In Vivo Two-Photon Imaging of Dendritic Spines in Marmoset Neocortex1,2,3

Abstract Two-photon microscopy in combination with a technique involving the artificial expression of fluorescent protein has enabled the direct observation of dendritic spines in living brains. However, the application of this method to primate brains has been hindered by the lack of appropriate labeling techniques for visualizing dendritic spines. Here, we developed an adeno-associated virus vector-based fluorescent protein expression system for visualizing dendritic spines in vivo in the marmoset neocortex. For the clear visualization of each spine, the expression of reporter fluorescent protein should be both sparse and strong. To fulfill these requirements, we amplified fluorescent signals using the tetracycline transactivator (tTA)–tetracycline-responsive element system and by titrating down the amount of Thy1S promoter-driven tTA for sparse expression. By this method, we were able to visualize dendritic spines in the marmoset cortex by two-photon microscopy in vivo and analyze the turnover of spines in the prefrontal cortex. Our results demonstrated that short spines in the marmoset cortex tend to change more frequently than long spines. The comparison of in vivo samples with fixed samples showed that we did not detect all existing spines by our method. Although we found glial cell proliferation, the damage of tissues caused by window construction was relatively small, judging from the comparison of spine length between samples with or without window construction. Our new labeling technique for two-photon imaging to visualize in vivo dendritic spines of the marmoset neocortex can be applicable to examining circuit reorganization and synaptic plasticity in primates.


Introduction
Direct observation of fine neuronal morphologies such as dendritic spines in living brains has been made possible with techniques involving the expression of fluorescent protein in neurons of living animals in combination with two-photon microscopy. Many researchers prefer to use transgenic mouse lines such as Thy1-GFP and Thy1-YFP mice for their imaging studies of dendritic spines, because of the stable and strong expression of fluorescent protein in a subpopulation of neurons in these mouse lines (Feng et al., 2000;Grutzendler et al., 2002;Trachtenberg et al., 2002;Holtmaat et al., 2005;Zuo et al., 2005;Kim and Nabekura, 2011;Fu et al., 2012). The advancement of two-photon microscopy was also a key factor for the application of spine-imaging techniques to living animals (Denk et al., 1990;Denk and Svoboda, 1997). The long-wavelength light used in two-photon microscopy penetrates a specimen with less scattering than the shortwavelength light used in other conventional microscopy techniques such as confocal microscopy; thus, it is possible to observe signals from deep regions in a certain thick tissue.
Dendritic spine imaging by two-photon microscopy has been almost exclusively performed on rodent brains, and there have been only a few studies in which the brains of larger species such as primates and carnivores have been examined. One major reason for the lack of studies of such animals with larger brains is that there has been no standard method for in vivo imaging in these species. Transgenic monkey studies have been generated Niu et al., 2010), but no primate model that strongly expresses fluorescent protein for observing signals in vivo has been established. Spine imaging of the primary visual cortex of ferrets using an expression system with a virus vector was reported (Yu et al., 2011). In the case of primates, one group studied the neuronal morphology in the primary visual cortex of macaque mon-keys by in vivo imaging using virus expression systems (Stettler et al., 2006;), but their observation and analysis focused not on dendritic spines but on axonal structures. Using virus vector-based methods for the labeling of neurons to study their morphology is difficult because the high density of labeled neurons around the injection site also makes the background signal intensity high, thus making such methods unsuitable for the observation of the morphology of fine structures such as dendritic spines. Although dendritic spines in the primate cortex have been extensively analyzed in fixed samples by dye injection methods (Elston et al., 1999;Oga et al., 2013), there has been no report in which dendritic spines were imaged and analyzed in vivo in primate brains.
In this report, we present a method of in vivo imaging of dendritic spines in the marmoset neocortex. The marmoset was chosen as a model animal in our study because the flat surface of a marmoset brain is advantageous for studying the entire cortical region. We mainly had to overcome two technical problems in the in vivo visualization of dendritic spines in the marmoset neocortex. We needed a stronger expression of fluorescent protein because the marmoset brain is more opaque than the brains of smaller animals such as mice. In addition, we needed neurons to be sparsely labeled, because the dense expression obtained by the virus vector method usually causes a high background signal intensity. We, therefore, used an adeno-associated virus (AAV) expression system for the sparse and strong expression of the fluorescent protein in cortical neurons of marmosets, and by twophoton microscopy we were able to observe more clearly dendritic spines labeled by fluorescent proteins.

