Investigating Salmonella biofilm responses to antibiotic treatment using optical photothermal infrared spectroscopy

TL;DR

This study uses a new imaging technique to track how Salmonella biofilms respond to antibiotics by analyzing their metabolism at different depths.

Contribution

A novel method combining optical-photothermal infrared spectroscopy and 13C isotope probing is introduced to study biofilm metabolism.

Findings

Metabolic gradients in biofilms show low activity in the core and higher activity in outer regions.

Antibiotic exposure alters metabolic responses in Salmonella biofilms depending on bacterial resistance profiles.

Abstract

Biofilms are microbial communities of aggregated cells encased in extracellular matrix that are a pressing healthcare concern. Since biofilms have complex metabolic dynamics, in this study a new approach for studying biofilm metabolism is developed that employs optical-photothermal infrared (O-PTIR) spectroscopy imaging combined with 13C stable isotope probing and cryosectioning to track the carbon metabolism of cells at different depths of the biofilm. This approach demonstrated that metabolic gradients can be visualised using O-PTIR imaging, revealing a core of cells with low metabolic activity at the centre of the biofilm, with outer regions showing significantly higher metabolic activity. By incorporating the heavy stable isotope of carbon into bacterial biomass, we monitored the metabolic activity of gentamicin-resistant Salmonella Typhimurium within the biofilm structure upon exposure to various antibiotics. O-PTIR imaging revealed altered metabolic responses at various depths of the biofilm, with variations that depend on the bacterial antibiotic susceptibility profile.

Optical-photothermal infrared spectroscopy imaging combined with 13C stable isotope probing reveals changing carbon metabolic gradients in Salmonella Typhimurium biofilms in response to antibiotic treatments.

Figures

Gentamicin treatment has limited impact on *Salmonella* Typhimurium 4/74 (pMRE-Tn7) carbon metabolism through biofilm depth.A PCA scores plot from the point spectral data collected through biofilm depth corresponding to the spectra in Fig. [3D](#Fig3). The clustering pattern indicates that there are three main groups separable in the PC1 and PC2 axes. B PC1 loadings plot showing which peaks have contributed the most to the separation seen in the scores plot, this indicates that amide I is responsible for the most separation in PC1, with other carbon-associated peaks contributing to the separat

Full text

Introduction

Biofilms are structures comprised of aggregated microorganisms, along with extracellular polymeric substances^1–3^, that consist of proteins, polysaccharides, lipids and extracellular DNA^4,5^. They are typically highly tolerant to environmental stress including antibiotic treatment and disinfection, extreme temperatures, variations in pH and nutrient starvation^1,6^. Using a range of techniques, biofilms have previously been shown to have stratified metabolic activity due to oxygen limitation at the centre of the biofilm, with higher activity found around the edges of the biofilm^7–12^. Biofilms are ubiquitous^13^, and pathogenic bacteria can form biofilms that are extremely difficult to treat using conventional methods^14,15^, making new methods for treating biofilm-related infections a priority. Developing these methods requires a better understanding of how biofilms respond to current drug treatments. Compounding the difficulties of treating biofilms, antimicrobial resistance (AMR) is also promoted in biofilms^16,17^. AMR is a pressing healthcare concern, and it is estimated that, unless steps are taken to address the issue, by 2050 there could be 10 million deaths annually caused by drug-resistant illnesses^18^.

Salmonella enterica subsp. enterica serovar Typhimurium (S. Typhimurium) causes invasive non-typhoidal Salmonella infections and readily forms biofilms on both abiotic and biotic surfaces^19,20^. Globally, there are 93.8 million human gastroenteritis cases caused by non-typhoidal Salmonella, resulting in a death toll of around 150,000 individuals^21,22^. The antimicrobial resistance rate within Salmonella and the diversity of resistance have been increasing^23^, with some reports indicating more than 90% of isolates tested (ranging from non-typhoidal to typhoidal serovars) are multi-drug resistant (MDR)^21,22^. It is well established that the biofilm phenotype is important in how Salmonella Typhi, responsible for typhoid fever, colonises gallstones in the gut and contributes to chronic infection^24^. The role that S. Typhimurium biofilm formation plays in infection is not as well understood, but it is thought that alternating between biofilm and planktonic states helps this organism to evade the host immune system^25^.

Most traditional microbiological techniques for biochemical analysis of multicellular communities rely on bulk measurements, but as biofilm architecture and cellular heterogeneity are likely responsible for many of the mechanisms of stress tolerance^26–28^, bulk measurements lack important biological spatial context. Fluorescence microscopy, particularly confocal laser scanning microscopy (CLSM), is a valuable technique for studying biofilms, and is used widely within this field for biofilm imaging, including determining responses to drug treatments^29^. CLSM provides spatially resolved 3D fluorescent images of biofilms through different depths in a non-destructive manner^30,31^. Although this is a powerful technique, several experimental limitations hinder the use of CLSM for explaining biofilm biochemistry and metabolism. These include the potential for fluorescent tags to alter biochemical processes, the need for suitable and distinguishable fluorescent markers^32,33^, and the requirement for such markers to penetrate the biofilm^9,34^. Importantly, CLSM and fluorescent techniques generate indirect images, visualising fluorescent markers rather than the actual biochemical components of the sample^35^.

