Live tracking metabolic networks and physiological responses within microbial assemblages at single-cell level Open Access

Abstract

Microbial interactions impact the functioning of both natural and engineered systems, yet our ability to directly monitor these highly dynamic and spatially resolved interactions in living cells is very limited. Here, we developed a synergistic approach coupling single-cell Raman microspectroscopy with ¹⁵N₂ and ¹³CO₂ stable isotope probing in a microfluidic culture system (RMCS-SIP) for live tracking of the occurrence, rate, and physiological shift of metabolic interactions in active microbial assemblages. Quantitative and robust Raman biomarkers specific for N₂ and CO₂ fixation in both model and bloom-forming diazotrophic cyanobacteria were established and cross-validated. By designing a prototype microfluidic chip allowing simultaneous microbial cultivation and single-cell Raman acquisition, we achieved temporal tracking of both intercellular (between heterocyst and vegetative cells of cyanobacteria) and interspecies N and C metabolite exchange (from diazotroph to heterotroph). Moreover, single-cell N and C fixation and bidirectional transfer rate in living cells were quantified via SIP-induced characteristic Raman shifts. Remarkably, RMCS captured physiological responses of metabolically active cells to nutrient stimuli through comprehensive metabolic profiling, providing multimodal information on the evolution of microbial interactions and functions under fluctuating conditions. This noninvasive RMCS-SIP is an advantageous approach for live-cell imaging and represents an important advancement in the single-cell microbiology field. This platform can be extended for real-time tracking of a wide range of microbial interactions with single-cell resolution and advances the understanding and manipulation of microbial interactions for societal benefit.

Introduction

Microbial interactions are ubiquitous in nature and have profound effects on the structure and function of microbial communities, biogeochemical cycles, and even impact global phenomena, like climate change (1–4). Different forms of microbial interactions have been documented in the environment, such as diazotroph–nondiazotroph (5, 6), phototroph–heterotroph (7), and bacteria–host (8). Metabolite exchange and the resultant element transfer are crucial for microbial interactions (4). They not only enable microbes to optimize the use of scarce nutrients and to keep microbial guilds in nutrient and energetic equilibrium but also act as fundamental drivers of global element cycling (2, 4). Metabolic interactions and the associated element fluxes are complex and dynamic processes that occur at the single-cell level and develop along various extrinsic factors, such as nutrients (9), light (10, 11), temperature (12, 13), and pH (ocean acidification) (14). To resolve such intimate and dynamic microbial interactions, it is necessary to track the metabolite exchange and physiological responses within and among living cells with single-cell resolution. Yet, our ability to monitor these processes non-destructively and quantitatively with high microscale resolution is very limited.

Currently, the only method competent to trace element fluxes and resolve metabolic interactions at the single-cell level is nanoscale secondary ion mass spectrometry coupled with stable isotope probing (NanoSIMS-SIP) (3, 5, 9, 15–17). NanoSIMS-SIP offers a snapshot of the metabolic processes taking place at a certain time with very high spatial and mass resolution. It has given important insights into trophic interactions and element (C, N, and S) fluxes in various microbial communities (3, 5, 9, 15–17). However, the analytical conditions of ultrahigh vacuum and the destructive nature of NanoSIMS do not allow the monitoring of living cells in real-time. A further challenge for NanoSIMS is to simultaneously resolve the rapid changes of cells physiology in response to external stimuli, which normally require retrieving comprehensive information of molecular events. Transcriptomics and proteomics allow a comprehensive analysis of cellular responses and give insights into the mechanism and biochemical pathway underlining microbial interactions (3, 18, 19). However, the bulk measurements of the total microbial population can average out the cell-to-cell variance which is the key strategy for microbes to adapt to fluctuating environments. Moreover, transcriptomics and proteomics depend on highly expressed genes and proteins and may overlook subtle metabolic and physiological changes.

Raman spectroscopy is a promising noninvasive technique for studying the metabolic and physiological state of living microbes at the single-cell level (20–23). Raman spectra of a microbial cell provide its intrinsic biochemical fingerprints, including nucleic acids, lipids, proteins, metabolites, and pigments, which are highly sensitive to cellular composition and physiological states under varying conditions (20–23). Raman spectroscopy is also flexible to integrate with stable isotope labelings such as ¹³C, ¹⁵N, and ²D (20, 24–29). Characteristic SIP-induced Raman shifts can be measured to quantify the metabolic assimilation of isotope-labeled substrates, thereby providing a way to reveal the associated microbial functions and activities. However, Raman-based techniques reported to date are not ready for tracking dynamic metabolic interactions and physiological states of living cells. Such developments require the exploration of SIP-specific Raman biomarkers (which are still scarce compared with the large diversity of microbial functions) and the design of devices suitable for both live-cell culturing and Raman monitoring.

