High efficiency beam splitter for multifocal multiphoton microscopy


Tim Nielsen. Tel.: + 49 521 1065434; fax: + 49 521 1062958; e-mail: nielsen@physik.uni-bielefeld.de


In this article we present the development of a multibeam two-photon laser scanning microscope. A new type of beam splitter to create the multitude of laser beams is described. This type of beam splitter has higher transmission and generates more uniform beams than can be achieved with the microlens approach used by other groups. No crosstalk exists between the different foci due to small temporal delays between the individual beams. The importance of dispersion compensation to obtain maximum efficiency of the microscope is discussed. With optimum compensation the fluorescence signal was raised by a factor of 14. Different modes of detecting the fluorescence signals and their effect on imaging speed and resolution are discussed.


Three-dimensional microscopy is an important tool for biomedical imaging. The ability to image thin sections of thick, intact specimens enables the study of live cells or tissue with high resolution. By fluorescence labelling techniques individual biomolecules or structures of cells can be visualized with high selectivity. Three-dimensional spatial resolution can be obtained either by confocal or non-linear laser scanning microscopy.

Two-photon microscopy has several advantages compared with confocal laser scanning microscopy: near infrared (NIR) lasers which are typically employed in two-photon systems have a higher penetration depth in scattering tissue than the visible lasers used in confocal microscopes. Because laser radiation is absorbed only in the focal region photobleaching and photodamage is confined to this small volume, whereas in confocal systems the entire part of the sample that is illuminated by the laser beam is subject to bleaching. With two-photon microscopy it is possible to study live cells for a longer period of time without light-induced damage (Squirrell et al., 1999).

One drawback of two-photon microscopy is the need for high peak intensities in the focal spot to get sufficient two-photon absorption. Usually pulsed lasers with a pulse length from pico- to femtoseconds are used.

Another drawback is the imaging speed. As it is a scanning technique where signal is acquired point-by-point it is time-consuming to image large volumes. For example, 10 s are needed for one frame if the acquisition time is 100 µs per point and 105 points are scanned. Using a stepwidth of 150 nm this corresponds to an area of about 50 × 50 µm. Imaging of a volume further increases the imaging time because multiple planes have to be scanned.

The use of laser scanning microscopy is limited to those processes occurring on a time-scale larger than the time needed to acquire one image. Increasing the imaging speed therefore means that the benefits of two-photon microscopy are made applicable to more areas of research. One example that illustrates the need for fast, three-dimensional microscopy is the study of fast processes in nerve cells: two-photon microscopy was used to study the activity of dendritic spines (Yuste & Denk, 1995). Here, on one hand two-photon microscopy was necessary to obtain high resolution images in nervous tissue, on the other hand the slow imaging speed prevented the acquisition of two-dimensional images. Only measurements along one line could be performed with the necessary speed.

One way to increase imaging speed is to reduce the time needed to acquire a sufficient amount of fluorescence light from each point in the sample. This can be done either by increasing the dye concentration, using dyes with better fluorescence properties (two-photon absorption cross-section, quantum efficiency) or increasing the laser intensity.

The concentration of the dye cannot be raised arbitrarily because it may alter the behaviour of the system under investigation. Often dyes have a toxic effect that limits their concentration. The development of new dyes (Bhawalkar et al., 1996) with improved fluorescence properties is certainly useful but the choice of the dye may be determined by the problem that is to be studied.

The fluorescence signal depends quadratically on the laser intensity. Therefore the signal can be improved by using higher intensity in the focus. This is limited by the onset of other non-linear effects such as self-focusing, multiphoton ionization, etc. which destroy the sample. Today only a little is known about the dependence of photodamage on laser power in two-photon microscopy (König et al., 1997). Typically only a small fraction (< 10 mW) of the output power of a laser can be focused into the sample without inducing damage.

Another way to increase imaging speed apart from maximizing the signal from each point is to illuminate multiple points of the sample simultaneously. The time needed to acquire one image is then reduced proportional to the number of beams employed. This so-called ‘multifocal multiphoton microscopy’ (MMM) has been demonstrated by Bewersdorf et al. (1998) and Buist et al. (1998).

In MMM two important changes compared to a single-beam laser scanning microscope have to be made: (1) a beam splitter device is needed to generate the multitude of focal spots, and (2) a detector capable of detecting the fluorescence signals of all foci is required.

In this article a multifocal multiphoton microscope is described that uses a new type of beam splitter and a different way of detecting the signals. Our set-up is compared to those in the literature. Different aspects of MMM will be discussed in the following sections in detail: the significance of the pulse duration in the sample, generation of the multitude of focal spots, and the detection of the fluorescence.

