Tracking heavy water (D₂O) incorporation for identifying and sorting active microbial cells

Significance

Measuring activity patterns of microbes in their natural environment is essential for understanding ecosystems and the multifaceted interactions of microorganisms with eukaryotes. In this study, we developed a technique that allows fast and nondestructive activity measurements of microbial communities on a single-cell level. Microbial communities were amended with heavy water (D₂O), a treatment that does not change the available substrate pool. After incubation, physiologically active cells are rapidly identified with Raman microspectroscopy by measuring cellular D incorporation. Using this approach, we characterized the activity patterns of two dominant microbes in mouse cecum samples amended with different carbohydrates and discovered previously unidentified bacteria stimulated by mucin and/or glucosamine by combining Raman microspectroscopy and optical tweezer-based sorting.

Abstract

Microbial communities are essential to the function of virtually all ecosystems and eukaryotes, including humans. However, it is still a major challenge to identify microbial cells active under natural conditions in complex systems. In this study, we developed a new method to identify and sort active microbes on the single-cell level in complex samples using stable isotope probing with heavy water (D₂O) combined with Raman microspectroscopy. Incorporation of D₂O-derived D into the biomass of autotrophic and heterotrophic bacteria and archaea could be unambiguously detected via C-D signature peaks in single-cell Raman spectra, and the obtained labeling pattern was confirmed by nanoscale-resolution secondary ion MS. In fast-growing Escherichia coli cells, label detection was already possible after 20 min. For functional analyses of microbial communities, the detection of D incorporation from D₂O in individual microbial cells via Raman microspectroscopy can be directly combined with FISH for the identification of active microbes. Applying this approach to mouse cecal microbiota revealed that the host-compound foragers Akkermansia muciniphila and Bacteroides acidifaciens exhibited distinctive response patterns to amendments of mucin and sugars. By Raman-based cell sorting of active (deuterated) cells with optical tweezers and subsequent multiple displacement amplification and DNA sequencing, novel cecal microbes stimulated by mucin and/or glucosamine were identified, demonstrating the potential of the nondestructive D₂O-Raman approach for targeted sorting of microbial cells with defined functional properties for single-cell genomics.

Microorganisms play a vital role in many environments. They mediate global biogeochemical cycles, catalyze biotechnological processes, and contribute to health and disease in the human body. The in situ study of microbial activity in natural and engineered ecosystems is therefore of great interest. For this purpose, several elegant methods have been established that use either transcriptional or translational activity of community members (i.e., metatranscriptomics, metaproteomics) (1–3) or the incorporation of isotopically labeled substrates into biomolecules (4–10) to infer the ecophysiology of microbes in such systems. However, these bulk techniques do not offer sufficient spatial resolution to study microbial activities at the micrometer scale. Therefore, important information can be overlooked because microbial communities are frequently spatially structured (e.g., biofilms) (11) and contain populations with life cycles (12, 13). Furthermore, even apparently identical cells in clonal populations can have strongly divergent activities (14).

Consequently, microbial ecophysiology is ideally studied also at the level of the single cell, but only a restricted number of approaches exist for determining physiological properties of individual cells in a microbial community. For example, one can observe the assimilation of radioactive or stable isotope-labeled substrates into cellular biomass using microautoradiography for the former or nanoscale resolution secondary ion mass spectrometry (NanoSIMS) or Raman microspectroscopy for the latter. These approaches can be directly combined with FISH to identify labeled organisms (15–17). However, many substrates are very expensive as isotopically labeled derivatives or are not even commercially available. Also, addition of labeled substrates easily changes the composition of the natural substrate pool, and thus potentially biases microbial community structures and functions. Furthermore, in almost all ecosystems, preferred substrates of most microbial community members are not known, highlighting the need for single-cell tools suitable to record the general physiological activity of microbes without prior knowledge. Microbiologists have tested fluorescent probes and stains to quantify cellular rRNA or its precursors (18, 19) and membrane potential (20) as activity measures, but cellular rRNA levels often do not correlate with activity (21) and membrane potential stains perform differently with different cell types (22). Recently, heterotrophic ¹⁴CO₂ fixation was used to infer the general activity of heterotrophic microbes semiquantitatively (23), but the extent of heterotrophic CO₂ fixation can vary widely between different species and also depends on the heterotrophic substrate used for growth (24–27). In addition, fluorescence-based methods have been developed for measuring single-cell biosynthesis of macromolecules (28–30), but the generalizability and usability of these new methods, which rely on the addition of substrates with modified functional groups, are still either largely untested in complex ecosystems or have known biases across different species.

