Raman 18O‐labeling of bacteria in visible and deep UV‐ranges

Raman stable isotope labeling with 2H, 13C or 15N has been reported as an elegant approach to investigate cellular metabolic activity, which is of great importance to reveal the functions of microorganisms in native environments. A new strategy termed Raman 18O‐labeling was developed to probe the metabolic activity of bacteria. Raman 18O‐labeling refers to the combination of Raman microspectroscopy with 18O‐labeling using H218O. At an excitation wavelength of 532 nm, the incorporation of 18O into the amide I group of proteins and DNA/RNA bases was observed in Escherichia coli cells, while for an excitation wavelength electronically resonant with DNA or aromatic amino acid absorption at 244 nm 18O assimilation was detected using chemometric tools rather than visual inspection. Raman 18O‐labeling at 532 nm combined with 2D correlation analysis confirmed the assimilation of 18O in proteins and nucleic acids and revealed the growth strategy of E. coli cells; they underwent protein synthesis followed by nucleic acid synthesis. Independent cultural replicates at different incubation times corroborated the reproducibility of these results. The variations in spectral features of 18O‐labeled cells revealed changes in physiological information of cells. Hence, Raman 18O‐labeling could provide a powerful tool to identify metabolically active bacterial cells.

isotope labeling is based on the replacement of single atoms (eg, 12 C, 14 N, 1 H and 16 O) by their stable heavier isotopes ( 13 C, 15 N, 2 H and 18 O) in biomolecules, such as proteins, lipids and nucleic acids [7,21]. This isotopic substitution leads to a red-shift of the vibrational frequency of the functional groups involved, that is, a shift toward lower wavenumbers, due to increased atomic mass [7,21]. Owing to the benefit of frequency shift, the heavier isotopes might serve as tracers to track the metabolic activity of microorganisms [7,[22][23][24][25][26][27].
The characterization of isotopically labeled bacteria using Raman spectroscopy with visible excitation wavelengths has been reported. For instance, Yang et al. amended clinical samples with heavy water (D 2 O) for a rapid antibiotic susceptibility testing of bacteria causing urinary tract infections [6]. By evaluating the amount of incorporated deuterium in bacteria, the authors could differentiate between antibiotics susceptible and resistant bacteria. The metabolic activity of resistant bacteria was not inhibited by the tested antibiotics thus, resistant bacteria exhibited more deuterium accumulation than sensitive bacteria [6]. Tao et al. used the same deuterium labeling approach to test the efficacy of drugs to inhibit the metabolic activity of oral bacteria [4]. The study reported on the detection of viable but not culturable (VBNC) cells, since the antimicrobial agents inhibited the bacterial growth of VBNC cells but not their metabolic activity [4]. Hence, the VBNC cells assimilated a considerable amount of deuterium from D 2 O. The study pointed the VBNC cells as the cause of many latent and recurring infections [4]. Olaniyi et al. identified cellulose degrading bacteria in a soil microbial community using deuterium labeling [12]. They found a strong correlation between ATP production and deuterium assimilation. From this they concluded that the larger the amount of accumulated deuterium in the bacterial cells, the higher the metabolic activity of cells [12]. Taubert et al. tracked active bacteria in groundwater using Raman deuterium labeling [14]. After a subsequent genomic analysis, the authors were able to link active bacteria to their ecological function in situ [14]. Kumar et al. demonstrated the mechanism of carbon catabolic repression in naphthalene degrading bacteria by combining two distinct labeled substrates, namely 13 C-glucose and fully deuterated glucose [10]. Cui et al. identified N 2 -fixing soil bacteria owing to 15 N induced Cyt c band shifts [28]. Based on the positive correlation found between the Cyt c band shift and the 15 N percentage, the authors could quantify the extent of N 2 fixation of various soil bacteria [28]. Angel et al. optimized the 15 N tracer method by tracing 15 N incorporation into microbial RNA [29]. The optimized 15 N-RNA method improved the sensitivity to identify soil bacteria involved in N 2 -fixation [29]. Up to now, only the isotopes 13 C, 2 H and 15 N have been employed in Raman stable isotope labeling of microorganisms. To the best of our knowledge studies with stable isotopes and bacteria using deep UV-resonance Raman spectroscopy at 244 nm have never been reported.
