1. Introduction

Coherent anti-Stokes Raman scattering (CARS) has received much attention as a microscopic method with spectroscopic resolution [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]. It involves the nonlinear dielectric response of matter due to the interaction of three intense electromagnetic fields. In most experimental realizations two laser beams are degenerate in frequency (ω_P) and act as so-called pump beams that are superimposed in the investigated sample volume with a Stokes beam of a lower frequency ω_S. The coherent anti-Stokes Raman signal is then generated at a blue-shifted frequency of ω_AS=2ω_P-ω_S. This signal is resonantly enhanced, if the energy difference of the laser beams ω_P-ω_S coincides with a vibrational transition ω_vib of the medium, representing a Raman resonance. A coherent signal amplification is obtained, if the so called wave-matching condition is fulfilled. There the sum of the wave-vectors of the two pump beams has to be equal to the sum of the Stokes and the anti-Stokes wave-vectors. In dispersive media such as liquids the laser beams have to include a certain angle, and the signal is emitted into a specific direction different from that of the excitation beams.

Recently a CARS setup was demonstrated [12] in which the sample is excited in a wide-field geometry with nanosecond pulses. There the signal is resonantly enhanced by fulfilling the phase matching condition over the whole field of view. This is achieved by an excitation geometry involving an ultra-dark-field condenser with a high numerical aperture range between 1.2 and 1.4. It produces a cone-shaped shell of light, sending the pump light in almost from the side. In the sample volume this generates a narrow ”sheet of light” that limits the axial interaction length of the pump and Stokes-beams to 5 µm, suppressing non-resonant CARS signals from the solvent in the regions below and above the imaged sample plane. Throughout the light-sheet the wave-matching condition is satisfied, which generates a coherently enhanced, resonant CARS signal.

Although using the same physical principle, wide-field and scanning CARS setups are totally different. Scanning systems have a higher spatial resolution and axial sectioning capability, due to their quasi-confocal excitation. They also do not suffer from speckle noise, and they are continuously advanced by sophisticated methods for noise and background suppression, like epi-detection [5], polarization control [8], heterodyning [13], or focus-engineered excitation [14, 15]. On the other hand, wide-field methods also have some interesting features: Excitation can be done by (but is not restricted to [16]) nanosecond laser systems. A bandwidth limited nanosecond system can in principle achieve a spectral resolution below 0.01 cm^-1 (although our present system is not optimized and has only a resolution of 5 cm^-1), exceeding the resolution of pico- or femtosecond lasers which are used for scanning CARS microscopy. Such a high spectral resolution may be beneficial in special cases of material research (e.g. investigations of low temperature systems [17]). Due to the high pulse energy, snapshot imaging is possible with one pair of laser pulses in a few nanoseconds (see inset in Fig. 4), although to date only with low intensity. In the future this might be useful to investigate fast chemical reactions. Conveniently, the wide-field setup uses only standard microscopy components (with the exception of the laser system), i.e. a fluorescence imaging port and a dark-field facility. Compared to scanning CARS experiments, a laser intensity control is not essential, since possible intensity fluctuations do not reduce the image quality. Also, temporal overlapping of the 1 m long light pulses in the sample plane is achieved easily.

First implementations of the method did not demonstrate its potential [12, 18, 19], since the bandwidth of the nanosecond excitation pulses emitted from a then used broad-band optical parametric oscillator (OPO) was rather high, and in the visible spectral range. The large bandwidth of 25 cm^-1 reduced the excitation efficiency, and the excitation wavelength in the visible range produced an undesired background of two-photon fluorescence and luminescence. In the present work we use an optical parametric oscillator (OPO), emitting in the near infrared (NIR) spectral range with a narrower bandwidth of 5 cm^-1, which approximately corresponds to the linewidth of the excited Raman transitions. Implementing near infrared CARS excitation with this OPO yielded a large improvement in signal efficiency, and a significant reduction of the undesired background due to two-photon processes [20].