In this study, we aimed at developing a universally applicable technique that (i) allows one to measure the activity of individual microbes within complex samples without prior knowledge of their physiology and without altering the natural substrate pool, (ii) can be combined with FISH to identify cells in situ, but (iii) can also be performed on living cells to facilitate postmeasurement analyses like single-cell genomics or cultivation. For this purpose, we explored the suitability of heavy water [deuterium oxide (D₂O)] as an additive to microbial communities. In all known lipid biosynthesis pathways, hydrogen from water is incorporated during the reductive steps of fatty acids synthesis (31–33). Interestingly, deuterium (D) can be used in lieu of hydrogen (H) during lipid biosynthesis, leading to deuterated lipids of active microbial community members after incubation with heavy water [in natural waters, the concentration of deuterium relative to hydrogen is at very low levels of ∼0.0156% (34)]. This effect has been exploited on the bulk level to monitor lipid biosynthesis of microbes in the environment (35, 36). In addition to lipid labeling, proteins of active community members will become deuterated, as demonstrated by metaproteomics (37). Here, we show that deuterium incorporation into individual active microbial cells can easily be detected within seconds by nondestructive Raman microspectroscopy and that the amount of deuterium labeling can reliably be quantified. In proof-of-principle experiments using mouse cecal microbial communities, we combined this simple and fast technique with FISH to demonstrate the activity of Bacteroides acidifaciens and Akkermansia muciniphila to addition of four different substrates. Furthermore, we identified novel microbes stimulated by glucosamine and mucin by combining Raman microspectroscopy with optical tweezer-based sorting, multiple displacement amplification (MDA), and 16S rRNA gene sequencing.

Results and Discussion

Raman Microspectroscopy Reveals Incorporation of Deuterium from D₂O in Active Microbial Cells.

Raman microspectroscopy produces a chemical “fingerprint” of the abundant molecular bonds in individual microbial cells. Because bulk analyses have demonstrated significant incorporation of D into the biomass of cells growing in the presence of heavy water (35, 37), we hypothesized that this incorporation might be detectable as a general metabolic activity marker in Raman spectra of single microbial cells. Indeed, Raman spectra of Escherichia coli cultivated in growth water with no, partial, or complete substitution with heavy water exhibited several distinctive shifts (Fig. 1 A and B; an overview of all experiments in this study is provided in SI Appendix, Table S1). Most dramatically, a broad peak appeared in a region between 2,040 and 2,300 cm⁻¹ that typically has no detectable peaks in Raman spectra of non–deuterium-labeled E. coli cells or cells from other microbes. Furthermore, the intensity of the prominent C-H peak (between 2,800 and 3,100 cm⁻¹) was reduced with increasing D incorporation (Fig. 1 A and B). C-D_X bonds of various types, including C-D₂ and symmetrical and asymmetrical C-D₃ bonds originating from various biomolecules, reportedly produce Raman bands between 2,013 and 2,300 cm⁻¹ (38–44). Furthermore, modeling of the Raman peaks induced by the C-D_X bond using deuterated palmitic acid as the model compound also predicted peaks in this region (SI Appendix, Fig. S1). This spectral pattern reflects a substitution of C-H_X by C-D_X in newly synthesized macromolecules in the presence of heavy water, and it is safe to assume that an important fraction of the D incorporation is found in lipids, which contribute roughly 25–30% to total cellular C-H (45, 46). Lipids from autotrophic and heterotrophic organisms show a predominance of water-derived protons (or D⁺ in the presence of heavy water) (33, 35, 47) as protons from water are transferred to NAD(P) during its reduction, and these protons are then incorporated into lipids via the known fatty acid biosynthesis pathways (31–33). D will, of course, also be incorporated to some extent via other processes, such as amino acid and carbohydrate biosynthesis (37, 48). Importantly, however, control experiments with paraformaldehyde-fixed cells demonstrated that abiotic H-D exchange reactions, such as occur in amide bonds of proteins (49) or other O- and N-bound H (32), do not contribute measurably to the C-D peak region (SI Appendix, Fig. S2), and therefore that C-D peaks in the Raman spectrum after growth of microbes in deuterated water are exclusively caused by biological processes.

Fig. 1.

E. coli incorporation of D from heavy water during growth as detected by Raman microspectroscopy and NanoSIMS. (A) Raman spectra of single E. coli cells grown to stationary phase in media with heavy water (0%, 2.5%, 5%, 10%, 15%, 20%, 30%, 50%, and 100% D₂O of growth water). Ten to 21 single-cell spectra were used to produce mean spectra. It should be noted that identical peak shifts were observed in E. coli cells that were grown in the presence of deuterated glucose instead of heavy water (SI Appendix, Fig. S11). AU, arbitrary unit. (B) Difference between mean spectra of E. coli cells from D₂O-containing media and cells grown without D₂O. Colors are the same as in A. (C) Quantification of D incorporation in individual E. coli cells from the same experiment as detected by NanoSIMS and shown as the isotope fraction D/(H + D) given as at%. All isotope fraction images are on the same scale (0–100 at%). ³¹P⁻ signal intensity distribution is displayed to indicate the location of cellular biomass. (D) Comparison of D content in single cells measured by NanoSIMS with respect to the D₂O percentage of growth water. Box plots show the quartiles for each population of cells. The detection limit, defined as the mean + 3 SD of unlabeled cells, is shown in gray (0.17 at%). A linear regression between D₂O concentration and cellular D is shown in red (R² = 0.84).

Using a supervised statistical method (principal components analysis in combination with linear discriminant analysis), a classification model was built for the presence or absence of D₂O in the growth water of E. coli. Using a 10-fold cross-validation scheme, the accuracy of the model was estimated to be 98%. This result indicates that Raman spectra are characteristic of growth in D₂O and an automatic discrimination based on a classification model is possible at all tested concentrations of D₂O in the growth medium (Fig. 1 and SI Appendix, Table S2). To quantify the deuterium content of the E. coli cells from growth experiments with different amounts of heavy water, the cells were subjected to NanoSIMS (Fig. 1C). As expected, the cellular D content was consistently lower than the D₂O concentration of the growth water (Fig. 1D), which is because H from organic substrates in the medium is also incorporated into the biomass during heterotrophic growth of E. coli. Although there was a clearly linear correlation (R² = 0.84) between the concentration of deuterated water in the growth medium and the deuterium content of the biomass, there was also some heterogeneity in the amount of incorporated deuterium between individual cells, especially at higher concentrations of deuterated water (Fig. 1D). This latter finding reflects physiological heterogeneity, which might be particularly pronounced at higher D₂O concentrations due to their potential inhibiting effect (below).