Experimental set-up

Figure 1 shows the set-up of our multibeam two-photon microscope. We use a femtosecond Ti:Sapphire laser (Coherent Vitesse, Santa Clara, CA, U.S.A.) to produce laser pulses at a wavelength of λ = 802 nm with an average power of P = 1 W and a repetition rate of 80 MHz. The FWHM of the laser spectrum is 34 nm. This limits the minimum pulse width that can be achieved theoretically to 29 fs.

Figure 1.

Experimental set-up.

An autocorrelator can be placed in the beam path to measure the pulse length and chirp. The autocorrelator is removed for normal operation of the microscope.

A pair of prisms (SF10) is used to compensate the group velocity dispersion of the system (Fork et al., 1984). The laser beam can be attenuated by a rotatable half-wave plate followed by a polariser. Then the laser beam is divided into an array of 8 × 8 beams by a beam splitter that will be described below. The beams converge to one point and are imaged by a telescope onto the aperture of the microscope objective lens. The telescope magnifies the beam diameter so that it slightly overfills the aperture of the objective lens. We use a Zeiss PlanNeofluar 63×/1.4 oil immersion objective lens to focus the beams into the sample. The sample is mounted on a piezo scanning stage allowing a scan range of 100 × 100 µm. The objective lens is attached to a piezo device (Physik Instrumente, PIFOC, Waldbraun, Germany) to perform scanning in the axial direction. The fluorescence is collected with the same objective lens, passes a dichroic mirror and an edge filter (1 mm BG39) and is imaged onto a CCD camera (Imager 3, LaVision GmbH, Göttingen, Germany). The whole setup is controlled by a personal computer that synchronizes camera readout and scanning.

Construction of the beam splitter

The principle of the beam splitter is shown in Fig. 2. The laser beam enters from the lower left and is divided into two beams by a dichroic 50% mirror (60 × 30 × 2 mm3 suprasil). Both beams are reflected back to the 50% mirror by high reflectivity mirrors (R0, S0) producing four beams. This is repeated, doubling the number of beams with every pass of the 50% mirror. All beams leave the device in a plane. The separation of the beams at the exit of the beam splitter is determined by the difference between the distances of the high reflectivity mirrors and the 50% mirror.

Figure 2.

Principle of the beam splitter.

In order to generate multiple foci inside the sample the different beams must have an angle with respect to each other. The angle between the beams determines the distance of the foci in the sample. The relationship between angle and distance can be calculated from the magnification of the objective lens and the intermediate telescope.

If all mirrors in Fig. 2 are parallel, all generated beams are parallel also. By tilting the mirrors S0, S1, S2, …, Si by an angle of α/2, α, 2α, …, 2i−1α all beams have a relative angle of α yielding equally spaced foci.

Each of the eight beams leaving the beam splitter is split again into eight beams by a second beam splitter of the same type. Before the beams enter the second beam splitter they are turned upright by a periscope. In this way, a matrix of 8 × 8 beams is generated. The same effect could be achieved without the periscope if the second beam splitter was tilted by 90°. As both 50% mirrors are perpendicular to each other (i.e. the periscope turns the polarization) one has to be coated for P polarization, the other for S polarization.


Significance of the pulse duration inside the sample

In this section the significance of the pulse duration inside the sample for MMM is discussed.

We measured pulse length and chirp with an autocorrelator placed between the laser and the dispersion compensation unit. The pulse parameters inside the sample were measured by recording the two-photon fluorescence (Wolleschensky et al., 1998). In this way the individual pulse parameters of all 64 beams were determined simultaneously.

Figure 3 shows two autocorrelation traces for one of the 64 beams recorded in the sample: (a) completely without dispersion compensation and (b) with optimal compensation of the dispersion. Without dispersion compensation the pulse length was τ = 575 ± 25 fs, with optimal compensation the pulse length was reduced by a factor of 14 to τ = 41 ± 0.2 fs.

Figure 3.

Averaged autocorrelation traces recorded in the sample. (a) without dispersion compensation unit (τ = 575 ± 25 fs), (b) with optimal compensation (τ = 41 ± 0.2 fs). The inset in (b) shows the oscillations at the maximum in a magnified view.

Given a fixed laser energy, the two-photon signal is inversely proportional to the pulse length. This means that the signal rises by a factor of 14 simply by ensuring short pulses inside the sample. It should be noted that the effect of dispersion compensation depends on the bandwidth of the laser pulses. The shorter the Fourier-limited pulses are the more important is the dispersion compensation.

