Real-time, high-resolution metabolic characterization of live bacteria using label-free optical metabolic imaging

Abstract

Label-free optical imaging provides non-invasive, high-speed, high-resolution metabolic characterization of live bacteria with single-cell resolution. Here, we demonstrate the ability of label-free multiphoton autofluorescence microscopy to characterize the fast (between 0 and 30 min) metabolic changes in bacteria in response to antibiotic treatments and observe the cell-to-cell metabolic heterogeneity of planktonic bacteria and biofilms. Results indicate that bacteria exhibit a distinct measurable response to bactericidal treatments within seconds. Furthermore, S. aureus biofilms exhibit metabolic heterogeneity, with local pockets of high metabolic activity. Bacteria in biofilms exhibit altered metabolic profiles compared to planktonic bacteria for all four species examined: S. aureus, P. aeruginosa, M. catarrhalis, and S. pneumoniae. These results shed light on the spatial and temporal metabolic heterogeneity of bacteria and the quantification possibilities using label-free nonlinear optical microscopy.

Introduction

Bacteria play a critical role in both health and disease. This includes the role of bacteria in homeostasis of the human microbiome, infectious diseases, the environment, food safety, and energy production. Microbiologists often employ established -omics methods to characterize bacteria. However, intercellular heterogeneity and temporal dynamics of bacteria are often inaccessible using established methods, commonly performed on populations of hundreds to millions of bacteria, extracting parameters on the average biochemical profile but lacking single-cell resolution. Additionally, most bacteria characterization methods cannot be performed in situ, and are unable to characterize live bacteria in diverse environments, such as in biofilms.

Bacteria metabolism is an important area for investigation and characterization. Both bacterial and mammalian cells use ATP as the primary source of chemical energy, and thus their energy metabolism pathways are focused on phosphorylation of ADP to ATP. There is a wide variety of bacteria species and a large diversity of bacterial metabolism; primarily, bacteria metabolism can be classified as: heterotrophic, autotrophic, and phototrophic¹; this paper focuses on heterotrophic metabolism, which generally relies on oxidation of organic compounds in order to phosphorylate ATP. A variety of intracellular metabolic pathways exist, including those related to glycolysis, fermentation, the citric acid cycle, and the electron transport chain, where most ATP is produced. In the electron transport chain, NADH is oxidized to NAD⁺ and FADH₂ is oxidized to FAD, pushing H⁺ ions across the cell membrane. One of the reasons why characterization of bacterial metabolism is important is that energy metabolism plays an essential role in the cellular stress and/or antibiotic response². For example, electron transport chain activity in E. coli was shown to decrease during the stress response³. Not only does cellular stress change cellular metabolism, but the current metabolic state of a cell at the time of the stress also impacts the response⁴.

Even within a colony of bacteria grown under constant conditions, cells exhibit heterogeneity. This natural heterogeneity is a key feature that spurs antibiotic resistance, a public health issue growing at an alarming rate⁵. In 2019, it was estimated that antimicrobial resistance was directly responsible for 1.27 million deaths and contributed to 4.95 million deaths⁶ worldwide, with the World Health Organization noting antimicrobial resistance as a top global threat⁷. Genetically resistant cells have a gene or genes that support the inhibition or evasion of an antibiotic. However, recent work has identified “persister” cells, which are not genetically resistant to antibiotics, but instead are functionally resistant. Persister cells can shut off their metabolism temporarily until conditions for growth become favorable, then return to a normal growth state and replicate⁸. To identify genetically-identical but phenotypically-distinct populations such as persister cells, functional and phenotypic bacteria characterization methods are needed, in addition to established genotypic methods. Thus, the ability to examine phenotypic heterogeneity in populations of bacteria provides essential information on viability and vitality and is pertinent to a variety of areas of intense microbiological research, such as antibiotic susceptibility.