Infrared (IR) spectroscopy measures the absorption of IR light by an analyte across the infrared wavelength range, producing a spectrum with characteristic peaks that reveal the functional groups present in a sample with their relative abundances. During the past two decades, IR spectroscopy has been employed for various microbiological applications, including rapid bacterial identification^36–38^, probing bacterial metabolic response to stressors^39^, and monitoring industrial bioprocesses^40^. Generally, the spatial resolution of IR imaging techniques in the mid infrared range is diffraction-limited to approximately 5–10 μm^41,42^, preventing the resolution of fine details at the cellular level. To achieve infrared imaging with a much-improved spatial resolution, optical- photothermal infrared (O-PTIR) spectroscopy has been developed, achieving submicron resolution (~500 nm; which is dependent on the frequency of the probe laser) and providing single-frequency imaging capability^35,42,43^. O-PTIR uses a tuneable mid-IR quantum cascade laser (QCL) beam to induce thermal expansion in a sample^35,42,43^, which is monitored by a continuous wave visible laser beam that is made collinear with the IR laser^35,42,43^. Using visible light (in our case 532 nm) to probe IR signatures significantly improves the spatial resolution compared to traditional FT-IR microspectroscopy, due to the shorter wavelength of visible light^35,42–44^.

One of the strengths of vibrational spectroscopy is its compatibility with the stable isotope probing (SIP) approach. This approach enables the tracking of metabolic activity in bacterial cells within biofilms by monitoring the active uptake and assimilation of stable heavy isotopes, such as ¹³C, ¹⁵N, ¹⁸O, or ²H, introduced through the growth medium^45–48^. As metabolically active cells incorporate these isotopes into their biomass, SIP provides a powerful means to identify and characterise functionally active individual cell members of complex microbial communities^9,44,49^. Using a ^13^C-labelled nutrient allows tracking of metabolism of various carbon-containing biomolecules in a sample, proteins are generally the easiest to track due to the high intensity of the protein-related amide I vibrational peaks^50^.

Earlier studies demonstrated the potential of combining SIP with spectroscopic techniques, such as Raman, IR, and O-PTIR spectroscopy, for the quantitative detection of heavy isotope-labelled compound incorporation at both the community and single-cell levels^46,51^. The quantitative nature of such an approach has also been validated by other studies using alternative analytical techniques such as secondary ion mass spectrometry (SIMS)^50^, as quantitative SIP has been widely used in SIMS and other mass spectrometry techniques^52,53^. In 2023, these advances led to a study which reported the combined application of O-PTIR and SIP for rapid identification of E. coli isolates that exhibit antimicrobial resistance at the single-cell level^44^, demonstrating the potential clinical applications of this approach. Here, we build on these advances to apply O-PTIR and SIP to biofilms, providing non-destructive label-free chemical imaging to highlight metabolic gradients present in biofilms, and how they change when treated with antibiotics.

While spectroscopic techniques have been used extensively on biofilms^54–57^, SIP has not yet been combined with spatially resolved O-PTIR imaging in a biofilm study. Most label-free molecular imaging approaches used to study biofilms have employed mass spectrometry imaging (MSI) to track distributions of target molecules within biofilms, with most studies focused on investigating the distribution of bacterial signalling molecule within biofilms^58,59^. In this study, the recent advances in SIP with vibrational spectroscopy are employed to address this significant gap in the field, allowing spatially resolved chemical imaging of biofilm’s metabolic activity at depth. ^13^C SIP is used for the first time in conjunction with O-PTIR spectroscopy to provide an insight into the protein metabolism throughout the depth of a biofilm.

Results

O-PTIR Imaging of Salmonella Biofilms

To assess the metabolic activity of bacterial cells at varying depths within the biofilm, the SIP approach was employed. An overnight culture of S. Typhimurium 4/74 (pMRE-Tn7) was spotted onto a polycarbonate membrane placed onto colonising factor antigen (CFA) agar and incubated at 30 ^◦^C for 24 h to facilitate the formation of early-stage biofilms^8^. The polycarbonate membrane was then transferred to M9 minimal medium containing uniformly labelled glucose ([U-^13^C_6_]-D-glucose) as the sole carbon source and incubated for an additional 24 h (Fig. 1A). The biofilms were sectioned using our optimised cryosectioning protocol (detailed in the Methods section) and placed onto CaF_2_ slides suitable for imaging with O-PTIR spectroscopy^60^. Cryosectioning was required as O-PTIR is a surface-based technique and cannot be operated confocally^35^.Fig. 1Salmonella Typhimurium 4/74 (pMRE-Tn7) biofilms exhibit gradients of protein metabolism.A Diagram describing biofilm growth conditions and demonstrating where in the biofilm the section has been obtained, in this case the edge of the biofilm. B Optical image of biofilm section with red markers highlighting point spectra that were collected every 5 μm and numbers indicating the collection order. C O-PTIR ratiometric image of single frequency images ratioed 1616 cm^-1^:1655 cm^-1^ indicating where ^13^C from the growth medium is incorporated into bacterial proteins. The image indicates a core of cells with low metabolic activity with much higher activity around the edges. D Point spectra collected through the depth of the biofilm arranged in order of collection. Each plotted spectrum comprises five averaged point spectra, which are shown on the biofilm image as red markers (B and C sections) and are each comprised of ten averages upon collection. These spectra demonstrate the shifts in amide I throughout the depth of the biofilm and confirm that the effect seen in the ratioed image is reflected in the full spectra.

