1. Introduction

Conventional epi-illumination fluorescence microscopes illuminate the focal plane of the detection objective as well as a large fraction of the sample outside the focal plane. This leads to background intensity due to out-of-focus fluorescence and reduces contrast and resolution. In confocal laser scanning microscopy a pinhole in a conjugate plane can be used to reject out-of-focus fluorescence since the sample is scanned point-wise but at the cost of low image acquisition rates. This type of sectioned imaging is advantageous for three-dimensional (3D) data acquisition but leads to high light exposure of the entire sample. Light Sheet Fluorescence Microscopy (LSFM) circumvents this disadvantage by illuminating the sample perpendicular to the detection direction. Typically, a laser beam with an elliptical cross section is formed by cylindrical lenses and focused into the sample by an illumination objective. Thus only fluorophores in a thin sheet near the focal plane are excited [1–4]. The result is an intrinsic optical sectioning effect, low photobleaching and reduced photo-toxicity. LSFM permits studies of bulk samples like whole embryonic mouse brains or the real time investigation of embryonic cell development in vivo, while keeping the total light exposure of the sample at a minimum.

An alternative method to generate a light sheet is realized in digitally scanned light sheet microscopy [5,6]. Here, a circular Gaussian beam is scanned across the focal plane by a mirror, which is positioned in the back focal plane of the illumination objective. The beam is focused into the sample with its main axis perpendicular to the focal plane of the illumination objective. When the beam is swept across the focal plane at least once during an image acquisition cycle a planar illumination equivalent to a sheet but with improved illumination efficiency and a uniform intensity distribution is achieved.

The fluorescence detection in LSFM does not differ from conventional wide-field microscopy. The image sensor detects ballistic photons as well as scattered photons which have lost their image information. In addition, excitation light is scattered in the sample and excites off-focal plane fluorophores. Especially in thick samples these effects can be noticed as image blur and reduction of contrast.

Previously published methods to reduce the amount of scattered light either rely on structured illumination patterns [7] or the application of confocal slit masks to images taken sequentially with line illumination [8]. Though image contrast is enhanced these methods suffer from long data acquisition times since the final contrast-enhanced image is numerically reconstructed from multiple raw images.

An elegant solution to speed up the data acquisition has recently been proposed [9]. The scanning of the illumination beam can be synchronized to the rolling shutter of a scientific CMOS (sCMOS) camera. An sCMOS is an active-pixel sensor with separate photodetector and amplification units for each pixel. In contrast to CCDs the pixels can be addressed and read out independently from each other. Currently available sCMOS can be operated in two data acquisition modes: global shutter and rolling shutter. When working with global shutter, all image pixels are simultaneously active, which is similar to image acquisition by CCD cameras. In rolling shutter mode, however, the active detection region is comprised of only a few adjacent pixel lines and thus equals a band of simultaneously exposed rows that moves across the image sensor. Choosing a sufficiently small exposure time results in a line-shaped detection region that represents a moving confocal slit detector, if synchronized to the illumination beam. In this manner fast line-scanning confocal light detection with a minimum number of mechanical parts can be realized.

We developed a microscope that uses a circular Gaussian beam scanned synchronously to the rolling shutter detection. It offers true confocal imaging at high frame rates. We showed that scattered fluorescence is rejected during the acquisition of an image. Further, we characterized the improvements in contrast, signal-to-noise ratio (SNR) and optical sectioning. The instrument presented here will be especially beneficial for 3D imaging of large specimen.

2. Methods

2.1 Illumination beam path

Our scanned light sheet microscope setup was designed to augment an inverted microscope. Custom built sample chambers with a coverslip bottom (0.17 mm) allowed us to use conventional objective lenses for observation. The sample was held by a motorized three axis translation stage [10].

