|Home | About | Journals | Submit | Contact Us | Français|
To understand the cellular and circuit mechanisms of experience-dependent plasticity, neurons and their synapses need to be studied in the intact brain over extended periods of time. Two-photon excitation laser scanning microscopy (2PLSM), together with expression of fluorescent proteins, enables high-resolution imaging of neuronal structure in vivo. In this protocol we describe a chronic cranial window to obtain optical access to the mouse cerebral cortex for long-term imaging. A small bone flap is replaced with a coverglass, which is permanently sealed in place with dental acrylic, providing a clear imaging window with a large field of view (~0.8–12 mm2). The surgical procedure can be completed within ~1 h. The preparation allows imaging over time periods of months with arbitrary imaging intervals. The large size of the imaging window facilitates imaging of ongoing structural plasticity of small neuronal structures in mice, with low densities of labeled neurons. The entire dendritic and axonal arbor of individual neurons can be reconstructed.
Neocortical circuits change in response to salient experiences. The cellular mechanisms underlying circuit plasticity are thought to involve use-dependent modifications of synapses. Linking in vivo experience-dependent plasticity to synaptic dynamics will ultimately require time-lapse imaging of synaptic structure and function as plasticity develops. Recent advances in high-resolution imaging are beginning to make such experiments feasible. Two-photon laser scanning microscopy (2PLSM)1,2 facilitates imaging of individual synapses deep in the scattering tissue. Genetically encoded fluorescent proteins (collectively referred to as XFPs; e.g., green fluorescent protein, GFP) can be used to measure the structural dynamics of living neurons (e.g., ref. 3) and the distribution of synaptic proteins4 in vivo. Transgenic mice expressing XFPs5,6 in subsets of neocortical neurons have enabled long-term imaging of neuronal structure in adult (e.g., refs. 7,8) and developing mice (e.g., refs. 3,9). GFP-based biosensors (e.g., phluorophorin, cameleon, TN-XXL)10–12 hold great promise as reporters of neuronal and synaptic functions.
As a first step in studying synaptic biology in vivo, several groups have measured the structural dynamics of dendritic and axonal arbors and their synaptic specializations, dendritic spines and axonal boutons. In developing mice, dendritic spines and filopodia were found to be highly dynamic over a period of several hours3. Subsequent long-term imaging in transgenic mice revealed complex, cell-type specific structural dynamics in the adult neocortex. Although the total length of most axonal and dendritic arbors remains stable in adult mice7,13,14, some cell-types display large-scale changes at the level of dendritic15 and axonal branches16. Similarly, although a large fraction of dendritic spines and axonal boutons remain stable in the adult neocortex7,8,17, a sub-population of dendritic spines appear and disappear in an experience-dependent manner7,18–21. Furthermore, the growth and retraction of spines and boutons have been linked to synapse formation and elimination16,22,23, indicating that structural changes underlie aspects of functional plasticity in vivo.
The opacity of the intact skull of adult mice precludes high-resolution imaging of neocortical neurons. As a result, for all imaging studies the overlying bone must be partially removed (Fig. 1). This can be achieved either by thinning the bone to a ~20 μm thick sheet (e.g., refs. 8,14,18,24–29) or by permanently replacing a small bone flap with a coverglass known as a cranial window (e.g., refs. 3,4,7,9,13–17,19–21,27–32). In this protocol we combine the converging experiences of six laboratories to describe the essential steps toward implantation of a chronic cranial-imaging window, with subsequent high-resolution imaging of dendritic spines and axonal boutons (for further details of our studies in which this preparation was used, see refs. 4,7,9,14–17,19–22,33,34).