The SIP method combined with O-PTIR analysis enables the identification and localisation of metabolically active cells by tracking shifts in carbon-associated vibrational peaks, resulting from ^13^C incorporation into various biomolecules and causing a redshift in the O-PTIR spectral bands. When ^13^C is incorporated into various functional groups, there is an associated change in reduced mass, as the mass of constituent atoms increases^46^. The increase in reduced mass changes the vibrational frequency of the bond, red-shifting the spectral bands^46^. Peaks that are redshifted upon ^13^C incorporation include amide I (1655 cm^-1^, mostly arising from C = O stretching vibrations in proteins, as well as contributions from C-N stretch and N-H bend^46,50^), amide II (1545 cm^-1^, C-N stretch, N-H bend^50^), amino acids and lipids (1400 cm^-1^, COO^-^ stretch^50^), amide III (1244 cm^-1^, N-H in-plane bend, C-N stretch^50^) and lipid backbone peaks (C-H stretches) that fall outside the O-PTIR QCL range^50^. Amide I exhibits the biggest shift when assimilating ^13^C, shifting to around 1616 cm^-1^ from 1655 cm^-1^ ^46,61^. This considerable peak shift in amide I can be exploited to indicate where in a sample active metabolism is occurring, indicating active growth and utilisation of the carbon source provided in the growth medium due to the incorporation of ^13^C into the proteins in bacteria^51^. This is particularly useful in biofilms, where there is known to be stratification of metabolic activity throughout the biofilm, due to oxygen limitation at the core of the biofilm^9,12,62^.

Single-frequency imaging of the biofilm was conducted by tuning the QCL to probe the amide I band at 1655 cm^-1^ (^12^C = O) and 1616 cm^-1^ (^13^C = O), followed by generating a final image based on the intensity-ratio of these O-PTIR signals. This approach produces chemical maps that highlight the most metabolically active regions of the biofilm, revealing areas with higher or lower incorporation of ^13^C into proteins associated with the amide I peak. Using this approach revealed a region of high metabolic activity concentrated around the edges of the biofilm, while a substantial central region exhibited low levels of ¹³C incorporation (Fig. 1C). This result agrees with findings reported in the literature, suggesting that biofilms are most metabolically active around the edges and have a core of cells with very low metabolic activity^8,9,11,62^. Collecting point spectra from the top to the bottom of the biofilm (Fig. 1D) provided a more detailed view, showing the spectral shifts induced by incorporation of ^13^C into biomolecules in individual point spectra sampled throughout the depth of the biofilm. A control biofilm grown on M9 medium containing ^12^C_6_-glucose was also prepared and analysed using the same experimental setup, with no shifts identified (Fig. S2). The lack of peak shifts in the control biofilm confirms that the previous shifts detected in the ^13^C-glucose-grown biofilm reflected the incorporation of ^13^C from the growth medium by the metabolically active cells.

The differences in ^13^C incorporation into the amide I band can be identified both in the ratioed image where the intensity of the 1616 cm^-1^ signal relative to the 1655 cm^-1^ signal varies by region, and in the line-spectral data which shows several peak shifts (Fig. 1D). To explore these data further, all the line-spectral data were subjected to principal component analysis (PCA) (Fig. 2A). The resulting PCA scores plot confirms that the spectra can be broadly separated into groups representing the different levels of ^13^C incorporation. The PC1 loadings plot (Fig. 2B) accounting for 82.31% of the total explained variance (TEV) of the data, clearly highlights the significant contribution of the amide I peak to the separation observed in the scores plot, with a split peak indicating that the peak is shifting due to ^13^C incorporation. There are also contributions from other peaks within the fingerprint region in the loadings plot, including peaks found at 1531 cm^-1^ (protein, amide II comprising of C-N stretching and N-H bending vibrations of proteins), 1359 cm^-1^ (lipids and amino acids, symmetric COO^-^ stretching), 1099 cm^-1^ (DNA, symmetric PO_2_^-^ stretching^50,63^) and 1029 cm^-1^ (carbohydrate)^64^.Fig. 2Salmonella Typhimurium 4/74 (pMRE-Tn7) carbon metabolism changes throughout the depth of the biofilm.A PCA scores plot from the point spectral data collected through the biofilm depth corresponding to the spectra in Fig. 1D, the clustering pattern indicates that there are two main groups separated in the PC1 axis. B PC1 loadings plot showing which peaks have contributed the most to the separation seen in the scores plot, this indicates that the amide I protein peak is responsible for the most separation in PC1 axis, with other carbon-associated peaks contributing to the separation to a lesser extent. C Averaged selections of 4 spectra from both the high and low ^13^C incorporation regions identified in the ratioed image illustrate the peak shifts of carbon-containing peaks.

