Time-resolved coherent anti-Stokes Raman scattering microscopy: Imaging based on Raman free induction decay
ABSTRACT A time-resolved coherent anti-Stokes Raman scattering (CARS) microscope allows three-dimensional imaging based on Raman free induction decay of molecular vibration with no requirement for labeling of the sample with natural or artificial fluorophores. A major benefit of the technique is the capability to completely remove nonresonant coherent background signal from the sample and the solvent, and thus increasing the detection sensitivity of CARS microscopy significantly. © 2002 American Institute of Physics.
Time-resolved coherent anti-Stokes Raman scattering microscopy:
Imaging based on Raman free induction decay
Andreas Volkmer,a)Lewis D. Book,b)and X. Sunney Xiec)
Harvard University, Department of Chemistry and Chemical Biology, 12 Oxford Street, Cambridge,
?Received 1 November 2001; accepted for publication 10 January 2002?
three-dimensional imaging based on Raman free induction decay of molecular vibration with no
requirement for labeling of the sample with natural or artificial fluorophores. A major benefit of
the technique is the capability to completely remove nonresonant coherent background signal from
the sample and the solvent, and thus increasing the detection sensitivity of CARS microscopy
significantly. © 2002 American Institute of Physics. ?DOI: 10.1063/1.1456262?
Recent advances in nonlinear coherent microscopy, such
as second-1–3and third-harmonic generation microscopy,4,5
microscopy6–9have attracted much interest. Among coherent
microscopies, CARS provides vibrational information intrin-
sic to and characteristic of chemical species. CARS is a
third-order nonlinear optical process involving three laser
beams ?Fig. 1 inset?, the pump, Stokes and probe beam with
frequencies at ?P1, ?S, and ?P2, respectively, which inter-
act with the sample and generate an anti-Stokes field at a
frequency ?AS??P1??P2??Sthat is higher than the exci-
tation frequencies. Therefore, CARS can be detected in the
presence of one-photon induced fluorescence. The anti-
Stokes signal is resonantly enhanced when the Raman shift,
?P1??S, coincides with the frequency of a Raman-active
molecular vibration, which provides the intrinsic vibrational
contrast mechanism.10Vibrational imaging based on CARS
has been previously demonstrated to be more sensitive than
infrared and confocal Raman microscopy techniques,6and
has three-dimensional sectioning capability and deep pen-
etration depth, similar to other nonlinear microscopy. Since
the molecules remain in the electronic ground state, pho-
tobleaching and damage to delicate biological samples is
minimized. However, CARS detection is not background
free. Electronic contributions to the third-order susceptibility
from the sample and solvent cause a nonresonant back-
ground signal, which provides no vibrational contrast. In ad-
dition, the solvent water has strong resonant signals of broad
spectral width. Both background signals often overwhelm
the CARS signal from small scatterers, and limit the sensi-
tivity. Recently, we have reported significant advances in the
sensitivity and spectral resolution of CARS microscopy and
its application to the imaging of live unstained cells.7–9
Here we use time-resolved CARS ?T-CARS?11to record
the Raman free induction decay ?RFID? of molecular vibra-
tions. This not only provides spectroscopic information in
the time domain, but also an alternative approach to separate
the nonresonant contribution of RFID, which is instanta-
neous, from the resonant contribution of RFID12in CARS
microscopy. The T-CARS experiment involves three incident
electric fields, Em(r,t), with frequencies at ?m
?P1,P2, and S? that induce a third-order nonlinear polariza-
tion, P(3)(r,t) at ?AS. Typically, the pair of the first pump
pulse, EP1(r,t), and the Stokes pulse, ES(r,t), are tempo-
rally overlapped, and impulsively polarizes the sample. The
third probe pulse, EP2(r,t), then interacts with the sample at
a certain delay time, ?, with respect to the previous pulse
pair, and probes the relaxation of the induced polarization.
A detailed account on the spatial field distribution of
tightly focused incident fields and the signal generation in
CARS microscopy has been given elsewhere.7,13Here, the
time-dependent field envelopes of the incident pulses and of
the coherent anti-Stokes field are assumed to be identical for
each point within the focal volume.