A laser beam with a wavelength of 633 nm was guided through a single mode optical fiber onto a two-axis scanning mirror, Fig. 1 . The scanning mirror was placed in the front focal plane of a relay lens system that imaged it into the back focal plane of a 10X achromat (Plan Apo Infinity-Corrected Long Working Distance, Mitutoyo Corp., Kawasaki, Japan) located inside the objective housing. The illuminated aperture radius in the back focal plane was 1.6 mm, which resulted in an effective illumination NA of 0.08. The relay system comprised a lens doublet and a single lens in a 4f configuration. It was designed to compensate spherical aberration that originates from the specimen chamber glass walls and the sample medium by ray-tracing using Zemax (Radiant Zemax, LLC, Redmond, WA, USA). The${\text{1/e}}^{\text{2}}$beam waist radius${\text{w}}_{\text{0}}$and the Rayleigh length ${\text{z}}_{\text{r}}$were ${\text{w}}_{\text{0}}\text{=}\text{4}\text{.5}\text{±0}\text{.2}\mu \text{m}$ and ${\text{z}}_{\text{r}}\text{=}\text{116}\text{±5}\mu \text{m}$ for 633 nm excitation wavelength as determined by imaging the beam in a fluorescent dye solution. A Nikon CFI Plan Fluor 10X objective lens with a NA of 0.3 was used for fluorescence detection. Its axial FWHM was estimated to be 11.6 µm for a wavelength of 650 nm [11].

 

Fig. 1 Illumination beam path. (A) Top view. The excitation beam exits a single mode fiber and is guided by mirrors onto a scanning mirror. The scanning position is imaged by a relay objective into the back focal plane of the illumination objective lens, which is located inside the objective housing. The relay lens assembly is composed of two lens systems with overlapping focal points. It also serves to expand the illumination beam. (B) Side view of the illumination optics.

Download Full Size | PDF

2.2 Data acquisition

We used a sCMOS camera (1920 x 1440 pixels, pixel size 3.63 µm, Orca Flash 2.8, Hamamatsu Photonics K.K., Hamamatsu City, Japan) for data acquisition. It was operated either in a global or a rolling shutter mode. In global shutter mode all pixels simultaneously detect photons for a predefined integration time after reset and begin of exposure. In rolling shutter mode the detector is activated, and the exposure starts sequentially with a time delay of 15 μs between neighboring rows beginning with the topmost pixel row. The activation process continues across the complete chip until the last row. The readout starts in a similar manner at the first pixel row after a predefined integration time. It is continued row by row again with a delay of 15 μs between subsequent rows. In this manner the time required to activate all 1440 rows of the sensor is 22 ms. We defined the integration time per line to be below this value. Therefore, the readout of the topmost rows began before the activation of the last row. In this manner a “rolling shutter” was realized by effectively scanning an active pixel region across the sCMOS chip. The principle of rolling shutter data acquisition is illustrated in Fig. 2 .

 

Fig. 2 Rolling shutter mode. An activated row was marked in red, and a read-out row was marked in yellow. The band of pixel rows in between is exposed to light simultaneously. This band moves from the top to the bottom of the sensor. Its width is defined by the single row exposure time, and can be adjusted from a minimum of one line up to the whole chip.

Download Full Size | PDF

2.3 Synchronization of illumination and detection

The scanning mirror in the illumination beam path was driven by a saw-tooth voltage signal generated by a NI USB-6221 data acquisition and generation device (DAQ) from National Instruments (National Instruments Corp., Austin, TX, USA). It also generated step function trigger signals with amplitude of 5V to control the data acquisition of the camera and the acousto-optical tunable filter which defined the laser intensity. It was crucial to synchronize the output of the analog scanning curve and the two further trigger signals with utmost precision. This was achieved by using internal signal routes of the DAQ. In short, the procedure was as follows. At first, the saw-tooth and the trigger waveforms were loaded into the buffer of the device, and the channels were initialized. This was done via the USB interface and was not time-critical. Then, the device generated the time-dependent analog output voltage driving the scanning mirror. Simultaneously the two digital signals for activating data acquisition and laser illumination were issued. Pixel-line reset and readout were controlled by the internal camera clock. The data acquisition of the camera started on the rising edge of the camera trigger signal and the laser was switched on while the laser trigger output was at 5V. The scanning mirror voltage curve and the two trigger signals shared the same sample clock of 65,376 Hz. For optimal imaging performance the illumination beam position in the focal plane had to match the conjugate rolling shutter position on the sCMOS exactly during the entire image acquisition process. To achieve this we imaged the beam position in a homogeneous fluorescent solution for a set of different mirror angles. The position of the beam main axis was determined by column-wise fitting of a Gaussian function to the image data. A linear fit to this data supplied the required amplitude, offset and step size for the voltage curve driving the mirrors. When the voltage curve was subsequently applied to the scanning mirror, the maximum beam deflection corresponded to the first and last pixel rows, and the intermediate beam positions increased linearly with scanning mirror angle.

