Synaptic remodeling at the heart of memory processes

(1)

Fundamental Medicine Section Department of Basic Neuroscience 

"Synaptic remodeling at the heart of memory processes"

Thesis submitted to the Faculty of Medicine of the University of Geneva  

for the degree of Privat-Docent by

--- Mathias DE ROO

(Geneva)

(2020) 

(2)

Acknowledgements

This thesis is dedicated to Pr. Dominique Muller. Working with him during these years was a great pleasure and a source of knowledge in multiple domains. My deep gratitude remains.

During this period, I was constantly stimulated by the inputs of several PhD students, post-docs and researchers who worked with me and I want to thank them for this: Paul Klauser, Lorenzo Poglia, Pablo Mendez, Irina Nikonenko, Sylvain Steen, Jakub Wlodarczyk, Laura Kehoe, Yann Bernardinelli, Carmen Flores, Enora Moutin, Thomas Marissal, Christina Bertollini, Mari Virtanen, Galil Mori and Greta Limoni.

I also wish to thank the professors and students in and outside the department of Neuroscience with whom I had the pleasure to collaborate: Pr. Jozsef Kiss, Eduardo Gascon, Ricardo Bocchi, Kristof Egervari, Pr. Laszlo Vutskits, Adrian Briner, Dr. Adema Ribic, Pr. Pierre Maechler, Dr. Melis Karaca, Pr. Christian Lüscher, Christina Bocklisch, Pr.

Hilal Lashuel, Filip Vercruysse, Dr. Anne-Laure Mahul-Mellier, Pr. Doron Merkler, Dr.

Nicolas Page, Dr. Giovanni Di Liberto, Pr. Pierre Magistretti, and Pr. Alan Carleton.

For essential technical support during all these years, I wish thank Marlis Moosmayer, Lorena Jourdain, René Umiker, Sébastien Pellat, Michelle Brunet and Philippe Correges.

After the tragic death of Pr. Dominique Muller, several Professors in the department have supported me and I wish to thank them especially for this: Pr. Jozsef Kiss, Pr. Laszlo Vutskits, Pr. Michel Muhlethaler, Pr. Laurent Bernheim, Pr. Anthony Holtmaat, Pr. Alan Carleton, Pr. Alexandre Dayer and Pr. Denis Jabaudon.

For the genesis of this thesis and the works it is based on, I wish to thank Pr. Rémy Schlichter, my PhD mentor who made me a scientist, and of course Pr. Dominique Muller without whom this habilitation thesis would simply not exist. I also wish to thank Richard Bohon for careful readings of this manuscript.

I am also extremely grateful to my wife, my children, my parents and my friends, who have given me invaluable support.

Lastly, I warmly thank Pr. Philippe Eigenmann, my rapporteur for this thesis, for his availability, his kindness and his advice. 

(3)

Summary

At the turn of the 21st century, the development of laser scanning microscopy combined with the methods of neuronal transfection with fluorescent proteins permitted the first ever real-time visualization of excitatory synapses following the induction of learning- related patterns of activity in living neurons in vitro, and later following sensory experience in living animals. These techniques, that are still improving today, led to the discovery of a causal link between learning and memory and structural plasticity of synapses, leading to long-term neuronal network remodeling. It has opened a new era in the study of the mechanisms that underlie learning and memory, and provided new therapeutic leads for brain disorders. This thesis describes my scientific contribution in this context.

(6)

Abbreviations:

2PLSM 2-Photon Laser Scanning Microscopy

AMPA α-Amino-3-hydroxy-5-Methyl-4-isoxazolepropionic Acid

CA Cornu Amonis

CamKII Calmodulin Kinase I

DAB 1,4-dideoxy-1,4-imino-d-arabinitol

DIV Day In Vitro

EGFP Enhanced Green Fluorescent Protein

EM Electron Microscopy

EPSP Excitatory Post-Synaptic Potential

GABA Gamma-Aminobutyric Acid

LPA Learning-related Pattern of Activity

LTD Long-Term Depression

LTP Long-Term Potentiation

MIP Maximum Intensity Projection

mRFP monomeric Red Fluorescent Protein

MRI Magnetic Resonance Imaging

NCad Neural Cadherin

NMDA N-Methyl D-Aspartate

PKC Protein Kinase C

PSD Post-Synaptic Denstity

TBS Theta Burst Stimulation

(7)

Preamble

« Synchronous stimulation of a pair of neurons induces a mechanism of cellular memory that will facilitate signal transmission between these two neurons, without changing the organization of their contacts »

This is what I would learn when I started my studies in neuroscience. Neurons that fire together would wire together during the first week of life, then these created networks would remain immobile in their structure. But at the turn of the 21 century, a few teams of researchers showed convincing evidence that a structural remodeling of the synapse network may occur following learning-related patterns of activity (Engert and Bonhoeffer, 1999; Maletic-Savatic et al., 1999; Toni et al., 1999). Soon after, growth and loss of synapses were observed in vivo in adult mice at basal state (Grutzendler et al., 2002; Trachtenberg et al., 2002) and following sensory experience (Trachtenberg et al., 2002), opening a new era of research.

I joined the lab of Pr Dominique Muller in order to participate in the study of the synapses structural plasticity and turnover that underly learning and memory. This has been a privilege.

(8)

1. Introduction

1.1. Synapses at dendritic spines

In the mammalian cerebral cortex, most excitatory synapses use glutamate as neurotransmitter and are made between a presynaptic axonal bouton and a post synaptic dendritic spine which contains AMPA and NMDA glutamatergic receptors (Kandel, 2013).

These spines are micrometer-sized protrusions that emerge from the dendrites of neurons (Figure 1).

Figure 1. Dendritic spines on a pyramidal neuron of the hippocampus.

Left: image obtained by laser scanning microscopy of a rat hippocampal organotypic slice culture (see 1.4.1) showing a neuron transfected with a plasmid expressing Enhanced Green Fluorescence Protein (EGPF).

Right: scan at high definition of the region represented in the orange rectangle. Note that an axonal bouton (not visible on the images) makes a synapse in front of each mature spine.

Scale bar: left: 100 µm, right : 5 µm

Personal material from my experiments.

Figure 1

dendrites

axons

dendritic spines

(9)

associated proteins, and faces an axonal presynaptic bouton. The neck of spines is variable in size, allowing different levels of electrical compartmentalization and filtering of excitatory post synaptic potential (Araya et al., 2006; Yuste, 2013). In some short spines, the neck is as large as the head. These particular spines are then called « stubby spines ». Some dendritic protrusions, called « filopodia », are very thin and devoid of a head and are thus not considered as spines, but rather as precursors of spines, searching for a presynaptic partner (Dailey and Smith, 1996; Ziv and Smith, 1996; Fiala et al., 1998; Yuste and Bonhoeffer, 2004). Nevertheless, in some publications, filopodia are sometimes also called

« spines » for simplification purposes. Importantly, filopodia are not systematically observed as a step of spine growth (Nagerl et al., 2007). These 3 main types of protrusions are shown in Figure 2.