Commercially available femto-/picosecond lasers achieve an average output power of up to approximately 2 W. This averaged power is practically independent of the pulse length of the lasers for pulses not too short (τ > 30 fs). As a consequence for MMM it is desirable to use pulses as short as possible because the same fluorescence signal can be generated with less laser energy which means, in turn, more individual beams can be employed increasing the imaging speed accordingly. Despite this, dispersion compensation has not been used in MMM setups before.

The influence of the pulse length on cell viability has been studied by König et al. (1999) for pulse durations in the range of ≈ 170 fs−2.2 ps. It was found that the damage behaviour follows approximately P2/τ, indicating that cell destruction in this region of pulse lengths is likely based on a two-photon excitation process. In this case, the ratio of fluorescence rate and damage rate is constant. Short pulses induce the same damage as longer pulses if both are compared at the same signal level. If this holds also for pulses shorter than 170 fs the output energy of a laser can be used most efficiently for MMM if pulses of the shortest possible duration are used.

It was pointed out by König et al. (1999) that for three-photon microscopy the ratio of fluorescence and damage rate increases as the pulse length decreases (assuming the damage rate is based on a two-photon process in this case too). This means short pulses are recommended for non-destructive three-photon microscopy.

For a multibeam three-photon microscope short pulses are a must because the signal depends inversely on the square of the pulse length (signal ∝ P32). Reducing the pulse length by a factor of 10 will allow 100 times more beams increasing the imaging speed accordingly.

The influence of the beam splitter on the pulse length

From Fig. 2 it is clear that the dispersion is not identical for every beam because the number of passes through the substrate of the 50% mirror is different. One beam (a in Fig. 2) does not pass the substrate at all, whereas another beam (b) makes five passes (once through, in and out twice). As the beam splitter is constructed of two stages, the maximum difference is 10 passes through 2 mm suprasil. The angle of the beam inside the substrate is 29° with respect to the surface normal, giving an effective path of 2.29 mm per pass. The dispersion of suprasil at 802 nm is 34.9 fs2 mm−1 (= 79.9 fs2 pass−1).

We investigated the influence of the beam splitter on the pulse length by varying the distance between the compensation prisms. For every position an autocorrelation trace inside the sample was recorded and the pulse length was calculated for every beam. If the dispersion of the system is ideally compensated the pulse length is minimal.

We determined the prism separation for minimal pulse length for every beam. In Fig. 4 the position of the minimum is displayed in dependence on the number of passes through the substrate of the 50% mirrors. There is a linear relationship between both variables. From linear regression we get a slope of 0.44 ± 0.02 cm pass−1.

Figure 4.

Dependence of the optimum distance between the compensation prisms on the number of passes through the substrate of the 50% mirrors of the beam splitter.

The dispersion of the prism sequence is changed by − 184 fs2 cm−1 if the distance between the prisms is varied. This can be calculated from material constants of SF10 and the geometry of the set-up. This means the slope corresponds to 0.44 cm pass−1·184 fs2 cm−1 = 81 ± 4 fs2 pass−1. This is in agreement with the calculated value of 79.9 fs2 pass−1. This shows there is a detectable effect of the dispersion of the beam splitter.

Assuming an optimal compensation for beams with a medium number of passes and a pulse length of 41 fs the pulse length of beams with minimum or maximum number of passes is 48 fs. This means that the two-photon signal is changed by approximately 10% due to variations of the pulse length between the different beams. This effect could be removed by a symmetric design of the beam splitter. For example, by putting 2 mm of suprasil in front of the mirrors S0, S1, S2.

Characterization of the beam splitter

Important parameters of the beam splitter for MMM are the transmission and the uniformity of the beams.

The overall transmission of the beam splitter was 91%, which is much higher than for the microlens approach for generating the beams. To characterize the uniformity of the beams the variation of the two-photon signals was measured by focusing all beams into a homogeneous solution of laser dye (Coumarin 500, Lambda Physik, Göttingen, Germany) in immersion oil. Figure 5 shows an image where the fluorescence of all beams is visible. The signals of the beams deviate from the mean with an rms of 24%. The maximum deviation from the mean is 60%. The reason for the non-uniformity of the beams is mainly that the reflectivity of the partially reflecting mirrors in the beam splitter is not exactly 0.5 but 0.56 (P polarization) and 0.495 (S polarization). In contrast to the microlens approach there is no influence of the laser beam profile on the intensity of the generated beams.

Figure 5.

Image of the fluorescence of all beams. The scalebar represents 5 µm.