When electronic dephasing is much faster than either the
nuclear dynamics of the molecular system or the optical
pulses, the time-resolved CARS signal can be written as,14
a?Present address: 3. Physikalisches Institut, Universita ¨t Stuttgart, Pfaffen-
waldring 57, 70550 Stuttgart, Germany.
b?Present address: BlueLeaf Networks, Sunnyvale, CA 94085.
c?Electronic mail: email@example.com
FIG. 1. Schematic of the time-resolved CARS microscope. Insert: Energy
diagram of the four-wave mixing process with three incident fields. PC,
pulse compression stage; VD, variable delay line; Pol, polarizer; and BC,
dichroic beam combiner.
APPLIED PHYSICS LETTERSVOLUME 80, NUMBER 94 MARCH 2002
15050003-6951/2002/80(9)/1505/3/$19.00© 2002 American Institute of Physics
The T-CARS experiment probes the correlation function of
the linear polarizability response, ?(t2), of the sample,
which in the case of parallel polarization of the three incident
and the anti-Stokes fields takes the following form:15–17
Here, the delta function ?scaled by Anr? represents the
nonresonant ?electronic? response of the system while the l
Raman-active vibrational modes are modeled as damped ex-
ponentials of amplitude Al
?10?2?l/2?c (in cm?1), vibrational dephasing time T2l,
rotational correlation time Tor,l, and depolarization ratio ?l.
These parameters can be independently obtained by record-
ing the corresponding spontaneous polarization-controlled
Raman spectra of the sample, ? ˜?(? ˜) and ? ˜?(? ˜), by comput-
ing their isotropic and anisotropic components, ? ˜iso(? ˜)
?? ˜?(? ˜)?
sequent decomposition into l isotropic and anisotropic
Lorentzian Raman line profiles with w ˜l
maximum ??FWHM? in cm?1? values, respectively.
Numerical simulation of the T-CARS measurement is
carried out using Eqs. ?1?–?3?. The only parameter that is
varied is Anr. The time-dependent field envelopes of the in-
cident laser pulses are assumed to be Gaussians, Em(t)
?exp(?2ln2t2/(?m)2). The FWHM of the laser pulses, ?m,
the laser pulse center wavenumbers, ? ˜m, and the spontane-
ous polarization Raman spectrum of the sample are obtained
from independent experiments.
The experimental layout of the RFID microscope is
schematically depicted in Fig. 1. A regeneratively amplified
Ti:Sapphire laser system that pumps an optical parametric
amplifier ?OPA? ?RegA 9000/OPA, 9400, Coherent Inc.? at a
repetition rate of 250 kHz was used to provide synchronized
femtosecond pulse trains at three different wavelengths that
are given by the fundamental (?1) of the Ti:Sapphire, the
signal (?2), and the idler (?3) output of the OPA. The dif-
ferent frequencies of the probe and pump beams avoids the
interference between the two beams in the colinear geometry.
Before being focused, all fields were independently ex-
panded to a beam diameter that matches the back aperture of
the objective lens ?Nikon Plan Apo oil, 60? NA?1.4? lin-
early polarized along the x axis, collinearly overlapped, and
coupled into an inverted optical microscope ?Nikon TE 300?.
The FWHM values of the laser pulses were ?1?185 fs,
?2?85 fs, and ?3?115 fs as obtained by intensity autocor-
relation measurements, with corresponding transform limited
spectral band widths of 80, 173, and 128 cm?1, respectively.
Two variable optical delay lines were used to control the
temporal overlap and delay between the different pulse
trains. The anti-Stokes signal was parfocally collected with
an identical objective lens in the forward direction, spectrally
isolated using interference band pass and holographic notch
aniso, center wavenumber ? ˜l
3? ˜?(? ˜) and ? ˜aniso(? ˜)?? ˜?(? ˜), and by their sub-
?1)/100?cT2,lbeing the full width of half
filters, passed through an analyzing polarizer, and detected
Canada?. Images were collected by raster scanning the
sample, respective to the fixed laser beams.
To demonstrate the ability of the T-CARS microscope to
record RFID, measurements on neat benzaldehyde have been
performed. Figure 2?a? displays the parallel and perpendicu-
lar polarized spontaneous Raman spectrum with respect to
the linear polarized incident field, which was taken with a
commercial confocal Raman microscope ?LabRam, Dilor?.
