Aberration compensation for confocal and STED imaging

In the next step we studied the impact of spherical aberrations on actual confocal and STED images. For this we again prepared bead samples, but instead of gold colloid we applied a thin layer of fluorescent beads. These beads are small polymeric spheres filled with a fluorescent dye; for these experiments we used 40nm large spheres filled with a yellow-green emitting dye. Because these beads tend to aggregate in the colloidal solution, they needed to be sonicated thoroughly before applying them to the coverslip. We imaged small 18µm×18µm areas covered in fluorescent beads, again using the entire range of the correction collar to compensate aberrations to a varying degree and to determine the optimal compensation. If optimally corrected we observed confocal resolutions of190nm to 200nm and STED resolutions of60nm to70nm at de-excitation powers of≈40mW

in the objective aperture. The STED resolution was slightly worse than can be expected from this laser power, because the wavelength of the de-excitation laser was optimized for use with EYFP and not for the yellow-green dye in the fluorescent beads. The yellow-green dye is blueshifted by roughly10nm, resulting in noticeably less STED efficiency of the595nm de-excitation beam.

The aberrations clearly distorted the images in the expected ways (Section 3.2.2).

When imaged confocally at constant laser power the recorded signal dropped significantly if the correction value was further than two steps away from the optimal position (Figure 3.2 a). This went hand in hand with a diminished signal-to-noise ratio. The full-width at half-maximum (FWHM) of the imaged beads increased slightly in the xy-plane, but was stretched significantly along the optical axis. With activated STED beam the effects were even more dramatic, due to the interplay of not one but two aberrated beams. Besides the observed dimming and lengthening of the FWHM along the axis, the subdiffraction resolution in the optical plane became severely compromised with increasing aberrations. Even when the reduced signal was compensated with increased excitation powers, the STED resolution decreased from optimal 60nm to 70nm to around 170nm to 180nm, almost to diffraction limited levels (Figure 3.2 b). Of importance is that the correction values were identical when determined by either confocal or STED imaging. The correction values were also identical when determined independently from either the dimming and z-lengthening effects and matched the correction values determined from the PSF measurements (Figure 3.3). This meant that the system could be calibrated using purely confocal images, thereby reducing the illumination intensity and concomitantly alleviating bleaching and photodamage problems inside the more delicate live samples. Furthermore, the calibration could be done in live, fluorescent samples either by maximizing the image brightness (whilst considering possible bleaching artifacts) or by minimizing the observed length along the z-axis, whichever proved to be more convenient.

The aberration compensation was more difficult when determined using structures in three-dimensional systems. One major problem was caused by system instability during operation of the correction collar. Because the correction ring is located on the objective lens itself, by necessity the objective will be moved slightly at

3.2. Measuring and compensating spherical aberrations

Figure 3.3. | Calibrating the correction collar by imaging fluorescent structures was possible either by maximizing the recorded signal of a structure, or by minimizing the apparent length in z. Here, fluorescent beads were imaged along the xz plane for different correction values. The peak intensity and the inverse length of the beads along the z-axis were normalized and plotted against the correction value (CV) of the objective. The maxima of both curves overlap, giving identical results of the optimal correction CV=6.5.

each step of the manual alignment process. Even in a fairly stable system, minute changes of the objective position could move the imaged structure outside the field of view, requiring the image frame to be recentered. In densely labeled systems the exact feature used for calibration might be difficult or time-consuming to relocate, especially if structures that are indistinguishable in the widefield were used, such as fluorescent or even reflecting beads. Further complications arose from the aberration-induced length distortions along the optical axis, which caused structures to shift in and out of the focal plane as the aberrations changed.⁸⁸ To compensate this effect, we recorded stacks of frames spanning several µm when adjusting the correction by maximizing the fluorescent signal of a structure best visible in the xy-plane. This way the imaged structure remained in the imaged volume, despite occurring drifts in z. Adjusting the correction by visualizing distortions in z was also not trivial, due to the instability of the system during correction. If the imaged structure, such as a dendritic shaft, varied more strongly in width and brightness along its length than the slight changes induced by spherical

Figure 3.4 | Dendritic spines resolution, but suffer from low im-age contrast. (C) When corrected, STED images of dendritic spines reveal prior hidden details, such as the curvature of the spine head cup and inhomogeneities in neck

aberrations, then manual shifts could ruin a measurement. Nevertheless, the correction procedures tested on two-dimensional samples could be applied in live brain samples. The benefits of proper aberration correction were clearly evident in the recorded images, as can be seen in Figure 3.4.

It was not possible to establish any universal calibration curve of imaging depth over correction value. Even for artificial systems, such as three-dimensional samples consisting of beads embedded in Moviol (refractive index of n=1.45), the measured calibration curve varied notably from sample to sample. A rough correction correlation of 1 unit / 10µm could be determined, but this was never more than a rule of thumb. This became increasingly clear when imaging deep inside live brain slices, which were by nature far more heterogeneous than the artificial systems.

Optimal correction values for individual locations at the same depth but at different lateral positions within a brain slice could vary considerably. The deviations were more prominent when comparing image locations from different anatomical areas inside the hippocampus, reflecting the high variability of the hippocampal structure.

Regions with a higher density of soma demanded different correction values than regions with dense axon pathways or a higher density of dendritic arbors. Not all locations in a slice were suitable for imaging: local anomalies in the brain slices, such as pockets of water between the coverslip and brain slice, possibly coinciding

3.2. Measuring and compensating spherical aberrations

with a strongly curved tissue interface could induce uncorrectable aberrations. Also, groups of dead or dying cells also hindered or even prevented imaging, as they rapidly became opaque and swelled, leading to severely increased scattering along with strong drifts of any structure in the immediate area. All in all, the strong inhomogeneity characteristic of the hippocampal brain slices necessitated separate aberration correction procedures for each individual image location. According to the nature of spherical aberrations, individual corrections could almost be ignored close to the coverslip but became increasingly obligatory at large imaging depths.