Single-cell measurement of microbial growth rate with Raman microspectroscopy

Microorganisms and growth conditions

We cultivated Thermodesulfovibrio hydrogeniphilus (Haouari et al. 2008) and Methanobacterium NSHQ04 (Miller et al. 2018), in media containing 0%, 10%, 20%, 30%, 40%, or 50% ²H₂O (0% = local ultrapure water which contains ∼140 ppm ²H₂O naturally). Base media (described below) were modified with different percentages (v/v) of ²H₂O (Cambridge Isotope Laboratories), depending on the final isotopic enrichment desired for the experiment. All cultures were grown in the dark in anaerobic 60 ml serum vials containing 25 ml of growth medium.

Thermodesulfovibrio hydrogeniphilus HBr5^T (DSM 18151) (Haouari et al. 2008), a sulfate-reducing bacterium (SRB), was obtained from the Deutsche Sammlung von Mikroorganismen und Zellkulturen. It was grown in a modified DSM 641 medium that contained (per liter): 1.0 g NH₄Cl, 2.0 g Na₂SO₄, 1.0 g Na₂S₂O₃ × 5H₂O, 1.0 g MgSO₄ × 7H₂O, 0.1 g CaCl₂ × 2H₂O, 0.5 g KH₂PO₄, 2.0 g NaHCO₃, 1 ml trace element solution SL-10, 1 ml selenite–tungstate solution, 1 g yeast extract, 0.5 ml sodium resazurin (0.1% w/v), 0.2 g sodium acetate, 10 ml 141 vitamin solution, and 0.1 g Na₂S × 9H₂O. The headspace was flushed and over-pressurized with 1 bar H₂:CO₂ (80:20).

Methanobacterium NSHQ04 (Miller et al. 2018) is a methanogenic archaeon enriched from groundwater isolated from the Samail ophiolite in Oman. The culture was grown in a synthetic site water (NSHQ04) medium (Miller et al. 2018) containing (per liter): 420 mg NaCl, 683 mg CaCl₂ × 2H₂O, 1.2 mg H₄SiO₄, 1.7 mg NaBr, 60 mg MgCl₂ × 6H₂O, 100 mg NH₄Cl, 10 ml 141 trace elements, 10 ml 141 vitamins, 0.5 ml sodium resazurin (0.1 w/v), 20 ml antibiotic cocktail [containing per 100 ml MilliQ water, 500 mg penicillin G, 500 mg streptomycin, 500ul ampicillin (10 mg ml⁻¹ stock)], 10 ml KH₂PO₄ (1.1 w/v), 4 ml yeast extract (5% w/v), 2 ml Fe(NH₄)₂(SO₄)₂ (0.2% w/v), 10 ml sodium formate (100 mM), 10 ml cysteine–HCl (2.5% w/v), and 10 ml Na₂S (2.5% w/v). The headspace was flushed and over-pressurized with 2 bar H₂:N₂ (5:95).

Cultures of T. hydrogeniphilus and M. NSHQ04 were incubated in forced-air incubators at 65°C and 40°C, respectively. Cultures were transferred three times into media of identical deuterium enrichment. Before each transfer, cultures were grown to stationary phase, which was reached in 3–4 days for T. hydrogeniphilus and 12–14 days for Methanobacterium NSHQ04. This ensured that the cells measured were at or near isotopic equilibrium with their respective growth water. After the third transfer, 1 ml of each culture was harvested and fixed by addition of paraformaldehyde to a final concentration of 2% (v/v). Fixed cells were washed by centrifugation in successively dilute phosphate-buffered saline (PBS) of rinses [1X, 1%, 0.1%, and 0.01% (v/v) PBS] to ensure the removal of excess salts and fixative. PBS was composed of Dulbecco’s formula: KCl 200 mg l⁻¹, KH₂PO₄ 200 mg l⁻¹, NaCl 8000 mg l⁻¹, Na₂HPO₄ 1150 mg l⁻¹. After the final wash, cells were resuspended in 50 µl of 0.01% (v/v) PBS.