About two different surgical preparations have been used for long-term high-resolution imaging in vivo: chronic cranial windows and thinned skull preparations. With the chronic cranial window, a bone flap is removed (while leaving the dura mater unperturbed) and replaced by a coverglass in a single surgical session (Fig. 1 left panel; e.g., refs. 7,17). The imaging window provides an excellent optical access, allowing repeated high-resolution imaging with essentially unlimited time points and at arbitrary imaging intervals (Figs. 2–4). The size of the craniotomy ranges from ~0.8 to 12 mm2. This large field of view permits imaging of mice with very sparse labeling (e.g., GFP-M line5,7; Fig. 1 left panel; see below). The chronic cranial window preparation is also suitable for larger animals, including rats and primates, in which the dura is opaque and thus has to be removed for imaging3,31 and for neonatal mice, in which the skull is too fragile for thinning9. It can be combined with intrinsic optical-signal imaging (e.g., refs. 13,20,21), injections of viral vectors30,31, and implantation of electrodes, probes or other devices. In addition, miniature microscopes for freely behaving mice35,36 and small diodes for photoactivation37 are best used in combination with this type of preparation.
The cranial window remains clear for several weeks to months, until regrowth of skull from the edges of the cranium, and thickening of the dura begins to degrade the image quality and ultimately terminates the experiment7,9,15–17. As the preparation is sealed off by glass and dental cement, it is challenging to re-open the cranium (see PROCEDURE). The chronic cranial window technique is highly operator-dependent, with success rates in the order of ~30–80%. Although the long-term optical clarity of the window depends on the quality of the surgery, the outcomes are sometimes unpredictable.
The thinned skull preparation leaves a thin (~20 μm thick), optically clear layer of bone in place (Fig. 1 right panel; e.g., refs. 8,25). As the remaining bone is devoid of vasculature it thickens and becomes inflamed and opaque within 1 d after the surgery. Thinning, therefore, needs to be repeated before every imaging session. This can result in variations in imaging quality between sessions. In addition, because of the increasing opacity after repeated thinning, the experiment must often be terminated after the second or third session. However, the procedure can be carried out at any given time, and therefore the first and last imaging time point can be separated by long intervals, including months or, theoretically, even years. Skull thinning is usually carried out over a small area (~0.1–0.3 mm2), as the shaving of larger parts of thin, unstable sheets of bone may contuse the underlying cortex (see ref. 25 and Supplementary Methods 1 online for a detailed description of the thinning procedure). This technique has therefore been used mainly on mice with dense and very bright labeling (e.g., YFP-H line5,8; Fig. 1 right panel; see below).
Transgenic mouse lines that express XFPs in sub-populations of neurons5,6 (Fig. 2) have become an important tool for high-resolution imaging in vivo. Often, the XFP expression patterns are stable and sparse, thereby permitting imaging of isolated neuronal structures with high signal to noise ratios over long periods of time. The most popular lines include GFP-M (e.g., refs. 5,7; Fig. 2b), mGFP-L15 (refs. 6,16), mGFP-L21 (refs. 6,9), YFP-H (e.g., refs. 5,8,28,38; Fig. 2a), YFP-G5,13 and GFP-S5,15. A chronic cranial window provides a large field of view, allowing imaging in mouse lines with sparse expression patterns (such as GFP-M and mGFP-L15 and L21; Figs. 2b,d and and4c).4c). As large parts of the dendritic and axonal arbor can be imaged, the identity of the parent cell can often be determined, either in vivo or by postmortem reconstruction (Figs. 4c,e)7,9,14–17,22. Imaging sparsely labeled cells also allows extensive sampling of individual cells (Figs. 2d–g), and thus detection of neuronal (sub)type- and region-dependent effects15,16,19,34. When combined with electrophysiological measurements7, intrinsic signal imaging20,21,32,39, or post hoc anatomy7,19, it allows for precise correlations between functional and structural plasticity. Although large data sets can be collected from mice with densely labeled neurons, such as the YFP-H line, it has not been possible to identify the neuronal subtype of the imaged dendrites and axons (Figs. 2a and 2c). Therefore, in these mice imaging is usually performed at the population level, mixing data from diverse cell-types8,28,38,40. Furthermore, detection of all spines and/or boutons is challenging because of the high density of processes and the resulting background fluorescence. In addition, long-term expression of high XFP levels might have toxic effects, as has been shown for the YFP-H line41.
