Neuron, Vol. 45, 279���291, January 20, 2005, Copyright ��2005 by Elsevier Inc. DOI 10.1016/j.neuron.2005.01.003 Transient and Persistent Dendritic Spines in the Neocortex In Vivo ity of excitatory synapses (Nimchinsky et al., 2002). Spines emerge to make synapses (Engert and Bonhoef- fer, 1999 Maletic-Savatic et al., 1999) or expand (Matsu- Anthony J.G.D. Holtmaat,1 Joshua T. Trachtenberg,1,3 Linda Wilbrecht,1 Gordon M. Shepherd,1 Xiaoqun Zhang,1 Graham W. Knott,2 and Karel Svoboda1,* zaki et al., 2004), in response to synaptic stimulation (Knott et al., 2002 Toni et al., 1999). Long-term in vivo 1Howard Hughes Medical Institute Cold Spring Harbor Laboratory imaging in S1 has revealed that some spines appear and disappear in an experience-dependent manner, as- Cold Spring Harbor, New York 11724 2 Institut de Biologie Cellulaire et de Morphologie sociated with synapse formation and elimination, while other spines persist for at least a month (Trachtenberg Universite �� de Lausanne Rue du Bugnon 9 et al., 2002). In V1, spines were found to be largely persistent (Grutzendler et al., 2002). However, a direct CH 1005 Lausanne Switzerland comparison of these studies is complicated because of differences in experimental methods, the age and strain of the mice used, and the cell types involved. Here, we imaged the apical dendritic tufts of layer (L) Summary 5B and 2/3 neurons in the developing and adult somato- sensory and visual cortex to measure the parameters Dendritic spines were imaged over days to months governing dendritic and spine structural plasticity. High- in the apical tufts of neocortical pyramidal neurons resolution, chronic, time-lapse imaging revealed that a (layers 5 and 2/3) in vivo. A fraction of thin spines fraction of spines appear and disappear over days, while appeared anddisappeared over afew days,while most other spines persist for months. The persistent fraction thick spines persisted for months. In the somatosen- increased gradually with development and continued to sory cortex, from postnatal day (PND) 16 to PND 25 increase in the adult. Spine turnover was more rapid in spine retractions exceeded additions, resulting in a the somatosensory than in the visual cortex. net loss of spines. The fraction of persistent spines (lifetime 8 days) grew gradually during development and into adulthood (PND 16���25, 35% PND 35���80, 54% Results PND 80���120, 66% PND 175���225, 73%), providing evi- dence that synaptic circuits continue to stabilize even In the majority of experiments, we used transgenic mice expressing EGFP in a subset of L2/3, L5, and L6 pyrami- in the adult brain, long after the closure of known critical periods. In 6-month-old mice, spines turn over dal neurons (line GFP-M) (Feng et al., 2000 Trachten- berg et al., 2002). In some experiments, we used a similar more slowlyin visualcompared tosomatosensory cor- tex, possibly reflecting differences in the capacity for transgenic mouse line in which a much larger fraction of the same neuronal subtypes are labeled (line YFP-H) experience-dependent plasticity in these brain re- gions. (Feng et al., 2000 Grutzendler et al., 2002). All neurons were reconstructed either from in vivo images or in fixed sections (Experimental Procedures). Introduction To determine whether the GFP-positive cells com- prise a normal subset of cortical neurons, we used laser- In the mammalian neocortex, neural circuits are sculpted scanning photostimulation in brain slices to map the by spontaneous and sensory-evoked activity (Katz and spatial distribution of their synaptic input (see the Sup- Shatz, 1996). Although plasticity is most rapid and ro- plemental Data at http://www.neuron.org/cgi/content/ bust during early postnatal development (Katz and full/45/2/279/DC1/) (Callaway and Katz, 1993 Shepherd Shatz, 1996), sensory representations in the adult cortex et al., 2003). GFP-positive L5B cells and GFP-negative remain dynamic (Darian-Smith and Gilbert, 1994 Daw neighbors received indistinguishable patterns of synap- et al., 1992 Diamond et al., 1994 Fox, 2002 Gilbert, tic