In-situ Raman spectroscopic-based microfluidic “labon-a-chip” platform for non-destructive and continuous characterization of Pseudomonas aeruginosa biofilms

Abstract

Pseudomonas aeruginosa biofilm was cultivated and characterized in a microfluidic “lab-on-a-chip” platform coupled with confocal Raman microscopy in a nondestructive manner. Biofilm formation could be quantified by this label-free platform and correlated well to confocal laser scanning microscopy. This Raman-microfluidic platform could also discriminate biofilms at different developmental stages.

A biofilm is a consortium of bacteria in which cells associate with each other and adhere to a surface ^{1, 2}. These sessile cells are embedded within a matrix of secreted extracellular polymeric substances (EPS), including exopolysaccharides, proteins, lipids and nucleic acids, which comprises up to 90% of the biofilm ³. EPS not only support biofilm structure mechanically, but also encase biofilm bacteria thus allowing them to more easily mediate cell-to-cell communication, and protecting them from environmental stresses, such as antimicrobials ⁴. Biofilms on medical devices and food processing surfaces (e.g. infusion tubes, water pipes) are highly resistant to antimicrobial treatments and disinfectants and thus represent major health risks ^{3, 5}.

The study of biofilm development is often limited to methods that involve static cultivation conditions, with low shear flow and no nutrient exchange. On the other hand, cultivation under flow conditions more closely mimics biofilms found in the environment and body due to hydrodynamic influences on biofilm structure and cell signalling ^{6, 7}. For example, biofilm formation of rpoS-deficient Escherichia coli is impaired under flow cultivation, compared to that under static conditions ⁸. The commonly used macro-scale flow cells require large volumes of media, resulting in a relatively expensive system that does not allow for spatial and temporal control of biofilm formation ⁹. By precise control of different hydrodynamic conditions, microfluidic platforms can closely simulate the appropriate environmental conditions and allow substantial reductions in reagents used. Microfluidic platforms have been recently applied to the study of bacterial biofilms in a high-throughput manner ^{10, 11}.

However, characterization of biofilms cultivated in a microfluidic platform has been challenging. Most studies have applied confocal laser scanning microscopy (CLSM) for quantitative studies after biofilm staining ^{12, 13}. However the staining process involved does not allow non-destructive characterization of biofilms in their natural state. Furthermore, chemical variations during biofilm development cannot be fully monitored in a continuous manner. We considered an alternative approach using Raman spectroscopy that can observe vibrational, rotational and other low-frequency energies in a molecule and/or biological system. When coupled with confocal imaging techniques, Raman spectroscopy can be used in situ to determine the chemical composition and localization of bacterial biofilms in three dimensions, without staining of bacterial cells ^{14, 15}.

Here, we report the first in-situ characterization of Pseudomonas aeruginosa biofilms cultivated in a microfluidic platform using confocal micro-Raman spectroscopy in a nondestructive and continuous manner (Fig. 1). The microfluidic platform was fabricated by soft lithography (Fig. S1, ESI†). The multi-channel microfluidic chip was connected to a pump with bio-compatible tubing. Nutrient broth was continuously infused from the inlet and expelled through the outlet during biofilm formation. Details of P. aeruginosa biofilm cultivation in the microfluidic chip are provided in the ESI†. The cultivation chamber containing the biofilm was loaded onto a sample stage, which motorized the chip location at a spatial resolution of 2 μm in the horizontal phase and a 3 μm in vertical phase. Confocal Raman microscope was integrated with the microfluidic platform (ESI†). Laser introduction and Raman signal collection was conducted through the same 50× objective. The Raman scattering signal was processed through an aperture, passed through a polarizer, and finally acquired by a CCD detector (Fig. 1).

Fig. 1.

Schematic illustration of the Raman spectroscopic-based microfluidic “lab-on-a-chip” platform for cultivation and characterization of bacterial biofilms.

To demonstrate the chemical composition of the biofilm, Raman signals from chip substrate (i.e. glass slide and polydimethylsiloxane (PDMS) layer) were firstly recorded when the microfluidic chip was filled with medium without inoculation and the focusing state was identical to that subsequently used for biofilm characterization. The chip substrate generated distinct Raman peaks at 485, 610, 703, 785, 860, 1125, and 1402 cm⁻¹. After initial attachment of bacterial cells to the substrate, biofilm started to develop and cover the substrate. The intensity of Raman peaks due to the substrate decreased significantly by 24 h (Fig. S2, ESI†). Although biofilm continuously accumulated, the peak intensities from the substrate remained relatively constant, indicating no overlap between biofilm peaks and substrate peaks. Therefore, peaks that were different from substrate peaks and that changed during biofilm development were regarded as deriving from P. aeruginosa biofilm.