Various studies have examined aspects of the spatial and temporal heterogeneity of bacteria. For example, the Raman signatures of two species of mycobacteria were shown to change significantly on the scale of 12–144 h as bacteria went through the four growth phases (lag, log, stationary, and death), primarily showing increased lipid signatures over time⁹. Additionally, one study found that a variety of sites within an individual including: gut, oral cavity, external auditory canal, nostrils, hair, and skin, that the external habitats (such as skin) exhibited the highest levels of intrapersonal variability in genetic composition over time, measured with 16S rRNA gene sequencing¹⁰. Furthermore, certain high-diversity skin locations showed significantly more phylogenetic diversity than the gut or oral cavity¹⁰. These results demonstrate the large scale spatial (~cm–m) and temporal (~days–months) genetic diversity of bacteria. However, little work exists examining the smaller-scale spatial (μm–mm) and temporal (s–min) heterogeneity of bacteria. Likely, this is due to the lack of available tools for bacteria characterization on those scales.

Bacteria heterogeneity is especially relevant in biofilms, and contributes to their capabilities for antibiotic resistance. Biofilms are (mono- or poly-species) communities of bacteria, containing diverse cells and extracellular components^11,12. Their physiology is distinct from planktonic bacteria; the structure and organization are species- and environment-dependent, and they are especially well-suited for antibiotic resistance and persistence due to their cellular heterogeneity, structure, and horizonal gene transfer¹². Biofilms have a large clinical impact due to their ability to evade antibiotics, often leading to chronic infections¹³. Bacteria in biofilms express different proteins to facilitate adhesion, and secrete additional extracellular components such as polysaccharides, proteins, and extracellular DNA^12,13. Biofilm heterogeneity encompasses many aspects, including metabolic heterogeneity, which has been observed in P. aeruginosa biofilms¹⁴. Quorum sensing is the ability of bacteria to sense the presence and concentration of other bacteria, and respond by adjusting their function (secretions, metabolism, virulence, etc.), thus enabling coordination among a large group of bacteria to form a functioning heterogenous biofilm¹³. Many species of bacteria exhibit metabolic flexibility in biofilms, and are able to adapt their metabolism to a specific environment to further promote virulence¹⁵. Due to the importance of metabolism in biofilm virulence, and the expected metabolic heterogeneity of biofilms, spatially mapping biofilm metabolism is of great interest for both improving our understanding of biofilm biology and for future work in clinical applications.

Thus, establishing new methods for dynamic characterization of bacteria phenotypic heterogeneity in diverse environments is essential. Label-free (and cultivation-free) optical methods for bacteria characterization are becoming increasingly popular, due to their adaptability, spatial and temporal resolution, and non-invasive nature^16,17. Light-matter interactions are rapid, enabling excellent temporal resolution for optical characterization of bacteria, on the seconds to minutes scale¹⁶. In terms of live imaging, multiple modalities including optical coherence tomography¹⁸, optical diffraction tomography¹⁹, and Raman spectroscopy²⁰ have been used for real-time optical characterization of bacteria. This enables not only an examination of the spatial heterogeneity, but additionally the temporal dynamics. Also, of recent interest is label-free optical metabolic imaging²¹, which relies on multiple endogenous contrast mechanisms to provide metabolic and biochemical information about biological samples with high spatial (<500 nm) and temporal (1 Hz) resolution. Metabolic contrast is achieved through multiphoton autofluorescence intensity and lifetime imaging of reduced nicotinamide adenine dinucleotide and reduced nicotinamide adenine dinucleotide phosphate (together termed NAD(P)H) and flavin adenine dinucleotide (FAD), essential metabolic cofactors. While the reduced form of NAD(P)H is autofluorescent, the oxidized NAD⁺ is not; conversely, oxidized FAD is autofluorescent, while reduced FADH₂ is not. Accordingly, the relative NAD(P)H and FAD intensity in cells has been used to approximate the reduction-oxidation (redox) state of cells^22–25. Autofluorescence intensity depends on the concentration and brightness of the fluorophores. Fluorescence lifetime, defined as the ns-scale time difference between excitation and emission, provides contrast based on whether the fluorophores are bound to proteins^22,26,27.

