Electromechanical and elastic probing of bacteria in a cell culture medium

BEPFM spectra were measured by repeated scanning of a single line (slow scan disabled, supplementary figure 2 available at stacks.iop.org/Nano/23/245705/mmedia) and spectra were averaged over an area on ML, PF and PLL. The amplitude of electrical excitation changed from 10 to 100 V, and dependence of the Fourier-transformed, frequency-domain electromechanical response spectra on the voltage applied from the tip were recorded for Millipore water (18 MΩ cm) and electrolyte media, DPBS and DMEM. In water one contact resonance peak on PLL appears at ∼38 kHz. The other contact resonance peak develops around 100 kHz at 60 V; then it shifts to lower frequencies as the tip bias increases (figure 1(b)). Previous considerations of the dynamics of PFM as a contact mode method [20, 24–26] discuss how the resonance frequencies are independent of relative electrostatic and electromechanical effects while providing information on voltage-dependent, local elastic properties of the sample (via contact stiffness) and on topography. The overall positive charge of PLL in water at the pH of the experiment results in a large Debye length with a highly diffuse electrical double layer (EDL), which permits an increase in probing volume with increasing bias, and a softening of the contact. Stiffness of probe–sample contact correlates with EDL thickness in two other solvents (DMEM and DPBS) used in the experiment (figures 1(e), (h)). As EDLs shorten and contact stiffens with increasing ionic strength, DMEM exhibits the next highest f of ∼60 kHz, and f is ∼70 kHz in DPBS (figures 1(e), (h); table 1 available at stacks.iop.org/Nano/23/245705/mmedia).

Zoom In Zoom Out Reset image size

The major peak on the ML bacterium for all three liquid environments appears near the resonance frequency of 50 kHz, and all peaks are broad, spanning 20 to 100 kHz. Applying Gaussian fits to peaks smoothed by adjacent averaging of 75 points (as in figure 1) provided maximum amplitudes, which are plotted against applied biases for ML bacteria (figure 2), revealing a quadratic increase in amplitude with increasing bias in water, and linear relationships of response amplitude with bias in DPBS and DMEM (figure 2 zoomed areas). In water, the nonlinear relationship of amplitude with voltage occurs because of electrostriction [27] and variations in electromechanical dissipation [28]. The linear relationships for DPBS and DMEM indicate that the predominant contribution to the signal comes from electromechanical coupling in the system. Although the linearity of response versus bias is similar for DPBS and DMEM, the differences in the respective spectra in different electrolyte solutions indicate that the EDL is responsible for a portion of the response signal, while similarities in amplitudes and peak resonances emphasize the contributions from the dynamical properties of the cantilever–tip–sample system.

Zoom In Zoom Out Reset image size

Electromechanical response of the ML bacterium on PLL is a convolution of the ML bacterium's response to electrical excitation, the PLL response to electrical excitation, and the tip transfer function (i.e., system dynamics). In liquid environments, the tip transfer function is very complicated, and the ratio of the convoluted response (assuming multiple layers of ML-with-PLL) to PLL electromechanical responses should be, to the first order approximation, independent of the tip transfer function. Also, taking the ratio of ML-with-PLL to PLL electromechanical responses removes the multiplicative component of the PLL response from the spectrum. Such a ratio shows that in DMEM, the ML-with-PLL resonance curve is independent of excitation amplitude within a wide range of the higher voltages. Such behavior suggests an absence of electromechanical response from ML, and image formation is dominated by the mechanical transfer function of ML. In water and DPBS, a convolution of electromechanical and mechanical properties of the ML bacterium is measured. Thus, the electromechanical response of ML changes depending on the solvent (figures 1(a), (d) and (g)). This solvent dependency becomes even more pronounced (figures 1(c), (f) and (i)) when there is compensation for the electromechanical response of PLL.