The Raman bands at 1585 and 1598 cm?1are assigned to the
two quadrant stretch components of CvC bonds in the ben-
zene ring, whereas the 1700 cm?1band reflects the noncon-
jugated CvO stretch vibration.18Their depolarization ratios
amount to 0.43, 0.53 and 0.33, respectively. Figure 2?b? dis-
plays the RFID measurement of benzaldehyde when tuned to
a Raman shift centered at 1636 cm?1. Due to the broad spec-
tral width of the pump and Stokes fields, the superposition of
all three vibrational resonances, shown in Fig. 2?a?, are co-
herently excited and contribute to the measured RFID of
benzaldehyde. An initial fast decay that follows the instru-
mental response, a multi-exponential decay, and superim-
posed quantum beats with period times at ?340 fs and
?2500 fs are observed. The Fourier transform of the oscil-
latory contributions to the measured decay curve into the
frequency domain ?see inset in Fig. 2?b?? reveals two distinct
difference frequencies at 13 and 97 cm?1, which can be as-
signed to the beating of the 1598 cm?1mode with the
1585 cm?1and 1700 cm?1mode, respectively. The numeri-
cal simulation of RFID is also depicted in Fig. 2?b?, satisfac-
FIG. 2. Information content of RFID measurements demonstrated for neat
benzaldehyde. ?a? Parallel ?? ˜
dashed line? components of the spontaneous Raman spectrum with respect
to the linear polarized orientation of the incident field; ?b? Measured
?circles? and simulated ?line? RFID decay curves. The instrumental response
function measured solely from the nonresonant coherent radiation originat-
ing from the glass substrate is also shown ?squares?. Insert: Fourier trans-
form of the oscillatory contributions to the measured RFID. ??P1??1
?800.0 nm, ?S??3?920.5 nm, ?P2??2?715.6 nm, corresponding to a
Raman shift centered at 1636 cm?1with average powers of each ?50 ?W?.
?(? ˜), solid line? and perpendicular ?? ˜?(? ˜),
1506Appl. Phys. Lett., Vol. 80, No. 9, 4 March 2002Volkmer, Book, and Xie
torily reproducing the characteristic features of the observed
decay curve. The temporal resolution of the T-CARS micro-
scope is given by the instrumental response function ?IRF?
that has been independently measured by detecting the solely
nonresonant coherent radiation originating from the glass
substrate. The FWHM of the IRF amounts to 200 fs.
Next, we apply the technique to demonstrate vibrational
imaging of a 1-?m diameter polystyrene bead that is embed-
ded in water. First, the spontaneous Raman depolarization
spectra were taken, with the parallel spectrum shown in the
inset of Fig. 3?a?. The spectrum is decomposed into a series
of Lorentzian line profiles, which correspond to the aromatic
CH stretching modes of the benzene ring at 3035, 3051, and
3061 cm?1.18Figure 3?a? reveals the RFID when focused
into the bead and tuned to a Raman shift centered at
3054 cm?1, and therefore coherently exciting all the vibra-
tional modes shown in the inset of Fig. 3?a?. The measured
RFID curve exhibits an initial fast decay of the IRF, followed
by a single exponential decay with a time constant of about
?390 fs, which is superimposed with a quantum beat that
recurs at ??1280 fs. The fastest feature in the time domain,
the decay constant of 390 fs, corresponds to the broadest
feature in the frequency domain, the line width (FWHM
?15.4 cm?1) of the 3035 cm?1resonance, which gives
T2/2?345 fs and appears to be the dominant contribution to
the observed RFID curve. The beating of the 3035 cm?1
mode with the 3061 cm?1mode with a difference frequency
of 26 cm?1accounts for the observed quantum beat period
of ?1280 fs. More quantitatively, the numerical simulation
?Fig. 3?a?? satisfactorily reproduces the experimental data.
Repeating this experiment in bulk water results in a de-
cay curve that solely resembles the IRF. This indicates that
the observed signal arises purely from the nonresonant
and/or spectrally very broad resonance contributions of water
in the 3000–3100 cm?1range. Figure 3?b? displays ultrafast
vibrational images of a polystyrene bead at zero-time delay
and at ??484 fs. Comparison of the two images clearly re-
veals that the time-delayed detection is capable of com-
pletely removing the nonresonant background contributions.