Autofluorescence imaging of NAD(P)H and FAD is well-characterized, and thoroughly established to characterize cellular metabolism in mammalian cells^22,23,28. Notably, studies have used autofluorescence to classify cancer cells^21,22 and cancer treatment response^28–30, immune cells^31,32, neural cells and neural activity^33,34, and more. Prior studies using autofluorescence to characterize the metabolism of mammalian cells have been validated using metabolic assays such as alamarBlue³⁰, oxygen consumption rate and extracellular acidification rate³⁵, and LC/MS-MS³⁶. However, applications to study bacteria metabolism have been sparse, yet promising^37–39. For example, Perinbam et al. found that surface-attached bacteria (which are considered to be virulence-activated) exhibited a significantly decreased NAD(P)H fluorescence lifetime than planktonic bacteria (low-virulence) in P. aeruginosa³⁷. Also, excitation-emission fluorescence spectroscopy with multivariate analysis was shown to be able to differentiate antibiotic resistant and susceptible populations of E. coli and K. pneumoniae⁴⁰. Previous work using fluorescence lifetime imaging microscopy (FLIM) of NAD(P)H showed some separation of six different species of bacteria on the phasor plot³⁸. The dose response of E. coli to ampicillin and nalidixic acid showed increased free NAD(P)H at higher concentrations of antibiotic, indicating a shift towards glycolytic metabolism. Recently, multiphoton FLIM of NAD(P)H showed that the NAD(P)H lifetime of E. coli has a dose- and mechanism-dependent response to antibiotics. Under high hydrostatic pressure, S. cerevisiae and E. coli both shifted towards more bound NAD(P)H on the phasor plot⁴¹. While this initial work shows the potential for FLIM of NAD(P)H for characterizing and differentiating bacteria, it also raised many new questions about the capabilities of multiphoton autofluorescence FLIM for characterizing bacteria³⁸.

Here, we explore the use of fast optical metabolic imaging with multiphoton autofluorescence intensity and lifetime of NAD(P)H and FAD to observe the temporal dynamics of bacteria in vitro in response to antibiotic treatment and the spatial heterogeneity of biofilms. Of particular interest is S. aureus, which is often present on human skin and is a major cause of infection, with growing concern due to resistant strains⁴². Label-free optical metabolic imaging presents a strong platform to characterize bacteria and has potential to address many needs of the field by enabling the ability to rapidly characterize single bacteria.

Results

Nonlinear optical microscopy can be used to observe fast metabolic dynamics

To examine fast dynamics, dual-channel multiphoton FLIM (0.231 fps) of NAD(P)H and FAD was continuously acquired on S. aureus immediately before 10% bleach addition and for 30 min after treatment to capture the immediate dynamics during treatment and enable tracking of single-cell responses (Fig. 1). S. aureus showed a dramatic decrease in NAD(P)H intensity, increase in FAD intensity, and decrease in NAD(P)H and FAD fluorescence lifetimes throughout the treatment.

Fig. 1. Fast dynamics of S. aureus in response to 10% bleach treatment.

In general, NAD(P)H intensity (I_N) fell, FAD intensity (I_F) increased, and NAD(P)H lifetime (τ_N) and FAD lifetime (τ_N) both decreased. Visualization of four small clusters of cells are shown over time (a−d), along with their locations within a larger field of view (e). For a–d, each sub-row represents a different modality: top row – NAD(P)H intensity, second row – FAD intensity, third row – NAD(P)H lifetime, bottom row – FAD lifetime, corresponding to the color bars in (e).

Single-cell segmentation enables heterogeneity analysis

The ability to segment single bacteria is important for enabling heterogeneity analysis. To examine capabilities for enumeration using label-free nonlinear optical microscopy, planktonic S. aureus in broth were imaged at different concentrations using dual-channel multiphoton FLIM of NAD(P)H and FAD (Fig. S1). Ground-truth CFU/mL numbers were determined using colony-counting of diluted bacteria after 24 h of growth on an agar plate, with accurate enumeration up to 6 × 10⁸ CFU/mL.