Here we developed a synergistic approach combining single-cell Raman microspectroscopy analysis with ¹⁵N and ¹³C SIP in a microfluidic culture system (RMCS-SIP) to live track metabolic interactions at both intercellular and interspecies levels. We first tested the approach with diazotrophic cyanobacteria capable of both N₂ and CO₂ fixation. Cyanobacteria were chosen due to their significant contributions to biogeochemical cycling, extensive interactions with other organisms, and ability to function in both local and global ecosystems (11, 15, 16, 30–33). In addition, N₂- and CO₂-fixing cyanobacteria offer a useful system to follow the transfer of both nitrogen and carbon-based nutrients simultaneously. Some CO₂-fixing vegetative cells (Vegs) in cyanobacteria differentiate into N₂-fixing heterocyst cells (Hets) under N-deficient conditions (34, 35). Hets and Vegs are spatially separated and functionally interdependent by exchanging N and C metabolites like amino acids and sugars. Here we used the Hets–Vegs and diazotrophs–heterotrophs interactions as model systems to investigate intercellular and interspecies interactions frequently found in nature, respectively. Quantitative and robust Raman-¹⁵N and ¹³C biomarkers specific for N₂ and CO₂ fixation were established and strictly validated. Using RMCS-SIP, we achieved a quantitative and time-resolved tracking of element fluxes and physiological changes within individual living cells. Moreover, interspecies element transfer between diazotrophs and heterotrophs was also temporally resolved. We recognize that this approach to resolving functional microbial interactions in real-time at single-cell precision could result in significant advances in disciplines including microbial ecology and ecophysiology.

Results

Raman spectra of heterocysts and vegetative cells in N₂-fixing cyanobacteria

Here, two heterocyst-bearing cyanobacteria, i.e. Anabaena sp. PCC 7120 and bloom-forming Raphidiopsis raciborskii XM1, were examined (33). These strains are morphologically distinct, with Anabaena sp. cells being more ellipsoidal while R. raciborskii is more columnar (Fig. 1A and C). Both contained Hets (red arrow) that were morphologically larger than Vegs (blue arrow). Single-cell Raman spectra from both Hets and Vegs cell types were acquired. When excited with visible lasers at 532 or 633 nm, strong fluorescence originating from photosynthetic pigments (e.g. chlorophyll, cyanophycin, scytonemin) (36–41) almost completely masked the Raman signal. Only slight carotenoid bands appeared due to the resonance Raman enhancement at 532 nm (Fig. S1A). The fluorescence was substantially reduced when a near-infrared 785 nm laser was used. High-quality Raman spectra containing more bands other than carotenoids were obtained from both individual Hets and Vegs cell types (Fig. 1B and D). These bands were mainly assigned to pigments including phycobilin, chlorophyll, and carotenoids (Table S1) (36–40). Notably, Hets and Vegs in both N₂-fixing cyanobacteria displayed prominent spectral differences with the two peaks at 668 cm⁻¹ and 1,637 cm⁻¹ (phycobilin) (36, 37, 40) in Vegs being much stronger than those in Hets (Fig. 1B and D), indicating a lower amount of photosynthetic pigments in Hets. This observation is consistent with previous work using a 967-nm laser for excitation (36). Considering that Hets are differentiated from Vegs under N-depletion conditions, these two Raman peaks can be selected as an internal biomarker for monitoring the differentiation process.

Quantitative identification of Raman-SIP biomarkers for N₂ and CO₂ fixation

Currently, there are no available Raman biomarkers associated with N₂ fixation in diazotrophic cyanobacteria, hindering the study of intercellular N and C exchange by this approach. To identify key Raman biomarkers, Anabaena sp. and R. raciborskii were cultured in an N-free medium supplemented with increasing concentrations of ¹⁵N₂ from natural abundance 0.36 at% to ∼99.36 at% ¹⁵N. From the mean Raman spectra of ∼20 randomly selected single cells, it was interesting to observe that the bands at 1,111, 1,146, and 1,285 cm⁻¹ in 0.36% ¹⁵N shifted to 1,103, 1,133, and 1,280 cm⁻¹ in 99.36% ¹⁵N, respectively (Fig. 2A and B). The same ¹⁵N-associated shifts were observed in both cyanobacteria species (Fig. 2B). These shifts were identical to those in cyanobacteria cultured with bioavailable ¹⁵NO₃⁻ (Fig. S2B and C) and were all assigned to N-related C–N vibrations (Table S1), thus indicating that the synthesis of identical ¹⁵N-biocomponents following ¹⁵N₂ fixation and ¹⁵NO₃⁻ assimilation. These findings demonstrated that ¹⁵N₂ fixation and incorporation in newly synthesized biomolecules accounted for the ¹⁵N-induced changes. In contrast, such shifts were not observed in Raman bands wherein N was absent, such as the strong C–C stretching band at 1,157 cm⁻¹ from carotenoids (Fig. 2A, B, and Fig. S2) (38). Hence, these ¹⁵N₂-induced Raman changes were selected as robust biomarkers for N₂ fixation in cyanobacteria.

