Live-cell imaging of dendritic spines by STED microscopy

To be able to directly compare the performance of STED with confocal imaging, we acquired images of dendritic spines in both modes by switching the STED laser beam on or off in an alternating fashion (Fig. 1B). The pulse peak intensity at the doughnut crest was I = 400 MW/cm², which in the case of negligible aberrations yields a focal plane resolution of Δr ≈ 69 nm. From electron microscopy studies, it is known that spine necks, which are the elongated structures that connect the heads of spines with their parent dendrite, are among the smallest structures enclosed by the plasmalemma of neurons. As a consequence, many spine necks are beyond the limit of what conventional light microscopy is capable of reproducing in detail.

To assess how faithfully each mode can measure the size of the diameter of spine necks, we determined the full width at half maximum (FWHM) of line profiles of pixel intensities across the most narrow part of the spine necks. Fig. 1C shows that the normalized line profile for a typical spine is sharper in the STED than in the confocal mode, reflecting the fact that spine necks appear thinner in the STED than in the confocal images. We then plotted the FWHM values for a population of randomly selected spines, which were imaged in both modes (Fig. 1D). The frequency histogram shows that spine necks appeared consistently and significantly narrower when imaged in the STED as compared to the confocal mode (FWHM_STED = 0.129 ± 0.025 μm vs. FWHM_confocal = 0.233 ± 0.022 μm; mean ± SD; n = 47; P < 10⁻³⁰; paired t test). Importantly, the STED distribution (coefficient of variation [CV] = 0.195) is broader than the confocal distribution (coefficient of variation = 0.094), which is slightly skewed to the right. The CV is the ratio between the standard deviation and the mean, and represents a normalized measure of the dispersion of a probability distribution. Given that the standard deviation of the values for confocal and STED are comparable and that the mean STED value is about half the size of the confocal value, the CV is accordingly greater for STED than for the confocal mode. This is consistent with the view that the STED mode captures the variability of the diameters of the spine neck more faithfully than the confocal mode for which the lower values are tightly bound by the diffraction limit. Consistently, the mean FWHM_STED value is about 70% larger than the mean of the three smallest values in the distribution (approximately 77 nm), while the mean FWHM_confocal value is only by about 15% larger than the smallest values in the confocal distribution (≈200 nm).

To assess the size of structures that can still be faithfully measured by confocal microscopy, we plotted the ratio (STED/confocal) of FWHMs measured in either mode against the width in the STED mode (FWHM_STED) (Fig. 1E) of a range of structures, including small ones such as spine necks and larger ones such as dendritic shaft diameters. The relationship is linear for small values of FWHM_STED (<0.2 μm), while it asymptotes to unity for larger values (more than −0.3 μm), reflecting the fact that the FWHM values bottom out earlier for the confocal than the STED mode. This analysis confirms that the performance of the confocal mode is severely compromised for structures with sizes approaching its minimum focal spot FWHM of approximately 200 nm.

Compared to the conventional light microscopic pictures, the clarity of the images of the dendrite and the associated dendritic spines is significantly enhanced, revealing structural details reminiscent of volume reconstructions of electron microscopic serial sections. Fig. 2 shows two examples of volume rendered images of stretches of dendrite, based on a 3D image stack acquired by STED microscopy (also see supporting information (SI) Movie S1). Note that in these experiments, the axial resolution is that of a standard confocal system. However, recent advances show that the axial resolution can also be substantially improved by STED microscopy (22).

The images were acquired in the range of 0–10 μm above the surface of the coverslip without showing any decrease in resolution as might have been expected for STED imaging in optically dense tissue. The use of water objectives will help improve the depth penetration, given that oil objectives (as used in our experiments) produce more aberration because of the mismatch of refractive indices.