In the next step, several bacterial and archaeal reference strains with differing physiologies were grown as pure cultures in the presence of different amounts of D₂O. The strains included heterotrophic bacilli (Bacillus thuringiensis and Bacillus subtilis), an autotrophic nitrite-oxidizing bacterium (Nitrospira moscoviensis), an autotrophic ammonia-oxidizing archaeon (Nitrosophaera gargensis), and autotrophic methanogenic archaea (Methanobrevibacter smithii and Methanocorpusculum labreanum). These species were selected to represent a wide range of phylogenetic and physiological diversity. Again, single-cell Raman spectra of all tested strains were marked by a peak appearing in the C-D region and a reduction in the C-H peak (Fig. 2A), indicating that this labeling strategy can be applied across diverse organisms. As expected heterotrophic organisms showed a weaker D incorporation than autotrophically growing microorganisms, which must reduce CO₂ to produce biomass, and therefore incorporate more hydrogen atoms from water (ANOVA, P < 0.0.001; Fig. 2B and SI Appendix, Fig. S6). The reduced D/(D + H) of measured autotrophs compared with growth water is at least partly caused by the large D/H fractionation by autotrophs, as has been observed, for example, in methanogenic sediments (50). Furthermore, for the slow-growing reference organisms used in our experiments, the applied incubation time might have been insufficient to achieve complete labeling.

Fig. 2.

C-D signature region of Raman spectra is conserved in diverse microorganisms. (A) Representative Raman spectra of microorganisms cultivated in media amended with various percentages of heavy water. Characteristic C-D and C-H regions are shaded in gray. N. moscoviensis and N. gargensis did not grow with 100% D₂O. Both methanogens were only grown at 30% D₂O and not tested at other levels. (B) NanoSIMS quantification of cellular D incorporation (at%) of different strains when grown in 30% D₂O (and E. coli with 0% D₂O). Each point is a measurement of a single cell, and box plots indicate quartiles. Note that only a few cells of N. moscoviensis became active after incubation (low-active/nonactive cells are shown in gray), a feature we have also observed by other techniques in medium without heavy water. M. smithii cultures produced extracellular substances enriched in D that were excluded from the analysis. Autotrophic organisms (M. smithii, N. gargensis, and N. moscoviensis) were significantly more enriched in D than heterotrophic organisms (E. coli, B. subtilis, and B. thuringiensis) when cultivated in the presence of the same D₂O concentration (30% D₂O; ANOVA, P < 0.001).

Although it is widely used in low doses for physiological studies (51), heavy water is known to inhibit cellular processes in mammals at higher concentrations (52). For bacteria, different responses of deuterated water on growth ranging from no effect to inhibition and even stimulation have been reported (53, 54). In our experiments, heavy water up to a concentration of 50% D₂O did not inhibit the growth of our test strains. However, the presence of 100% D₂O in the growth medium significantly reduced the specific growth rate of exponentially growing E. coli and B. thuringiensis (P = 0.002 and P = 0.02, respectively) but did not significantly affect the growth of B. subtilis (SI Appendix, Fig. S3 A–C). Additionally, the ammonia oxidation activity of N. gargensis was only slightly affected in the initial phase of incubation by up to 50% D₂O (SI Appendix, Fig. S3D), but 100% D₂O fully inhibited its growth. Taken together, these data suggest that heavy water up to 50% of total growth water can be applied in labeling studies of natural microbial communities without causing major changes in the activity of individual community members.

Raman spectra of microbial cells vary between different taxa (55, 56), and even between individual cells of the same microbial strain that have experienced different growth conditions (57), because they reflect the cellular chemical composition. Consequently, using information from the entire Raman spectrum to quantify cellular C-D abundance of a microbial cell in a complex sample will not be trivial if the study is not focused on a specific strain and sufficiently defined growth conditions. Thus, we were interested in determining if the C-D and C-H signature regions alone could be used to quantify the amount of cellular deuterium incorporation. The metric “%CD” was defined as the integrated spectral intensity of the C-D signature region as a percentage of the integrated spectral intensity of both the C-D and C-H signature regions defined above. Repeated measurement of single cells revealed that the %CD quantification was reproducible and varied much less than the variation observed within a clonal population of cells (relative SD = 0.09; SI Appendix, Fig. S4). Remarkably, in fast-growing E. coli cells in the presence of 40% D₂O, some labeled cells could already be detected by Raman spectroscopy after 20 min of incubation, demonstrating that this approach is very sensitive and does not necessarily require cell division (cell doubling time was 47 min), because synthesis of macromolecules already leads to deuterium incorporation (SI Appendix, Fig. S5). The %CD was linearly associated with the percentage of D₂O in the growth water for each given strain, and for all strains, the inferred detection limit (mean + 3 SD of unlabeled cells) was less than 8%CD. (SI Appendix, Fig. S6).