To determine whether these shifts can be used to quantify the N₂-fixing activity, we conducted a validation process. First, principal component analysis (PCA) was employed to discern and quantify subtle spectral variations by avoiding deconvolution errors due to the overlap of ¹⁵N-assoicated bands with neighboring bands. PCA scores accounting for ¹⁵N-induced spectral variations displayed a linear relationship with the amount of ¹⁵N₂ supplied to both Anabaena sp. (R² = 0.9803) (Fig. 2C) and R. raciborskii (R² = 0.9103) (Fig. S3B). Meanwhile, bulk analysis via isotope ratio mass spectrometry (IRMS) was employed to assess the reliability of Raman biomarkers in quantifying ¹⁵N abundance in cells. IRMS results showed increasing ¹⁵N abundance of 0.37, 9.91, 22.06, 39.66, 59.62, and 79.19% with increasing ¹⁵N₂ incubations concentrations in Anabaena sp. (Fig. 2D). Remarkably, the ¹⁵N abundance quantified by IRMS correlated linearly with the PCA scores representing ¹⁵N-induced Raman spectral changes (R² = 0.9794) (Fig. 2D), clearly illustrating a larger spectral change at a higher intracellular ¹⁵N abundance. As such, these ¹⁵N-induced Raman changes can not only act as robust biomarkers for ¹⁵N₂ fixation but also be used to quantify the fixation activity/rate of cyanobacteria.

Both carbon and nitrogen fixation are the primary processes by which photoautotrophic cyanobacteria assure its C and N demands and fuel metabolic interactions with other species. To identify Raman biomarkers for CO₂ fixation, cyanobacteria were cultured with ¹³C-NaHCO₃ as the sole carbon source. For both cyanobacteria strains, prominent ¹³CO₂-induced band changes were observed, including 1,157 to 1,129 cm⁻¹ (C–C stretching of carotenoids) (38), 1,525 to 1,486 cm⁻¹ (C=C stretching of carotenoid) (38), and 1,637 to 1,579 cm⁻¹ (C=C stretching of phycobilin) (Fig. 2E and F) (36, 40). Consecutive shifts along with an increase in ¹³C from 1.11% (natural abundance) to 99.00% were obvious (Fig. 2E). All of these vibrations were C related, confirming that ¹³CO₂ fixation induced the shifts. The spectral PCA scores of Anabaena sp. increased linearly with the ¹³C percentages in medium (R² = 0.9493) (Fig. 2G) and correlated well with the ¹³C enrichment in N₂-fixing cyanobacteria measured by IRMS (R² = 0.9506) (Fig. 2H), demonstrating that these shifts can quantify CO₂ fixation.

To determine the detection limit of ¹⁵N and ¹³C, the threshold of PCA scores that can consider a cell labeled was calculated based on the mean + 3 standard deviation of PCA scores from natural abundance cells following previous reports (28, 42). The corresponding threshold PCA scores were then used to determine the Raman detection limit based on the linear relationship between PCA scores and isotope contents (Fig. 2D and H). The detection limit was determined as 9% for ¹⁵N and 7% for ¹³C enrichment in cells, respectively.

Construction of the microfluidic culture system

An RMCS was designed to enable cell cultivation, microscopic observation, and in situ Raman measurements simultaneously in one chip (Fig. 3). Several strategies were adopted to construct RMCS as polydimethylsiloxane (PDMS) and glass that are typically used to fabricate microfluidic chips can produce strong Raman/fluorescence signals and interfere or even mask Raman of microbes. As shown in Fig. 3A–C, firstly, quartz (white) with a low Raman background was used as the bottom substrate to bind PDMS (yellow). Secondly, channels (green) were specially designed and fabricated on PDMS for cells and medium to flow in and out. The height of channels from the surface of the quartz to the bottom of PDMS was 100 μm after proving that this height would allow a neglectable Raman signal of PDMS to be excited by the laser focusing on the quartz below (Fig. S4). The shapes of the channels at the inlet and outlet were designed differently. A narrow neck structure (50 μm wide) was fabricated as an outlet, to allow medium to flow out while holding cells in. The inlet was 500 μm wide and allowed the overflow of surplus cells during cultivation. Fig. 3D–F shows the RMCS mounted on a Raman microscopic stage and the microscopic image of cyanobacteria cultured inside. With this RMCS, living cells can be observed and measured by singe-cell Raman spectroscopy, and culture media amended with ¹³C-NaHCO₃, ¹⁵N₂ or nitrate can be flexibly changed.