Animals
We used six marmosets (all males; body weight, 310 -420 g; age, 13-22 months); five animals for in vivo imaging and one animal for dye injection. All the protocols used in this study were approved by the Institutional Animal Care and Use Committee of the National Institutes of Natural Sciences, Japan. The experiment was also conducted in accordance with the Guide for the Care and Use of Laboratory Animals of the U.S. National Institutes of Health.

Plasmid construction and AAV preparation
The constructs used in this study are schematically shown in Figure 1A. The Thy1S promoter was cloned from pThy1S-GFP for the sparse labeling of cortical neurons as previously reported (Ako et al., 2011). Owing to capacity limitations of the AAV vector, we truncated ϳ1.3 kb of the 5' region of the Thy1S promoter, which is reported to be nonessential for the activity of the promoter (Vidal et al., 1990;Caroni, 1997). The plasmid AAV (pAAV):Thy1Stetracyline transactivator (tTA) was constructed by subcloning the DNA fragments containing the truncated Thy1S promoter and tTA in pAAV-MCS (Agilent Technologies). The pAAV:tetracycline-responsive element (TRE)humanized renilla GFP (hrGFP)-humanized renilla GFP (hrGFP) was constructed by replacing the TurboRed flu-orescent protein (tRFP) sequence of AAV:TRE-tRFP (Watakabe et al., 2014) with hrGFP.
The AAV vectors used in this study have capsids of serotype 1. They were produced in HEK 293 cells using a helper-virus-free system and were purified twice by CsCl 2 density gradient centrifugation and titrated by quantitative PCR, as described previously (Konishi et al., 2008). The final preparations obtained were dialyzed against PBS and diluted as described in the Results section. To prevent adhesion of the AAV vector to glass micropipettes, Pluronic-F68 (Sigma-Aldrich) was added to the vector stock at 0.001%.

Virus injection
The marmosets were treated by intramuscular injections of ketamine (20 mg/kg; Daiichi Sankyo) and xylazine (1 mg/kg; Bayer Health Care). Under deep anesthesia induced and maintained by isoflurane (1-2%) inhalation (Abbott Laboratories), the head of the animal was fixed to a stereotaxic apparatus. Pulse rate, O 2 saturation (SpO 2 ), and rectal temperature were continuously monitored. A small hole was drilled in the skull using a dental drill. To inject the viruses into the cortex, the dura was punctured using the tip of a 27 ga needle, through which a glass pipette was slowly inserted to a depth of 500 m from the cortical surface. Approximately 0.5 l of a viral solution was injected at a rate of 0.1 l/min. For the imaging of the prefrontal cortex (PFC), our injections were targeted anteroposterior ϩ18.5 mm, 2 mm to the right from the midline. Following the viral injection, the hole was filled with Spongel, an absorbable gelatin sponge (Astellas Pharma Inc.), and the scalp was sutured. Then the animal was returned to a cage and remained there until the imaging sessions started. To prevent infection, ampicillin (40 mg/kg; Meiji Seika Pharma) was administered intra-muscularly. Carprofen (5 mg/kg; Pfizer) was administered intramuscularly as an analgesic and an anti-inflammatory agent. Ampicillin and carprofen were administered immediately after surgery and 2 subsequent days. We waited until an adequate level of gene expression was obtained, which took at least 2 weeks.