Figure 2. The three main types of dendritic protrusions.

Image obtained as in Figure 1 showing the 3 main types of protrusions on the same dendritic branch: a mushroom spine (M), a stubby spine (S), and a filopodium (F). Scale bar: 1 µm.

Personal material from my experiments.

Figure 2

M

S

F

(10)

contain receptors of excitatory transmitters, scaffolding proteins and other functionally related proteins (Yamauchi, 2002). The main scaffolding protein of the AMPA receptor (which binds glutamate, the main excitatory neurotransmitter in the brain) is called PSD-95 (Friedman et al., 2000; Chen et al., 2005), and has been shown to regulate spine size and spine turnover upon increased activity (Ehrlich et al., 2007). Importantly, dendritic spines also contain actin filaments and numerous proteins that participate in molecular cascades that underlie synaptic plasticity and synapse stability (for review, see Yasuda, 2012). Importantly, some of these molecules appear to be regularly redistributed among nearby spines in an activity-dependent way, and this rule is also valid for presynaptic elements (Tsuriel et al., 2006).

Figure 3. Dendritic spines making synapses with axonal boutons.

Electron microscopic view of a section trough a dendrite (highlighted in green) and an axon (highlighted in red). Axonal boutons (B) make synapses with dendritic spines (S). Magenta arrow shows presynaptic vesicles containing neurotransmitter. Cyan arrow shows the post-synaptic density. Scale bar: 0.5 µm.

Adapted from an electron microscopy work kindly provided by Dr Irina Nikonenko, University of Geneva.

Of note, although the receptors of inhibitory transmitters are more often found on

S

B

(11)

the main inhibitory transmitter in the brain), is the gephyrin (Tyagarajan and Fritschy, 2014).

Before having the technical knowledge to allow the visualization of dendritic spines in living neurons, most scientists would argue that they were formed during development, as wiring and switches are set while constructing a building, and would stay still until they decline with advanced age. Our capacity to learn and remember was considered to rely only on the capability of neurons to modify their individual synaptic weight following specific patterns of activity.

1.2. Synaptic weight and memory.

1.2.1 Long term potentiation and long term depression at synapses.

The hypothesis that synaptic weight (or synaptic strength) increases during learning and memory was first proposed by Ramon y Cajal more than a century ago (Ramón y Cajal, 1911) and further modeled by Hebb in 1949 (Hebb, 1949). In his postulate, Hebb proposed that this increase in synaptic strength would result from a coincidence of pre and post-synaptic activity.

The first experimental evidence of this postulate came about 20 years later and was called Long Term Potentiation (LTP). Since the publication of its discovery in vivo in 1973 by Bliss and Lømo (Bliss and Lomo, 1973) LTP has been by far the most studied form of change in synaptic weight in the context of learning and memory (Nicoll, 2017). From these studies, we learn that LTP is made possible by the addition of new AMPA receptors at the synapse through exocytosis (Lledo et al., 1998; Makino and Malinow, 2009; Patterson et al., 2010;

Kennedy and Ehlers, 2011; Opazo et al., 2012; Jurado et al., 2013) and lateral diffusion

(12)

and/or post-synaptic molecular cascades that differ depending on the neurons studied (Zakharenko et al., 2001; Yasuda et al., 2003; Palmer et al., 2004; Nicoll and Schmitz, 2005).

In this thesis, we triggered LTP at the hippocampal CA3-CA1 synapse, the most studied site for LTP. Here LTP requires the activation of NMDA receptors as coincidence detectors (MacDermott et al., 1986; Ascher and Nowak, 1988), a fine tuning of free intracellular calcium levels (Tigaret et al., 2016), molecular cascades involving, non exhaustively, the calmodulin kinase CamKII (Malenka et al., 1989), PKC for the induction of LTP (Malinow et al., 1989), and PSD-95 (Ehrlich and Malinow, 2004), dynamic actin filaments (Krucker et al., 2000) and protein synthesis (Steward and Schuman, 2001) for the maintenance of LTP.

The link between LTP and learning & memory has been shown in several publications (Nicoll, 2017) and, in hippocampal neurons, LTP was shown to be induced by learning in vivo during a hippocampus-dependent learning task (Whitlock et al., 2006). Several patterns of activity other than TBS can induce LTP in vivo and in vitro. Thus, together with TBS, these patterns of activity are often referred to « learning-related patterns of activity ».

Of course, synaptic strength is not only able to be increased. Long term depression (LTD), which decreases synaptic strength through internalization of AMPA receptors, is also involved in learning and memory processes where input strength has to be selectively tuned in both ways (Ito et al., 1982; Bear, 1996; Collingridge et al., 2010). Therefore, a memory can be engineered with LTP and LTD (Nabavi et al., 2014).

Importantly, glial cells, in particular, astrocytes, also play a critical role in learning and memory processes (see also 1.4.5). In one way, the coverage of activated synapses by astrocytic processes is enhanced following LTP (Bernardinelli et al., 2014). In the other way, by fueling the synapses with L-lactate deriving from astrocytic glycogenolysis (Pellerin et al., 2007), astrocytes ensure the maintenance of LTP, and disrupting this mechanism induces amnesia (Suzuki et al., 2011). Also, astrocyte calcium signaling is required for

(13)

Despite several thousands of publications already written on this topic, scientific studies on synaptic plasticity still reveal new insights every year.

1.2.2. Spine morphological changes associated with LTP and LTD.

One of the first striking things one can observe when looking at a spiny dendrite is the variability of spine shape. Indeed, one may not be able to find two identical spines (see Figure 1). Additionally, immunohistochemical studies showed that dendritic spines contain high levels of actin filaments, a hallmark of cellular motility (Matus et al., 1982).