Live imaging N and C fluxes in cyanobacteria and derivation of single-cell-based rate

To track N and C fixation and transfer in and between living cyanobacteria, cells were cultured separately with either ¹⁵N₂ or ¹³C-NaHCO₃ and measured in RMCS, i.e. ¹⁵N and ¹³C traces were applied in parallel. Four continuously connected cells including one Het and three Vegs from the same filament of Anabaena sp. were subjected to Raman mapping at 0, 48, 72, and 96 h. Based on the linear relationship between cellular ¹⁵N and ¹³C contents and PCA scores representing isotope-induced spectral variations (Fig. 2D and H), both the temporal and spatial dynamics of ¹⁵N and ¹³C enrichments in these individual Hets and Vegs were determined, and visualized by color coding in Raman mapping images (Fig. 4). Single-cell inspection indicated that Vegs and Hets displayed distinct temporal dynamics toward ¹³C and ¹⁵N enrichment. Significant enrichment in ¹³C was initially detected only in the Vegs, followed by a progressive increase in both Vegs and Het where more newly fixed ¹³C were allocated to Vegs until a balanced distribution was reached (Fig. 4A and C). This result indicated that Vegs not only performed active CO₂ fixation for their use but also transferred newly fixed ¹³C to the Het. In contrast, ¹⁵N fixation and enrichment were detected initially in the Het, followed by observation of elevated levels of ¹⁵N in both Het and Vegs, with the Het having a higher ¹⁵N level than the three adjacent Vegs (Fig. 4B and D), demonstrating that Hets fixed N₂ and actively transferred the newly fixed ¹⁵N to Vegs. The observed gradual N and C transfer is due to that Hets and Vegs rely on each other for energy and growth. Hets lack a functional photosystem and thus rely on Vegs to supply the C and energy required for N fixation and exportation. At the same time, Vegs require new N import to maintain their photosynthesis, growth, and differentiation (35). After a 96-h incubation, the ¹⁵N or ¹³C level of the Het and Vegs converged, implying that the newly fixed ¹⁵N and ¹³C reached an even distribution among the four cyanobacterial cells. These findings demonstrated that RMCS-SIP was competent in quantifying N and C fixation and bidirectional transfer within and between individual live cells. Notably, although the three Veg cells were located at different distances from the Het, no obvious ¹⁵N gradient was observed at any time point which could be attributed to the rapid transport of N among Vegs to keep pace with the N demand for their growth as formerly proposed (17). In addition to the in situ measurement of these four cells, ex situ Raman measurement of around 20 Vegs and Hets randomly selected from cyanobacterial filaments taken out at different time points was also performed. Similar temporal patterns of N and C fixation and transfer were observed in both Anabaena sp. and R. raciborskii (Fig. S5).

Because the same living individual cells were measured using RMCS at different time points, single-cell flux rates can be calculated based on the ¹⁵N or ¹³C enrichment over time. The single-cell fixation rates were calculated as the amount of ¹⁵N (¹³C) fixed in the individual Hets (Vegs) over the incubation time (Fig. 4A and B). Because the fixed N released as ammonia was not considered, the net instead of gross fixation rates were determined (16). The single-cell ¹⁵N (¹³C) transfer rates were calculated as the amount of ¹⁵N (¹³C) transferred to Veg (Het) over time (17, 31). All the rate values of fixation and transfer are presented in Table S2. The C fixation started slowly (∼0.30% cell⁻¹ h⁻¹ at 48 h), but increased rapidly from 48 to 72 h (∼1.60% cell⁻¹ h⁻¹), then slowed down (∼0.70% cell⁻¹ h⁻¹) ending with 69.2% ¹³C net fixation at 96 h. The transfer of ¹³C to Het was also initiated slowly but kept increasing (2.05% cell⁻¹ h⁻¹ at 96 h) until a balanced ¹³C enrichment was reached. In contrast, N fixation initiated rapidly (0.86% cell⁻¹ h⁻¹ at 48 h), and the rate was almost constant except for a small drop at 96 h (∼0.56% cell⁻¹ h⁻¹), ending with 76.8% ¹⁵N net fixation. The transfer trend of ¹⁵N from Het to Veg was similar to that of ¹³C from Veg to Het, namely, a slow start in the initial 48 h followed by a rapid increase until an even distribution of ¹⁵N or ¹³C was reached among the four cells.

Impact of nutrients on cellular physiology and element fluxes in living cells

We further studied the impact of nutrient additions on cell physiology in living cyanobacterial cells and the related element fluxes. Time wise impact of NO₃⁻ on living Hets and Vegs was studied by adding NO₃⁻ containing medium into the microfluidic channels. Following NO₃⁻ addition, although Hets can still be morphologically distinguished from Vegs (red circles in Fig. 5A), the Raman band at 1,637 cm⁻¹ (assigned to phycobilin) (36, 37, 40) specific to Vegs appeared in Hets at 2 h and increased in intensity with the incubation time (Fig. 5B). At 72 h, Hets were no longer recognized by their morphology, and the intensity of 1,637 cm⁻¹ band became almost the same as those of Vegs (Fig. 5C), indicating Raman-based approach can capture the occurrence of physiological changes much earlier (2 h) than morphological changes (>48 h). Moreover, with nitrate addition, ¹⁵N-induced spectral redshifts were shifted back over time in both Hets and Vegs, such as those from 1,103 to 1,111, 1,113 to 1,146, and 1,280 to 1,285 cm⁻¹ (Fig. 5B and C). Accordingly, the cellular ¹⁵N at% derived from spectral changes decreased, indicating that ¹⁵N in living cells was gradually replaced with ¹⁴N due to nitrate assimilation (Fig. 5D). Moreover, contrary to the distinct ¹⁵N uptake dynamics in Hets and Vegs (Fig. 4B and D), the ¹⁴NO₃⁻ assimilation dynamics (Fig. 5D) were very similar in Hets and Vegs with the original Hets being even faster in NO₃⁻ substitution. The reason could be that the original Hets required more NO₃⁻ to support state transition. Therefore, sensitively tracing cellular physiological changes by Raman provides an important single-cell means to decipher the state and function of living cells under varying conditions.

