Conceived and designed the experiments: II CGS AT MD. Performed the experiments: II CGS. Analyzed the data: II CGS. Contributed reagents/materials/analysis tools: XD ML CZ. Wrote the paper: II CGS.
The authors have declared that no competing interests exist.
The actin cytoskeleton of dendritic spines plays a key role in morphological aspects of synaptic plasticity. The detailed analysis of the spine structure and dynamics in live neurons, however, has been hampered by the diffraction-limited resolution of conventional fluorescence microscopy. The advent of nanoscopic imaging techniques thus holds great promise for the study of these processes. We implemented a strategy for the visualization of morphological changes of dendritic spines over tens of minutes at a lateral resolution of 25 to 65 nm. We have generated a low-affinity photoconvertible probe, capable of reversibly binding to actin and thus allowing long-term photoactivated localization microscopy of the spine cytoskeleton. Using this approach, we resolve structural parameters of spines and record their long-term dynamics at a temporal resolution below one minute. Furthermore, we have determined changes in the spine morphology in response to pharmacologically induced synaptic activity and quantified the actin redistribution underlying these changes. By combining PALM imaging with quantum dot tracking, we could also simultaneously visualize the cytoskeleton and the spine membrane, allowing us to record complementary information on the morphological changes of the spines at super-resolution.
Fluorescence microscopy using genetically encoded fluorescent proteins has greatly advanced our understanding of many functional biological systems over the last decade. However, the precision at which cellular structures can be visualized has been limited by the spatial resolution imposed by the diffraction limit of light (∼250 nm). Novel super-resolution imaging methods using photoactivatable proteins or photoswitchable fluorophores bypass this limitation and have the potential to revolutionize the experimental scope of light microscopy
Dendritic spines are small cellular structures that compartmentalize the sites of excitatory neurotransmission in neurons
Recent studies have combined the use of photoactivatable probes and single particle tracking (SPT) to study the kinetics of the actin cytoskeleton of spines
To study the organization of the actin cytoskeleton we designed an actin probe that combines an actin-binding peptide (ABP) sequence
When expressed in COS-7 cells, ABP-tdEosFP labels filamentous actin structures with high specificity, as judged by the colocalization with phalloidin labeled F-actin (
The spatial resolution of PALM imaging depends on the ability to accurately determine the position of each single molecule, and ultimately on the number of detected photons emitted by the fluorophore, for EosFP typically in the range of hundreds of photons
To confirm the reversibility of the binding of ABP-tdEosFP to F-actin we also performed photoactivation experiments in live COS-7 cells (
Taken together, these experiments show that 1) ABP-tdEosFP has a high specificity for F-actin, 2) that its low affinity allows ABP-tdEosFP to exchange between different F-actin binding sites and 3) that PALM imaging of ABP-tdEosFP yields super-resolution images of the actin cytoskeleton down to a spatial resolution of ∼25 nm.
We then transfected rat hippocampal neurons with ABP-tdEosFP and fixed the cultures at 3 to 4 weeks
(A) Conventional fluorescence microscopy of a rat hippocampal neuron expressing ABP-tdEosFP and fixed at DIV 25 (unconverted form of the fluorophore). (B) Super-resolution PALM image of the same dendritic segment, acquired with a frame rate of 50 ms (8000 frames). (C) Profiles across the dendritic shaft (a), the spine neck (b), and the spine head (c), as indicated in (B). (D) Detail of a single dendritic spine. (E) Quantification of the spine parameters indicated in (D): neck length (panel 1; N = 48 spines from 4 cells and 3 independent experiments), neck width (2; N = 48, 4 cells, 3 experiments), spine head diameter (3; N = 48, 4 cells, 3 experiments), and diameter of actin-free regions (‘hole’, 4; N = 21, 4 cells, 3 experiments). (F) Two-color PALM/dSTORM reconstruction of the spine actin cytoskeleton (in red) in relation to the postsynaptic density protein Shank2 (green).