Measurement of deuterium incorporation on a single-cell level by Raman microspectroscopy can also be directly combined with FISH. Probed cells were identified using the fluorescence microscope attached to the Raman microspectrometer, and high signal-to-noise Raman spectra of such cells could be consistently recorded (SI Appendix, Fig. S7). The standard storage conditions used for FISH did not significantly affect %CD (SI Appendix, Fig. S8), but hybridization conditions used for FISH did have an effect, leading to an average reduction in %CD by 42% (SI Appendix, Fig. S9). Loss of cellular D will increase the deuterium detection limit of the Raman method, which is most relevant for experiments using very low-percentage additions of heavy water or for targeting microbes with a low activity. It should be noted that some loss of labeled compounds is also expected or has already been reported in all other single-cell isotope labeling methods, for example, microautoradiography or NanoSIMS for isotope detection if either is combined with FISH or catalyzed reporter deposition (CARD)-FISH (58, 59).

Determining Compound Utilization Patterns of Members of the Gut Microbiota.

The gut microbiota is important to human and animal health and nutrition (60, 61). Competition for nutrients is thought to shape the overall composition of the gut microbiota as well as the colonization success of enteropathogens (62, 63). For example, the availability of mucosally derived sugars has been shown to be a key factor in the colonization of E. coli, Salmonella typhimurium, and Clostridium difficile (64, 65). Although important, the compound utilization activities of individual microbes in the complex gut community are not well understood (66). In a proof-of-principle experiment, we applied the heavy water labeling approach to study how microbes in a complex microbial community respond to the addition of different nutrients with single-cell resolution. Mouse cecum contents were incubated anaerobically in the presence of 50% D₂O and were amended with sugars that are components of dietary- and/or host-derived polysaccharides (glucose, glucosamine, and mannose) or a secreted mucosal glycoprotein with mixed glycans (porcine gastric mucin) as a model complex substrate. Subsequently, single microbial cells were probed by Raman microspectroscopy for deuterium incorporation. This probe was done either by a nontargeted approach of random selection of cells or by targeted FISH-based profiling of two important degraders of host-derived compounds, A. muciniphila and B. acidifaciens (67–69) (Fig. 3A).

Fig. 3.

Application of D₂O to monitor the activity of mouse cecum microbiota on a single-cell level after in vitro incubation with various unlabeled substrates. Experiments were performed on cecal biomass amended with glucose, glucosamine, mannose, mucin, or nothing and were incubated overnight. (A) Representative FISH images of all Bacteria (blue), B. acidifaciens (red), A. muciniphila (green), and composite. (Scale bar: 5 μm.) (B) Percentage of cells from the respective target population for which a C-D peak could be detected by Raman microspectroscopy under each incubation condition. For each target and incubation condition, 68–92 cells were measured. (C) Contributions of B. acidifaciens and A. muciniphila to all labeled (Left) and highly labeled (Right, >10%CD) cells in the gut community in the different experiments are displayed. These contributions were calculated as follows: [Relative abundance of target population as measured by quantitative FISH using a specific probe] * [Proportion of target population labeled]/[Relative abundance of all labeled cells]. (D) Intensity of deuterium incorporation in single cells of randomly selected (DAPI-positive) and specific FISH probe-defined populations, measured by %CD. Single-cell spectra are denoted by black points, and population quartiles are shown as box plots. The red horizontal line at 2.78%CD indicates the threshold for considering a cell labeled. It was determined by calculating the mean + 3 SD of %CD in randomly selected cells from the cecum sample that were incubated without addition of heavy water. All comparisons are statistically significantly (P < 0.05), except for those comparisons denoted by a black bar connecting them and labeled by N.S. (not significant).

Overall, a large percentage of the cecal microbial community was stimulated by addition of glucose (81%), mannose (75%), and mucin (71%), whereas the response to glucosamine was lower (41%) (Fig. 3B). Additionally, a fraction of the community (∼9%) was active in the incubation without substrate addition, and thus thrived due to endogenous metabolism or on decay products of other microbes. The B. acidifaciens population was stimulated to the greatest extent by mannose, followed by mucin, glucose, glucosamine, and, lastly, no amendment (P < 0.05; Fig. 3B). In contrast, the A. municiphila population was stimulated primarily by mucin, with much lower activity in mannose and glucose amendments and no increase in activity due to glucosamine compared with no amendment controls (P < 0.05). B. acidifaciens and other Bacteroides spp. are capable of fermenting a wide range of simple and complex carbohydrates and using mixed substrates (70–74). Previous studies have shown that in pure culture, members of the Bacteroides prioritize mannose utilization over other sugars (74, 75), and our results suggest that mannose may be a preferred substrate in vivo as well. Both B. acidifaciens and A. muciniphila were stimulated by mucin. These two species are known to degrade mucin in pure cultures (76, 77) and are important host-compound degraders in vivo (68). A. muciniphila has a phosphotransferase transporter system that is likely capable of taking up a range of simple sugars, including glucose, glucosamine, and mannose (78). Interestingly, however, this organism was stimulated in situ almost exclusively by mucin (Fig. 3 C and D). This result reinforces the concept that in vivo, A. muciniphila is primarily active as a mucin degrader (67) in contrast to B. acidifaciens, which is more versatile in its nutrient utilization strategy. The flexible foraging activity of B. acidifaciens raises the possibility that the presence of alternative nutrients could modify its activity in degrading host-derived compounds in the gut. Future research is needed to determine how this versatile foraging behavior might affect the mucus layer and if altered mucus degradation due to changed foraging patterns has an impact on immunological tolerance to the gut microbiota. By combining quantitative FISH with quantification of active cells using the Raman-based detection of deuterated cellular compounds, it could be demonstrated that B. acidifaciens represented an important but not dominant fraction of all highly labeled cells for all amendments (7–19% of all cells that had >10%CD), whereas A. muciniphila accounted for less than 1% of highly labeled cells under all conditions except for the mucin amendment, in which it represented 15% of highly labeled cells (Fig. 3C). These quantitative inferences demonstrated that although B. acidifaciens and A. muciniphila contributed to community activity, other unidentified microbes dominated the active microbial community after amendment of the various substrates.