From intercellular to interspecies metabolic exchange

To demonstrate the wide applicability of the presented approach, the interspecies metabolic exchange between N₂-fixing Azotobacter chroococcum and heterotrophic Escherichia coli was also traced by co-culturing the two species in media using ¹⁵N₂ as the only N source. Individual A. chroococcum and E. coli can be easily distinguished based on their distinct Raman spectral features and morphologies. Azotobacter chroococcum displayed characteristic resonance Raman bands assigned to cytochrome c at 749, 1,114, 1,312, and 1,589 cm⁻¹ (Fig. S6A), which has essential role in protecting the activity of oxygen-labile nitrogenase (26, 43, 44). These cytochrome c bands were absent in E. coli (Fig. S6B). Following the incubation with ¹⁵N₂ in the co-culture system, characteristic ¹⁵N-induced Raman band shifts were observed in both species (Fig. 6A and B), i.e. 1,129 to 1,114 (C–N vibration of cytochrome c) and 1,245 to 1,232 cm⁻¹ (Amide III) in A. chroococcum, and 1,245 to 1,232 cm⁻¹ (Amide III) and 1,579 to 1,570 cm⁻¹ (ring stretching of guanine and adenine) in E. coli. These ¹⁵N-induced shifts were all from chemical bonds containing N and consistent with previous works (24, 26, 27), indicating the uptake and substitution of ¹⁴N with ¹⁵N. By comparison, when the E. coli were cultured alone in ¹⁵N₂ without A. chroococcum, no band shifts were observed from those in ¹⁴N₂ (Fig. S7), indicating the inability of E. coli to utilize ¹⁵N₂ without A. chroococcum. Additionally, the time-course results indicated that the prominent shift in A. chroococcum happened at 24 h (Fig. 6C), much earlier than E. coli after 72 h (Fig. 6D). Because only A. chroococcum can fix ¹⁵N₂ but E. coli cannot, these results clearly demonstrated the fixation of ¹⁵N₂ by A. chroococcum and the following interspecies transfer of ¹⁵N from A. chroococcum to E. coli.

Discussion

Microbial interactions impact the functioning of natural and engineered systems, yet our ability to see these dynamic and highly spatially resolved interactions directly in living cells is extremely limited. Here quantitatively live tracking of N and C fixation and bidirectional transfer at the single-cell level was achieved via RMCS-SIP. This approach is applicable for both intercellular metabolic exchange within the same organism and interspecies metabolic exchange between organisms of different physiologies. At the same time, RMCS-SIP reveals physiological responses to nutrient stimuli through a comprehensive metabolic profiling, providing multimodal information to understand the dynamics of microbial functions and interactions under fluctuating conditions.

Major methodological innovations of RMCS-SIP are the establishment of robust Raman-¹⁵N and ¹³C biomarkers specific to N₂ and CO₂ fixation and the development of the microfluidic culture chip. Despite the recognized potential of Raman-SIP in revealing microbial functions, available SIP-associated Raman biomarkers have been very limited, particularly for ¹⁵N when compared with the more frequently used ¹³C and ²D. This is because ¹⁵N substitution causes a significantly smaller redshift than ¹³C and ²D (24). Here we showed ¹⁵N-induced characteristic Raman shifts of chlorophyll can be used as quantitative markers for N₂ fixation in diazotrophic cyanobacteria. These biomarkers were confirmed with distinct cyanobacteria species and validated with IRMS, revealing the relatively common applicability for different diazotrophic cyanobacteria. Together with our previous reports on ¹⁵N₂-induced cytochrome c band shift for the identification of non-photosynthetic diazotrophic bacteria in the soil such as Rhizobium (26), these identified Raman biomarkers can be used to study the metabolism of different N₂-fixing microorganisms and their interactions with other bacteria in artificial or natural ecosystems. Furthermore, the designed microfluidic culture chip enables the simultaneous cultivation of microorganisms and Raman detection of metabolically interconnected living cells with minimal interference from PDMS due to its sufficiently large distance from cells on quartz. RMCS is flexible to couple with other state-of-the-art approaches, like the application of laser tweezer Raman (45, 46), to capture and detect microbial cells that may move rapidly in solution. The design of the microfluidic culture system can be further tailored to accommodate special cultivation demands such as anaerobic, dark, and high-pressure conditions expanding uses for microbial interactions across many habitats. Presently, one limitation of RMCS-SIP is that ¹⁵N and ¹³C labeling cannot be always applied and detected simultaneously, for instance when Raman shifts caused by ¹³C masked those caused by ¹⁵N. This limitation could be resolved by using advanced chemometric or machine-learning methods to recognize subtle spectral changes. Additionally, the bleaching effect induced by a 785-nm laser, especially on photosynthetic bacteria, limits the time points for spatiotemporal Raman imaging. This could be resolved by the future use of a 1,064 nm laser which may minimize bleaching and autofluorescence effects (36). Finally, the sensitivity of Raman in detecting isotopes is lower than that of NanoSIMS. A high enrichment of ¹⁵N (>9%) and ¹³C (>7%) in cells is required, rendering the analysis of slower processes and low-activity microbial cells difficult. This can be partially avoided by using a higher percentage of labeling during incubation and a longer exposure time.

