Wavelength division scanning for two-photon excitation fluorescence imaging

Authors


Z. Liu. Tel: +1 814 8652362; fax: +1 814 8657065; e-mail: zliu@psu.edu

Summary

We investigate wavelength division scanning for two-photon excitation fluorescence imaging. Two-photon imaging using lateral wavelength division scanning is demonstrated. In addition, we theoretically analyse the spatial and temporal properties of a femtosecond laser beam focused by a Fresnel lens and investigate the feasibility of axial scanning using wavelength division.

Introduction

Two-photon excitation fluorescence microscopy (Denk et al., 1990) has become an important tool for biological and biomedical applications, due to its large penetration depth in scattering samples and intrinsic optical sectioning capability. Conventional two-photon microscopes, however, have limited capability to image fast biological events, due to their slow point-by-point scanning speed. In the past several years, much progress has been made in developing high-speed lateral scanning techniques such as line scanning (Guild & Webb, 1995; Brakenhoff et al., 1996; Tal et al., 2005), multipoint multiphoton imaging (Bewersdorf et al., 1998; Buist et al., 1998; Sacconi et al., 2003), and polygonal mirror scanning (Kim et al., 1999). However, a high-speed axial scanning method for two-photon imaging is still lacking. As a result, it is currently still a challenge to visualize fast biological processes in full three-dimension (3D) or xz and yz cross-section. On the other hand, wavelength division-based techniques are known to have the unique capability of allowing high-speed axial scanning. For instance, in chromatic confocal microscopy (Molesini et al., 1984; Browne et al., 1992; Maly & Boyde, 1994; Cha et al., 2000; Shi et al., 2004), different wavelengths of a broadband light source are focused to different axial positions through deliberately introduced chromatic aberration. Because different wavelengths are directed to interrogate different axial positions axial scanning can be effectively achieved by analysing the spectrum of the reflected light. A spectrally encoded confocal microscope (Tearney et al., 1998) was also demonstrated which uses wavelength division to effectively realize one lateral scanning. However, wavelength division-based methods have mainly been applied to reflection-type confocal microscopy, which has limited applications in biology. In this article, we intend to investigate the feasibility of applying wavelength division to fluorescence imaging, in particular, two-photon excitation fluorescence imaging.

Experimental procedures and Discussion

Our experimental setup is shown in Fig. 1. A femtosecond laser (KMlabs, Boulder, CO) beam is first sent to a tunable filter, which consists of a grating, a lens, a movable slit, and a mirror. The grating disperses the incident pulse, and different wavelengths are focused to different positions at the back focal plane of the lens, where the slit is placed to filter the spectrum of the incident pulse. The mirror reflects the filtered spectrum back, and the beam is recollimated. The spectrally filtered excitation pulses are then dispersed with a prism and subsequently relayed to the specimen by a telescope imaging system to achieve desired lateral and axial resolutions. Finally, the fluorescence signal is separated by a dichroic mirror and detected by a CCD camera (Apogee AP32ME). Since different wavelengths travel along different directions and are therefore focused to different lateral positions, lateral scanning can be effectively achieved by tuning the centre wavelength of the excitation pulses (i.e. moving the slit). Figure 2(a) shows a pseudocolour image of the excitation beam when all the wavelengths are let through (i.e. without the slit). As expected, a narrow line is obtained. When we block all but a small spectral window of wavelengths (bandwidth: ∼2 nm) using the slit, a focused excitation point can be observed, as shown in Fig. 2(b). As the wavelength is tuned, the excitation point scans laterally. Figure 2(c–e) shows the fluorescence (in pseudocolour) corresponding to three different excitation wavelengths when the laser is focused into a fluorophore solution (1 µm Rhod-123). It is evident that lateral scanning can be achieved by tuning the wavelength. A small effective lateral scanning range of about 10 µm is obtained in our experiment, partly because of the limited angular dispersion of the prism used. In order to show that the fluorescence is indeed excited by a two-photon process, we measured the dependence of fluorescence signal on the average excitation power. The result is given in Fig. 2(f). The quadratic relationship clearly demonstrates that the process is two-photon. The inset plots the square root of the fluorescence signal as a function of the excitation power, which is linear, as expected. Next, we used this system to image 1-µm-diameter and 2-µm-diameter fluorescent microspheres (Duke Scientific, Fremont, CA), immobilized on a cover slip glass. One lateral direction was scanned through wavelength division, and the other lateral direction was scanned by moving the cover slip, and hence the microspheres. The results are shown in Fig. 3(a) and (b). The fluorescence signal was normalized to the square of the excitation power to compensate for the nonuniform spectrum of the femtosecond laser pulses. Finally, we also used the system to image a neurone, which was stained with 10 µm FM 1-43. In this experiment, the slit was removed, and therefore a line image was obtained each time. The specimen was mechanically translated along the other lateral direction. The result is given in Fig. 3(c). Nerve terminals can be observed. Although the scanning speed in the above proof-of-concept experiments is limited, due to the use of a slow tunable filter, the speed can be greatly improved if ultrafast tunable filters are used. Note that tunable filters used in optical communication have already achieved nanosecond tuning speed; see, for example, Matsuo et al. (2005). The wavelength scanning technique can potentially result in a speed much higher than that of mechanical scanning methods.