To examine cell-to-cell heterogeneity, two single colonies of S. aureus were separately taken from an agar plate and grown in Tryptic Soy Borth (TSB) overnight prior to imaging. Autofluorescence metrics were calculated for each cell: NAD(P)H intensity, NAD(P)H lifetime, FAD intensity, and FAD lifetime (Fig. 2). Using a two-sample t-test, S. aureus from the two colonies showed significant differences (p < 0.05) for FAD intensity, but not for NAD(P)H intensity, NAD(P)H lifetime, or FAD lifetime. In addition to inter-colony heterogeneity, the intra-colony single bacteria heterogeneity is also visible due to the nonzero widths of the histograms in Fig. 2, indicating that within a single colony suspension, single bacteria exist in different metabolic states.

Fig. 2. Quantification of S. aureus optical metabolic heterogeneity across single-colony suspensions.

Histograms of number of segmented cells based on: a NAD(P)H intensity, b FAD intensity, c NAD(P)H lifetime, and d FAD lifetime. P values indicate significant differences between colonies A and B with a two-sample t-test; number of cells (N) segmented in each group indicated in the legend.

Optical metabolic response to antibiotics depends on dose and mechanism

To track dynamics, single S. aureus cells were segmented and then tracked over time using the FAD channel, which was consistently bright (Fig. S2). Figure 3 shows the collection of single-cell responses in NAD(P)H and FAD intensity and lifetimes over 30 min for four different treatments (H₂O/control, 1% bleach, 35 mM clindamycin, 20× penicillin/streptomycin). Each treatment perturbs the cell in a slightly different way: clindamycin disrupts protein synthesis, penicillin disrupts cell wall synthesis, streptomycin disrupts protein synthesis, and bleach disrupts the redox balance and denatures proteins. The treatment was added into the imaging dish using a pipette approximately 10 frames after the start of imaging. Treatment with 1% bleach induced a much larger magnitude increase in the FAD channel, displayed with a different intensity scale for Fig. 3b, compared to the other subfigures. The dose dependence for clindamycin with additional lower concentrations is shown in Fig. S3, with the initial dip in NAD(P)H and FAD intensities vanishing as the dose decreases, indicating that this is a signature related to the metabolic stress of the added antibiotics. The mean values of phasor components s and g of all segmented cells are shown over time in Fig. S4, which shows the trajectory of NAD(P)H and FAD lifetimes over 30 min.

Fig. 3. Single-bacterium traces over time.

NAD(P)H (blue) and FAD (yellow) intensity and lifetime for single-bacteria (thin lines) and means (thick lines) following treatment with a H₂O, b 1% bleach, c 35 mM clindamycin, and d 20× penicillin-streptomycin. Note the unique intensity scale for 1% bleach FAD intensity.

Bacteria in biofilms exhibit an altered and spatially heterogenous optical metabolic profile

To examine intra-biofilm heterogeneity, a large mosaic of a S. aureus biofilm is shown in Fig. 4. Regions of highly autofluorescent cells are visible, varying in size from ~1 μm above 20 μm in diameter. These bright autofluorescence regions also display bright signal from CH-region (2700–3200 cm⁻¹) coherent anti-Stokes Raman scattering (CARS) signal (Fig. S4), indicating a large and heterogenous presence of lipids and proteins.

Fig. 4. S. aureus biofilm spatial metabolic heterogeneity.

A 5 × 5 frame mosaic of four different channels: a NAD(P)H intensity, b FAD intensity, c NAD(P)H lifetime, and d FAD lifetime; and zoomed-in single frames of each channel: e NAD(P)H intensity, f FAD intensity, g NAD(P)H lifetime, and h FAD lifetime.