Proof of principle application showed that RCMS-SIP can temporally track both intercellular and interspecies (from autotroph to heterotroph) N and C transfer, and simultaneously determine the single-cell nutrient uptake and transfer rates. Although N and C bidirectional transfer between individual Het and Veg of cyanobacteria is known (17), here the occurrence and rate of this process are determined in living single cells. Microbial interactions based on N₂ and CO₂ fixation are widespread in nature, including associations of N₂⁻ and CO₂⁻ fixing microorganisms with bacteria, archaea, algae, protists, and plants, with implications in ecosystem functioning (e.g. induction of recurrent bloom) (11, 16, 32, 47) or nature-based services (e.g. enhanced wastewater treatment and carbon capture) (48, 49). Given the similarities to the system investigated here, we anticipate that RCMS could be used to live track metabolic interactions within such associations and advance our understanding of how complex communities operate and serve to harness microbial interactions for human benefits.

Furthermore, RMCS enabled the measurement of physiological responses of living cells to nutrient stimuli, a feature that can be extended to understand how environmental fluctuations impact metabolic processes. An increase in the phycobilin characteristic band was rapidly detected in Hets following only 2 h of nitrate addition, much earlier than morphological changes at 48 h. Phycobilin is an important pigment for photosynthesis and was previously reported as absent or at a lower level in Hets than in Vegs (50), consistent with the Raman observation (Figs. 1 and 5). The increase in Hets may indicate a physiological and functional shift toward an elevated photosynthetic activity before their morphology changes. It is noteworthy that the ability of Raman in chemically profiling of multiple endogenous molecules plays a key role in capturing such subtle physiological changes in individual cells (22, 23, 51, 52). This is also applicable to many cases wherein cell type and function cannot be readily determined by their morphology as observed in Hets and Vegs here (10, 53). For example, in some cyanobacteria, N- and C-fixation occur in the same cells but are temporarily separated as a function of the diel cycle, without morphological changes (10). Therefore, sensitively tracing of ongoing physiological evolution by Raman provides a useful way to decipher the competitive advantage from a physiological perspective for microbes encountering periodic or episodic environmental alterations, such as nutrients, temperature (climate change), and pH (ocean acidification). The observed rapid response to nitrate addition in physiological states of cyanobacteria indicates their advantage in trophic strategy to acclimate a wide range of nutrients. This may also partly explain the continuous expansion of R. raciborskii toward higher latitudes in past decades (32).

In summary, RMCS-SIP represents an important advancement for single-cell microbiology in both natural and engineered systems. It offers a non-destructive single-cell means for real-time monitoring the occurrence, rate and physiological shifts of metabolic interactions in metabolically active microbial populations. It complements the destructive NanoSIMS-SIP-based chemical imaging approaches, which are currently the only single-cell methods used to quantify element uptake and fluxes within microbial guilds (3, 5, 9, 11, 15–17, 53). We believe this innovative non-destructive technique can be expanded for real-time unraveling a wide range of metabolic interactions in situ. This will advance our understanding and harnessing of microbial interactions for societal benefits like reducing algal bloom and climate change and improving wastewater treatment and crop productivity.

Materials and methods

Bacterial strains, media, and culture conditions

Two types of heterocyst-forming diazotrophic cyanobacteria i.e. Anabaena sp. and R. raciborskii were used in this study. A model strain Anabaena sp. PCC 7120 was obtained from the Freshwater Algae Culture Collection of the Institute of Hydrobiology, China. The bloom-forming cyanobacterium of R. raciborskii XM1 was isolated in a Xiamen reservoir during a summer bloom in 2018 (33). To further confirm the N₂ fixation potential of these strains, dinitrogenase reductase genes (nifH) were amplified and visualized on agarose gels. Briefly, extracted cyanobacterial DNAs were polymerase chain reaction amplified with primers nifH1 (5′-TGYGAYCCNAARGCNGA-3′) and nifH2 (5′-ADNGCCATCATYTCNCC-3′). The appearance of nifH at around 359 bp in agarose gel electrophoresis is shown in Fig. S2a. These two cyanobacterial strains were cultured in BG11 or N-free BG11 (BG11₀) medium at ∼28°C and 90 rpm under constant light condition. They were cultured in BG11₀ to ensure heterocyst formation before isotope labeling. N₂-fixing A. chroococcum (ACCC10096) and E. coli MG1655 were purchased from the Guangdong Culture Collection Center of Microbiology, China, and cultured in N-free azotobacter media and Luria−Bertani broth, respectively before co-culture. All culture medium recipes are provided in Supplementary information.