Direct gene delivery to brain tissue can be performed using electroporation and viral vectors. For example, in utero electroporation in mouse embryos has been used to label large groups of neurons in neocortex42,43. Recombinant and replication-defective viral vectors, such as Lenti- and AAV based constructs, can be injected directly into the postnatal or adult mouse cortex and combined with long-term imaging. Depending on the injection volume, DNA concentration, vector backbone and promoter sequences, high expression levels of fluorescent proteins can be achieved in a relatively sparse population of neurons, permitting in vivo imaging4,29–31,37,39,44.
Fluorescent neurons can be imaged in vivo using 2PLSM. The authors' laboratories typically use custom-built microscopes with detectors that are placed as close as possible to the objective, to increase the detection efficiency45,46. A tunable Ti:sapphire laser is typically used as a light source. The wavelength of the excitation light depends on the fluorophore, but typically the cross sections of intrinsically fluorescent proteins are relatively broad (e.g., EGFP can be excited with reasonable efficiencies from 830 nm to 1,020 nm). High NA (×0.8–0.9, ×20–60) water immersion objectives are well suited for imaging of small structures, such as dendritic spines and axonal boutons. Sensitive photomultiplier tubes with high quantum efficiencies and low dark currents (e.g., Hamamatsu R3896 or H7422P) are adequate for detection of recurrent fluorescence.
The appearance or disappearance of dendritic spines and terminaux boutons should be scored on the basis of consistent, preset criteria (Fig. 5). As en passant boutons are continuous with the axon shaft, and smaller boutons are difficult to distinguish from nonsynaptic swellings47, their appearance or disappearance can only be assessed by changes in brightness using threshold intensity-ratio measurements (see ref. 16 for details; Fig. 5l). To consistently score appearances and disappearances of spine and terminaux boutons, one has to use similar lookup tables at each time point. As small movements of the tissue (e.g., because of pulsation of nearby arterioles) can shift images in different sections, images need to be analyzed in three dimensions rather than in projections. Owing to the limited numerical apertures of long working distance water immersion objectives, and spherical aberrations in the brain, the spatial resolution along the axial dimension is poor relative to the sizes of typical synaptic structures. Therefore, small protrusions cannot always be detected or resolved if they emanate from the dendritic shaft perpendicular to the focal plane. We usually only include protrusions that emanate laterally from the dendritic shaft and if their lengths span more than 1 resolution unit (~0.4 μm, typically 5 pixels; Fig. 5a), which exceeds the width of the haze of fluorescence around the dendrite. It is also important to verify that the intensity of the dimmest protrusion exceeds the noise in the images. The noise can be estimated by the standard deviation of the background fluorescence. We only include images for analysis if the fluorescence intensity of the (neck of the) dimmest protrusion is ~5 times higher than the standard deviation of the background around the dendrite. In images with many fluorescent structures (axons and dendrites crossing each other), dim filopodia-like dendritic protrusions are often lost in the noise, and in addition can become indistinguishable from thin axons in the vicinity of dendrites. Furthermore, to compare spines over several imaging sessions one has to keep the fluorescence levels constant across images. All these criteria can considerably influence quantification of structural plasticity, and therefore contribute to differences in results reported in different studies from different labs (Fig. 6).
In general, we feel that measurements of absolute numbers of turnover of small synaptic structures are subject to significant systematic differences between different observers (perhaps as large as a factor of two; see ANTICIPATED RESULTS) and should therefore be treated with caution. On the other hand, the scoring of changes in structural plasticity, e.g., before and after sensory deprivation7,19–21,32, is more robust if turnover is scored in a blind manner using consistent criteria. In addition, post hoc analysis by super-resolution reconstruction of the imaged structures (e.g., retrospective serial section EM22, see also ref. 48) can be performed to verify scoring criteria.