To better visualise shifts beyond the prominent amide I shift, four spectra from both the high and low ^13^C incorporation regions were selected and averaged for comparison (Fig. 2C). The resulting spectra clearly reveal shifts in all carbon-associated vibrations due to ^13^C incorporation, consistent with the significant peaks highlighted in the PC1 loadings plot (Fig. 2B). The results presented indicate that the O-PTIR platform can provide a direct biochemical map reflecting variation in metabolic activity across a biofilm.

O-PTIR Imaging of Kanamycin and Gentamicin Treated Salmonella Biofilms

After demonstrating that biofilms exhibit a gradient in carbon metabolism, most notably into proteins, the effect of antibiotic treatment on the pattern of carbon incorporation was investigated. To measure the metabolic changes induced by antibiotic treatment, Salmonella biofilms were cultured on polycarbonate membranes placed on CFA agar for 24 h, then transferred to M9 agar containing uniformly labelled glucose ([U-^13^C_6_]-D-glucose) as the sole carbon source, this time supplemented with gentamicin (Figs. 3–4) or kanamycin (Figs. 5–6). Biofilm samples were then sectioned and transferred onto CaF_2_ slides prior to O-PTIR analysis. Doses of 0.08 mg/mL (5ₓMIC) of kanamycin and 0.02 mg/mL gentamicin were selected based on the results from antibiotic susceptibility testing performed to calculate minimal inhibitory concentration (MIC) (Fig. S1). These values were selected to allow analysis of the metabolic response to antibiotics to which the bacteria are susceptible (kanamycin) and non-susceptible (gentamicin).Fig. 3Salmonella Typhimurium 4/74 (pMRE-Tn7) biofilms show minimal changes in protein metabolism when treated with gentamicin.A Diagram describing biofilm growth conditions and demonstrating where in the biofilm the section has been obtained, in this case a central region of the biofilm. B Optical image of biofilm section grown with gentamicin, with red markers highlighting point spectra that were collected every 5 μm and numbers indicating the collection order. C O-PTIR ratiometric image of single frequency images ratioed 1616 cm^-1^:1655 cm^-1^ indicating where ^13^C from the growth medium is incorporated into proteins (amide I). This indicates a similar pattern of metabolic activity to the untreated example. D Point spectra collected through the depth of the biofilm arranged in order of collection. Each plotted spectrum comprises five averaged point spectra, which are shown on the biofilm image as red markers (B and C sections) and are each comprised of ten averages upon collection. These spectra demonstrate the shifts in amide I throughout the depth of the biofilm and confirm that the effect seen in the ratioed image is reflected in the full spectra.Fig. 4. Gentamicin treatment has limited impact on Salmonella Typhimurium 4/74 (pMRE-Tn7) carbon metabolism through biofilm depth.A PCA scores plot from the point spectral data collected through biofilm depth corresponding to the spectra in Fig. 3D. The clustering pattern indicates that there are three main groups separable in the PC1 and PC2 axes. ** B** PC1 loadings plot showing which peaks have contributed the most to the separation seen in the scores plot, this indicates that amide I is responsible for the most separation in PC1, with other carbon-associated peaks contributing to the separation to a lesser extent. C Averaged selections of four spectra from both the high and low ^13^C incorporation regions identified in the ratioed image illustrate the peak shifts of carbon-containing peaks.Fig. 5. Treatment of Salmonella Typhimurium 4/74 (pMRE-Tn7) with kanamycin alters protein metabolism through biofilm depth.A Diagram describing biofilm growth conditions and demonstrating where in the biofilm the section has been obtained, in this case a central region of the biofilm. B Optical image of biofilm section grown with kanamycin with red markers highlighting point spectra that were collected every 5 μm, and numbers indicating the collection order. C O-PTIR ratiometric image of single frequency images ratioed 1616 cm^-1^:1655 cm^-1^ indicating where ^13^C from the growth medium is incorporated into proteins (amide I). The image indicates a changed carbon incorporation pattern compared to the untreated sample, with no visible carbon incorporation at the bottom of the biofilm, and reduced incorporation at the top of the biofilm. D Point spectra collected through the depth of the biofilm arranged in order of collection. Each plotted spectrum comprises of five averaged point spectra, which are shown on the biofilm image as red markers (B and C sections) and are each comprised of ten averages upon collection. These spectra demonstrate the shifts in amide I throughout the depth of the biofilm and confirm that the effect seen in the ratioed image is reflected in the full spectra.Fig. 6. Kanamycin treatment impacts Salmonella Typhimurium 4/74 (pMRE-Tn7) carbon metabolism through biofilm depth.A PCA scores plot from the point spectral data collected through biofilm depth corresponding to the spectra in Fig. 5D. The clustering pattern indicates that there are two main groups separated according to the PC1 axis. B PC1 loadings plot showing which peaks have contributed the most to the separation seen in the scores plot, indicating that amide I is responsible for most of the separation in PC1 with other carbon-associated peaks contributing to a lesser extent. C Averaged selections of four spectra from both the high and low ^13^C incorporation regions identified in the ratioed image illustrate the peak shifts of carbon-containing peaks.