Several reports mentioned anomalies in spine shapes in mental diseases (Fiala et al., 2002) indicating a crucial role of spine morphology in cognition. Also, the demonstration by electron microscopy that spine size correlates to the size of the PSD (Harris et al., 1992) and to the number of AMPA receptors (Nusser et al., 1998; Takumi et al., 1999) suggested early that spine shape and LTP might be related (see 2.1). This link was further confirmed by Nicolas Toni, in the group of Dominique Muller, showing that LTP was followed by a transient increase of the proportion of spines having a partitioned PSD (Toni et al., 1999). A few years later, the progress of laser scanning microscopy coupled with transfection of neurons with fluorescent dyes allowed the repetitive visualization of spine structure as a function of time, and revealed that dendritic filopodia and spines undergo continuous morphological variations throughout their lifetime in vitro (Dailey and Smith, 1996; Fischer et al., 1998; Dunaevsky et al., 1999; Matus, 2000) and in vivo (Lendvai et al., 2000). The relationship between spine structural change and spine function was then established by these imaging techniques, linking spine geometry to AMPA receptor expression in hippocampal pyramidal neurons (Matsuzaki et al., 2001). And of importance for our work, a fundamental study finally demonstrated the association between spine enlargement and LTP (Matsuzaki et al., 2004). This finding was since followed by the demonstration that a

(14)

increased our knowledge of the mechanism underlying morphological spine dynamics (for an extensive review, see (Kasai et al., 2010). See also 1.4.3).

1.3. From re-weighting to re-wiring.

In late 90’s and early 2000’s, the development of fluorescent probes (Tsien, 1989;

Zhang et al., 2002) associated with the development of repetitive imaging by 2-Photon laser scanning microscopy (2PLSM) (Denk and Svoboda, 1997), allowed researchers to explore the possibility that neuronal networks could structurally re-wire upon activity in organotypic slice cultures and in vivo. 3 publications came out in 1999, confirming this hypothesis. The group of Karel Svoboda first showed in hippocampal organotypic slice cultures that LTP induces a growth of filopodia within a 50 µm area around the stimulation, in an NMDA-dependent manner (Maletic-Savatic et al., 1999). Almost at the same time, the group of Bonhoeffer detected an increase in the growth of new spines during a few hours in the area of LTP induction (Engert and Bonhoeffer, 1999). Following this, the group of Muller showed by post-hoc EM on identified synapses that LTP is followed by an increase in the number of axonal boutons contacted by several dendritic spines at the same time (Toni et al., 1999).

A few years after, using repetitive 2PLSM imaging on transgenic mice expressing GFP in a subset of Layer-5 pyramidal neurons of the barrel cortex, Joshua Trachtenberg observed a spine turnover for weeks and showed that the trimming of every other whisker (chessboard trimming) was followed by an increase in spine turnover specifically in the barrel cortex which receives sensory inputs from whiskers (Trachtenberg et al., 2002). Four years later, Anthony Holtmaat, from the same group, used again chessboard whisker trimming and in vivo repetitive imaging to show that this novel sensory experience was followed by a stabilization of newly generated spines together with an increasing loss of

(15)

memory, probably by allowing a re-wiring of neuronal networks. Therefore, a better understanding of the rules of spine dynamics became of utmost importance for scientists in this field, and crucial questions needed to be addressed.

1.4. The questions we addressed and the methods we applied to better understand learning-related network structural remodeling.

This section introduces some of our work in this field (see Chapter 2). Other important discoveries made during this period are discussed in Chapter 3.

1.4.1. What are the basics of spine turnover and network remodeling?

Dendritic protrusions are believed to appear most of the time under the form of filopodia and driven by actin motility (Zito et al., 2004) before becoming dendritic spines, acquiring a PSD to become functional and eventually getting stabilized at a low rate (Yuste and Bonhoeffer, 2004). This phenomenon continues to occur in adult life and is compensated by the loss of some spines (Trachtenberg et al., 2002).

1.4.1.1. What is the daily rate of protrusion formation and elimination in hippocampal organotypic slice culture?

1.4.1.2. Is this rate influenced by spine-type?

1.4.1.3. What is the success rate of new protrusion stabilization?

1.4.1.4. What is the time course of this stabilization?

1.4.1.5. Is there a critical period for a filopodium to become a spine and to get stabilized?

(16)

To answer this first set of questions, we needed to be able to observe with a high throughput and for several days the same dendritic branches of undoubtedly healthy neurons in their original immediate environment while having full pharmacological access.

The full pharmacological access and the high throughput implies the use of an in vitro preparation. The observation for several days of neurons in their original immediate environment (neuronal partners and glial cells) narrowed the choice to the hippocampal organotypic slice culture.

The technique of the organotypic slice culture was already mastered in the lab since it was developed by the team of Dominique Muller more than a decade before (Stoppini et al., 1991). With this model of study, the hippocampus is isolated from the brain and cut in thick slices in a way that allows it to culture for about a month. Very importantly, because of its laminar organization the main neuronal networks of the hippocampus are maintained in organotypic slice culture. Together with Paul Klauser and Lorenzo Poglia, two PhD students of the lab, we managed to select an appropriate transfection method that allowed a stable expression of fluorescent protein in neurons. Following this, I set up a 2-photon laser scanning microscope equipped with an electrophysiological recording device and two single photon lasers (to allow 2-channel imaging of dyes of different absorption and emission wavelength). We then determined with preliminary experiments the ideal paradigm of observation to fully preserve the health of imaged neurons for up to one month, and we adapted our timing of observation to the pace of observed spine turnover. Lastly, we chose to use for our analysis a free open source Software developed by Antoine Rosset, a radiologist at the Geneva University Hospital. As this extremely powerful software was initially made for scanner and MRI images we asked Joël Spaltenstein, a talented programmer and MD student of our Faculty, to implement it for our needs. This method is illustrated in Figures 4 and 5. Finally, I developed macros to

(17)

Figure 4. Culture, transfection and imaging of a hippocampal organotypic slice culture.

(A) Picture of a hippocampal thick slice prepared from a neonate rat brain. Scale bar: 1 mm. A schematic drawing indicates the CA1 and the CA3 areas in the hippocampus.

(B) Transfection with a plasmid expressing the enhanced green fluorescent protein (EGFP) with a gene gun after 7 days in vitro (DIV). Scale bar: 400 µm.

(C) 5×-image obtained at DIV 11 with a confocal microscope and a 488 nm laser. Arrow indicates a transfected neuron in CA1. Scale bar: 400 μm.

(D) Maximum Intensity Projection (MIP) image of the neuron indicated in (C), obtained with the 40×

objective lens. Box shows a dendrite of interest. Scale bar: 50 μm.

(E) MIP of the dendrite-of-interest observed with the 40× objective lens with 10× additional zoom. Scale bar:

Figure 12

Figure 4. Culture, transfection and laser scanning imaging of a hippocampal organotypic slice.

A. Picture of a hippocampal thick slice prepared from a neonate rat brain. Scale bar: 1 mm B. Transfection with a plasmid expressing the enhanced green fluorescen protein (EGFP) with a gene gun after 7 days in vitro (DIV). Scale bar: 400 µm.