The appearance and disappearance of small structures can be expressed by the survival function (SF) and the turnover ratio (TOR). The survival function describes the fraction of structures that remain present over time; SF(t) = N(t)/N0, where N0 is the number of structures at t = 0, and N(t) is the number of structures of the original set surviving after time t. By definition SF(0) = 1. The TOR describes the fraction of structures that appear and disappear from one time point to the next TOR (t1, t2) = (Ngained + Nlost)/(N(t1) + N(t2)), where N(t1) and N(t2) are the total number of structures at the first and second time point, respectively. If the number of appearances and disappearances are similar, TOR (t1, t2) = (Ngained + Nlost)/(2xN(t1)) and [1−SF(t2−t1)] ≈TOR (t1, t2).
To control for the detection sensitivity of the preparation, the results from the in vivo imaging can be compared with images from fixed brains (naive or experimental mice). Spine and bouton density and the distribution of their sizes should be comparable under all conditions16,17. To control for the stability of the preparation image series taken shortly after surgery can be compared with image series taken several weeks later. Any surgical procedure (craniotomy and skull thinning alike) carries the potential risk of acute and transient effects on the brain due to mechanical stress and prolonged anesthesia. Thus, data collection immediately after the surgery should be avoided unless the experimental design demands this. To illustrate such control experiments, we provide examples from several of our laboratories, in which spine and bouton densities and their turnover were measured with different intervals from day 1 until several weeks after the surgery20,21,34 (see ANTICIPATED RESULTS; Figs. 7 and and8).8). Similarly, the ultrastructure and glial cell reactivity can be estimated at several time points after the surgery by EM and immunohistochemical detection of glial cell markers, respectively7,25,34 (see ANTICIPATED RESULTS; Figs. 9 and and1010).
Chronic cranial windows have been used by several research groups to image the structure of cortical neurons over periods of weeks–months4,7,9,13–17,19–21,28,29,31,32,34,38. A variety of the following parameters have been measured: dendritic and axonal growth or stability7,9,13–16,29,31,34; dendritic spine and axonal bouton turnover7,16,17,19–21,28,32,38; dendritic spine, axonal bouton volume changes16,19,21 and dynamics of GFP-tagged synaptic proteins4,39. Imaged neurons are easily identified in vivo and recovered in the postmortem material7,15,19, axons can be traced back to their cell body9,16 and neuronal structures can even be reconstructed with serial-section electron microscopy7,16,22,48. The lengths and complexities of dendritic and axonal arbors, as well as their spine and bouton densities, remain stable for months after the surgery7,13–17,19,34. Similarly, baseline structural dynamics of synaptic structures remain constant7,13,15–17,19. The brain underlying chronic cranial windows display only minor and transient changes in the expression of glial-cell markers during the first 2 weeks after the surgery32,34.
The ability to image and reconstruct entire neuronal structures in sparsely labeled transgenic XFP mice over arbitrary and unlimited imaging time points for several months represent a distinct advantage of the chronic cranial window. The chronic cranial window is therefore exquisitely suitable for studies aiming at a detailed investigation of the relationship between structural dynamics and functionality of neuronal networks.
Mix 13 mg ml−1 of ketamine and 1 mg ml−1 of xylazine in distilled, sterile H2O. This mixture can be made in advance and stored in the dark at room temperature for few weeks.
Dissolve 2.5 g of 2,2,2–tribromoethanol in 5 ml 2-methyl-2-butanol, heat to 40 °C and stir. Add distilled, sterile H2O to a final volume of 200 ml. This mixture can be made in advance and stored in the dark at 4 °C for 2 weeks.
Dissolve 125 mM NaCl (7.21 g NaCl), 5 mM KCl (0.372 g KCl), 10 mM glucose (1.802 g glucose), 10 mM HEPES (2.38 g HEPES), 2 mM CaCl2 (2 ml 1M CaCl2) and 2 mM MgSO4 (2 ml 1M MgSO4) in distilled H2O (1 l of dH2O). The buffer should be maintained at pH 7.4. Sterilize the solution by passing it through a sterilization filter. The solution can be prepared in advance, aliquotted and stored at 4 °C for at least a month.
Warm 0.3 g of agarose in 25 ml of cortex buffer until fully dissolved. This solution must be freshly prepared and sterile for each experiment, and kept hot. It must be stirred frequently during the procedure. Before applying the hot agarose to the cortex, allow it to cool briefly and check that the temperature is not above ~37 °C.