Identifying Active Cells Using Raman-Based Sorting and DNA Sequencing.

To demonstrate the power of heavy water labeling with Raman as a discovery tool, we further investigated cells that were highly labeled by deuterium after glucosamine (16% of all cells in the community) and mucin (9% of all cells in the community) amendments. Nonfixed samples from the incubations were directly added to a microcapillary and single cells were trapped with an optical tweezer (79) for determining their deuterium content via Raman microspectroscopy. Because neither H₂O nor D₂O has Raman peaks in the C-D region, background from water did not interfere in this approach (80). From each incubation, 40 cells with a high C-D peak (corresponding to the top decile of labeled cells) were identified and sorted by moving them individually to the sterile end of the capillary (representative spectra are shown in Fig. 4A). After MDA, the sorted cells were identified by 16S rRNA gene amplification, cloning, and sequencing (Fig. 4 and SI Appendix, Fig. S10). The obtained 16S rRNA gene sequences reflect the diversity of the substrate-stimulated cells, but it should be kept in in mind that due to well-described biases of the analysis pipeline (cell lysis, MDA, PCR, and cloning) (81), the number of clones recovered may not reflect the relative abundance of active cells in the sample. Several taxa (Akkermansia, Allobaculum, Bacteroides, Barnesiella, Clostridiales, Desulfovibrioaceae, and Parabacteroides) were found among the sorted cells from both incubations, demonstrating that these microbes responded to both added substrates. Many of these taxa are saccharolytic, although glucosamine and mucin utilization has not been demonstrated before for most of them (82–85). Consistent with the experiments depicted in Fig. 3, active A. muciniphila cells also belonged to this group. Although quantitative inferences from obtained clone numbers have to be interpreted with great caution, it is interesting to note that 15 of 39 clones were affiliated with A. muciniphila in the mucin incubation experiment, whereas only two of 43 clones belonged to this important mucin degrader after glucosamine incubation (Fig. 4B). The detection of multiple Bacteroides populations stimulated by mucin suggests that multiple Bacteroides species in the same community respond to the same compound, which could indicate either cooperative degradation of or competition for host-derived substrates among members of this genus. Future work is needed to unravel the nature of possible interactions among different Bacteroides species in the gut microbiota. The detection of active Lactobacillus cells after glucosamine detection confirms the notion that the use of this substrate is an important strategy of these microbes for survival in the gut (86). Interestingly, members of the less well-characterized TM7 candidate phylum, which were recently postulated to be saccharolytic (87), were also active after glucosamine addition. Putatively sulfate- or sulfite-reducing Desulfovibrionaceae responded to glucosamine and mucin amendments. This response likely reflects indirect stimulation of this group due to cooperative interactions with saccharolytic bacteria that produce metabolic substrates for these sulfur-compound reducers (88).

Fig. 4.

Targeted sorting and phylogenetic analysis of deuterium-labeled cells from two gut microbiota incubations. (A) Raman spectra of 40 highly labeled cells from the glucosamine and mucin incubations, respectively, that were sorted using optical tweezers. Arrows indicate the strong C-D peaks in the spectra of the sorted cells. A few representative spectra from cells that were not sorted from the mucin incubation are also shown in gray to allow for comparison. (B) Phylogenetic analysis of partial 16S rRNA gene sequences recovered from sorted cells [the number of sequences is indicated and colored blue for mucin (Muc) incubation and red for glucosamine (GlcN) incubation]. The tree was produced using near-full-length sequences from organisms closely related to clone sequences (indicated in black) using RAxML with 500 bootstrap resamplings (black circles indicate >90% support, white circles indicate >75% support). Clone sequences were then added to the tree using the quick-add parsimony method in ARB. The taxonomic classification of sequences is indicated. Coverage of A. muciniphila and B. acidifaciens FISH probes (estimated from full-length sequences) is indicated. The scale bar indicates the number of substitutions per site. A fully expanded tree is presented in SI Appendix, Fig. S10.