The characteristic Raman peaks for biofilm were averaged and evident in the wavenumber range of 400 to 1800 cm⁻¹ in Fig. 2. Peaks at 746, 1123, 1307 and 1580 cm⁻¹ were prominent and could be clearly observed after 24 h cultivation (Fig. 2A). Peaks at 746 and 1580 cm⁻¹ were assigned as specific ring structures in nucleic acids (i.e. adenine, thymine, guanine), while peaks at 1123 and 1580 cm⁻¹ referred to C-C stretching and CH₃/CH₂ twisting or bending respectively which exist widely in carbohydrates, proteins and lipids (Table S1, ESI†). The laser (532 nm) used in this study was shown to enhance the Raman signal of cytochrome c by resonance in other organisms ^{16, 17}. As a highly conserved heme protein, cytochrome c can also be found in P. aeruginosa. The prominent peaks at 746, 1123, 1307, and 1580 cm⁻¹ showed at similar locations as major bands of pure cytochrome c. Thus, cytochrome c may contribute to Raman spectral pattern of P. aeruginosa biofilm. It should be noted that cytochrome c in P. aeruginosa is significantly supressed under biofilm conditions ¹⁸. Thus, nucleic acids, proteins, lipids and carbohydrates should still be the major components of biofilm. Irrespective of peak specificity, the increase in Raman peak intensities correlates to biofilm maturation. We defined the first 24 h as the “early stage” of biofilm development. The results demonstrated that nucleic acids, proteins, lipids, carbohydrates, and cytochrome c were synthesized when biofilm started to form. After 36 h cultivation, the intensity of the major early peaks (i.e. at 746, 1123, 1307 and 1580 cm⁻¹) increased significantly and then remained relatively constant to 45 h. The significant change that occurred at 36 h could be regarded as a stage switch of biofilm development. Accordingly, we defined 36 h to 45 h as the “mid stage” of biofilm development. The major compositions of biofilm at “mid stage” were almost the same as that of “early stage”, but with a higher amount. This indicated that the biofilm had been preliminarily established and synthesis of basic components was temporarily ceased. The intensity of the major early peaks (i.e. at 746, 1123, 1307 and 1580 cm⁻¹) increased significantly up to a maximum at 48 h and then remained relatively constant (Fig. 2B), indicating that 48 h was another critical time point for a switch in the stage of biofilm development, and the synthesis of basic components in P. aeruginosa biofilm did not increase after 48 h. This result indicated that accumulation of basic components into the biofilm was periodic rather than continuous.

Fig. 2. Raman spectroscopy monitoring of the development of P. aeruginosa biofilm in the microfluidic chamber.

(A) Prominent peaks appearing within the first 24 h (early stage) are shaded; (B) Variations in intensities of the corresponding Raman peaks (746, 1123, 1307, and 1580 cm⁻¹) over time. (C) Prominent peaks appearing after 48 h (late stage) are shaded; (D) Variations in intensities of the corresponding Raman peaks (918, 968, 1167, 1223, 1333, and 1357 cm⁻¹) over time.

P. aeruginosa biofilm also showed Raman peaks at 918, 968, 1167, 1223, 1333, and 1357 cm⁻¹ (Fig. 2). Peaks at 918, 1167, 1223, and 1333 cm⁻¹ were assigned to carbohydrates and proteins, while peaks at 968 and 1357 cm⁻¹ were likely due to lipids and nucleic acids (Table S1, ESI†). The intensity of these peaks was too weak to be observed until 45 h, but increased substantially from 45 h to 48 h, and then kept relatively constant until 72 h (Fig. 2C). After 48 h of cultivation, biofilm was considered to have entered “late stage” development using these peaks as indicators. Compared to the counterpart peaks shown in the spectra of biofilm at 24 h cultivation, peaks at 918, 968, 1167, 1223, 1333, and 1357 cm⁻¹ in the spectra of biofilm at 48 h cultivation demonstrated 3.1, 3.2, 6.0, 3.7, 3.5 and 1.7 fold increases in intensity, respectively (Fig. 2D). Peaks assigned to proteins and carbohydrates (918, 1167, 1223, and 1333 cm⁻¹) had higher increases than those for nucleic acids (968 and 1357 cm⁻¹), indicating proteins and carbohydrates accounted for the largest proportion of synthesized substances at this stage. Taken together, we defined biofilm developmental stages according to the appearance of prominent peaks and the significant increases in peak intensity. The results indicated that P. aeruginosa biofilm produced different substances at two different stages (i.e. early stage and late stage).

To confirm the variations of P. aeruginosa biofilms developed at different time points shown in the Raman spectra, principle component analysis (PCA) was utilized. PCA was able to cluster samples on the basis of chemical information and enable further differentiation. Fig. 3 demonstrates analytical groups for different biofilm samples according to the score plot of principle component 1 (PC1) and component 2 (PC2) from the PCA model. Loading the profiles of PC1 and PC2 accounted for 92.85% and 4.78% of the total variation in spectra, respectively (Fig. S3, ESI†). The separating lines, derived from a support vector machine classification algorithm, clearly separated clusters due to biofilms at 24 h, biofilms at 36 h to 45 h, and biofilms at 48 h to 72 h into three groups. The Mahalanobis distance between these groups ranged from 5.57 to 12.83. Clusters with interclass distance values higher than 3 are considered to be significantly different (P < 0.05) from each other ¹⁹. This result further confirmed the obvious stage switches occurred at 24-36 h (early stage) and 45-48 h (late stage) during biofilm development within the first 72 h (Fig. 2). We accordingly defined the biofilm grown between 24 h and 48 h as the “mid stage”.