In vivo imaging
Before an imaging session, we constructed an imaging window on the head of the animal. The hole in the skull used for virus injection was expanded to a size of ϳ2 ϫ 3 mm 2 using a dental drill, and part of the dura above the injection site was deflected and resected, yielding an incision ϳ1 mm in diameter ( Fig. 2A). A small coverglass ϳ4 ϫ 4 mm 2 in size was fixed with dental cement on top of the skull, and the space between the coverglass and the cortex was filled with an agarose gel (1.5% in artificial CSF; type III-A; Sigma-Aldrich) to minimize vibration. A custom-made metal plate with a hole having an 11 mm inner diameter was glued to the skull (Figs. 2B,C). This plate was used to fix the head of the animal during the imaging sessions. The same antibiotic, analgesic, and anti-inflammatory agents as those used for the virus injections were administered immediately after window construction and on 2 subsequent days. We started the imaging sessions from 1 to 7 d after the imaging window construction (Fig. 2D).
In vivo two-photon imaging was performed using a FV1000MPE Multiphoton Laser Scanning Microscope (Olympus) and a water immersion objective lens [25ϫ; numerical aperture (NA), 1.05; Olympus]. Two-photon excitation (920 nm) was provided by a mode-locked Ti: Sapphire laser (MaiTai Deep See, Spectra-Physics). Fluorescence was detected using a multialkali photomultiplier tube (PMT) without any filter in front of the PMT. The regions with fluorescently labeled dendrites were identified without digital zoom (field of view, 508 ϫ 508 m 2 ). Then we used an 8ϫ digital zoom to acquire images of magnified sites including dendritic spines at a resolution of 0.124 ϫ 0.124 ϫ 0.2 m 3 (field of view, 63 ϫ 63 m 2 ). These sites were identified the following day, and the images obtained on 2 consecutive days were analyzed. Imaged sites were scattered around a 508 ϫ 508 m 2 region and were presumed to contain dendrites originating from different neurons. During imaging sessions, the marmosets were anesthetized with isoflurane (1-2%). Pulse rate, SpO 2 , and rectal temperature were continuously monitored.

Dye injection
A marmoset was sedated with ketamine hydrochloride (25 mg/kg, i.m.; Daiichi Sankyo) and were overdosed with sodium pentobarbital (75 mg/kg, i.p.; Dainippon Sumitomo Pharma). The animal was perfused intracardially with 0.1 M potassium PBS, pH 7.2, followed by 4% paraformaldehyde (Merck) in 0.1 M PB. A block of tissue was excised from the PFC. Coronal slices of 250 m thickness were prepared from the block. Slices were incubated in DAPI (Sigma-Aldrich) solution to visualize the cell bodies. Pyramidal cells were individually injected with 10 mM Alexa Fluor 568 (Alexa Fluor 568 Hydrazide in 200 mM KCl, A-10441; Thermo Fisher Scientific) under visual guidance with a triple-band fluorescence filter (Semrock). The dyeinjected neurons were imaged using a Leica SP-8 confocal laser scanning microscope and a water-immersion lens (63ϫ; NA, 1.2; Leica) at a resolution of 0.045 ϫ 0.045 ϫ 0.336 m 3 .

Image analysis
We used ImageJ (National Institutes of Health), Neurolucida (MBF Bioscience), and custom-made software written on MATLAB (R2009a; MathWorks) for our image analysis. Images were processed with a median filter (2.0 pixel radius) to reduce noise. Dendrites and spines were traced and marked manually in a three-dimensional space. The loss or gain rate of dendritic spines was calculated as the percentage of spines that appeared or disappeared on day 1, relative to the total number of spines on day 0. The length of spines was measured from the tip of the spine to the interface with the dendritic stalk .

Statistical analysis
All statistical analyses were performed using R (R Development Core Team, 2012). We used the Wilcoxon rank sum test to compare between groups, and we corrected for multiple comparisons when required. Differences were considered to be significant at p Ͻ 0.05. Measurements are reported as the mean and SD. The values of statistical power were calculated using G‫ء‬Power (http://www.gpower.hhu.de/en.html) and are presented in Table 1. Ex- perimental animals were randomly assigned to in vivo or ex vivo conditions.