Figure 1.

Schematic diagram of the lateral wavelength division scanning two-photon fluorescence imaging setup.

Figure 2.

(a) Image of the excitation beam when all wavelengths are let through. (b) A focused excitation point when all but a small spectral window of wavelengths are blocked. (c–e) The observed two-photon excited fluorescence corresponding to three different excitation wavelengths when the laser is focused into a fluorophore solution (1 µm Rhod-123). All colours are pseudocolour. (f) Dependence of fluorescence signal on excitation power. The inset plots the square root of fluorescence signal as a function of the average excitation power.

Figure 3.

Two-photon fluorescence imaging. (a, b) Images of immobilized fluorescent microspheres with 1-µm diameter and 2-µm diameter, respectively. (c) Result of imaging a neurone (sample obtained from Dr Gong Chen's laboratory).

A unique advantage of the wavelength division technique is that it can be used to implement axial scanning in two-photon imaging, which still lacks a high-speed axial scanning method. Usually, slow mechanical translation of the objective lens or the specimen itself along the axial direction is still required, which limits the xz or yz cross-section and the overall 3D imaging speed. To achieve axial scanning, the prism and the lens that follows it in Fig. 1 may be replaced with a Fresnel lens (Guenther, 1990), which can produce large chromatic aberration. Since different wavelengths are now focused to different axial positions, axial scanning can be similarly realized by tuning the centre wavelength of the spectrally filtered laser pulses. In the following, we analyse the problem of focusing ultrashort femtosecond pulses with a Fresnel lens and show that axial scanning can be effectively achieved through chromatic aberration. We consider a femtosecond laser beam uniformly incident on a Fresnel lens (design wavelength λ0, design focal length f0, K phase levels, aperture radius a). Let us first consider a particular frequency component (ω) of the incident pulse. The field immediately behind the Fresnel lens is given by

image(1)

where

inline image

n (n0) is the refractive index of the lens material at λ (λ0), (ω) is the spectrum of the incident field, and

inline image

is the pupil function. The field near the focal point is given by (Goodman, 1996; Ashman & Gu, 2003)

image(2)

If a spectrally filtered excitation pulse (centre wavelength λ, bandwidth δλ) is focused by the Fresnel lens, the intensity distribution near the focal point is then given by

image(3)

Figure 4 shows the calculated radial and axial intensity distribution of the focused beam when the centre wavelength is shifted from 780 nm to 820 nm (pulse bandwidth δλ = 1 nm, a = 5 mm, K = 8, design wavelength 800 nm, design focal length 100 mm). An effective scanning range of about 5 mm can be obtained. Note that there is no appreciable aberration of the focused beam, as the resultant chromatic aberration of the filtered pulse (with just 1-nm bandwidth) is smaller than the depth of focus. This point will become more apparent if we consider the spatial and temporal distribution of the field near the focal point. To this end, we consider a pulse that has a rectangular spectrum, i.e. (ω) = 1 for | ω − ω0 | £ Ω/2 or equivalently a pulse width of 2π/Ω. From Eq. (2), at position z on the optical axis the field can be simplified to (Wilson & Sheppard, 1984)

Figure 4.