3. Results

3.1 Contrast improvement in scattering media

A focused Gaussian beam provides illumination in scanned LSFM. This pattern could directly be imaged when it illuminated a weakly scattering homogeneously fluorescent solution of Alexa Fluor 647 (Life Technologies Corp., Carlsbad, CA, USA) in phosphate buffered saline (PBS), Fig. 3(A) . However, when the same beam profile illuminated a strongly scattering sample such as fluorescent polystyrene beads (diameter 200 nm, Molecular Probes, Eugene, OR, USA) embedded in an agarose gel containing additional fluorescent molecules, an immense deterioration of the profile became visible, Fig. 3(B).

 

Fig. 3 Image of the illumination beam entering from the left into a (A) homogeneously fluorescent solution of Alexa Fluor 647 in PBS buffer and (B) sample of fluorescent polysterene beads embedded in agarose containing Alexa Fluor 647. A significant amount of light was scattered leading to fluorescence excitation outside the directly illuminated probe region. Imaging conditions and image contrast settings were kept identical. Scale bar 100 μm.

Download Full Size | PDF

A large amount of light was scattered into the sample along the Gaussian illumination beam path. The scattered light excited in-focus and out-of-focus fluorescence at large distances off the illumination beam. This greatly reduced the image quality, when images were acquired in a global shutter mode during beam scanning. Fluorescence photons that were scattered on their way through the sample to the detection device added to a uniform background. These effects were shown in the image in Fig. 4(A) . This situation could be greatly improved, when the image was acquired in a rolling shutter mode during beam scanning as shown in Figs. 4(B) and 4(C), with rolling shutter widths of 128 px. and 32 px., respectively. Figure 4(D) shows the same data taken with a rolling shutter size of 2 px. Here the rejection of light leads to a reduced signal to noise ratio.

 

Fig. 4 Images of beads in an agarose gel demonstrating the contrast difference between global shutter and rolling shutter image acquisition. (A) Image taken in global shutter mode, (B) in rolling shutter mode with a slit width of 128 rows, (C) in rolling shutter mode with a slit width of 32 rows and (D) in rolling shutter mode with a slit width of 2 rows. The image acquisition time was 22 ms in each image. Scale bar, 50 μm.

Download Full Size | PDF

The qualitative difference was most striking when performing 3D reconstructions. Again we used the sample with fluorescent beads, and imaged a 50 μm thick volume in steps of 1 μm. Figure 5 shows a comparison between 3D reconstructions of the sample imaged in global shutter and in rolling shutter mode, respectively. The global shutter image in Fig. 5(A) shows a strong background, and substantial post-processing would be required to remove it. In the confocal image in Fig. 5(B) the background is almost absent.

 

Fig. 5 3D reconstruction of a 120x120x50 $\mu {\text{m}}^{\text{3}}$volume of beads fixed in agarose. The sample volume was imaged in (A) global shutter mode and (B) with a rolling shutter of two lines. The reduction of background in the reconstructed rolling shutter data is evident. The rolling shutter data were averaged 40x to yield an SNR comparable to the global shutter images.

Download Full Size | PDF

The confocal data shown Fig. 5 was taken with a rolling shutter width of 2 pixels (8% of the width of the illumination beam) corresponding to an effective exposure time of 30 µs. A large amount of ballistic photons was excluded from detection, which reduced the overall intensity. In global shutter mode the effective exposure time approximated the time the beam needed to sweep over a certain point in the sample. This was approximately 375 μs, according to a line activation time of approximately 15 μs and a beam diameter equal to 25 pixels. Additionally, in-focus emitters were not only excited by the scanned Gaussian beam but also by light scattered from the excitation beam at other positions in the sample. Thus scattered light also caused illumination of fluorophores, in addition to the Gaussian beam profile. To achieve a comparable SNR in both data sets, and to emphasize the attainable sectioning improvement, the confocal data was averaged 40-fold.