Sparse and strong expression system for spine visualization
To observe the dendritic spines of neurons in the marmoset cortex in vivo, there are the following two requirements: strong and sparse expression of fluorescent protein. The scattering of light in living tissues prevents weak signals from being captured by the detector, particularly in the marmoset brain, which is more opaque than the mouse brain; thus, a strong expression is required. Moreover, to observe fine structures such as spines, the expression of fluorescent protein should be sufficiently strong to adequately label these fine structures. Sparse expression is also required to reduce the intensity of background signals. Even when the expression level of fluorescent protein in each neuron is sufficiently strong, a dense expression of fluorescent protein in neighboring neuronal structures such as dendrites and axons makes the intensity of background signals high, thus preventing the clear observation of dendritic spines.
Our strategy to achieve the requirements described above was to combine the tTA-TRE system and Thy1S promoter (Ako et al., 2011) using two virus vectors. We constructed two AAV vectors: one had the tTA component under the control of the Thy1S promoter (AAV: Thy1S-tTA) and the other had hrGFP under the control of TRE (AAV:TRE-hrGFP). Figure 1A   oped the Thy1S promoter to sparsely label the fine structures of neurons in the mouse neocortex. When expressed by electroporation in the mouse neocortex, the Thy1S promoter drives the gene expression only in a small number of pyramidal neurons in layers 2/3 and 5. In our expression system, the tTA-TRE system was driven only in Thy1S-positive cells. Hioki et al. (2009) showed that the tTA-TRE system is effective in amplifying transgene expression. The tTA-TRE system has two components, one is tTA and the other is TRE. When tTA binds to the TRE component, the transcription of hrGFP that is under the control of the TRE component is strongly activated.
We injected a mixture of these virus vectors into the marmoset neocortex and first examined the expression in fixed brain samples (Fig. 1B). We were interested in the density of hrGFP-positive neurons and the visibility of each dendrite. In our preliminary mouse experiment, we found that maintaining the concentration of AAV:TRE-hrGFP at 4.6 ϫ 10 12 vector genomes (vg)/ml and reducing the concentration of AAV:Thy1S-tTA to ϳ1/500 of the TRE vector (8.8 ϫ 10 9 vg/ml) leads to the sparse but strong expression of the fluorescent protein (data not shown). On the basis of this finding, we tested various concentrations of AAV:Thy1S-tTA relative to that of AAV:TRE-hrGFP in marmosets. At an AAV:Thy1S-tTA to AAV:TRE-hrGFP concentration ratio of 1:200, the number of hrGFPpositive neurons was relatively high (Fig. 1B, left). At a concentration ratio of 1:2000, the number of hrGFPpositive neurons was too small (Fig. 1B, right). An AAV: Thy1S-tTA to AAV:TRE-hrGFP concentration ratio of 1:500 yielded the most desirable signals in terms of intensity and density (Fig. 1B, middle).
We emphasize the importance of amplifying the expression level of hrGFP using the tTA-TRE system. Simply reducing the titer of the virus leads to a low expression level, thus making it difficult to observe spines in vivo. The combination of titration and amplification was required for the clear visualization of dendritic spines in the marmoset neocortex.

In vivo visualization of dendritic spines in neocortex
To test the feasibility of our viral expression system in vivo, we injected the virus into the marmoset neocortex. After 2 weeks of an expression period, we acquired images from living animals under anesthesia induced by isoflurane. Figure 3 shows in vivo captured images. We were able to visualize each dendritic spine using our virus constructs (Fig. 3C-E). We observed the dendritic spines of apical dendrites located in layer 1. We were also able to observe the cell bodies located at a depth of ϳ300 m from the pia (Fig. 3F, at 220 m; Fig. 3G, at 330 m).
We then acquired images repeatedly from the same region over time. Figure 4 shows time-lapse images of the dorsolateral prefrontal cortex, presumably area 8B or 9. The images of the same region of the dendrites were taken over time at 24 h intervals. During these imaging sessions, the clarity of the imaging window was maintained. Because our samples contained a relatively small number of hrGFP-positive neurons per injected site, we were able to easily identify the same dendrite that we observed in the previous imaging session. The overall shapes of dendrites did not change over this imaging period (Fig. 4A, top and bottom). We marked each spine on the two images (day 0 and day 1) by comparing these images side by side, and identified the spines that were "gained" or "lost" during this time interval (Fig. 4B,C). In our experiments, we analyzed 779 spines (12 sites; 34 dendrites; 3 animals; total dendrite length, 2238 m); of these spines, 51 were gained (mean across sites, 6.4%; SD, 4.2%) and 49 were lost (mean across sites, 5.6%; SD, 3.6%; Fig. 4D). The loss or gain rate at the 1 d interval observed in this study was similar to those in previous studies of layer 5 neurons of the somatosensory cortex of transgenic mice (ϳ12% in 3 d for both loss and gain; Kim and Nabekura, 2011) and layer 2/3 neurons of ferret V1 by the virus vector method (ϳ4% in 1 d for both loss and gain; Yu et al., 2011). We measured spine length by manual tracing using Neurolucida software, and examined the difference between the distribution of the spines that persisted and that of the spines that changed (gained and lost) over the period of imaging. We observed the tendency that the changed spines were shorter than those that persisted (Wilcoxon rank sum test, changed vs persisted, p ϭ 0.0009 a ), which means that shorter spines tend to be gained or lost (Fig. 4E).
One of the concerns raised in the two-photon imaging of dendritic spines in mice studies was the activation of glial cells under invasive procedures. Because our methods in the marmoset neocortex include invasive procedures of virus injection and dura opening, we checked the activation of glial cells in our sample of the prefrontal cortex that was used for the in vivo imaging study. We observed the activation of both astrocytes (GFAP; Fig.  5A-C) and microglias (Iba1; Fig. 5D-F) around the injection site. This observation indicated that experimenters should carefully choose the experimental paradigm when applying the method presented in this article (see Discussion).