Imaging can be started immediately after the surgery. We usually prefer to wait for 10–14 d to allow the mice to recover from the surgery. This also enables selection of those mice that have a viable and durable window (Figs. 3g,h, Supplementary Fig. 1). In addition, in many instances the window can become somewhat opaque during the first week after the surgery, temporarily reducing the imaging quality (see TROUBLESHOOTING).
Superficial blood vessel patterns can be used as a guide (Figs. 3g,h and 4a,c, and Supplementary Fig. 1) to repeatedly image regions of interest with high magnification (Fig. 4d). Entire dendritic or axonal arbors can be imaged and reconstructed (Figs. 4c,e; Supplementary Movie 1). Overall branching patterns of pyramidal cell apical dendrites do not change over the time course of weeks in the adult neocortex (see ref. 7). Axons and interneuron dendrites can display length and trajectory changes over tens of micrometers over weeks15,16. Dendritic spines and axonal boutons can be imaged with high resolution (Figs. 2d–g, ,4d,4d, ,6,6, 7a,c and Supplementary Fig. 1), often at depths of more than 100–200 μm, and sometimes even deeper (Figs. 2e–g). Imaging windows can remain transparent for several months (Supplementary Fig. 1). A subset of spines and boutons can be seen to appear and disappear between imaging sessions (Figs. 4d, 7a,c and Supplementary Fig. 1).
Structural stability and dynamics can be quantified e.g., by measuring spine and bouton densities and their turnover over time. In Figures 7 and and88 we show examples of quantitative imaging data that can typically be obtained by long-term imaging of neurons using chronic cranial-window preparations. Data from six laboratories were combined (indicated by different colors in Figs. 7 and and88 and Supplementary Fig. 2), representing dendrites of 51 layer 5B (L5B) pyramidal cells in somatosensory and visual cortices of GFP-M mice (Fig. 7; 19 mice, age > 3 months), and multiple axons of layer 6 (L6) pyramidal cells in the somatosensory cortex of mGFPL15 mice (Fig. 8; 7 cells; 7 mice, age > 3 months). Imaging was started within 0–2 d or 10–14 d after implantation of the cranial window, and spines and boutons were tracked over the following weeks and months. Two groups reported small decreases in average spine densities in the somatosensory cortex over the first week after the surgery (13% ± 16% between day 1 and 30, red lines, bootstrap correlation analysis (BCA): r = −0.31 ± 0.11, P(no or pos corr) = 0.001; and 9% ± 6% over the first 3–4 weeks, black lines, BCA: r = −0.29 ± 0.18, P(no or pos corr) = 0.05; Fig. 7b). However, spine densities remained unchanged thereafter, to 3 months of imaging (data not shown). It is important to note that spine densities under chronic cranial windows were similar to spine densities in fixed, naive somatosensory cortex (naive, green: 0.28 ± 0.06 μm−1; cranial window, black: 0.30 ± 0.07 μm−1, 10–28 d post surgery, t-test P = 0.44; measured by the same observer; Fig. 7b). Thus, the small drop in spine densities that was observed in a subset of experiments over the first week, likely reflects a loss of protrusions that were formed in response to the surgical procedure. In the visual cortex, spine densities remained very stable under cranial windows over a 2-week period (2% ± 7% increase between day 2 and day 10–15, blue lines, BCA: r = 0.34 ± 0.14, P(no or neg corr) = 0.02; Figs. 7b,c). Similarly, L6 terminaux bouton densities in the somatosensory cortex remained unchanged over a 4-week period after the surgery (0% ± 12% change, black lines with triangles, BCA: r = 0.03 ± 0.18, P(no or neg corr) = 0.44; Figs. 8a,b), and were equal to densities in the naive fixed cortex (cranial window: 0.077 ± 0.021 μm−1, 28 d post surgery, naive: 0.063 ± 0.016 μm−1, t-test P = 0.19; Fig. 8b). Large cell-to-cell variability in spine and bouton densities was observed under all experimental conditions (Figs. 7b and and8b),8b), and is inherent to regional and cell subtype-dependent differences in L5B and L6 pyramidal dendritic and axonal morphology16,17,51,52.