Next we examined whether this technique also lends itself to time lapse imaging. Since the peak intensity for STED (I = 400 MW/cm²) is larger than the typical focal intensity of an excitation beam by about three orders of magnitude, we examined potential effects of phototoxicity and bleaching, which could make it impossible to acquire multiple time points. Fig. 3A illustrates that, at a practical level, neither phototoxicity nor bleaching associated with STED imaging were prohibitive. The panels show nine successive image frames of YFP-labeled dendrites out of a time lapse series of 20 image frames in total (see Movie S2), all of which were individually normalized to the respective peak value of pixel intensity. We extended the time lapse imaging to series up to 100 frames (data not shown), as well as to high-resolution 3D stacks (Fig. 2). These data demonstrate that the morphology of spines can be studied in time lapse experiments with a resolution <100 nm without compromising the integrity of the synaptic structures or the signal-to-noise ratio. This is not surprising given that (a) the STED beam intensity (I = 400 MW/cm²) is much lower than intensity typically used for two-photon microscopy (200 GW/cm²), and (b) that the absorption by water is much lower at the wavelength of 598 nm than in the near infrared range, resulting in negligible temperature rises in the focal region (<0.2K) (23).

To determine the degree of bleaching, we measured the pixel intensity averaged over the area of spine heads (n = 14 in total) and plotted this value over consecutive image frames acquired without interruption at 20 sec/frame (Fig. 3C). The graph illustrates that the pixel intensity decayed over the first few frames, before settling on a new level of around 50% of the initial value. This time course is consistent with the idea that the fraction of YFP that bleaches becomes replenished by diffusible YFP from areas within the dendrite that were not exposed to the beams. Therefore, one would expect the level of fluorescence to drop initially, converging on a steady-state value determined by the counteracting rates of bleaching and diffusion. Taken together, these data indicate that high-resolution STED microscopy is compatible with time lapse imaging, in particular if YFP is used as a volume marker. Our experiments indicate that STED imaging of neurons should work with any fluorescent indicator that is freely diffusible and which is quickly and continuously replenished after latent bleaching. This is likely to be the case for other genetically encoded indicators as well as small synthetic indicators such as Ca²⁺ sensitive dyes.

Finally we used STED microscopy to image morphological plasticity of dendritic spines. To induce the structural changes we used the well-established synaptic plasticity paradigm of “chemical LTP,” which involves raising neuronal activity in the entire slice by briefly perfusing it with a modified ACSF solution that globally potentiates synapses. Visual inspection showed that in the respective field of view, 40–60% of the spines changed shape after the stimulation (number of experiments = 4), while fewer than 10% of the spines appeared plastic under unstimulated conditions (n = 5). Alternatively, we used local tetanic stimulation of afferent fibers projecting into the field of view which is a standard protocol for the induction of LTP in a synaptic pathway, which induced similar changes as the chemical LTP protocol (data not shown).

While it confirmed previous reports on activity-dependent size increases of dendritic spines (5), time lapse imaging using STED revealed details that are very difficult to detect by established light microscopy approaches, such as subtle changes in the shape of spine heads. In fact, due to the lack of resolution it had not been possible to examine structural changes other than simple changes in spine size. STED imaging now overcomes this limitation and permits studying the fine structure of spines in living tissue.

Fig. 4 illustrates the types of structural changes that can be observed by STED imaging. The images were acquired before (leftmost image) and after the plasticity-inducing stimulation at the times indicated. Fig. 4 shows examples of plastic spines, which change in size and shape considerably. Remarkably, the changes in shapes usually led from smaller and amorphous structures toward larger and more differentiated ones, often taking on cup-like shapes. For example, Fig. 4C shows a series of images of round spine heads morphing into cup-like structures. In addition, Fig. 4D shows a spine with a large head, which grows out before settling in a new position, where it changes shape from bulbous to cup-shaped. While some of these changes may represent instances of spine heads splitting into two parts, as reported by electron microscopy, others rather suggest a scenario whereby the spine changes shape after the plasticity-inducing stimulation to come into contact with its presumptive presynaptic partner.

These examples highlight the potential of STED microscopy for live cell imaging of genetically encoded fluorescence within the important neurobiological context of the structural plasticity of spines and synapses. STED microscopy may even make it possible to assess functional synaptic connectivity simply from the synaptic morphology alone, replacing more complicated approaches such as correlative electron microscopy or functional labeling.