input (Supplemental Figure S1). We conclude that 1998 Wang et al., 1995). GFP-positive neurons constitute a functional and repre- Functional rewiring of cortical circuits may involve sentative subset of cortical cells. structural plasticity with synapse formation and elimina- Animalswerepreparedfor imagingbyimplantingasmall tion (Antonini and Stryker, 1993 Chklovskii et al., 2004 glass window centered over somatosensory or visual cor- Knott et al., 2002 Lendvai et al., 2000 Lowel and Singer, tex (Experimental Procedures). We used 2-photon laser- 1992 Ramon y Cajal, 1893 Stepanyants et al., 2002 scanning microscopy (2PLSM) (Denk et al., 1990 Denk Trachtenberg et al., 2002 Turner and Greenough, 1985 and Svoboda, 1997) to repeatedly image dendrites and Ziv and Smith, 1996). In particular, recent work has fo- their spines during development and in the adult brain. cused on dendritic spines as a possible substrate of circuit plasticity (Bonhoeffer and Yuste, 2002). Spines are tiny dendritic protrusions that receive the vast major- Pruning of Dendritic Spines during Development Previous studies in young adult (2 months old) neocortex found that a fraction of dendritic spines appear and *Correspondence: svoboda@cshl.edu disappear over days, while a subpopulation (50%���60%) 3Present address: Department of Neurobiology, Box 951761, 695 Charles Young Drive South, Los Angeles, California 90095. of spines persists for at least a month of time-lapse

Neuron 280 imaging (Trachtenberg et al., 2002). How does spine Age-Dependent Regulation of Spine Plasticity stability change with developmental age? To begin to and Stability in the Adult Brain address this question, we performed chronic imaging Could spine plasticity be regulated by age even in experiments in developing S1 spanning the third and the adult brain? To answer this question, we imaged the fourth postnatal week. These experiments are compli- apical dendritic arbors of individual L5B neurons in the cated by the developmental regulation of expression of S1 cortex of mature adult (age 6 months) GFP-M mice fluorescent proteins under control of the Thy-1 promoter (Figure 2A) and compared the results to our experiments in the transgenic mice (Feng et al., 2000) (Figures 1A in developing mice (Figure 1) and to previous experi- and 1B). Expression in sensory neocortex starts toward ments in young adult mice (Trachtenberg et al., 2002). the end of the second postnatal week, but at this time In the mature adult animals, a subpopulation of spines only the YFP-H line had adequate numbers of brightly appeared and disappeared from day to day (n 5 1286 labeled cells for in vivo imaging (Figures 1A and 1B). At spines) (Figure 2B Supplemental Figures S2A and S2B postnatal day (PND) 16, a sparse subset of L5B pyrami- at http://www.neuron.org/cgi/content/full/45/2/279/ dal cells were labeled in S1 (Figure 1B). Over the time DC1/). These spines were usually thin, as judged by their course of the experiment, the field of view became low fluorescence level, compared to bright mushroom crowded with dendrites and axons that began to ex- spines on the same dendrites, which tended to persist press YFP at later time points (Figures 1A and 1B). We (Figure 2B). Spine densities and daily turnover ratios focused our analysis on dendritic segments that were (TOR, the fraction of spines appearing and disappearing separated from neighboring branches and could be from day to day) were constant over time (density, 0.29 tracked over all experimental days. 0.08 m 1 TOR, 15.4% 1.6% Figures 2C and 2D) To monitor spine turnover and stability, we collected (Trachtenberg et al., 2002). Also in experiments that high-resolution image stacks of several (five to ten) re- lasted for several months, spines were seen to appear gions of interest per dendritic tuft (Figure 1C). Around and disappear during the entire imaging period (Figure 2 weeks of age, dendritic branches were studded with 2F Supplemental Figure S2C). These observations ar- numerous spines (0.49 0.13 m 1 n 5 PND 16) gue against the possibility that