The present wide-field CARS setup provides a signal contrast which allows one to record CARS images also at weak CARS resonances. The non-resonant CARS background generated by the solvent is for ”pure” systems (e.g. micro-droplets of oil in water) sufficiently low, such that the CARS spectra can be quantitatively evaluated without the need for background subtraction or data post-processing with numerical algorithms [7, 10]. For quantitative investigations of weak CARS resonances in “messy” biological samples, a background correction by subtracting a reference image recorded under off-resonant CARS conditions was performed.

Here the setup is used to differentiate between various vegetable oils at the signatures of the weak =C-H stretching vibration around 3015cm^-1, and to obtain quantitative spectra that measure the ratio between saturated and unsaturated fatty acids in vegetable oils. In the first part, the system is calibrated using micro-droplets of pure saturated and unsaturated vegetable oils, and it is demonstrated that mixtures of different oil micro-droplets can be distinguished.

The setup was then used to image various lipids within adipocyte cell cultures. Different populations of the same cell line had previously been fed on various saturated and unsaturated fatty acids. This ”diet” led to different average concentrations of 5.8% and 9.8% of polyunsaturated fatty acids in the internal fat vesicles, respectively, as measured with a high performance liquid (hpl) chromatograph using centrifugally enriched amounts of liposomes. With the wide-field CARS system it was possible to reproduce the chromatographically obtained results by investigating individual fat vesicles in single cells, despite the low concentration difference of unsaturated fatty acids, and the weak corresponding =C-H CARS resonance.

Although CARS investigations on the fat metabolism of adipocytes have been already reported [21], using deuterated oleic acid for simplifying its distinction from polyunsaturated fish oil, in our experiment we use - to our knowledge - for the first time the weak =C-H Raman resonance at 3015cm^-1 for the chemical distinction in CARS imaging, i.e. it is applicable to unprepared samples in their natural environment. The possibility to selectively measure the concentration of saturated and unsaturated lipids in the liposomes of individual, unstained cells offers new opportunities for the investigation of the metabolism of adipocytes, which is an important topic in biomedical research.

2. Experiment

The setup of the wide-field CARS microscope is sketched in Fig. 1. Two nanosecond laser pulses (3 ns) with controlled incidence angles that satisfy the phase matching condition are homogeneously illuminating the sample. The light pulses are guided to the microscope inputs through multi-mode fibers that act as ”mode scramblers”, and distribute the light homogeneously over the sample. The pump beam enters the sample plane from above through an ultra-dark-field condenser, whereas the Stokes beam comes from below through the objective (Zeiss ACROPLAN 40× water immersion, numerical aperture NA=0.8). In this case the phase matching condition is satisfied if the resulting CARS signal counter-propagates with respect to the Stokes beam back through the objective. Imaging is then done with an intensified CCD camera. The usable field-of-view, where a rather homogeneous CARS signal is emitted is approximately 40 µm broad.

 

Fig. 1. Wide-field CARS microscopy corresponds to a combination of epi-fluorescence imaging with dark-field illumination. The near-infrared Pump (green) and Stokes beam (red) enter the microscope from different directions such that phase matching leads to an anti-Stokes signal beam (blue) travling anti-parallel with respect to the Stokes beam, i.e. back through the objective to the ICCD camera. The insert shows in more detail how the cone of light illumination by the ultra-dark-field condenser is achieved.

Download Full Size | PDF

In the current setup the wavelengths of the laser pulses can be chosen in the near-infrared region, thus reducing the background from two-photon processes. The Stokes pulse (1064 nm) is emitted from a narrow-band Nd:YAG laser (Coherent Infinity, bandwidth <0.03cm^-1), and the pump pulse from a mid-band OPO (GWU Lasertechnik). The spectral resolution of the CARS setup is now limited by the 5 cm^-1 bandwidth of the OPO, rather than by the 25 cm^-1 bandwidth in our earlier experiments [12, 18, 19]. The pulse energies of the pump- and Stokes beams emerging from the fibers were 1 mJ and 1,5 mJ, respectively, and the average power at 10 Hz repetition rate on the whole sample was less than 10 mW for the Stokes beam, and 2 mW for the pump beam due to losses in the dark-field condenser.