Fitting a damped harmonic oscillator (DHO) model to the spectra from each voxel of the BEPFM data set generates maps of four parameters: peak amplitude (A), resonance frequency (f), quality factor (Q), and phase (φ) [29]. The frequency ranges of the peaks that were fit are given by the ranges of the color bars in the figures. Each of the four maps from BEPFM show contrast between an isolated ML cell and the underlying PLL-coated mica in the three liquid environments tested: water (figure 3(a)), DPBS (figure 3(b)), and DMEM (figure 3(c)). The amount of contrast is highest in water, lower in DMEM, and lowest in DPBS. It is noted here that contrast decreases yet is maintained at lower applied biases, and similar results have been obtained with other AFM tips and separately prepared samples of the same type of bacteria on PLL-coated mica in the same conditions ([9] and data not published). The ionic strength, pH, osmolarity and main components of DPBS and DMEM are presented in supplemental table 1 (available at stacks.iop.org/Nano/23/245705/mmedia). Although the inorganic components and ionic strengths for DPBS and DMEM are similar, DMEM contains a host of organic components, mainly glucose and including various amino acids and vitamins, which can affect the biological activity of the bacteria [30, 31] as well as BEPFM image formation. The contrast observed in DHO fits of BEPFM images can consist of contributions from the electromechanical properties of the sample material (per the discussion in section 2.1), from the EDLs present between the tip and sample [18, 32], from viscous effects of the liquid on the cantilever [4], and from heterogeneity of mechanical properties within varying tip–surface contact area [24, 33], especially at the edges between the bacteria and substrate surface.

Zoom In Zoom Out Reset image size

One goal of this paper is to distinguish Gram-positive (e.g., ML) and Gram-negative (e.g., PF) bacteria based on electromechanical response, for the first time in biologically relevant media, which is accomplished by comparison of response spectra. ML and PF demonstrated distinctive BEPFM spectra while the PLL surface adjacent to each bacterium elicited similar responses, especially frequency responses. The results shown (figure 4, again normalized to PLL as in figures 1(c), (f) and (i)) were obtained at 50 V. The spectra from water and DPBS that are presented were obtained with slow scan disabled and were averaged from five successive lines (as in supplementary figure 2 available at stacks.iop.org/Nano/23/245705/mmedia). Averaged areas did not include edge effects, which arise from rapid transitions in tip–surface contact mechanics and are observed especially in maps of f. Results for DMEM were obtained from a single image as discussed in the next paragraph. Immediately evident in figure 4 are some differences in the frequency domain. In water, ML has a f around 60 kHz, whereas PF has a peak centered about 80 kHz. DPBS attenuates f for ML more than for PF, while the relative amplitude decreases more for PF. Because the amplitude–frequency spectra in water and DPBS are a convolution of electromechanical and mechanical properties (section 2.1), the amplitude change suggests that the more negatively charged PF experiences more shielding from the EDL in DPBS, yet retains stiffer contact than for ML in DPBS per the shift in f. Although most types of bacteria demonstrate relatively long-term survivability in water and PBS [34], ML and PF are non-spore-forming bacteria that tend to convert into dormant, cyst-like forms with altered ultrastructures [35, 36], whereas in DMEM similar bacteria maintain a natural form for a longer time because of the presence of glucose in the solution. Also, the possibility of distinguishing between ML and PF in pure water has already been shown by Nikiforov et al [21]. Here, large scale mapping of co-seeded ML and PF bacteria was performed in DMEM.

Zoom In Zoom Out Reset image size

Capturing both ML and PF within the same image in DMEM allows for direct comparison between the cells in the electrolyte culture medium (figure 5). The AFM height image confirms that the bacteria speculated to be alive are 400–600 nm tall, as expected (figure 5(a)). Bacteria that appear damaged or dead (D-PF) are clearly discerned in the deflection image (figure 5(b)), which also contains what seems to be PF reproducing and undergoing binary fission in the upper left corner. Groups of ML are in the center of the image, while a ML and a PF adjacent to one another are in the lower left corner. figure 5(c) shows the first principal component of the BEPFM amplitude image, which again shows contrast between bacteria and the PLL surface. Principal component analysis (PCA) of the BE data [37] serves as the initial analytical step in performing artificial neural network analysis (ANN) for functional recognition imaging (FRI) [21] of each type of bacteria on the PLL. However, FRI using ANN analysis does not distinguish readily between intact ML and PF. Functional recognition of bacteria also is possible by fitting with a DHO model to the two peaks within the spectra, resulting in two maps each of amplitude, f and Q (supplementary figure 3(b) available at stacks.iop.org/Nano/23/245705/mmedia). By selecting several spectra from voxels on intact ML, intact PF, D-PF and PLL (figure 5(d)), average bacteria spectra divided by the PLL spectrum are shown to differ mainly in amplitudes, especially for the speculatively damaged cells (figure 5(e)). Histograms compiled for each response parameter (supplementary figure 3(a) available at stacks.iop.org/Nano/23/245705/mmedia) show that the distribution of responses is similar for ML and PF, yet it is apparent that damaged cells and PLL have distinguishing responses. Notably, each spectrum in DMEM exhibits similar resonances and peak shapes (figure 5(e)), suggesting that any electromechanical response is attributable mainly to the constituents of the liquid environment and similar EDL interactions, while the majority of the response depends upon mechanical transfer function of the system, as investigated further in section 2.4.