C. 5×-image obtained at DIV 11 with a confocal microscope and a 488 nm laser. Arrow indicates a transfected neuron in CA1. Scale bar: 400 μm.

D. Maximum Intensity Projection (MIP) image of the neuron indicated in (C), obtained with the 40× objective lens. Box shows a dendrite of interest. Scale bar: 50 μm.

E. MIP of the dendrite-of-interest observed with the 40× objective lens with 10× additional zoom. Scale bar: 2.5 μm. Images are presented as raw from the acquisition software.

Repetitive imaging is done on the same dendritic branch a 5h, 24 h, 48h, 72h and more if 7 days

Gene-gun transfection

A B

C D

E

CA1 CA3

(18)

Figure 5. Analysis of spine dynamics.

(A) Screen capture of the analysis software showing the dendrite imaged in Figure 4 at time-points 0, 5, 24, 48, and 72 h. Images are MIP with volume rendering.

(B) Individual z-axis optical sections of the top part of the dendrite shown in (A) during analysis of spine dynamics. The software allows fast navigating through the z-axis to identify and label individual spines. ROI labels are shown for two spines. Note that the spine labeled 1 M is lost at 24 h, whereas the spine labeled 2 M is conserved until the end of the experiment (72 h time-point). ROI labels are made bigger for illustration purposes.

Adapted form (De Roo and Ribic, 2017)

Figure 10

Figure 5. Analysis of spine dynamics.

(A) Screen capture of the analysis software showing the dendrite imaged in Figure 4 at time-points 0, 5, 24, 48, and 72 h. Images are MIP with volume rendering.

(B) Individual z-axis optical sections of the top part of the dendrite shown in (A) during analysis of spine dynamics. The software allows fast navigating through the z-axis to identify and label individual spines.

ROI labels are shown for two spines. Note that the spine labeled 1 M is lost at 24 h, whereas the spine labeled 2 M is conserved until the end of the experiment (72 h time-point). ROI labels are made bigger for illustration purposes.

A

B

1 M 1 M

2 M 2 M 2 M

2 M 2 M

(19)

We used this technique for the first time to calculate the rate of spine turnover in hippocampal neurons of organotypic slice cultures (1.4.1.1). We showed that the rate of formation and elimination of protrusion is different across spine-type (1.4.1.2). We demonstrated that dendritic spines stabilize at a very low success rate (1.4.1.3), and that this stabilization required (1) a transformation from filopodia or stubby spines to mushroom spines, (2) to occur within the first 24 h and (3) the enlargement of the spine- head and the acquisition of a PSD (1.4.1.3, 1.4.1.4 and 1.4.1.5). Following this, we showed that this stabilization process was significantly impaired when synaptic activity is reduced by blocking AMPA and NMDA receptors (1.4.1.6). Lastly, we showed that the spine turnover rate is lowered with the age of the culture (1.4.1.7). Importantly, spine density always remained stable throughout our imaging sessions and neurons were always shown to stay healthy until the end of our imaging experiments, confirming the validity of our approach.

These results are described in the first article included in this thesis.

Knowing much better about the basics of spine dynamics, we decided to address fundamental questions more directly related to learning and memory.

1.4.2. What is the sequence of events that dendritic spines undergo following the induction of learning-related patterns of activity and what are the functional consequences of these events?

Induction of learning-related patterns of activity (LPA) can trigger new protrusion growth within tens of minutes, such as observed in organotypic slice culture, and new spine growth is also observed in vivo following sensory experience. Also, sensory activity has been shown to increase the stabilization of a pool of newly formed spines and to destabilize a pool of stable spines in vivo (see chapter 1.3). But several questions required answers to determine the specificity of these changes and to determine if they have

(20)

1.4.2.1. For how long can we observe the generation of new protrusions following learning-related patterns of activity?

1.4.2.2. Can we trigger a lasting increase in spine stability by inducing an LTP?

1.4.2.3. What is then the impact of the generation of new spines on the spine density over days? Is it eventually compensated by an equivalent loss of spine?

1.4.2.4. Are the LPA-induced mechanisms of spine destabilization and spine stabilization input-specific?

1.4.2.5. Can we observe an input-specific change in spine morphology following LPA, and what is the following sequence of morphological changes over days?

1.4.2.6. Is the generation of new spines a general reaction for the target neuron or is there any rule for their localization?

1.4.2.7. Do these new spines become integrated functionally in the pre- existing network, and if yes, how fast?

To answer these questions, we decided to upgrade our imaging setup in order to allow the application of LPA in hippocampal neurons and to the study the consecutive spine turnover over the following hours and days in an input specific manner. The design of it is detailed in Figure 6.

(21)

Figure 6. Study of spine dynamics following the induction of a long-term potentiation.

(A) Transfection of a hippocampal slice culture as in Figure 4 but with the red fluorescent protein mRFP.

(B) Bolus loading of the mRFP-transfected neuron with the calcium indicator Fluo-4 AM and placement of an electrode of stimulation on the axonal departure of a group of CA3 neurons (red arrow). These axons make synapses with some dendritic spines of CA1 neurons.

(C) Imaging of a dendrite of interest. Each of its spines are line-scanned in both green and red channel while electrical stimulation pulses are triggered. Spines that respond to the stimulation will be qualified as

« activated spines » by the subsequent Theta Burst Stimulation (TBS). An example of a scanned spine (yellow line) and its response (black arrow) is illustrated.

(D) Induction of an LTP with a TBS is done immediately after Step C. LTP is monitored electrophysiologically.

(E) Repetitive imaging is performed as in Figures 4 & 5. Calcium imaging can be done again to test the functionality of new spines. Further analysis of spine turnover is done blind, without the knowledge of

Figure 11

Figure 6. Study of spine dynamics following the induction of a long-term potentiation.

(A) Transfection of a hippocampal slice culture as in Figure 4 but with the red fluorescent protein mRFP.

(B) Bolus loading of the mRFP-transfected neuron with the calcium indicator Fluo-4 AM and placement of an electrode of stimulation on the axonal departure of a group of CA3 neurons (red arrow). These axons make synapses with some dendritic spines of CA1 neurons.

(C) Imaging of a dendrite of interest. Each of its spines are line-scanned in both green and red channel while electrical stimulation pulses are triggered. Spines that respond to the stimulation will be qualified as

« activated spines » by the subsequent Theta Burst Stimulation (TBS). An example of a scanned spine (yellow line) and its response (black arrow) is illustrated.

(D) Induction of an LTP with a TBS is done immediately after Step C. LTP is monitored electrophysiologically.