Typical image acquisition times are on the order of 1-10 s for images at strong CARS resonances of pure oil drops or polystyrene test beads. In principle the present setup allows to decrease the imaging time by a factor of 10 by increasing the pulse repetition rate of the laser from 10 Hz to its maximal rate of 100 Hz, however this is normally avoided since it increases the risk of optical damage if the beam is accidentally misaligned. The minimal diameter of polystyrene test beads which can be imaged with a reasonable signal-to-noise ratio (SNR) of 2 is currently 500 nm (test beads with 350 nm diameter are just visible with a SNR of 1.2; 700 nm beads have a SNR of 4). The axial sectioning resolution of the setup corresponds approximately to the 5 µm thickness of the ”sheet-of-light” illumination. We hope to be able to improve the axial resolution further by employing structured illumination methods.

In order to measure CARS spectra, the pump wavelength was tuned in a range between 803nm and 823 nm. This corresponds to a tuning range of the Raman shift between 3050cm^-1 and 2750cm^-1, respectively. For obtaining spectra from small regions within extended samples, several CARS images of the whole sample were recorded at equally spaced pump beam wavelengths. Afterwards, the CARS signal intensities within the selected regions of interest were numerically determined from the image sequences.

Table 1. Measured CARS signal intensities of various vegetable oils at 3015 cm^-1 (normalized to the strongest peak of linseed oil). Saturated fatty acids (SFA), mono-unsaturated fatty acids (MUFA), twofold-unsaturated fatty acids (2FUFA), threefold-unsaturated fatty acids (3FUFA), corresponding to stearic acid, oleic acid, linoleic acid, alpha-linoleic acid, respectively, and calculated number of C=C double bonds per milliliter (N/ml).

View Table

In order to evaluate the setup, we recorded spectra of various vegetable oils. These oils are a mixture of saturated and unsaturated fatty acids. Since ”saturated” refers to the situation that all carbons (apart from the carboxylic acid [-COOH] group) contain as many hydrogens as possible, saturated fatty acids do not contain any double bonds. Unsaturated fatty acids have at least one singly-bonded -CH₂-CH₂- part of the chain substituted by a doubly-bonded -CH=CH-portion and are named mono-unsaturated, twofold-unsaturated, threefold-unsaturated and so on, referring to the number of double bonds.

 

Fig. 2. (A) CARS spectrum representing the typical CARS intensity distribution of lipids in the Raman shift region between 2750 and 3050 cm^-1 showing a superposition of several different vibrational states of symmetric and asymmetric CH₂ and CH₃ stretching vibrations. The inset shows the Raman spectrum of linseed oil for comparison. (B) The resonance at 3015 cm^-1 depends on the number of C=C double bonds, since it excites C-H stretching vibrations of a hydrogen attached to a C=C group and therefore varies strongly for different vegetable oils. (C) Dependence of the CARS signal intensity ration between the =C-H peaks at 3015 cm^-1, and the -C-H peak at 2850 cm^-1 on the corresponding ratio of the numbers of C=C double bindings and C-C single bindings.

Download Full Size | PDF

Figure 2(A) shows the measured spectra for various vegetable oils in the Raman shift region between 2750 and 3050cm^-1. Each point in the spectrum represents the total CARS signal of a micrometer sized droplet in an oil-in-water emulsion, normalized by the cross sectional area of the droplet. Each spectrum was acquired in 4 minutes and consists of 50 points, separated by 5 cm^-1. The line profiles from 2750cm^-1 to 2950cm^-1 are a superposition of several vibrational states of symmetric and asymmetric CH₂ and CH₃ stretching vibrations with the maxima at 2850cm^-1 and are thus very similar for each vegetable oil. Hence, a differentiation of vegetable oils in this frequency region is not possible. On the contrary, in the Raman shift region between 2990 and 3050cm^-1 a peak emerges that varies strongly for different vegetable oils, since it is the C-H stretching frequency of a hydrogen attached to a C=C group, and therefore the CARS signal intensities depend on the number of C=C double bonds of the oils. The inset shows a Raman spectrum of linseed oil for comparison with the CARS micro-spectrum. It demonstrates that the CARS signal reproduces the Raman signal without undesired peak-shifts or spectrum deformations, indicating an efficient suppression of the non-resonant background from the solvent.