The strongly scattering microbead-sample was used to characterize the SNR and contrast for global shutter and rolling shutter detection, respectively. We systematically varied the width of the rolling shutter from 2 to 256 rows and compared the imaging result with the global shutter mode. For the acquisition of an image the beam was only swept once across the sample resulting in an acquisition time of 22 ms for all data points.

The SNR was measured by fitting two-dimensional Gaussian functions to determine the signal amplitude ${\text{I}}_{\text{S}}$ of single fluorescent beads. Regions with no signal were used for determination of the background standard deviation${\sigma }_{\text{bg}}$. The SNR was then calculated as $\text{SNR=}{\text{I}}_{\text{S}}\text{/}{\sigma }_{\text{bg}}$ [12]. Image contrast C was measured by determining the highest and lowest intensity values in each row${\text{(I}}_{\text{max}}{\text{,I}}_{\text{min}}\text{)}$and applying the Michelson formula ${\text{C=(I}}_{\text{max}}\text{-}{\text{I}}_{\text{min}}{\text{)/(I}}_{\text{max}}\text{+}{\text{I}}_{\text{min}}\text{)}$ [13]. Finally we determined the mean of all rows. The results are given in Fig. 6 . The width of the rolling shutter was expressed in units normalized to the illumination beam diameter. Unity corresponded to the ${\text{1/e}}^{\text{2}}$-diameter of the illumination beam on the chip (25 rows). As expected, the detection slit width determined the SNR, which showed a distinct maximum around a slit size of unity. The image contrast C for rolling shutter was about twice as large as that for global shutter detection.

 

Fig. 6 SNR and contrast. (A) SNR for different rolling shutter sizes in units of the illumination beam diameter, and for global shutter. (B) Contrast as a function of rolling shutter size, and for global shutter. The rolling shutter sizes corresponding to each point are 2, 4, 8, 16, 32, 64, 128 and 256 pixels, starting from the left.

Download Full Size | PDF

3.2 Optical sectioning

We measured the microscope point spread function (PSF) by fitting a 3D Gaussian function to images of single diffraction limited particles of the data shown in Fig. 5 and averaged the results of 100 particles. The full width at half maximum (FWHM) of the PSF in axial direction was ${\text{w}}_{\text{0}}\text{=}\text{5}\text{.7±1}\text{.0}\mu \text{m}$and ${\text{w}}_{\text{0}}\text{=}\text{4}\text{.5±1}\text{.0}\mu \text{m}$for global shutter and rolling shutter (2 pixels) detection, respectively. The latter is comparable to the FWHM of the illumination beam of $\text{4}\text{.4±0}\text{.2}\mu \text{m}$.

3.3 Confocal imaging of biological samples

In order to demonstrate the superior imaging performance of the line scanning light sheet microscope with rolling shutter detection we imaged a 3D extended, scattering biological sample. To this end we used salivary gland cells of Chironomus tentans that were fixed in 4% formaldehyde solution. The transcription sites along the polytene chromosomes were fluorescently labeled using monoclonal mouse antibodies ARNA-3 CBL221 (Millipore, Billerica, MA, USA) against RNA polymerase II as primary, and Alexa Fluor 647 labeled goat anti-mouse IgG1 (Life Technologies Corp., Carlsbad, CA, USA) as secondary antibody.

Figure 7 shows the transcription sites inside the nucleus of a gland cell. The sample was imaged in global shutter mode, Fig. 7(A), and rolling shutter mode, Fig. 7(B), with a width corresponding to the illumination beam width of approximately 9 µm (${\text{1/e}}^{\text{2}}$-diameter). Hence, with an effective pixel size of 0.363 μm, the illumination beam diameter corresponded to about 25 rows. The acquisition time for each frame was 22 ms for global and rolling shutter mode.