The present study combines for the first time 18 O labeling and Raman microspectroscopy for the characterization of metabolically active bacterial cells. We hypothesized that the accumulation of 18 O could be detected in proteins and nucleic acids. In order to do so, we combined the advantages of non-resonant Raman microspectroscopy at 532 nm and UV-resonance Raman spectroscopy at 244 nm excitation wavelength, to analyze the phenotypic features of bacterial cells exposed to H 2 18 O. The Raman excitation wavelength of 532 nm is chosen to obtain information about proteins and the excitation wavelength of 244 nm to obtain information about both nucleic acids and the aromatic amino acid composition in bacterial cells. For single cells Raman measurements in the visible range (532 nm) 1 mL of the bacterial culture was centrifuged for 5 min at 5000 g and 4 C. The supernatant was removed, and the pellet was resuspended in 1 mL of distilled water to discard the medium. Three washing steps with distilled water were performed. The resulting pellet was resuspended in 1 mL distilled water. Then 10 μL of the resulting suspension was spotted onto nickel foil and allowed to air dry.
For bulk Raman measurements in the deep UV range (244 nm) 5 mL of the bacterial culture was centrifuged for 5 min at 5000 g and 4 C. The washing steps were analog to those of the measurements at 532 nm. The pellet resulting from the last washing step was resuspended in 10 μL distilled water and the cell suspension was spread on a quartz slide and dried at room temperature.

| Raman measurements
The Raman measurements in the VIS region were performed at a single cell level using a Raman microscope (BioParticleExplorer, Rap.ID Particle Systems GmbH, Berlin, Germany). The setup was equipped with a 532 nm solid-state frequency-doubled Nd:YAG laser (LCM-S-111-NNP25; Laser-export Co.Ltd.) and a 100x air objective (MPLFLN-BD, Olympus, NA = 0.90) with a spot size below 1 μm focusing the laser light of approximately 9 mW onto single bacterial cells. A single-stage monochromator (HE 532, Horiba Jobin Yvon) equipped with a 920 lines/mm grating allowed the 180 backscattered Raman light to be diffracted and then detected by a thermoelectrically cooled charge-coupled device (CCD) camera (DV401-BV; Andor Technology). The spectral resolution was approximately 8 cm −1 and the integration time of 15 s per single bacterial cell was chosen.
For the bulk biomass measurements in the deep UV region, the Raman spectra were recorded using a Raman microscope (HR800, Horiba/Jobin-Yvon, Bensheim, Germany) with a 244 nm frequency-doubled Argon-ion laser (Innova 300, FReD, Coherent, Dieburg, Germany) and a focal length of 800 mm. The setup was equipped with a 2400 lines/mm grating and a ×20 magnification antireflection coated objective (LMU UVB) with a numerical aperture of 0.4. The width of the entrance slit was set to 400 μm and the exposure time was 15 s with 10 accumulations. The maximal laser power of about 18 mW was chosen, leading to about 0.5 mW on a sample. To minimize possible photodegradation, the sample was rotated at a speed of 30 rpm and moved in the x, y direction after each rotation to cover a large sample area. The Raman scattered light was detected by a nitrogen-cooled CCD camera. The spectral resolution was approximately 2 cm −1 .

| Data preprocessing
The data preprocessing of spectra acquired in the VIS and UV regions were carried out with the software Gnu R [30] using in-house written scripts. First the cosmic spikes were removed from the spectra [31]. Then the wavenumber axis of VIS-spectra was calibrated using acetylaminophenol as reference [32]. Subsequently, the sensitive nonlinear iterative peak (SNIP) clipping algorithm [33] with 30 and 40 iterations was applied to remove the fluorescence background in the spectra recorded in the VIS and UV regions, respectively. The spectra were then vector normalized and averaged across groups. Principal component analysis (PCA) was subsequently applied to all spectra to reduce the data dimensionality. Linear discriminant analysis (LDA) was then performed to classify 18 O-labeled and non-labeled cells. The classification performance was evaluated by the F I G U R E 1 Single cells mean Raman spectra of Escherichia coli cells recorded with an excitation wavelength of 532 nm. Raman spectra from 3200 to 750 cm −1 ; A, from 1740 to 1530 cm −1 ; B, and from 800 to 740 cm −1 ; C recorded after 24, 48 and 72 h of incubation in water (blue spectra) and H 2 18 O (red spectra). Assignments of represented bands are given in Table S1 leave-one-batch-out cross-validation with the optimal number of principal components. A 2D correlation analysis was applied to the time-dependent Raman spectra of E. coli cells using the R package corr2D [34]. The 2D synchronous and asynchronous maps were interpreted according to the Noda's rules [35].