(E) Repetitive imaging is performed as in Figure 4. Calcium imaging can be done again to test the functionality of new spines. Further analysis of spine turnover is done blind, without the knowledge of

A B

D E

TBS

Recording electrode Gene-gun

transfection with mRFP

DIV 7

Repetitive imaging as in Figures 4 & 5

5 h 24 h 48 h 72 h

C

On the entire dendritic branch:

identification of activated and non-activated spines with mRFP imaging + calcium imaging

Time (s) Stimulation

" !

Bolus loading with the calcium indicator Fluo-4 AM

Stimulation

DIV 11

(22)

Our experiments allowed us to successfully answer these questions and provided the first direct demonstration that the stability of spines is functionally linked to the plasticity mechanisms that underlie memory.

Specifically, after inducing LTP, we observed: an increase in protrusion formation and elimination that lasted for several days, in which the protrusion formation was dependent of protein synthesis (1.4.2.1), a lasting increase in spine stability (1.4.2.2), a transient increase in protrusion density, followed by a return to previous density at day 3 after the induction of LTP (1.4.2.3), a selective stabilization of spines activated by TBS accompanied by a selective destabilization of non-activated spines (1.4.2.4), and a selective and transient enlargement of the spine-head of activated spines, followed within 24 h by a decrease to values comparable with non-activated spines (1.4.2.5).

Importantly, all these mechanisms were dependent on the activity of NMDA receptor.

Finally, we showed that new spines grow preferentially close to activated pre-existing ones (1.4.2.6), and become functional within 24 h (1.4.2.7).

These results are presented in the second publication inserted in this thesis.

We next wondered what molecular mechanisms could participate in the implementation of spine stabilization induced by learning-related patterns of activity.

1.4.3. Are transmembrane adhesion molecules a part of the molecular mechanisms that mediate the LPA-induced long term stabilization of spines?

N-cadherin (NCad) is a transmembrane adhesion protein that has been shown to be essential for the maintenance of LTP (Tang et al., 1998; Okuda et al., 2007; Arikkath and Reichardt, 2008). In addition, it regulates spine shape (Togashi et al., 2002; Okamura et al., 2004; Elia et al., 2006; Xie et al., 2008; Arikkath et al., 2009). We therefore postulated that

(23)

1.4.3.1. Is NCad expression at spines a determinant of spine shape and PSD expression?

1.4.3.2. Is NCad playing a role in spine turnover rate?

1.4.3.3. Is there a correlation between NCad expression in spine and spine stability?

1.4.3.4. Is NCad expression regulated to selectively target spines activated by TBS?

1.4.3.5. Is NCad function necessary for plasticity-induced spine stabilization?

To answer these questions, we applied the methods that we developed for the study of LPA-induced network remodeling (see 4.2) on hippocampal slice cultures in which we either over-expressed NCad, in order to increase endogenous NCad function, or we over- expressed a defective mutant form of NCad, in order to create a dominant negative effect on endogenous NCad function. We also used a tagged version of NCad, NCad-EGFP to allow localization and repetitive imaging of the NCad protein in dendritic spines (Figure 7).

In this work, detailed in publication 3, we demonstrate that the adhesion molecule NCad plays several determinant roles in structural plasticity. We demonstrate that: NCad is required for the regulation of the spine-head volume and of the PSD size (1.4.3.1), NCad plays a stabilizing role in spine turnover by improving the stabilization of nascent spines through the enhancement of PSD-95 expression in these spines and by maintaining the stability of pre-existing spines (1.4.3.2), NCad is selectively expressed in spines activated by TBS (1.4.3.3), and finally, that NCad is required for TBS-induced selective stabilization of activated spines (1.4.3.4).

(24)

Figure 7. Imaging of NCadherin and spine dynamics.

Left: Maximum Intensity Projection (obtained by imaging done as in Figure 4) of a pyramidal neuron with a plasmid expressing mRFP (in red) and a plasmid expressing NCad-EGFP (in green).

Scale bar: 50 µm

Right: High magnification of a dendritic branch (yellow insert in A) showing spines with, and without, NCad-eGFP. Scale bar: 2 µm

Personal material

We next wondered if we could use our approach to better understand the etiology of the 22q11 deletion syndrome, an autosomal dominant pathology associated with a learning deficit and a high risk of developing schizophrenia.

1.4.4. What are the rewiring mechanisms and the molecular determinants of the spine density deficit observed in the mouse model of the 22q11 deletion syndrome?

As a member of the consortium NCCR-Synapsy, I had the opportunity to participate in the effort towards a better understanding of the 22q11 deletion syndrome (22q11DS), or Di George syndrome, known as the most important genetic cause of schizophrenia (Karayiorgou et al., 1995; Arguello and Gogos, 2012). Transgenic mice

(25)

Arguello and Gogos, 2010; Karayiorgou et al., 2010; Meechan et al., 2015). Interestingly, a spine density deficit has been described in this mouse (Mukai et al., 2008; Fenelon et al., 2013; Xu et al., 2013) but questions about the mechanism behind this deficit remained open in these studies.

Among the genes missing in 22q11DS, zdhhc8 retained our attention because the ZDHHC8 enzyme it codes for regulates, through palmitoylation (Young et al., 2012), the trafficking and the stabilization of an array of other proteins, among which are PSD-95 and the brain- specific cdc42-palm, enriched with dendritic spines. Also, ZDHHC8 over-expression has been shown to almost rescue spine density deficits in dissociated cultures of a mouse model of 22q11DS (Mukai et al., 2008).

We thus applied our method of spine turnover imaging (see 1.4.1 - 1.4.3) on the murine model of the 22q11DS (LgDel Mice) in order to answer the following questions:

1.4.4.1. Is the spine density deficit of LgDel mice due to an impairment of the generation of new spines or to a failure in spine stabilization processes, or both?

1.4.4.2. Once we have identified which impairment of spine dynamics is behind the spine density deficit, can we correct this specific impairment, then eventually the spine density deficit by a genetic rescue of ZDHHC8 activity?

1.4.4.3. Is the haploinsufficiency of ZDHHC8 in LgDel mice sufficient to result in a deficit of PSD-95 palmitoylation level? If yes, can we rescue the spine dynamics impairment and the spine density deficit with a PSD-95 constitutively palmitoylated form?

1.4.4.4. Is the haploinsufficiency of ZDHHC8 in LgDel mice sufficient to result in a deficit of cdc42 palmitoylation level? If yes, can we rescue the spine dynamics impairment and the spine density deficit with a cdc42 constitutively palmitoylated form?

1.4.4.5. Based on our discoveries, can we rescue the spine density deficit of LgDel mice in vivo using a viral vector bringing the missing gear into the neurons of these mice?

