1. Introduction

State of the art microscopy of biological materials has to reveal structures with nanometer dimensions. Due to the diffraction limit (≈λ/2), conventional microscopy would require short wavelengths (λ) in the ultraviolet to X-ray involving potentially damaging photon energies. New imaging techniques using fluorophors have been developed to beat the diffraction limit such as 4Pi-microscopy. A resolution of less than 100 nm has been obtained, allowing the observation of mitochondrial compartments in a yeast cell [1]. Fluorescence imaging techniques can also follow the path of a single virus penetrating into a cell [2].

In many applications, a non-invasive, non-destructive label-free imaging technique is desired to obtain an unaltered view of molecular processes. Chemical microscopy identifies label-free molecules and materials by spectral, and therefore chemical, fingerprints in the infrared region (IR). IR spectroscopy allows the analysis of biochemical components and spectroscopic changes during distinct cell division cycles [3]. To obtain nanometer chemical information with optical lasers (see Ref. [4] for sub-15 nm resolution using the Raman effect) and for infrared light sources with even longer wavelengths from 3 to 11 µm, near-field methods are needed.

Scanning near-field infrared microscopy (SNIM), with nanometer resolution, combines the advantages of IR spectroscopy with high resolution atomic force microscopy (AFM). Further in-depth information on the development of scanning near-field microscopy can be found in the following reviews by Courjon and Bainier [5], Dragena and Leone [6], Keilmann and Hillenbrand [7]. In addition, the feasibility of sub-surface nanoscale imaging was previously demonstrated [8, 9, 10]. Very recent work provides a theoretical formalism to model near-field contrast and line shapes [11, 12] (and references within).

Previously, using an all solid state, tunable, table-top opto-parametric oscillator (2500–4000 cm^-1), we demonstrated chemical microscopy with 30 nm lateral resolution [10]. With a liquid nitrogen-cooled, tunable CO-laser (1500–2000 cm^-1) [13], we recorded the IR spectrum in the amid region of a biotin monolayer supported on a gold surface [14]. A detection limit of 30000 molecules corresponding to 5 attoliter was achieved.

In this paper, we present measurements of supported lipid bilayers and vesicles on silicon and mica. Especially, the sample preparation on mica led to controlled single bilayers and bilayer stacks of different heights. The contrast as a function of lipid height allowed to analyze the contributions of sample and substrate, lipids and mica, respectively.

2. Experimental

The set-up of our SNIM is shown in Fig. 1(a). The CO-laser power of up to 5 W is attenuated to approximately 100 mW at the entrance of the telescope. The IR light is focused by an off-axis parabolic mirror with a 30 µm diameter onto the metallized tip (Fig. 1(b)) of an AFM cantilever (curvature radius: R=40 nm). A topographic and near-field image is simultaneously recorded by lateral scanning of the sample while the tip is oscillating at a frequency of f=170 kHz. The amplitude modulated scatter light is collected with a CaF₂ lens and focused on a mercury cadmium telluride (MCT) detector using phase-sensitive, lock-in, detection. The near-field signal increases non-linearly as a function of the sample-tip distance [11], whereas the scatter light, e.g. from the shaft of the cantilever, shows a weaker non-linearity. Therefore, we recorded the signal at the first harmonic of the oscillation frequency (2f=340 kHz) to considerably reduce topographic and background artifacts (cf. Ref. [7]). Further details of our set-up can be found in the references [10, 14].

 

Fig. 1. (a) Set-up of the grating tunable CO-laser illuminated SNIM. The IR scatter light from the tip-sample system is collected with a CaF₂-lens and MCT-detector. A HeNe-laser is used for optical alignment. (b) Illustration of the 5 nm thick bilayer and stack formation by fusion of multilamellar vesicles to a substrate. (c) Topography image of a lipid patch with 4 bilayers (20 nm height) and 3 additional bilayers in a smaller patch on top corresponding to 35 nm total height above the mica substrate.

Download Full Size | PDF

Our preparation of dimyristoylphosphatidylcholine (DMPC) vesicles is detailed in Ref. [15]. The vesicles were diluted in distilled water and filtered with a 0.1 µm filter. The vesicle solution (5 µl of 0.16 mg/ml), including 5×10¹⁰/ml gold particles of 30 nm diameter, were deposited on either a 5×5 mm² freshly cleaved mica substrate, an optically polished or Au-coated Si wafer. Mica is a typical AFM-substrate since it can be easily cleaved to provide a ultra-flat surface. Mica has a very low near-field signal while Au has a very high value due to its polarizability which we have confirmed by approach curves to the pure substrates. The near-field signals for each material change marginally in the investigated frequency range providing a minimum-maximum scale after normalization from 0 to 1. In-between we find the investigated DMPC as a function of frequency.

