Microscopy Matters

Insights from the Advanced Imaging Center at Janelia

Structured-Illumination Microscopy: What types of samples are not suitable and why

Structured-Illumination Microscopy: What types of sample are not suitable and why

Needless to say that, just like in any type of microscopy, low signal samples in general do not work well for SIM. Besides that, there are other factors unrelated to the general signal level that will determine if a sample, either fixed or living, is suitable for SIM. The following is by no means a complete list of those factors.

Thick (> 15 µm) samples

The issue with thick samples is two-fold. 3D SIM relies on the interference of three laser beams to form the illumination pattern at the in-focus plane in the sample and that the higher contrast of the pattern (defined as (peak-trough)/peak), the more reliably SIM can extract high-resolution information. Before hitting the sample at the sample-cover glass interface (Fig. 1A), the three beams are nearly perfect plane waves (i.e., their wavefront is a plane) and all s-polarized (i.e., polarized along the direction perpendicular to the plane of incidence formed by the three beams). That is the ideal condition for producing interference fringes of the highest contrast in which the troughs of the fringes have zero intensity. After the beams travel into the sample (Fig. 1B), the plane waves would experience changes to their wavefronts and polarization state because of the inhomogeneity of the sample; the deeper they travel the more distorted from the original s-polarized plane waves they become and the condition for high-contrast interference becomes less ideal. At some point the fringe contrast would become so low that no meaningful high-resolution information can be obtained by SIM.

thick_sample_aberration

Figure 1

The other problem caused by thick samples pertains to the fact that 3D SIM is fundamentally a wide-field epi-fluorescence microscope in which out-of-focus background is the norm. Thicker samples with fluorophores throughout would produce stronger out-of-focus background. In theory 3D SIM should be able to differentiate the out-of-focus from the in-focus signal (i.e., capable of optical sectioning), but in reality out-of-focus background can deteriorate SIM quality quickly because of shot noise. Imagine your sample is a layer of beads flanked axially by a thick layer of uniform fluorescent solution. When the bead layer is in focus and being imaged, for sure there will be high background but in theory one should be able to discern the beads signal standing out from the background. In reality, though, the high background comes with high shot noise because shot noise is proportional to the square root of the number of photons. If the fluorescent solution layer is thick enough to make the background high enough, the beads signal would be completely inundated in shot noise and indistinguishable from the background. This is an extreme scenario, of course, for which the only imaging technique that may help is confocal microscopy. In general, higher background results in lower signal-to-noise ratio and consequently lower SIM quality.

For fast 3D SIM imaging of live samples, thicker samples would require more slices to be imaged and thus longer acquisition time, which would not only slow down acquisition but also exacerbate photobleaching.

Samples with low contrast

Each individual fluorophore in a sample with low contrast can be considered being surrounded by its similar peers in all directions in a three-dimensional space. In a manner very similar to in-focus signals being immersed in out-of-focus background, the surrounding fluorophores create a high background for the central fluorophore because the former greatly outnumber the latter and a point source’s image creeps into its neighbors in the form of the point spread function. Therefore, borrowing the same reasoning used above, images of samples with low contrast, however bright they may be, are inherently low in signal-to-noise ratio and thus not suitable for SIM, 3D or 2D.

Another way to look at this is by considering the essence of SIM image formation – structured illumination helps bring conventionally unobservable high-resolution information into the observable region and later the algorithm high-pass filters those information components and then rearranges them spatially in reciprocal space based on where they originate. Since low-contrast samples by definition have little high-resolution information, what is in the high-resolution components is simply noise. After noise contained in those components is high-pass filtered, it loses its uniformly fine-grained appearance most people are familiar with from raw images; instead it appears more coarsely granulated and more like artifacts. The final reconstruction would thus simply be a conventional image overlaid with artifacts.

Samples with high dynamic range

Here high dynamic range means sparsely labeled (i.e., dim) parts of the sample coexist with densely labeled (i.e., bright) parts in the immediate vicinity of each other. The bright spots can act as noise polluting source for their dimly fluorescent neighbors because, again, bright light results in high background in neighbors and hence high shot noise that can easily inundate dim signals. Another harm produced by bright spots is due to the inevitable ringing artifacts as a result of digital signal processing – a peak after high-pass filtering becomes a narrower peak but with oscillating side lobes surrounding the main peak (Fig. 2) with the side lobes’ intensity proportional to the peak intensity. The bright spots in the final SIM images therefore spill into their dim neighbors, both laterally and axially, in the form of side lobes, whose intensity can overwhelm the dim signals that rightfully reside there.

Figure 2. Blue: raw data; Red: after high-pass filtering

High dynamic range is partly why actins are more challenging to image than microtubules (the other main reason is that actin produces a higher diffuse background because it is everywhere in cytoplasm). Actins can appear as single fibers, moderately bundled fibers, or highly bundled stress fibers (Fig. 3A), whereas microtubules (except in mitosis) mostly come in isolated single fibers with nearly equal brightness (Fig. 3B).

tubulin

Figure 3. A: Mouse embryo fibroblast (MEF) labeled with phalloidin-Alexa568. B: MEF whose tubulin is immunofluorescence-labeled with Alexa488
Lin Shao, Ph.D.

Author: Lin Shao, Ph.D.

Applications Scientist

Comments are closed.