Comparison between in vivo and fixed samples
To evaluate our in vivo two-photon microscope images of dendritic spines in the marmoset neocortex, we compared them with those of fixed samples. We wanted to determine whether we observed all existing spines or only a subpopulation of these spines. To clarify this point, we used the method that was used to identify the morphological differences of basal dendrites of pyramidal neurons in different cortical areas from various species, including marmosets and macaques (Elston et al., 1999;Oga et al., 2013). We injected the dye Alexa Fluor 568 into the neurons in the coronal sections of the marmoset prefrontal cortex, corresponding to the region we imaged under the in vivo condition, and observed the spines of apical dendrites of these dye-injected neurons under a confocal microscope (Fig. 6A; one animal; 15 sites; 22 dendrites; 1219 spines; total dendrite length, 1125 m). Our injection labeled dendritic spines well in the distal region of apical dendrites. We quantified the density and shape of spines in our fixed samples by manual tracing using Neurolucida software, and we compared the results with our in vivo data. The density of spines under our in vivo condition (mean, 0.36 spines/m; SD, 0.14) was significantly lower than that in dye-stained samples (mean, 1.12 spines/m; SD, 0.21; Wilcoxon rank sum test with Holm correction, p ϭ 0.00038 b ; Fig. 6B). The spines observed by in vivo two-photon imaging were shorter than those in dye-stained fixed tissues (Wilcoxon rank sum test with Holm correction, p ϭ 7.6 ϫ 10 Ϫ12c ; Fig. 6C). This comparison of results between in vivo two-photon imaging and dye staining of fixed samples indicated that the observation of dendritic spines by the in vivo imaging system still has certain limitations.
As for the causes of these limitations, there are two possibilities. The most likely possibility is that because of inefficient visualization, we underestimated the number of spines in in vivo imaging. The second possibility is that the damage caused by AAV injection decreased the number of spines. To determine which of these two possibilities is true, we immunohistochemically stained our samples used in vivo imaging with the antibody to hrGFP immunohistochemically stained [immunohistochemistory (IHC) samples], and then observed these fixed samples by confocal microscopy (two animals; 15 sites; 20 dendrites; 839 spines; total dendrite length, 951 m). Although one may consider the comparison between the observation in vivo and the same dendritic segments identified in fixed condition, it was difficult for us, in a practical way, to identify the same dendritic segments in the more crowded fixed samples because immunohistochemical staining amplified the signals even in dendrites that express hrGFP weakly. Thus, here we compared the observation in vivo with the dendritic segments at the matching regions from the same samples. The spine density of IHC samples was much higher than that of samples in in vivo imaging (Wilcoxon rank sum test with Holm correction, p ϭ 0.00038 d ; Fig. 6B), suggesting that we were not able to detect weaker signals under in vivo two-photon imaging. However, the spine density of IHC samples was lower than that of dye-stained fixed samples (Wilcoxon rank sum test with Holm correction, p ϭ 0.0028 e ), which also suggests that there may have been some losses of spines owing to tissue damage, as mentioned above. However, the distribution of spine length of IHC samples largely overlapped with that of dye-stained fixed samples, but not with that of in vivo samples (Fig. 6C). This indicates that the window construction only mildly affected dendritic spines (Wilcoxon rank sum test with Holm correction; IHC vs dye, p ϭ 0.64 f ; IHC vs in vivo, p ϭ 2.2 ϫ 10 Ϫ10g ).