The fraction of spines and terminaux boutons that survived over the first week of imaging was comparable between experiments that were started immediately after the surgery and those that were started after 10–14 d of recovery (compare open and solid markers in Figs. 7f and and8c;8c; black circles: 0.60 ± 0.14 and 0.65 ± 0.09, 20 d; red: 0.55 ± 0.09 and 0.52 ± 0.09, 28 and 30 d; blue: 0.89 ± 0.06 and 0.88 ± 0.12, 11 and 9 d; black triangles: 0.53 ± 0.16 and 0.47 ± 0.17, 24 d), indicating that spine and terminaux bouton stability does not change with time after the implantation of a chronic cranial window.
Appearances and disappearances of spines and terminaux boutons can be observed over many weeks to months after the surgery. Under baseline conditions in adult mice the number of appearances balance the disappearances (Figs. 7a,c,e and 8a,d and Supplementary Fig. 1)7,16,17,20,21,38. Figures 7e and and8d8d show examples of TOR of spines and terminaux boutons, which are constant over very long imaging experiments (to 3 months; red lines, BCA day 5–94: r = −0.13 ± 0.09, P(no or pos corr) = 0.06; black lines with round markers, BCA day 14–110: r = −0.039 ± 0.10, P(no or pos corr) = 0.36; blue lines, BCA day 16–48: r = 0.13 ± 0.09, P(no or neg corr) = 0.09; black lines with triangles, BCA day 10–46: r = −0.15 ± 0.17, P(no or pos corr) = 0.21), indicating that spine and bouton dynamics do not change over time after implantation of a chronic cranial window. The TORs under cranial windows in visual cortex were significantly lower than in the somatosensory cortex. These differences are likely due to the different locations (visual versus barrel cortex17) and differences in scoring criteria.
To illustrate the influence of scoring criteria on the reported turnover of dendritic spines, a data set collected and analyzed by the CSHL group was also analyzed by the MPI group. The CSHL estimates of TORs were ~75% higher than the MPI-measurements. A detailed comparison of the scoring criteria revealed seven subtle causes of bias underlying this difference (Fig. 6):
(1) The cut-off for spine lengths and fluorescence intensities that is used to determine whether a protrusion should be counted as a spine. This can influence the estimates of dynamics, as the smallest (dimmest) protrusions are the most dynamic (Fig. 6c)17. (2) Spines often arise from thickenings in the dendritic shaft. These protrusions could be interpreted as new spines or as extensions from a pre-existing spine (Fig. 6e). (3) If spines are included when emanating perpendicular to the focal plane (i.e., above or below the dendrite), this can result in an underestimation of the dynamics because only large (and thus relatively stable) spines can be detected in this direction17. (4) Some spines can be absent in one of the imaging sessions, whereas present in all others. This is interpreted by some observers as the spine having been transiently lost, whereas others view this as a failure to detect the spine at one particular time point (e.g., due to movement or lower image quality; Fig. 6a). (5) Similarly, big stubby spines can disappear, but the dendritic shaft retains increased curvature at the site (Fig. 6d). If this remnant curvature is continued to be counted as a spine, dynamics will be lower than if scored as a disappearance. (6) In regions with numerous labeled axons, thin filopodia-like protrusions are often indistinguishable from axons (Fig. 6b), and scoring of particular structures is dependent on the interpretation of the observer. Failure to detect all thin protrusions will reduce turnover, as smaller spines are more plastic. (7) Estimates of stability can also vary as a function of the number of analyzed images in a time-lapse sequence. Regions of interest typically move a little from one imaging session to another imaging session, and persistent spines are only scored if they are present in all analyzed images. In contrast, transient spines are scored if they (or the relevant stretch of parent dendrite) are contained in the field of view over a subset of images (when the spines are present and the image immediately before and after). Therefore, regions of the dendrite close to the edge of the field of view are more likely to contribute to the count of transient spines. Although this is a subtle effect, it can be sizeable for small regions of interest containing only a few spines (~10).
