Practical Guide for Excellent GSDIM Super-Resolution Images

Kees:I have always been interested in what makes a collection of dead chemicals ‘alive’, in other words, how does a cell work. Life is in the number and types of molecules, in their detailed localization and in their interaction.

Most molecules cannot be seen, not even with an ideal microscope. They have no color and are too small. Over the past years, tremendous advancements have been made in coloring the molecules: we have Fluorescent Proteins in all colors to tag proteins, we have probes to stain specific parts of the DNA and we have sensors that are able to see small-molecule messengers like Ca²⁺ and many more. With these, live-cell imaging allowed us to follow these invisible molecules in time and space. But only to the resolution of light microscopy, that is about 100 times lower resolution than what we would need to see molecules.

Functional imaging techniques like Fluorescence Resonance Energy Transfer (FRET), Fluorescence Recovery after Photobleaching (FRAP) and Fluorescence Lifetime Imaging (FLIM) allow us to see molecular interactions. But again, only with 0.25 µm resolution, and the experiments are labor-intensive. And one must also realize that the outcome of such experiments could be negative, even if the two molecules interact.

GSD and Super-Resolution are sort of the logical next step in obtaining sufficient resolution to study molecules and their interactions (Figure 1).

Kees: When we started working with the Leica SR GSD localization microscope we extensively studied sample drift using different commercially available dishes and mounting procedures. By analyzing long time-lapse imaging series taken from coverslips with sub-resolution fluorescent beads (Invitrogen) the Chamlide CMB magnetic chamber (Live Cell Instrument, Seoul, South Korea) was found to have minimal drift. In addition we place the sample on the microscope stage 20 min prior to imaging to let the sample settle down. If x/y drift is notable, we make a correction routine for that by using a home-built “de-jittering” routine or alternatively the analysis package ThunderSTORM.

We found that during image acquisition artefacts can be introduced in these already impressive images, especially when the background is very high. Existing background subtractions only partially solved that. So we worked together with Eelco Hoogendoorn and Marten Postma from the University of Amsterdam (UvA) who implemented a new background subtraction method: a running time-averaged median filter (Figure 3). This works great. [2]

We then found that at such high magnification, small image errors that go unnoticed in the diffraction limited world can no longer be ignored. Chromatic aberrations are notably visible as was shown by Daniela Leyton in our lab, who was working on a protein that translocated to the plasma membrane after stimulation of cells. The question was whether it translocated to the plasma membrane lipid bilayer, or merely associated with the cortical actin just adjacent to that, a question that seems tailored for super-resolution.

Initially, Daniela found Clic4 almost always just next to the lipid bilayer, but depending on the orientation of the cell, they appeared sometimes in and sometimes just outside the cell. Suspecting chromatic aberrations to be the cause, together with physicist Bram van den Broek, 0.1 µm Tetraspec microspheres embedded in a matrix were imaged in the red, green and blue channel. With these data, we quantified chromatic aberrations and were able to precisely correct them with an affine transformation matrix using a home-built Image J macro (Figure 4). [3]

Please keep in mind that chromatic aberration is dependent on the objective in use and has to be corrected individually, but once characterized, the calibration remains valid throughout the years.

Daniela also spent much time to compare fixation protocols for super-resolution microscopy. The occurrence of fixation artifacts - and how to deal with them - is well documented in electron microscopy. Yet, these protocols cannot be simply copied for SR imaging because preparations are often imaged in aqueous buffers.

For the combined imaging of the actin cytoskeleton and associated actin-binding proteins, optimal results were obtained by the protocol in Leyton-Puig et. al, 2016. [4]

Kees: Many! Too many to mention them all. For example, we observed that the molecular makeup of hemidesmosomes, the adhesion structures that make sure your skin doesn’t come off the rest of your body, are not quite as they were thought to be based on earlier electron microscopy and widefield fluorescence microscopy.

We found that stretches of DNA may enter empty spaces (a sort of pockets) in the lamin network at the inner leaflet of the nuclear membrane. And we are studying cytoskeletal components, cell surface receptors and vesicular proteins, all of which bring up new scientific insights, and also new scientific questions. A striking observation was also that the intermediate filament network seems to closely associate with microtubules, and this is seen predominantly in the periphery of the cell. The technique is so powerful that we receive requests constantly from groups that want to collaborate.

Another important insight is that quantitative image analysis is urgently needed – and more within reach than even before. With resolution far superior to fluorescence microscopy, and labeling density far superior to EM, we can start to count the number of receptors in a cluster, to quantitate the distance between filaments and the proteins that anchor them, we can quantitate spatial separation of two different molecular species and much more. But traditional tools like the Manders coefficient or Pearson correlation coefficient do no longer apply! Single-molecule localizations by definition do not overlap, so rather we have to ask the question: how close are the two molecular species, on average? That requires new analysis tools. [5,6,7,8,9]

Kees: The new sCMOS camera in our hands has several advantages. It delivers (almost) the same image quality at ~500 Hz, i.e. about 5 times faster than CCD. The camera can do 1500Hz but in that case we are not collecting enough photons for the highest quality images.

5 times faster means 5 times more experiments before your buffer runs out, much less drift and much less bleaching. It also works extremely well in low-power mode (40 x 40 µm field of view).

But the most exciting is that we start to do GSD in living cells. We found early on that mVenus blinks quite well when excited with 488 nm laser.