As such, the solvent background signal is suppressed by a
factor of ?570, with the remaining background signal in the
time-delayed image being limited by the dark count level of
the detection system only. In contrast, the maximum signal
from the polystyrene bead at time zero is reduced by only a
factor of 50. Consequently, the signal-to-background ratio,
S/B, increases from S/B???0??3 to S/B???484 fs??35. In
this way, the vibrational contrast in the time-delayed image
arises exclusively from the Raman-resonant modes. The lack
of interference with any nonresonant signal allows one to
quantitatively relate the recorded image pixel intensity to the
squared concentration of the vibrational modes within the
In conclusion, the T-CARS microscope is capable of
completely removing nonresonant background signal from
the sample and the solvent, and thus significantly increases
the detection sensitivity of coherent Raman microscopy.
Time-domain experiments on vibrational imaging are com-
plimentary to their analogous frequency domain experiments
of chemical and biological systems.
The authors thank Professor R. M. Hochstrasser for
stimulating discussions, Dr. E. J. Sanchez and Dr. J. X.
Cheng for technical assistance with the data acquisition soft-
ware. This work is supported by a NIH grant ?GM62536-01?
and a Harvard start-up fund. One of the authors ?A.V.? ac-
knowledges support from the Deutsche Forschungsgemein-
1J. N. Gannaway and C. J. R. Sheppard, Opt. Quantum Electron. 10, 435
2P. J. Campagnola, M. D. Wei, A. Lewis, and L. M. Loew, Biophys. J. 77,
3L. Moreaux, O. Sandre, and J. Mertz, J. Opt. Soc. Am. B 17, 1685 ?2000?.
4Y. Barad, H. Eisenberg, M. Horowitz, and Y. Silberberg, Appl. Phys. Lett.
70, 922 ?1997?.
5M. Mu ¨ller, J. Squier, K. R. Wilson, and G. J. Brakenhoff, J. Microsc. 191,
6A. Zumbusch, G. R. Holtom, and X. S. Xie, Phys. Rev. Lett. 82, 4142
7A. Volkmer, J.-X. Cheng, and X. S. Xie, Phys. Rev. Lett. 87, 023901
8J.-X. Cheng, A. Volkmer, L. D. Book, and X. S. Xie, J. Phys. Chem. B
105, 1277 ?2001?.
9J.-X. Cheng, L. D. Book, and X. S. Xie, Opt. Lett. 26, 1341 ?2001?.
10Y. R. Shen, The Principles of Nonlinear Optics ?Wiley, New York, 1984?.
11A. Laubereau and W. Kaiser, Rev. Mod. Phys. 50, 607 ?1978?.
12F. M. Kamga and M. G. Sceats, Opt. Lett. 5, 126 ?1980?.
13J.-X. Cheng, A. Volkmer, and X. S. Xie, J. Opt. Soc. Am. B ?to be pub-
14S. Mukamel, Principle of Nonlinear Optical Spectroscopy ?Oxford Uni-
versity Press, New York, 1995?.
15R. F. Loring and S. Mukamel, J. Chem. Phys. 83, 2116 ?1985?.
16D. McMorrow and W. T. Lotshaw, Chem. Phys. Lett. 174, 85 ?1990?.
17W. Li, H.-G. Purucker, and A. Laubereau, Opt. Commun. 94, 300 ?1992?.
18D. Lin-Vien, N. B. Colthup, W. G. Fateley, and J. G. Grasselli, The Hand-
book of Infrared and Raman Characteristic Frequencies of Organic Mol-
ecules ?Academic, San Diego, 1991?.
FIG. 3. Temporally and spatially resolved CARS signals from a 1-?m di-
ameter polystyrene bead embedded in water. ?a? Measured and simulated
?line? decay curves when focused on the bead ?circles? and into bulk water
?squares?, and the parallel component of the spontaneous Raman spectra
with its decomposition into Lorentzian line profiles ?inset?; ?b? RFID images
and the lateral intensity profiles along the white line at time zero and at
??484 fs, demonstrating the complete removal of non-resonant background
contributions from the sample and solvent to the image contrast in the latter.
The image size amounts to 200?70 ?pixel?2with an integration time of
4.88 ms per pixel. ??P1??1?714.6 nm, ?P2??2?798.1 nm, ?S??3
?914.1 nm, corresponding to a Raman shift centered at 3054 cm?1with
average powers of each ?43 ?W?.
1507Appl. Phys. Lett., Vol. 80, No. 9, 4 March 2002Volkmer, Book, and Xie