Sample preparation

Sample coupons were prepared by raster engraving of aluminum-coated glass slides (Deposition Research Lab Inc.) with an Epilog Mini 24 CO₂ laser machine operating at 80 W laser power, 10% etching speed, to create a 3 × 6 grid. These etchings were used to manually break the glass slide, producing 7 mm² coupons that are compatible with both Raman and nanoSIMS sample mounts. Aluminum coupons were used instead of silicon wafers (often used for nanoSIMS studies) because silicon exhibits a large peak at 520 cm⁻¹ in the Raman spectrum. All coupons were sterilized and rendered organic-clean by combustion for 8 h at 450°C in a muffle furnace. 5 µl of fixed and washed cells was spotted on individual coupons and allowed to air-dry.

Raman spectroscopy, fitting, and SIP calculations

Raman spectroscopy was conducted at the Raman Microspectroscopy Laboratory, Department of Geological Sciences, University of Colorado-Boulder (RRID:SCR_019305) on a Horiba LabRAM HR Evolution Raman spectrometer equipped with a 100 mW 532 nm excitation laser. The laser was focused with a 100X (NA = 0.90) air objective lens, resulting in a spot size of ∼1 µm. Single-cell spectra were captured in the 200–3150 cm⁻¹ range using 100% laser power (2.55 mW) over 2 acquisitions of 45 s. We note that, at time of analysis, the laser employed here was near end-of-life, thereby requiring relatively long acquisition times (45 s). After the collection of the original dataset a new laser was installed, which resulted in acquisition times of 10 s at 10% laser power. Future researchers attempting to replicate our methods should note the effective laser power of their instrument and adjust acquisition time accordingly.

A 100 µm confocal pinhole and 600 lines mm⁻¹ diffraction grating were used, resulting in a spectral resolution of ∼4.5 cm⁻¹. Duplicate acquisitions were averaged to remove noise and cosmic ray spikes. Spectra were baseline-subtracted using a polynomial fit in LabSpec 6 (Horiba Scientific). Fitting of bands in the 1800–3150 cm⁻¹ region was performed using Gaussian–Lorentzian curve fitting algorithms to integrate C–D (2040–2300 cm⁻¹) and C–H (2800–3100 cm⁻¹) bands. Raman-derived ²F was determined by calculating the CD% value of each spectrum: ²F = CD% = CD/(CH + CD) × 100% where CD and CH are the areas of the C–D and C–H bands, respectively.

To correlate Raman spectra with nanoSIMS data, extensive context maps of the sample coupons were acquired using reflected light microscopy on the Raman instrument. Before cell spotting, fiducial markings were etched into the aluminum coupon surface with a Leica LMD7000 microdissection system.

As described in the section “Results and Discussion,” modeling of growth rate was carried out with Eqs (1–3). (Kopf et al. 2016, Caro et al. 2023). Using these formulae, we modeled µ across a range of F_L, t, and a_w. We define an acceptable cutoff using a relative error term defined as the fraction of uncertainty in calculated growth rate over the growth rate itself (σ_µ/µ) where quantifiable growth rate is where relative error = (σ_µ/µ) × 100% <50% (Fig. 3).

The value of the water hydrogen assimilation constant (a_w) was determined for both organisms from the slopes of the linear regression relating ²F_biomass versus ²F_water. This value, a_w = X_w × α_biomass/w, corresponds to the isotopic offset between microbial biomass relative to its growth water. This term includes both X_w, the mole fraction of lipid H sourced from growth water, and α_biomass/w, the isotopic fractionation between whole-cell biomass and growth water (Zhang et al. 2009).