Another advantage of the beam splitter is that every beam that is generated has the same beam profile as the incoming laser beam, i.e. the size of the foci is not affected by the beam splitter.

It has been reported by Buist et al. (1998) that crosstalk between the excitation beams can be a problem in MMM depending on the distance between the individual beams. Because the beams are tightly focused, the cones of neighbouring beams overlap above and below the plane of focus. If there is two-photon excitation in this overlapping region too, the PSF is not the same as for single beam two-photon microscopy and the resolution will be reduced. This crosstalk in the excitation process must not be confused with crosstalk in the detection of the fluorescence, which will be discussed in a following section.

This is a problem only if the pulses overlap not only spatially but also temporally. If there is a delay between the pulses that is greater than their duration no interference between the pulses can occur.

With our method of creating the multitude of beams there is an intrinsic delay of several picoseconds between the beams because the optical paths are different. The delay between the beams is approximately given by the distance of the beams at the exit of the beam splitter. For neighbouring beams this was ≈ 3 mm, corresponding to 10 ps, which is far more than the pulse duration. No crosstalk between the excitation beams exists, no matter how close they are spatially. This has been confirmed by axial scanning through a thin fluorescent layer. No difference was observed if the scan was performed with one beam only or with all beams.

On the other hand, it would be possible to place the foci so close together that they are no longer resolved by the CCD camera as individual points. In this case one would get an image of the sample completely without scanning. The axial sectioning capability would be unaffected because all beams pass the focal plane at different times. This will be discussed in a future publication.

Detection of the fluorescence

The fluorescence signals of all beams have to be detected simultaneously. Therefore, a CCD camera is used instead of a avalanche photodiode typically employed in single beam laser scanning microscopes. From single beam laser scanning microscopes two techniques are known: descanned and non-descanned detection.

In the non-descanned detection scheme the focus is scanned in a plane through the sample and the fluorescence is imaged onto a camera. This directly yields a two-dimensional image of the section of the sample without any image processing. This approach has been used for MMM with a rotating microlens disc by Bewersdorf et al. (1998).

In our set-up we used a scanning stage to image the sample. This means that non-descanned detection of the fluorescence is not possible. Instead, we recorded one image with the CCD camera for every position of the sample. The image of the sample is calculated from the raw data, as will be described below. The descanned detection scheme is slower due to the detector readout time but has the advantage that the resolution is given by the size of the focus only, and not by the image of the fluorescence emitted from the focus, which is affected more strongly than the NIR laser radiation by scattering inside the sample.

First, the signals of the individual beams are calculated by averaging over a small region covering the fluorescence spot. These values are corrected for the camera dark current and are multiplied with a normalization factor that takes account of the different intensities of the beams. As a result, 64 intensity values (S) are obtained from every recorded image.

From calibration measurements the (x, y, z) position of the scanning stage as well as the distance between the beams are known. With these data the absolute position of every beam inside the sample can be calculated. This means that a scan can be represented by a set of (x, y, z, S) values. These can be converted to an image by image processing software.

Two problems that can occur have to be noted:

1. Each beam scans a certain area of the sample. The areas of adjacent beams overlap if the piezo stage is scanned more than the distance between the foci.

2. Depending on the distance of the foci and the shape of the image of the fluorescence from one focus there may be crosstalk between the averaged intensity values of the beams if the size of the averaging region is too large.

The first problem can be overcome simply by averaging the data in the overlapping regions. The second problem is more serious and will be discussed in the following section.

Crosstalk in the detection of the fluorescence

The fluorescence from every focus is not imaged onto a point but is spread out over a small region (FWHM of the peaks is 0.6 µm). Depending on the distance of the foci, a fraction of the fluorescence from one beam falls into the averaging region for the neighbouring beams. This means that the observed signal inline image for beam i is the weighted sum of all signals inline image that would be observed if the fluorescence of all foci could be separated completely.



inline image

In the ideal case with no crosstalk the matrix of weights would be diagonal with wij = δij. For a real set-up the weights can be calculated from the function that characterizes the image of the fluorescence of one focus, the distance between the foci, their relative intensities and the size of the averaging region. The true signals can then be calculated from the observed signals by inversion of Eq. (1).

The weights can also be determined experimentally by recording a series of 64 images where in image j only the beam j enters the microscope. This can be achieved by blocking some of the mirrors of the beam splitter. The observed signal for a beam i in the image j is directly proportional to the weight wij.