Discussion
In this study, we established a method of visualizing dendritic spines in the marmoset cortex by in vivo twophoton imaging using a virus expression system. To the best of our knowledge, this is the first demonstration of two-photon imaging of dendritic spines in the neocortex of a living primate.

Merits of this method
By virus injection, we can control gene expression and monitor the morphology of transduced neurons without affecting other parts of the brain. Even though transgenic marmoset lines in which a specific neuronal population expresses fluorescent protein may be available for in vivo imaging in the near future, virus expression systems have their own merits. One advantage is the shorter time required for their preparation. Generating transgenic primate lines requires a much longer time. Another merit is region-specific manipulation. The use of the combination of transgenic animals and virus expression systems will be beneficial in future studies. For example, a specific cell type can be targeted by the transgenic insertion of an appropriate promoter, and the local expression of a gene may be induced by an appropriate virus expression system.

Technical considerations
Our method presented in this article still has some limitations. First, there may have been tissue damage due to the procedures used in our study. The dura of the marmoset brain is more opaque than that of the mouse brain, so we dissected the dura to observe dendritic spines in vivo by two-photon imaging. This procedure may be more invasive than that used in mice. In mice, some researchers argued that even removing the skull could perturb the underlying brain tissues and make dendritic spines more unstable (Xu et al., 2007). Indeed, as shown in Figure 5, we observed the activation of glial cells around the injection site. Since a previous study in the primary visual cortex of the ferret (Yu et al., 2011) was conducted under conditions comparable to those in our study (using the Sindbis virus), and examined the activity-dependent morphological plasticity of dendritic spines by comparing active and inactive ocular dominance column, likewise our methods presented in this article may be applicable to studies that compare morphological plasticity in different conditions, such as the level of sensory input or the state of learning. Therefore, we believe that our method is an important technical improvement toward the understanding of the function of the primate neocortex. However, we have to further improve our methods so that the tissue will be less damaged. One possible improvement is to construct an imaging window that permits the longer imaging period, and allows waiting for time until the activity of glial cells subsides. Another improvement could be developing procedures to acquire images through the dura. As part of the process of developing the methods described in this article, we are now on the way to acquire images of axons through the intact dura of the marmoset cortex. In future studies, the improvement of the method for imaging dendritic spines through the dura will lead to less invasive imaging.
The second limitation is in detecting weak signals, as shown in Figure 6. Therefore, we need to amplify such signals or use a detector with high sensitivity.

Importance of examining spines in brains of living primates
The majority of imaging studies of spines were performed in mice, because many molecular biological techniques are available for mice. Although experiments using mice are important, there are certain functions and structures that only primates have acquired during the course of their evolution. For example, the area specialization of the neocortex is far more evolved in primates than in rodents. In previous studies, the area-specific gene expression profiles in the primate neocortex were examined (Yamamori and Rockland, 2006;Yamamori, 2011;Bernard et al., 2012). Genes selectively expressed in association areas (Komatsu et al., 2005;Sasaki et al., 2010) and the primary visual cortex (Takahata et al., 2006; were reported. Importantly, areaselective expressions of these genes are not observed in rodents (Yamamori, 2011). In relation to the gene expression patterns in different cortical areas, previous studies have shown that the density of spines significantly differs among different areas of the primate neocortex (Elston et al., 1999(Elston et al., , 2005Elston and Rockland, 2002). There is a smaller difference in the density of spines among different cortical areas in mice than in primates (Ballesteros-Yáñez et al., 2006). In addition, previous studies by in vivo twophoton imaging of spines in mice showed that there is no significant difference in morphological plasticity among different areas (Zuo et al., 2005). It is, therefore, of great interest to us to determine whether there is a difference in morphological plasticity among different areas in the primate brain, which we are going to investigate using the method described in this report.