Conclusions and Outlook

This study shows that incubation of samples containing complex microbial communities with heavy water subsequently allows microbiologists to identify individual metabolically active bacterial and archaeal cells by Raman microspectroscopy without any further sample pretreatment. The C-D signature peak for deuterium incorporation into single microbial cells is easily detectable and is already visible in rapidly growing cells after 20 min of incubation. Compared with many other isotope-labeled substrates, heavy water is inexpensive and its addition to samples enables functional studies of microbial communities with minimal perturbation of the natural environmental conditions. Interestingly, the tested microbial strains showed no significant change in growth rate in the presence of up to 50% D₂O, and because much lower D₂O concentrations were sufficient for Raman-based activity measurements, toxicity of heavy water is not a limiting factor for the application of this technique. As demonstrated with incubation experiments using mouse cecum microbial communities (Fig. 3), heavy water labeling can be combined with the addition of unlabeled substrates to investigate which members of a microbial community are stimulated by which compound. This approach is particularly important for those researchers interested in the metabolic conversion of substrates that are not commercially available as isotope-labeled derivatives or for which isotope labeling is extremely expensive.

Heavy water could also be used to label microbial cells living within or attached to plants, animals, and possibly even humans. Although consumption of high amounts of heavy water is toxic or even lethal to plants, animals, and humans (89), the D₂O concentrations required for activity labeling of microbial communities inferred in this study are below the toxicity threshold for body water deuteration of several important model animals. For example, mice and rats have been observed to be asymptomatic at D levels of 20–25% (90). Consequently, it should become possible to perform in vivo activity labeling experiments with microbial gut communities in mice by providing heavy water as drinking water for a short time period. However, the application range of the developed approach might even be much broader because it seems likely that fast-growing neoplastic cells also incorporate significantly more D from heavy water than other cells, a feature that would enable their straightforward detection via Raman microscopy or sensors (91).

Furthermore, this study demonstrated that heavy water labeling of microbial cells and Raman microspectroscopy can be combined with optical tweezing for targeted sorting of individual microbial cells with certain functional traits for subsequent characterization via MDA and DNA sequencing. Because Raman microspectroscopy and optical tweezing are nondestructive techniques that work in water, no pretreatment of the cells (other than the isotope labeling) is required; thus, this approach will not only allow single-cell genomics experiments targeted to members of defined microbial guilds but might also prove to be useful for targeted cultivation of members of microbial communities with interesting ecophysiological properties.

Materials and Methods

Microorganisms and Growth Conditions.

For all strains tested, media were prepared with different percentages (vol/vol) of D₂O [99.9% atom % (at%) D; Sigma–Aldrich) to test D incorporation. Pure cultures of the Gram-negative bacterium E. coli DH10B and the Gram-positive strains B. subtilis W23 and B. thuringiensis American Type Culture Collection 10792 were grown aerobically in LB with shaking at 37 °C. B. thuringiensis was selected due to a report indicating that this species might be stimulated by D₂O (54). Fifty microliters of stationary-phase cultures was inoculated into 5 mL of LB, and growth was monitored spectrophotometrically (at 578 nm; Camspec M107 Spectrophotometer) every 30 min until stationary phase (OD at 578 nm of 0.7–0.8) to calculate the growth rate of cultures. Additionally, to determine short-term incorporation of deuterium, samples were collected at 20, 40, 60, 90, 120, 180, and 240 min. For serial growth measurements of E. coli, an overnight culture of cells was inoculated into fresh medium and grown overnight. For growth of E. coli on deuterated glucose (SI Appendix, Fig. S11), cells were grown in M9 medium (Sigma-Aldrich) with 10 mM of glucose as the sole carbon source. Cells were incubated with different ratios of unlabeled/deuterated glucose aerobically with shaking at 37 °C for 16 h in biological triplicate. Cells were then harvested, washed three times with deionized water, and measured with Raman microspectroscopy. N. gargensis was grown aerobically at 46 °C in ammonia-oxidizing archaea (AOA) medium (all chemicals from Sigma-Aldrich) (92), amended with 2 mM NH₄Cl, and regularly fed with 1 mM NH₄Cl when ammonium was depleted. Ammonium was detected by Nessler’s reagent (Sigma–Aldrich), and nitrite was quantified with the sulfanilamide N-1-napthylethylenediamine dihydrochloride reagent spectrophotometric method (93). N. moscoviensis was grown in mineral medium (94) amended with 1 mM NaNO₂ and shaking (90 rpm, KS 4000i, IKA) at 29 °C in the dark. Nitrite was regularly checked with colorimetric test strips (Merck) and was replenished every 2 d by injection, and cultures were sampled after 6 d. The methanogen strains Methanobrevibacter gottschalkii (DSM 11977), M. smithii (DSM 861), and M. labreanum strain Z (DSM 4855) were grown anaerobically in Hungate tubes at 39 °C for 7 d in basal medium with vitamins according to Leahy et al. (95), including ampicillin, erythromycin, and vancomycin at 50 μg/mL. Methane production was monitored by gas chromatography (95). Cells from all bacterial and archaeal reference strains were harvested by centrifugation and fixed in 3% formaldehyde for 3 h at 4 °C and stored either in 1:1 PBS/ethanol solution or in 1:1 PBS/glycerol at −20 °C until measurement.

Mouse Cecum Incubations.