Using PALM image reconstruction we obtained high-resolution images of mature dendritic spines (
Large spine heads were either cup-shaped or mushroom-shaped, in which case they frequently enclosed a region in which no ABP-tdEosFP signals were detected (
Closer inspection of the high-resolution PALM images shows that actin is not homogeneously distributed within the spine (arrowheads in
Given the role of the actin cytoskeleton in morphological aspects of synaptic plasticity
In order to determine the best conditions for live PALM imaging, we adjusted the image acquisition time to optimize the signal-to-noise-ratio (SNR) of the individual fluorophores. To ensure the fastest possible sampling of the structures, we chose the maximal excitation laser power (561 nm at 4 kW/cm2) to bleach the activated fluorophores effectively. We then recorded individual fluorophores at different frame acquisition rates and measured their SNR (
(A) Optimization of the SNR as a function of the image acquisition time. The left panels show the histograms of the amplitudes of the recorded signals at different acquisition times (12.5 ms to 500 ms); the right panels show their respective SNRs. The mean values are indicated as red lines. (B) PALM reconstruction of fixed dendritic spines from 500, 1000, 2000, and 5000 frames of 25 ms. (C) Standard deviation of the distribution of signals across a thin filopodium from the live recording shown in
For the reconstruction of live super-resolution images the spatial and the temporal resolution are intimately correlated. The number of frames for the reconstruction of the PALM image (that determines the temporal resolution) must be adjusted such that the Nyquist-Shannon criterion is satisfied. This criterion states that the sampling frequency must be at least twice as fine as the desired spatial resolution
We then imaged live hippocampal neurons expressing ABP-tdEosFP for periods of up to 30 min of constant illumination. Since cellular structures move during live experiments, the time required for the reconstruction of one PALM image must not exceed the time scale of the morphological changes. Hence, there is a compromise between the spatial and the temporal resolution of the recording. In order to determine the best conditions for PALM image reconstruction, i.e. the highest possible temporal and spatial resolution, we chose cultured neurons at day
During acquisition, ABP-tdEosFP molecules are continuously photoconverted, imaged and finally bleached, meaning that the pool of unconverted fluorophores is slowly depleted. This effect causes a reduction of the number of events that are detected per image frame at constant illumination (
In summary, ABP-tdEosFP serves as an ideal tool to achieve a high coverage of the actin cytoskeleton for relatively long periods of recording despite ongoing photobleaching. The reconstruction of PALM images from 2000 frames (50 s at 40 Hz) was generally sufficient to achieve a good sampling while providing an adequate temporal resolution to visualize the movements of dendritic spines or filopodia.
Having established the best conditions for live PALM imaging, we then recorded the baseline dynamics of mature dendritic spines in hippocampal cultures at DIV 20 to 30 under reduced levels of synaptic activity (MEM-based imaging medium containing 1 µM TTX). These recordings were done at constant illumination for up to 15 minutes at 40 Hz and images were reconstructed from 2000 individual frames (
(A) Super-resolution time-lapse imaging of dendritic spines from a neuron expressing ABP-tdEosFP at DIV 27. Each PALM image was reconstructed from 2000 frames recorded at 25 ms. (B) Quantification of morphological spine parameters (as in
We then quantified morphological parameters of the spine actin cytoskeleton in living neurons (
In the experiments described thus far we inferred the morphology of the spine from the distribution of the actin cytoskeleton as judged by the localization of the ABP-tdEosFP probe. We therefore sought to relate our measurements of the actin cytoskeleton to the position of the plasma membrane, in order to study how temporal changes of the actin cytoskeleton are translated into a modification of the spine membrane.
We co-expressed a GFP-tagged membrane construct (GFP-GPI) together with ABP-tdEosFP in hippocampal neurons. By attaching quantum dots (QDs, 705 nm emission) to the extracellular GFP domain using specific antibodies, we could visualize, in parallel and with a localization accuracy below the diffraction limit of light, the ABP-tdEosFP labeling the actin cytoskeleton and the position of the QDs at the cell membrane. The lateral diffusion of the QD-tagged GPI was determined by single particle tracking (SPT) and served as a readout for the plasma membrane (
(A) PALM imaging of ABP-tdEosFP combined with QD tracking of a diffusing membrane construct (DIV 23). The QD trajectory is depicted as a green line (middle panel) and its outline is superimposed in the overlay with the actin cytoskeleton (right). Both images were reconstructed from 2000 frames recorded at 25 ms. (B) Measurement of the distance between the cytoskeleton and the membrane at the neck (left histogram, N = 16 spines from 9 cells and 2 independent experiments) and the spine head (right, N = 19 spines, 12 cells, 2 experiments).
We could then measure the distance between the plasma membrane outlined by the QD trajectories and the actin cytoskeleton, as reconstructed by the PALM images (
We found that in the spine neck, the distance between the two structures was generally below 50 nm (24±28 nm, mean ± standard deviation, N = 16); in contrast, in the spine head we observed a larger apparent distance between the cytoskeleton and the membrane (46±40 nm, N = 19). Nonetheless, in most of the measured spine heads the membrane was within 50 nm of the cytoskeleton. These data show that the super-resolution imaging of the actin cytoskeleton is an accurate readout of the spine morphology. Although both approaches are to a certain extent complementary, our measurements of the actin cytoskeleton provide additional information on the internal organization of synaptic spines.