As expected, gentamicin treatment had minimal impact on the pattern of ^13^C incorporation within the bacterial biofilm community, as displayed by the 1616 cm⁻¹:1655 cm⁻¹ ratio image (Fig. 3B–D). Following our previous approach, the pre-processed O-PTIR spectral data from the biofilm samples were subjected to PCA to identify any potential clustering patterns (Fig. 4A). The resulting PCA scores plot revealed three distinct clusters, with clear separation along the PC1 and PC2 axes, with a combined TEV of 88.17%. These clusters correspond to the regions at the edge of the biofilm with high ^13^C incorporation, the biofilm core with low ^13^C incorporation, and at the boundary between these two regions with intermediate ^13^C incorporation. Inspecting the PC1 loadings plot (Fig. 4B) indicates the amide I peak as the primary contributor to the separation observed in the scores plot. The presence of a split peak suggests that the variation in the amide I band, driven by differing levels of ^13^C incorporation, is the dominant factor distinguishing between the biofilm regions. While other minor peaks also contributed to the separation, they were less prominent than those observed in the kanamycin-treated biofilm. This is likely due to the higher overall ^13^C incorporation in the gentamicin-treated samples, amplifying the influence of amide I band shifts in the PCA separation. These peaks are found at 1531 cm^-1^ (protein, amide II), 1359 cm^-1^ (lipids and amino acids, symmetric COO^-^ stretching), 1219 cm^-1^ (DNA, asymmetric PO_2_^-^ stretching), 1133 cm^-1^ (polysaccharides, COC deformation^65,66^), 1095 cm^-1^ (DNA, symmetric PO_2_^-^ stretching) and 1021 cm^-1^ (carbohydrates)^64^. Similar to the results of the untreated conditions, the average spectra from the high and low ^13^C incorporation regions (Fig. 4C), clearly displayed the shifts in amide I (1655 cm^-1^ shifts to 1619 cm^-1^), and other peaks such as: amide II (1539 cm^-1^ shifts to 1535 cm^-1^) and symmetric COO^-^ stretching (1391 cm^-1^ shifts to 1367 cm^-1^).

In contrast, kanamycin treatment significantly disrupted the incorporation of ^13^C into the bacterial cells, as observed through the 1616 cm^-1^:1655 cm^-1^ ratioed image of the amide I band in the biofilm (Fig. 5C), with no notable incorporation seen at the base of the biofilm, where the antibiotic concentration was potentially the highest, and only limited incorporation was evident at the top layer. This indicates that kanamycin has a pronounced inhibitory effect on the metabolic activity of the bacterial cells, as active ^13^C incorporation would typically be expected in metabolically active cells around the biofilm periphery. The presence of a thin layer of ^13^C incorporation at the top of the biofilm furthest away from the antibiotic source highlights the protective properties of the biofilm against antibiotic penetration, resulting in exposure to sub-inhibitory concentration levels of the antibiotic. Previous studies have shown that diffusion of antibiotics through biofilms typically results in a gradient forming in which the antibiotic becomes more dilute the further it travels through the biofilm, with the physicochemical properties of the antibiotic and how it interacts with biofilm components determining this gradient^67–69^.

To explore these effects further, the pre-processed O-PTIR spectral data from the kanamycin-treated biofilm samples were subjected to PCA to identify any potential clustering patterns (Fig. 6A). The PCA scores plot revealed two distinct clusters, clearly separated along the PC1 axis, with a TEV of 75.9%. The corresponding PC1 loadings plot (Fig. 6B) revealed that while the separation is largely attributable to the shifting peak at amide I, as seen by a split peak in the amide I region, several other relevant peaks contribute to the separation between these groups. These include peaks at 1533 cm^-1^ (protein, amide II), 1359 cm^-1^ (lipids and amino acids, symmetric COO^-^ stretching) and 1097 cm^-1^ (DNA, symmetric PO_2_^-^ stretching)^64^. To better visualise these spectral differences, four spectra from regions with high and low ^13^C incorporation were averaged and overlaid (Fig. 6C), corresponding to significant peaks from the PC1 loadings plot (Fig. 6B). Interestingly, contribution from the symmetric PO_2_^-^ stretching peak from DNA found at 1097 cm^-1^ is notable in the loadings plot, despite not containing carbon and hence not shifting upon incorporation of ^13^C into the biofilm. This may reflect broader biochemical changes in response to antibiotic stress beyond metabolic changes that impact on ^13^C incorporation. Collectively, these findings demonstrate that O-PTIR imaging is a powerful approach for assessing biofilm metabolic responses to antibiotic treatment, providing spatial insight into how protein metabolism shifts throughout biofilms facing different treatments.

