Skip to content
Skip to navigation menu


Dr Iestyn Pope 

Coherent Anti-Stokes Raman Scattering (CARS)

CARS energy level diagram

In CARS, two near infra-red laser beams of different frequencies (called “pump” (ω1) and “Stokes” (ω2)) are used to drive the molecular vibrations at the frequency ωD = ω1 - ω2 (where, ω1>ω2).  By adjusting the frequency of the Stokes beam the driving frequency can be tuned to specific vibrational resonance (ωD = Ω).  A third beam (ω3) is then used to generate the emission of a photon (ω4) at the frequency ω4 = ω1 - ω2 + ω3, which is detected.  Usually ω1 and ω3 are chosen to be degenerate therefore, ω4 = 2ω1 - ω2 = ω1 + Ω.  Hence ω4 is the anti-Stokes Raman frequency relative to ω1 .  Since all bonds within the focal volume with a vibrational resonance equal to Ω are drive coherently, the anti-Stokes Raman emission constructively interferes with itself resulting in a non-linear increase in signal.  Since the CARS signal scales non-linearly with the number of bonds within the focal volume, CARS is ideally suited to imaging compounds with a large number of similar bonds within their structure, such as lipids, which contain a large number of C-H bonds within their acyl chains.

Right - CARS energy level diagram.  E1 and E0 are electronic levels, with E1 - E0 typically UV for molecules studied with CARS. Ω is the energy separation between vibrational modes, typically mid-infrared. Dashed lines represent virtual energy states.

Our Single Source CARS/TPF/SHG microscope 

We have developed a multimodal multiphoton laser-scanning microscope for cell imaging featuring simultaneous acquisition of D-CARS, TPF and SHG using a single 5 fs Ti:Sa broadband (660-970 nm) laser. The spectral and temporal pulse requirements of these modalities were optimized independently by splitting the laser spectrum into three parts: TPF/SHG excitation (> 900 nm), CARS Pump excitation (< 730 nm), and CARS Stokes excitation (730-900 nm).

CARS Beam Optimisation

After splitting the initial broadband laser pulse up into the pump and Stokes beams each generated femtosecond pulse is still much broader than the typical Raman linewidths (i.e., the duration of pump and Stokes pulses are much shorter than the vibrational coherence time which are on the order of a picosecond).  However, the vibrational excitation in CARS is governed by the interference of the pump and Stokes fields; therefore, the spectral resolution is not determined by these individual spectra, but by the spectrum of their temporal interference.  Hence by carefully shaping the pump and Stokes pulses in time, it is possible to drive a narrow vibrational frequency range, even though the individual pulses are spectrally broad.  We recently demonstrated a simple, highly efficient, alignment insensitive, and cost-effective method which utilizes glass elements of known dispersion (Langbein et al., 2009a; Rocha-Mendoza et al., 2008) through a process known as spectral focusing or chirp.

Spectral Focusing

The refractive index (n) of a dispersive medium is not constant but varies with wavelength, n(lambda) (and thus frequency). This means that when a short pulse travels through a dispersive medium, its different wavelength (or frequency) components travel at different speeds Typically, the longer wavelengths travel faster and thus emerge from the medium first.  This stretches out the pulse from its initial width.  The key point for spectrally selective CARS is to introduce the same chirp parameter in pump and Stokes pulses so that the instantaneous frequency difference (IFD) between the two pulses remains constant.  The molecular vibrations are then driven at the beat frequency (w1-w2) which will be centred at the IFD.  Using glass dispersion to chirp the beams we achieve a CARS spectral resolution of 10 cm1.  Tuning to different Raman frequencies is achieved by simply by adjusting the arrival time of the pump beam relative to the Stokes.  Using this method we are able to acquire CARS images over the spectral range 1200-3800 cm1.

Differential CARS (D-CARS)

D-CARS is implemented with few passive optical elements and enables simultaneous excitation and detection of two vibrational frequencies with a separation adjustable from 20 cm1 to 150 cm1 for selective chemical contrast and background suppression.


Broadband sub-10fs lasers diagram  756.3 Kb

Above - (a) Broadband sub-10 fs lasers produce a broad range of wavelengths, that can be separated (b) into pump and Stokes beams through the use of a dichroic mirror (DM).  (c) Illustration demonstrating that the IFD between two equally chirped pulses is constant with time. (e) Illustration demonstrating how varying the overlap of the two pulses changes the IFD, thus allowing spectral tuning via simple delay scanning. t, time; ω, frequency.


TPF/SHG Beam Optimisation

A prism pulse compressor in the TPF/SHG excitation is used to achieve Fourier limited 30 fs pulses at the sample for optimum TPF and SHG.