| RESULTS AND DISCUSSION
To compare the wavelength dependent Raman spectra of E. coli cells, samples from the same batch culture were used. In Figure 1A (Table S1) on bacteria [19,[36][37][38][39][40][41]. The most prominent difference between the spectra of E. coli cells grown in H 2 O (nonlabeled) and those grown in 18 O-water ( 18 O-labeled) is observed in the amide I band. For more detailed examination, the amide I band has been enlarged and depicted in Figure 1B cells. The band superposition in the fingerprint region of Raman spectra makes it more challenging to detect small spectral changes related to the assimilation of 18 O. The application of Raman spectroscopy with electronically resonant excitation wavelengths in the deep UV region reduces the fluorescence background and provides an optimal signal-to-noise ratio due to the resonance Raman enhancement effect of biomolecules [40,[42][43][44]. For an excitation wavelength at 244 nm in particular vibrations due to aromatic amino acids and nucleic acids are resonantly enhanced. Figure 2 shows the Raman spectra of cells excited at 244 nm. The spectral assignment is in accordance with previously published UV-Raman spectroscopic studies of bacteria (Table S2) [40,43,45].
The band at 1640 cm −1 is attributed to the ν(C O) of thymine and cytosine, whereas the signal at 1617 cm −1 is assigned to the ν(C C) of the aromatic amino acids tyrosine, tryptophan and phenylalanine. Additional contributions arising from the ring breathing modes of aromatic amino acids were observed at 1365 cm −1 (tyrosine) and 1011 cm −1 (tryptophan and phenylalanine). The bands at 1176 and 762 cm −1 can also be attributed to tyrosine and tryptophan, respectively. The signal at 1575 cm −1 is assignable to the ν(C C) and ν(C N) modes of guanine and cytosine, while the signal at 1485 cm −1 is due to the ring vibrations of all nucleobases. The combination of adenine and tyrosine gives rise to the broad band at about 1335 cm −1 , whereas the combination of adenine and thymine contributes to the band at about 1241 cm −1 . The signal at 1530 cm −1 can be assigned to cytosine. The differences between labeled and nonlabeled spectra for the same excitation wavelength are difficult to detect by simple visual inspection (Figures 1  and 2). For better visualization, multivariate data analysis was performed on the 532 and 244 nm-spectra of each time point (24,48, and 72 h). PCA was first applied for F I G U R E 2 Mean Raman spectra of Escherichia coli cells excited at 244 nm after 24, 48 and 72 h of incubation in water (blue spectra) and H 2 18 O (red spectra). The spectra were shifted vertically for clarity. Corresponding peak assignments are provided in Table S2 dimensionality reduction. Then LDA was performed with leave-one-batch-out cross-validation to classify 18 Olabeled and non-labeled cells. At each time point both classes were very well separated with more than 92% accuracy. The classification results are summarized in Table S3. The loadings plots resulting from the PCA-LDA classification are depicted in Figure 3. Negative loadings are characteristics of non-labeled cells, whereas positive difference signals represent 18 Figure 1C.