One metric commonly used to characterize metabolism is the optical redox ratio (ORR), defined here as the FAD fluorescence intensity over the sum of FAD and NAD(P)H fluorescence intensity. This approximates the redox state of the cell by examining the relative portion of FAD, which is oxidized, relative to reduced NAD(P)H. Pixelwise distributions of the ORR of both planktonic bacteria and biofilms are shown in Fig. 5 for four different species of bacteria: S. aureus, P. aeruginosa, M. catarrhalis, and S. pneumoniae. P. aeruginosa and M. catarrhalis (both gram-negative) show very low NAD(P)H intensity in biofilm form (and thus a high ORR), indicating a much more oxidized cell redox state. This shift toward a more oxidized state is additionally visible in S. aureus and S. pneumonia, but is much less dramatic in these two gram-positive species. Biofilms also show a shift in CARS signature towards lower wavenumbers (relatively more protein and lipid content), likely due to increased extracellular components (Fig. S5). Segmentation of single bacteria was not possible in biofilm images due to their high density and heterogeneity.

Fig. 5. Optical metabolic profile of planktonic bacteria and biofilms for: S. aureus (blue), P. aeruginosa (green), M. catarrhalis (orange), and S. pneumoniae (purple).

Violin plots contain one point per pixel, originating from 25 original fields of view each.

Discussion

One of the key benefits of nonlinear optical microscopy is its high resolution, which enables single-cell segmentation when aggregation is not occurring. Most bacteria characterization methods are performed in bulk, and lack the ability to observe and quantify this heterogeneity. Here, we show that single cells and clusters of cells can be visualized and segmented (Fig. 1, Figs. S1 and S2). Single S. aureus could be segmented at concentrations up to 6 × 10⁸ CFU/mL using the FAD channel for segmentation (Fig. S1). In previous work, label-free single-photon autofluorescence intensity and lifetime of S. aureus, K. pneumoniae and E. coli bacteria could be detected at concentrations as low as 2 × 10² CFU/mL using confocal microscopy⁴³. Locke et al. estimated that the limit of detection for optical methods ranges from 1 (fluorescence imaging with flow cytometry) to 10⁷ (elastic light scattering) CFU/mL¹⁶.

Segmentation enabled analysis of the heterogeneity of single colony suspensions (Fig. 2) and the tracking of the metabolism of single cells over time during treatments (Fig. S2, Fig. 3). Single colony suspensions showed a heterogeneity in intensity-based metrics of single cells, indicating a variety of metabolic states (Fig. 2). NAD(P)H and FAD lifetime were not significantly different between colonies, suggesting that lifetime is a more robust metric for comparing different suspensions of the same species. The significant difference in FAD intensity between colonies may stem from differences in concentration, or alternatively it could be from a slight metabolic difference between the two different colonies.