Figure 2(B) shows a magnification of the CARS spectra in the range between 2990cm^-1 and 3050cm^{-1. The corresponding maximal values of the signal intensities at the peak around 3015cm-1 are listed in Tab. 1, as well as the fractions of saturated and unsaturated fatty acids which were taken from the literature [24]. The last row of the table shows the calculated number of the double bonds per milliliter for the various vegetable oils. Figure 2(C) shows the ratio between the intensities of the ”unsaturated peak” at 3015cm-1 and the ”saturated peak” at 2850cm-1 as a function of the ratio between the numbers of the corresponding bindings. As expected for the non-linear CARS signal, the curve shows a quadratic dependence on the concentration of the bindings.}

 

Fig. 3. (A) Dark-field illumination of various oil droplets inside a multi-component oil-in-water emulsion. (B) CARS signal generated at 2850 cm^-1. (C) CARS signal generated at 3015 cm^-1. The olive oil droplet at the bottom generates a weaker signal than the thistle oil droplet, the latter containing more of the targeted unsaturated fatty acids. (D) Off-resonant CARS signal at 3050 cm^-1. The negative contrast arises from the displacement of the solvent in the illuminated sample slice. The image intensities in (C) and (D) are amplified by a factor of 3 with respect to (B).

Download Full Size | PDF

An advantageous feature of CARS microscopy is that it combines spectroscopy and imaging, generating a spectrum for each sample point in parallel. This allows chemically selective imaging within an extended sample area. An example is presented in Fig. 3, where an emulsion consisting of a mixture of micrometer sized linseed and olive oil droplets in water is spectrally mapped. Picture (A) shows an image obtained with dark-field illumination, where the oil droplets are visible as a result of the refractive index mismatch of water and oil. Image (B) shows a CARS image of the same droplets, recorded at the strong 2850cm^-1 Raman resonance. There both oil droplets deliver a strong CARS signal, the lower droplet having a slightly higher intensity. The signal to water-background ratio at this Raman resonance is 50:1, thus additional background subtraction or image processing is not necessary. The exposure time was 10 sec at a laser repetition rate of 10 Hz corresponding to 100 laser pulses. The relative signal intensities of the two oil droplets change drastically, when tuned to 3015cm^-1 (C). Now linseed oil with a high fraction of polyunsaturated fatty acids shows a strong signal, whereas the signal of olive oil decreases due to a lack of polyunsaturated fatty acids. Analysis of the two CARS images at these two Raman peaks therefore shows that the two drops are of a different oil species, with the upper drop consisting of linseed oil, and the lower one of olive oil. As a control, image (D) presents the CARS signal intensities when the laser frequencies are detuned by 35 cm^-1 to 3050cm^-1. There the CARS signal of the two oil droplet vanishes, as expected, and the signal of the surrounding water becomes dominant. Note that the droplets are now imaged with negative contrast, since they displace the surrounding water. This is an indication that the CARS generation is confined to an optical illumination ”slice” that is smaller than the diameter of a droplet, i.e. it is on the order of some microns. The suppression of the non-resonant CARS signal from the solvent below and above the droplets is due to the axial confinement of the beam overlap region caused by the slanting-angle-illumination from the dark-field condenser.