spine addition and sub- comprising all commonly described classes, including traction were caused by implanting the imaging window. mushroom-type, stubby, and thin spines (Peters and To compare spine dynamics and stability between Kaiserman-Abramof, 1970), as well as long filopodia- different ages and different cortical regions, we calcu- like protrusions (Dailey and Smith, 1996 Lendvai et al., lated the spine survival fraction for each cell as a func- 2000 Portera-Cailliau et al., 2003). tion of time (survival function [SF]). SFs could be fit with The development of dendritic spines was analyzed in an exponential function and a constant term. The time daily time-lapse images (922 spines n 5) (Figures constant, , of the exponential (a few days) is a measure 1C���1F). Spine densities were higher at younger ages of the rate of spine turnover. The constant term reflects (Figure 1C), implying a net loss of dendritic spines with the persistent fraction (Figure 6A see Experimental Pro- developmental age. The decrease in spine density did cedures). not proceed simply by retraction of dendritic spines For 6-month-old mice, the SF revealed that spines rather, at young ages (PND 16) more than 30% of the that ultimately disappeared lasted at most a few days spines disappeared, while 25% appeared, between im- after they were formed ( 1.9 days Figures 2E and aging sessions from one day to the next (Figures 1C 6A). We defined spines that persisted for four days or and 1D). This resulted in a gradual decrease in the spine less as ���transient��� spines. Spines that persisted for 8 density (to 0.30 0.08 m 1 at PND 26) (Figure 1E). days or more are ���persistent��� spines. Although addition A similar decrease in spine density was observed in and subtraction of persistent spines was observed (Fig- measurements from naive perfusion-fixed tissue derived ure 2G), these events were rare spines that survived for from YFP-H mice (0.44 0.10 m 1, PND 16���18, n more than 8 days were highly likely (93%) to persist for 10 0.32 0.02 m 1, PND 25���33, n 3 indistinguish- the rest of the imaging session, more than 3 weeks able from the decrease seen in vivo analysis of covari- (Figure 6A). We further imaged two mice for over 3 ance). This implies that the developmental decrease in months with longer sampling intervals. These experi- spine density was not due to chronic imaging or devel- ments confirmed that spines that persisted for 8 days opmental changes in detection efficiency of small were highly likely to persist for 3 months (Figure 2F). spines. Rates of spine retraction and addition in vivo Under conditions of constant sensory experience, tran- decreased with developmental age at different rates, so sient and persistent spines therefore constitute largely that at 4 weeks of age addition and subtraction were distinct populations. balanced (Figure 1D). Thereafter, spine densities were In 6-month-old mice, the fraction of transient spines relatively stable (Trachtenberg et al., 2002). was significantly lower (21.4% 4.3% versus 30.2% We tracked the fates of individual spines observed in 6.9% p 0.05)���and the fraction of persistent spines the first image and calculated the fraction of spines was significantly larger (72.5% 2.6% versus 53.9% surviving (survival fraction) as a function of time (Figure 8.5% p 0.01)���compared to young adult (5���11 weeks) 1E). The fraction of spines that persisted for at least 8 animals. We noticed that dendrites with low spine densi- days was 35.0% 9.9% (n 5), much less than that ties tend to have higher proportions of transient spines previously observed in young adult mice (Trachtenberg than dendrites with high spine densities (Figure 6C). et al., 2002) (53.9% 8.5% n 6 p 0.05). This Therefore, we quantified the density of transient spines indicates that spine plasticity and stability is regulated per cell as a measure of spine dynamics, which is inde- during development, after the closure of critical periods in S1 ( PND 15) (Fox, 1992 Stern et al., 2001). pendent of total spine density and has lower variance.