NanoSIMS

Samples were analyzed with a CAMECA NanoSIMS 50 L (CAMECA, Gennevilliers, France) housed in the Caltech Microanalysis Center at the California Institute of Technology. Before, analysis, cells were sputter-coated with 25 nm of Au. Cells were analyzed using a 2.5-pA primary Cs+ beam current, and a presputter time of 6–20 min depending on the size of the raster area. Two masses were collected in parallel (¹H^{−, 2}H⁻) using electron multipliers. Individual samples were identified using the NanoSIMS CCD camera and correlated with Raman-measurements using previously generated image maps. For all analyses, at least two frames were collected. All ion images were recorded at 512 × 512 pixel area. Single-cell isotope values were quantified using a custom script (see the section “Data availability”) relying on the sims Python library (https://github.com/zanpeeters/sims). Cell areas were manually defined using the GNU Image Manipulation Program. Isotope ratios (²R) were converted to fractional abundances (²F) with the relations: ²F = ²R/(1 + ²R) and ²R = ²H/¹H, where ¹H and ²H represent total ion counts in detectors EM1 and EM2, respectively, averaged across a cell area. Additionally, a deadtime correction (Kilburn and Wacey 2014) was applied: N = (N_m)/(1–N_m × τ/T), where N is the real number of secondary ions, N_m is the number of secondary ions measured, is deadtime (in seconds), and T is dwell time (seconds per pixel). This instrument’s detector deadtime was 44 ns. Deadtime correction had a negligible effect on isotopic values (Supplementary Text, Supplementary Fig. S7).

Single-cell isotopic measurements

To assess the sensitivity and precision of Raman microspectroscopy as a reporter of cellular ²H incorporation, we generated a large isotopic dataset of bacterial and archaeal cells grown to equilibrium with deuterated (²H₂O) media of varying hydrogen isotope compositions (0–50 at. % ²H). We focused our study on two anaerobic organisms isolated from anaerobic subsurface aquifers: a SRB T. hydrogeniphilus (Haouari et al. 2008), and a methanogenic archaeon Methanobacterium NSHQ04 (Miller et al. 2018). We correlatively acquired individual measurements of cellular deuterium enrichment (n = 351) with Raman spectroscopy and nanoSIMS (Fig. 1). We refer to hydrogen isotope content using both Raman and nanoSIMS as the fractional abundance of deuterium, ²F, reported in atom % (at. %) (see the section “Materials and methods”), where ²F = ²H/(H + ²H) × 100%. To generate hydrogen isotope calibrations, we fit simple linear models relating the hydrogen isotopic composition of microbial biomass (²F_biomass) to the organism's growth water (²F_gw) (Fig. 2).

With Raman microspectroscopy, biomass ²H content (²F_biomass) increased linearly and exhibited strong positive correlation (R² = 0.94, 0.95, for T. hydrogeniphilus and M. NSHQ04, respectively) with the ²H content of the growth water (²F_gw) (Fig. 2, Table 1B). Raman measurements of biomass deuterium content exhibited an average standard deviation of σ_2F = 2.89 at. %. The variability of Raman-derived ²F_biomass increased with the isotopic composition of the growth water: a maximum standard deviation was observed at the ²F_gw = 50% ²H₂O condition, where σ_2F = 3.77 at. %. Corresponding measurements by nanoSIMS of the same cells followed a similar trend: ²F_Biomass increased linearly with the growth water applied (R² = 0.84, 0.96 for T. hydrogeniphilus and M. NSHQ04, respectively) and the average standard deviation of nanoSIMS ²F was σ_2F = 1.86 at. %. Similar to Raman, the standard deviation of nanoSIMS measurements increased along with the isotopic composition of growth water, reaching a maximum at the 50% label where σ_2F = 3.20 at. % (Table 1A). We emphasize that these measures of variance are conservative over-estimates of analytical error due to heterogeneity in the hydrogen isotopic enrichment of individual cells. Our data support previous observations by Berry et al. (2015) that point to high reproducibility of single-cell measurements as we observe a typical analytical uncertainty of less than 1 CD% (Supplementary Text, Supplementary Fig. S11). We apply the larger error term for a more conservative estimation of growth rate measurement range in the proceeding section.