Figure 6 shows the weights wij for a central beam (i = 18) in a logarithmic scale for an arrangement of the beams as in Fig. 5. The integration region was a circle centred at each peak with a diameter of twice the FWHM of the peaks. It can be seen from Fig. 6 that only the weights of the adjacent beams (j = 10, 17, 19, 26) have a noticeable value of ≈ 0.10.

Figure 6.

Contribution from neighbouring beams to the signal of one beam, i.e. the wij from Eq. (1) for i = 18. Beam no. 18 is circled in Fig. 5.

The effect of crosstalk can be more severe if the fluorescence of individual beams is no longer well resolved, e.g. because of scattering of the fluorescence inside the sample.

If a sample is investigated that scatters the fluorescence, the width of the image of the fluorescence can be significantly broadened. This is shown in Fig. 7, where horizontal profiles of the fluorescence of one focus are displayed for different depths in a sample of multiple layers of Medicago sativa stained with SyBR Green.

Figure 7.

Horizontal profile of the detected fluorescence for different depths in a stack of multiple cells showing the influence of scattering of the fluorescence.

Even if the fluorescence is scattered on its way out of the sample it is possible with our method of detecting the fluorescence to image the specimen with the resolution that is determined by the diameter of the laser focus and not by the diameter of the image of the fluorescence, as long as the distance between the individual beams is large enough to separate the fluorescence from different foci. In this case the image resolution is not lowered by scattering of the fluorescence.

The correction for crosstalk between the beams is not possible in the non-descanned detection mode. Here the width of the fluorescence of a single beam is the limit for the resolution that can be achieved in the scanned image. Hence, non-descanned detection yields low resolution images if the specimens scatter the fluorescence light too much.

As a demonstration image, Fig. 8 shows a stack of images from a Chinese hamster ovary (CHO) cell stained with ethidium bromide and fluorescein.

Figure 8.

A stack of images from a CHO cell recorded at different depths. The cell was stained with ethidium bromide and fluorescein.

Imaging speed

A drawback of our approach to detecting the fluorescence is a lower imaging speed as compared to non-descanned detection. It is necessary for our method to readout the camera once for every position of the scanner. This means in addition to the integration time that is needed to accumulate a sufficient amount of light, the readout time of the CCD camera must be added to the pixel dwell time.

To minimize the deadtime due to camera readout the camera should be as fast as possible. For the CCD camera we used the readout time for a full frame is 33 ms. This can be reduced to 4.2 ms if the camera is operated in a binning mode where multiple pixels are summed before the signal is digitized. Even this reduced time is far longer than the integration time, which was typically 100 µs.

Two hundred and forty frames per second could be recorded, giving 64 intensity values each. The total pixel rate of the microscope is 15000 s−1. This is only on the order of a single beam microscope but it should be noted that the imaging speed is dominated by the slow readout speed of our CCD camera. By improving the set-up with a highspeed camera or non-descanned detection a pixel rate 10–100 times higher could be achieved.

Summary and future prospects

We described the set-up of a stage-scanning, multi-beam, two-photon microscope. A novel beam splitter was used to create the multitude of beams. This beam splitter has some advantages compared to the microlens approach that has been used previously:

• a high transmission of 91%,

• good uniformity of the beams (rms of the two-photon signal is 0.26 of the mean signal),

• no influence of the laser beam profile on the energy distribution of the generated beams,

• a simple set-up with few standard optical components,

• no crosstalk between the excitation beams due to inherent delays of some picoseconds between the individual beams,

• short pulse operation is possible without problems

The crucial parameter for maximum two-photon excitation is the total laser intensity that is put into the sample. This means the total energy must be maximized (i.e. transmission) and the pulse length must be minimized at the same time by the use of dispersion compensation. Two-photon excitation is maximized if every individual beam has as much intensity as can be allowed without damaging the sample. Both high energy and short pulses are possible with our beam splitter set-up.

Short pulses might be a problem with a microlens array because chromatic aberration of the lenses could reduce the resolution if broadband pulses are used (Buist et al., 1998).

To detect the fluorescence we used a CCD camera that recorded one image for every scanning position. From these images the final image of the sample is calculated by image processing routines.

A disadvantage of this technique is the lower imaging speed due to the additional deadtime for camera readout. To enable video rate imaging it is planned to use beam scanning and non-descanned detection.

One advantage of this detection method is that crosstalk between the images of the fluorescence from the individual beams can be detected and corrected for. Crosstalk of the fluorescence can lower the resolution of the images (in non-descanned detection) or lead to the formation of artefacts. Especially if scattering specimens are to be investigated with MMM crosstalk of the fluorescence is a severe problem that can be managed with our method of detecting the signals.