Discussion

Biofilms represent highly structured, surface-associated microbial communities that are embedded within an extracellular polymeric substance  (EPS) matrix. These systems exhibit complex metabolic behaviour, which is often spatially and temporally heterogeneous due to gradients in nutrients, oxygen, waste products and signalling molecules^70^. Studying biofilm metabolism at high spatial resolution remains a significant challenge, primarily due to the dense and protective matrix that limits access and hinders the application of many conventional biochemical and analytical techniques. Additionally, the intrinsic complexity of biofilms, comprising a mixture of diverse chemical species and metabolic states, further complicates interpretation of metabolism using bulk analytical methods^70,71^. Aside from bulk studies, microscopy techniques such as CLSM and light sheet microscopy are also commonly applied to biofilms, however fluorescence-based approaches rely on using labels that may impact analysis. Additionally, mass spectrometry approaches can provide rich chemical information but are typically destructive, with complex and time-consuming sample preparation processes. In this study we introduce a new implementation of O-PTIR spectroscopy to provide spatially resolved, label-free and non-destructive chemical analysis of biofilms in cross-section, enabling high-resolution mapping of metabolic activity while preserving the structural and biological information. This implementation considers the metabolic heterogeneity within biofilms, thus overcoming one of the major limitations of traditional bulk measurements, where spatial information from within biofilms is lost. The compatibility of O-PTIR spatially resolved chemical mapping with cross-sectional biofilm samples helps to overcome the considerable difficulties in optically sampling biofilms at depth by providing physical access to the biofilm contents. By combining O-PTIR with stable isotope probing using ^13^C-labelled substrates, we achieve simultaneous visualisation and detection of isotopically enriched biomolecules within the native biofilm architecture. Isotopic enrichment of proteins was tracked by spectroscopic single-frequency imaging and provided insights into protein metabolism within the biofilm. This integrated capability not only streamlines the experimental workflow but also enables the in-situ investigation of metabolic heterogeneity within microbial communities without perturbing their native organisation.

To validate our approach, S. Typhimurium biofilms were grown in the presence of ^13^C-glucose and subjected to different antibiotic treatments. We observed spatially distinct patterns of ^13^C incorporation across the biofilm depth, with significantly higher enrichment detected at the biofilm periphery compared to the interior. An important question arising from these results is whether these patterns of ^13^C incorporation are predominantly driven by the oxygen gradient, the nutrient gradient, or a combination of both factors. In the colony biofilm model oxygen is primarily available from the air above the biofilm, while nutrients are available by diffusion from the growth medium at the base of the biofilm^72^. This results in oxygen and nutrient gradients forming in opposing directions with relation to the biofilm, with oxygen more available near the surface and nutrients more available near the base. The observation of higher ^13^C incorporation around the peripheral areas of the biofilm including the surface, which is the furthest area from the ^13^C source, indicates that the dominant driver of metabolic activity in this instance is oxygen availability. This is because ^13^C incorporation at the biofilm surface indicates that glucose from the growth medium can reach the upper level of the biofilm, but is only metabolised and actively incorporated into the biofilm cells at the periphery where oxygen is readily available. These findings align with previous reports indicating the presence of physiologically stratified subpopulations in biofilms, including metabolically dormant cells located at the centre and more active cells at the edges^7–12^. This metabolic stratification is well recognised and has been observed in biofilms using stimulated Raman spectroscopy and D_2_O SIP to show how biofilm cellular arrangement affects the metabolic activity in Pseudomonas aeruginosa^9^. This paper also indicated that in a colony biofilm model metabolic activity is highest near the surface, despite nutrients provided from agar at the base. Previous studies utilising fluorescent tags to study biofilm gene expression patterns have also indicated higher activity near the base and surface of colony biofilms, also attributed to an integration of opposing nutrient and oxygen gradients forming through the biofilm^73,74^. Another recent study predicted similar effects with computational simulations, which was experimentally validated within the study^75^.

Our study is the first, to our knowledge, to visualise carbon metabolic gradients at high spatial resolution (~500 nm) using ratiometric O-PTIR imaging. The biofilm metabolic gradient contributes to the tolerance of biofilms to antibiotic treatment, due to dormant cells downregulating key metabolic pathways that are important antibiotic targets, notably protein synthesis pathways^76,77^.

In this study, we also investigated how patterns of metabolic activity within biofilms change following antibiotic treatment, indicating that antibiotic treatment suppresses carbon metabolism throughout the biofilm. Previous research has primarily focused on examining the distribution of antibiotics within biofilms after treatment, rather than exploring how such treatments impact the metabolic gradients of biofilms^54,78^. Antibiotics have been shown to exhibit limited penetration into biofilm structures^54,78^, with multiple factors influencing the diffusion gradient^67,68^. Given that antibiotics are known to have constrained diffusion through a biofilm, it is likely that this diffusion gradient contributes at least in part to the metabolic gradient we observe in this study. Since biofilms have been shown to regrow after antibiotic exposure^54,79^ it is also unclear whether the observed metabolic arrest reflects cell death or a dormant state. Future studies could apply a multimodal imaging approach to address both the question of whether the observed metabolic gradient is linked to an antibiotic diffusion gradient and the question of whether the metabolically inactive cells are dead or simply dormant. The SIP strategy, combined with the cryosectioning method described in this study, is compatible with multimodal imaging techniques. This compatibility opens the door for future integration of MSI methods like MALDI-MS imaging into the current workflow to study the distribution of antibiotic signatures within the same biofilm sample^58,59^.