Applying Raman microspectroscopy to SIP growth rate measurements

Having generated a Raman-based hydrogen isotopic calibration of microbial biomass, we sought to test the applicability of this analytical method to measurements of cell-specific growth rate. Microbial growth rate can be inferred from SIP incubations because the degree of an organism’s biomass isotopic enrichment over time depends on the isotopic composition of a given substrate and its rate of biomass synthesis (Kopf et al. 2016). Therefore, if measurements of ²H incorporation are reliable, then these values should translate to cell-specific growth rates. To this end, we sought to parameterize sources of uncertainty and propagation to downstream calculations to define acceptable experimental conditions and analytical uncertainties for quantitative measurement of microbial growth rate.

The growth rate of an organism in the presence of a ²H₂O is calculated with Eq. (1) (Kopf et al. 2016). In brief, microbial growth rate is a natural logarithmic relationship between an organism’s isotopic composition at the start of an incubation (F₀) and its isotopic composition at the time (t) of sampling (F_T). Cellular isotopic enrichment is compared to the enrichment of the isotopic label (F_L) offset by a water hydrogen assimilation efficiency constant (a_w)—a value representing the fraction of biomass hydrogen that a cell acquires from its growth water, as opposed to other metabolic sources and associated fractionation effects.

(1)

We modeled growth rates (µ) (Eq. 1) across a wide range of SIP experimental conditions (varying F_L, t, and a_w,) and applied standard error propagation to determine the associated uncertainty (σ_µ) (Eq. 2) that result from compounding uncertainty of Raman–SIP experimental parameters (Kopf et al. 2016, Caro et al. 2023):

(2)

where sigma (σ) terms represent uncertainties of each subscripted parameter (e.g. σ_FT represents uncertainty in biomass isotopic composition at time of sampling T). We convert growth rate to biomass generation time using the relationship:

(3)

where µ is growth rate in units of t⁻¹ and T_G is generation time in units of t.

To a first order, the range of growth rates (µ) that can be measured with SIP depends on the isotopic composition of the label and the incubation duration. Uncertainty in growth rate (σ_µ) depends on uncertainty in label strength (σ_FL), biomass isotopic enrichment (σ_FT), and microbial water hydrogen assimilation efficiency (σ_aw). Our model constrains the region of quantification for both isotopic enrichment (Fig. 3C) and corresponding growth rate (Fig. 3D and E) in Raman SIP experiments. We propagate the uncertainty estimates of ²F_biomass estimated from our hydrogen isotope calibration as σ_FT to estimate associated uncertainty in growth rate (σ_µ) that results from a given SIP experiment. We report uncertainty in growth rate relative to the growth rate itself as Relative Error = (σ_µ/µ) × 100%. We set our limit for relative error to be 50% such that errors below this cutoff result in growth rate differences that are distinguishable with >2σ confidence.

In Fig. 3, we display example outputs from our model using incubation times of t = 5 days and t = 30 days across a range of ²H₂O label strengths. For example, the range of quantification for an experiment applying a 35% label is 6.9–32.6 at. % deuterium enrichment. Ranges of quantification and error optima for additional isotopic label strengths are reported in Table 2. For a 5-day incubation, this range of ²F quantification corresponds to growth rates between 0.53 and 0.04 d⁻¹, or generation times of 1.3 and 17 days. For a 30-day incubation, this range of ²F quantification corresponds to growth rates between 0.09 and 0.007 d⁻¹ or generation times between 7.7 and 101.9 days (Fig. 3E).

Asymptotic rises in uncertainty at high and low biomass isotopic enrichment, respectively, correspond to (i) uncertainty in growth rate as cell isotopic enrichment approaches that of the label solution and (ii) uncertainty resulting from biomass isotopic signal not significantly exceeding that of the noise inherent to the analytical method. The first condition (i) represents cells that grow faster than can be discriminated using the set incubation time and label strength, while (ii) represents cells that grow too slowly to be distinguished from noise. It should be noted that our ranges of quantification should be viewed conservative estimates, as the source of uncertainty we apply corresponds to heterogeneity of ²H-incorporation across cell populations, as opposed to the error across individual Raman acquisitions. This latter source of error is far smaller both in our dataset and in previously reported datasets (Berry et al. 2015) (Supplementary Fig. S11).