 

Fig. 7 Images of polytene chromosomes in C. tentans salivary gland cell nucleus. (A) Image taken in global shutter mode. The beam was swept once across the sample. (B) Nucleus imaged with a rolling shutter width corresponding to the illumination beam diameter. The bands are crisper and the background in dark areas of the sample is lower as compared to (A). Scale bar, 30 μm. (C) Normalized intensity along the blue and red line in (A) and (B), respectively.

Download Full Size | PDF

In both images the prominent Balbiani ring 2 transcription site and further transcription bands were visible. In comparison to the global shutter image the rolling shutter image was sharper and had a higher contrast. Thus, it was easier to distinguish transcription sites from the background. Furthermore, the total background signal in the dark regions of the image was lower in the confocal mode as compared to the global shutter mode. Figure 7(C) shows an intensity plot normalized according to ${\text{I}}_{\text{norm}}\text{=}{\text{I/(I}}_{\text{max}}\text{+}{\text{I}}_{\text{min}}\text{)}$along a vertical line in the images as marked in Figs. 7(A) and 7(B). The plot illustrates that the chromosome bands showed a better contrast when imaged in rolling shutter detection.

4. Discussion

We developed a confocal line scanning light microscope combining the benefits of selective plane illumination and confocal line detection. A digitally scanned Gaussian laser beam, which was focused into the sample perpendicular to the detection axis, was used for fluorescence excitation. Rolling shutter data acquisition using a scientific CMOS camera provided synchronized confocal line detection.

Similar results can be achieved by taking multiple images that were illuminated row by row, cutting out the illuminated rows and stitching them back together [8]. Obviously, this is a time and computationally intensive process. Alternatively, structured illumination created with a scanned laser beam can be used to enhance image contrast. Here, the illumination beam is switched on and off repeatedly during scanning to produce a sinusoidal illumination pattern [14]. The sample has to be illuminated three times with the pattern phase shifted by 60 degrees between each acquisition. The final contrast-enhanced image is then calculated from the structured illumination data. A related technique uses a uniformly illuminated image and a structured illuminated image to numerically remove contrast-degrading background [7]. However, contrast-enhancing techniques based on illuminating the sample with a structured pattern ultimately depend on the contrast between bright and dark parts of the pattern and thus may suffer from low SNR or high background signals. Compared to these existing approaches to confocal or contrast improved light sheet microscopy, our approach has several advantages. Only a single, intrinsically confocal image has to be acquired. This reduces the risk of artifacts due to sample drift between subsequent image acquisitions. No post-processing or numerical background elimination from multiple images is required. Notably, no special optical and mechanical elements for de-scanning like in conventional confocal microscopes are required, since the rolling shutter represents a slit detector that moves synchronously to the illumination beam in a conjugate image plane. This simplifies the setup, improves mechanical stability and most importantly improves fluorescence intensity. However it should be mentioned that the Hamamatsu Orca Flash 2.8 is an early prototype of a scientific CMOS camera where the rolling shutter moves continuously from the top to the bottom of the image sensor. Its successor, the Hamamatsu Orca Flash 4.0 as well as current models from other manufacturers, have a considerably larger image sensor. The trade-off is that the chip consists of two parts, where each half has its own rolling shutter. These are moving synchronously outwards from the center. Thus, in order to be able to take advantage of the whole image sensor, the sample would have to be illuminated from both sides with counter-directed moving beams, which might lead to artifacts at their overlap. Of course, the most elegant solution would be to utilize a camera with rolling shutters moving in the same direction and two parallel beams. Furthermore, the time needed for the rolling shutter to run from the top to the bottom of the chip is set to 22 ms by the camera electronics. In order to achieve a rolling shutter width of one line, the exposure time had to be set to 22 ms divided by the total number of pixel rows, which is approximately 15 μs and equals the line cycle time. Thus the only way to adjust the rolling shutter size is by setting the exposure time to the desired width multiplied by 15 μs. A straight forward solution would be to make the cycle time between activation of rows adjustable by the user.

Despite the fact that optical confocality is enhanced only perpendicular to the direction of rolling shutter movement, we could significantly improve the image quality. We measured an increase in optical sectioning capability for small rolling shutter sizes and could demonstrate the improvement in contrast and SNR without the need of repetitive data acquisition like in structured illumination techniques.