Zoom In Zoom Out Reset image size

Changing applied force by means of the deflection set point in DMEM on ML during slow scan disabled BEPFM (figure 6) supports the interpretation that electromechanical response is weakly dependent upon contact area mechanics. Again, ML-with-PLL spectra are divided by PLL spectra. Off-surface, the spectrum loses observable, dominant peaks (supplementary figure 4 available at stacks.iop.org/Nano/23/245705/mmedia), thus confirming the limits of BEPFM at low biases as a contact mode technique. An effect from indentation is observed by the values of amplitude for each step in applied force. At increased strains [38] and greater deflection values of 162 nm (9.26 nN) and 204 nm (11.6 nN), relative amplitudes decrease with the increased contact stiffness, whereas at lower indentation forces when contact is still retained (at 0 to −0.468 nN), the relative amplitudes increase. When retracting from the surface in the attractive regime at −2.32 nN, which was much lower than the contact set point during approach, the response of ML is relatively the same as for PLL. At the contact point where the force is −0.926 nN, the relative response is largest, consistent with electrostatic coupling in an EDL [32]. Although resonance peaks do not change noticeably, a minor contribution from mechanical properties of the samples and from stiffening of the contact [24] or surface penetration upon increased indentation cannot be discounted without a complete consideration of the complex, heterogeneous mechanical, dielectric and piezoelectric properties of the systems.

Zoom In Zoom Out Reset image size

The relative structures of Gram-positive (e.g., ML) and Gram-negative (e.g., PF) bacteria cell walls (supplementary figure 5 available at stacks.iop.org/Nano/23/245705/mmedia) provide insight into the results presented here; furthermore, results obtained with other methods in the literature qualitatively agree with the preceding analysis of BEPFM data. Using electrophoretic mobility, Sonohara et al [39] found Gram-positive S. aureus to possess a surface that was less rigid and less negatively charged than Gram-negative E. coli. However, Vadillo-Rodriguez et al [40] performed AFM force-curve creep relaxation experiments to show that the entire Gram-positive B. subtilis cell is stiffer and more viscous than an E. coli cell, in accordance with the relatively thicker peptidoglycan cell wall layers of the Gram-positive bacteria (supplementary figure 5(a) available at stacks.iop.org/Nano/23/245705/mmedia). Our data from BEPFM in liquids of differing ionic strengths corroborates with the surface characteristic results of Sonohara et al in that the Gram-positive ML displays a lower f in water and DPBS suggesting a less rigid surface. The smaller changes in f for PF in different liquids are attributed to the stiff lipopolysaccharide (LPS) layer on their surface, as compared to the more compliant, charged 'brushy' coating of ML that makes it more susceptible to changes in overall cell stiffness with changes in ionic strength of the liquid because of turgor pressure [41]. Decreases in amplitudes with increasing ionic strength, if normalized to an average PLL spectrum, are greater for PF than ML, consistent with the more dramatic screening of the more negatively charged PF. In DMEM, the spectra for each type of cell are dominated by the tip transfer function and are nearly the same because the organic components of DMEM can coat the tip and interact strongly with the surface molecules of the bacteria.

For comparison, a quasi-static force–volume map of Young's modulus for the same group of bacteria in DMEM with the same tip was acquired at different force set points (figure 7). Although each map took approximately the same amount of time to acquire as the BEPFM image (∼10 min), the resolution is obviously lower. This makes identification of the bacteria based on force–distance data difficult. The Young's modulus quantity can be calculated from force curves given accurate knowledge of contact geometry and Poisson's ratio of the sample, yet this requirement cannot be met here because the contact locations on each bacterium cannot be discerned (e.g., the tip could be interacting with the side of a cell) and Poisson's ratios are generally unknown. However, the apparent Young's modulus values calculated by the Hertzian model (assuming a Poisson's ratio of 0.5) of approximately 0.5 to 1.2 MPa are reasonable for the high strain rates of 50 μm s⁻¹ and the large strain gradients induced by a sharp tip [38] and fall within a range seen in the literature [42, 43]. Also as expected, the apparent Young's moduli of the bacteria increased as the tip delved closer to the underlying PLL surface. Yet, more accurate results require better resolution of the probed locale, and imaging techniques such as multi-harmonic AFM [44] and BEPFM, which are based upon different physical mechanisms and offer unique information sets, can provide this complement.