The peak shifts observed and the differences in the loading vectors hint changes in the biochemical composition of bacterial cells incubated with the labeled substrate. Hence, these observations from the non-resonant Raman difference spectra excited at 532 nm indicate that E. coli cells used 18 O-labeled water (H 2 18 O) during metabolic processes. The increase of intensity of the peaks at 2983 and 2890 cm −1 at 72 h incubation could be due to nutrient depletion in the medium and the adaptation of the cell metabolism to the present conditions. At 244 nm, the loadings plot showed that the differentiation between 18 O-labeled and non-labeled cells was mainly based on DNA signatures due to a resonant excitation ( Figure 3B). The feature differences in the guanine and adenine ring breathing vibrations at 1479 and 1491 cm −1 and the tryptophan symmetric stretching vibration at 780 and 762 cm −1 could represent red-shifts that were not visible in the mean spectra. In general, the contributions of ring breathing modes of guanine, cytosine, adenine and thymine appear at about 1485 cm −1 ( Table S2). The peak at about 1491 cm −1 was assigned to the guanine, cytosine, and thymine contribution of non-F I G U R E 3 Loadings vectors resulting from the principal component analysis (PCA)-linear discriminant analysis (LDA) classification of 18 O-labeled and non-labeled Escherichia coli cells after 24, 48 and 72 h of incubation. A, 532 nm excitation and B, 244 nm excitation labeled cells. In contrast to adenine, both guanine and cytosine have a carbonyl group (C O) which is bound to the purine and pyrimidine ring, respectively. Thymine, however, comprises two carbonyl groups ( Figure S1). Hence, the signal at about 1479 cm −1 can be attributed to the substitution of 16 O by 18 O in guanine, cytosine and thymine bases. In UV-resonance Raman spectroscopy, only the Raman intensities of vibrational modes associated with the electronic transition (Franck-Condon active modes) are enhanced [18]. Similarly, to the 532 nm excitation, a possible red-shift is present in the nucleic acid band at 780 cm −1 indicating that the same vibrational modes were detected by both excitation wavelengths. Another red-shift is present in the loadings of the 48 h incubation, at 828 cm −1 , which is not present in the other studied time points and is assigned to tryptophan. The absence of this shift in the other time points could indicate adaptations of cellular metabolism to the consumption of available nutrients in the medium. Unlike DNA and aromatic amino acids, the amide I group of proteins absorbs light at even shorter wavelengths, that is, in the deeper UV-range at 197-206 nm [46]; hence, the signal of amide I is not resonantly enhanced for 244 nm excitation. Consequently, the incorporation of 18 O into the amide I group could in contrast to non-resonant excitation at 532 nm not easily be detected in the UV resonance Raman spectra of E. coli cells. However, a peak of weak intensity (1659 cm −1 ) assignable to the amide I signal was present in the loadings plot and characteristic for 18 O-labeled cells ( Figure 3B). Although no peak shift was detected by the visual inspection of the UV resonance Raman spectra of 18 O-labeled, it is noteworthy that chemometric methods such as PCA-LDA classification allowed to visualize the differences between 18 O-labeled cells and non-labeled cells. Hence, Raman 18 O labeling in deep-UV range can be combined with chemometrics to study the changes in the genotype of bacteria.
Since a peak shift was already detected in the 532 nm spectra at 24 h, we checked whether 18 O-assimilation was time-dependent. For this purpose, we recorded the Raman spectra of earlier incubation times (4, 8, 12 and 18 h) using 532 nm excitation wavelength, to check at which time point 18 O was assimilated in E. coli cells. The mean spectra of earlier incubations are depicted in Figure 4A. It is observed that a red-shift occurred in the amide I band ( Figure 4B) and the nucleic acid signal ( Figure 4C than the shifts induced by other stable isotopes such as 2 H (D), 13 C and 15 N. It could be that only a small fraction of amide I has been converted and that we see a mixed signal in the amide I region. A possible reason for this could be the different metabolic pathways used by bacteria to assimilate each substrate. It is likely that the chemical pathway of 18 O from H 2 18 O into biomolecules is more restricted than that of 13 C from glucose (commonly used in labeling of bacteria), resulting in a lower labeling efficiency. This would explain both the low Raman intensity and the small shift observed in the amide I band.
To investigate the incorporation of 18 O into proteins and nucleic acids in more details and to resolve overlapping bands, 2D-correlation analysis was applied using all the spectra recorded at 532 nm excitation wavelength. The fundamentals of 2D-correlation spectroscopy were described by Noda and others [35,47]. Figure 5 depicts synchronous and asynchronous 2D spectra in the range 1750 to 1520 cm −1 generated from the time-dependent (4-72 h) spectral variations of E. coli cells. The red and blue areas in the 2D maps represent positive and negative correlation, respectively, and the cross-peaks are read in the upper part of the diagonal. The synchronous map of 18 O-labeled cells ( Figure 5A) is dominated by a strong autopeak at about 1650 cm −1 , due to the C O stretching mode of proteins. This autopeak means that the amide I signal changed significantly over time. The signal at 1650 cm −1 has a positive correlation with the bands at about 1620 cm −1 (tyrosine) and 1560 cm −1 (DNA), indicating that the amide I signal changed in the same direction with tyrosine and DNA signals. The corresponding asynchronous map developed a "fourleave-clover cluster" pattern in the amide I band with a positive (at about 1666 cm −1 ) and a negative (at about 1648 cm −1 ) cluster (highlighted by the dashed circles) near the diagonal. The appearance of well-resolved lines in the clusters indicates that peaks were present at different positions, suggesting that the composition of amide I signal was heterogeneous, due to the substitution of 16 O by 18 O atoms. The sign of cross-peaks in the asynchronous map suggested that the spectral changes at 1648 cm −1 occurred earlier than those at 1666 cm −1 . This is to say the assimilation of 18 O into the C O group first causes the red-shift of the peak before affecting the signal intensity. The same information resulted from the 2D maps of the nucleic acid peak at about 781 cm −1 , which showed that the DNA/RNA peak was heterogeneous with a positive and a negative cross-peak ( Figure S2A); the corresponding asynchronous map ( Figure S2B); corroborated that the peak shift preceded the change in signal intensity.