In the following, the method was applied to measure the uptake of different fatty acids of cultivated mouse adipocytes that were fed on different saturated and unsaturated oils. For this purpose, murine fibroblasts 3T3-L1 cells were cultivated in 5% CO₂ at 37°C. The cells were maintained in Dulbeccos Modified Eagles Medium (DMEM, Gibco,Germany) supplemented with 5 mM glucose, 10% heat-inactivated bovine-serum, 2 mM L-glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin, respectively. The medium was changed every 2 days. Two days after reaching confluence, the pre-adipocytes were treated with a medium to induce differentiation by standard procedures [23]. Briefly, two-day post-confluent cells were supplemented with 0.5 mmol/l 1-methyl-3-isobutylmethylxanthine +1.0 µmol/l dexamethasone +10 µg/ml insulin +10% fetal bovine serum (FBS) for two days. Then the cells were kept for two more days in culture medium with 10 µg/ml insulin +10% FBS, and after this cycle of differentiation the medium was replaced every second day using culture media with 10% FBS. The cells used for experimentation were over 80% differentiated, as determined by light microscopy. All experiments were performed within 10 - 14 days post differentiation. At that time the cells were incubated with different free fatty acids (FFAs) in an overnight procedure.

For test purposes we used two cell populations that were fed on stearic acid (saturated) and on arachidonic acid (polyunsaturated), respectively. Chromatographic measurements on ”macroscopic” amounts of the liposomes of the two cell cultures showed that the cells fed on stearic acid contained 23.9% saturated, 41.4% mono-unsaturated, and 5.5% polyunsaturated fatty acids, whereas the second cell population fed on arachidonic acid contained 23.3% saturated, 37.5% mono-unsaturated, and 9.9% polyunsaturated lipids. The remaining substances were not identified by the chromatograph. The chromatographic experiments were reproducible to within a relative standard deviation below 5%. Although the second cell population has a 3.9% smaller degree of mono-unsaturated acids, it has a 4.4% higher degree of the four-fold unsaturated arachidonic acid. Therefore it is supposed to give a higher signal at the =C-H CARS resonance, as compared to the first cell population.

In order to check whether the two cell cultures can be distinguished by CARS microscopy, we imaged them at the strong resonance of the saturated -C-H bindings at 2850cm^-1, and at the weak peak of the unsaturated =C-H bindings at 3015cm^-1. The results are shown in Fig. 4. Images (A) and (D) show dark field images of two representative cells of the populations fed on stearic acid (A) and arachidonic acid (D), respectively. The next row (B) and (E) shows CARS images of the two cells recorded at the strong CARS resonance at 2850cm^-1. The images were taken with 100 shots from the laser system in 10 s. The inset in (B) shows a single shot image of the same cell, demonstrating that the liposomes can be detected even with one laser pulse corresponding to a total excitation energy of 1 mJ, at the cost of an increased noise level. The last row shows CARS images of the two cells, (C) and (F), that were taken on the weak =C-H resonance at 3015cm^-1, using 600 laser shots in 60 s exposure time. The two images are intensified by a factor of 7 with respect to the images in (B) and (E). Since here the signal is already quite low, all of the displayed CARS images were background corrected by subtracting an off-resonant CARS image that was recorded with a pump beam wavelength that was blue-detuned by 2.3 nm (corresponding to 35 cm^-1) from the 3015cm^-1 resonance.

A quantitative comparison shows that the signal in (F) is by 95% higher than in (C). The ratio between the weak =C-H, and the strong -C-H peaks is measured to be 1.7% and 3.3% for the two cells fed of stearic acid and on arachidonic acid, respectively. Comparing these ratios to the calibration data in Fig. 2(C) shows that the ratio of unsaturated to saturated bindings in the liposomes of the two cells corresponds roughly to 0.065 and 0.047, which is similar to the measured peak ratios in milk (lowest curve in Fig. 2(B)), and in olive oil (second lowest curve in Fig. 2(B)), respectively. However, from the chromatographic data it can be estimated that the density of unsaturated bindings in the cells fed on arachidonic acid should be 22% higher than that in the cell fed on stearic acid, corresponding to a signal increase of approximately 50% (due to the quadratic dependence of the CARS signal on the number of CARS active bindings). This is lower than the experimentally observed increase of 95%, suggesting that the measured cell has consumed more polyunsaturated acid than the ”average” of the cell culture. This is indeed possible: measuring various cells of the different cell cultures, we observed that the cells fed on arachidonic acid separated in two classes, namely in one population that did not show any increase in the =C-H signal strength, and a second class that showed a considerable increase by a factor of two, whereas all cells of the first culture (fed on stearic acid) showed similar signal intensities [25]. This suggests that the uptake of arachidonic acid in the second cell population did not take place in every cell, probably due to the fact that the cells were in different states of cell differentiation when they were fed. This effect is not visible in the cells given stearic acid, since an uptake of this lipid does not considerably change the ”normal” lipid composition within the cell, as confirmed by reference measurements on a ”normally” nourished control group.