Incubation time (t) shifts the dynamic range of SIP (Supplementary Fig. S8): shorter incubations can resolve faster-growing organisms from each other but may not capture slower-growing organisms due to insufficient label incorporation. Conversely, longer incubations can capture slower-growing organisms, but faster-growing organisms may become saturated with the label and will no longer be resolvable from each other (Fig. 3). Growth rates of fast-growing organisms, those that are readily saturated with deuterium, could lead to underestimates of aggregate microbial turnover.

Increasing the strength of the label (F_L) expands the dynamic range of the SIP method i.e. both slower and faster growth rates can be captured as there is more isotopic “space” that can be measured (Table 2, Fig. 3). However, increasing F_L must be weighed against the risk that higher deuterium concentrations can impose toxicity or unintended physiological effects. Deuterated water at high levels of isotopic labeling have been shown to have no effect, inhibition and even stimulation among different microorganisms (Lester et al. 1960, Berry et al. 2015, Kopf et al. 2015). In our experiments, we observed depressed isotopic enrichment in growth water containing 50 at. % ²H₂O. These effects were noted for both organisms and both the Raman and nanoSIMS data (gray point intervals in Fig. 2). We suspect that this result stems from isotopic toxicity affecting microbial growth or metabolic switching at high isotopic enrichment. Deuterium toxicity may result from decreased reaction rate of deuterium relative to protium and by physiological stress induced by altered solvent characteristics. Recent molecular dynamics analyses suggest that ²H₂O as a solvent can have profound impacts on lipid membrane packing and protein organization (Tempra et al. 2023). Thus, it is conceivable that increased protein and membrane rigidity at high ²H₂O concentrations inhibits growth, induces metabolic switching (i.e. acquiring more H from nonwater sources) or stress responses. We emphasize that the exact mechanism of D₂O toxicity cannot be explained by our study and that more targeted studies of deuterium toxicity mechanisms are warranted. We caution against applying ²H₂O label strengths at or above 50 at. % due to varying physiological effects that may arise at this threshold, especially when applying this approach to natural microbial communities of unknown and mixed composition. In addition, we encourage future researchers applying ²H₂O–SIP at high concentrations to include an unlabeled control to monitor changes in physiology, metabolite production, transcription, and so on. Uncertainties in isotopic label strength (σ_FL) due to pipetting or gravimetric dilution contribute relatively minor components to total uncertainty, especially at high isotopic enrichment. Unintended evaporation of a water-based tracer may lead to increased uncertainty in label strength via the different evaporation rates of deuterated and nondeuterated water. Evaporative effects would contribute to a gradual isotopic enrichment of the growth medium in a manner predicted by a Rayleigh distillation model. Thus, though isotopic distillation can be numerically modeled, conducting water-based SIP in a closed system is preferable for many experimental designs.