All animal experiments were discussed and approved by the University of Veterinary Medicine, Vienna, Institutional Ethics Committee and conducted in accordance with protocols approved by Austrian laws (BMWF-66.006/0002-II/10b/2010). Three adult C57BL/6 mice were killed, and their ceca were harvested anaerobically in an anaerobic tent. Contents from each cecum were suspended in 7.8 mL of 50% D₂O-containing PBS and homogenized by vortexing. The homogenate was then distributed into glass vials and amended with glucose (2.5 mg/mL), glucosamine (2.5 mg/mL), mannose (2.5 mg/mL), porcine gastric mucin (1% mass/vol), or nothing (no-amendment control) (all amendment chemicals were from Sigma–Aldrich). Vials were crimp-sealed with rubber stoppers and incubated anaerobically overnight (18 h) at 37 °C without shaking. At the end of the incubation, the biomass was washed with PBS to remove D₂O and subsamples of the biomass were fixed with either 100% ethanol alone or 3% formaldehyde for 2.5 h at 4 °C and stored in 50% PBS/50% ethanol solution at −20 °C or 100% ethanol.

FISH.

DNA oligonucleotide probes were used to target most bacteria (EUB338I-III mix) (96, 97), as well as the genus Akkermansia [Akk1437 (S-G-Akk-1437-a-A-20), 5′-CCT TGC GGT TGG CTT CAG AT-3′] (98) and B. acidifaciens [Baci731AB (two probes used together: S-*-Baci-0087-a-A-24, 5′-GCG CCG GTC GCC ATC AAA AGT TTG-3′ and S-*-Baci-0087-a-A-22, 5′-G CCG GTC GCC ATC GGA AGT TTG-3′), 35% formamide]. Although Akk1437 targets the entire genus Akkermansia, previous sequencing-based studies have shown that A. muciniphila is the only species of the genus in these mice (69). Probes double-labeled with either the fluorescent cyanide dye Cy3 or Cy5 (only Cy5 was used for Raman experiments) were hybridized to fixed cells of the reference strains or to fixed biomass samples at 46 °C for 2 h using a standard liquid-FISH protocol (15, 99).

Confocal Raman Microspectroscopy and Spectral Processing.

Samples were spotted on aluminum-coated slides (Al136; EMF Corporation) and washed by dipping into ice-cold Milli-Q (MQ) water (Millipore) to remove traces of buffer components. Samples were stained with DAPI (1 μg/mL; Sigma–Aldrich) for 5 min and washed with MQ water. Single microbial cell spectra were acquired using either a LabRAM HR800 or an HR Evolution confocal Raman microscope (both from Horiba Jobin-Yvon) equipped with a 532-nm neodymium-yttrium aluminium garnet laser and either 300 or 600 grooves/mm diffraction grating. Spectra were acquired in the range of 400–3,200 cm⁻¹. For hybridized samples, fluorescence images of each field of view were acquired using Horiba software. Raman spectra were preprocessed as described previously (100). First, a background correction with the sensitive nonlinear iterative peak-clipping algorithm was carried out using optimized parameters (smoothing = true, iteration = 70, window = “15”) (101). Thereafter, each spectrum was normalized to the sum of its absolute spectral intensity. For quantification of the degree of D substitution in C-H bonds (%CD), the bands assigned to C-D (2,040–2,300) and C-H (2,800–3,100) were calculated using integration of the specified region. Preprocessed Raman spectra were also used to construct a classification model for the presence or absence of D₂O in the growth media. Here, a principal component analysis was carried out to reduce the dimension of the data (102). Seven principal components were used to construct the classification model. As a classification model, linear discriminant analysis (103) was applied, and the predictive performance was evaluated using a 10-fold cross-validation (104). The result of this procedure can be presented in a confusion matrix, where true class labels are compared with the predicted ones. Here, these class labels are either “presence of D₂O” or “absence of D₂O.”

Raman-activated cell sorting.

Capillary tubes (Rectangle Glass Capillary, 0.10 × 1.00 mm; CM Scientific) were baked at 500 °C overnight before use and half-filled with sterile PBS buffer with 0.5% Tween 20. Samples from the mouse cecum incubation experiments were diluted 1:50 and introduced into capillary tubes. The end of the capillary tube at which the sample was added was sealed with petroleum jelly to prevent evaporation. The capillary tube was then placed under a 60× water immersion objective (UPlanSApo; Olympus) mounted on the Raman spectrometer. Single cells were optically trapped with a 1,064-nm laser (mpc6000; Laser Quantum), and spectra were measured over a range of 600–3,000 cm⁻¹ to cover the most relevant Raman peaks in spectra of microbial cells, including the C-D peaks. When a deuterium-labeled cell was identified, it was moved to the sterile end of the capillary. After repeating this procedure for multiple labeled cells, the end of the capillary was broken off and placed in a sterile 0.2-μL tube. The lid of the tube was closed, and the sorted cells were recovered into the tube by briefly centrifuging (<10 s) with a microcentrifuge. To wash off any cells that might have stuck to the walls of the capillary, 2 μL of sterile PBS buffer was then added to the capillary piece and the tube was centrifuged again.

MDA, 16S rRNA gene clone library construction, and phylogenetic analysis.