 

Fig. 4. Imaging of liposomes in two adipocytes that were previously fed on different lipid oils, i.e. on saturated stearic acid and on polyunsaturated arachidonic acid, respectively. (A) Dark-field image of a cell fed on stearic acid. (B) CARS image of the same cell at the strong -C-H resonance at 2850 cm^-1. The images were recorded within 10 s, integrating over 100 laser pulse pairs. For comparison the inset shows a single shot image of the same cell, recorded within 3 ns. (C) CARS signal at the weak =C-H resonance at 3015 cm^-1. (D) Dark-field image of another pre-adipocyte fed on arachidonic acid. (E) CARS signal of the same cell at 2850 cm^-1. (F) CARS signal of the same cell at 3015 cm^-1. The image intensities in (C) and (F) are amplified by a factor of 7 with respect to (B) and (E). The signal of the second cell at the =C-H resonance at 3015 cm^-1 in (F) is about twice as intense, as the respective signal of the first cell in (C).

Download Full Size | PDF

Since the chromatographic measurements can only give an average over the whole cell culture, they underestimate the amount of arachidonic acid that has been consumed by the active cells. This agrees with our observation of a stronger-than-expected CARS signal from these adipocytes. In fact, this points out an advantage of CARS micro-spectroscopy with respect to ”macroscopic” methods such as chromatography: The CARS images display the variability of the cell metabolism within a population by avoiding to just average over all cells.

3. Conclusion

Chemical mapping of various fatty acids with different degrees of unsaturation is demonstrated by means of CARS micro-spectroscopy. The low non-resonant background of the solvent in wide-field CARS microscopy allows to image particles with strong Raman resonances (such as the -C-H vibration of lipids) without background correction, and even to obtain meaningful snapshot images with a single laser pulse exposure, though at the cost of increased noise. In pure systems, as for instance oil-in-water emulsions, it is possible to record quantitative CARS spectra that are almost free of a solvent background. The good signal-to-noise level also allows chemical mapping of oils not only at the strong Raman resonance at 2850cm^-1, but also at weaker Raman resonances that have - to our knowledge - not previously been used for CARS imaging. Thus various oil droplets of a multi-component oil-in-water emulsion could be distinguished at the weak Raman resonance at 3015cm^-1 due to their different fractions of unsaturated fatty acids. A quantitative distinction of saturated and unsaturated fatty acids could be also demonstrated within biological samples, i.e. in liposomes of cultivated mouse pre-adipocytes. The CARS measurements are in rough quantitative agreement with control measurements done with a HPLC, and they demonstrate that the variability of lipid metabolism can be investigated at the level of individual cells, which is not possible in chromatography. CARS microscopy can deliver spectroscopic information with a high spatial microscopic resolution at average powers below 10mW, allowing to investigate individual living cells. Application of this method in biomedical studies promises new insights into mammalian lipid metabolism and metabolic disease. For example, it can be investigated whether the distribution of lipid vesicles in cells is homogeneous or inhomogeneous. Especially the degree of unsaturation has recently attracted high interest [21], since a diet with poly-unsaturated fatty acids such as omega 3 fatty acids stimulates lipid oxidation and represses lipid synthesis in adipose tissue [22].