Hippocampal slices (300 μm thick) from postnatal day 5–7 wild-type C57/BL6J mice were prepared, embedded in a plasma clot on glass cover slips, and incubated up to two weeks in a roller incubator at 35 °C, according to the Gähwiler method (24). The age of the slice cultures for the experiments ranged between 14 to 21 days in vitro (DIV) after the preparation. For the experiments, cultures were transferred into a heated chamber (35 °C), where they were maintained in standard ACSF solution containing (in mM): NaCl 126, KCl 2.5, CaCl₂ 2.5, MgCl₂ 1, glucose 10, and Hepes 20. The pH was adjusted with NaOH to 7.3.

Slices were imaged over a baseline period (typically acquiring three time points) before they were subjected to two different experimental paradigms that are commonly used to induce functional and structural plasticity at hippocampal synapses. The ACSF designed to induce “chemical LTP” contained (in mM): NaCl 124, KCl 5, MgCl₂ 0.1, CaCl₂ 5, glucose 10, TEA 25, and Hepes 29, and its pH was adjusted to 7.3. This modified ACSF solution was washed in for a period of 5 min before it was washed out by the standard ACSF solution. In the experiments using electrical stimulation, brief (0.2 ms) current pulses (10 μA) at 100 Hz for 1 sec (repeated 5 times at 60 sec intervals) were delivered via a patch pipette positioned under visual control and with the help of a micromanipulator within 20 μm of the field of view that contained the dendrites that were imaged.

We used excitation pulses of 100 ps and 490 nm that were synchronously followed by 300 ps STED pulses having a central wavelength of 598 nm. The 80MHz pulse trains were focused by a 1.4 NA objective lens (PL APO, 100×, oil, Leica). The excitation pulses were delivered by a pulsed laser diode (Toptica) at an average focal power of 1.7 μW. The fluorescence collected by the lens was separated by a dichroic mirror, filtered via a 525/60 bandpass and imaged onto a multimode optical fiber. The diameter of the fiber amounted to 80% of the back-projected Airy disk and therefore functioned as a confocal pinhole. The STED pulses originated from an optical parametric oscillator (OPO, APE) fed from a Ti:Sapphire laser (MaiTai, Spectra-Physics) operating at 800 nm. Originally of 200 fs duration, the OPO pulses were stretched by dispersion in a polarization preserving fiber (OZ Optics) of 120 m length to approximately 300 ps. The time-averaged power P_s of the STED beam reducing the fluorescence by half was identified in a separate measurement as 0.9 mW. The excitation pulses were synchronized with the STED pulses via external triggering of the laser diode; the delay was tuned using a home-built electronic delay generator. The expanded STED beam passed through a polymeric phase plate imprinting on the wavefront a helical phase delay of exp(iφ), with 0 < φ < 2π (RPC Photonics) which delivers a doughnut at the focal region of the lens. The count rate was compensated for the dead time of the avalanche photodiode (APD, SPCM-AQR-13, PerkinElmer) by multiplying the measured count rate with a correction factor of 1/(1-t_d*CR); t_d = 50 ns is the dead time of the APD and CR the measured count rate.

Images were acquired with a pixel size of 29 nm over a field of view of up to 25 μm × 25 μm in size. At least one image frame was taken in the confocal mode before STED imaging began. Depending on the size of the field of view that was imaged, the frame rate was 20 or 40 sec/frame. Images or image stacks (Δz = 0.5 μm) were acquired repeatedly (typically 20 frames in a series, but up to 60) at variable time points (at the maximal frame rate or at 10 min intervals) without any signs of photo damage. The imaging depth typically varied between 0 and 10 μm below the surface of the cover slips.

Images were analyzed using ImageJ. Values of FWHM were extracted from line intensity profiles of the narrowest part of spine necks using a fitting routine written in Matlab (Mathworks). Statistical significance was tested using SigmaPlot 8.0 (SPSS Inc.). All images shown are raw, unprocessed images, except for the volume-rendered image stack in Fig. 1B, which was generated using Imaris (Bitplane AG).