The REPLI-g Single Cell Kit (Qiagen) was used for cell lysis and MDA according to the manufacturer’s instructions. MDA was performed for 8 h at 30 °C and inactivated for 3 min at 65 °C. Finally, MDA products were diluted 50-fold in sterile nuclease-free water. Bacterial 16S rRNA genes were amplified with primers 341F (5′-CCTACGGGNGGCWGCAG-3′) and 785R (5′-GACTACHVGGGTATCTAATCC-3′) (105, 106). PCR mixtures (25 μL) contained 5 μL of diluted template, 2.5 μL of 10× Taq Buffer with KCl, 2 μL of 25 mM MgCl₂, 2.5 μL of 2 mM dNTP, 0.5 μL of 0.5 μM each primer, and 1 μL of Taq DNA polymerase (5 U/μL) (all reagents from Thermo Scientific). PCR analysis was performed according to the following program: 95 ° for 4 min and 35 cycles of 94 °C for 40 s, 51 °C for 30 s, and 72 ° for 1 min, with a final elongation step at 72 °C for 10 min. The PCR product was purified using the QIAQuick PCR Purification Kit (Qiagen), and a clone library was produced using the TOPO TA cloning kit (Invitrogen) according to the manufacturer’s instructions. Forty-eight clones were picked from each clone library and were sent to a company (Microsynth) for Sanger sequencing. The flow grams of all 16S rRNA gene sequences were inspected manually, and only-high quality sequences were retained. Cloning vector- and primer-derived sequences were removed, and the trimmed 16S rRNA gene sequences were chimera-checked with the UCHIME program (107). All nonchimeric sequences having at least 250 nt were analyzed with ARB software and the SILVA SSU Ref NR 111 database (108). A tree was constructed from near-full-length 16S rRNA gene sequences closely related to clone sequences using the maximum likelihood method RAxML (109). Bootstrap analysis was performed (rapid bootstrap algorithm) using 500 bootstrap iterations. Clone sequences were then added to the tree using the quick-add parsimony method in ARB. Sequences have been deposited in the National Center for Biotechnology Information Nucleotide database under accession numbers KP055097–KP055178.

NanoSIMS Sample Preparation, Measurements, and Data Evaluation.

Samples were deposited onto boron-doped silicon wafers (7 × 7 × 0.7 mm; Active Business Company GmbH), washed with ddH₂O, and air-dried. Measurements were carried out on a NanoSIMS NS50L (Cameca). Data were acquired by scanning a finely focused Cs⁺ primary ion beam over 30 × 30–40 × 40-μm² areas (512 × 512 pixels and ∼80-nm probe size). The primary ion beam current and the dwell time were 1.2–2.2 pA and 1.5–10 ms per (layer * pixel), respectively. The base pressure in the analysis chamber was <1.8E-10 torr, the minimized residual gas adsorption. The secondary ions ¹²C¹H⁻, ¹²C²H⁻, ¹⁶O¹H⁻, ¹⁶O²H⁻, ¹²C¹⁴N⁻, ³¹P⁻, and ³²S⁻ were detected in parallel and with a mass resolving power of >10,000 for OH⁻ and CH⁻. ¹⁶O¹H⁻ and ¹⁶O²H⁻ or ¹²C¹H⁻ and ¹²C²H⁻ were used for determination of hydrogen isotope composition. To discriminate between isobaric species, detectors were positioned to blank ¹²CH₂⁻ and ¹⁶O¹H₂⁻ selectively. The contributions of ¹³C¹H⁻ and ¹⁷O¹H⁻ were subtracted from ¹²C²H⁻ and ¹⁶O²H⁻ by estimating their values from ¹²C¹H⁻ and¹⁶O¹H⁻ and multiplying by the natural abundance ¹³C/¹²C and ¹⁷O/¹⁶O values [0.0109 and 0.000380, respectively (110)]. Measurements with unlabeled E. coli cells were used to validate this approach, and resulted in D/(H + D) values of 0.0214 ± 0.0495 at% (1σ) and 0.0181 ± 0.0243 at% (1σ) based on CH⁻ and OH⁻ detection, respectively. These values are well within the range of natural variation and did not significantly differ from each other (P = 0.67) (110). Because the analytical precision of measurements was dominated by counting statistics, values obtained from the ion species with the higher intensity were used for quantification of the isotopic composition. For data intercomparison, images acquired from regions close to the center of the cells were evaluated using the WinImage software package (Cameca). Image stacks were aligned to correct for drift of the primary ion beam and/or sample stage, dead time-corrected on a per-pixel basis, and accumulated. Regions of interest (ROIs), referring to individual cells, were manually defined by ¹²C¹⁴N⁻ and ³¹P⁻ intensity as indicators of cellular biomass and cross-checked by the topographical/morphological appearance of the sampled areas in secondary electron images. The isotope fraction D/(H + D) was calculated per pixel by

$$\text{D}/\left(\text{H}+\text{D}\right)=\frac{{}^{12}\text{C}{}^2\text{H}_{corr}^{\mathrm{-}}}{\left({}^{12}\text{C}^1{\text{H}}^{-}+{}^{12}\text{C}{}^2\text{H}_{corr}^{\mathrm{-}}\right)}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\text{or}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\text{D}/\left(\text{H}+\text{D}\right)=\frac{{}^{16}\text{O}{}^2\text{H}_{corr}^{\mathrm{-}}}{\left({}^{16}\text{O}^1{\text{H}}^{-}+{}^{16}\text{O}{}^2\text{H}_{corr}^{\mathrm{-}}\right)},$$

where the subscript corr refers to the subtraction of the intensity originating from the coeval detection of the isobaric ¹³C¹H⁻ and ¹⁷O¹H⁻ ions, respectively. Mean D/(H + D) values for each ROI were then calculated.