The chemicals used for stable isotope incubation included ¹⁵N-NaNO₃ (98 atom% ¹⁵N, Sigma−Aldrich, United Kingdom), ¹³C-NaHCO₃ (99 atom%; Cambridge Isotope Laboratories, Inc., UK), and ¹⁵N₂ (99 atom%; Aladdin, China). Unless otherwise stated, all chemicals were purchased from Sinopharm Chemical Reagent Co., China.

Incubation of N₂-fixing cyanobacteria with ¹⁵N and ¹³C stable isotopes

¹⁵N₂ incubation was performed according to our previous work (26). Briefly, an aliquot of 20 mL of BG11₀ medium was filled into a 120-mL crimp-top vial and inoculated with 500 μL of Anabaena sp. PCC 7120 or R. raciborskii XM1 solution. The vials were sealed, and air in the headspace was replaced with Ar gas after 10 min of continual aeration. Then, a gas mixture consisting of N₂ and O₂ (4:1 v/v) was aerated into the vials to achieve different volume percentages of ¹⁵N₂ (¹⁵N₂/(¹⁵N₂ + ¹⁴N₂)). Considering that ¹⁵N content is 0.36% in natural abundance and 99% in commercial areas, the absolute atmospheric ¹⁵N abundances in the incubations were calculated as 0.36, 10.22, 25.02, 49.68, 74.52, and 99.36%. To temporally track N₂ fixation and transfer, cyanobacteria were incubated in the prepared ¹⁵N₂-containing atmosphere and harvested after 6 or 12 h time intervals.

In parallel incubation experiments, to perform ¹³C labeling, BG11₀ medium without NaHCO₃ was aerated with Ar gas to remove dissolved CO₂. Premixed ¹²C and ¹³C-NaHCO₃ solutions with ¹³C percentages (¹³C/(¹³C + ¹²C)) of 1.11, 10.90, 25.58, 50.06, 74.53, and 99.00% (calculated after considering the ¹³C natural abundance) were immediately added to the medium. The vials were sealed, and air in the headspace was replaced with Ar gas after 10 min of continual aeration. Then, a mixture gas consisting of N₂ (as a sole source of nitrogen) and O₂ (4:1 v/v) was aerated into the vials. After inoculating 500 μL of Anabaena sp. PCC 7120 or R. raciborskii XM1 solution, cyanobacteria were cultured in these media and harvested at 6 or 12 h time intervals.

Co-culturing of A. chroococcum and E. coli with ¹⁵N₂ stable isotopes

Azotobacter chroococcum and E. coli were inoculated simultaneously into Azotobacter-specific media using ¹⁵N₂ as the sole nitrogen source and co-cultured at 28°C and 180 rpm (26). The control consisted of E. coli grown in the same media with ¹⁵N₂ and ¹⁴N₂ as the only nitrogen sources in the absence of A. chroococcum.

Single-cell Raman measurement and data analysis

Single-cell Raman measurements were conducted using a LabRAM Aramis confocal micro-Raman system (HORIBA Jobin-Yvon, Japan). For cyanobacteria, excitation was mainly provided by a 785-nm near-infrared laser (Diode laser, Sacher Lasertechnik). For A. chroococcum and E. coli, excitation was provided by a 532-nm laser (Nd:YAG, UK). In ex situ measurement, cyanobacteria were washed with deionized water twice, loaded on an aluminum foil substrate, and air-dried before Raman measurements (54). A 100× objective with a numerical aperture of 0.9 (Olympus, Japan) was used for bacteria visualization and Raman measurement on Al foil. In in situ measurement, a 50× objective with a numerical aperture of 0.55 and a longer working distance of 8 mm (Olympus, Japan) was used to detect Raman signals of cyanobacterial cells in the microfluidic chip. At least 20 spectra were acquired from individual cells of each bacterial sample.

LabSpec5 software (HORIBA Jobin-Yvon) was used for Raman spectra preprocessing, including baseline subtraction, mean spectra calculation, normalization, and peak fitting. To quantify ¹⁵N- or ¹³C-induced spectral variations, PCA was performed via the IRootLab toolbox running on MATLAB 2012a (https://code.google.com/p/irootlab/) (26, 55). Spectral regions from 1,098 to 1,120 cm⁻¹ and from 1,465 to 1,545 cm⁻¹ were extracted for PCA of ¹⁵N- and ¹³C-related spectral variations in cyanobacteria, respectively. Spectral regions from 1,200 to 1,270 cm⁻¹ were used for PCA of the ¹⁵N-induced spectral variation in A. chroococcum and E. coli. The resulting PC1 scores were plotted against the ¹⁵N₂ or ¹³C-NaHCO₃ percentages in the medium or the incubation time. The processed Raman spectra and linear correlations were plotted using Python 3. The distribution map of isotopes in cells via Raman detection was plotted using MATLAB.