Zoom In Zoom Out Reset image size

Line-mode Band Excitation PFM was implemented using an Asylum Research (Santa Barbara, CA) MFP-3D Atomic Force Microscope (AFM) with an in-house developed MatLab/LabView data acquisition and control system. During the trace raster, a constant-voltage excitation band ranging approximately 16.5–145 kHz (increasing chirp) was applied to the microscope tip at each voxel; each BEPFM data set consists of 128 × 256 voxels. The excitation voltage amplitude was amplified by a factor of ten (FLC Electronics F10A voltage amplifier) to a maximum of 100 V. The mechanical response of the system was recorded by measuring and digitally storing the motion of the tip, taking the Fourier transform of the response during the retrace raster. The amplitude, resonance frequency, phase, and quality factor were extracted from individual voxels using a damped harmonic oscillator model. A gold-coated, pyramidal tip (Olympus, TR400PB) with a measured stiffness of 0.057 N m⁻¹ and an inverse optical lever sensitivity (InvOLS) of 42 nm V ⁻¹ in the aqueous solutions was used for the experiments [45]. The cantilever had a triangular shape with 13.4 μm width of each leg, 100 μm total length and 0.4 μm thickness.

Gram-positive bacteria M. lysodeikticus (Sigma-Aldrich # M3770) and Gram-negative bacteria P. fluorescens (ATCC # 11150) were grown in Trypticase Soy Broth (BD # 211768) and Difco™ Nutrient Broth (BD # 234000), respectively, for at least 24 h at 37 °C. The suspensions of bacteria were purified by centrifugations at 1000xg for 5 min, and then the bacteria were resuspended in Millipore water. For bacteria immobilization we used mica coated with poly-l-lysine (PLL, Sigma-Aldrich #P4707) −50 μl of a sterile 0.01% PLL solution was incubated on freshly cleaved mica (EMS #71851-05) at room temperature for 30 min and then washed with Millipore water. Then 50 μl of each suspension of bacteria in water was adsorbed simultaneously onto adjacent spots on the PLL-coated mica for 15 min, followed by washing with copious amounts of Millipore water.

A closed fluid cell (Asylum Research) consisting of a round glass slide screwed into a poly(ether ether ketone) PEEK holder with a Viton rubber O-ring was used for all imaging. The mica sample was epoxied onto the glass slide, and 2.5 ml of liquid was placed into the fluid cell. The metal stage was separated from the tip electrode by approximately 1.5 cm of Millipore water (18 MΩ cm) and 4 mm PEEK (10¹⁰ MΩ cm [46]).

The same sample and same tip were used for acquisition of all the data presented here except for in figure 1(b) (a full BEPFM image set of M. lysodeikticus in DPBS), for which the same type of sample and tip were used. Slow scan disabled voltage steps in DPBS used the same sample and tip as all other data. BEPFM responses are typically similar for multiple cells of the same type imaged with the same tip, as averaged spectra from a group of ML on PLL in water exhibit similar variation to that of a single ML bacterium in water. Repeatability of BEPFM in liquid was revealed by slow scan disabled images of ML in water, which retained characteristic spectra with repeated scanning, although some instrumental drift or movement of the soft sample resulted in large standard deviations across any single column of voxels. The bacteria sample was prepared three days prior to BEPFM imaging in Millipore water. For the first 24 h, the bacteria were immobilized on the PLL surface in Millipore water, and for the next 48 h they were in Dulbecco's phosphate-buffered saline (DPBS). Then, immediately before imaging in the open fluid cell, the sample was washed with copious amounts of Millipore water and imaged in 2.5 ml Millipore water for several hours. The same day the sample was rinsed with copious amounts of DPBS and imaged in 2.5 ml DPBS for 1.5 days, after which the sample was rinsed with copious amounts of Dulbecco's modified Eagle's medium (DMEM) and imaged for 2 days in 2.5 ml DMEM.