Actin fibers in fixed U2OS cells acquired in dual-view geometry on a reflective coverslip. Credit: Image adapted with permission from Wu et al. (Springer Nature).

In light-sheet fluorescence microscopy (LSFM), a thin layer of a sample is illuminated perpendicularly to the observation objective. This reduces photodamage and out-of-focus fluorescence while providing diffraction-limited resolution. These features, in combination with its high speed, make LSFM an attractive method for time-lapse imaging in a wide range of samples—from cells to living animals and thick tissue samples.

In the popular diSPIM LSFM configuration, two objectives are placed at 45° angles above the sample. One objective generates the light sheet, while the second objective is used for imaging. This process is repeated with reversed objectives to generate a 3D image with similar resolution in all directions. Often, high numerical aperture (NA) objectives are used to achieve a high resolution. But these short-working-distance objectives are bulky and can be difficult or impossible to arrange in the small available space around the sample, which is a restriction especially when imaging small samples in LSFM.

Hari Shroff and colleagues from the NIH National Institute of Biomedical Imaging and Bioengineering in Bethesda, Maryland, have come up with a simple trick to address this issue—they utilized additional views of the specimen without the need for further expensive objectives or complicating the microscope design. “If one just grew the sample on a mirrored coverslip, [one] would very naturally get these extra views of the sample,” says Shroff. The light sheet would reflect off the mirrored coverslip and create a second orthogonal light sheet. This doubles the acquisition speed, as it allows simultaneous rather than sequential diSPIM imaging. In addition, the fluorescence that would normally be lost through the coverslip would also be reflected, leading to two additional images (one for the original and one for the reflected light sheet). By fusing these images, signal-to-noise ratio and resolution are considerably improved compared to conventional LSFM.

However, the fusion of the four views requires sophisticated modeling and computational postprocessing. “We constructed a very simple model which turned out to be not enough,” explains Shroff. At that point, his group teamed up with Patrick La Riviere's computational group at the University of Chicago. La Riviere's group developed an algorithm that modeled the imaging process better than the initial approach. “And then we [went] back and forth [to see] what kinds of features of the model were important for capturing the data,” explains Shroff. The model was especially important to remove background fluorescence and account for the fact that the light sheets spread diffractively. The latter also affects nonmirrored LSFM images but has to date been largely ignored, as mathematical tools to model the spreading of the light sheet and correct for it have not been available, according to Shroff. The new deconvolution algorithm can now be used for this purpose. The only price one has to pay is additional computational time. But the researchers have already tuned the algorithm and substantially reduced the postprocessing time since the publication of the paper. And Shroff expects further improvements in this area with the use of faster graphic processing units.

Shroff and colleagues demonstrate the benefits of their approach in a variety of examples, one of them being a freely moving Caenorhabditis elegans embryo. They employed a strain that expresses GCaMP3, thus enabling functional imaging of Ca2+ transients in cells. Extending earlier work from Shroff's group, the researchers were able to visualize Ca2+ transients even in neurites while maintaining a scanning speed of more than 1 Hz.