The strength of synaptic neurotransmission is regulated by activity-dependent processes, and includes morphological alterations of dendritic spines. These dynamic changes are brought about by rearrangements of the actin cytoskeleton
In response to the application of 10 µM AMPA we observed systematic changes in the distribution of actin within the spine head (
(A) Upper panels: super-resolution time-lapse imaging of a single spine from a DIV 28 neuron expressing ABP-tdEosFP. AMPA was added at t = 0 at a final concentration of 10 µM (arrowhead). Lower panels: average projection of the raw frames used for PALM reconstruction. (B) Quantification of the change of the area of the spine heads over time. (C) Redistribution of the actin cytoskeleton in response to the bath application of AMPA: ratio of number of counts in the spine head relative to the counts in the shaft of the spine. In both, (B) and (C), full circles represent the spine shown in panel (A) and the black line represents the averaged data (N = 5 spines from 3 cells and 3 independent experiments). In the insets to the graphs, measured control spines without treatment (N = 3 spines, 2 cells, 2 experiments). (D) PALM imaging of spines during bath application of 2 µM AMPA (DIV 27), the arrowhead indicates a spine that rounds up after treatment. (E) Quantification of the change of the area of the spine heads over time under control conditions (N = 22 spines from 6 cells and 3 independent experiments) and 2 µM AMPA treatment (35 spines, 4 cells, 2 experiments). (F) Simultaneous PALM/QD imaging of the spine cytoskeleton and the plasma membrane, before and after application of 2 µM AMPA. The outline of the membrane is superimposed as a green trace to the PALM image (left). Histogram of the distance between the cytoskeleton and the spine head during bath application of 2 µM AMPA (right, N = 13 spines, 2 cells, 2 independent experiments).
We quantified the AMPA-induced morphological changes of the spine cytoskeleton to determine the temporal profile and the magnitude of these events. When we measured the spine head area as a function of time, we observed a pronounced reduction of the spine size after AMPA treatment, with an average decay time of 67±12 s (N = 5 spines from 3 independent experiments) (
We also conducted experiments using a lower AMPA concentration of 2 µM that had been reported to lead to rounding and the loss of motility of dendritic spines
In this study, we have implemented PALM microscopy for long-term super-resolution dynamic imaging. The necessary tools and techniques are widely available, which makes this a relatively low-cost approach that may be easily set up in many laboratories using fluorescence microscopy.
We initially validated our experimental strategy by studying the organization of the actin cytoskeleton in dendritic spines in fixed neurons. From the high-resolution PALM images of mature dendrites it was possible to quantify morphological parameters of spines that are comparable with the results obtained by electron microscopy
Subsequently, we performed live PALM experiments to observe the baseline dynamics of the spine morphology over tens of minutes. The quantification of the morphological spine parameters in living neurons is in very good agreement with the values obtained in fixed neurons, suggesting that the chemical fixation with paraformaldehyde did not affect the cytoskeletal structure noticeably. Furthermore, we observed that the temporal rearrangement of the spine cytoskeleton was more pronounced in the spine head and the length of the neck, while the width of the neck remained remarkably constant. This could reflect the fact that the actin filaments in the spine head are organized in a complex manner in contrast to the neck where the F-actin appears to have a linear organization
In this context it should be noted that our measurements refer to the dimension of the cytoskeleton as determined by the detection of the ABP-tdEosFP expression construct. In order to relate our findings to the localization of the plasma membrane, we combined PALM imaging with QD tracking of a diffusing membrane construct (GFP-GPI). This technique allowed us to visualize both, the actin cytoskeleton and the plasma membrane simultaneously at super-resolution. Our measurements showed a close correspondence between the cytoskeleton and the membrane, typically within a range of 50 nm. This distance includes the width of the plasma membrane (5–10 nm) in addition to the size of the antibodies (∼5 nm) and the streptavidin-coupled QDs (∼25 nm) used for labeling. The measured distances in the spine heads had a broader distribution compared to the neck. The likely cause of this difference is that F-actin filaments are distributed very unequally within the spine head and that consequently low density regions of the cytoskeleton at the periphery of the spine are sampled less efficiently than the core of the spine.