The radial and axial intensity distribution produced by focusing spectrally filtered laser pulses (bandwidth 1 nm) with a Fresnel lens. As the centre wavelength is tuned from 780 nm to 820 nm, the focus spot moves accordingly by about 5 mm and effectively realizes axial scanning.

image(4)

where

image

Here we have dropped the nearly constant amplitude factor, the constant phase, and the linear phase, which merely corresponds to a time delay. By performing the inverse Fourier transform, we obtain the temporal evolution of the pulse at position z.

image(5)

where ωc = 2πcz0f0 is the centre angular frequency at position z. We can see that the pulse width is now broadened by a factor of b, which is given by

image(6)

Note that Ω/ω0 = −Δλ/λ0 = Δf/f0, where Δλ is the spectral width of the incident pulse and Δf is the chromatically extended depth of focus (CEDOF). Putting this into Eq. (6), we obtain

image(7)

where NA is the numerical aperture of the Fresnel lens and DOF is the nominal depth of focus. Intuitively, wavelength λc is focused by the Fresnel lens to the position z = λ0f0c and therefore is the centre wavelength of the pulse there [compare Eq. (5)]. Because the spectrum of the incident pulse is spread over a chromatically extended depth of focus region (Δf), and at z only a portion of the original spectrum that falls into the local depth of focus region can be ‘seen’, the pulse width at z is therefore broadened accordingly by a factor equal to the ratio between the chromatically extended depth of focus and the nominal depth of focus. Since excitation pulses with different wavelengths are focused to different axial positions, similar to the lateral scanning demonstrated using a prism, we can then effectively scan the specimen axially by rapidly scanning the centre wavelength. It can be easily shown that the effective scanning range is given by nmM2Δf, where M is the magnification of the telescope system, and nm is the refractive index of the immersion medium (water or oil) for the objective lens in proximity to the specimen. Considering that femtosecond laser pulses with large bandwidth are available (see, for example, http://www.kmlabs.com), we envision that an effective scanning range up to 100 µm can be potentially achieved.

Conclusion

In summary, we have investigated the feasibility of applying the wavelength division technique to two-photon fluorescence imaging. It is worth pointing out that the bandwidth of femtosecond laser pulses can be made large enough (e.g. ∼50 nm) to achieve a large effective scanning range but still within the two-photon excitation bandwidth (Xu & Webb, 1996; Albota et al., 1998) of many fluorophores. The spectrally filtered excitation pulses can typically have a bandwidth of about 1 nm and a pulse width of about 1 ps, and can produce a peak intensity greater than 1 GW cm−2 with just a few milliwatts average power. The average power of two-photon excited fluorescence is approximately proportional to inline image (Konig et al., 1999; Koester et al., 1999), where Pav is the average excitation power and τ is the pulse width [Note that the two-photon fluorescence yield may also depend on the phase of the field and the detailed pulse shape. See, for example, Meshulach & Silberberg (1998) and Ogilvie et al. (2006).] It has been reported that the dominating mechanism of cell damage in two-photon fluorescence imaging is also due to two-photon or multiphoton processes (Koester et al., 1999; Konig et al., 1999). The two-photon processes responsible for photodamage follow the same dependence on average power and pulse width (∝ inline image) as two-photon excited fluorescence. Therefore, for the spectrally filtered pulses described in this article, a higher average power (scaled by the square root of the pulse width ratio) can indeed be used to produce similar image quality without damaging the specimen.

Acknowledgements

The authors thank Dr Gong Chen for preparing the neurone specimen and Dr Ahmed A. Heikal for providing the fluorophore and helpful discussions. This work is supported by the Lehigh/Penn State Center for Optical Technology.

Ancillary