Below is a sketch of the microscope set-up. M: mirror; DM: dichroic mirror; SF57: glass blocks; R: reflecting prism; λ /2: half-wave plate; BE: beam expander; PBS: polarizing beam splitter; F: filter. The side view of the optics between by the two indicated arrows shows the beam height difference. Graph: typical spectra of the laser, pump, Stokes and TPE beams.

Sketch of the microscope set-up  889.9 Kb
Mouse tail  197.1 Kb

Above - Mouse tail tissue section fluorescently labeled with FITC-Concanavalin A. Top Left: Stitched epi-fluorescence image, scale bar = 300μm. Bottom Left: Single epi-fluorecence image cropped to 150×150μm (the boxed region shown on the Top Leftimage). Far Right: False-colour image generated from simultaneously acquired TPF, CARS and SHG images.  CARS from subcutaneous lipid deposits at 2850 cm-1 (red) and SHG from collagen (blue), scale bar = 20μm. 150×150μm, 501×501 pixels, 0.01 ms pixel dwell time (2.5 s total image acquisition).

Hyperspectral Imaging

The images below were generated by taking a hyperspectral CARS stack of images at different molecular vibrations.  Through the analysis of this image stack a spectral profile for each pixel in the image may be generated.  Further analysis of these profiles allows a false-colour chemical map may be generated. (red: Lipids, blue: DNA, green: protein)

Image of metastatic colorectal carcinomaImage of mouse intestinal organoid

Left shows a human metastatic colorectal carcinoma imaged live using our CARS microscopy. The 3D culture system used to grow the carcinoma spheres aims to reproduce the architecture of the cancer, allowing us to employ CARS microscopy to investigate the composition of these spheres outside the patient’s body.

Right shows a mouse intestinal organoid, images live using our CARS microscope.  The 3D culture system recapitulates the architecture of the small intestine with crypt and villi structures, comprising both stem cells and differentiated cells.

Arnica KarunaAndrew Hollins, Iestyn Pope, Anika Offergeld, Francesco Masia, Wolfgang Langbein, Trevor DalePaola Borri


Nanoparticles are attracting enormous attention for biomedical applications as optical labels, drug delivery vehicles and contrast agents in vivo.  Nanoparticles made of various types of organic and inorganic materials have been investigated, and recently diamond has emerged as one of the best material choices due to its bio-compatibility and unique structural, chemical, mechanical, and optical properties.  So far, diamond nanoparticles (nanodiamonds) have been visualised optically mainly via the existence of fluorescing ‘defects’ in the crystal lattice. However, the production of these defects is costly and not very well controlled hence limiting their use. Moreover, these defects might irreversibly change their fluorescence properties upon optical illumination and when close to the surface of a nanodiamond they might become unstable.  We have shown that single non-fluorescing nanodiamonds exhibit a strong CARS signal at the sp3 vibrational resonance of diamond.  Using our in-house developed CARS microscope we have measured the CARS signal strength on a series of single nanodiamonds of different sizes. The nanodiamond size was accurately determined by means of electron microscopy and correlative quantitative optical contrast methods developed in-house. In this way, we were able to quantify the relationship between the CARS signal strength and nanoparticle size. The calibrated CARS signal in turn enables us to analyse the number and size of nanodiamonds internalized in living cells in situ. Owing to the high bio-compatibility of nanodiamonds, this imaging modality opens the exciting prospect of following complex cellular trafficking pathways quantitatively.

3D CARS image of a HeLe cell  307.5 Kb

Above - A 3D CARS image of a HeLa cell that has taken up nanodiamonds.&#160; The image consists as an overlay of the CARS signal at the nanodiamond resonance (1332 cm-1) and the CARS signal from intracellular structures (2900 cm-1) where the relative intensities have been adjusted to enhance the cellular contrast for visualisation purposes. A false colour map has then been used where the nanodiamonds can be seen as the bright orange spots within the magenta cell.


We have demonstrated the applicability of CARS micro-spectroscopy for quantitative chemical imaging of saturated and unsaturated lipids in human stem cell derived adipocytes. Comparing dual-frequency/differential CARS (D-CARS), which enables rapid imaging and simple data analysis, with broadband hyperspectral CARS microscopy analysed using an unsupervised phase-retrieval and factorization method recently developed within our group for quantitative chemical image analysis.

Through a ratiometric analysis, both D-CARS and phase-retrieved hyperspectral CARS determine the concentration of unsaturated lipids with comparable accuracy in the fingerprint region, while in the CH stretch region D-CARS provides only a qualitative contrast owing to its non-linear behaviour. When analysing hyperspectral CARS images using the blind factorization into susceptibilities and concentrations of chemical components recently demonstrated by us, we are able to determine vol:vol concentrations of different lipid components and spatially resolve inhomogeneities in lipid composition with superior accuracy compared to state-of-the art ratiometric methods.