The water hydrogen assimilation efficiency constant (a_w) is essential for estimation of microbial growth rate, as it sets the upper limit of biomass deuterium enrichment. This limit is represented by the terms in Eq. (1) (Fig 3A). Therefore, similar to increasing the strength of the label, an organism with a greater a_w has more isotopic “space” in which to become enriched. In our study, we use Raman-derived ²F measurements to empirically determine the a_w of T. hydrogeniphilus and M. NSHQ04 to be a_w = 0.76 ± 0.02 and a_w = 0.88 ± 0.02, respectively (Table 1C, Fig. 2). The higher a_w observed with the methanogen M. NSHQ04 is in line with primarily autotrophic growth (Miller et al. 2018) with an offset from a_w = 1 due to hydrogen isotopic fractionation. The slightly lower a_w for the sulfate reducing T. hydrogeniphilus is indicative of a hydrogen contribution from its acetotrophic metabolism. This result supports prior observations by Berry et al. (2015) who noted substantial differences in ²H incorporation between heterotrophs and autotrophs arising from differences in hydrogen sources between organisms and metabolisms. We note that Raman-derived values represent a_w of whole-cell biomass, which may be distinct from those calculated via compound-specific analysis (Zhang et al. 2009, Wijker et al. 2019, Caro et al. 2023). It was previously reported that uncertainty in a_w is a significant driver of uncertainty in hydrogen SIP-derived growth rate estimates (Caro et al. 2023). In studies of mixed communities, this uncertainty must be propagated through growth rate calculations. For organisms with known metabolisms, the a_w term can be estimated as common metabolic modes often constrain a_w factor to a degree (Zhang et al. 2009, Wijker et al. 2019, Caro et al. 2023). In an ideal scenario, this uncertainty would be constrained or minimized by prior knowledge of dominant metabolisms present in an environmental sample (e.g. via marker gene or metagenomic sequencing).

Raman and nanoSIMS measure different pools of cellular hydrogen

Deuterium abundances measured by Raman were greater than those measured by nanoSIMS when compared to the culture growth water (Fig. 2). When we correlated cell-specific ²F_biomass measurements we observed that, cell-to-cell, nanoSIMS-derived hydrogen isotope values were severely depressed compared to corresponding Raman-derived values (Supplemental Fig. S2). A depression in expected isotopic content relative to an expected value can be expressed as a dilution factor, where the decrease in isotope abundance is expressed as a percentage. From this correlative dataset, we calculated an average approximate hydrogen isotope dilution factor of 58.7 ± 6.0% across all growth water labeling conditions (Supplementary Text).

Depression of isotopic enrichment measured by nanoSIMS due to sample preparation has been widely reported for multiple stable isotope tracers (¹³C, ¹⁸O, ¹⁵ N, ²H, and ³⁴S) under various preparation conditions, as stains, fixatives, and nucleic acid probes can overprint cellular isotopic signals (Musat et al. 2014, Kopf et al. 2015, Pernice et al. 2015, Woebken et al. 2015, Stryhanyuk et al. 2018, Meyer et al. 2021). Our results point to a separate phenomenon: the replacement of exchangeable hydrogen during sample washing. Such a significant dilution of deuterium abundance speaks to a fundamental difference in the nature of nanoSIMS- and Raman-based measurements. Owing to how the primary ion beam ionizes and ablates a cell, nanoSIMS is mostly agnostic to the sources of ²H/¹H it measures, setting aside potential differences in hydrogen ionization efficiency across different classes of molecules. Therefore, ¹H⁻ and ²H⁻ ions reaching the instrument detectors can derive from a variety of cellular hydrogen sources including relict cytosolic water, adsorbed extracellular water, and biomass hydrogen in both nonexchangeable sites and exchangeable sites that are readily mixed with natural-abundance washing buffers during sample preparation. Protonated sites in biomolecules (e.g. O–H, N–H, S–H, and so on) experience rapid exchange with aqueous wash solutions on the order of picoseconds to minutes (Bai et al. 1993, Englander et al. 1996), meaning that anabolically produced O–²H, N–²H, S–²H, and so on bonds would be rapidly overprinted by natural-abundance hydrogen during washing procedures (Fig. 4). As described in the section “Materials and methods,” no stains or probes were applied to the cells in this study. Therefore, the hydrogen dilution factors reported here result primarily from sample washing. This significant and persistent isotopic dilution may confound efforts to use deuterium tracers as measures of biosynthesis in nanoSIMS-based studies (Kopf et al. 2015). This problem is compounded at lower-isotopic enrichment, where substantial variability in ²H/¹H values results from low ²H ion counts as well as variation in which cell subcomponents (which may display distinct ²H enrichments) are ablated (Kopf et al. 2015). Preparatory methods that forego or reduce washing may suffer from an opposite effect: protic sites in microbial biomass, still in isotopic equilibrium with the ²H-enriched growth medium, could contribute to a biomass ²H/¹H in excess of what was generated through anabolic processes. Future work should apply caution when applying dilution factors to back-calculate isotopic abundance, as these can vary widely between the sample preparation and organism (Meyer et al. 2021), an effect that may be pronounced in environmental systems with mixed microbial communities.