Treating biofilm samples with antibiotics allowed our new implementation of O-PTIR for visualising biofilm metabolic gradients to be extended to evaluate antibiotic efficacy against biofilms. This method has the potential to be utilised as a model system, improving insight into how biofilm treatments affect the metabolic dynamics of the community. Although this study focused specifically on the effects of antibiotics on carbon metabolism within biofilms, the same method can be readily adapted for other organisms of interest or different external treatments. An example application could be use of studying cross-feeding mechanisms in a mixed species biofilm. Providing different species of bacteria within a mixed biofilm with a stable isotope-labelled nutrient that only one can metabolise and measuring the isotopic incorporation could shed light on how biofilms can facilitate bacterial cross-feeding in different species and with different nutrients. Additionally, mechanisms of shared resistance may be another key application. Preparing mixed species biofilms from strains with differing resistance profiles and treating with an antibiotic to which one strain is susceptible, and one strain is non-susceptible can be used to investigate mechanisms whereby resistance is conferred throughout the whole biofilm from one resistant strain. Different resistance mechanisms are likely to differ in their ability to confer resistance within a biofilm community, so analysing a range of different strains in coculture with differing resistance profiles could help uncover mechanisms of shared resistance. In this study, the colony biofilm model was employed as a relatively simple model to demonstrate the applicability of ^13^C-SIP to biofilm research, however integration of this platform into more complex biofilm models including flow cell devices could be further investigated to study differences in metabolic activity throughout more complex biofilm structures. Other examples may be found in specific industrial applications, for instance a previous study used nanoSIMS in combination with SIP to unveil metabolic stratification in current-producing biofilms^80^. The method we outline in this publication could be implemented in similar studies where biofilms of specific industrial interest are isolated and subjected to O-PTIR spectroscopic imaging to understand their metabolic heterogeneity. Since it is a non-destructive technique, this would be of particular use in studies that require multimodal biochemical imaging approaches.

In summary this study implements O-PTIR spectroscopy for a new application, providing label-free non-destructive chemical imaging of functional groups throughout biofilm depth. Our work showed that biofilms have protein synthesis gradients using this chemical imaging capability. Using the O-PTIR platform to investigate antimicrobial treatments gives detailed information about the effect a treatment has on biofilm protein metabolism, demonstrating the impact antibiotic treatment has on this metabolism as well as the protective properties biofilms confer against treatment.

Methods

Bacterial Strains and Growth Conditions

The Salmonella strain used in this study was a derivative of Salmonella enterica serovar Typhimurium strain 4/74. A strain with engineered non-susceptibility to gentamicin was prepared by introducing a pMRE-Tn7 plasmid^81^ into S. Typhimurium 4/74 to confer non-susceptibility to gentamicin^82^. To prepare cultures S. Typhimurium 4/74 (pMRE-Tn7) was grown from glycerol stocks by streaking and growing overnight for 18 h at 37 ^◦^C on an adapted biofilm-promoting^1^ CFA agar (w/v 1.5% agar, 1.5% Yeast Extract, 0.1% Casamino acids, 0.05% MgSO_4_ and 0.005% MnCl_2_^1^) with plates containing 40 μg/mL gentamicin. Axenic colonies were obtained by repeating subcultures on gentamicin-containing CFA agar plates, and were used to inoculate CFA broth cultures which were grown shaking overnight for 18 h at 37 ^◦^C.

Minimal Inhibitory Concentration (MIC) Assay

To establish concentrations of gentamicin and kanamycin that elicit non-susceptible and susceptible responses to treatment respectively, MICs were calculated. Bacterial cultures were prepared from glycerol stocks as described previously. Antibiotic solutions of kanamycin and gentamicin were prepared in deionised water to an initial stock solution of 1.024 mg/mL. This was filtered with a filter pore size of 0.22 μm, then serially diluted in CFA broth 2-fold 14 times down to 6.25 × 10^-6 ^mg/mL. The overnight culture optical density at 600 nm (OD_600nm_) was measured and adjusted to 0.13 in CFA medium, then added to the dilutions to a final OD_600nm_ of 0.1 with final antibiotic concentrations ranging from 0.256 mg/mL to 1.5625 × 10^-5 ^mg/mL. Aliquots (200 μL) of these dilutions, as well as blank growth medium and a positive control with water instead of antibiotic solution, were pipetted in triplicate into a 100-well honeycomb Bioscreen plate. These were incubated overnight using a Bioscreen C growth profiler at 37 ^◦^C with continuous medium shaking until stationary phase was reached and maintained (16 h), with OD_600nm_ measurements recorded at 12 min intervals. From the resulting growth profiles (Fig. S1), the minimum inhibitory concentrations (MICs) were determined to be 0.016 mg/mL for kanamycin and 0.128 mg/mL for gentamicin. The gentamicin value is above the EUCAST breakpoint for bacteria in the order Enterobacterales of 0.002 mg/mL, so this strain can be considered non-susceptible to gentamicin^83^. A dosage of 0.02 mg/mL was selected for gentamicin, since the bacteria are non-susceptible at this concentration. Conversely, for kanamycin a value of 5 x MIC (0.08 mg/mL) was selected, which is in agreement with other Salmonella biofilm studies^84–86^, enabling the metabolic response of susceptible biofilms to be analysed.

