity, we rather used the fact that each isotopologue must have gained a certain number of neutrons, for example M8 has gained a total of 8 neutrons from 13 C, 2 H etc. We used the isotopologue distribution to calculate by how many neutrons the distribution had shifted with respect to the monoisotopic mass M0. This approach eliminates inaccuracies. It also avoids the complicating fact that the mass difference between a D and 1 H is not exactly 1.

GC pyrolysis isotoperatio mass spectrometry
20 mg of frozen and lyophilized cell pellet was transesterified and extracted in hexane/anhydrous methanol/acetyl chloride at 100°C for 10 minutes [55] . The extract was concentrated under N 2 . Fatty acid methyl esters (FAMEs) were first analyzed by gas chromatography mass spectrometry The δD of FAMEs was measured by gas chromatography pyrolysis isotoperatio mass spectrometry (GC/P/IRMS) on a ThermoScientific DELTA plus XP with methane of known isotopic composition as the calibration standard [37] . Chromatographic conditions were identical as for regular GCMS, and peaks were identified by retention order and relative height. Samples were analyzed in triplicate. All data were corrected for methyl H originating from methanol by analyzing the dimethyl derivative of a phthalic acid standard, for which the δD value of ring H is known. For comparison with LCMS, δD values were converted into fractional abundances ( 2 F ).   to be performed to achieve deuterium enrichment of +0.03 at% D in lipids, assuming that the newly made lipid fraction ( 2 F NEW LIPID ) gets labeled at 1 % (red) or 5 % (blue). For a microbe growing with a doubling time of one day, incubation would need to be performed for 12.5 min or 1.1 hours, respectively, to reach en enrichment of 0.03 % over natural 2 F LIPID . We here estimate the enrichment at time t using the following equation: 2 F LIPID (t) = 2 F NEW LIPID * (1 2 t/GT ) .  synthesized from two molecules of PG and often increased in stationary phase [56] . In this time course CL was 3.5 mol% in the inoculum and 1.52.5 mol% during outgrowth. (D) Time course for PE and PG lipids with 0 (green), 1 (brown) or 2 cyclopropyl rings (violet) in cultures uL and lU .
Size of the data points represents deuterium abundance in the lipid ( 2 F LIPID ).  and unlabeled sputum (red). From the D/H of 0.05% generation times of S. aureus on the order of one cell doubling per day can be inferred [3] .  Protein abundance grew disproportionately during initial cell expansion, likely owing to the fact that stationary phase cells were smaller than those in exponential phase. In exponential phase, the three methods yield generation times of 43.5 minutes for protein (green squares) as well as lipids (magenta circles) and 50 minutes for OD 600 (blue triangles). Data from a single culture ( uU culture) are shown.