Differences in these biases can produce significant systematic differences in reported turnover between different observers (perhaps as large as a factor of two). When analyzing changes in structural plasticity in response to a manipulation, such as sensory deprivation, it is therefore critical to score structural changes in a blind manner and to use consistent criteria.
The condition of the cortex under a chronic cranial window can further be assessed by comparing the cellular ultrastructure and the expression levels of markers of injury/inflammation between operated and non-operated hemispheres in perfusion-fixed brains. In routine cranial window preparations, large-scale morphological features are normally indistinguishable from control brains (data not shown)17. In Figure 9 we show a qualitative assessment of the ultrastructure in cortical layer 1. The EM sections, taken from chronic cranial-window preparations 15 d after the surgery and from controls did not reveal differences in neuronal morphology or the fraction of the tissue occupied by glia (compare Figs. 9a,c and 9b,d). Figure 10 shows examples of an assessment of the number and morphology of astrocytes and microglia using anti-GFAP and anti-Iba1 immunostaining, respectively53. The data were obtained by two different groups, and include immunostainings on day 2, 10, 30 and 90 after implantation of a cranial window above the somatosensory cortex (RM and CP-C), and quantifications of the number of immunopositive cells in the visual cortex at different time points after surgery (WCAL and EN). An increase in GFAP immunostaining was typically observed between 2 and 14 d after the surgery (Figs. 10a,c), which could have been caused by an increase in the number of GFAP-positive astrocytes (Fig. 10c, asterisks indicate P < 0.05, ANOVA). GFAP immunoreactivity returned to control levels within 4 weeks and remained unchanged over subsequent time points (Figs. 10a,c). An increase in immunoreactivity for Iba1 was observed within the first day after the surgery (arrowhead in Fig. 10b; RM and CP-C). This increase was transient and returned to normal levels after 7–10 d (Fig. 10b). The number of Iba1 immunopositive cells typically remained stable and their morphology remained unchanged (Figs. 10b,d).
Studies using the protocol presented here6,7,14,17,19–21,29–34,38,39 show that if surgeries are performed carefully, and if windows remain transparent beyond the first week post-surgery, spine and bouton densities are largely stable over the first 2 weeks after the surgery as well as over the following months and essentially comparable to naive brains (Figs. 7 and and8)8) and thinned skull preparations (Supplementary Fig. 2). Similarly, the rate of turnover of spines and boutons is reported to be stable under the chronic imaging window for several months7,15–17,19–21,32,34 (Figs. 7 and and8).8). The upregulation of glial cell protein markers under cranial windows is mild and transient, whereas structural plasticity remains unchanged34 (Figs. 8 and and9).9). This indicates that there is no correlation between the rate of spine turnover and the intensity of glial marker staining. These findings are consistent across all six laboratories. A recent study compared chronic cranial windows with thinned skull preparations25, suggesting that the implantation of a cranial window is associated with substantial loss of dendritic spines (~35% drop in spine density over the first 2 weeks after surgery), transiently elevated turnover, and extensive activation of microglia and astrocytes25. This study was in contradiction to an earlier study that had shown indistinguishable structural plasticity with the two types of preparation28 and studies reporting spine and terminaux bouton densities under cranial windows that were similar to spine densities in naive tissue, both immediately after the surgery and after long-term imaging16,17. More recent studies that included additional control experiments have further been unable to reproduce all of the key aspects of the study of Xu et al.14,20,21,25,32,34,39. The ability to image and reconstruct entire neuronal structures in sparsely labeled fluorescent mice over arbitrary and unlimited imaging time points for several months renders the chronic cranial window an excellent preparation for studies of neuron-subtype specific structural and functional plasticity.
We thank K. Masback, X. Zhang and A. Canty for helping with data analysis, and S. Song and T. O'Conner for writing image analysis software. This work was supported by HHMI, NIH, the IRP Foundation, the Swiss National Science Foundation, the Max Planck Society, the Larry L. Hillblom Foundation and the March of Dimes Foundation.