3. Results and discussion

We recorded IR spectra of DMPC multilamellar vesicles adsorbed on Au-coated Si and pure mica substrates by Fourier-transform infrared (FTIR) spectroscopy between 700 and 8000 cm^-1 using either grazing incidence (GI) or attenuated total reflection (ATR). The spectra show characteristic absorption bands, the carbonyl stretching mode at 1739 cm^-1 and absorption features due to incorporated water in vesicles (for GI). From the spectra, we derived [16, 17] the components of the complex refractive index n̂=n+ik (Fig. 2(d)). The imaginary part, k, was calculated from the FTIR absorbance of adsorbed multilamellar vesicles on Au and Ge which was used in ATR. In the GI case, the sample contained residual water similar to the samples studied by SNIM. A k-value of 0.12 at 1739 cm^-1 was used for the calculation [16]. The refractive index, n, varies for the different measurements and was estimated by using Kramers-Krönig relation with the DMPC refractive index at infinity n _∞≈1.5 [17].

 

Fig. 2. (a) Topography of DMPC bilayers stacks and vesicles on Si. (b) Marked section of (a) showing mainly the substrate Si (black). (c) Height, h, profile showing ≈1 µm thick lipid coverage. (d) Near-field contrast, experimental C and theoretical K, of DMPC on Si as a function of laser frequency, ν. The error bar represents the maximum error of the whole data set (connecting line as a guide to the eye). The FTIR absorption is shown as absorbance A with the refractive index n for both types of measurement (ATR in blue and GI in red).

Download Full Size | PDF

With SNIM we investigatedDMPC bilayer stacks and vesicles on Si. The sample preparation led to thick lipid coverage up to 1 µm. Near-field data, evaluated in each image, was scaled to the values obtained in a region where we expect uncovered Si (Fig. 2(b)–(c)). The Si dielectric constant does not vary significantly in the investigated frequency range serving as constant reference. The contrast, C=(I_DMPC-I_Si)/I_Si, was calculated from measured intensities of DMPC and uncovered Si, I_DMPC and I_Si, at each laser frequency. The frequency dependence of the near-field contrast is found to resemble mostly the dispersion of the carbonyl absorption at 1739 cm^-1 (Fig. 2(d)) in accordance for thick samples (cf. Ref. [12]). We have calculated the contrast K similar to the formalism of Ref. [11] calculating the effective polarizability and demodulating the s ₂ with Fourier decomposition. The result here takes not into account the Fresnel factors or complex multilayers as investigated recently [11, 12].

Other features are attributed to the residual water content within intact vesicles and bound water in the layers shown in the difference of the ATR and GI measurement. The contrast is due to the interplay of substrate, sample and tip geometry, and the materials dielectric constant as outlined by Aizpurua et al. [12]. Our set-up uses non-interferometric detection [18], so may obtain a mixture of absorption and dispersion-like frequency dependence, therefore accounting for the broader peak.

Sample preparation of DMPC on mica led to controlled formation of single bilayers and stacks of up to 25 bilayers. Images were recorded at various frequencies (Fig. 3). A topographic image is shown in Fig. 3(b). Figure 3(c) shows the simultaneously recorded near-field image at 1744 cm^-1, close to the maximum absorption of DMPC. The raw data in Fig. 3(b)–(c) was corrected by subtracting a linear function in each line, the “flatten function” in the imaging software WSxM [19]. The smallest resolved bilayer stacks had lateral dimensions of approximately 100 nm. Single Au nanoparticles with a size of 30 nm (≈λ/200) were found at various positions, mostly at lipid patch edges. Figure 3(a) shows the contrast as a function of topographic height, h, at 1744 and 1754 cm^-1. The contrast C=(I_DMPC-I_MICA)/(I_Au-I_MICA) is defined, as previously mentioned to form a frequency independent scale from 0 to 1, with the measured lock-in voltage amplitude of the lipids, mica and 30 nm Au particles, i.e. I_DMPC, I_MICA and I_Au, respectively. The DMPC contrast varies from 0.1 to 0.6 (Fig. 3(a)) between the value of mica and Au, i.e. C(I→I_MICA)→0 and C(I→I_Au)→1. Selected lipid patches at different constant height were analyzed, and standard deviations in topography and near-field signal calculated. For simplicity, the size of the data points reflects only the maximum standard deviation of all points in the whole data set.