A key advantage of Raman–SIP is that the C–H (2800–3100 cm⁻¹) and C–D (2040–2300 cm⁻¹) bands report solely nonexchangeable carbon–hydrogen and carbon–deuterium stretching modes associated with mostly lipids, proteins, and nucleic acids (Berry et al. 2015, Cui et al. 2022). These spectral regions are oblivious to exchangeable bonding environments including O–H, N–H, S–H, and so on, as well as relict water. Raman microspectroscopy, therefore focuses on anabolically produced organic bonds without confounding hydrogen sources. It is important to note that the application of organic nucleic acid probes, such as stains, FISH, or CARD-FISH, as well as organic fixatives such as formaldehyde, may interfere with Raman-derived measurements via the introduction of nonbiomass C–H bonds (Berry et al. 2015) (Supplementary Text, Supplementary Fig. S9). Therefore, despite the utility linking microbial identity to function, the application of nucleic acid probes to both Raman–SIP and nanoSIMS–SIP studies may be unsuitable for optimum quantification of microbial growth. However, because of its nondestructive nature, Raman can be readily applied as a mid-point, rather than end-point analysis, and so the application of FISH or stains after Raman measurement is plausible.

Future directions and limitations for Raman–SIP

Because Raman microspectroscopy is a standard technique for mineralogical/material characterization, it brings with it several aspects that complement SIP practices. First, it is a nondestructive, rapid method. Like nanoSIMS, Raman is readily coregistered with other techniques such as fluorescence microscopy, FISH, scanning electron microscopy, or cell sorting (Berry et al. 2015, Schaible et al. 2022). A single Raman spectrum takes between seconds to 1 min to acquire, and many Raman instruments are equipped with epifluorescence modules that can enable simultaneous visualization of cells in addition to transmitted or reflected light microscopy. Second, Raman is comparatively accessible relative to nanoSIMS and allows analysis with minimal preparation, foregoing the need for conductive surfaces and/or metallic sputter coating, vacuum chamber, and secondary electron optics. Raman instruments like those employed in this study are similar in operation to epifluorescence microscopes, which lowers user barrier-to-entry and improves the capacity for direct deployment on various forms of environmental samples including filters, rock faces, plant material, and so on. Third, Raman microspectroscopy has microscale spatial resolution and provides diagnostic information about cellular organic composition. The fingerprint region (200–1800 cm⁻¹) can identify specific biomolecular composition (Cui et al. 2022), which can be useful for differentiating taxa and growth stage (Supplementary Fig. S1) (Mosier-Boss 2017, Novelli-Rousseau et al. 2018, Wang et al. 2020). Fourth, the Raman fingerprint region enables identification of biologically precipitated minerals and the mineralogical/material context of a cell (Klein et al. 2015, Suzuki et al. 2020). Raman SIP with deuterium, specifically, makes identification of individual cell and mineral components in organic–mineral assemblages feasible because the C–D band occupies a typically silent region in inorganic Raman spectra. Coregistered mineral and cell-activity measurements can support investigations into the cell–mineral and cell–cell relationships that drive microbial activity (Templeton and Caro 2023).