(26)

ZDHHC8 in neurons, which also restores spine density to levels measured in wild-type mice (1.4.4.2). We also indicated that the palmitoylation level of PSD-95 is normal, but the palmitoylation level of cdc42 is dramatically lower (1.4.4.3). Following this we demonstrated that the over-expression of a constitutively palmitoylated form of cdc42 (cdc42-palm-CA) rescues the long-term stabilization of spines and rescues the spine density deficit (1.4.4.4). Lastly, we successfully rescued the spine density deficit in LgDel mice in vivo by injecting LgDel pups with a viral construct carrying cdc42-palm-CA, the constitutively active form of cdc42 (1.4.4.5).

1.4.5. Is network structural plasticity only a neuro-neuronal business?

In only a few years, spine dynamics have become the focus of many studies, and it is now well accepted that dendritic spines undergo a continuous turnover in vivo, with a fraction of spines that constantly grow and are often rapidly eliminated, while others get stabilized for months to lifespan (Grutzendler et al., 2002). Therefore, it implies that the cost in energy for triggering learning processes and encoding of long term memory may be higher than expected with simple weight changes upon synaptic plasticity. Although neurons can uptake glucose as energy fuel (Pellerin et al., 2007) the main form of energy storage, glycogen, is located in astrocytes. This knowledge has naturally driven the attention of research on the products of glycogenolysis and their transport from astrocytes to neurons in the context of learning and memory. This led to the discovery that, in rat hippocampus, learning leads to an increase in extracellular levels of L-lactate, that this lactate exclusively derives from astrocytic glycogen breakdown, and is essential for the maintenance LTP in vivo and the formation of long-term memory, but not of short-term memory (Suzuki et al., 2011). Interestingly, Astrocytic L-lactate has also been shown to induce the expression of plasticity-related genes (Yang et al., 2014). Based on this

(27)

1.4.5.1. Using post-hoc electron microscopy, can we detect higher spine density in hippocampal neurons 24 h after a hippocampus-dependent learning task in mice compared to control mice? (see 1.4.2).

1.4.5.2. If yes, can we prevent this spine density increase and impair its associated learning performance by blocking astrocytic glycogen breakdown?

1.4.5.3. Is astrocytic L-Lactate required for some specific changes in spine dynamics induced by learning-related patterns of activity and that underlie long-term memory? (see 1.4.2)

We answered these questions In the last publication incorporated in this thesis. In this article, we showed that an increase in spine density can be detected post-hoc in hippocampal neurons 24 h after a novel object recognition learning task in mice (1.4.5.1).

Next, we demonstrated that when astrocytic gylcogenolysis is blocked, the performance of novel object recognition is impaired and the associated increase in spine density is not detectable anymore. These parameters were rescued when injecting exogenous L-lactate in mice (1.4.5.2). Finally, we revealed that although astrocytic L-lactate is not required by neurons to ensure basal spine turnover, nor for the general decrease in spine stabilization induced by LPA, it is crucial for the LPA-induced generation of new spine, and for the LPA-induced selective stabilization of activated spines (1.4.5.3).

(28)

2. Published studies

Activity-dependent PSD formation and stabilization of newly formed spines in hippocampal slice cultures.

The context of this research article is presented in detail in chapter 1.4.1. In this work, we developed a method to allow long-term repetitive imaging of hippocampal organotypic slice cultures made from rat pus. We aimed to obtain high quality images to undoubtedly discriminate spine types while keeping the neuronal network healthy for weeks. Under these conditions, we could follow individual spines for several days to unravel the fundamentals of spine turnover. Mainly, based on our observations, we calculated the spine formation and elimination daily rate and indicated that these parameters are dependent on the developmental stage. We also revealed under what morphological form new spines are formed and demonstrated that the enlargement of their head and the acquisition of the post-synaptic density protein PSD-95 are two key determinants of their stabilization. Lastly, we showed that the stabilization processes for a newly formed protrusion must occur within 24 h after its birth otherwise the protrusion is eliminated.

This publication was evaluated in Faculty of 1000 and was granted with a « Must Read » mention⁽¹⁾.

(29)

Activity-Dependent PSD Formation and Stabilization of Newly Formed Spines in Hippocampal Slice Cultures

Mathias De Roo, Paul Klauser, Pablo Mendez, Lorenzo Poglia and Dominique Muller

University of Geneva Medical School, Department of Neurosciences, Centre Me´dical Universitaire, 1211 Gene`ve 4, Switzerland

Mathias De Roo and Paul Klauser contributed equally to this work.

Development and remodeling of synaptic networks occurs through a continuous turnover of dendritic spines. However, the mechanisms that regulate the formation and stabilization of newly formed spines remain poorly understood. Here, we applied repetitive confocal imaging to hippocampal slice cultures to address these issues. We find that, although the turnover rate of protrusions progressively decreased during development, the process of stabilization of new spines remained comparable both in terms of time course and low level of efficacy. Irrespective of the developmental stage, most new protrusions were quickly eliminated, in particular filopodia, which only occasionally lead to the formation of stable dendritic spines. We also found that the stabilization of new protrusions was determined within a critical period of 24 h and that this coincided with an enlargement of the spine head and the expression of tagged PSD-95.

Blockade of postsynaptic AMPA and NMDA receptors significantly reduced the capacity of new spines to express tagged PSD-95 and decreased their probability to be stabilized. These results suggest a model in which synaptic development is associated with an extensive, nonspecific growth of protrusions followed by stabilization of a few of them through a mechanism that involves activity- driven formation of a postsynaptic density.

Keywords:confocal imaging, dendritic spine, hippocampus, plasticity, postsynaptic density, synaptogenesis

Introduction

Neuronal activity modulates excitatory synaptic properties and function in many different ways. Inhibition or enhancement of firing or synaptic activity results in an up scaling or down scaling of glutamate sensitivity, an effect that appears to be mainly mediated by changes in receptor expression mechanisms (Turrigiano et al. 1998; Turrigiano and Nelson 2004). Also patterns of synaptic activation, that induce properties of synaptic plasticity, such as long-term potentiation or depression, result in modifications of receptor cycling and expression at the synapse (Nicoll 2003; Kennedy and Ehlers 2006; Nicoll et al.

2006). However, in addition to these homeostatic and activity- dependent regulations of receptor expression, several recent studies have provided evidence for mechanisms of activity- dependent synaptogenesis and synapse remodeling. Confocal analysis of dendritic spines, which are the sites of most excitatory synapses in the brain, showed that they are not as stable structures as previously thought (Yuste and Bonhoeffer 2004;

Segal 2005). In vitro experiments showed that they can be formed de novo within short periods of time as a result of synaptic activation or induction of synaptic plasticity (Engert and Bonhoeffer 1999; Maletic-Savatic et al. 1999; Toni et al. 1999).