To provoke morphological changes of the spine cytoskeleton, we induced a sustained synaptic depolarization by bath application of AMPA and measured the concomitant alterations of the spine shape. We found that this treatment reproducibly led to the shrinkage of synaptic spines, followed by a reduction of the actin levels in spines. This is in agreement with the data by Halpain and colleagues, who found that glutamate receptor activation triggered a loss of F-actin and the collapse of the spines
In addition to the morphological changes, the levels of actin in the spine head decreased during AMPA application. This redistribution of actin could also be observed in the diffraction-limited images, where the fluorescence intensity of diffusing fluorophores contributes to the overall fluorescence. In contrast, in the PALM image reconstruction only fluorophores that are bound to F-actin are likely to be detected as well focused spots (with 2D-Gaussian shape) and thus represented in the image. Therefore, the decrease in the signal intensity of the spine head relative to the shaft is due not only to depolymerization of actin but also to an actual redistribution of actin through the spine neck. Since the observed redistribution of depolymerized actin from the spine head into the dendritic shaft occurred gradually and on a somewhat slower time scale, our findings also imply that the spine neck represents a barrier that limits the diffusion of actin monomers to the dendrite, and underlines the importance of the spine neck for the compartmentalization of the spine head as the site of synaptic neurotransmission.
The use of a low-affinity probe (ABP-tdEosFP) for live PALM imaging confers several beneficial features: 1) bleached fluorophores are replenished, in contrast to probes that are fused to the protein of interest, thereby enabling long-term recording of a given cellular structure; 2) the probe is not directly incorporated into the actin filaments, reducing the risk of altering the cytoskeletal organization; 3) the cell may tolerate higher levels of expression, meaning that there is a larger pool of photoactivatable probes. This also implies an effective increase of the sampling of the structure since a large number of binding sites are available for ABP-tdEosFP. Our experimental strategy also represents a step forward in comparison to the indirect measurements achieved by SPT techniques
In conclusion, our work brings together several innovative approaches to study the dynamics of cellular structures at super-resolution, namely long-term dynamic PALM imaging and the use of a low-affinity probe. Furthermore, we have combined live PALM imaging with QD-based single particle tracking in order to visualize two cellular structures (membrane and cytoskeleton) simultaneously at super-resolution. We believe that this novel approach is a powerful combination of techniques that can be applied to correlate single molecule dynamics with cellular structures at the nanometer scale, such as the movement of neurotransmitter receptors in relation to the postsynaptic density
COS-7 cells (ATCC, cat. No. CRL-1651,
Rat primary hippocampal neurons were prepared as previously described
The actin-binding sequence of ABP140 from
Cells were fixed in 0.1 M phosphate buffer (pH 7.4), containing 4% paraformaldehyde and 1% sucrose at 36°C for 10 min and rinsed in phosphate buffered saline (PBS, pH 7.4). The F-actin cytoskeleton in COS cells was labeled using 13 nM Alexa Fluor 647-phalloidin (Invitrogen, 1∶500 dilution) in PBS containing 1% bovine serum albumin (BSA, Sigma) for 1 h and washed in PBS. For Shank2 immunolabeling, neurons were post-fixed in methanol for 5 min at −20°C and blocked over night at 4°C in PBS containing 3% BSA. A polyclonal rabbit antibody directed against the PSD protein Shank2 (ProSAP1,
Coverslips with rat hippocampal neurons were imaged in MEM medium without phenol red (Invitrogen) containing 2% B27, 2 mM glutamine, 1 mM pyruvate, 33 mM glucose, 20 mM HEPES, 1 µM tetrodotoxin (TTX) and 5 µM glycine ( = imaging buffer) at 35°C for durations of less than 1 h. AMPA receptors (AMPAR) were activated by bath application of 2 µM or 10 µM α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA). To correct for the x/y-drift of the microscope stage multicolored beads (TetraSpeck) were used as fiducial markers.
Super-resolution imaging was performed on an inverted Nikon Ti Eclipse microscope. Activation (405 nm/532 nm) and excitation (561 nm/641nm) lasers for PALM/dSTORM were combined in an external platform; the combined beam was expanded and re-collimated with a beam expander, directed through free air into the microscope and focused in the rear plane of a 100× objective (N.A. 1.49) using an achromatic converging doublet-lens with a focal length of 500 mm. Images were taken in wide-field configuration and, unless otherwise stated, the experiments were acquired under continuous illumination of both, activation and excitation laser light. Activation laser power was finely tuned during acquisition, in order to maintain the density of activated fluorophores constant; typical values of activation and imaging laser power densities on the sample were 7×10−3 to 3×10−2 kW/cm2 (405 nm), 4 kW/cm2 (561 nm), 1 kW/cm2 (532 nm), and 2.7 kW/cm2 (641 nm). Single-molecule tdEosFP signals were separated with a 561 nm dichroic (Di01-R561-25×36) and a single band 617 nm emission filter (FF01-617/73), expanded through a 1.5× lens in the tube-lens of the microscope and acquired with an Andor iXon EMCCD camera (512×512 pixels, pixel size 16 µm) at frame rates of 25 or 50 ms for up to 30 minutes. The low-resolution, conventional fluorescence images of the pre-converted form of tdEosFP were taken using a mercury lamp for illumination (excitation: Semrock FF01-485/20, emission: FF01-525/30).