Characterizing the viability and vitality of bacteria provides essential information in a variety of scenarios, including: antibiotic susceptibility testing, examining host-microbe interactions in vivo, and quantifying biofilm growth dynamics. Previous work has examined the optical metabolic signatures of cell death in mammalian cells, observing apoptosis-related increases in NAD(P)H fluorescence lifetime and intensity^30,44–46 and changes in lipid-droplet dynamics⁴⁷. One study examined the NAD(P)H fluorescence lifetime of different species of bacteria 10–30 min after addition of antibiotics^38,39. Previous work showed that E. coli exhibited a decrease in NAD(P)H fluorescence lifetime after 30 min of treatment with either a bacteriostatic (nalidixic acid) or a bactericidal (ampicillin) antibiotic agent³⁸. Recent work examined that the effects of antibiotics on E. coli could be detected as soon as 10 min after treatment³⁹. However, there is a lack of data examining the shorter (<10 min) timescale of this process. Here, we examine the NAD(P)H and FAD-related redox dynamics of S. aureus in response to three different treatments (Fig. 3), and with different concentrations of clindamycin that range from bacteriostatic to bactericidal (Fig. S3). Antibiotic treatment with clindamycin and penicillin-streptomycin induced an immediate drop in NAD(P)H and FAD intensity, likely a halting of cellular metabolism due to the antibiotic presence, followed by a recovery on the ~60 s timescale. As shown in Figs. 1 and 3, bleach induces an immediate and dramatic shift in FAD intensity, and starts to show a shift towards lower NAD(P)H and FAD fluorescence lifetimes around 15 min (900 s). The initial increase in FAD intensity is likely indicative of increased intracellular oxidation of FADH₂ to FAD, which rapidly occurs since bleach is a strong oxidizing agent. Previous work showed E. coli treated with nalidixic acid (bacteriostatic) and ampicillin (bactericidal) showed a drop in NAD(P)H lifetime and in increase in phasor component g values³⁸, consistent with our bleach treatment of S. aureus. Additionally, a previous study using quantitative phase imaging found that the optical signature of E. coli changed within 19–27 min after ampicillin exposure when cells began to rupture⁴⁸. This may suggest that this shift in lifetime coincides with cells rupturing, exposing intracellular NAD(P)H and FAD to the broth solution which could change the fluorescence lifetime. FAD has also shown pH-related fluorescence lifetime decrease⁴⁹. Thus, the decrease in NAD(P)H lifetime is likely a combination of cell-death related changes and pH-related response. These results indicate that nonlinear optical microscopy can be used to quantify fast metabolic changes in bacteria with single-cell resolution, and that the optical metabolic response of bacteria is sensitive to the dose and action mechanism of applied antibiotics.

Bacteria in biofilms exhibit an altered optical metabolic profile compared to planktonic bacteria of the same species. Figure 5 suggests a redox shift to a more oxidized state occurs in biofilms, especially in gram-positive bacteria. Each of the four species shows a slightly different planktonic and biofilm profile in Fig. 5; while not the focus of this work, previous work has shown that planktonic bacteria can be differentiated by species using NAD(P)H FLIM³⁸. Additionally, previous work has shown that surface-attached P. aeruginosa has decreased bound NAD(P)H and decreased concentration/production of NAD(P)H³⁷ and used carbon tracing to determine that P. aeruginosa biofilms have significantly slower metabolism than planktonic bacteria⁵⁰, which supports the presented results which show a relative decrease in NAD(P)H compared to FAD in the biofilms. Furthermore, NADH oxidase related metabolic pathways play a role in biofilm formation⁵¹. Nonlinear optical microscopy thus provides an avenue for examining biofilm spatial heterogeneity and temporal dynamics on previously unexplored scales, unlocking the ability for future work observing biofilm metabolism under different treatments and conditions. Furthermore, the non-invasive, label-free nature of nonlinear optical microscopy enables the examination of native, undisturbed biofilms for real-time functional characterization, unlike other methods that require mechanical disruption of the biofilm to characterize it.

The ability to image and characterize viable but non-culturable (VBNC) bacteria is especially elusive. These VBNC bacteria exist only in specialized conditions, and methods to cultivate/propagate them in laboratory settings have not yet been established. Thus, they cannot to be studied using the current characterization gold standard methods such as colony counting, which requires bacterial colonies to grow on agar plates. Furthermore, methods like colony counting require 1–5 days of culturing in dishes, which is a significant time limitation in the research process¹⁷. Similar to the benefits with biofilm characterization, cultivation-free methods such as label-free optical imaging could be an enabling technology, providing the ability to characterize VNBC bacteria.