Under in vivo conditions, they undergo a continuous turnover and replacement that are developmentally regulated, appear to

be region specific, and modulated by sensory activity (Grutzen- dler et al. 2002; Trachtenberg et al. 2002; Zuo, Lin, et al. 2005;

Zuo, Yang, et al. 2005; Holtmaat et al. 2006). These results have thus strengthened the possibility that learning processes not only involve synapse-specific modifications of synaptic strength but may also be associated with a remodeling of synaptic con- nections through competitive mechanisms of synapse or spine formation, stabilization, or elimination (Stepanyants et al. 2002).

These processes, however, remain poorly understood (Garner et al. 2006). It remains unclear how exactly are new spines formed, what is the efficacy of the process, how fast are new, stable spines generated, what regulates the stabilization of a new protrusion? To address these issues, we developed here a time-lapse confocal analysis applied to hippocampal organotypic slice cultures. We find that turnover in this developmental system affects a high proportion of protrusions over 24 h and shares many similarities with what has been reported in vivo, including its developmental regulation. Furthermore, the ac- cessibility of the system made it possible to analyze important properties of newly formed protrusions. We find that most new protrusions are only transient and only occasionally lead to the formation of stable spines and this with markedly different efficacies depending upon protrusion type. We find also that the stabilization of a new protrusion occurs over a critical period of 24 h, that this process is associated with an increase in the size of the spine head and correlates with activity-driven mechanisms of postsynaptic density (PSD) expression.

Materials and Methods Cultures

Transverse hippocampal organotypic slice cultures (400 lm thick) were prepared from 6- to 7-day-old rats using a protocol approved by the Geneva Veterinarian Office (authorization 31.1.1007/3129/0) and maintained for 11--29 days in culture as described (Stoppini et al. 1991).

Slice cultures of 2 different ages (11 and 25 days in vitro [DIV]) were specifically analyzed in this study because these periods correspond to different phases of synaptic development in slice cultures (Buchs et al.

1993; Collin et al. 1997). In order to facilitate transfer of slice cultures to recording conditions, they were cultured on a small membrane confetti (6--8 mm in diameter, Millipore, Billerica, MA, USA) placed on top of a Millipore insert. In all experiments, slice cultures were maintained in a CO2incubator at 33!C. For the visualization of spines, slice cultures were transfected with a pcDNA3-EGFP (enhanced green fluorescent protein) plasmid using a biolistic method (Helios Gene Gun, Bio-Rad, Hercules, CA, USA) 3 days before the first observation. The number of transfected CA1 pyramidal cells varied between 0 and 5 per slice culture; fluorescence usually started to be expressed after 24--48 h and then remained stable for at least 10--15 days. To examine PSD formation in newly formed spines, slice cultures were cotransfected with a pcDNA--EGFP plasmid and a pcDNA--PSD-95--DsRed2 plasmid. Co- transfection was effective in 15 out of 16 cases.

Cerebral Cortex January 2008;18:151--161 doi:10.1093/cercor/bhm041 Advance Access publication May 20, 2007

For permissions, please e-mail: journals.permissions@oxfordjournals.org

(30)

Confocal Imaging and Analysis

Laser scanning microscopy was realized with an Olympus Fluoview 300 system using a 488-nm Argon laser or a 2-photon laser set at 920 nm (Coherent, Santa Clara, CA, USA). Transfected slice cultures were maintained in a recording chamber under immersion conditions with culture serum for the time of observation (10 min, 25!C) and then transferred back to the incubator. In each experiment, a complete image of the slice was initially obtained with a 53objective to allow precise recognition and localization of the dendritic segment under analysis (one segment per slice culture). Also an image of the entire CA1 pyramidal neuron with steps of 3lm between scans was realized each day using the 403 objective to check for the general morphology and health of the selected neuron. All analyses were carried out on dendritic segments of about 35lm in length and located between 150 and 300lm from the soma and imaged with a 403objective using a 103additional zoom (final definition: 25 pixels per micron; steps between scans: 0.25--0.5 lm). Control experiments showed that this procedure could be applied up to 10 consecutive times without deleterious effects on cell viability, as indicated by absence of cell death, dendritic beadings, or propidium iodide staining.

PSD-95--DsRed2 was imaged in all experiments using a spinning-disk confocal system (Visitron Systems, Puchheim, Germany) and an excita- tion laser set at 568 nm, and imaging was only carried out at the end of the experiment to avoid possibilities of phototoxicity. Images were obtained with a 403objective using z-steps of 0.4 microns with Metamorph software. Maximal intensity projections of the z-stacks were used to analyze the presence or absence of PSD-95--DsRed2 staining. The average intensity of the spine head area (determined in the EGFP image) was measured in the red channel (PSD-95--DsRed2). Positive spines were in average 17%

above the threshold level. Spines were considered positive for PSD-95

staining if this average intensity was more than 5% higher than the back- ground determined in an adjacent area. The minimal average intensity for a positive spine was 11% and the maximal 30%.

All analyses were carried out blind by at least 2 independent observers on z-stacks of raw images using a plugin specifically developed for OsiriX software (http://homepage.mac.com/rossetantoine/osirix/).

Protrusions were classified as filopodia (protrusions without enlargement at the tip), stubby spines (short protrusions without neck and in direct continuity with the dendritic surface), and mushroom spines (protrusions with a neck and an enlargement at the tip). Determination of spine behavior over time was done by systematic analysis of individual z-stacked images. Also quantifications of spines width were carried out by measuring the largest diameter of the spine head observed on any of the z-stacked confocal images. All statistics are given with the standard error of the mean. Standardt-tests were performed to compare Gaussian distributions, and Mann--Whitney tests were used for non-Gaussian distributions. For all tests,awas set to 5%.