Fig. 3. Representative two-dimensional principle component analysis for the segregation of P. aeruginosa biofilms at different development stages.

The boundary lines could be used to cluster different groups.

As a parallel study, CLSM was applied to quantify the extent of formation of P. aeruginosa biofilm. CLSM uses Cyto-9 dye added at various times to identify biofilm mass and optical sectioning and reconstruction to give a 3D perspective. The amount of biofilm increased as a function of cultivation time (Fig. 4A). At 24 h, the biofilm consisted of a thin, nearly confluent layer of bacteria without large structures. At 36 h, the confluent layer had increased more than double in depth. Between 42 and 45 h, the biofilm remained relatively similar to that seen at 36 h, although it must be pointed out that the necessity to stain biofilms for this method means that different biofilms were imaged at each time point. At 48 h, the biofilm increased in thickness and denser microcolonies started to form. Biofilms were very mature after 72 h of growth, which is consistent to the results obtained using a more conventional large volume flow cell apparatus ²⁰. Based upon CLSM images, the mean biofilm thickness was calculated to quantifiably reflect the physiological changes (Fig. 4B). The total height of cultivation chamber in the microfluidic chip was 50 μm. At 24 h, mean biofilm thickness was 5.24 μm. Up to 36 h, it had a 2.4-fold increase to 12.71 μm and remained constant until 45 h. The mean thickness of biofilm reached a maximum of 24.75 μm at 48 h cultivation, which was 1.95-fold greater than that of biofilm at 36 h.

Fig. 4. Representative confocal laser scanning microscopic images of P. aeruginosa biofilms grown in a microfluidic chamber.

(A) Each panel shows reconstructions from the top in the large panel and sides in the right and bottom panels (xy, yz, and xz dimensions). Biofilm thickness (μm) increased over time. (B) All the values were measured by confocal laser scanning microscopy. Statistical significance was determined using Student's t-test (* denotes P < 0.05).

Both the Raman and CLSM results demonstrated that the formation of P. aeruginosa biofilm increased significantly (P < 0.05) after 24 h (end point of the early stage) and 48 h (starting point of the late stage). Therefore, the correlation between the Raman and CLSM results was examined. Partial least squares regression (PLSR) model was constructed (Fig. 5). When fitted the calculated biofilm thickness (from Raman spectra) to the measured biofilm thickness (from CLSM) in the PLSR model, these two sets of data showed a close correlation, with a regression coefficient (R-square) of 0.9783. Thus, the PLSR result confirmed that chemical variations (Raman) and physiological variations (CLSM) in P. aeruginosa biofilms cultivated in the microfluidic chip occurred concurrently.

Fig. 5.

Correlation of biofilm thickness measured by confocal laser scanning microscopy and calculated by Raman spectroscopy coupled with partial least-squares regression.

According to previous reports, biofilm development has four major stages, including attachment, accumulation, maturation and dispersion ^{21, 22}. Our current study depicted two important stage switches from attachment to early accumulation (early stage to mid stage) and then to maturation (mid stage to late stage) based on chemical and physiological assessments. However, no evidence was obtained to demonstrate any stage switch from maturation to dispersion. This is most likely due to the fact that dispersion usually occurs at the end of a long period of cultivation, and is induced by environmental stresses (e.g. starvation and accumulation of toxic wastes) and regulated by signalling molecules ^{23, 24}. In the current study, biofilm was developed in a fluidic environment in which nutrients were supplied continuously. In addition, wastes were continuously expelled to reduce the accumulation of toxic substances and signalling molecules. Hence, it is logical to expect a longer maturation time than for static cultivation. Cultivation for 72 h may not be sufficient time to monitor any biofilm switch to the dispersion stage. Another explanation is that accumulation and dispersion within a biofilm might occur simultaneously when the biofilm is fully mature ²⁵. Hence, the biomass and thickness are in a dynamic balance for a long period, and the maturation and dispersion stages could not be differentiated. A long-term monitoring might be necessary to enable completion of the biofilm dispersion stage. Identification of a specific molecule synthesized at the dispersion stage would provide further evidence for identifying this important stage.

Conclusions

To the best of our knowledge, this study presents a first attempt to develop a non-destructive and label-free in situ platform to characterize bacterial biofilms in a precisely controlled hydrodynamic environment. The Raman spectroscopy-based microfluidic “lab-on-a-chip” platform demonstrates the ability to characterize and distinguish P. aeruginosa biofilms in different developmental stages (i.e. early, mid, and late stages). In addition, Raman results were well correlated with CLSM analyses, demonstrating their feasibility for quantifying biofilm thickness and determining chemical composition simultaneously. Future work is required to apply this platform to screen antimicrobials for the inactivation of biofilms in a high-throughput manner.