To avoid a substantial degradation of the spatial resolution, sample drift needed to be corrected. The position along the optical axis (Z axis) was actively stabilized during the acquisition by the Nikon perfect focus system (PFS) integrated in the scope. Drift in the XY plane, which may be as high as 50–100 nm in 10 min, was corrected in the post-processing of the images using the multicolor beads described above as fiducial markers.
Dual-color dSTORM/PALM imaging was performed successively. First, we reversibly switched Alexa 647 fluorophores (labeled Shank2) between a dark state and a fluorescent state, under reducing buffer conditions (PBS, pH 7.4, containing 10% glucose, 50 mM β-mercaptoethylamine, 0.5 mg/ml glucose oxidase and 40 µg/ml catalase, degassed with N2;
Live dual-color measurements of the actin cytoskeleton and the spine membrane were done by combining PALM imaging of ABP-tdEosFP with QD-tracking of a GFP-GPI membrane expression construct with a Photometrics dual-view detection system, using the 561 nm laser as excitation for both, the converted form of tdEosFP and the QDs. Co-transfected neurons were labeled with a rabbit antibody directed against GFP (Synaptic Systems, 1∶104 in imaging buffer, 4 min), followed by a secondary biotinylated anti-rabbit Fab fragment (Jackson ImmunoResearch, 1∶2000 in imaging buffer, 4 min) and streptavidin-coupled QDs emitting at 705 nm (Invitrogen Q10161MP, 1 nM in QD binding buffer, 1 min; see
To quantify the fluorescence loss after photoactivation (FLAP), the 405 nm laser beam was focused in the center of the field of view, covering an area of ∼4 µm2. ABP-tdEosFP transfected COS-7 cells (in MEM imaging buffer at 35°C) were exposed to a short beam of the 405 nm laser, while images were taken every 30 s for 18 min using the mercury lamp (560/25 nm excitation and 614/70 nm emission filters).
Raw images of single molecule signals were analyzed with an adapted version of the Multi-Trace Tracking MTT algorithm
Sample images of dendritic protrusions, showing the morphological variety of spines described in
(TIF)
Pointillist representation of live PALM imaging of a dendritic segment of an immature hippocampal neuron (DIV 9) at time 0 (black points), 12 min (blue) and 25 min (red) under continuous illumination and recording. The 405 nm laser power was continuously adjusted to yield a constant number of single molecule events. The chosen time-window for image reconstruction was 50 s (2000 frames). The data plotted in
(TIF)
Live PALM imaging of a dendritic segment of a mature hippocampal neuron (DIV 27) under baseline conditions. The total length of the movie is 12.5 minutes. Each PALM frame was reconstructed from 2000 image frames of 25 ms, hence the temporal resolution is 50 s. The movie is rendered with a temporal sliding window of a step of 2.5 s. The width of the field of view is 12 µm. The drift of the movie was corrected with fiducial markers, therefore the actual movement of the dendrite is visualized.
(AVI)
Simultaneous PALM imaging and QD tracking. ABP-tdEosFP (top) and QD signals (bottom) were simultaneously imaged in a hippocampal neuron at DIV 27. The detected ABP-tdEosFP fluorophores were used to reconstruct the actin cytoskeleton of the dendritic spine (cumulative movie, middle panel); the QD trajectory (bottom) served to outline the shape of the spine membrane. Field of view is 5 µm×3 µm.
(AVI)
Live PALM imaging of a large dendritic segment of a mature hippocampal neuron (DIV 28), treated with 10 µM AMPA, as indicated in the movie. Movie rendered with a sliding window of 2000 frames of 25 ms ( = 50 s) and a step of 2.5 s. Field of view, 35 µm×16 µm.
(AVI)
Movie of an individual spine during 10 µM AMPA treatment (detail from
(AVI)
We would like to thank Marianne Renner, Géraldine Gouzer, Yann Nadjar, Fabien Pinaud, Serge Marty and Ricardo Henriques for technical help and their comments on the manuscript. Antibodies and cDNA constructs were kindly provided by Tobias Böckers, Satyajit Mayor, Mike Davidson and Vladislav Verkhusha.