Results

Spine Turnover Rate in Hippocampal Slice Cultures To visualize dendritic spines, CA1 pyramidal neurons in slice cultures were transfected with EGFP at either 8 or 22 DIV and imaged repetitively over the next 3--7 days using confocal microscopy. Control experiments showed that repetitive imaging of the same cells and/or dendritic segments for up to 10 times under the conditions used resulted in no obvious signs of toxicity such as beadings or cell death and no staining

Figure 1. Developmental regulation of protrusion turnover in hippocampal slice cultures. (A) Illustration of a typical CA1 pyramidal neuron imaged 3 (D1) and 8 (D5) days after transfection in an 11 DIV hippocampal organotypic culture. (B) Illustration of protrusion turnover obtained by daily confocal imaging of the same dendritic segment. The arrowhead indicates one stable spine; plus and minus signs are examples of new and lost spines, respectively. (C) Bar graph showing the percent of unchanged, stable spines (white bars), stable spines with changes in morphology (dashed bars), newly formed spines (black bars), and lost spines (gray bars) detected over a 24-h period in 11 DIV and 25 DIV hippocampal cultures. Data are mean ± standard error of the mean of analyses made on 10 dendritic segments from 10 pyramidal cells per group;P\0.05. (D) Changes in protrusion density expressed in each experiment as percents of initial values in 11 (open squares) and 25 (black circles) DIV cultures (n510;P\0.05; scale bars:A100lm;

B1lm).

(31)

for propidium iodide. Figure 1Aillustrates one CA1 pyramidal neuron imaged on days 1 and 5, following a daily analysis of the spine changes occurring on one of its dendritic segments (Fig.

1B). As apparent in this example, numerous changes could be detected over a 24-h period, including transformation or modifications of the shape of spines, appearance of filopodia, formation of new spines, and disappearance of preexisting protrusions. Analyses were carried out on 20 different cells from either 11 or 25 DIV hippocampal slice cultures with daily observations made for 5 consecutive days on a total of 978 protrusions. The data, summarized in Figure 1C, indicate that the changes detected over a 24-h period were considerably larger in all aspects examined in 11 as compared with 25 DIV slice cultures, thus clearly pointing to a developmental regulation of spine turnover and plasticity in slice cultures. The proportion of new protrusions averaged 19±2% in 11 DIV slice cultures, but only 8±2% in 25 DIV tissue (n=10 per group;P^<

0.05), whereas the proportion of lost protrusions decreased from 19±3% to 12±2% (n=10;P ^<0.05). Accordingly, the proportion of stable protrusions increased from 81±3% to 88± 2% (n=10;P ^<0.05). Overall, the turnover rate calculated as half of the sum of the ratios of new and lost protrusions observed over 24 h decreased from 19±2% to 10±1% between 11 and 25 DIV (n=10;P ^<0.05).

Interestingly, in addition to new and lost protrusions, we also observed a significant proportion of transformations or changes in spine shape. These included not only transformations of filopodia into stubby or mushroom spines but also transforma-

tions of stubby into mushroom spines (Fig. 3). The proportion of these shape changes also considerably decreased with developmental maturation of slice cultures between DIV 11 and 25 (from 16±2% to 8±2%;n=10;P^<0.05). Finally, as indicated in Figure 1D, all these modifications occurred without detectable changes in protrusion density that remained constant both during the periods of observation (Fig. 1D) and by comparison of 11 and 25 DIV slice cultures (1.02±0.11 vs. 1.06± 0.11 protrusions/1lm, respectively;n=10). All together, these data indicated that dendritic spines in hippocampal slice cultures exhibit a high degree of turnover that is developmentally regulated.

Characteristics of Protrusion Changes

In order to more precisely define the characteristics of new protrusions, we classified them into 3 main categories: filopodia, defined as long protrusions without head or widening at the tip, stubby spines, defined as protrusions without neck, and mushroom spines, protrusions with a neck and a widening at the tip (Fig. 2A). Overall, the proportion between these different protrusions did not vary greatly with development, as the ratios were comparable in 11 and 25 DIV slice cultures, although, as shown in Figure 2B, the proportion of filopodia was slightly but not significantly smaller at 25 DIV (4±2% vs. 8± 3%). To verify this further and test the accuracy of the confocal analyses, we also carried out an electron microscopic determination of the proportion of these different protrusions. For this,

Figure 2. Categorization of spine types in hippocampal cultures. (A) Typical examples of a mushroom spine (left), a stubby spine (center), and a filopodium (right) imaged with confocal microscopy (left panel) and electron microscopy (right panel). (B) Proportion of protrusion types (mushroom spines: black columns; stubby spines: gray columns; filopodia:

white columns) detected on dendritic segments imaged with confocal microscopy in 11 and 25 DIV cultures (n510 cells; 720 protrusions analyzed) and with electron microscopy in 22 DIV cultures (n53 cultures; 99 protrusions analyzed). (C) Proportion of the different types of new protrusions (mushroom spines: black columns; stubby spines: gray columns;

filopodia: white columns) detected using different observation intervals in 11 DIV cultures. Data were obtained through analysis of 9 different cells (25--66 new protrusions analyzed). Note the nonlinearity of results indicating low stability of new protrusions and the high rates of formation of both new spines and filopodia with the shortest interval (scale bar:A0.5lm).

(32)

dendritic segments from 3 hippocampal slice cultures of 22 DIV were analyzed from serial sections (Fig. 2A, right panel) and the proportion of filopodia, stubby, and mushroom spines determined. As illustrated in Figure 2B, the ratios so obtained nicely correlated with the data obtained by confocal microscopy.

Next, we used these categories to analyze the mechanisms through which new spines were generated. For this, we classified all new protrusions detected over different periods of time. As shown on Figure 2C, the number of new protrusions

detected by repetitive imaging showed some variability, but clearly depended upon the period of observation, indicating that some protrusions were likely to have a rather short lifetime.

For this reason, we also carried out analyses with a short time interval (2.5 h). These data indicated that new protrusions mainly appeared as new mushroom spines (50%) or as filopodia (40%) and that the rates of formation for both types of protrusions were in the range of 2% of existing protrusions per hour, which is rather high, but consistent with previous

Figure 3.Fate and stability of new protrusions. (A--D) Illustration of new protrusions fate. (A) New spine that remained stable for[24 h (arrowhead). (B) Filopodium that transformed into a mushroom spine (arrowhead). (C) Transient filopodium. (D) Stubby spine that transformed into a mushroom spine. (E) Quantitative analysis of the fate of a pool of 33 new filopodia analyzed in 7 experiments. Data are expressed as percent of the number of initial filopodia still present as filopodia (open bars) or mushroom spines (black bars) on the next 4 consecutive days. (F) Same, but for a pool of 29 new stubby spines (gray bars) and their transformations in mushroom spines (black bars). (G) Same, but for a pool of 59 new mushroom spines (black bars), among which some could transform back into stubby spines (gray bars). (H) Proportion of protrusion types (F: filopodia; S: stubby spines; M:

mushroom spines) resulting in the formation of stable dendritic spines exhibiting a minimum 48 h stability in 11 DIV cultures. Note the low efficiency of filopodia in generating stable spines (scale bar: 0.5lm).

Synaptic remodeling at the heart of memory processes