Despite the promise of Raman SIP, this method has key caveats that must be considered and/or addressed. The major disadvantage of ²H–Raman–SIP is sensitivity. Classical or spontaneous Raman microspectroscopy of microbial cells is limited by low signal intensities, particularly in samples with high autofluorescence (Hatzenpichler et al. 2020). A key outcome of our study is the definition of Raman–SIP ranges of quantification for growth rate estimation. Our results conservatively define the minimum cutoff for quantitative growth rate measurement as ∼7 at. % (Table 2), depending on the label strength applied. Microbial ²H incorporation that does not exceed this threshold cannot be quantitatively distinguished from noise. Similarly, microbial ²H incorporation in excess of the upper range of quantification cannot yield quantitative growth rates. We emphasize that our estimates of uncertainty (and by extension, growth rate quantification ranges) should serve as guidelines and we encourage future researchers to estimate these parameters for their own instruments and experiments. Further computational tool development and optimization may further improve Raman’s detection limit and growth rate quantification capabilities (Schaible et al. 2024).

While we have demonstrated severe dilution of hydrogen isotopic measurements by nanoSIMS, this methodology remains well-suited to sensitively measure isotopes of C, N, O, S, and so on at the single-cell level. While measurement of ¹³C and ¹⁵ N has been demonstrated with Raman, the precision of these measurements is far below that of nanoSIMS and is traceable only within specific biomolecules (Cui et al. 2017, Weber et al. 2021). In the coming years, advancements in Raman technology and data processing (Schaible et al. 2024) could conceivably increase sensitivity and subsequently improve isotopic measurements. Specifically, surface- or tip-enhanced Raman spectroscopy (Efrima and Zeiri 2009, Mosier-Boss et al. 2016, Chisanga et al. 2017, 2018, Mosier-Boss 2017), stimulated Raman scattering, resonance Raman spectroscopy, or coherent anti-Stokes Raman spectroscopy, which are shown to improve acquisition times and enhance signal, could be evaluated for isotopic measurements (Ivleva et al. 2010, 2017, Camp and Cicerone 2015, Kubryk et al. 2015, Cicerone 2016, Cui et al. 2017, Chisanga et al. 2018, Hu et al. 2019, Weiss et al. 2019). However, at present, validation of these methods for SIP is required (Supplementary Text, Supplementary Figs S4 andS5). As noted earlier, Raman–SIP may suffer dilution effects related to staining or the use of nucleic acid probes (FISH), similar to nanoSIMS. Similarly, fixatives such as formaldehyde, in stabilizing the cell membrane, introduce C–H bonds that can affect isotopic measurements (Supplementary Text, Supplementary Fig. S9).

Compared to the surveys of microbial identity and functional potential, relatively few measurements of cellular growth rates in nature exist (Koch et al. 2018, Caro et al. 2023, Templeton and Caro 2023, Foley et al. 2024). Growth rates of microorganisms in complex systems can inform theories of microbial ecology, as well as elucidate preferences of microbial taxa for specific environmental niches, substrates, and microscale locality. In the rock-hosted biosphere, for example, measurements of microbial anabolic activity can clarify how microbial activity relates to mineralogical context, or how the activities of microbial partners/consortia are correlated. While we present a calibration of environmentally relevant taxa, further validation of Raman–SIP and correlative measurements of mineralogical/material composition is required in complex environmental samples.

For the benefit of future researchers, we designed an interactive GUI, Shiny R-SIP, that allows users to implement our SIP model and optimize their experiments by considering how the relevant sources of uncertainty impact the design of a Raman–SIP experiment (Supplementary Data or online: https://apps.kopflab.org/login with login credentials username: shiny-rSIP, password: public-access). The GUI allows users to adjust parameters of label strength, incubation time, and water–hydrogen assimilation efficiency, as well associated uncertainties in these terms, in order to design effective SIP experiments. The parameters we define in our microbial growth model are specific to the study organisms used and our Raman instrument—they are intended to provide reasonable estimates of uncertainty that may be encountered. Users of this GUI are encouraged to define their own uncertainty terms by constraining the technical variation in individual CD% measurements inherent to their Raman instrument. For samples where microbial growth may exhibit large variation, we suggest users consider multiple incubation times (or subsampling efforts) to effectively capture a wide range of biomass growth rates in experiments targeting natural microbial communities, while also quantifying their threshold of tracer saturation as an upper-limit of growth-rate quantification.