The amide I signal of 18 O-labeled cells showed a positive cross-peak with the nucleic acid peak at 781 cm −1 ( Figure S3). The sign of the cross-peak in the asynchronous map revealed that the amide I signal changed prior to the nucleic acid peak. This means that E. coli cells incorporated 18 O into proteins first and then into nucleic acids. This indicates that E. coli cells started with protein synthesis during the adaptation phase (within 4 h of incubation) and later performed the DNA replication or RNA synthesis, allowing protein synthesis from 24 h onward, where E. coli cells should have reached the steady state. This growth strategy may explain the variations observed in the amide I signal (Figures 1 and 4A), and thus indicate that the observed decrease in the intensity of the amide I band of 18 O-labeled cells ( Figure 4A) was not only due to 18 O-uptake.
Unlike the synchronous map of 18 O-labeled cells, which developed cross-peaks only with the amide I signal, the synchronous map of water treatment shows a block of autopeaks and positive correlated cross-peaks in the region of 1750 to 1520 cm −1 ( Figure 5B). The corresponding asynchronous map (Figure 5B*) developed no cross-peak near the diagonal, indicating that the amide I signal of nonlabeled cells was homogenous. The patterns of both synchronous and asynchronous maps of 18 O-labeled cells ( Figure 5A, A*) were significantly different from those of non-labeled cells ( Figure 5B, B*), indicating that the variations caused by 18 O assimilation in bacterial cells can be distinguished from those resulting from normal bacterial growth with water. Therefore, 2D correlation analysis can be combined with Raman 18 O-labeling to study the metabolic dynamics of bacteria. The results of Raman 18 O-labeling at 532 and 244 nm demonstrated that the variations in spectral features were related to changes in the biochemical composition of bacterial cells. Thus, this labeling approach might allow the access to physiological and genetical information of bacterial cells.

| CONCLUSION
The present study demonstrates for the first time the use of 18 O stable isotope in the labeling strategies of bacteria. The results showed that 4 h of incubation are sufficient to label E. coli cells with 18 O. The Raman spectra of 18 Olabeled bacterial cells excited at 532 nm showed the accumulation of 18 O in the amide I group and DNA/RNA bases (cytosine/thymine/uracil), suggesting that protein and nucleic acid metabolisms were active in E. coli cells. Hence, the presence of a peak-shift is a reliable indicator for 18 O-uptake. The combination of Raman 18 O-labeling at 532 nm with 2D correlation analysis revealed the growth strategy of E. coli cells, which performed protein synthesis in the latent and steady phases, and DNA replication in the exponential phase. Hence Raman 18 O labeling combined with 2D correlation analysis can serve to monitor the metabolic activity of single bacterial cells.
However, for electronic resonant excitation at 244 nm 18 O uptake was not detected by visual inspection. The non-detection of 18 O uptake in the UV resonance Raman spectra leads to the following question: Are Raman modes sensitive to 18 O uptake not resonantly enhanced at 244 nm, or in other words was the selected UV excitation wavelength at 244 nm not suitable to detect 18 O incorporation in nucleic acids and aromatic amino acids? Future work also applying different UV excitation wavelengths is needed to answer this question. The results of this study showed that Raman 18 O-labeling approach at 532 nm is sensitive to the metabolic activity of both proteins and nucleic acids. Hence, this labeling approach could detect fluctuations in physiological information of bacteria and thus reveal the presence of metabolically active cells. The presented approach might offer opportunities for future research and the development of application in different fields. For instance, Raman 18 O-labeling will be of great advantage to investigate phosphate solubilizing bacteria, since phosphate lacks stable isotopes. Furthermore, Raman 18 O-labeling could be applied to identify antimicrobial resistant bacteria by monitoring the metabolic dynamics of proteins and nucleic acids in response to antimicrobial agents. We believe that Raman 18 O-labeling might represent a promising tool to probe the general metabolic activity of bacterial cells in a very effective and simple way.