 

Fig. 3. (a) Near-field contrast, C, of DMPC referenced to Au and mica as a function of height, h, and bilayer number (each 5 nm thick) for different frequency. The contour line encloses the most frequent data pairs of topography height and near-field contrast at 1744 cm^-1 (orange). The 2f-sensitivity limit (C≈0.15) corresponds to 3 bilayers. Selected bilayer patches are given as data points. The blue line represents the contrast calculated from the cross-section σ (see inset). The inset shows Au nanoparticles visible at 1658 cm^-1 due to low DMPC contrast. (b) Topography of DMPC stacked bilayers on mica (5×5 µm²). In the color scale the maximum topography corresponds to 128 nm (noise level: 2 nm). (c) Near-field image measured at 1744 cm^-1 with a maximum voltage of 138 mV (noise level: 20 mV). Selected image areas (1–10) are shown in a bar diagram.

Download Full Size | PDF

The frequency of the near-field image data at 1744 cm^-1 as a function of corresponding topography height was counted to obtain a two-dimensional histogram. The shown contour line encloses the most frequent data in the histogram (Fig. 3(a)). The measured 2f-lock-in voltages for areas, numbered from 1 to 10, are shown in a bar diagram (Fig. 3(c)) corresponding to selected 30 nm Au particles, DMPC patches and mica substrate. The measured voltages were used to calculate the contrast.

The near-field interaction can be approximated by considering the tip as a spherical dipole induced by the infrared electric light field [14, 20]. Recent work has extended the theoretical modeling to extended dipoles and multilayers [11, 12]. The inset in Fig. 3(a) shows the calculation of the near-field cross-section σ=σ(z)-σ(z→∞) at the carbonyl band as a function of the tip distance and subtracting the far-field contribution [20]. For a tip with R=40 nm, we expect significant near-field contributions up to a distance of approximately 2R, i.e. 80 nm. It was previously shown that for a lateral resolution of <120 nm, a material contrast can be obtained for objects >80 nm deep below the surface [8]. In this study, we were able to evaluate the contrast as a function of the depth variation of the dielectric constant due to layer by layer build-up of lipids into stacks. With increasing layer thickness, the penetration depth, 2R, probes an increasing number of lipid layers, and less and less of the mica substrate. We obtain a larger contrast for lipids relative to mica but a decreasing contrast relative to gold. The contrast tends to saturate at 0.6, at approximately 3R or 120 nm. We calculated the contrast from the theoretical cross-sections (σ) of DMPC, Au, and mica at 1744 cm^-1 using an average dielectric constant

with weighing factors 2R-h and h, for mica and the respective DMPC layer height. The calculated curve is in good agreement with the main weight of the data (blue line in Fig. 3(a)).

The non-interferometric set-up measures the intensity |E_B+E_N|² with E_B, the background component due to direct reflection at the cantilever shaft (cf. Fig. 1(b)), and E_N, the near-field component of the electric field. The background intensity |E_B|² marginally contributes at a 2f-demodulation. The near-field contribution |E_N|² of mica and a few DMPC bilayers is small. Aizpurua et al. [12] suggest only a few percent contrast change for increasing layer thicknesses of a polymer. The signal values in our application may indicate that the interference term proportional to |E_B||E_N| is beneficial for contrast formation although difficult to interpret. For single bilayers it would be desirable to deposit on an Au-coated substrate to benefit from the possible high contrast (Fig. 3). Au nanoparticles are expelled from lipid membranes so that they are mostly found at the edges of bilayers. This behavior is related to the unfortunate tendency to form vesicles on Au surfaces so that new sample preparation techniques are required.

4. Conclusion

We have demonstrated chemical imaging of lipid bilayer stacks. Patches of 100 nm have been resolved at the peak absorption of the carbonyl band with high contrast. The limit sensitivity was reached at 3 bilayers corresponding to 1 attomole. The penetration depth of more than 100 nm is determined by tip curvature and DMPC polarizability. The contrast of DMPC on various substrates was evaluated and the contributions of the substrate mica and the lipids were calculated and separated. The latter is especially important for future investigations of thin samples with a thickness of less than 120 nm or complex biological systems, like cell membranes.

The results show that SNIM provides a non-invasive technique and that cell membrane investigations are feasible. The phase transitions of lipid mixtures, for example by the addition of cholesterol inserting itself into the bilayer [15], will be interesting future targets. SNIM does not require a labeled molecule thus preventing any possible alteration of the molecule’s natural response. Due to the strong IR absorption of water, the next major challenge will be the study of membranes in physiological solutions requiring high power IR lasers. Understanding the processes of well-defined artificial membrane systems mimicking cell membranes will have immediate medical impact; for example, high concentrations of cholesterol leading to the formation of cholesterol crystals, considered a hallmark of advanced atherosclerotic plaque [21].