Optical metabolic imaging of metabolic cofactors NAD(P)H and FAD inherently reports on cellular metabolism, and thus are considered phenotypic characterization. NAD(P)H- and FAD-related metabolism has previously been reported to give important insight into bacteria. For example, it was previously found using HPLC-MS/MS that NAD was the most important metabolite detected for differentiation of methicillin-resistant S. aureus (MRSA) and methicillin-sensitive S. aureus (MSSA)⁵². Phenotypic analysis cannot replace genotypic analysis, but provides complementary information. For example, it could be possible to identify genetically-identically but functionally different cells within a biofilm, or persister cells that evade antibiotics. However, for definitive species identification, genetic markers will likely remain the gold standard. Future work could examine the ability to spatially co-register genetic markers with nonlinear optical imaging, although many genetic examinations require the amplification of specific nucleic acid sequences, which is not feasible for high-resolution, live imaging. Furthermore, single-cell RNA sequencing methods can be used to build functional profiles of single bacteria cells, however these methods are destructive and cannot be used to monitor populations over time⁵³. Future work could correlate the transcriptomic profile of single cells with their optical metabolic profiles, and then examine changes in optical metabolic profiles over time with label-free imaging.

Translation of nonlinear optical microscopy to in vivo studies to assess the microbiome of various tissues is additionally possible since many studies have performed nonlinear optical microscopy of skin on live animals⁵⁴ and human subjects⁵⁵. This type of in vivo characterization is not possible for other biochemical techniques that require assays, imaging labels, or require other extensive sample preparations, and there is a need to be able to study the effect of host environmental factors on the microbiome⁵⁶.

Various other groups have performed multimodal optical characterizations, for example, multimodal imaging using modalities such as OCT and Raman spectroscopy have shown promise for characterizing bacteria and biofilms for application in otitis media¹⁶ and dental placque⁵⁷. Additionally, the combination of OCT and single-photon autofluorescence has been used for characterizing the oral environment, including associated bacteria⁵⁸. Compared to OCT, nonlinear optical microscopy is significantly more expensive, however it provides biochemical specificity and finer spatial resolution. As optical methods are further developed for application in microbiology, the choice of ideal method for a given task will depend on various factors, with each method bringing unique advantages and disadvantages.

In this manuscript, we demonstrate fast, label-free, non-destructive, multimodal analysis of bacteria using label-free optical metabolic imaging. This method provides multidimensional data on bacteria metabolism and their environment, enabling live characterization in diverse conditions. Single bacteria can be segmented and enumerated up to 6 × 10⁸ CFU/mL, enabling single-bacterium analysis under the right conditions. Optical metabolic imaging provides a phenotypic description of the bacteria that can be used to characterize different states and conditions, such as observing the dynamic response of bacteria to antibiotics. Future work is needed to untangle the exact metabolic pathways responsible for these responses. Compared to planktonic bacteria, bacteria in biofilms exhibited a much more oxidized metabolic state, with high spatial heterogeneity. These results indicate that quantitative, high-resolution, noninvasive metabolic analysis of bacteria is possible with label-free multiphoton microscopy of NAD(P)H and FAD.

The human microbiome is greatly heterogenous, between and within individuals, and has been linked to a variety of diseases including cancer, metabolic disease, inflammatory bowel disease, and depression^59,60. The development of non-invasive methods using label-free multimodal nonlinear optical microscopy to image and characterize the microbiome will open up new future opportunities to better characterize and understand the spatial and temporal complexities of microbiomes.

Methods

Bacteria growth

Bacteria was grown in TSB at 37˚C and 5% CO₂ in a stationary incubator. For examining single-colony bacteria, bacteria were streaked onto agar plates and left to incubate for 24 h. One single colony was then removed from the plate and resuspended in 2 mL of broth, then left to grow for another 24 h. The following species and st rains were used: S. aureus (Newman), P. aeruginosa (ATCC 14203), S. pneumoniae (ATCC 6301), and M. catarrhalis (ATCC 49143).

Biofilm growth

Once in log growth phase, 10 μL of single-colony planktonic bacteria in broth was transferred to a poly-d-lysine coated glass-bottom imaging dish, along with 990 μL of sterile TSB. Dishes were then incubated at 37 ˚C and 5% CO₂ in a stationary incubator. Every 24 h, the media was carefully aspirated and new media was added, for a total of five days of growth.