Quantification of ¹⁵N and ¹³C isotope incorporation by IRMS

Isotope ratio mass spectrometry was used to measure the bulk isotope contents of ¹⁵N or ¹³C in Anabaena sp. PCC 7120 was incubated with various percentages of ¹⁵N₂ or ¹³C-NaHCO₃. For ¹⁵N or ¹³C measurement, 0.01–0.03 mg of lyophilized ¹⁵N- or ¹³C-labeled Anabaena sp. PCC 7120 and 0.05–0.17 mg of urea (standards) were mixed in a tin capsule. About 0.10–0.20 mg of non-labeled lyophilized Anabaena sp. PCC 7120 used as control groups were placed in tin capsules. The samples were analyzed with an elemental analyzer (Flash HT 2000 Thermo Fisher) coupled via a ConFlo IV device to the IRMS (Delta V advantage). The bulk isotope contents were calculated using the equation below:

where At ¹⁵N% (At ¹³C%) is the percentage of ¹⁵N (¹³C) relative to the total N (C) measured by IRMS; M_urea and M₇₁₂₀ are the weights of urea and lyophilized cyanobacteria, respectively; 0.36% (1.11%) is the natural abundance of ¹⁵N (¹³C) in the urea; 46.67 and 11.3% (19.98 and 51.4%) are the total N (C) contents in urea and cyanobacteria, respectively; and ¹⁵N% (¹³C%) is the abundance of ¹⁵N (¹³C) in cyanobacteria that needs to be calculated.

Raman microfluidic culture system for cyanobacteria

The RMCS was constructed by binding PDMS with specially designed chambers to a quartz slide substrate. The RMCS production processes were as follows. First, a mask to cover the selected area was designed and prepared before the subsequent photolithography process. Second, photolithography technology was used to transfer the pattern on the mask to a silicon wafer with a suitable thickness of photoresist. Subsequently, a chip formpiston was obtained via photolithography, and PDMS was poured on it and solidified. The PDMS and quartz substrates were placed in anhydrous ethanol for ultrasonic cleaning. Finally, the dried PDMS and quartz substrates were treated with oxygen plasma and bombarded for 15 s. The bonding and sealing of the PDMS on a quartz plate were performed to produce a microfluidic chip.

Five parallel micro-chambers (1 mm wide, 100 μm high) were used for cyanobacterial culture, microscopic observation, and Raman detection. The entrance of the micro-chamber was 500 μm wide, which was used to load the strain and media solution, and the exit was designed as a narrow neck structure with a width of 50 μm to block cyanobacteria from flowing out during media input. Two wide channels (100 μm high, 2 mm wide) that connected to the chamber were used to deliver media and discharge waste solution during cyanobacterial culture. Cyanobacteria that were diluted in fresh BG11/BG11₀ medium (or amended with ¹⁵N₂ or ¹³C-NaHCO₃) were immediately loaded into the microfluidic chip through a polytetrafluoroethylene tube. The chip was sealed by linking the outlet tube to a 15-mL crimp-top vial. Thereafter, the medium flowed into the chip at a rate of 0.5 μL min⁻¹ for 15 min, and the flow was maintained at 0.3 μL min⁻¹ to continuously provide nutrients to the cells and simultaneously eliminate metabolic waste. All the cyanobacteria in the RMCS were cultured at ∼28°C under light conditions. Fig. S8 shows the continuous growth and reproduction of the cyanobacteria in the RMCS.

Raman mapping at a rectangular area of 6 μm × 14 μm containing the same four cyanobacterial cells was performed over time. The PC1 scores representing the isotope-induced spectral variations were obtained in the same way as before. The linear relationship between PC1 scores and IRMS-measured cellular ¹⁵N and ¹³C contents were used to determine the ¹⁵N and ¹³C enrichments in the four cells analyzed, which were further visualized as color coding in Raman mapping images.

Funding

This work was supported by the National Natural Science Foundation of China (91851101, 42021005, 22241603, 92251306) and the Chinese Academy of Sciences (ZDBS-LY-DQC027).

Supplementary material

Supplementary material is available at PNAS Nexus online.

Author contributions

L.C., Y.X., K.Y., and H.L. performed the experiment, analyzed the data, and plotted the figures. L.C. designed the whole experiment, wrote, and revised the manuscript. N.M. and F.M. codesigned the experiment, participated in writing and revised the manuscript. L.C., Y.X., and Y.Z. designed and verified the culture chip. X.L. and Z.Z. fabricated the culture chip. F.T. and J.Y. isolated and cultured bloom-forming cyanobacteria. S.K., B.R., and Y.G.Z. discussed the results, interpretation, and contributed to revising the manuscript. L.C., Y.X., and K.Y. contributed equally to this work. All authors provided feedback on the manuscript.

Data availability

All study data are included in the article and/or supporting information.

References

Author notes