Biofilm Formation Conditions

Bacterial cultures were prepared from glycerol stocks as described previously. The OD_600nm_ of these cultures was measured and adjusted to 0.05, then 10 μL was spotted onto 0.2μm polycarbonate membranes placed onto CFA agar plates, which were then incubated at 30^◦^C^8^. Polycarbonate membranes were selected due to the ease of removing from the growth medium for sectioning, other studies have previously used this method for Pseudomonas biofilms^8^, or used polycarbonate coupons in the CDC biofilm reactor for Salmonella biofilms^87,88^. After 24 h growth, these membranes were transferred to fresh M9 agar plates (w/v 0.5% glucose, 0.1% NH_4_Cl, 0.05% MgSO_4_.7H_2_O, 0.3% Na_2_HPO_4_, 0.3% KH_2_PO_4_, 0.0001% FeSO_4_.7H_2_O, 0.0001% CaCl_2_) of relevant experimental conditions (^12^C-labelled glucose without antibiotic, ^13^C-labelled glucose without antibiotic, ^13^C-labelled kanamycin treated, ^13^C-labelled gentamicin treated) and incubated at 30 ^◦^C for a further 24 h. Kanamycin-treated conditions were subjected to a concentration of 0.08 mg/mL, or 5 x MIC, and gentamicin-treated conditions were subjected to a concentration of 0.02 mg/mL.

Cryosectioning of Biofilm Samples

Cryosectioning was performed on biofilms after preparation of samples using the above protocol. Biofilms on polycarbonate membranes were carefully detached from the surface of the agar and the section of membrane containing the biofilm was cut out using scissors. The biofilms were then placed into cryomolds (Tissue-Tek), and tissue-freezing medium (Leica) was layered onto them to embed them for cryosectioning. They were then frozen using dry ice and 100% ethanol slurry. Liquid nitrogen was not considered, as due to the Leidenfrost effect it can freeze slowly and result in uneven freezing and sample damage^78,89^. The frozen blocks were transferred to a Leica CM1850 cryostat set to −21 ^◦^C equipped with MX35 ultra blades (Epredia) and cut at 5 μm thickness onto CaF_2_ slides (Crystran Ltd.) suitable for O-PTIR analysis.

O-PTIR Analysis

Once biofilm samples were sectioned onto CaF_2_ slides, they were analysed using O-PTIR spectroscopy. O-PTIR analysis was carried out using the mIRage infrared microscope (Photothermal Spectroscopy Corp., Santa Barbara, USA). Single-frequency images were collected by tuning the quantum cascade laser (QCL) to amide I (1655 cm^-1^) and ^13^C-labelled amide I (1616 cm^-1^). These peaks from the C = O vibration were ratioed using the PTIR Studio version 4.6 software (Photothermal Spectroscopy Corp.) with cross-correlated inputs. Based on the ratioed images, point spectra (in the 937–1805 cm^-1^ and 1973–2323 cm^-1^ regions) were collected in reflection mode with 2 cm^-1^ spectral resolution, using the inbuilt line array function, from the top to the bottom of the biofilm section. Each point spectrum comprised of 10 averages.

Data Analysis

After acquisition of O-PTIR spectra, the data were processed using Matlab Software Version 2021a (Mathworks). Where appropriate, spectra from the very top and bottom of the line of acquired data were excluded, as in some situations these spectra went beyond where the biofilm was found and had contributions from other elements (tissue-freezing medium, polycarbonate membrane) which would interfere with the processing of the spectra. Additionally, excessively noisy spectra were manually removed. All spectral data were scaled using the extended multiplicative signal correction (EMSC) algorithm^90^, then smoothed using Savitsky-Golay filtering with an order of 2 and a frame length of 11. Principal component analysis (PCA) was then used as an unsupervised multivariate analysis to identify clustering patterns within the data. Both the PCA scores plot, which shows spectra as points and demonstrates similarities and differences between groups, and the PCA loadings plot which shows the regions that contribute most to this separation, were plotted and investigated.

Statistics and Reproducibility

Discrete biofilm images were collected for analysis with point spectra collected from top to bottom. In this instance point spectra numbers collected from each image ranged from 35 to 56. Due to the risk of damaging the sample with the laser, replicate point spectra were not collected; however each point spectrum was comprised of 10 averages to ensure reproducibility.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information Description of Additional Supplementary files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Reporting Summary