Antibiotic treatment

Antibiotic treatments and their efficacy on four bacteria species are listed in Table 1. The four treatments used were: clindamycin, a penicillin/streptomycin combination, bleach, and distilled H₂O as a control. When verified with streaking on an agar plate, neither 35 mM clindamycin or 20× Pen/Strep had full efficacy on S. aureus, with small amounts of visible bacteria after plate incubation for 24 h.

Table 1.

Sensitivity of sample bacteria strains to antibiotics, tested using colony counting

	S. aureus (gram-positive)	P. aeruginosa (gram-negative)	M. catarrhalis (gram-negative)	S. pneumoniae (gram-positive)
Clindamycin (bacteriostatic)	Susceptible	Resistant	Resistant	Susceptible
Penicillin (bactericidal) Streptomycin (bactericidal)	Susceptible	Resistant	Resistant	Susceptible
Bleach (bactericidal)	Effective	Effective	Effective	Effective
Distilled H2O (control)	No effect	No effect	No effect	No effect

Effective refers to 0% growth in an agar plate 24 h after treatment, susceptible means visibly reduced growth, and resistant means no visible reduction in from growth compared to the control.

Propidium iodide staining

Propidium iodide is a membrane-impermeable DNA indicator, which shows when the membranes of bacteria have been compromised, indicative of cell death. Two-photon propidium iodide imaging (900 nm excitation, 625–665 nm emission) was co-registered with two-photon FAD imaging of S. aureus (Fig. S6), showing that indeed the cells treated with antibiotics exhibited increased cell death, whereas the H₂O-treated bacteria did not. Co-registration is approximate, since addition of propidium iodide was performed after autofluorescence imaging to prevent spectral bleed through. Propidium iodide fluorescence was not visible in a bleach-treated sample, likely due to bleach breaking down and/or quenching the propidium iodide.

Nonlinear optical microscopy

Imaging was performed on a previously described custom nonlinear optical microscope^33,61,62 (Fig. S7). Briefly, an 80 MHz tunable laser (InSight X3 + , Spectra-Physics) was used for 750 nm two-photon excitation of both NAD(P)H and FAD, which were spectrally separated with a 506 nm dichroic filter. The microscopy was in an inverted, epi-detection geometry with a 1.05 NA objective lens (XLPLN25XWMP2, Olympus), using laser scanning with a pair of galvanometer mirrors. Detection was performed using two hybrid photodetectors (R10467U-40, Hamamatsu) using computational photon counting to achieve simultaneous dual-channel intensity and lifetime with accuracy up to approximately 180 Mega-counts-per-second per channel^33,61. The lateral spatial resolution of the system is approximately 390 nm. The frame size of the system is variable, depending on the voltage applied to the scanning galvanometer mirrors (Fig. S7); generally a 100 × 100 μm, 512 × 512 pixel, field of view was acquired. If a larger area was required, multiple adjacent FOVs were collected and stitched together as a mosaic.

Prior to bacteria imaging, the optimal imaging parameters were determined by characterizing background broth signature (Fig. S8), photobleaching (Fig. S9), and the effect of plating conditions (Fig. S10). These results established that safe imaging of multiple bacteria species using 750 nm and 5 mW multiphoton excitation was appropriate, with non-negligible but low background.

Image processing and statistical analysis

Image/data analysis and visualization were performed using custom Matlab scripts.

Author contributions

J.E.S. and S.A.B designed the study. J.E.S., L.Y., and R.R.I developed the custom microscopy system. J.E.S., F.R.Z., A.M., G.L.M., and E.A. prepared bacteria and biofilm samples. J.E.S. performed the imaging and image analysis. J.E.S., M.M., and S.A.B. interpreted the results. J.E.S. and S.A.B. wrote the manuscript with input from all coauthors. S.A.B. obtained funding to conduct the research.

Data availability

The data generated in this study are available from the authors upon formation of a collaborative research agreement.

Code availability

The code generated in this study is available from the authors upon formation of a collaborative research agreement.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41522-026-00920-0.